solana/programs/bpf_loader/src/serialization.rs

1199 lines
46 KiB
Rust
Raw Normal View History

#![allow(clippy::arithmetic_side_effects)]
2022-03-02 14:50:16 -08:00
use {
byteorder::{ByteOrder, LittleEndian},
solana_program_runtime::invoke_context::SerializedAccountMetadata,
solana_rbpf::{
aligned_memory::{AlignedMemory, Pod},
ebpf::{HOST_ALIGN, MM_INPUT_START},
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
memory_region::{MemoryRegion, MemoryState},
},
solana_sdk::{
bpf_loader_deprecated,
entrypoint::{BPF_ALIGN_OF_U128, MAX_PERMITTED_DATA_INCREASE, NON_DUP_MARKER},
instruction::InstructionError,
pubkey::Pubkey,
system_instruction::MAX_PERMITTED_DATA_LENGTH,
transaction_context::{
BorrowedAccount, IndexOfAccount, InstructionContext, TransactionContext,
},
},
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
std::mem::{self, size_of},
};
/// Maximum number of instruction accounts that can be serialized into the
/// SBF VM.
const MAX_INSTRUCTION_ACCOUNTS: u8 = NON_DUP_MARKER;
enum SerializeAccount<'a> {
Account(IndexOfAccount, BorrowedAccount<'a>),
Duplicate(IndexOfAccount),
}
struct Serializer {
pub buffer: AlignedMemory<HOST_ALIGN>,
regions: Vec<MemoryRegion>,
vaddr: u64,
region_start: usize,
aligned: bool,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
}
impl Serializer {
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
fn new(size: usize, start_addr: u64, aligned: bool, copy_account_data: bool) -> Serializer {
Serializer {
buffer: AlignedMemory::with_capacity(size),
regions: Vec::new(),
region_start: 0,
vaddr: start_addr,
aligned,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data,
}
}
fn fill_write(&mut self, num: usize, value: u8) -> std::io::Result<()> {
self.buffer.fill_write(num, value)
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
pub fn write<T: Pod>(&mut self, value: T) -> u64 {
self.debug_assert_alignment::<T>();
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let vaddr = self
.vaddr
.saturating_add(self.buffer.len() as u64)
.saturating_sub(self.region_start as u64);
// Safety:
// in serialize_parameters_(aligned|unaligned) first we compute the
// required size then we write into the newly allocated buffer. There's
// no need to check bounds at every write.
//
// AlignedMemory::write_unchecked _does_ debug_assert!() that the capacity
// is enough, so in the unlikely case we introduce a bug in the size
// computation, tests will abort.
unsafe {
self.buffer.write_unchecked(value);
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
vaddr
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
fn write_all(&mut self, value: &[u8]) -> u64 {
let vaddr = self
.vaddr
.saturating_add(self.buffer.len() as u64)
.saturating_sub(self.region_start as u64);
// Safety:
// see write() - the buffer is guaranteed to be large enough
unsafe {
self.buffer.write_all_unchecked(value);
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
vaddr
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
fn write_account(
&mut self,
account: &mut BorrowedAccount<'_>,
) -> Result<u64, InstructionError> {
let vm_data_addr = if self.copy_account_data {
let vm_data_addr = self.vaddr.saturating_add(self.buffer.len() as u64);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
self.write_all(account.get_data());
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
vm_data_addr
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
} else {
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
self.push_region(true);
let vaddr = self.vaddr;
self.push_account_data_region(account)?;
vaddr
};
if self.aligned {
let align_offset =
(account.get_data().len() as *const u8).align_offset(BPF_ALIGN_OF_U128);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if self.copy_account_data {
self.fill_write(MAX_PERMITTED_DATA_INCREASE + align_offset, 0)
.map_err(|_| InstructionError::InvalidArgument)?;
} else {
// The deserialization code is going to align the vm_addr to
// BPF_ALIGN_OF_U128. Always add one BPF_ALIGN_OF_U128 worth of
// padding and shift the start of the next region, so that once
// vm_addr is aligned, the corresponding host_addr is aligned
// too.
