Line data Source code
1 : //!
2 : //! VirtualFile is like a normal File, but it's not bound directly to
3 : //! a file descriptor. Instead, the file is opened when it's read from,
4 : //! and if too many files are open globally in the system, least-recently
5 : //! used ones are closed.
6 : //!
7 : //! To track which files have been recently used, we use the clock algorithm
8 : //! with a 'recently_used' flag on each slot.
9 : //!
10 : //! This is similar to PostgreSQL's virtual file descriptor facility in
11 : //! src/backend/storage/file/fd.c
12 : //!
13 : use crate::context::RequestContext;
14 : use crate::metrics::{StorageIoOperation, STORAGE_IO_SIZE, STORAGE_IO_TIME_METRIC};
15 :
16 : use crate::page_cache::PageWriteGuard;
17 : use crate::tenant::TENANTS_SEGMENT_NAME;
18 : use camino::{Utf8Path, Utf8PathBuf};
19 : use once_cell::sync::OnceCell;
20 : use pageserver_api::shard::TenantShardId;
21 : use std::fs::File;
22 : use std::io::{Error, ErrorKind, Seek, SeekFrom};
23 : use tokio_epoll_uring::{BoundedBuf, IoBuf, IoBufMut, Slice};
24 :
25 : use std::os::fd::{AsRawFd, FromRawFd, IntoRawFd, OwnedFd, RawFd};
26 : use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
27 : use tokio::sync::{RwLock, RwLockReadGuard, RwLockWriteGuard};
28 : use tokio::time::Instant;
29 :
30 : pub use pageserver_api::models::virtual_file as api;
31 : pub(crate) mod io_engine;
32 : pub use io_engine::feature_test as io_engine_feature_test;
33 : pub use io_engine::FeatureTestResult as IoEngineFeatureTestResult;
34 : mod metadata;
35 : mod open_options;
36 : use self::owned_buffers_io::write::OwnedAsyncWriter;
37 : pub(crate) use io_engine::IoEngineKind;
38 : pub(crate) use metadata::Metadata;
39 : pub(crate) use open_options::*;
40 :
41 : pub(crate) mod owned_buffers_io {
42 : //! Abstractions for IO with owned buffers.
43 : //!
44 : //! Not actually tied to [`crate::virtual_file`] specifically, but, it's the primary
45 : //! reason we need this abstraction.
46 : //!
47 : //! Over time, this could move into the `tokio-epoll-uring` crate, maybe `uring-common`,
48 : //! but for the time being we're proving out the primitives in the neon.git repo
49 : //! for faster iteration.
50 :
51 : pub(crate) mod write;
52 : pub(crate) mod util {
53 : pub(crate) mod size_tracking_writer;
54 : }
55 : }
56 :
57 : ///
58 : /// A virtual file descriptor. You can use this just like std::fs::File, but internally
59 : /// the underlying file is closed if the system is low on file descriptors,
60 : /// and re-opened when it's accessed again.
61 : ///
62 : /// Like with std::fs::File, multiple threads can read/write the file concurrently,
63 : /// holding just a shared reference the same VirtualFile, using the read_at() / write_at()
64 : /// functions from the FileExt trait. But the functions from the Read/Write/Seek traits
65 : /// require a mutable reference, because they modify the "current position".
66 : ///
67 : /// Each VirtualFile has a physical file descriptor in the global OPEN_FILES array, at the
68 : /// slot that 'handle points to, if the underlying file is currently open. If it's not
69 : /// currently open, the 'handle' can still point to the slot where it was last kept. The
70 : /// 'tag' field is used to detect whether the handle still is valid or not.
71 : ///
72 : #[derive(Debug)]
73 : pub struct VirtualFile {
74 : /// Lazy handle to the global file descriptor cache. The slot that this points to
75 : /// might contain our File, or it may be empty, or it may contain a File that
76 : /// belongs to a different VirtualFile.
77 : handle: RwLock<SlotHandle>,
78 :
79 : /// Current file position
80 : pos: u64,
81 :
82 : /// File path and options to use to open it.
83 : ///
84 : /// Note: this only contains the options needed to re-open it. For example,
85 : /// if a new file is created, we only pass the create flag when it's initially
86 : /// opened, in the VirtualFile::create() function, and strip the flag before
87 : /// storing it here.
88 : pub path: Utf8PathBuf,
89 : open_options: OpenOptions,
90 :
91 : // These are strings becase we only use them for metrics, and those expect strings.
92 : // It makes no sense for us to constantly turn the `TimelineId` and `TenantId` into
93 : // strings.
94 : tenant_id: String,
95 : shard_id: String,
96 : timeline_id: String,
97 : }
98 :
99 : #[derive(Debug, PartialEq, Clone, Copy)]
100 : struct SlotHandle {
101 : /// Index into OPEN_FILES.slots
102 : index: usize,
103 :
104 : /// Value of 'tag' in the slot. If slot's tag doesn't match, then the slot has
105 : /// been recycled and no longer contains the FD for this virtual file.
106 : tag: u64,
107 : }
108 :
109 : /// OPEN_FILES is the global array that holds the physical file descriptors that
110 : /// are currently open. Each slot in the array is protected by a separate lock,
111 : /// so that different files can be accessed independently. The lock must be held
112 : /// in write mode to replace the slot with a different file, but a read mode
113 : /// is enough to operate on the file, whether you're reading or writing to it.
114 : ///
115 : /// OPEN_FILES starts in uninitialized state, and it's initialized by
116 : /// the virtual_file::init() function. It must be called exactly once at page
117 : /// server startup.
118 : static OPEN_FILES: OnceCell<OpenFiles> = OnceCell::new();
119 :
120 : struct OpenFiles {
121 : slots: &'static [Slot],
122 :
123 : /// clock arm for the clock algorithm
124 : next: AtomicUsize,
125 : }
126 :
127 : struct Slot {
128 : inner: RwLock<SlotInner>,
129 :
130 : /// has this file been used since last clock sweep?
131 : recently_used: AtomicBool,
132 : }
133 :
134 : struct SlotInner {
135 : /// Counter that's incremented every time a different file is stored here.
136 : /// To avoid the ABA problem.
137 : tag: u64,
138 :
139 : /// the underlying file
140 : file: Option<OwnedFd>,
141 : }
142 :
143 : /// Impl of [`tokio_epoll_uring::IoBuf`] and [`tokio_epoll_uring::IoBufMut`] for [`PageWriteGuard`].
144 : struct PageWriteGuardBuf {
145 : page: PageWriteGuard<'static>,
146 : init_up_to: usize,
147 : }
148 : // Safety: the [`PageWriteGuard`] gives us exclusive ownership of the page cache slot,
149 : // and the location remains stable even if [`Self`] or the [`PageWriteGuard`] is moved.
150 : unsafe impl tokio_epoll_uring::IoBuf for PageWriteGuardBuf {
151 111465 : fn stable_ptr(&self) -> *const u8 {
152 111465 : self.page.as_ptr()
153 111465 : }
154 334395 : fn bytes_init(&self) -> usize {
155 334395 : self.init_up_to
156 334395 : }
157 111465 : fn bytes_total(&self) -> usize {
158 111465 : self.page.len()
159 111465 : }
160 : }
161 : // Safety: see above, plus: the ownership of [`PageWriteGuard`] means exclusive access,
162 : // hence it's safe to hand out the `stable_mut_ptr()`.
163 : unsafe impl tokio_epoll_uring::IoBufMut for PageWriteGuardBuf {
164 111465 : fn stable_mut_ptr(&mut self) -> *mut u8 {
165 111465 : self.page.as_mut_ptr()
166 111465 : }
167 :
168 111465 : unsafe fn set_init(&mut self, pos: usize) {
169 111465 : assert!(pos <= self.page.len());
170 111465 : self.init_up_to = pos;
171 111465 : }
172 : }
173 :
174 : impl OpenFiles {
175 : /// Find a slot to use, evicting an existing file descriptor if needed.
176 : ///
177 : /// On return, we hold a lock on the slot, and its 'tag' has been updated
178 : /// recently_used has been set. It's all ready for reuse.
179 195433 : async fn find_victim_slot(&self) -> (SlotHandle, RwLockWriteGuard<SlotInner>) {
180 195433 : //
181 195433 : // Run the clock algorithm to find a slot to replace.
