blob: 4c6e063fa114d4c003322c49cc1ac70a1cc37fdd [file] [log] [blame]
// Copyright (c) 2011 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
// For 64-bit file access (off_t = off64_t, lseek64, etc).
#define _FILE_OFFSET_BITS 64
#include "net/base/file_stream.h"
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>
#include "base/basictypes.h"
#include "base/callback.h"
#include "base/eintr_wrapper.h"
#include "base/file_path.h"
#include "base/logging.h"
#include "base/message_loop.h"
#include "base/metrics/histogram.h"
#include "base/string_util.h"
#include "base/task.h"
#include "base/threading/thread_restrictions.h"
#include "base/threading/worker_pool.h"
#include "base/synchronization/waitable_event.h"
#include "net/base/net_errors.h"
namespace net {
// We cast back and forth, so make sure it's the size we're expecting.
#if defined(__BIONIC__) && defined(ANDROID)
COMPILE_ASSERT(sizeof(int32) == sizeof(off_t), off_t_32_bit);
#else
COMPILE_ASSERT(sizeof(int64) == sizeof(off_t), off_t_64_bit);
#endif
// Make sure our Whence mappings match the system headers.
COMPILE_ASSERT(FROM_BEGIN == SEEK_SET &&
FROM_CURRENT == SEEK_CUR &&
FROM_END == SEEK_END, whence_matches_system);
namespace {
// Map from errno to net error codes.
int64 MapErrorCode(int err) {
switch (err) {
case ENOENT:
return ERR_FILE_NOT_FOUND;
case EACCES:
return ERR_ACCESS_DENIED;
default:
LOG(WARNING) << "Unknown error " << err << " mapped to net::ERR_FAILED";
return ERR_FAILED;
}
}
// ReadFile() is a simple wrapper around read() that handles EINTR signals and
// calls MapErrorCode() to map errno to net error codes.
int ReadFile(base::PlatformFile file, char* buf, int buf_len) {
base::ThreadRestrictions::AssertIOAllowed();
// read(..., 0) returns 0 to indicate end-of-file.
// Loop in the case of getting interrupted by a signal.
ssize_t res = HANDLE_EINTR(read(file, buf, static_cast<size_t>(buf_len)));
if (res == static_cast<ssize_t>(-1))
return MapErrorCode(errno);
return static_cast<int>(res);
}
void ReadFileTask(base::PlatformFile file,
char* buf,
int buf_len,
CompletionCallback* callback) {
callback->Run(ReadFile(file, buf, buf_len));
}
// WriteFile() is a simple wrapper around write() that handles EINTR signals and
// calls MapErrorCode() to map errno to net error codes. It tries to write to
// completion.
int WriteFile(base::PlatformFile file, const char* buf, int buf_len) {
base::ThreadRestrictions::AssertIOAllowed();
ssize_t res = HANDLE_EINTR(write(file, buf, buf_len));
if (res == -1)
return MapErrorCode(errno);
return res;
}
void WriteFileTask(base::PlatformFile file,
const char* buf,
int buf_len,
CompletionCallback* callback) {
callback->Run(WriteFile(file, buf, buf_len));
}
// FlushFile() is a simple wrapper around fsync() that handles EINTR signals and
// calls MapErrorCode() to map errno to net error codes. It tries to flush to
// completion.
int FlushFile(base::PlatformFile file) {
base::ThreadRestrictions::AssertIOAllowed();
ssize_t res = HANDLE_EINTR(fsync(file));
if (res == -1)
return MapErrorCode(errno);
return res;
}
} // namespace
// CancelableCallbackTask takes ownership of the Callback. This task gets
// posted to the MessageLoopForIO instance.
class CancelableCallbackTask : public CancelableTask {
public:
explicit CancelableCallbackTask(Callback0::Type* callback)
: canceled_(false), callback_(callback) {}
virtual void Run() {
if (!canceled_)
callback_->Run();
}
virtual void Cancel() {
canceled_ = true;
}
private:
bool canceled_;
scoped_ptr<Callback0::Type> callback_;
};
// FileStream::AsyncContext ----------------------------------------------
class FileStream::AsyncContext {
public:
AsyncContext();
~AsyncContext();
// These methods post synchronous read() and write() calls to a WorkerThread.
void InitiateAsyncRead(
base::PlatformFile file, char* buf, int buf_len,
CompletionCallback* callback);
void InitiateAsyncWrite(
base::PlatformFile file, const char* buf, int buf_len,
CompletionCallback* callback);
CompletionCallback* callback() const { return callback_; }
// Called by the WorkerPool thread executing the IO after the IO completes.