self.fill_write(MAX_PERMITTED_DATA_INCREASE + BPF_ALIGN_OF_U128, 0)
.map_err(|_| InstructionError::InvalidArgument)?;
self.region_start += BPF_ALIGN_OF_U128.saturating_sub(align_offset);
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
// put the realloc padding in its own region
self.push_region(account.can_data_be_changed().is_ok());
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
}
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
Ok(vm_data_addr)
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
fn push_account_data_region(
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
&mut self,
account: &mut BorrowedAccount<'_>,
) -> Result<(), InstructionError> {
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
if !account.get_data().is_empty() {
let region = match account_data_region_memory_state(account) {
MemoryState::Readable => MemoryRegion::new_readonly(account.get_data(), self.vaddr),
MemoryState::Writable => {
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
MemoryRegion::new_writable(account.get_data_mut()?, self.vaddr)
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
MemoryState::Cow(index_in_transaction) => {
MemoryRegion::new_cow(account.get_data(), self.vaddr, index_in_transaction)
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
};
self.vaddr += region.len;
self.regions.push(region);
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
Ok(())
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
fn push_region(&mut self, writable: bool) {
let range = self.region_start..self.buffer.len();
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let region = if writable {
MemoryRegion::new_writable(
self.buffer.as_slice_mut().get_mut(range.clone()).unwrap(),
self.vaddr,
)
} else {
MemoryRegion::new_readonly(
self.buffer.as_slice().get(range.clone()).unwrap(),
self.vaddr,
)
};
self.regions.push(region);
self.region_start = range.end;
self.vaddr += range.len() as u64;
}
fn finish(mut self) -> (AlignedMemory<HOST_ALIGN>, Vec<MemoryRegion>) {
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
self.push_region(true);
debug_assert_eq!(self.region_start, self.buffer.len());
(self.buffer, self.regions)
}
fn debug_assert_alignment<T>(&self) {
debug_assert!(
!self.aligned
|| self
.buffer
.as_slice()
.as_ptr_range()
.end
.align_offset(mem::align_of::<T>())
== 0
);
}
}
pub fn serialize_parameters(
transaction_context: &TransactionContext,
instruction_context: &InstructionContext,
should_cap_ix_accounts: bool,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
) -> Result<
(
AlignedMemory<HOST_ALIGN>,
Vec<MemoryRegion>,
Vec<SerializedAccountMetadata>,
),
InstructionError,
> {
let num_ix_accounts = instruction_context.get_number_of_instruction_accounts();
if should_cap_ix_accounts && num_ix_accounts > MAX_INSTRUCTION_ACCOUNTS as IndexOfAccount {
return Err(InstructionError::MaxAccountsExceeded);
}
let (program_id, is_loader_deprecated) = {
let program_account =
instruction_context.try_borrow_last_program_account(transaction_context)?;
(
*program_account.get_key(),
*program_account.get_owner() == bpf_loader_deprecated::id(),
)
};
let accounts = (0..instruction_context.get_number_of_instruction_accounts())
.map(|instruction_account_index| {
if let Some(index) = instruction_context
.is_instruction_account_duplicate(instruction_account_index)
.unwrap()
{
SerializeAccount::Duplicate(index)
} else {
let account = instruction_context
.try_borrow_instruction_account(transaction_context, instruction_account_index)
.unwrap();
SerializeAccount::Account(instruction_account_index, account)
}
})
// fun fact: jemalloc is good at caching tiny allocations like this one,
// so collecting here is actually faster than passing the iterator
// around, since the iterator does the work to produce its items each
// time it's iterated on.
.collect::<Vec<_>>();
if is_loader_deprecated {
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
serialize_parameters_unaligned(
accounts,
instruction_context.get_instruction_data(),
&program_id,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data,
)
} else {
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
serialize_parameters_aligned(
accounts,
instruction_context.get_instruction_data(),
&program_id,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data,
)
}
}
pub fn deserialize_parameters(
transaction_context: &TransactionContext,
instruction_context: &InstructionContext,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
buffer: &[u8],
accounts_metadata: &[SerializedAccountMetadata],
) -> Result<(), InstructionError> {
let is_loader_deprecated = *instruction_context
.try_borrow_last_program_account(transaction_context)?
.get_owner()
== bpf_loader_deprecated::id();
let account_lengths = accounts_metadata.iter().map(|a| a.original_data_len);
if is_loader_deprecated {
deserialize_parameters_unaligned(
transaction_context,
instruction_context,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data,
buffer,
account_lengths,
)
} else {
deserialize_parameters_aligned(
transaction_context,
instruction_context,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data,
buffer,
account_lengths,
)
}
}
fn serialize_parameters_unaligned(
accounts: Vec<SerializeAccount>,
instruction_data: &[u8],
program_id: &Pubkey,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
) -> Result<
(
AlignedMemory<HOST_ALIGN>,
Vec<MemoryRegion>,
Vec<SerializedAccountMetadata>,
),
InstructionError,
> {
// Calculate size in order to alloc once
let mut size = size_of::<u64>();
for account in &accounts {
size += 1; // dup
match account {
SerializeAccount::Duplicate(_) => {}
SerializeAccount::Account(_, account) => {
size += size_of::<u8>() // is_signer
+ size_of::<u8>() // is_writable
+ size_of::<Pubkey>() // key
+ size_of::<u64>() // lamports
+ size_of::<u64>() // data len
+ size_of::<Pubkey>() // owner
+ size_of::<u8>() // executable
+ size_of::<u64>(); // rent_epoch
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if copy_account_data {
size += account.get_data().len();
}
}
}
}
size += size_of::<u64>() // instruction data len
+ instruction_data.len() // instruction data
+ size_of::<Pubkey>(); // program id
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
let mut s = Serializer::new(size, MM_INPUT_START, false, copy_account_data);
let mut accounts_metadata: Vec<SerializedAccountMetadata> = Vec::with_capacity(accounts.len());
s.write::<u64>((accounts.len() as u64).to_le());
for account in accounts {
match account {
SerializeAccount::Duplicate(position) => {
accounts_metadata.push(accounts_metadata.get(position as usize).unwrap().clone());
s.write(position as u8);
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
SerializeAccount::Account(_, mut account) => {
s.write::<u8>(NON_DUP_MARKER);
s.write::<u8>(account.is_signer() as u8);
s.write::<u8>(account.is_writable() as u8);
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let vm_key_addr = s.write_all(account.get_key().as_ref());
let vm_lamports_addr = s.write::<u64>(account.get_lamports().to_le());
s.write::<u64>((account.get_data().len() as u64).to_le());
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let vm_data_addr = s.write_account(&mut account)?;
let vm_owner_addr = s.write_all(account.get_owner().as_ref());
s.write::<u8>(account.is_executable() as u8);
2022-11-09 11:39:38 -08:00
s.write::<u64>((account.get_rent_epoch()).to_le());
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
accounts_metadata.push(SerializedAccountMetadata {
original_data_len: account.get_data().len(),
vm_key_addr,
vm_lamports_addr,
vm_owner_addr,
vm_data_addr,
});
}
};
}
s.write::<u64>((instruction_data.len() as u64).to_le());
s.write_all(instruction_data);
s.write_all(program_id.as_ref());
let (mem, regions) = s.finish();
Ok((mem, regions, accounts_metadata))
}
pub fn deserialize_parameters_unaligned<I: IntoIterator<Item = usize>>(
transaction_context: &TransactionContext,
instruction_context: &InstructionContext,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
buffer: &[u8],
account_lengths: I,
) -> Result<(), InstructionError> {
let mut start = size_of::<u64>(); // number of accounts
for (instruction_account_index, pre_len) in (0..instruction_context
.get_number_of_instruction_accounts())
.zip(account_lengths.into_iter())
{
let duplicate =
instruction_context.is_instruction_account_duplicate(instruction_account_index)?;
start += 1; // is_dup
if duplicate.is_none() {
let mut borrowed_account = instruction_context
.try_borrow_instruction_account(transaction_context, instruction_account_index)?;
start += size_of::<u8>(); // is_signer
start += size_of::<u8>(); // is_writable
start += size_of::<Pubkey>(); // key
let lamports = LittleEndian::read_u64(
2022-03-10 11:48:33 -08:00
buffer
.get(start..)