182 195433 : //
183 195433 : let num_slots = self.slots.len();
184 195433 : let mut retries = 0;
185 : let mut slot;
186 : let mut slot_guard;
187 : let index;
188 2697259 : loop {
189 2697259 : let next = self.next.fetch_add(1, Ordering::AcqRel) % num_slots;
190 2697259 : slot = &self.slots[next];
191 2697259 :
192 2697259 : // If the recently_used flag on this slot is set, continue the clock
193 2697259 : // sweep. Otherwise try to use this slot. If we cannot acquire the
194 2697259 : // lock, also continue the clock sweep.
195 2697259 : //
196 2697259 : // We only continue in this manner for a while, though. If we loop
197 2697259 : // through the array twice without finding a victim, just pick the
198 2697259 : // next slot and wait until we can reuse it. This way, we avoid
199 2697259 : // spinning in the extreme case that all the slots are busy with an
200 2697259 : // I/O operation.
201 2697259 : if retries < num_slots * 2 {
202 2591567 : if !slot.recently_used.swap(false, Ordering::Release) {
203 2382628 : if let Ok(guard) = slot.inner.try_write() {
204 89741 : slot_guard = guard;
205 89741 : index = next;
206 89741 : break;
207 2292887 : }
208 208939 : }
209 2501826 : retries += 1;
210 : } else {
211 105692 : slot_guard = slot.inner.write().await;
212 105692 : index = next;
213 105692 : break;
214 : }
215 : }
216 :
217 : //
218 : // We now have the victim slot locked. If it was in use previously, close the
219 : // old file.
220 : //
221 195433 : if let Some(old_file) = slot_guard.file.take() {
222 191952 : // the normal path of dropping VirtualFile uses "close", use "close-by-replace" here to
223 191952 : // distinguish the two.
224 191952 : STORAGE_IO_TIME_METRIC
225 191952 : .get(StorageIoOperation::CloseByReplace)
226 191952 : .observe_closure_duration(|| drop(old_file));
227 191952 : }
228 :
229 : // Prepare the slot for reuse and return it
230 195433 : slot_guard.tag += 1;
231 195433 : slot.recently_used.store(true, Ordering::Relaxed);
232 195433 : (
233 195433 : SlotHandle {
234 195433 : index,
235 195433 : tag: slot_guard.tag,
236 195433 : },
237 195433 : slot_guard,
238 195433 : )
239 195433 : }
240 : }
241 :
242 : /// Identify error types that should alwways terminate the process. Other
243 : /// error types may be elegible for retry.
244 0 : pub(crate) fn is_fatal_io_error(e: &std::io::Error) -> bool {
245 0 : use nix::errno::Errno::*;
246 0 : match e.raw_os_error().map(nix::errno::from_i32) {
247 : Some(EIO) => {
248 : // Terminate on EIO because we no longer trust the device to store
249 : // data safely, or to uphold persistence guarantees on fsync.
250 0 : true
251 : }
252 : Some(EROFS) => {
253 : // Terminate on EROFS because a filesystem is usually remounted
254 : // readonly when it has experienced some critical issue, so the same
255 : // logic as EIO applies.
256 0 : true
257 : }
258 : Some(EACCES) => {
259 : // Terminate on EACCESS because we should always have permissions
260 : // for our own data dir: if we don't, then we can't do our job and
261 : // need administrative intervention to fix permissions. Terminating
262 : // is the best way to make sure we stop cleanly rather than going
263 : // into infinite retry loops, and will make it clear to the outside
264 : // world that we need help.
265 0 : true
266 : }
267 : _ => {
268 : // Treat all other local file I/O errors are retryable. This includes:
269 : // - ENOSPC: we stay up and wait for eviction to free some space
270 : // - EINVAL, EBADF, EBADFD: this is a code bug, not a filesystem/hardware issue
271 : // - WriteZero, Interrupted: these are used internally VirtualFile
272 0 : false
273 : }
274 : }
275 0 : }
276 :
277 : /// Call this when the local filesystem gives us an error with an external
278 : /// cause: this includes EIO, EROFS, and EACCESS: all these indicate either
279 : /// bad storage or bad configuration, and we can't fix that from inside
280 : /// a running process.
281 0 : pub(crate) fn on_fatal_io_error(e: &std::io::Error, context: &str) -> ! {
282 0 : tracing::error!("Fatal I/O error: {e}: {context})");
283 0 : std::process::abort();
284 : }
285 :
286 : pub(crate) trait MaybeFatalIo<T> {
287 : fn maybe_fatal_err(self, context: &str) -> std::io::Result<T>;
288 : fn fatal_err(self, context: &str) -> T;
289 : }
290 :
291 : impl<T> MaybeFatalIo<T> for std::io::Result<T> {
292 : /// Terminate the process if the result is an error of a fatal type, else pass it through
293 : ///
294 : /// This is appropriate for writes, where we typically want to die on EIO/ACCES etc, but
295 : /// not on ENOSPC.
296 22 : fn maybe_fatal_err(self, context: &str) -> std::io::Result<T> {
297 22 : if let Err(e) = &self {
298 0 : if is_fatal_io_error(e) {
299 0 : on_fatal_io_error(e, context);
300 0 : }
301 22 : }
302 22 : self
303 22 : }
304 :
305 : /// Terminate the process on any I/O error.
306 : ///
307 : /// This is appropriate for reads on files that we know exist: they should always work.
308 1950 : fn fatal_err(self, context: &str) -> T {
309 1950 : match self {
310 1950 : Ok(v) => v,
311 0 : Err(e) => {
312 0 : on_fatal_io_error(&e, context);
313 : }
314 : }
315 1950 : }
316 : }
317 :
318 : /// Observe duration for the given storage I/O operation
319 : ///
320 : /// Unlike `observe_closure_duration`, this supports async,
321 : /// where "support" means that we measure wall clock time.
322 : macro_rules! observe_duration {
323 : ($op:expr, $($body:tt)*) => {{
324 : let instant = Instant::now();
325 : let result = $($body)*;
326 : let elapsed = instant.elapsed().as_secs_f64();
327 : STORAGE_IO_TIME_METRIC
328 : .get($op)
329 : .observe(elapsed);
330 : result
331 : }}
332 : }
333 :
334 : macro_rules! with_file {
335 : ($this:expr, $op:expr, | $ident:ident | $($body:tt)*) => {{
336 : let $ident = $this.lock_file().await?;
337 : observe_duration!($op, $($body)*)
338 : }};
339 : ($this:expr, $op:expr, | mut $ident:ident | $($body:tt)*) => {{
340 : let mut $ident = $this.lock_file().await?;
341 : observe_duration!($op, $($body)*)
342 : }};
343 : }
344 :
345 : impl VirtualFile {
346 : /// Open a file in read-only mode. Like File::open.
347 1832 : pub async fn open(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
348 1832 : Self::open_with_options(path, OpenOptions::new().read(true)).await
349 1832 : }
350 :
351 : /// Create a new file for writing. If the file exists, it will be truncated.
352 : /// Like File::create.
353 1089 : pub async fn create(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
354 1089 : Self::open_with_options(
355 1089 : path,
356 1089 : OpenOptions::new().write(true).create(true).truncate(true),
357 1089 : )
358 563 : .await
359 1089 : }
360 :
361 : /// Open a file with given options.
362 : ///
363 : /// Note: If any custom flags were set in 'open_options' through OpenOptionsExt,
364 : /// they will be applied also when the file is subsequently re-opened, not only
365 : /// on the first time. Make sure that's sane!
366 4495 : pub async fn open_with_options(
367 4495 : path: &Utf8Path,
368 4495 : open_options: &OpenOptions,
369 4495 : ) -> Result<VirtualFile, std::io::Error> {
370 4495 : let path_str = path.to_string();
371 4495 : let parts = path_str.split('/').collect::<Vec<&str>>();
372 4495 : let (tenant_id, shard_id, timeline_id) =
373 4495 : if parts.len() > 5 && parts[parts.len() - 5] == TENANTS_SEGMENT_NAME {
374 3077 : let tenant_shard_part = parts[parts.len() - 4];
375 3077 : let (tenant_id, shard_id) = match tenant_shard_part.parse::<TenantShardId>() {
376 3077 : Ok(tenant_shard_id) => (
377 3077 : tenant_shard_id.tenant_id.to_string(),
378 3077 : format!("{}", tenant_shard_id.shard_slug()),
379 3077 : ),
380 : Err(_) => {
381 : // Malformed path: this ID is just for observability, so tolerate it
382 : // and pass through
383 0 : (tenant_shard_part.to_string(), "*".to_string())
384 : }
385 : };
386 3077 : (tenant_id, shard_id, parts[parts.len() - 2].to_string())
387 : } else {
388 1418 : ("*".to_string(), "*".to_string(), "*".to_string())
389 : };
390 4495 : let (handle, mut slot_guard) = get_open_files().find_victim_slot().await;
391 :
392 : // NB: there is also StorageIoOperation::OpenAfterReplace which is for the case
393 : // where our caller doesn't get to use the returned VirtualFile before its
394 : // slot gets re-used by someone else.
