use log::*;
use memmap2::MmapMut;
use serde::{Deserialize, Serialize};
use solana_sdk::{
account::Account,
clock::{Epoch, Slot},
hash::Hash,
pubkey::Pubkey,
};
use std::{
fs::{remove_file, OpenOptions},
io,
io::{Seek, SeekFrom, Write},
mem,
path::{Path, PathBuf},
sync::atomic::{AtomicUsize, Ordering},
sync::Mutex,
};
const ALIGN_BOUNDARY_OFFSET: usize = mem::size_of::<u64>();
macro_rules! u64_align {
($addr: expr) => {
($addr + (ALIGN_BOUNDARY_OFFSET - 1)) & !(ALIGN_BOUNDARY_OFFSET - 1)
};
}
const MAXIMUM_APPEND_VEC_FILE_SIZE: usize = 16 * 1024 * 1024 * 1024;
#[derive(Clone, PartialEq, Debug)]
pub struct StoredMeta {
pub write_version: u64,
pub pubkey: Pubkey,
pub data_len: u64,
}
#[derive(Serialize, Deserialize, Clone, Debug, Default, Eq, PartialEq)]
pub struct AccountMeta {
pub lamports: u64,
pub owner: Pubkey,
pub executable: bool,
pub rent_epoch: Epoch,
}
impl<'a> From<&'a Account> for AccountMeta {
fn from(account: &'a Account) -> Self {
Self {
lamports: account.lamports,
owner: account.owner,
executable: account.executable,
rent_epoch: account.rent_epoch,
}
}
}
#[derive(PartialEq, Debug)]
pub struct StoredAccountMeta<'a> {
pub meta: &'a StoredMeta,
pub account_meta: &'a AccountMeta,
pub data: &'a [u8],
pub offset: usize,
pub stored_size: usize,
pub hash: &'a Hash,
}
impl<'a> StoredAccountMeta<'a> {
pub fn clone_account(&self) -> Account {
Account {
lamports: self.account_meta.lamports,
owner: self.account_meta.owner,
executable: self.account_meta.executable,
rent_epoch: self.account_meta.rent_epoch,
data: self.data.to_vec(),
}
}
fn sanitize(&self) -> bool {
self.sanitize_executable() && self.sanitize_lamports()
}
fn sanitize_executable(&self) -> bool {
self.ref_executable_byte() & !1 == 0
}
fn sanitize_lamports(&self) -> bool {
self.account_meta.lamports != 0 || self.clone_account() == Account::default()
}
fn ref_executable_byte(&self) -> &u8 {
let executable_bool: &bool = &self.account_meta.executable;
let executable_byte: &u8 = unsafe { &*(executable_bool as *const bool as *const u8) };
executable_byte
}
}
#[derive(Debug, AbiExample)]
#[allow(clippy::mutex_atomic)]
pub struct AppendVec {
path: PathBuf,
map: MmapMut,
#[allow(clippy::mutex_atomic)]
append_offset: Mutex<usize>,
current_len: AtomicUsize,
file_size: u64,
remove_on_drop: bool,
}
impl Drop for AppendVec {
fn drop(&mut self) {
if self.remove_on_drop {
if let Err(_e) = remove_file(&self.path) {
}
}
}
}
impl AppendVec {
#[allow(clippy::mutex_atomic)]
pub fn new(file: &Path, create: bool, size: usize) -> Self {
let initial_len = 0;
AppendVec::sanitize_len_and_size(initial_len, size).unwrap();
if create {
let _ignored = remove_file(file);
}
let mut data = OpenOptions::new()
.read(true)
.write(true)
.create(create)
.open(file)
.map_err(|e| {
panic!(
"Unable to {} data file {} in current dir({:?}): {:?}",
if create { "create" } else { "open" },
file.display(),
std::env::current_dir(),
e
);
})
.unwrap();
data.seek(SeekFrom::Start((size - 1) as u64)).unwrap();
data.write_all(&[0]).unwrap();
data.seek(SeekFrom::Start(0)).unwrap();
data.flush().unwrap();
let map = unsafe { MmapMut::map_mut(&data) };
let map = map.unwrap_or_else(|e| {
error!(
"Failed to map the data file (size: {}): {}.\n
Please increase sysctl vm.max_map_count or equivalent for your platform.",
size, e
);
std::process::exit(1);
});
AppendVec {
path: file.to_path_buf(),
