foundationdb/fdbrpc/AsyncFileEIO.actor.h

565 lines
16 KiB
C++

/*
* AsyncFileEIO.actor.h
*
* This source file is part of the FoundationDB open source project
*
* Copyright 2013-2022 Apple Inc. and the FoundationDB project authors
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#pragma once
#if defined(__unixish__)
#define Net2AsyncFile AsyncFileEIO
// When actually compiled (NO_INTELLISENSE), include the generated version of this file. In intellisense use the source
// version.
#if defined(NO_INTELLISENSE) && !defined(FLOW_ASYNCFILEEIO_ACTOR_G_H)
#define FLOW_ASYNCFILEEIO_ACTOR_G_H
#include "fdbrpc/AsyncFileEIO.actor.g.h"
#elif !defined(FLOW_ASYNCFILEEIO_ACTOR_H)
#define FLOW_ASYNCFILEEIO_ACTOR_H
#include <fcntl.h>
#include <sys/stat.h>
#include "fdbrpc/libeio/eio.h"
#include "flow/flow.h"
#include "flow/ThreadHelper.actor.h"
#include "fdbrpc/IAsyncFile.h"
#include "flow/TDMetric.actor.h"
#include "flow/actorcompiler.h" // This must be the last #include.
class AsyncFileEIO : public IAsyncFile, public ReferenceCounted<AsyncFileEIO> {
public:
static void init() {
eio_set_max_parallel(FLOW_KNOBS->EIO_MAX_PARALLELISM);
if (eio_init(&eio_want_poll, nullptr)) {
TraceEvent("EioInitError").detail("ErrorNo", errno);
throw platform_error();
}
}
static void stop() { eio_set_max_parallel(0); }
static bool should_poll() { return want_poll; }
static bool lock_fd(int fd) {
// Acquire a "write" lock for the entire file
struct flock lockDesc;
lockDesc.l_type = F_WRLCK;
lockDesc.l_whence = SEEK_SET;
lockDesc.l_start = 0;
lockDesc.l_len =
0; // "Specifying 0 for l_len has the special meaning: lock all bytes starting at the location specified by
// l_whence and l_start through to the end of file, no matter how large the file grows."
lockDesc.l_pid = 0;
if (fcntl(fd, F_SETLK, &lockDesc) == -1) {
return false;
}
return true;
}
ACTOR static Future<Reference<IAsyncFile>> open(std::string filename, int flags, int mode, void* ignore) {
std::string open_filename = filename;
if (flags & OPEN_ATOMIC_WRITE_AND_CREATE) {
ASSERT((flags & OPEN_CREATE) && (flags & OPEN_READWRITE) && !(flags & OPEN_EXCLUSIVE));
open_filename = filename + ".part";
}
state Promise<Void> p;
state eio_req* r = eio_open(open_filename.c_str(), openFlags(flags), mode, 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
eio_cancel(r);
throw;
}
if (r->result < 0) {
errno = r->errorno;
bool notFound = errno == ENOENT;
Error e = notFound ? file_not_found() : io_error();
TraceEvent(notFound ? SevWarn : SevWarnAlways, "FileOpenError")
.error(e)
.GetLastError()
.detail("File", filename)
.detail("Flags", flags)
.detail("Mode", mode);
throw e;
}
TraceEvent("AsyncFileOpened")
.suppressFor(1.0)
.detail("Filename", filename)
.detail("Fd", r->result)
.detail("Flags", flags);
if ((flags & OPEN_LOCK) && !lock_fd(r->result)) {
TraceEvent(SevError, "UnableToLockFile").detail("Filename", filename).GetLastError();
throw io_error();
}
return Reference<IAsyncFile>(new AsyncFileEIO(r->result, flags, filename));
}
static Future<Void> deleteFile(std::string filename, bool mustBeDurable) {
::deleteFile(filename);
if (mustBeDurable) {
TEST(true); // deleteFile and fsync parent dir
return async_fsync_parent(filename);
} else
return Void();
}
ACTOR static Future<Void> renameFile(std::string from, std::string to) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state eio_req* r = eio_rename(from.c_str(), to.c_str(), 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
try {