395 4495 : let file = observe_duration!(StorageIoOperation::Open, {
396 4495 : open_options.open(path.as_std_path()).await?
397 : });
398 :
399 : // Strip all options other than read and write.
400 : //
401 : // It would perhaps be nicer to check just for the read and write flags
402 : // explicitly, but OpenOptions doesn't contain any functions to read flags,
403 : // only to set them.
404 4495 : let mut reopen_options = open_options.clone();
405 4495 : reopen_options.create(false);
406 4495 : reopen_options.create_new(false);
407 4495 : reopen_options.truncate(false);
408 4495 :
409 4495 : let vfile = VirtualFile {
410 4495 : handle: RwLock::new(handle),
411 4495 : pos: 0,
412 4495 : path: path.to_path_buf(),
413 4495 : open_options: reopen_options,
414 4495 : tenant_id,
415 4495 : shard_id,
416 4495 : timeline_id,
417 4495 : };
418 4495 :
419 4495 : // TODO: Under pressure, it's likely the slot will get re-used and
420 4495 : // the underlying file closed before they get around to using it.
421 4495 : // => https://github.com/neondatabase/neon/issues/6065
422 4495 : slot_guard.file.replace(file);
423 4495 :
424 4495 : Ok(vfile)
425 4495 : }
426 :
427 : /// Async version of [`::utils::crashsafe::overwrite`].
428 : ///
429 : /// # NB:
430 : ///
431 : /// Doesn't actually use the [`VirtualFile`] file descriptor cache, but,
432 : /// it did at an earlier time.
433 : /// And it will use this module's [`io_engine`] in the near future, so, leaving it here.
434 28 : pub async fn crashsafe_overwrite<B: BoundedBuf<Buf = Buf> + Send, Buf: IoBuf + Send>(
435 28 : final_path: Utf8PathBuf,
436 28 : tmp_path: Utf8PathBuf,
437 28 : content: B,
438 28 : ) -> std::io::Result<()> {
439 28 : // TODO: use tokio_epoll_uring if configured as `io_engine`.
440 28 : // See https://github.com/neondatabase/neon/issues/6663
441 28 :
442 28 : tokio::task::spawn_blocking(move || {
443 28 : let slice_storage;
444 28 : let content_len = content.bytes_init();
445 28 : let content = if content.bytes_init() > 0 {
446 28 : slice_storage = Some(content.slice(0..content_len));
447 28 : slice_storage.as_deref().expect("just set it to Some()")
448 : } else {
449 0 : &[]
450 : };
451 28 : utils::crashsafe::overwrite(&final_path, &tmp_path, content)
452 28 : })
453 28 : .await
454 28 : .expect("blocking task is never aborted")
455 28 : }
456 :
457 : /// Call File::sync_all() on the underlying File.
458 2139 : pub async fn sync_all(&self) -> Result<(), Error> {
459 2139 : with_file!(self, StorageIoOperation::Fsync, |file_guard| {
460 2139 : let (_file_guard, res) = io_engine::get().sync_all(file_guard).await;
461 2139 : res
462 : })
463 2139 : }
464 :
465 : /// Call File::sync_data() on the underlying File.
466 0 : pub async fn sync_data(&self) -> Result<(), Error> {
467 0 : with_file!(self, StorageIoOperation::Fsync, |file_guard| {
468 0 : let (_file_guard, res) = io_engine::get().sync_data(file_guard).await;
469 0 : res
470 : })
471 0 : }
472 :
473 1180 : pub async fn metadata(&self) -> Result<Metadata, Error> {
474 1180 : with_file!(self, StorageIoOperation::Metadata, |file_guard| {
475 1180 : let (_file_guard, res) = io_engine::get().metadata(file_guard).await;
476 1180 : res
477 : })
478 1180 : }
479 :
480 : /// Helper function internal to `VirtualFile` that looks up the underlying File,
481 : /// opens it and evicts some other File if necessary. The passed parameter is
482 : /// assumed to be a function available for the physical `File`.
483 : ///
484 : /// We are doing it via a macro as Rust doesn't support async closures that
485 : /// take on parameters with lifetimes.
486 406593 : async fn lock_file(&self) -> Result<FileGuard, Error> {
487 406593 : let open_files = get_open_files();
488 :
489 190938 : let mut handle_guard = {
490 : // Read the cached slot handle, and see if the slot that it points to still
491 : // contains our File.
492 : //
493 : // We only need to hold the handle lock while we read the current handle. If
494 : // another thread closes the file and recycles the slot for a different file,
495 : // we will notice that the handle we read is no longer valid and retry.
496 406593 : let mut handle = *self.handle.read().await;
497 513203 : loop {
498 513203 : // Check if the slot contains our File
499 513203 : {
500 513203 : let slot = &open_files.slots[handle.index];
501 513203 : let slot_guard = slot.inner.read().await;
502 513203 : if slot_guard.tag == handle.tag && slot_guard.file.is_some() {
503 : // Found a cached file descriptor.
504 215655 : slot.recently_used.store(true, Ordering::Relaxed);
505 215655 : return Ok(FileGuard { slot_guard });
506 297548 : }
507 : }
508 :
509 : // The slot didn't contain our File. We will have to open it ourselves,
510 : // but before that, grab a write lock on handle in the VirtualFile, so
511 : // that no other thread will try to concurrently open the same file.
512 297548 : let handle_guard = self.handle.write().await;
513 :
514 : // If another thread changed the handle while we were not holding the lock,
515 : // then the handle might now be valid again. Loop back to retry.
516 297548 : if *handle_guard != handle {
517 106610 : handle = *handle_guard;
518 106610 : continue;
519 190938 : }
520 190938 : break handle_guard;
521 : }
522 : };
523 :
524 : // We need to open the file ourselves. The handle in the VirtualFile is
525 : // now locked in write-mode. Find a free slot to put it in.
526 190938 : let (handle, mut slot_guard) = open_files.find_victim_slot().await;
527 :
528 : // Re-open the physical file.
529 : // NB: we use StorageIoOperation::OpenAferReplace for this to distinguish this
530 : // case from StorageIoOperation::Open. This helps with identifying thrashing
531 : // of the virtual file descriptor cache.
532 190938 : let file = observe_duration!(StorageIoOperation::OpenAfterReplace, {
533 190938 : self.open_options.open(self.path.as_std_path()).await?
534 : });
535 :
536 : // Store the File in the slot and update the handle in the VirtualFile
537 : // to point to it.
538 190938 : slot_guard.file.replace(file);
539 190938 :
540 190938 : *handle_guard = handle;
541 190938 :
542 190938 : return Ok(FileGuard {
543 190938 : slot_guard: slot_guard.downgrade(),
544 190938 : });
545 406593 : }
546 :
547 0 : pub fn remove(self) {
548 0 : let path = self.path.clone();
549 0 : drop(self);
550 0 : std::fs::remove_file(path).expect("failed to remove the virtual file");
551 0 : }
552 :
553 3552 : pub async fn seek(&mut self, pos: SeekFrom) -> Result<u64, Error> {
554 3552 : match pos {
555 3542 : SeekFrom::Start(offset) => {
556 3542 : self.pos = offset;
557 3542 : }
558 4 : SeekFrom::End(offset) => {
559 4 : self.pos = with_file!(self, StorageIoOperation::Seek, |mut file_guard| file_guard
560 4 : .with_std_file_mut(|std_file| std_file.seek(SeekFrom::End(offset))))?