.ok_or(InstructionError::InvalidArgument)?,
);
if borrowed_account.get_lamports() != lamports {
borrowed_account.set_lamports(lamports)?;
}
start += size_of::<u64>() // lamports
+ size_of::<u64>(); // data length
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if copy_account_data {
let data = buffer
.get(start..start + pre_len)
.ok_or(InstructionError::InvalidArgument)?;
// The redundant check helps to avoid the expensive data comparison if we can
match borrowed_account
.can_data_be_resized(data.len())
.and_then(|_| borrowed_account.can_data_be_changed())
{
Ok(()) => borrowed_account.set_data_from_slice(data)?,
Err(err) if borrowed_account.get_data() != data => return Err(err),
_ => {}
}
start += pre_len; // data
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
start += size_of::<Pubkey>() // owner
+ size_of::<u8>() // executable
+ size_of::<u64>(); // rent_epoch
}
}
Ok(())
}
fn serialize_parameters_aligned(
accounts: Vec<SerializeAccount>,
instruction_data: &[u8],
program_id: &Pubkey,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
) -> Result<
(
AlignedMemory<HOST_ALIGN>,
Vec<MemoryRegion>,
Vec<SerializedAccountMetadata>,
),
InstructionError,
> {
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let mut accounts_metadata = Vec::with_capacity(accounts.len());
// Calculate size in order to alloc once
let mut size = size_of::<u64>();
for account in &accounts {
size += 1; // dup
match account {
SerializeAccount::Duplicate(_) => size += 7, // padding to 64-bit aligned
SerializeAccount::Account(_, account) => {
let data_len = account.get_data().len();
size += size_of::<u8>() // is_signer
+ size_of::<u8>() // is_writable
+ size_of::<u8>() // executable
+ size_of::<u32>() // original_data_len
+ size_of::<Pubkey>() // key
+ size_of::<Pubkey>() // owner
+ size_of::<u64>() // lamports
+ size_of::<u64>() // data len
+ MAX_PERMITTED_DATA_INCREASE
+ size_of::<u64>(); // rent epoch
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if copy_account_data {
size += data_len + (data_len as *const u8).align_offset(BPF_ALIGN_OF_U128);
} else {
size += BPF_ALIGN_OF_U128;
}
}
}
}
size += size_of::<u64>() // data len
+ instruction_data.len()
+ size_of::<Pubkey>(); // program id;
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
let mut s = Serializer::new(size, MM_INPUT_START, true, copy_account_data);
// Serialize into the buffer
s.write::<u64>((accounts.len() as u64).to_le());
for account in accounts {
match account {
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
SerializeAccount::Account(_, mut borrowed_account) => {
s.write::<u8>(NON_DUP_MARKER);
s.write::<u8>(borrowed_account.is_signer() as u8);
s.write::<u8>(borrowed_account.is_writable() as u8);
s.write::<u8>(borrowed_account.is_executable() as u8);
s.write_all(&[0u8, 0, 0, 0]);
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let vm_key_addr = s.write_all(borrowed_account.get_key().as_ref());
let vm_owner_addr = s.write_all(borrowed_account.get_owner().as_ref());
let vm_lamports_addr = s.write::<u64>(borrowed_account.get_lamports().to_le());
s.write::<u64>((borrowed_account.get_data().len() as u64).to_le());
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
let vm_data_addr = s.write_account(&mut borrowed_account)?;
2022-11-09 11:39:38 -08:00
s.write::<u64>((borrowed_account.get_rent_epoch()).to_le());
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
accounts_metadata.push(SerializedAccountMetadata {
original_data_len: borrowed_account.get_data().len(),
vm_key_addr,
vm_owner_addr,
vm_lamports_addr,
vm_data_addr,
});
}
SerializeAccount::Duplicate(position) => {
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
accounts_metadata.push(accounts_metadata.get(position as usize).unwrap().clone());
s.write::<u8>(position as u8);
s.write_all(&[0u8, 0, 0, 0, 0, 0, 0]);
}
};
}
s.write::<u64>((instruction_data.len() as u64).to_le());
s.write_all(instruction_data);
s.write_all(program_id.as_ref());
let (mem, regions) = s.finish();
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
Ok((mem, regions, accounts_metadata))
}
pub fn deserialize_parameters_aligned<I: IntoIterator<Item = usize>>(
transaction_context: &TransactionContext,
instruction_context: &InstructionContext,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
copy_account_data: bool,
buffer: &[u8],
account_lengths: I,
) -> Result<(), InstructionError> {
let mut start = size_of::<u64>(); // number of accounts
for (instruction_account_index, pre_len) in (0..instruction_context
.get_number_of_instruction_accounts())
.zip(account_lengths.into_iter())
{
let duplicate =
instruction_context.is_instruction_account_duplicate(instruction_account_index)?;
start += size_of::<u8>(); // position
if duplicate.is_some() {
start += 7; // padding to 64-bit aligned
} else {
let mut borrowed_account = instruction_context
.try_borrow_instruction_account(transaction_context, instruction_account_index)?;
start += size_of::<u8>() // is_signer
+ size_of::<u8>() // is_writable
+ size_of::<u8>() // executable
+ size_of::<u32>() // original_data_len
+ size_of::<Pubkey>(); // key
let owner = buffer
.get(start..start + size_of::<Pubkey>())
.ok_or(InstructionError::InvalidArgument)?;
start += size_of::<Pubkey>(); // owner
let lamports = LittleEndian::read_u64(
2022-03-10 11:48:33 -08:00
buffer
.get(start..)