state int result = r->result;
if (result == -1) {
TraceEvent(SevError, "FileRenameError").detail("Errno", r->errorno);
throw internal_error();
} else {
wait(delay(0, taskID));
return Void();
}
} catch (Error& e) {
state Error _e = e;
wait(delay(0, taskID));
throw _e;
}
}
ACTOR static Future<std::time_t> lastWriteTime(std::string filename) {
EIO_STRUCT_STAT statdata = wait(stat_impl(filename));
return statdata.st_mtime;
}
void addref() override { ReferenceCounted<AsyncFileEIO>::addref(); }
void delref() override { ReferenceCounted<AsyncFileEIO>::delref(); }
int64_t debugFD() const override { return fd; }
Future<int> read(void* data, int length, int64_t offset) override {
++countFileLogicalReads;
++countLogicalReads;
return read_impl(fd, data, length, offset);
}
Future<Void> write(void const* data, int length, int64_t offset) override // Copies data synchronously
{
++countFileLogicalWrites;
++countLogicalWrites;
// Standalone<StringRef> copy = StringRef((const uint8_t*)data, length);
return write_impl(fd, err, StringRef((const uint8_t*)data, length), offset);
}
Future<Void> truncate(int64_t size) override {
++countFileLogicalWrites;
++countLogicalWrites;
return truncate_impl(fd, err, size);
}
Future<Void> sync() override {
++countFileLogicalWrites;
++countLogicalWrites;
auto fsync = sync_impl(fd, err);
if (flags & OPEN_ATOMIC_WRITE_AND_CREATE) {
flags &= ~OPEN_ATOMIC_WRITE_AND_CREATE;
return waitAndAtomicRename(fsync, filename + ".part", filename);
}
return fsync;
}
Future<int64_t> size() const override {
++countFileLogicalReads;
++countLogicalReads;
return size_impl(fd);
}
std::string getFilename() const override { return filename; }
ACTOR static Future<Void> async_fsync_parent(std::string filename) {
std::string folder = parentDirectory(filename);
TraceEvent("FSyncParentDir").detail("Folder", folder).detail("File", filename);
state int folderFD = ::open(folder.c_str(), O_DIRECTORY | O_CLOEXEC, 0);
if (folderFD < 0)
throw io_error();
try {
wait(async_fsync(folderFD)); // not sure if fdatasync on the folder has the same effect
} catch (...) {
close(folderFD);
throw;
}
close(folderFD);
return Void();
}
static Future<Void> async_fdatasync(int fd) {
// Used by AsyncFileKAIO, since kernel AIO doesn't really implement fdatasync yet
return sync_impl(fd, makeReference<ErrorInfo>());
}
static Future<Void> async_fsync(int fd) {
// Used by AsyncFileKAIO, since kernel AIO doesn't really implement fsync yet
return sync_impl(fd, makeReference<ErrorInfo>(), true);
}
ACTOR static Future<Void> waitAndAtomicRename(Future<Void> fsync,
std::string part_filename,
std::string final_filename) {
// First wait for the data in the part file to be durable
wait(fsync);
// rename() is atomic
if (rename(part_filename.c_str(), final_filename.c_str())) {
TraceEvent("AsyncFileEIORenameError").detail("Filename", final_filename).GetLastError();
throw io_error();
}
// fsync the parent directory to make it durable as well
wait(async_fsync_parent(final_filename));
return Void();
}
// Run the given function on the EIO thread pool and return its result
template <class R>
static Future<R> dispatch(std::function<R()> const& func) {
return dispatch_impl(func);
}
~AsyncFileEIO() override { close_impl(fd); }
private:
struct ErrorInfo : ReferenceCounted<ErrorInfo>, FastAllocated<ErrorInfo> {
Error err;
void set(const Error& e) {
if (err.code() == invalid_error_code)
err = e;
}
void report() {
if (err.code() != invalid_error_code)
throw err;
}
};