map,
append_offset: Mutex::new(initial_len),
current_len: AtomicUsize::new(initial_len),
file_size: size as u64,
remove_on_drop: true,
}
}
pub fn set_no_remove_on_drop(&mut self) {
self.remove_on_drop = false;
}
#[allow(clippy::mutex_atomic)]
pub fn new_empty_map(current_len: usize) -> Self {
let map = MmapMut::map_anon(1).unwrap_or_else(|e| {
error!(
"Failed to create VM map for snapshot. {:?}\n
Please increase sysctl vm.max_map_count or equivalent for your platform.",
e
);
std::process::exit(1);
});
AppendVec {
path: PathBuf::from(String::default()),
map,
append_offset: Mutex::new(current_len),
current_len: AtomicUsize::new(current_len),
file_size: 0,
remove_on_drop: true,
}
}
fn sanitize_len_and_size(current_len: usize, file_size: usize) -> io::Result<()> {
if file_size == 0 {
Err(std::io::Error::new(
std::io::ErrorKind::Other,
format!("too small file size {} for AppendVec", file_size),
))
} else if file_size > MAXIMUM_APPEND_VEC_FILE_SIZE {
Err(std::io::Error::new(
std::io::ErrorKind::Other,
format!("too large file size {} for AppendVec", file_size),
))
} else if current_len > file_size {
Err(std::io::Error::new(
std::io::ErrorKind::Other,
format!("current_len is larger than file size ({})", file_size),
))
} else {
Ok(())
}
}
pub fn flush(&self) -> io::Result<()> {
self.map.flush()
}
#[allow(clippy::mutex_atomic)]
pub fn reset(&self) {
let mut offset = self.append_offset.lock().unwrap();
self.current_len.store(0, Ordering::Relaxed);
*offset = 0;
}
pub fn len(&self) -> usize {
self.current_len.load(Ordering::Relaxed)
}
pub fn is_empty(&self) -> bool {
self.len() == 0
}
pub fn capacity(&self) -> u64 {
self.file_size
}
pub fn get_relative_path<P: AsRef<Path>>(append_vec_path: P) -> Option<PathBuf> {
append_vec_path.as_ref().file_name().map(PathBuf::from)
}
pub fn new_relative_path(slot: Slot, id: usize) -> PathBuf {
PathBuf::from(&format!("{}.{}", slot, id))
}
#[allow(clippy::mutex_atomic)]
pub fn new_from_file<P: AsRef<Path>>(path: P, current_len: usize) -> io::Result<(Self, usize)> {
let data = OpenOptions::new()
.read(true)
.write(true)
.create(false)
.open(&path)?;
let file_size = std::fs::metadata(&path)?.len();
AppendVec::sanitize_len_and_size(current_len, file_size as usize)?;
let map = unsafe { MmapMut::map_mut(&data)? };
let new = AppendVec {
path: path.as_ref().to_path_buf(),
map,
append_offset: Mutex::new(current_len),
current_len: AtomicUsize::new(current_len),
file_size,
remove_on_drop: true,
};
let (sanitized, num_accounts) = new.sanitize_layout_and_length();
if !sanitized {
return Err(std::io::Error::new(
std::io::ErrorKind::Other,
"incorrect layout/length/data",
));
}
Ok((new, num_accounts))
}
fn sanitize_layout_and_length(&self) -> (bool, usize) {
let mut offset = 0;
let mut num_accounts = 0;
while let Some((account, next_offset)) = self.get_account(offset) {
if !account.sanitize() {
return (false, num_accounts);
}
offset = next_offset;
num_accounts += 1;
}
let aligned_current_len = u64_align!(self.current_len.load(Ordering::Relaxed));
(offset == aligned_current_len, num_accounts)
}
fn get_slice(&self, offset: usize, size: usize) -> Option<(&[u8], usize)> {
let (next, overflow) = offset.overflowing_add(size);
if overflow || next > self.len() {
return None;
}
let data = &self.map[offset..next];
let next = u64_align!(next);
Some((
unsafe { std::slice::from_raw_parts(data.as_ptr() as *const u8, size) },