561 : }
562 6 : SeekFrom::Current(offset) => {
563 6 : let pos = self.pos as i128 + offset as i128;
564 6 : if pos < 0 {
565 2 : return Err(Error::new(
566 2 : ErrorKind::InvalidInput,
567 2 : "offset would be negative",
568 2 : ));
569 4 : }
570 4 : if pos > u64::MAX as i128 {
571 0 : return Err(Error::new(ErrorKind::InvalidInput, "offset overflow"));
572 4 : }
573 4 : self.pos = pos as u64;
574 : }
575 : }
576 3548 : Ok(self.pos)
577 3552 : }
578 :
579 331667 : pub async fn read_exact_at<B>(&self, buf: B, offset: u64) -> Result<B, Error>
580 331667 : where
581 331667 : B: IoBufMut + Send,
582 331667 : {
583 331667 : let (buf, res) =
584 729025 : read_exact_at_impl(buf, offset, None, |buf, offset| self.read_at(buf, offset)).await;
585 331667 : res.map(|()| buf)
586 331667 : }
587 :
588 30196 : pub async fn read_exact_at_n<B>(&self, buf: B, offset: u64, count: usize) -> Result<B, Error>
589 30196 : where
590 30196 : B: IoBufMut + Send,
591 30196 : {
592 30196 : let (buf, res) = read_exact_at_impl(buf, offset, Some(count), |buf, offset| {
593 30196 : self.read_at(buf, offset)
594 30196 : })
595 15369 : .await;
596 30196 : res.map(|()| buf)
597 30196 : }
598 :
599 : /// Like [`Self::read_exact_at`] but for [`PageWriteGuard`].
600 111465 : pub async fn read_exact_at_page(
601 111465 : &self,
602 111465 : page: PageWriteGuard<'static>,
603 111465 : offset: u64,
604 111465 : ) -> Result<PageWriteGuard<'static>, Error> {
605 111465 : let buf = PageWriteGuardBuf {
606 111465 : page,
607 111465 : init_up_to: 0,
608 111465 : };
609 111465 : let res = self.read_exact_at(buf, offset).await;
610 111465 : res.map(|PageWriteGuardBuf { page, .. }| page)
611 111465 : .map_err(|e| Error::new(ErrorKind::Other, e))
612 111465 : }
613 :
614 : // Copied from https://doc.rust-lang.org/1.72.0/src/std/os/unix/fs.rs.html#219-235
615 4 : pub async fn write_all_at<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
616 4 : &self,
617 4 : buf: B,
618 4 : mut offset: u64,
619 4 : ctx: &RequestContext,
620 4 : ) -> (B::Buf, Result<(), Error>) {
621 4 : let buf_len = buf.bytes_init();
622 4 : if buf_len == 0 {
623 0 : return (Slice::into_inner(buf.slice_full()), Ok(()));
624 4 : }
625 4 : let mut buf = buf.slice(0..buf_len);
626 8 : while !buf.is_empty() {
627 : let res;
628 4 : (buf, res) = self.write_at(buf, offset, ctx).await;
629 0 : match res {
630 : Ok(0) => {
631 0 : return (
632 0 : Slice::into_inner(buf),
633 0 : Err(Error::new(
634 0 : std::io::ErrorKind::WriteZero,
635 0 : "failed to write whole buffer",
636 0 : )),
637 0 : );
638 : }
639 4 : Ok(n) => {
640 4 : buf = buf.slice(n..);
641 4 : offset += n as u64;
642 4 : }
643 0 : Err(e) if e.kind() == std::io::ErrorKind::Interrupted => {}
644 0 : Err(e) => return (Slice::into_inner(buf), Err(e)),
645 : }
646 : }
647 4 : (Slice::into_inner(buf), Ok(()))
648 4 : }
649 :
650 : /// Writes `buf.slice(0..buf.bytes_init())`.
651 : /// Returns the IoBuf that is underlying the BoundedBuf `buf`.
652 : /// I.e., the returned value's `bytes_init()` method returns something different than the `bytes_init()` that was passed in.
653 : /// It's quite brittle and easy to mis-use, so, we return the size in the Ok() variant.
654 40991 : pub async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
655 40991 : &mut self,
656 40991 : buf: B,
657 40991 : ctx: &RequestContext,
658 40991 : ) -> (B::Buf, Result<usize, Error>) {
659 40991 : let nbytes = buf.bytes_init();
660 40991 : if nbytes == 0 {
661 32 : return (Slice::into_inner(buf.slice_full()), Ok(0));
662 40959 : }
663 40959 : let mut buf = buf.slice(0..nbytes);
664 81916 : while !buf.is_empty() {
665 : let res;
666 40959 : (buf, res) = self.write(buf, ctx).await;
667 2 : match res {
668 : Ok(0) => {
669 0 : return (
670 0 : Slice::into_inner(buf),
671 0 : Err(Error::new(
672 0 : std::io::ErrorKind::WriteZero,
673 0 : "failed to write whole buffer",
674 0 : )),
675 0 : );
676 : }
677 40957 : Ok(n) => {
678 40957 : buf = buf.slice(n..);
679 40957 : }
680 2 : Err(ref e) if e.kind() == std::io::ErrorKind::Interrupted => {}
681 2 : Err(e) => return (Slice::into_inner(buf), Err(e)),
682 : }
683 : }
684 40957 : (Slice::into_inner(buf), Ok(nbytes))
685 40991 : }
686 :
687 40959 : async fn write<B: IoBuf + Send>(
688 40959 : &mut self,
689 40959 : buf: Slice<B>,
690 40959 : ctx: &RequestContext,
691 40959 : ) -> (Slice<B>, Result<usize, std::io::Error>) {
692 40959 : let pos = self.pos;
693 40959 : let (buf, res) = self.write_at(buf, pos, ctx).await;
694 40959 : let n = match res {
695 40957 : Ok(n) => n,
696 2 : Err(e) => return (buf, Err(e)),
697 : };
698 40957 : self.pos += n as u64;
699 40957 : (buf, Ok(n))
700 40959 : }
701 :
702 362307 : pub(crate) async fn read_at<B>(&self, buf: B, offset: u64) -> (B, Result<usize, Error>)
703 362307 : where
704 362307 : B: tokio_epoll_uring::BoundedBufMut + Send,
705 362307 : {
706 563471 : let file_guard = match self.lock_file().await {
707 362307 : Ok(file_guard) => file_guard,
708 0 : Err(e) => return (buf, Err(e)),
709 : };
710 :
711 362307 : observe_duration!(StorageIoOperation::Read, {
712 362307 : let ((_file_guard, buf), res) = io_engine::get().read_at(file_guard, offset, buf).await;
713 362307 : if let Ok(size) = res {
714 362305 : STORAGE_IO_SIZE
715 362305 : .with_label_values(&[
716 362305 : "read",
717 362305 : &self.tenant_id,
718 362305 : &self.shard_id,
719 362305 : &self.timeline_id,
720 362305 : ])
721 362305 : .add(size as i64);
722 362305 : }
723 362307 : (buf, res)
724 : })
725 362307 : }
726 :
727 40963 : async fn write_at<B: IoBuf + Send>(
728 40963 : &self,
729 40963 : buf: Slice<B>,
730 40963 : offset: u64,
731 40963 : _ctx: &RequestContext, /* TODO: use for metrics: https://github.com/neondatabase/neon/issues/6107 */
732 40963 : ) -> (Slice<B>, Result<usize, Error>) {
733 40963 : let file_guard = match self.lock_file().await {
734 40963 : Ok(file_guard) => file_guard,
735 0 : Err(e) => return (buf, Err(e)),
736 : };
737 40963 : observe_duration!(StorageIoOperation::Write, {
738 40963 : let ((_file_guard, buf), result) =
739 40963 : io_engine::get().write_at(file_guard, offset, buf).await;
740 40963 : if let Ok(size) = result {
741 40961 : STORAGE_IO_SIZE
742 40961 : .with_label_values(&[
743 40961 : "write",
744 40961 : &self.tenant_id,
745 40961 : &self.shard_id,
746 40961 : &self.timeline_id,
747 40961 : ])
748 40961 : .add(size as i64);
749 40961 : }
750 40963 : (buf, result)
751 : })
752 40963 : }
753 : }
754 :
755 : // Adapted from https://doc.rust-lang.org/1.72.0/src/std/os/unix/fs.rs.html#117-135
756 361873 : pub async fn read_exact_at_impl<B, F, Fut>(
757 361873 : buf: B,
758 361873 : mut offset: u64,
759 361873 : count: Option<usize>,
760 361873 : mut read_at: F,
761 361873 : ) -> (B, std::io::Result<()>)
762 361873 : where
763 361873 : B: IoBufMut + Send,
764 361873 : F: FnMut(tokio_epoll_uring::Slice<B>, u64) -> Fut,
765 361873 : Fut: std::future::Future<Output = (tokio_epoll_uring::Slice<B>, std::io::Result<usize>)>,
766 361873 : {
767 361873 : let mut buf: tokio_epoll_uring::Slice<B> = match count {
768 30198 : Some(count) => {
769 30198 : assert!(count <= buf.bytes_total());
770 30198 : assert!(count > 0);
771 30198 : buf.slice(..count) // may include uninitialized memory
772 : }
773 331675 : None => buf.slice_full(), // includes all the uninitialized memory
774 : };
775 :
776 723748 : while buf.bytes_total() != 0 {
777 : let res;
778 744394 : (buf, res) = read_at(buf, offset).await;
779 0 : match res {
780 2 : Ok(0) => break,
781 361875 : Ok(n) => {
782 361875 : buf = buf.slice(n..);
783 361875 : offset += n as u64;
784 361875 : }
785 0 : Err(ref e) if e.kind() == std::io::ErrorKind::Interrupted => {}
786 0 : Err(e) => return (buf.into_inner(), Err(e)),
787 : }
788 : }
789 : // NB: don't use `buf.is_empty()` here; it is from the
790 : // `impl Deref for Slice { Target = [u8] }`; the &[u8]
791 : // returned by it only covers the initialized portion of `buf`.
