Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(591)

Side by Side Diff: net/disk_cache/file_ios.cc

Issue 103533012: Disk Cache: Use the old WorkerPool for Async IO on iOS. (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/src
Patch Set: Created 6 years, 11 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
OLDNEW
1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be 2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file. 3 // found in the LICENSE file.
4 4
5 #include "net/disk_cache/file.h" 5 #include "net/disk_cache/file.h"
6 6
7 #include "base/bind.h" 7 #include "base/bind.h"
8 #include "base/lazy_instance.h"
9 #include "base/location.h" 8 #include "base/location.h"
10 #include "base/logging.h" 9 #include "base/logging.h"
11 #include "base/run_loop.h" 10 #include "base/threading/worker_pool.h"
12 #include "base/task_runner_util.h"
13 #include "base/threading/sequenced_worker_pool.h"
14 #include "net/base/net_errors.h" 11 #include "net/base/net_errors.h"
15 #include "net/disk_cache/disk_cache.h" 12 #include "net/disk_cache/disk_cache.h"
13 #include "net/disk_cache/in_flight_io.h"
16 14
17 namespace { 15 namespace {
18 16
19 // The maximum number of threads for this pool. 17 // This class represents a single asynchronous IO operation while it is being
20 const int kMaxThreads = 5; 18 // bounced between threads.
19 class FileBackgroundIO : public disk_cache::BackgroundIO {
20 public:
21 // Other than the actual parameters for the IO operation (including the
22 // |callback| that must be notified at the end), we need the controller that
23 // is keeping track of all operations. When done, we notify the controller
24 // (we do NOT invoke the callback), in the worker thead that completed the
25 // operation.
26 FileBackgroundIO(disk_cache::File* file, const void* buf, size_t buf_len,
27 size_t offset, disk_cache::FileIOCallback* callback,
28 disk_cache::InFlightIO* controller)
29 : disk_cache::BackgroundIO(controller), callback_(callback), file_(file),
30 buf_(buf), buf_len_(buf_len), offset_(offset) {
31 }
21 32
22 class FileWorkerPool : public base::SequencedWorkerPool { 33 disk_cache::FileIOCallback* callback() {
34 return callback_;
35 }
36
37 disk_cache::File* file() {
38 return file_;
39 }
40
41 // Read and Write are the operations that can be performed asynchronously.
42 // The actual parameters for the operation are setup in the constructor of
43 // the object. Both methods should be called from a worker thread, by posting
44 // a task to the WorkerPool (they are RunnableMethods). When finished,
45 // controller->OnIOComplete() is called.
46 void Read();
47 void Write();
48
49 private:
50 virtual ~FileBackgroundIO() {}
51
52 disk_cache::FileIOCallback* callback_;
53
54 disk_cache::File* file_;
55 const void* buf_;
56 size_t buf_len_;
57 size_t offset_;
58
59 DISALLOW_COPY_AND_ASSIGN(FileBackgroundIO);
60 };
61
62
63 // The specialized controller that keeps track of current operations.
64 class FileInFlightIO : public disk_cache::InFlightIO {
23 public: 65 public:
24 FileWorkerPool() : base::SequencedWorkerPool(kMaxThreads, "CachePool") {} 66 FileInFlightIO() {}
67 virtual ~FileInFlightIO() {}
68
69 // These methods start an asynchronous operation. The arguments have the same
70 // semantics of the File asynchronous operations, with the exception that the
71 // operation never finishes synchronously.
72 void PostRead(disk_cache::File* file, void* buf, size_t buf_len,
73 size_t offset, disk_cache::FileIOCallback* callback);
74 void PostWrite(disk_cache::File* file, const void* buf, size_t buf_len,
75 size_t offset, disk_cache::FileIOCallback* callback);
25 76
26 protected: 77 protected:
27 virtual ~FileWorkerPool() {} 78 // Invokes the users' completion callback at the end of the IO operation.
79 // |cancel| is true if the actual task posted to the thread is still
80 // queued (because we are inside WaitForPendingIO), and false if said task is
81 // the one performing the call.