.ok_or(InstructionError::InvalidArgument)?,
);
if borrowed_account.get_lamports() != lamports {
borrowed_account.set_lamports(lamports)?;
}
start += size_of::<u64>(); // lamports
2022-03-10 11:48:33 -08:00
let post_len = LittleEndian::read_u64(
buffer
.get(start..)
.ok_or(InstructionError::InvalidArgument)?,
) as usize;
start += size_of::<u64>(); // data length
if post_len.saturating_sub(pre_len) > MAX_PERMITTED_DATA_INCREASE
|| post_len > MAX_PERMITTED_DATA_LENGTH as usize
{
return Err(InstructionError::InvalidRealloc);
}
// The redundant check helps to avoid the expensive data comparison if we can
let alignment_offset = (pre_len as *const u8).align_offset(BPF_ALIGN_OF_U128);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if copy_account_data {
let data = buffer
.get(start..start + post_len)
.ok_or(InstructionError::InvalidArgument)?;
match borrowed_account
.can_data_be_resized(post_len)
.and_then(|_| borrowed_account.can_data_be_changed())
{
Ok(()) => borrowed_account.set_data_from_slice(data)?,
Err(err) if borrowed_account.get_data() != data => return Err(err),
_ => {}
}
start += pre_len; // data
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
} else {
// See Serializer::write_account() as to why we have this
// padding before the realloc region here.
start += BPF_ALIGN_OF_U128.saturating_sub(alignment_offset);
let data = buffer
.get(start..start + MAX_PERMITTED_DATA_INCREASE)
.ok_or(InstructionError::InvalidArgument)?;
match borrowed_account
.can_data_be_resized(post_len)
.and_then(|_| borrowed_account.can_data_be_changed())
{
Ok(()) => {
borrowed_account.set_data_length(post_len)?;
let allocated_bytes = post_len.saturating_sub(pre_len);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
if allocated_bytes > 0 {
borrowed_account
.get_data_mut()?
.get_mut(pre_len..pre_len.saturating_add(allocated_bytes))
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
.ok_or(InstructionError::InvalidArgument)?
.copy_from_slice(
data.get(0..allocated_bytes)
.ok_or(InstructionError::InvalidArgument)?,
);
}
}
Err(err) if borrowed_account.get_data().len() != post_len => return Err(err),
_ => {}
}
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
start += MAX_PERMITTED_DATA_INCREASE;
start += alignment_offset;
start += size_of::<u64>(); // rent_epoch
if borrowed_account.get_owner().to_bytes() != owner {
// Change the owner at the end so that we are allowed to change the lamports and data before
borrowed_account.set_owner(owner)?;
}
}
}
Ok(())
}
direct mapping: misc fixes (#32649) * transaction_context: update make_data_mut comment * bpf_loader: cpi: pass SerializeAccountMetadata to CallerAccount::from* We now have a way to provide CallerAccount with trusted values coming from our internal serialization code and not from untrusted vm space * bpf_loader: direct_mapping: enforce account info pointers to be immutable When direct mapping is enabled, we might need to update account data memory regions across CPI calls. Since the only way we have to retrieve the regions is based on their vm addresses, we enforce vm addresses to be stable. Accounts can still be mutated and resized of course, but it must be done in place. This also locks all other AccountInfo pointers, since there's no legitimate reason to make them point to anything else. * bpf_loader: cpi: access ref_to_len_in_vm through VmValue Direct mapping needs to translate vm values at each access since permissions of the underlying memory might have changed. * direct mapping: improve memory permission tracking across CPI calls Ensure that the data and realloc regions of an account always track the account's permissions. In order to do this, we also need to split realloc regions in their own self contained regions, where before we had: [account fields][account data][account realloc + more account fields + next account fields][next account data][...] we now have: [account fields][account data][account realloc][more account fields + next account fields][next account data][...] Tested in TEST_[FORBID|ALLOW]_WRITE_AFTER_OWNERSHIP_CHANGE* Additionally when direct mapping is on, we must update all perms at once before doing account data updates. Otherwise, updating an account might write into another account whose perms we haven't updated yet. Tested in TEST_FORBID_LEN_UPDATE_AFTER_OWNERSHIP_CHANGE. * bpf_loader: serialization: address review comment don't return vm_addr from push_account_region * bpf_loader: rename push_account_region to push_account_data_region * cpi: fix slow edge case zeroing extra account capacity after shrinking an account When returning from CPI we need to zero all the account memory up to the original length only if we know we're potentially dealing with uninitialized memory. When we know that the spare capacity has deterministic content, we only need to zero new_len..prev_len. This fixes a slow edge case that was triggerable by the following scenario: - load a large account (say 10MB) into the vm - shrink to 10 bytes - would memset 10..10MB - shrink to 9 bytes - would memset 9..10MB - shrink to 8 bytes - would memset 8..10MB - ... Now instead in the scenario above the following will happen: - load a large account (say 10MB) into the vm - shrink to 10 bytes - memsets 10..10MB - shrink to 9 bytes - memsets 9..10 - shrink to 8 bytes - memset 8..9 - ... * bpf_loader: add account_data_region_memory_state helper Shared between serialization and CPI to figure out the MemoryState of an account. * cpi: direct_mapping: error out if ref_to_len_in_vm points to account memory If ref_to_len_in_vm is allowed to be in account memory, calles could mutate it, essentially letting callees directly mutate callers memory. * bpf_loader: direct_mapping: map AccessViolation -> InstructionError Return the proper ReadonlyDataModified / ExecutableDataModified / ExternalAccountDataModified depending on where the violation occurs * bpf_loader: cpi: remove unnecessary infallible slice::get call