// This method queues RunAsynchronousCallback() on the MessageLoop and signals
// |background_io_completed_callback_|, in case the destructor is waiting. In
// that case, the destructor will call RunAsynchronousCallback() instead, and
// cancel |message_loop_task_|.
// |result| is the result of the Read/Write task.
void OnBackgroundIOCompleted(int result);
private:
// Always called on the IO thread, either directly by a task on the
// MessageLoop or by ~AsyncContext().
void RunAsynchronousCallback();
// The MessageLoopForIO that this AsyncContext is running on.
MessageLoopForIO* const message_loop_;
CompletionCallback* callback_; // The user provided callback.
// A callback wrapper around OnBackgroundIOCompleted(). Run by the WorkerPool
// thread doing the background IO on our behalf.
CompletionCallbackImpl<AsyncContext> background_io_completed_callback_;
// This is used to synchronize between the AsyncContext destructor (which runs
// on the IO thread and OnBackgroundIOCompleted() which runs on the WorkerPool
// thread.
base::WaitableEvent background_io_completed_;
// These variables are only valid when background_io_completed is signaled.
int result_;
CancelableCallbackTask* message_loop_task_;
bool is_closing_;
DISALLOW_COPY_AND_ASSIGN(AsyncContext);
};
FileStream::AsyncContext::AsyncContext()
: message_loop_(MessageLoopForIO::current()),
callback_(NULL),
background_io_completed_callback_(
this, &AsyncContext::OnBackgroundIOCompleted),
background_io_completed_(true, false),
message_loop_task_(NULL),
is_closing_(false) {}
FileStream::AsyncContext::~AsyncContext() {
is_closing_ = true;
if (callback_) {
// If |callback_| is non-NULL, that implies either the worker thread is
// still running the IO task, or the completion callback is queued up on the
// MessageLoopForIO, but AsyncContext() got deleted before then.
const bool need_to_wait = !background_io_completed_.IsSignaled();
base::TimeTicks start = base::TimeTicks::Now();
RunAsynchronousCallback();
if (need_to_wait) {
// We want to see if we block the message loop for too long.
UMA_HISTOGRAM_TIMES("AsyncIO.FileStreamClose",
base::TimeTicks::Now() - start);
}
}
}
void FileStream::AsyncContext::InitiateAsyncRead(
base::PlatformFile file, char* buf, int buf_len,
CompletionCallback* callback) {
DCHECK(!callback_);
callback_ = callback;
base::WorkerPool::PostTask(FROM_HERE,
NewRunnableFunction(
&ReadFileTask,
file, buf, buf_len,
&background_io_completed_callback_),
true /* task_is_slow */);
}
void FileStream::AsyncContext::InitiateAsyncWrite(
base::PlatformFile file, const char* buf, int buf_len,
CompletionCallback* callback) {
DCHECK(!callback_);
callback_ = callback;
base::WorkerPool::PostTask(FROM_HERE,
NewRunnableFunction(
&WriteFileTask,
file, buf, buf_len,
&background_io_completed_callback_),
true /* task_is_slow */);
}
void FileStream::AsyncContext::OnBackgroundIOCompleted(int result) {
result_ = result;
message_loop_task_ = new CancelableCallbackTask(
NewCallback(this, &AsyncContext::RunAsynchronousCallback));
message_loop_->PostTask(FROM_HERE, message_loop_task_);
background_io_completed_.Signal();
}
void FileStream::AsyncContext::RunAsynchronousCallback() {
// Wait() here ensures that all modifications from the WorkerPool thread are
// now visible.
background_io_completed_.Wait();
// Either we're in the MessageLoop's task, in which case Cancel() doesn't do
// anything, or we're in ~AsyncContext(), in which case this prevents the call
// from happening again. Must do it here after calling Wait().
message_loop_task_->Cancel();
message_loop_task_ = NULL;
if (is_closing_) {
callback_ = NULL;
return;
}
DCHECK(callback_);
CompletionCallback* temp = NULL;
std::swap(temp, callback_);
background_io_completed_.Reset();
temp->Run(result_);
}
// FileStream ------------------------------------------------------------
FileStream::FileStream()
: file_(base::kInvalidPlatformFileValue),
open_flags_(0),
auto_closed_(true) {
DCHECK(!IsOpen());
}
FileStream::FileStream(base::PlatformFile file, int flags)
: file_(file),
open_flags_(flags),
auto_closed_(false) {
// If the file handle is opened with base::PLATFORM_FILE_ASYNC, we need to
// make sure we will perform asynchronous File IO to it.