next,
))
}
fn append_ptr(&self, offset: &mut usize, src: *const u8, len: usize) {
let pos = u64_align!(*offset);
let data = &self.map[pos..(pos + len)];
unsafe {
let dst = data.as_ptr() as *mut u8;
std::ptr::copy(src, dst, len);
};
*offset = pos + len;
}
fn append_ptrs_locked(&self, offset: &mut usize, vals: &[(*const u8, usize)]) -> Option<usize> {
let mut end = *offset;
for val in vals {
end = u64_align!(end);
end += val.1;
}
if (self.file_size as usize) < end {
return None;
}
let pos = u64_align!(*offset);
for val in vals {
self.append_ptr(offset, val.0, val.1)
}
self.current_len.store(*offset, Ordering::Relaxed);
Some(pos)
}
fn get_type<'a, T>(&self, offset: usize) -> Option<(&'a T, usize)> {
let (data, next) = self.get_slice(offset, mem::size_of::<T>())?;
let ptr: *const T = data.as_ptr() as *const T;
Some((unsafe { &*ptr }, next))
}
pub fn get_account<'a>(&'a self, offset: usize) -> Option<(StoredAccountMeta<'a>, usize)> {
let (meta, next): (&'a StoredMeta, _) = self.get_type(offset)?;
let (account_meta, next): (&'a AccountMeta, _) = self.get_type(next)?;
let (hash, next): (&'a Hash, _) = self.get_type(next)?;
let (data, next) = self.get_slice(next, meta.data_len as usize)?;
let stored_size = next - offset;
Some((
StoredAccountMeta {
meta,
account_meta,
data,
offset,
stored_size,
hash,
},
next,
))
}
pub fn get_account_test(&self, offset: usize) -> Option<(StoredMeta, Account)> {
let (stored_account, _) = self.get_account(offset)?;
let meta = stored_account.meta.clone();
Some((meta, stored_account.clone_account()))
}
pub fn get_path(&self) -> PathBuf {
self.path.clone()
}
pub fn accounts(&self, mut start: usize) -> Vec<StoredAccountMeta> {
let mut accounts = vec![];
while let Some((account, next)) = self.get_account(start) {
accounts.push(account);
start = next;
}
accounts
}
#[allow(clippy::mutex_atomic)]
pub fn append_accounts(
&self,
accounts: &[(StoredMeta, &Account)],
hashes: &[Hash],
) -> Vec<usize> {
let mut offset = self.append_offset.lock().unwrap();
let mut rv = Vec::with_capacity(accounts.len());
for ((stored_meta, account), hash) in accounts.iter().zip(hashes) {
let meta_ptr = stored_meta as *const StoredMeta;
let account_meta = AccountMeta::from(*account);
let account_meta_ptr = &account_meta as *const AccountMeta;
let data_len = stored_meta.data_len as usize;
let data_ptr = account.data.as_ptr();
let hash_ptr = hash.as_ref().as_ptr();
let ptrs = [
(meta_ptr as *const u8, mem::size_of::<StoredMeta>()),
(account_meta_ptr as *const u8, mem::size_of::<AccountMeta>()),
(hash_ptr as *const u8, mem::size_of::<Hash>()),
(data_ptr, data_len),
];
if let Some(res) = self.append_ptrs_locked(&mut offset, &ptrs) {
rv.push(res)
} else {
break;
}
}
rv.push(u64_align!(*offset));
rv
}
pub fn append_account(
&self,
storage_meta: StoredMeta,
account: &Account,
hash: Hash,
) -> Option<usize> {
let res = self.append_accounts(&[(storage_meta, account)], &[hash]);
if res.len() == 1 {
None
} else {
res.first().cloned()
}
}
}
pub mod test_utils {
use super::StoredMeta;
use rand::distributions::Alphanumeric;
use rand::{thread_rng, Rng};
use solana_sdk::account::Account;
use solana_sdk::pubkey::Pubkey;
use std::fs::create_dir_all;
use std::path::PathBuf;
pub struct TempFile {
pub path: PathBuf,
}
impl Drop for TempFile {
fn drop(&mut self) {
let mut path = PathBuf::new();