2023-08-30 02:57:24 -07:00
pub(crate) fn account_data_region_memory_state(account: &BorrowedAccount<'_>) -> MemoryState {
if account.can_data_be_changed().is_ok() {
if account.is_shared() {
MemoryState::Cow(account.get_index_in_transaction() as u64)
} else {
MemoryState::Writable
}
} else {
MemoryState::Readable
}
}
#[cfg(test)]
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
#[allow(clippy::indexing_slicing)]
mod tests {
use {
super::*,
solana_program_runtime::with_mock_invoke_context,
solana_sdk::{
account::{Account, AccountSharedData, WritableAccount},
account_info::AccountInfo,
bpf_loader,
entrypoint::deserialize,
transaction_context::InstructionAccount,
},
std::{
cell::RefCell,
mem::transmute,
rc::Rc,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
slice::{self, from_raw_parts, from_raw_parts_mut},
},
};
#[test]
fn test_serialize_parameters_with_many_accounts() {
struct TestCase {
num_ix_accounts: usize,
append_dup_account: bool,
should_cap_ix_accounts: bool,
expected_err: Option<InstructionError>,
name: &'static str,
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
for copy_account_data in [true] {
for TestCase {
num_ix_accounts,
append_dup_account,
should_cap_ix_accounts,
expected_err,
name,
} in [
TestCase {
name: "serialize max accounts without cap",
num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS),
should_cap_ix_accounts: false,
append_dup_account: false,
expected_err: None,
},
TestCase {
name: "serialize max accounts and append dup without cap",
num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS),
should_cap_ix_accounts: false,
append_dup_account: true,
expected_err: None,
},
TestCase {
name: "serialize max accounts with cap",
num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS),
should_cap_ix_accounts: true,
append_dup_account: false,
expected_err: None,
},
TestCase {
name: "serialize too many accounts with cap",
num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS) + 1,
should_cap_ix_accounts: true,
append_dup_account: false,
expected_err: Some(InstructionError::MaxAccountsExceeded),
},
TestCase {
name: "serialize too many accounts and append dup with cap",
num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS),
should_cap_ix_accounts: true,
append_dup_account: true,
expected_err: Some(InstructionError::MaxAccountsExceeded),
},
// This test case breaks parameter deserialization and can be cleaned up
// when should_cap_ix_accounts is enabled.
//
// TestCase {
// name: "serialize too many accounts and append dup without cap",
// num_ix_accounts: usize::from(MAX_INSTRUCTION_ACCOUNTS) + 1,
// should_cap_ix_accounts: false,
// append_dup_account: true,
// expected_err: None,
// },
] {
let program_id = solana_sdk::pubkey::new_rand();
let mut transaction_accounts = vec![(
program_id,
AccountSharedData::from(Account {
lamports: 0,
data: vec![],
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
owner: bpf_loader::id(),
executable: true,
rent_epoch: 0,
}),
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
)];
for _ in 0..num_ix_accounts {
transaction_accounts.push((
Pubkey::new_unique(),
AccountSharedData::from(Account {
lamports: 0,
data: vec![],
owner: program_id,
executable: false,
rent_epoch: 0,
}),
));
}
let mut instruction_accounts: Vec<_> = (0..num_ix_accounts as IndexOfAccount)
.map(|index_in_callee| InstructionAccount {
index_in_transaction: index_in_callee + 1,
index_in_caller: index_in_callee + 1,
index_in_callee,
is_signer: false,
is_writable: false,
})
.collect();
if append_dup_account {
instruction_accounts.push(instruction_accounts.last().cloned().unwrap());
}
let program_indices = [0];
let instruction_data = vec![];
with_mock_invoke_context!(
invoke_context,
transaction_context,
transaction_accounts
);
invoke_context
.transaction_context
.get_next_instruction_context()
.unwrap()
.configure(&program_indices, &instruction_accounts, &instruction_data);
invoke_context.push().unwrap();
let instruction_context = invoke_context
.transaction_context
.get_current_instruction_context()
.unwrap();
let serialization_result = serialize_parameters(
invoke_context.transaction_context,
instruction_context,
should_cap_ix_accounts,
copy_account_data,
);
assert_eq!(
serialization_result.as_ref().err(),
expected_err.as_ref(),
"{name} test case failed",
);
if expected_err.is_some() {
continue;
}
let (mut serialized, regions, _account_lengths) = serialization_result.unwrap();
let mut serialized_regions = concat_regions(&regions);
let (de_program_id, de_accounts, de_instruction_data) = unsafe {
deserialize(
if copy_account_data {
serialized.as_slice_mut()
} else {
serialized_regions.as_slice_mut()
}
.first_mut()
.unwrap() as *mut u8,
)
};