template <class R>
struct Dispatch {
std::function<R()> func;
ErrorOr<R> result;
Promise<Void> done;
explicit Dispatch(std::function<R()> const& func) : func(func) {}
};
int fd, flags;
Reference<ErrorInfo> err;
std::string filename;
mutable Int64MetricHandle countFileLogicalWrites;
mutable Int64MetricHandle countFileLogicalReads;
mutable Int64MetricHandle countLogicalWrites;
mutable Int64MetricHandle countLogicalReads;
AsyncFileEIO(int fd, int flags, std::string const& filename)
: fd(fd), flags(flags), err(new ErrorInfo), filename(filename) {
if (!g_network->isSimulated()) {
countFileLogicalWrites.init(LiteralStringRef("AsyncFile.CountFileLogicalWrites"), filename);
countFileLogicalReads.init(LiteralStringRef("AsyncFile.CountFileLogicalReads"), filename);
countLogicalWrites.init(LiteralStringRef("AsyncFile.CountLogicalWrites"));
countLogicalReads.init(LiteralStringRef("AsyncFile.CountLogicalReads"));
}
}
static int openFlags(int flags) {
int oflags = O_CLOEXEC;
ASSERT(bool(flags & OPEN_READONLY) != bool(flags & OPEN_READWRITE)); // readonly xor readwrite
if (flags & OPEN_EXCLUSIVE)
oflags |= O_EXCL;
if (flags & OPEN_CREATE)
oflags |= O_CREAT;
if (flags & OPEN_READONLY)
oflags |= O_RDONLY;
if (flags & OPEN_READWRITE)
oflags |= O_RDWR;
if (flags & OPEN_ATOMIC_WRITE_AND_CREATE)
oflags |= O_TRUNC;
#if defined(__linux__)
if (flags & OPEN_UNBUFFERED && FLOW_KNOBS->EIO_USE_ODIRECT)
oflags |= O_DIRECT;
#endif
return oflags;
}
static void error(const char* context,
int fd,
eio_req* r,
Reference<ErrorInfo> const& err = Reference<ErrorInfo>()) {
Error e = io_error();
errno = r->errorno;
TraceEvent(context).error(e).detail("Fd", fd).detail("Result", r->result).GetLastError();
if (err)
err->set(e);
else
throw e;
}
ACTOR static void close_impl(int fd) {
state Promise<Void> p;
state eio_req* r = eio_close(fd, 0, eio_callback, &p);
wait(p.getFuture());
if (r->result)
error("CloseError", fd, r);
TraceEvent("AsyncFileClosed").suppressFor(1.0).detail("Fd", fd);
}
ACTOR static Future<int> read_impl(int fd, void* data, int length, int64_t offset) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
// fprintf(stderr, "eio_read (fd=%d length=%d offset=%lld)\n", fd, length, offset);
state eio_req* r = eio_read(fd, data, length, offset, 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
try {
state int result = r->result;
// printf("eio read: %d/%d\n", r->result, length);
if (result == -1) {
error("ReadError", fd, r);
throw internal_error();
} else {
wait(delay(0, taskID));
return result;
}
} catch (Error& _e) {
state Error e = _e;
wait(delay(0, taskID));
throw e;
}
}
ACTOR static Future<Void> write_impl(int fd, Reference<ErrorInfo> err, StringRef data, int64_t offset) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state eio_req* r = eio_write(fd, (void*)data.begin(), data.size(), offset, 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
if (r->result != data.size())
error("WriteError", fd, r, err);
wait(delay(0, taskID));
return Void();
}
ACTOR static Future<Void> truncate_impl(int fd, Reference<ErrorInfo> err, int64_t size) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state eio_req* r = eio_ftruncate(fd, size, 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
if (r->result)
error("TruncateError", fd, r, err);
wait(delay(0, taskID));
return Void();
}
static eio_req* start_fsync(int fd, Promise<Void>& p, bool sync_metadata) {
#ifdef __APPLE__
// Neither fsync() nor fdatasync() do the right thing on OS X!
