proxmox-backup/src/backup/chunk_store.rs

394 lines
12 KiB
Rust
Raw Normal View History

use failure::*;
use std::path::{Path, PathBuf};
use std::io::{Read, Write};
use std::time::Duration;
2018-12-07 13:44:56 +00:00
2018-12-22 13:31:59 +00:00
use openssl::sha;
2018-12-07 15:12:45 +00:00
use std::sync::Mutex;
2018-12-19 09:02:24 +00:00
use std::fs::File;
use std::os::unix::io::AsRawFd;
2018-12-19 09:02:24 +00:00
use crate::tools;
pub struct GarbageCollectionStatus {
pub used_bytes: usize,
pub used_chunks: usize,
pub disk_bytes: usize,
pub disk_chunks: usize,
}
impl Default for GarbageCollectionStatus {
fn default() -> Self {
GarbageCollectionStatus {
used_bytes: 0,
used_chunks: 0,
disk_bytes: 0,
disk_chunks: 0,
}
}
}
/// File system based chunk store
pub struct ChunkStore {
2018-12-19 12:40:26 +00:00
name: String, // used for error reporting
pub (crate) base: PathBuf,
chunk_dir: PathBuf,
2018-12-07 15:12:45 +00:00
mutex: Mutex<bool>,
2018-12-19 11:49:23 +00:00
_lockfile: File,
2018-12-07 13:44:56 +00:00
}
2018-12-22 14:39:05 +00:00
// TODO: what about sysctl setting vm.vfs_cache_pressure (0 - 100) ?
pub fn verify_chunk_size(size: u64) -> Result<(), Error> {
static SIZES: [u64; 7] = [64*1024, 128*1024, 256*1024, 512*1024, 1024*1024, 2048*1024, 4096*1024];
if !SIZES.contains(&size) {
bail!("Got unsupported chunk size '{}'", size);
}
Ok(())
}
2018-12-19 08:51:33 +00:00
fn digest_to_prefix(digest: &[u8]) -> PathBuf {
2018-12-07 13:44:56 +00:00
let mut buf = Vec::<u8>::with_capacity(2+1+2+1);
2018-12-07 13:44:56 +00:00
const HEX_CHARS: &'static [u8; 16] = b"0123456789abcdef";
2018-12-07 13:44:56 +00:00
buf.push(HEX_CHARS[(digest[0] as usize) >> 4]);
buf.push(HEX_CHARS[(digest[0] as usize) &0xf]);
buf.push(HEX_CHARS[(digest[1] as usize) >> 4]);
2018-12-07 13:44:56 +00:00
buf.push(HEX_CHARS[(digest[1] as usize) & 0xf]);
buf.push('/' as u8);
let path = unsafe { String::from_utf8_unchecked(buf)};
path.into()
}
impl ChunkStore {
fn chunk_dir<P: AsRef<Path>>(path: P) -> PathBuf {
let mut chunk_dir: PathBuf = PathBuf::from(path.as_ref());
chunk_dir.push(".chunks");
chunk_dir
}
2018-12-19 12:40:26 +00:00
pub fn create<P: Into<PathBuf>>(name: &str, path: P) -> Result<Self, Error> {
let base: PathBuf = path.into();
if !base.is_absolute() {
bail!("expected absolute path - got {:?}", base);
}
let chunk_dir = Self::chunk_dir(&base);
2018-12-08 07:38:42 +00:00
if let Err(err) = std::fs::create_dir(&base) {
2018-12-19 12:40:26 +00:00
bail!("unable to create chunk store '{}' at {:?} - {}", name, base, err);
2018-12-08 07:38:42 +00:00
}
if let Err(err) = std::fs::create_dir(&chunk_dir) {
2018-12-19 12:40:26 +00:00
bail!("unable to create chunk store '{}' subdir {:?} - {}", name, chunk_dir, err);
2018-12-08 07:38:42 +00:00
}
// create 64*1024 subdirs
let mut last_percentage = 0;
for i in 0..64*1024 {
2018-12-16 12:52:16 +00:00
let mut l1path = chunk_dir.clone();
l1path.push(format!("{:04x}", i));
2018-12-08 07:38:42 +00:00
if let Err(err) = std::fs::create_dir(&l1path) {
2018-12-19 12:40:26 +00:00
bail!("unable to create chunk store '{}' subdir {:?} - {}", name, l1path, err);
2018-12-08 07:38:42 +00:00
}
let percentage = (i*100)/(64*1024);
if percentage != last_percentage {
eprintln!("Percentage done: {}", percentage);
last_percentage = percentage;
}
}
2018-12-19 12:40:26 +00:00
Self::open(name, base)
}
2018-12-19 12:40:26 +00:00
pub fn open<P: Into<PathBuf>>(name: &str, path: P) -> Result<Self, Error> {
let base: PathBuf = path.into();
if !base.is_absolute() {
bail!("expected absolute path - got {:?}", base);
}
let chunk_dir = Self::chunk_dir(&base);
2018-12-09 08:42:17 +00:00
if let Err(err) = std::fs::metadata(&chunk_dir) {
2018-12-19 12:40:26 +00:00
bail!("unable to open chunk store '{}' at {:?} - {}", name, chunk_dir, err);