assert_eq!(de_program_id, &program_id);
assert_eq!(de_instruction_data, &instruction_data);
for account_info in de_accounts {
let index_in_transaction = invoke_context
.transaction_context
.find_index_of_account(account_info.key)
.unwrap();
let account = invoke_context
.transaction_context
.get_account_at_index(index_in_transaction)
.unwrap()
.borrow();
assert_eq!(account.lamports(), account_info.lamports());
assert_eq!(account.data(), &account_info.data.borrow()[..]);
assert_eq!(account.owner(), account_info.owner);
assert_eq!(account.executable(), account_info.executable);
assert_eq!(account.rent_epoch(), account_info.rent_epoch);
}
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
}
}
#[test]
fn test_serialize_parameters() {
for copy_account_data in [false, true] {
let program_id = solana_sdk::pubkey::new_rand();
let transaction_accounts = vec![
(
program_id,
AccountSharedData::from(Account {
lamports: 0,
data: vec![],
owner: bpf_loader::id(),
executable: true,
rent_epoch: 0,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 1,
data: vec![1u8, 2, 3, 4, 5],
owner: bpf_loader::id(),
executable: false,
rent_epoch: 100,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 2,
data: vec![11u8, 12, 13, 14, 15, 16, 17, 18, 19],
owner: bpf_loader::id(),
executable: true,
rent_epoch: 200,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 3,
data: vec![],
owner: bpf_loader::id(),
executable: false,
rent_epoch: 3100,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 4,
data: vec![1u8, 2, 3, 4, 5],
owner: bpf_loader::id(),
executable: false,
rent_epoch: 100,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 5,
data: vec![11u8, 12, 13, 14, 15, 16, 17, 18, 19],
owner: bpf_loader::id(),
executable: true,
rent_epoch: 200,
}),
),
(
solana_sdk::pubkey::new_rand(),
AccountSharedData::from(Account {
lamports: 6,
data: vec![],
owner: bpf_loader::id(),
executable: false,
rent_epoch: 3100,
}),
),
];
let instruction_accounts: Vec<InstructionAccount> = [1, 1, 2, 3, 4, 4, 5, 6]
.into_iter()
.enumerate()
.map(
|(index_in_instruction, index_in_transaction)| InstructionAccount {
index_in_transaction,
index_in_caller: index_in_transaction,
index_in_callee: index_in_transaction - 1,
is_signer: false,
is_writable: index_in_instruction >= 4,
},
)
.collect();
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
let instruction_data = vec![1u8, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11];
let program_indices = [0];
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
let mut original_accounts = transaction_accounts.clone();
with_mock_invoke_context!(invoke_context, transaction_context, transaction_accounts);
invoke_context
.transaction_context
.get_next_instruction_context()
.unwrap()
.configure(&program_indices, &instruction_accounts, &instruction_data);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
invoke_context.push().unwrap();
let instruction_context = invoke_context
.transaction_context
.get_current_instruction_context()
.unwrap();
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
// check serialize_parameters_aligned
let (mut serialized, regions, accounts_metadata) = serialize_parameters(
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
invoke_context.transaction_context,
instruction_context,
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
true,
copy_account_data,
)
.unwrap();
let mut serialized_regions = concat_regions(&regions);
if copy_account_data {
assert_eq!(serialized.as_slice(), serialized_regions.as_slice());
}
let (de_program_id, de_accounts, de_instruction_data) = unsafe {
deserialize(
if copy_account_data {
serialized.as_slice_mut()
} else {
serialized_regions.as_slice_mut()
}
.first_mut()
.unwrap() as *mut u8,
)
};
assert_eq!(&program_id, de_program_id);
assert_eq!(instruction_data, de_instruction_data);
assert_eq!(
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
(de_instruction_data.first().unwrap() as *const u8).align_offset(BPF_ALIGN_OF_U128),
0
);
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
for account_info in de_accounts {
let index_in_transaction = invoke_context
.transaction_context
.find_index_of_account(account_info.key)
.unwrap();
let account = invoke_context
.transaction_context
.get_account_at_index(index_in_transaction)
.unwrap()
.borrow();
assert_eq!(account.lamports(), account_info.lamports());
assert_eq!(account.data(), &account_info.data.borrow()[..]);
assert_eq!(account.owner(), account_info.owner);
assert_eq!(account.executable(), account_info.executable);
assert_eq!(account.rent_epoch(), account_info.rent_epoch);
assert_eq!(
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
(*account_info.lamports.borrow() as *const u64).align_offset(BPF_ALIGN_OF_U128),
0
);
assert_eq!(
account_info
.data
.borrow()
.as_ptr()
.align_offset(BPF_ALIGN_OF_U128),
0
);
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
deserialize_parameters(
invoke_context.transaction_context,
instruction_context,
copy_account_data,
serialized.as_slice(),
&accounts_metadata,
)
.unwrap();
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
for (index_in_transaction, (_key, original_account)) in
original_accounts.iter().enumerate()
{
let account = invoke_context
.transaction_context
.get_account_at_index(index_in_transaction as IndexOfAccount)
.unwrap()
.borrow();
assert_eq!(&*account, original_account);