792 : // Whereas we're interested in ensuring that we filled the entire
793 : // buffer that the user passed in.
794 361873 : if buf.bytes_total() != 0 {
795 2 : (
796 2 : buf.into_inner(),
797 2 : Err(std::io::Error::new(
798 2 : std::io::ErrorKind::UnexpectedEof,
799 2 : "failed to fill whole buffer",
800 2 : )),
801 2 : )
802 : } else {
803 361871 : assert_eq!(buf.len(), buf.bytes_total());
804 361871 : (buf.into_inner(), Ok(()))
805 : }
806 361873 : }
807 :
808 : #[cfg(test)]
809 : mod test_read_exact_at_impl {
810 :
811 : use std::{collections::VecDeque, sync::Arc};
812 :
813 : use tokio_epoll_uring::{BoundedBuf, BoundedBufMut};
814 :
815 : use super::read_exact_at_impl;
816 :
817 : struct Expectation {
818 : offset: u64,
819 : bytes_total: usize,
820 : result: std::io::Result<Vec<u8>>,
821 : }
822 : struct MockReadAt {
823 : expectations: VecDeque<Expectation>,
824 : }
825 :
826 : impl MockReadAt {
827 14 : async fn read_at(
828 14 : &mut self,
829 14 : mut buf: tokio_epoll_uring::Slice<Vec<u8>>,
830 14 : offset: u64,
831 14 : ) -> (tokio_epoll_uring::Slice<Vec<u8>>, std::io::Result<usize>) {
832 14 : let exp = self
833 14 : .expectations
834 14 : .pop_front()
835 14 : .expect("read_at called but we have no expectations left");
836 14 : assert_eq!(exp.offset, offset);
837 14 : assert_eq!(exp.bytes_total, buf.bytes_total());
838 14 : match exp.result {
839 14 : Ok(bytes) => {
840 14 : assert!(bytes.len() <= buf.bytes_total());
841 14 : buf.put_slice(&bytes);
842 14 : (buf, Ok(bytes.len()))
843 : }
844 0 : Err(e) => (buf, Err(e)),
845 : }
846 14 : }
847 : }
848 :
849 : impl Drop for MockReadAt {
850 10 : fn drop(&mut self) {
851 10 : assert_eq!(self.expectations.len(), 0);
852 10 : }
853 : }
854 :
855 : #[tokio::test]
856 2 : async fn test_basic() {
857 2 : let buf = Vec::with_capacity(5);
858 2 : let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
859 2 : expectations: VecDeque::from(vec![Expectation {
860 2 : offset: 0,
861 2 : bytes_total: 5,
862 2 : result: Ok(vec![b'a', b'b', b'c', b'd', b'e']),
863 2 : }]),
864 2 : }));
865 2 : let (buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
866 2 : let mock_read_at = Arc::clone(&mock_read_at);
867 2 : async move { mock_read_at.lock().await.read_at(buf, offset).await }
868 2 : })
869 2 : .await;
870 2 : assert!(res.is_ok());
871 2 : assert_eq!(buf, vec![b'a', b'b', b'c', b'd', b'e']);
872 2 : }
873 :
874 : #[tokio::test]
875 2 : async fn test_with_count() {
876 2 : let buf = Vec::with_capacity(5);
877 2 : let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
878 2 : expectations: VecDeque::from(vec![Expectation {
879 2 : offset: 0,
880 2 : bytes_total: 3,
881 2 : result: Ok(vec![b'a', b'b', b'c']),
882 2 : }]),
883 2 : }));
884 2 :
885 2 : let (buf, res) = read_exact_at_impl(buf, 0, Some(3), |buf, offset| {
886 2 : let mock_read_at = Arc::clone(&mock_read_at);
887 2 : async move { mock_read_at.lock().await.read_at(buf, offset).await }
888 2 : })
889 2 : .await;
890 2 : assert!(res.is_ok());
891 2 : assert_eq!(buf, vec![b'a', b'b', b'c']);
892 2 : }
893 :
894 : #[tokio::test]
895 2 : async fn test_empty_buf_issues_no_syscall() {
896 2 : let buf = Vec::new();
897 2 : let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
898 2 : expectations: VecDeque::new(),
899 2 : }));
900 2 : let (_buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
901 0 : let mock_read_at = Arc::clone(&mock_read_at);
902 2 : async move { mock_read_at.lock().await.read_at(buf, offset).await }
903 2 : })
904 2 : .await;
905 2 : assert!(res.is_ok());
906 2 : }
907 :
908 : #[tokio::test]
909 2 : async fn test_two_read_at_calls_needed_until_buf_filled() {
910 2 : let buf = Vec::with_capacity(4);
911 2 : let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
912 2 : expectations: VecDeque::from(vec![
913 2 : Expectation {
914 2 : offset: 0,
915 2 : bytes_total: 4,
916 2 : result: Ok(vec![b'a', b'b']),
917 2 : },
918 2 : Expectation {
919 2 : offset: 2,
920 2 : bytes_total: 2,
921 2 : result: Ok(vec![b'c', b'd']),
922 2 : },
923 2 : ]),
924 2 : }));
925 4 : let (buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
926 4 : let mock_read_at = Arc::clone(&mock_read_at);
927 4 : async move { mock_read_at.lock().await.read_at(buf, offset).await }
928 4 : })
929 2 : .await;
930 2 : assert!(res.is_ok());
931 2 : assert_eq!(buf, vec![b'a', b'b', b'c', b'd']);
932 2 : }
933 :
934 : #[tokio::test]
935 2 : async fn test_eof_before_buffer_full() {
936 2 : let buf = Vec::with_capacity(3);
937 2 : let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
938 2 : expectations: VecDeque::from(vec![
939 2 : Expectation {
940 2 : offset: 0,
941 2 : bytes_total: 3,
942 2 : result: Ok(vec![b'a']),
943 2 : },
944 2 : Expectation {
945 2 : offset: 1,
946 2 : bytes_total: 2,
947 2 : result: Ok(vec![b'b']),
948 2 : },
949 2 : Expectation {
950 2 : offset: 2,
951 2 : bytes_total: 1,
952 2 : result: Ok(vec![]),
953 2 : },
954 2 : ]),
955 2 : }));
956 6 : let (_buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
957 6 : let mock_read_at = Arc::clone(&mock_read_at);
958 6 : async move { mock_read_at.lock().await.read_at(buf, offset).await }
959 6 : })
960 2 : .await;
961 2 : let Err(err) = res else {
962 2 : panic!("should return an error");
963 2 : };
964 2 : assert_eq!(err.kind(), std::io::ErrorKind::UnexpectedEof);
965 2 : assert_eq!(format!("{err}"), "failed to fill whole buffer");
966 2 : // buffer contents on error are unspecified
967 2 : }
968 : }
969 :
970 : struct FileGuard {
971 : slot_guard: RwLockReadGuard<'static, SlotInner>,
972 : }
973 :
974 : impl AsRef<OwnedFd> for FileGuard {
975 406593 : fn as_ref(&self) -> &OwnedFd {
976 406593 : // This unwrap is safe because we only create `FileGuard`s
977 406593 : // if we know that the file is Some.