2018-12-09 08:42:17 +00:00
}
let mut lockfile_path = base.clone();
lockfile_path.push(".lock");
// make sure only one process/thread/task can use it
let lockfile = tools::open_file_locked(
lockfile_path, Duration::from_secs(10))?;
2018-12-08 07:21:20 +00:00
Ok(ChunkStore {
2018-12-19 12:40:26 +00:00
name: name.to_owned(),
2018-12-08 07:21:20 +00:00
base,
chunk_dir,
_lockfile: lockfile,
2018-12-08 07:21:20 +00:00
mutex: Mutex::new(false)
})
}
2019-01-04 11:51:43 +00:00
pub fn touch_chunk(&self, digest:&[u8]) -> Result<(), Error> {
2018-12-18 10:06:03 +00:00
2018-12-25 12:29:27 +00:00
let mut chunk_path = self.chunk_dir.clone();
2018-12-19 08:51:33 +00:00
let prefix = digest_to_prefix(&digest);
2018-12-18 10:06:03 +00:00
chunk_path.push(&prefix);
let digest_str = tools::digest_to_hex(&digest);
2018-12-18 10:06:03 +00:00
chunk_path.push(&digest_str);
2018-12-25 10:59:02 +00:00
const UTIME_NOW: i64 = ((1 << 30) - 1);
const UTIME_OMIT: i64 = ((1 << 30) - 2);
2018-12-25 12:29:27 +00:00
let times: [libc::timespec; 2] = [
2018-12-25 10:59:02 +00:00
libc::timespec { tv_sec: 0, tv_nsec: UTIME_NOW },
libc::timespec { tv_sec: 0, tv_nsec: UTIME_OMIT }
];
use nix::NixPath;
let res = chunk_path.with_nix_path(|cstr| unsafe {
libc::utimensat(-1, cstr.as_ptr(), &times[0], libc::AT_SYMLINK_NOFOLLOW)
})?;
if let Err(err) = nix::errno::Errno::result(res) {
bail!("updata atime failed for chunk {:?} - {}", chunk_path, err);
}
2018-12-18 10:06:03 +00:00
Ok(())
}
pub fn read_chunk(&self, digest:&[u8], buffer: &mut Vec<u8>) -> Result<(), Error> {
let mut chunk_path = self.chunk_dir.clone();
let prefix = digest_to_prefix(&digest);
chunk_path.push(&prefix);
let digest_str = tools::digest_to_hex(&digest);
chunk_path.push(&digest_str);
let f = std::fs::File::open(&chunk_path)?;
let mut decoder = lz4::Decoder::new(f)?;
decoder.read_to_end(buffer)?;
Ok(())
}
pub fn get_chunk_iterator(
&self,
print_percentage: bool,
) -> Result<
impl Iterator<Item = Result<tools::fs::ReadDirEntry, Error>> + std::iter::FusedIterator,
Error
> {
use nix::dir::Dir;
use nix::fcntl::OFlag;
use nix::sys::stat::Mode;
let base_handle = match Dir::open(
&self.chunk_dir, OFlag::O_RDONLY, Mode::empty()) {
Ok(h) => h,
Err(err) => bail!("unable to open store '{}' chunk dir {:?} - {}",
self.name, self.chunk_dir, err),
};
let mut verbose = true;
let mut last_percentage = 0;
Ok((0..0x10000).filter_map(move |index| {
if print_percentage {
let percentage = (index * 100) / 0x10000;
if last_percentage != percentage {
last_percentage = percentage;
eprintln!("percentage done: {}", percentage);
}
}
let subdir: &str = &format!("{:04x}", index);
match tools::fs::read_subdir(base_handle.as_raw_fd(), subdir) {
Err(e) => {
if verbose {
eprintln!("Error iterating through chunks: {}", e);
verbose = false;
}
None
}
Ok(iter) => Some(iter),
}
})
.flatten()
.filter(|entry| {
// Check that the file name is actually a hash! (64 hex digits)
let entry = match entry {
Err(_) => return true, // pass errors onwards
Ok(ref entry) => entry,
};
let bytes = entry.file_name().to_bytes();
if bytes.len() != 64 {
return false;
}
for b in bytes {
if !b.is_ascii_hexdigit() {
return false;
}
}
true
}))
}
pub fn sweep_unused_chunks(&self, status: &mut GarbageCollectionStatus) -> Result<(), Error> {
use nix::sys::stat::fstatat;
2018-12-19 08:51:33 +00:00
let now = unsafe { libc::time(std::ptr::null_mut()) };
for entry in self.get_chunk_iterator(true)? {
let (dirfd, entry) = match entry {
Ok(entry) => (entry.parent_fd(), entry),
Err(_) => continue, // ignore errors
2018-12-19 11:49:23 +00:00
};
2018-12-19 11:49:23 +00:00
let file_type = match entry.file_type() {
Some(file_type) => file_type,
2018-12-19 12:40:26 +00:00
None => bail!("unsupported file system type on chunk store '{}'", self.name),
2018-12-19 11:49:23 +00:00