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
// check serialize_parameters_unaligned
original_accounts
.first_mut()
.unwrap()
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
.1
.set_owner(bpf_loader_deprecated::id());
invoke_context
.transaction_context
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
.get_account_at_index(0)
.unwrap()
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
.borrow_mut()
.set_owner(bpf_loader_deprecated::id());
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
let (mut serialized, regions, account_lengths) = serialize_parameters(
invoke_context.transaction_context,
instruction_context,
true,
copy_account_data,
)
.unwrap();
let mut serialized_regions = concat_regions(&regions);
let (de_program_id, de_accounts, de_instruction_data) = unsafe {
deserialize_unaligned(
if copy_account_data {
serialized.as_slice_mut()
} else {
serialized_regions.as_slice_mut()
}
.first_mut()
.unwrap() as *mut u8,
)
};
assert_eq!(&program_id, de_program_id);
assert_eq!(instruction_data, de_instruction_data);
for account_info in de_accounts {
let index_in_transaction = invoke_context
.transaction_context
.find_index_of_account(account_info.key)
.unwrap();
let account = invoke_context
.transaction_context
.get_account_at_index(index_in_transaction)
.unwrap()
.borrow();
assert_eq!(account.lamports(), account_info.lamports());
assert_eq!(account.data(), &account_info.data.borrow()[..]);
assert_eq!(account.owner(), account_info.owner);
assert_eq!(account.executable(), account_info.executable);
assert_eq!(account.rent_epoch(), account_info.rent_epoch);
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
deserialize_parameters(
invoke_context.transaction_context,
instruction_context,
copy_account_data,
serialized.as_slice(),
&account_lengths,
)
.unwrap();
for (index_in_transaction, (_key, original_account)) in
original_accounts.iter().enumerate()
{
let account = invoke_context
.transaction_context
.get_account_at_index(index_in_transaction as IndexOfAccount)
.unwrap()
.borrow();
assert_eq!(&*account, original_account);
}
}
}
// the old bpf_loader in-program deserializer bpf_loader::id()
#[deny(unsafe_op_in_unsafe_fn)]
pub unsafe fn deserialize_unaligned<'a>(
input: *mut u8,
) -> (&'a Pubkey, Vec<AccountInfo<'a>>, &'a [u8]) {
// this boring boilerplate struct is needed until inline const...
struct Ptr<T>(std::marker::PhantomData<T>);
impl<T> Ptr<T> {
const COULD_BE_UNALIGNED: bool = std::mem::align_of::<T>() > 1;
#[inline(always)]
fn read_possibly_unaligned(input: *mut u8, offset: usize) -> T {
unsafe {
let src = input.add(offset) as *const T;
if Self::COULD_BE_UNALIGNED {
src.read_unaligned()
} else {
src.read()
}
}
}
// rustc inserts debug_assert! for misaligned pointer dereferences when
// deserializing, starting from [1]. so, use std::mem::transmute as the last resort
// while preventing clippy from complaining to suggest not to use it.
// [1]: https://github.com/rust-lang/rust/commit/22a7a19f9333bc1fcba97ce444a3515cb5fb33e6
// as for the ub nature of the misaligned pointer dereference, this is
// acceptable in this code, given that this is cfg(test) and it's cared only with
// x86-64 and the target only incurs some performance penalty, not like segfaults
// in other targets.
#[inline(always)]
fn ref_possibly_unaligned<'a>(input: *mut u8, offset: usize) -> &'a T {
#[allow(clippy::transmute_ptr_to_ref)]
unsafe {
transmute(input.add(offset) as *const T)
}
}
// See ref_possibly_unaligned's comment
#[inline(always)]
fn mut_possibly_unaligned<'a>(input: *mut u8, offset: usize) -> &'a mut T {
#[allow(clippy::transmute_ptr_to_ref)]
unsafe {
transmute(input.add(offset) as *mut T)
}
}
}
let mut offset: usize = 0;
// number of accounts present
let num_accounts = Ptr::<u64>::read_possibly_unaligned(input, offset) as usize;
offset += size_of::<u64>();
// account Infos
let mut accounts = Vec::with_capacity(num_accounts);
for _ in 0..num_accounts {
let dup_info = Ptr::<u8>::read_possibly_unaligned(input, offset);
offset += size_of::<u8>();
if dup_info == NON_DUP_MARKER {
let is_signer = Ptr::<u8>::read_possibly_unaligned(input, offset) != 0;
offset += size_of::<u8>();
let is_writable = Ptr::<u8>::read_possibly_unaligned(input, offset) != 0;
offset += size_of::<u8>();
let key = Ptr::<Pubkey>::ref_possibly_unaligned(input, offset);
offset += size_of::<Pubkey>();
let lamports = Rc::new(RefCell::new(Ptr::mut_possibly_unaligned(input, offset)));
offset += size_of::<u64>();
let data_len = Ptr::<u64>::read_possibly_unaligned(input, offset) as usize;
offset += size_of::<u64>();
let data = Rc::new(RefCell::new(unsafe {
from_raw_parts_mut(input.add(offset), data_len)
}));
offset += data_len;
let owner: &Pubkey = Ptr::<Pubkey>::ref_possibly_unaligned(input, offset);
offset += size_of::<Pubkey>();
let executable = Ptr::<u8>::read_possibly_unaligned(input, offset) != 0;
offset += size_of::<u8>();
let rent_epoch = Ptr::<u64>::read_possibly_unaligned(input, offset);
offset += size_of::<u64>();
accounts.push(AccountInfo {
2021-04-18 10:27:36 -07:00
key,
is_signer,
is_writable,
lamports,
data,
owner,
executable,
rent_epoch,
});
} else {
// duplicate account, clone the original
2022-03-10 11:48:33 -08:00
accounts.push(accounts.get(dup_info as usize).unwrap().clone());
}
}
// instruction data