978 406593 : self.slot_guard.file.as_ref().unwrap()
979 406593 : }
980 : }
981 :
982 : impl FileGuard {
983 : /// Soft deprecation: we'll move VirtualFile to async APIs and remove this function eventually.
984 203349 : fn with_std_file<F, R>(&self, with: F) -> R
985 203349 : where
986 203349 : F: FnOnce(&File) -> R,
987 203349 : {
988 203349 : // SAFETY:
989 203349 : // - lifetime of the fd: `file` doesn't outlive the OwnedFd stored in `self`.
990 203349 : // - `&` usage below: `self` is `&`, hence Rust typesystem guarantees there are is no `&mut`
991 203349 : let file = unsafe { File::from_raw_fd(self.as_ref().as_raw_fd()) };
992 203349 : let res = with(&file);
993 203349 : let _ = file.into_raw_fd();
994 203349 : res
995 203349 : }
996 : /// Soft deprecation: we'll move VirtualFile to async APIs and remove this function eventually.
997 4 : fn with_std_file_mut<F, R>(&mut self, with: F) -> R
998 4 : where
999 4 : F: FnOnce(&mut File) -> R,
1000 4 : {
1001 4 : // SAFETY:
1002 4 : // - lifetime of the fd: `file` doesn't outlive the OwnedFd stored in `self`.
1003 4 : // - &mut usage below: `self` is `&mut`, hence this call is the only task/thread that has control over the underlying fd
1004 4 : let mut file = unsafe { File::from_raw_fd(self.as_ref().as_raw_fd()) };
1005 4 : let res = with(&mut file);
1006 4 : let _ = file.into_raw_fd();
1007 4 : res
1008 4 : }
1009 : }
1010 :
1011 : impl tokio_epoll_uring::IoFd for FileGuard {
1012 203240 : unsafe fn as_fd(&self) -> RawFd {
1013 203240 : let owned_fd: &OwnedFd = self.as_ref();
1014 203240 : owned_fd.as_raw_fd()
1015 203240 : }
1016 : }
1017 :
1018 : #[cfg(test)]
1019 : impl VirtualFile {
1020 20200 : pub(crate) async fn read_blk(
1021 20200 : &self,
1022 20200 : blknum: u32,
1023 20200 : ) -> Result<crate::tenant::block_io::BlockLease<'_>, std::io::Error> {
1024 20200 : use crate::page_cache::PAGE_SZ;
1025 20200 : let buf = vec![0; PAGE_SZ];
1026 20200 : let buf = self
1027 20200 : .read_exact_at(buf, blknum as u64 * (PAGE_SZ as u64))
1028 10255 : .await?;
1029 20200 : Ok(crate::tenant::block_io::BlockLease::Vec(buf))
1030 20200 : }
1031 :
1032 224 : async fn read_to_end(&mut self, buf: &mut Vec<u8>) -> Result<(), Error> {
1033 224 : let mut tmp = vec![0; 128];
1034 : loop {
1035 : let res;
1036 444 : (tmp, res) = self.read_at(tmp, self.pos).await;
1037 2 : match res {
1038 222 : Ok(0) => return Ok(()),
1039 220 : Ok(n) => {
1040 220 : self.pos += n as u64;
1041 220 : buf.extend_from_slice(&tmp[..n]);
1042 220 : }
1043 2 : Err(ref e) if e.kind() == std::io::ErrorKind::Interrupted => {}
1044 2 : Err(e) => return Err(e),
1045 : }
1046 : }
1047 224 : }
1048 : }
1049 :
1050 : impl Drop for VirtualFile {
1051 : /// If a VirtualFile is dropped, close the underlying file if it was open.
1052 3809 : fn drop(&mut self) {
1053 3809 : let handle = self.handle.get_mut();
1054 3809 :
1055 3809 : fn clean_slot(slot: &Slot, mut slot_guard: RwLockWriteGuard<'_, SlotInner>, tag: u64) {
1056 3809 : if slot_guard.tag == tag {
1057 3809 : slot.recently_used.store(false, Ordering::Relaxed);
1058 3809 : // there is also operation "close-by-replace" for closes done on eviction for
1059 3809 : // comparison.
1060 3809 : if let Some(fd) = slot_guard.file.take() {
1061 3331 : STORAGE_IO_TIME_METRIC
1062 3331 : .get(StorageIoOperation::Close)
1063 3331 : .observe_closure_duration(|| drop(fd));
1064 3331 : }
1065 3809 : }
1066 3809 : }
1067 3809 :
1068 3809 : // We don't have async drop so we cannot directly await the lock here.
1069 3809 : // Instead, first do a best-effort attempt at closing the underlying
1070 3809 : // file descriptor by using `try_write`, and if that fails, spawn
1071 3809 : // a tokio task to do it asynchronously: we just want it to be
1072 3809 : // cleaned up eventually.
1073 3809 : // Most of the time, the `try_lock` should succeed though,
1074 3809 : // as we have `&mut self` access. In other words, if the slot
1075 3809 : // is still occupied by our file, there should be no access from
1076 3809 : // other I/O operations; the only other possible place to lock
1077 3809 : // the slot is the lock algorithm looking for free slots.
1078 3809 : let slot = &get_open_files().slots[handle.index];
1079 3809 : if let Ok(slot_guard) = slot.inner.try_write() {
1080 3809 : clean_slot(slot, slot_guard, handle.tag);
1081 3809 : } else {
1082 0 : let tag = handle.tag;
1083 0 : tokio::spawn(async move {
1084 0 : let slot_guard = slot.inner.write().await;
1085 0 : clean_slot(slot, slot_guard, tag);
1086 0 : });
1087 0 : };
1088 3809 : }
1089 : }
1090 :
1091 : impl OwnedAsyncWriter for VirtualFile {
1092 : #[inline(always)]
1093 3 : async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
1094 3 : &mut self,
1095 3 : buf: B,
1096 3 : ctx: &RequestContext,
1097 3 : ) -> std::io::Result<(usize, B::Buf)> {
1098 3 : let (buf, res) = VirtualFile::write_all(self, buf, ctx).await;
1099 3 : res.map(move |v| (v, buf))
1100 3 : }
1101 : }
1102 :
1103 : impl OpenFiles {
1104 126 : fn new(num_slots: usize) -> OpenFiles {
1105 126 : let mut slots = Box::new(Vec::with_capacity(num_slots));
1106 1260 : for _ in 0..num_slots {
1107 1260 : let slot = Slot {
1108 1260 : recently_used: AtomicBool::new(false),
1109 1260 : inner: RwLock::new(SlotInner { tag: 0, file: None }),
1110 1260 : };
1111 1260 : slots.push(slot);
1112 1260 : }
1113 :
1114 126 : OpenFiles {
1115 126 : next: AtomicUsize::new(0),
1116 126 : slots: Box::leak(slots),
1117 126 : }
1118 126 : }
1119 : }
1120 :
1121 : ///
1122 : /// Initialize the virtual file module. This must be called once at page
1123 : /// server startup.
1124 : ///
1125 : #[cfg(not(test))]
1126 0 : pub fn init(num_slots: usize, engine: IoEngineKind) {
1127 0 : if OPEN_FILES.set(OpenFiles::new(num_slots)).is_err() {
1128 0 : panic!("virtual_file::init called twice");
1129 0 : }
1130 0 : io_engine::init(engine);
1131 0 : crate::metrics::virtual_file_descriptor_cache::SIZE_MAX.set(num_slots as u64);
1132 0 : }
1133 :
1134 : const TEST_MAX_FILE_DESCRIPTORS: usize = 10;
1135 :
1136 : // Get a handle to the global slots array.
1137 414897 : fn get_open_files() -> &'static OpenFiles {
1138 414897 : //
1139 414897 : // In unit tests, page server startup doesn't happen and no one calls
1140 414897 : // virtual_file::init(). Initialize it here, with a small array.