82 virtual void OnOperationComplete(disk_cache::BackgroundIO* operation,
83 bool cancel) OVERRIDE;
84
85 private:
86 DISALLOW_COPY_AND_ASSIGN(FileInFlightIO);
28 }; 87 };
29 88
30 base::LazyInstance<FileWorkerPool>::Leaky s_worker_pool = 89 // ---------------------------------------------------------------------------
31 LAZY_INSTANCE_INITIALIZER; 90
91 // Runs on a worker thread.
92 void FileBackgroundIO::Read() {
93 if (file_->Read(const_cast<void*>(buf_), buf_len_, offset_)) {
94 result_ = static_cast<int>(buf_len_);
95 } else {
96 result_ = net::ERR_CACHE_READ_FAILURE;
97 }
98 NotifyController();
99 }
100
101 // Runs on a worker thread.
102 void FileBackgroundIO::Write() {
103 bool rv = file_->Write(buf_, buf_len_, offset_);
104
105 result_ = rv ? static_cast<int>(buf_len_) : net::ERR_CACHE_WRITE_FAILURE;
106 NotifyController();
107 }
108
109 // ---------------------------------------------------------------------------
110
111 void FileInFlightIO::PostRead(disk_cache::File *file, void* buf, size_t buf_len,
112 size_t offset, disk_cache::FileIOCallback *callback) {
113 scoped_refptr<FileBackgroundIO> operation(
114 new FileBackgroundIO(file, buf, buf_len, offset, callback, this));
115 file->AddRef(); // Balanced on OnOperationComplete()
116
117 base::WorkerPool::PostTask(FROM_HERE,
118 base::Bind(&FileBackgroundIO::Read, operation.get()), true);
119 OnOperationPosted(operation.get());
120 }
121
122 void FileInFlightIO::PostWrite(disk_cache::File* file, const void* buf,
123 size_t buf_len, size_t offset,
124 disk_cache::FileIOCallback* callback) {
125 scoped_refptr<FileBackgroundIO> operation(
126 new FileBackgroundIO(file, buf, buf_len, offset, callback, this));
127 file->AddRef(); // Balanced on OnOperationComplete()
128
129 base::WorkerPool::PostTask(FROM_HERE,
130 base::Bind(&FileBackgroundIO::Write, operation.get()), true);
131 OnOperationPosted(operation.get());
132 }
133
134 // Runs on the IO thread.
135 void FileInFlightIO::OnOperationComplete(disk_cache::BackgroundIO* operation,
136 bool cancel) {
137 FileBackgroundIO* op = static_cast<FileBackgroundIO*>(operation);
138
139 disk_cache::FileIOCallback* callback = op->callback();
140 int bytes = operation->result();
141
142 // Release the references acquired in PostRead / PostWrite.
143 op->file()->Release();
144 callback->OnFileIOComplete(bytes);
145 }
146
147 // A static object tha will broker all async operations.
148 FileInFlightIO* s_file_operations = NULL;
149
150 // Returns the current FileInFlightIO.
151 FileInFlightIO* GetFileInFlightIO() {
152 if (!s_file_operations) {
153 s_file_operations = new FileInFlightIO;
154 }
155 return s_file_operations;
156 }
157
158 // Deletes the current FileInFlightIO.
159 void DeleteFileInFlightIO() {
160 DCHECK(s_file_operations);
161 delete s_file_operations;
162 s_file_operations = NULL;
163 }
32 164
33 } // namespace 165 } // namespace
34 166
35 namespace disk_cache { 167 namespace disk_cache {
36 168
37 File::File(base::PlatformFile file) 169 File::File(base::PlatformFile file)
38 : init_(true), 170 : init_(true),
39 mixed_(true), 171 mixed_(true),
40 platform_file_(file), 172 platform_file_(file),
41 sync_platform_file_(base::kInvalidPlatformFileValue) { 173 sync_platform_file_(base::kInvalidPlatformFileValue) {
(...skipping 44 matching lines...) Expand 10 before | Expand all | Expand 10 after
86 offset > static_cast<size_t>(kint32max)) { 218 offset > static_cast<size_t>(kint32max)) {
87 return false; 219 return false;
88 } 220 }
89 221
90 int ret = base::WritePlatformFile(platform_file_, offset, 222 int ret = base::WritePlatformFile(platform_file_, offset,
91 static_cast<const char*>(buffer), 223 static_cast<const char*>(buffer),
92 buffer_len); 224 buffer_len);
93 return (static_cast<size_t>(ret) == buffer_len); 225 return (static_cast<size_t>(ret) == buffer_len);
94 } 226 }
95 227
228 // We have to increase the ref counter of the file before performing the IO to
229 // prevent the completion to happen with an invalid handle (if the file is
230 // closed while the IO is in flight).