std::mem::swap(&mut path, &mut self.path);
let _ignored = std::fs::remove_file(path);
}
}
pub fn get_append_vec_dir() -> String {
std::env::var("FARF_DIR").unwrap_or_else(|_| "farf/append_vec_tests".to_string())
}
pub fn get_append_vec_path(path: &str) -> TempFile {
let out_dir = get_append_vec_dir();
let rand_string: String = thread_rng().sample_iter(&Alphanumeric).take(30).collect();
let dir = format!("{}/{}", out_dir, rand_string);
let mut buf = PathBuf::new();
buf.push(&format!("{}/{}", dir, path));
create_dir_all(dir).expect("Create directory failed");
TempFile { path: buf }
}
pub fn create_test_account(sample: usize) -> (StoredMeta, Account) {
let data_len = sample % 256;
let mut account = Account::new(sample as u64, 0, &Pubkey::default());
account.data = (0..data_len).map(|_| data_len as u8).collect();
let stored_meta = StoredMeta {
write_version: 0,
pubkey: Pubkey::default(),
data_len: data_len as u64,
};
(stored_meta, account)
}
}
#[cfg(test)]
pub mod tests {
use super::test_utils::*;
use super::*;
use assert_matches::assert_matches;
use rand::{thread_rng, Rng};
use solana_sdk::timing::duration_as_ms;
use std::time::Instant;
impl AppendVec {
fn append_account_test(&self, data: &(StoredMeta, Account)) -> Option<usize> {
self.append_account(data.0.clone(), &data.1, Hash::default())
}
}
impl<'a> StoredAccountMeta<'a> {
#[allow(clippy::cast_ref_to_mut)]
fn set_data_len_unsafe(&self, new_data_len: u64) {
unsafe {
*(&self.meta.data_len as *const u64 as *mut u64) = new_data_len;
}
}
fn get_executable_byte(&self) -> u8 {
let executable_bool: bool = self.account_meta.executable;
let executable_byte: u8 = unsafe { std::mem::transmute::<bool, u8>(executable_bool) };
executable_byte
}
#[allow(clippy::cast_ref_to_mut)]
fn set_executable_as_byte(&self, new_executable_byte: u8) {
unsafe {
*(&self.account_meta.executable as *const bool as *mut u8) = new_executable_byte;
}
}
}
#[test]
#[should_panic(expected = "too small file size 0 for AppendVec")]
fn test_append_vec_new_bad_size() {
let path = get_append_vec_path("test_append_vec_new_bad_size");
let _av = AppendVec::new(&path.path, true, 0);
}
#[test]
fn test_append_vec_new_from_file_bad_size() {
let file = get_append_vec_path("test_append_vec_new_from_file_bad_size");
let path = &file.path;
let _data = OpenOptions::new()
.read(true)
.write(true)
.create(true)
.open(&path)
.expect("create a test file for mmap");
let result = AppendVec::new_from_file(path, 0);
assert_matches!(result, Err(ref message) if message.to_string() == *"too small file size 0 for AppendVec");
}
#[test]
fn test_append_vec_sanitize_len_and_size_too_small() {
const LEN: usize = 0;
const SIZE: usize = 0;
let result = AppendVec::sanitize_len_and_size(LEN, SIZE);
assert_matches!(result, Err(ref message) if message.to_string() == *"too small file size 0 for AppendVec");
}
#[test]
fn test_append_vec_sanitize_len_and_size_maximum() {
const LEN: usize = 0;
const SIZE: usize = 16 * 1024 * 1024 * 1024;
let result = AppendVec::sanitize_len_and_size(LEN, SIZE);
assert_matches!(result, Ok(_));
}
#[test]
fn test_append_vec_sanitize_len_and_size_too_large() {
const LEN: usize = 0;
const SIZE: usize = 16 * 1024 * 1024 * 1024 + 1;
let result = AppendVec::sanitize_len_and_size(LEN, SIZE);
assert_matches!(result, Err(ref message) if message.to_string() == *"too large file size 17179869185 for AppendVec");