if (flags & base::PLATFORM_FILE_ASYNC) {
async_context_.reset(new AsyncContext());
}
}
FileStream::~FileStream() {
if (auto_closed_)
Close();
}
void FileStream::Close() {
// Abort any existing asynchronous operations.
async_context_.reset();
if (file_ != base::kInvalidPlatformFileValue) {
if (close(file_) != 0) {
NOTREACHED();
}
file_ = base::kInvalidPlatformFileValue;
}
}
int FileStream::Open(const FilePath& path, int open_flags) {
if (IsOpen()) {
DLOG(FATAL) << "File is already open!";
return ERR_UNEXPECTED;
}
open_flags_ = open_flags;
file_ = base::CreatePlatformFile(path, open_flags_, NULL, NULL);
if (file_ == base::kInvalidPlatformFileValue) {
return MapErrorCode(errno);
}
if (open_flags_ & base::PLATFORM_FILE_ASYNC) {
async_context_.reset(new AsyncContext());
}
return OK;
}
bool FileStream::IsOpen() const {
return file_ != base::kInvalidPlatformFileValue;
}
int64 FileStream::Seek(Whence whence, int64 offset) {
base::ThreadRestrictions::AssertIOAllowed();
if (!IsOpen())
return ERR_UNEXPECTED;
// If we're in async, make sure we don't have a request in flight.
DCHECK(!async_context_.get() || !async_context_->callback());
off_t res = lseek(file_, static_cast<off_t>(offset),
static_cast<int>(whence));
if (res == static_cast<off_t>(-1))
return MapErrorCode(errno);
return res;
}
int64 FileStream::Available() {
base::ThreadRestrictions::AssertIOAllowed();
if (!IsOpen())
return ERR_UNEXPECTED;
int64 cur_pos = Seek(FROM_CURRENT, 0);
if (cur_pos < 0)
return cur_pos;
struct stat info;
if (fstat(file_, &info) != 0)
return MapErrorCode(errno);
int64 size = static_cast<int64>(info.st_size);
DCHECK_GT(size, cur_pos);
return size - cur_pos;
}
int FileStream::Read(
char* buf, int buf_len, CompletionCallback* callback) {
if (!IsOpen())
return ERR_UNEXPECTED;
// read(..., 0) will return 0, which indicates end-of-file.
DCHECK(buf_len > 0);
DCHECK(open_flags_ & base::PLATFORM_FILE_READ);
if (async_context_.get()) {
DCHECK(open_flags_ & base::PLATFORM_FILE_ASYNC);
// If we're in async, make sure we don't have a request in flight.
DCHECK(!async_context_->callback());
async_context_->InitiateAsyncRead(file_, buf, buf_len, callback);
return ERR_IO_PENDING;
} else {
return ReadFile(file_, buf, buf_len);
}
}
int FileStream::ReadUntilComplete(char *buf, int buf_len) {
int to_read = buf_len;
int bytes_total = 0;
do {
int bytes_read = Read(buf, to_read, NULL);
if (bytes_read <= 0) {
if (bytes_total == 0)
return bytes_read;
return bytes_total;
}
bytes_total += bytes_read;
buf += bytes_read;
to_read -= bytes_read;
} while (bytes_total < buf_len);
return bytes_total;
}
int FileStream::Write(
const char* buf, int buf_len, CompletionCallback* callback) {
// write(..., 0) will return 0, which indicates end-of-file.
DCHECK_GT(buf_len, 0);
if (!IsOpen())
return ERR_UNEXPECTED;
if (async_context_.get()) {
DCHECK(open_flags_ & base::PLATFORM_FILE_ASYNC);
// If we're in async, make sure we don't have a request in flight.
DCHECK(!async_context_->callback());
async_context_->InitiateAsyncWrite(file_, buf, buf_len, callback);
return ERR_IO_PENDING;
} else {
return WriteFile(file_, buf, buf_len);
}
}
int64 FileStream::Truncate(int64 bytes) {
base::ThreadRestrictions::AssertIOAllowed();
if (!IsOpen())
return ERR_UNEXPECTED;
// We better be open for reading.
DCHECK(open_flags_ & base::PLATFORM_FILE_WRITE);
// Seek to the position to truncate from.
int64 seek_position = Seek(FROM_BEGIN, bytes);
if (seek_position != bytes)
return ERR_UNEXPECTED;
// And truncate the file.
int result = ftruncate(file_, bytes);
return result == 0 ? seek_position : MapErrorCode(errno);
}
int FileStream::Flush() {
if (!IsOpen())
return ERR_UNEXPECTED;
return FlushFile(file_);
}
} // namespace net