1141 414897 : //
1142 414897 : // This applies to the virtual file tests below, but all other unit
1143 414897 : // tests too, so the virtual file facility is always usable in
1144 414897 : // unit tests.
1145 414897 : //
1146 414897 : if cfg!(test) {
1147 414897 : OPEN_FILES.get_or_init(|| OpenFiles::new(TEST_MAX_FILE_DESCRIPTORS))
1148 : } else {
1149 0 : OPEN_FILES.get().expect("virtual_file::init not called yet")
1150 : }
1151 414897 : }
1152 :
1153 : #[cfg(test)]
1154 : mod tests {
1155 : use crate::context::DownloadBehavior;
1156 : use crate::task_mgr::TaskKind;
1157 :
1158 : use super::*;
1159 : use rand::seq::SliceRandom;
1160 : use rand::thread_rng;
1161 : use rand::Rng;
1162 : use std::future::Future;
1163 : use std::io::Write;
1164 : use std::os::unix::fs::FileExt;
1165 : use std::sync::Arc;
1166 :
1167 : enum MaybeVirtualFile {
1168 : VirtualFile(VirtualFile),
1169 : File(File),
1170 : }
1171 :
1172 : impl From<VirtualFile> for MaybeVirtualFile {
1173 6 : fn from(vf: VirtualFile) -> Self {
1174 6 : MaybeVirtualFile::VirtualFile(vf)
1175 6 : }
1176 : }
1177 :
1178 : impl MaybeVirtualFile {
1179 404 : async fn read_exact_at(&self, mut buf: Vec<u8>, offset: u64) -> Result<Vec<u8>, Error> {
1180 404 : match self {
1181 202 : MaybeVirtualFile::VirtualFile(file) => file.read_exact_at(buf, offset).await,
1182 202 : MaybeVirtualFile::File(file) => file.read_exact_at(&mut buf, offset).map(|()| buf),
1183 : }
1184 404 : }
1185 8 : async fn write_all_at<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
1186 8 : &self,
1187 8 : buf: B,
1188 8 : offset: u64,
1189 8 : ctx: &RequestContext,
1190 8 : ) -> Result<(), Error> {
1191 8 : match self {
1192 4 : MaybeVirtualFile::VirtualFile(file) => {
1193 4 : let (_buf, res) = file.write_all_at(buf, offset, ctx).await;
1194 4 : res
1195 : }
1196 4 : MaybeVirtualFile::File(file) => {
1197 4 : let buf_len = buf.bytes_init();
1198 4 : if buf_len == 0 {
1199 0 : return Ok(());
1200 4 : }
1201 4 : file.write_all_at(&buf.slice(0..buf_len), offset)
1202 : }
1203 : }
1204 8 : }
1205 36 : async fn seek(&mut self, pos: SeekFrom) -> Result<u64, Error> {
1206 36 : match self {
1207 18 : MaybeVirtualFile::VirtualFile(file) => file.seek(pos).await,
1208 18 : MaybeVirtualFile::File(file) => file.seek(pos),
1209 : }
1210 36 : }
1211 8 : async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
1212 8 : &mut self,
1213 8 : buf: B,
1214 8 : ctx: &RequestContext,
1215 8 : ) -> Result<(), Error> {
1216 8 : match self {
1217 4 : MaybeVirtualFile::VirtualFile(file) => {
1218 4 : let (_buf, res) = file.write_all(buf, ctx).await;
1219 4 : res.map(|_| ())
1220 : }
1221 4 : MaybeVirtualFile::File(file) => {
1222 4 : let buf_len = buf.bytes_init();
1223 4 : if buf_len == 0 {
1224 0 : return Ok(());
1225 4 : }
1226 4 : file.write_all(&buf.slice(0..buf_len))
1227 : }
1228 : }
1229 8 : }
1230 :
1231 : // Helper function to slurp contents of a file, starting at the current position,
1232 : // into a string
1233 442 : async fn read_string(&mut self) -> Result<String, Error> {
1234 442 : use std::io::Read;
1235 442 : let mut buf = String::new();
1236 442 : match self {
1237 224 : MaybeVirtualFile::VirtualFile(file) => {
1238 224 : let mut buf = Vec::new();
1239 226 : file.read_to_end(&mut buf).await?;
1240 222 : return Ok(String::from_utf8(buf).unwrap());
1241 : }
1242 218 : MaybeVirtualFile::File(file) => {
1243 218 : file.read_to_string(&mut buf)?;
1244 : }
1245 : }
1246 216 : Ok(buf)
1247 442 : }
1248 :
1249 : // Helper function to slurp a portion of a file into a string
1250 404 : async fn read_string_at(&mut self, pos: u64, len: usize) -> Result<String, Error> {
1251 404 : let buf = vec![0; len];
1252 404 : let buf = self.read_exact_at(buf, pos).await?;
1253 404 : Ok(String::from_utf8(buf).unwrap())
1254 404 : }
1255 : }
1256 :
1257 : #[tokio::test]
1258 2 : async fn test_virtual_files() -> anyhow::Result<()> {
1259 2 : // The real work is done in the test_files() helper function. This
1260 2 : // allows us to run the same set of tests against a native File, and
1261 2 : // VirtualFile. We trust the native Files and wouldn't need to test them,
1262 2 : // but this allows us to verify that the operations return the same
1263 2 : // results with VirtualFiles as with native Files. (Except that with
1264 2 : // native files, you will run out of file descriptors if the ulimit
1265 2 : // is low enough.)
1266 206 : test_files("virtual_files", |path, open_options| async move {
1267 206 : let vf = VirtualFile::open_with_options(&path, &open_options).await?;
1268 206 : Ok(MaybeVirtualFile::VirtualFile(vf))
1269 206 : })
1270 530 : .await
1271 2 : }
1272 :
1273 : #[tokio::test]
1274 2 : async fn test_physical_files() -> anyhow::Result<()> {
1275 206 : test_files("physical_files", |path, open_options| async move {
1276 206 : Ok(MaybeVirtualFile::File({
1277 206 : let owned_fd = open_options.open(path.as_std_path()).await?;
1278 206 : File::from(owned_fd)
1279 206 : }))
1280 206 : })
1281 104 : .await
1282 2 : }
1283 :
1284 4 : async fn test_files<OF, FT>(testname: &str, openfunc: OF) -> anyhow::Result<()>
1285 4 : where
1286 4 : OF: Fn(Utf8PathBuf, OpenOptions) -> FT,
1287 4 : FT: Future<Output = Result<MaybeVirtualFile, std::io::Error>>,
1288 4 : {
1289 4 : let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
1290 4 : let testdir = crate::config::PageServerConf::test_repo_dir(testname);
1291 4 : std::fs::create_dir_all(&testdir)?;
1292 :
1293 4 : let path_a = testdir.join("file_a");
1294 4 : let mut file_a = openfunc(
1295 4 : path_a.clone(),
1296 4 : OpenOptions::new()
1297 4 : .write(true)
1298 4 : .create(true)
1299 4 : .truncate(true)
1300 4 : .to_owned(),
1301 4 : )
1302 4 : .await?;
1303 4 : file_a.write_all(b"foobar".to_vec(), &ctx).await?;
1304 :
1305 : // cannot read from a file opened in write-only mode
1306 4 : let _ = file_a.read_string().await.unwrap_err();
1307 :
1308 : // Close the file and re-open for reading
1309 4 : let mut file_a = openfunc(path_a, OpenOptions::new().read(true).to_owned()).await?;
1310 :
1311 : // cannot write to a file opened in read-only mode
1312 4 : let _ = file_a.write_all(b"bar".to_vec(), &ctx).await.unwrap_err();
1313 4 :
1314 4 : // Try simple read
1315 4 : assert_eq!("foobar", file_a.read_string().await?);
1316 :
1317 : // It's positioned at the EOF now.
1318 4 : assert_eq!("", file_a.read_string().await?);
1319 :
1320 : // Test seeks.