}
#[test]
fn test_append_vec_sanitize_len_and_size_full_and_same_as_current_len() {
const LEN: usize = 1024 * 1024;
const SIZE: usize = 1024 * 1024;
let result = AppendVec::sanitize_len_and_size(LEN, SIZE);
assert_matches!(result, Ok(_));
}
#[test]
fn test_append_vec_sanitize_len_and_size_larger_current_len() {
const LEN: usize = 1024 * 1024 + 1;
const SIZE: usize = 1024 * 1024;
let result = AppendVec::sanitize_len_and_size(LEN, SIZE);
assert_matches!(result, Err(ref message) if message.to_string() == *"current_len is larger than file size (1048576)");
}
#[test]
fn test_append_vec_one() {
let path = get_append_vec_path("test_append");
let av = AppendVec::new(&path.path, true, 1024 * 1024);
let account = create_test_account(0);
let index = av.append_account_test(&account).unwrap();
assert_eq!(av.get_account_test(index).unwrap(), account);
}
#[test]
fn test_append_vec_data() {
let path = get_append_vec_path("test_append_data");
let av = AppendVec::new(&path.path, true, 1024 * 1024);
let account = create_test_account(5);
let index = av.append_account_test(&account).unwrap();
assert_eq!(av.get_account_test(index).unwrap(), account);
let account1 = create_test_account(6);
let index1 = av.append_account_test(&account1).unwrap();
assert_eq!(av.get_account_test(index).unwrap(), account);
assert_eq!(av.get_account_test(index1).unwrap(), account1);
}
#[test]
fn test_append_vec_append_many() {
let path = get_append_vec_path("test_append_many");
let av = AppendVec::new(&path.path, true, 1024 * 1024);
let size = 1000;
let mut indexes = vec![];
let now = Instant::now();
for sample in 0..size {
let account = create_test_account(sample);
let pos = av.append_account_test(&account).unwrap();
assert_eq!(av.get_account_test(pos).unwrap(), account);
indexes.push(pos)
}
trace!("append time: {} ms", duration_as_ms(&now.elapsed()),);
let now = Instant::now();
for _ in 0..size {
let sample = thread_rng().gen_range(0, indexes.len());
let account = create_test_account(sample);
assert_eq!(av.get_account_test(indexes[sample]).unwrap(), account);
}
trace!("random read time: {} ms", duration_as_ms(&now.elapsed()),);
let now = Instant::now();
assert_eq!(indexes.len(), size);
assert_eq!(indexes[0], 0);
let mut accounts = av.accounts(indexes[0]);
assert_eq!(accounts.len(), size);
for (sample, v) in accounts.iter_mut().enumerate() {
let account = create_test_account(sample);
let recovered = v.clone_account();
assert_eq!(recovered, account.1)
}
trace!(
"sequential read time: {} ms",
duration_as_ms(&now.elapsed()),
);
}
#[test]
fn test_relative_path() {
let relative_path = AppendVec::new_relative_path(0, 2);
let full_path = Path::new("/tmp").join(&relative_path);
assert_eq!(
relative_path,
AppendVec::get_relative_path(full_path).unwrap()
);
}
#[test]
fn test_new_from_file_crafted_zero_lamport_account() {
let file = get_append_vec_path("test_append");
let path = &file.path;
let mut av = AppendVec::new(&path, true, 1024 * 1024);
av.set_no_remove_on_drop();
let pubkey = solana_sdk::pubkey::new_rand();
let owner = Pubkey::default();
let data_len = 3_u64;
let mut account = Account::new(0, data_len as usize, &owner);
account.data = b"abc".to_vec();
let stored_meta = StoredMeta {
write_version: 0,
pubkey,
data_len,
};
let account_with_meta = (stored_meta, account);
let index = av.append_account_test(&account_with_meta).unwrap();