};
if file_type != nix::dir::Type::File {
continue;
}
2018-12-19 11:49:23 +00:00
let filename = entry.file_name();
if let Ok(stat) = fstatat(dirfd, filename, nix::fcntl::AtFlags::AT_SYMLINK_NOFOLLOW) {
2018-12-19 11:49:23 +00:00
let age = now - stat.st_atime;
//println!("FOUND {} {:?}", age/(3600*24), filename);
2018-12-19 11:49:23 +00:00
if age/(3600*24) >= 2 {
println!("UNLINK {} {:?}", age/(3600*24), filename);
let res = unsafe { libc::unlinkat(dirfd, filename.as_ptr(), 0) };
2018-12-19 12:40:26 +00:00
if res != 0 {
let err = nix::Error::last();
bail!(
"unlink chunk {:?} failed on store '{}' - {}",
filename,
self.name,
err,
);
2018-12-19 12:40:26 +00:00
}
} else {
status.disk_chunks += 1;
status.disk_bytes += stat.st_size as usize;
2018-12-19 08:51:33 +00:00
}
2018-12-19 11:49:23 +00:00
}
}
2018-12-19 12:40:26 +00:00
Ok(())
2018-12-19 08:51:33 +00:00
}
pub fn insert_chunk(&self, chunk: &[u8]) -> Result<(bool, [u8; 32], u64), Error> {
2018-12-07 13:44:56 +00:00
2018-12-22 13:31:59 +00:00
// fixme: use Sha512/256 when available
let mut hasher = sha::Sha256::new();
hasher.update(chunk);
let digest = hasher.finish();
//println!("DIGEST {}", tools::digest_to_hex(&digest));
2018-12-07 13:44:56 +00:00
2018-12-16 12:52:16 +00:00
let mut chunk_path = self.chunk_dir.clone();
2018-12-19 08:51:33 +00:00
let prefix = digest_to_prefix(&digest);
2018-12-07 15:12:45 +00:00
chunk_path.push(&prefix);
let digest_str = tools::digest_to_hex(&digest);
2018-12-07 15:12:45 +00:00
chunk_path.push(&digest_str);
let lock = self.mutex.lock();
if let Ok(metadata) = std::fs::metadata(&chunk_path) {
if metadata.is_file() {
return Ok((true, digest, metadata.len()));
2018-12-07 15:12:45 +00:00
} else {
2018-12-19 12:54:22 +00:00
bail!("Got unexpected file type on store '{}' for chunk {}", self.name, digest_str);
2018-12-07 15:12:45 +00:00
}
}
2018-12-07 13:44:56 +00:00
2018-12-07 15:12:45 +00:00
let mut tmp_path = chunk_path.clone();
tmp_path.set_extension("tmp");
let f = std::fs::File::create(&tmp_path)?;
// fixme: what is the fasted lz4 encoder available (see lzbench)?
let mut encoder = lz4::EncoderBuilder::new().level(1).build(f)?;
encoder.write_all(chunk)?;
let (f, encode_result) = encoder.finish();
encode_result?;
2018-12-07 13:44:56 +00:00
2018-12-07 15:12:45 +00:00
if let Err(err) = std::fs::rename(&tmp_path, &chunk_path) {
if let Err(_) = std::fs::remove_file(&tmp_path) { /* ignore */ }
bail!(
"Atomic rename on store '{}' failed for chunk {} - {}",
self.name,
digest_str,
err,
);
2018-12-07 15:12:45 +00:00
}
// fixme: is there a better way to get the compressed size?
let stat = nix::sys::stat::fstat(f.as_raw_fd())?;
let compressed_size = stat.st_size as u64;
//println!("PATH {:?}", chunk_path);
2018-12-07 13:44:56 +00:00
2018-12-07 15:12:45 +00:00
drop(lock);
Ok((false, digest, compressed_size))
2018-12-07 13:44:56 +00:00
}
pub fn relative_path(&self, path: &Path) -> PathBuf {
let mut full_path = self.base.clone();
full_path.push(path);
full_path
}
2018-12-18 10:06:03 +00:00
pub fn base_path(&self) -> PathBuf {
self.base.clone()
}
}
#[test]
fn test_chunk_store1() {
let mut path = std::fs::canonicalize(".").unwrap(); // we need absulute path
path.push(".testdir");
if let Err(_e) = std::fs::remove_dir_all(".testdir") { /* ignore */ }
let chunk_store = ChunkStore::open("test", &path);
assert!(chunk_store.is_err());
let chunk_store = ChunkStore::create("test", &path).unwrap();
let (exists, _, _) = chunk_store.insert_chunk(&[0u8, 1u8]).unwrap();
assert!(!exists);
let (exists, _, _) = chunk_store.insert_chunk(&[0u8, 1u8]).unwrap();
assert!(exists);
2018-12-07 13:44:56 +00:00
let chunk_store = ChunkStore::create("test", &path);
assert!(chunk_store.is_err());
if let Err(_e) = std::fs::remove_dir_all(".testdir") { /* ignore */ }
}