1321 4 : assert_eq!(file_a.seek(SeekFrom::Start(1)).await?, 1);
1322 4 : assert_eq!("oobar", file_a.read_string().await?);
1323 :
1324 4 : assert_eq!(file_a.seek(SeekFrom::End(-2)).await?, 4);
1325 4 : assert_eq!("ar", file_a.read_string().await?);
1326 :
1327 4 : assert_eq!(file_a.seek(SeekFrom::Start(1)).await?, 1);
1328 4 : assert_eq!(file_a.seek(SeekFrom::Current(2)).await?, 3);
1329 4 : assert_eq!("bar", file_a.read_string().await?);
1330 :
1331 4 : assert_eq!(file_a.seek(SeekFrom::Current(-5)).await?, 1);
1332 4 : assert_eq!("oobar", file_a.read_string().await?);
1333 :
1334 : // Test erroneous seeks to before byte 0
1335 4 : file_a.seek(SeekFrom::End(-7)).await.unwrap_err();
1336 4 : assert_eq!(file_a.seek(SeekFrom::Start(1)).await?, 1);
1337 4 : file_a.seek(SeekFrom::Current(-2)).await.unwrap_err();
1338 4 :
1339 4 : // the erroneous seek should have left the position unchanged
1340 4 : assert_eq!("oobar", file_a.read_string().await?);
1341 :
1342 : // Create another test file, and try FileExt functions on it.
1343 4 : let path_b = testdir.join("file_b");
1344 4 : let mut file_b = openfunc(
1345 4 : path_b.clone(),
1346 4 : OpenOptions::new()
1347 4 : .read(true)
1348 4 : .write(true)
1349 4 : .create(true)
1350 4 : .truncate(true)
1351 4 : .to_owned(),
1352 4 : )
1353 2 : .await?;
1354 4 : file_b.write_all_at(b"BAR".to_vec(), 3, &ctx).await?;
1355 4 : file_b.write_all_at(b"FOO".to_vec(), 0, &ctx).await?;
1356 :
1357 4 : assert_eq!(file_b.read_string_at(2, 3).await?, "OBA");
1358 :
1359 : // Open a lot of files, enough to cause some evictions. (Or to be precise,
1360 : // open the same file many times. The effect is the same.)
1361 : //
1362 : // leave file_a positioned at offset 1 before we start
1363 4 : assert_eq!(file_a.seek(SeekFrom::Start(1)).await?, 1);
1364 :
1365 4 : let mut vfiles = Vec::new();
1366 404 : for _ in 0..100 {
1367 400 : let mut vfile =
1368 400 : openfunc(path_b.clone(), OpenOptions::new().read(true).to_owned()).await?;
1369 400 : assert_eq!("FOOBAR", vfile.read_string().await?);
1370 400 : vfiles.push(vfile);
1371 : }
1372 :
1373 : // make sure we opened enough files to definitely cause evictions.
1374 4 : assert!(vfiles.len() > TEST_MAX_FILE_DESCRIPTORS * 2);
1375 :
1376 : // The underlying file descriptor for 'file_a' should be closed now. Try to read
1377 : // from it again. We left the file positioned at offset 1 above.
1378 4 : assert_eq!("oobar", file_a.read_string().await?);
1379 :
1380 : // Check that all the other FDs still work too. Use them in random order for
1381 : // good measure.
1382 4 : vfiles.as_mut_slice().shuffle(&mut thread_rng());
1383 400 : for vfile in vfiles.iter_mut() {
1384 400 : assert_eq!("OOBAR", vfile.read_string_at(1, 5).await?);
1385 : }
1386 :
1387 4 : Ok(())
1388 4 : }
1389 :
1390 : /// Test using VirtualFiles from many threads concurrently. This tests both using
1391 : /// a lot of VirtualFiles concurrently, causing evictions, and also using the same
1392 : /// VirtualFile from multiple threads concurrently.
1393 : #[tokio::test]
1394 2 : async fn test_vfile_concurrency() -> Result<(), Error> {
1395 2 : const SIZE: usize = 8 * 1024;
1396 2 : const VIRTUAL_FILES: usize = 100;
1397 2 : const THREADS: usize = 100;
1398 2 : const SAMPLE: [u8; SIZE] = [0xADu8; SIZE];
1399 2 :
1400 2 : let testdir = crate::config::PageServerConf::test_repo_dir("vfile_concurrency");
1401 2 : std::fs::create_dir_all(&testdir)?;
1402 2 :
1403 2 : // Create a test file.
1404 2 : let test_file_path = testdir.join("concurrency_test_file");
1405 2 : {
1406 2 : let file = File::create(&test_file_path)?;
1407 2 : file.write_all_at(&SAMPLE, 0)?;
1408 2 : }
1409 2 :
1410 2 : // Open the file many times.
1411 2 : let mut files = Vec::new();
1412 202 : for _ in 0..VIRTUAL_FILES {
1413 200 : let f = VirtualFile::open_with_options(&test_file_path, OpenOptions::new().read(true))
1414 101 : .await?;
1415 200 : files.push(f);
1416 2 : }
1417 2 : let files = Arc::new(files);
1418 2 :
1419 2 : // Launch many threads, and use the virtual files concurrently in random order.
1420 2 : let rt = tokio::runtime::Builder::new_multi_thread()
1421 2 : .worker_threads(THREADS)
1422 2 : .thread_name("test_vfile_concurrency thread")
1423 2 : .build()
1424 2 : .unwrap();
1425 2 : let mut hdls = Vec::new();
1426 202 : for _threadno in 0..THREADS {
1427 200 : let files = files.clone();
1428 200 : let hdl = rt.spawn(async move {
1429 200 : let mut buf = vec![0u8; SIZE];
1430 200 : let mut rng = rand::rngs::OsRng;
1431 200000 : for _ in 1..1000 {
1432 199800 : let f = &files[rng.gen_range(0..files.len())];
1433 655859 : buf = f.read_exact_at(buf, 0).await.unwrap();
1434 199800 : assert!(buf == SAMPLE);
1435 2 : }
1436 200 : });
1437 200 : hdls.push(hdl);
1438 200 : }
1439 202 : for hdl in hdls {
1440 200 : hdl.await?;
1441 2 : }
1442 2 : std::mem::forget(rt);
1443 2 :
1444 2 : Ok(())
1445 2 : }
1446 :
1447 : #[tokio::test]
1448 2 : async fn test_atomic_overwrite_basic() {
1449 2 : let testdir = crate::config::PageServerConf::test_repo_dir("test_atomic_overwrite_basic");
1450 2 : std::fs::create_dir_all(&testdir).unwrap();
1451 2 :
1452 2 : let path = testdir.join("myfile");
1453 2 : let tmp_path = testdir.join("myfile.tmp");
1454 2 :
1455 2 : VirtualFile::crashsafe_overwrite(path.clone(), tmp_path.clone(), b"foo".to_vec())
1456 2 : .await
1457 2 : .unwrap();
1458 2 : let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
1459 2 : let post = file.read_string().await.unwrap();
1460 2 : assert_eq!(post, "foo");
1461 2 : assert!(!tmp_path.exists());
1462 2 : drop(file);
1463 2 :
1464 2 : VirtualFile::crashsafe_overwrite(path.clone(), tmp_path.clone(), b"bar".to_vec())
1465 2 : .await
1466 2 : .unwrap();
1467 2 : let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
1468 2 : let post = file.read_string().await.unwrap();
1469 2 : assert_eq!(post, "bar");
1470 2 : assert!(!tmp_path.exists());
1471 2 : drop(file);
1472 2 : }
1473 :
1474 : #[tokio::test]
1475 2 : async fn test_atomic_overwrite_preexisting_tmp() {
1476 2 : let testdir =
1477 2 : crate::config::PageServerConf::test_repo_dir("test_atomic_overwrite_preexisting_tmp");
1478 2 : std::fs::create_dir_all(&testdir).unwrap();
1479 2 :
1480 2 : let path = testdir.join("myfile");
1481 2 : let tmp_path = testdir.join("myfile.tmp");
1482 2 :
1483 2 : std::fs::write(&tmp_path, "some preexisting junk that should be removed").unwrap();
1484 2 : assert!(tmp_path.exists());
1485 2 :
1486 2 : VirtualFile::crashsafe_overwrite(path.clone(), tmp_path.clone(), b"foo".to_vec())
1487 2 : .await
1488 2 : .unwrap();
1489 2 :
1490 2 : let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
1491 2 : let post = file.read_string().await.unwrap();
1492 2 : assert_eq!(post, "foo");
1493 2 : assert!(!tmp_path.exists());
1494 2 : drop(file);
1495 2 : }
1496 : }
|