eio_req* req = (eio_req*)calloc(1, sizeof *req);
req->type = EIO_CUSTOM;
req->pri = 0;
req->finish = eio_callback;
req->data = &p;
req->destroy = free_req;
req->int1 = fd;
req->feed = apple_fsync;
eio_submit(req);
return req;
#else
if (sync_metadata)
return eio_fsync(fd, 0, eio_callback, &p);
else
return eio_fdatasync(fd, 0, eio_callback, &p);
#endif
}
ACTOR static Future<Void> sync_impl(int fd, Reference<ErrorInfo> err, bool sync_metadata = false) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state eio_req* r = start_fsync(fd, p, sync_metadata);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
try {
// Report any errors from prior write() or truncate() calls
err->report();
if (r->result)
error("SyncError", fd, r);
wait(delay(0, taskID));
return Void();
} catch (Error& _e) {
state Error e = _e;
wait(delay(0, taskID));
throw e;
}
}
ACTOR static Future<int64_t> size_impl(int fd) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state eio_req* r = eio_fstat(fd, 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
if (r->result)
error("FStatError", fd, r);
EIO_STRUCT_STAT* statdata = (EIO_STRUCT_STAT*)r->ptr2;
if (!statdata)
error("FStatBufferError", fd, r);
state int64_t size = statdata->st_size;
wait(delay(0, taskID));
return size;
}
ACTOR static Future<EIO_STRUCT_STAT> stat_impl(std::string filename) {
state TaskPriority taskID = g_network->getCurrentTask();
state Promise<Void> p;
state EIO_STRUCT_STAT statdata;
state eio_req* r = eio_stat(filename.c_str(), 0, eio_callback, &p);
try {
wait(p.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
if (r->result)
error("StatError", 0, r);
if (!r->ptr2)
error("StatBufferError", 0, r);
statdata = *EIO_STAT_BUF(r);
wait(delay(0, taskID));
return statdata;
}
ACTOR template <class R>
static Future<R> dispatch_impl(std::function<R()> func) {
state Dispatch<R> data(func);
state TaskPriority taskID = g_network->getCurrentTask();
state eio_req* r = eio_custom(
[](eio_req* req) {
// Runs on the eio thread pool
auto data = reinterpret_cast<Dispatch<R>*>(req->data);
try {
data->result = data->func();
req->result = 0;
} catch (Error& e) {
data->result = e;
req->result = -1;
} catch (...) {
data->result = unknown_error();
req->result = -1;
}
},
0,
[](eio_req* req) {
// Runs on the main thread, in eio_poll()
if (EIO_CANCELLED(req))
return 0;
auto data = reinterpret_cast<Dispatch<R>*>(req->data);
Promise<Void> p = std::move(data->done);
p.send(Void());
return 0;
},
&data);
try {
wait(data.done.getFuture());
} catch (...) {
g_network->setCurrentTask(taskID);
eio_cancel(r);
throw;
}
wait(delay(0, taskID));
if (data.result.isError())
throw data.result.getError();
return data.result.get();
}
static std::atomic<int32_t> want_poll;
ACTOR static void poll_eio() {
while (eio_poll() == -1)
wait(yield());
want_poll = 0;
}
static void eio_want_poll() {
want_poll = 1;
// SOMEDAY: nullptr for deferred error, no analysis of correctness (itp)
onMainThreadVoid([]() { poll_eio(); }, nullptr, TaskPriority::PollEIO);
}
static int eio_callback(eio_req* req) {
if (EIO_CANCELLED(req))
return 0;
Promise<Void> p = std::move(*(Promise<Void>*)req->data);
p.send(Void());
return 0;
}
#ifdef __APPLE__
static void apple_fsync(eio_req* req) { req->result = fcntl(req->int1, F_FULLFSYNC, 0); }
static void free_req(eio_req* req) { free(req); }
#endif
};
#ifdef FILESYSTEM_IMPL
std::atomic<int32_t> AsyncFileEIO::want_poll = 0;
#endif
#include "flow/unactorcompiler.h"
#endif
#endif