assert_eq!(av.get_account_test(index).unwrap(), account_with_meta);
av.flush().unwrap();
let accounts_len = av.len();
drop(av);
let result = AppendVec::new_from_file(path, accounts_len);
assert_matches!(result, Err(ref message) if message.to_string() == *"incorrect layout/length/data");
}
#[test]
fn test_new_from_file_crafted_data_len() {
let file = get_append_vec_path("test_new_from_file_crafted_data_len");
let path = &file.path;
let mut av = AppendVec::new(&path, true, 1024 * 1024);
av.set_no_remove_on_drop();
let crafted_data_len = 1;
av.append_account_test(&create_test_account(10)).unwrap();
let accounts = av.accounts(0);
let account = accounts.first().unwrap();
account.set_data_len_unsafe(crafted_data_len);
assert_eq!(account.meta.data_len, crafted_data_len);
let accounts = av.accounts(0);
let account = accounts.first().unwrap();
assert_eq!(account.meta.data_len, crafted_data_len);
av.flush().unwrap();
let accounts_len = av.len();
drop(av);
let result = AppendVec::new_from_file(path, accounts_len);
assert_matches!(result, Err(ref message) if message.to_string() == *"incorrect layout/length/data");
}
#[test]
fn test_new_from_file_too_large_data_len() {
let file = get_append_vec_path("test_new_from_file_too_large_data_len");
let path = &file.path;
let mut av = AppendVec::new(&path, true, 1024 * 1024);
av.set_no_remove_on_drop();
let too_large_data_len = u64::max_value();
av.append_account_test(&create_test_account(10)).unwrap();
let accounts = av.accounts(0);
let account = accounts.first().unwrap();
account.set_data_len_unsafe(too_large_data_len);
assert_eq!(account.meta.data_len, too_large_data_len);
let accounts = av.accounts(0);
assert_matches!(accounts.first(), None);
av.flush().unwrap();
let accounts_len = av.len();
drop(av);
let result = AppendVec::new_from_file(path, accounts_len);
assert_matches!(result, Err(ref message) if message.to_string() == *"incorrect layout/length/data");
}
#[test]
fn test_new_from_file_crafted_executable() {
let file = get_append_vec_path("test_new_from_crafted_executable");
let path = &file.path;
let mut av = AppendVec::new(&path, true, 1024 * 1024);
av.set_no_remove_on_drop();
av.append_account_test(&create_test_account(10)).unwrap();
{
let mut executable_account = create_test_account(10);
executable_account.1.executable = true;
av.append_account_test(&executable_account).unwrap();
}
let accounts = av.accounts(0);
assert_eq!(*accounts[0].ref_executable_byte(), 0);
assert_eq!(*accounts[1].ref_executable_byte(), 1);
let account = &accounts[0];
let crafted_executable = u8::max_value() - 1;
account.set_executable_as_byte(crafted_executable);
let accounts = av.accounts(0);
let account = accounts.first().unwrap();
{
let executable_bool: &bool = &account.account_meta.executable;
assert_eq!(*executable_bool, false);
const FALSE: bool = false;
if *executable_bool == FALSE {
panic!("This didn't occur if this test passed.");
}
assert_eq!(*account.ref_executable_byte(), crafted_executable);
}
{
let executable_bool: bool = account.account_meta.executable;
assert_eq!(executable_bool, false);
assert_eq!(account.get_executable_byte(), 0);
}
av.flush().unwrap();
let accounts_len = av.len();
drop(av);
let result = AppendVec::new_from_file(path, accounts_len);
assert_matches!(result, Err(ref message) if message.to_string() == *"incorrect layout/length/data");
}
}