let instruction_data_len = Ptr::<u64>::read_possibly_unaligned(input, offset) as usize;
offset += size_of::<u64>();
let instruction_data = unsafe { from_raw_parts(input.add(offset), instruction_data_len) };
offset += instruction_data_len;
// program Id
let program_id = Ptr::<Pubkey>::ref_possibly_unaligned(input, offset);
(program_id, accounts, instruction_data)
}
Account data direct mapping (#28053) * AccountSharedData: make data_mut() private This ensures that the inner Vec is never handed out. This is in preparation of enforcing that the capacity of the inner vec never shrinks, which is required for direct mapping. * Adds the feature bpf_account_data_direct_mapping. * Remaps EbpfError::AccessViolation into InstructionError::ReadonlyDataModified. * WIP: Memory regions for each instruction account in create_vm(). * Fix serialization benches, run both copy and !copy variants * rbpf-cli: fix build * BorrowedAccount: ensure that account capacity is never reduced Accounts can be directly mapped in address space. Their capacity can't be reduced mid transaction as that would create holes in vm address space that point to invalid host memory. * bpf_load: run serialization tests for both copy and !copy account data * bpf_loader: add Serializer::write_account * fix lints * BorrowedAccount: make_data_mut is host only * Fix unused import warning * Fix lints * cpi: add explicit direct_mapping arg to update_(callee|caller)_account * cpi: rename account_data_or_only_realloc_padding to serialized_data * cpi: add CallerAccount::original_data_len comment * cpi: add update_callee_account direct_mapping test * cpi: add test_update_caller_account_data_direct_mapping and fix bug We used to have a bug in zeroing data when shrinking account, where we zeroed the spare account capacity but not the realloc padding. * cpi: add tests for mutated readonly accounts * cpi: update_caller_account doesn't need to change .serialized_data when direct_mapping is on * cpi: update_caller_account: ensure that account capacity is always enough Introduce a better way to ensure that account capacity never goes below what might be mapped in memory regions. * cpi: zero account capacity using the newly introduced BorrowedAccount::spare_data_capacity_mut() Before we were using BorrowedAccount::get_data_mut() to get the base pointer to the account data, then we were slicing the spare capacity from it. Calling get_data_mut() doesn't work if an account has been closed tho, since the current program doesn't own the account anymore and therefore get_data_mut() errors out. * bpf_loader: fix same lint for the umpteenth time * bpf_loader: map AccessViolation to ReadonlyDataModified only for account region violations * programs/sbf: realloc: add test for large write after realloc Add a test that after a realloc does a large write that spans the original account length and the realloc area. This ensures that memory mapping works correctly across the boundary. * programs/sbf: run test_program_sbf_realloc with both direct_mapping on and off By default test banks test with all features on. This ensures we keep testing the existing code until the new feature is enabled. * bpf_loader: tweak memcmp syscall Split the actual memcmp code in a separate function. Remove check indexing the slices since the slices are guaranteed to have the correct length by construction. * bpf_loader: tweak the memset syscall Use slice::fill, which is effectively memset. * bpf_loader: syscalls: update mem syscalls to work with non contiguous memory With direct mapping enabled, accounts can now span multiple memory regions. * fix lint, rebase mem_ops * Implement CoW for writable accounts * Fix CI * Move CoW to the MemoryMapping level * Update after rbpf API change * Fix merge screwup * Add create_vm macro. Fix benches. * cpi: simplify update_caller_account Simplify the logic to update a caller's memory region when a callee causes an account data pointer to change (eg during CoW) * benches/bpf_loader: move serialization out of create_vm bench * benches/bpf_loader: don't copy accounts when direct mapping is on * Fix review nits * bpf_loader: mem_ops: handle u64 overflow in MemoryChunkIterator::new When starting at u64::MAX, the chunk iterator would always return the empty sequence (None on the first next()) call, instead of returning a memory access violation. Use checked instead of saturating arithmetic to detect the condition and error out. This commit also adds more tests around boundary conditions. * Fix loader-v3 tests: data_mut => data_as_mut_slice * Fix CI * bpf_loader: fix tuner bench: account must be writable With direct mapping on, invalid writes are caught early meaning the tuner would fail on the first store and not consume the whole budget like the benchmark expects. --------- Co-authored-by: Alexander Meißner <AlexanderMeissner@gmx.net>
2023-04-28 13:54:39 -07:00
fn concat_regions(regions: &[MemoryRegion]) -> AlignedMemory<HOST_ALIGN> {
let len = regions.iter().fold(0, |len, region| len + region.len) as usize;
let mut mem = AlignedMemory::zero_filled(len);
for region in regions {
let host_slice = unsafe {
slice::from_raw_parts(region.host_addr.get() as *const u8, region.len as usize)
};
mem.as_slice_mut()[(region.vm_addr - MM_INPUT_START) as usize..][..region.len as usize]
.copy_from_slice(host_slice)
}
mem
}
}