96 bool File::Read(void* buffer, size_t buffer_len, size_t offset, 231 bool File::Read(void* buffer, size_t buffer_len, size_t offset,
97 FileIOCallback* callback, bool* completed) { 232 FileIOCallback* callback, bool* completed) {
98 DCHECK(init_); 233 DCHECK(init_);
99 if (!callback) { 234 if (!callback) {
100 if (completed) 235 if (completed)
101 *completed = true; 236 *completed = true;
102 return Read(buffer, buffer_len, offset); 237 return Read(buffer, buffer_len, offset);
103 } 238 }
104 239
105 if (buffer_len > static_cast<size_t>(kint32max) || 240 if (buffer_len > ULONG_MAX || offset > ULONG_MAX)
106 offset > static_cast<size_t>(kint32max)) {
107 return false; 241 return false;
108 }
109 242
110 base::PostTaskAndReplyWithResult( 243 GetFileInFlightIO()->PostRead(this, buffer, buffer_len, offset, callback);
111 s_worker_pool.Pointer(), FROM_HERE,
112 base::Bind(&File::DoRead, this, buffer, buffer_len, offset),
113 base::Bind(&File::OnOperationComplete, this, callback));
114 244
115 *completed = false; 245 *completed = false;
116 return true; 246 return true;
117 } 247 }
118 248
119 bool File::Write(const void* buffer, size_t buffer_len, size_t offset, 249 bool File::Write(const void* buffer, size_t buffer_len, size_t offset,
120 FileIOCallback* callback, bool* completed) { 250 FileIOCallback* callback, bool* completed) {
121 DCHECK(init_); 251 DCHECK(init_);
122 if (!callback) { 252 if (!callback) {
123 if (completed) 253 if (completed)
124 *completed = true; 254 *completed = true;
125 return Write(buffer, buffer_len, offset); 255 return Write(buffer, buffer_len, offset);
126 } 256 }
127 257
128 if (buffer_len > static_cast<size_t>(kint32max) || 258 return AsyncWrite(buffer, buffer_len, offset, callback, completed);
129 offset > static_cast<size_t>(kint32max)) {
130 return false;
131 }
132
133 base::PostTaskAndReplyWithResult(
134 s_worker_pool.Pointer(), FROM_HERE,
135 base::Bind(&File::DoWrite, this, buffer, buffer_len, offset),
136 base::Bind(&File::OnOperationComplete, this, callback));
137
138 *completed = false;
139 return true;
140 } 259 }
141 260
142 bool File::SetLength(size_t length) { 261 bool File::SetLength(size_t length) {
143 DCHECK(init_); 262 DCHECK(init_);
144 if (length > kuint32max) 263 if (length > kuint32max)
145 return false; 264 return false;
146 265
147 return base::TruncatePlatformFile(platform_file_, length); 266 return base::TruncatePlatformFile(platform_file_, length);
148 } 267 }
149 268
150 size_t File::GetLength() { 269 size_t File::GetLength() {
151 DCHECK(init_); 270 DCHECK(init_);
152 int64 len = base::SeekPlatformFile(platform_file_, 271 int64 len = base::SeekPlatformFile(platform_file_,
153 base::PLATFORM_FILE_FROM_END, 0); 272 base::PLATFORM_FILE_FROM_END, 0);
154 273
155 if (len > static_cast<int64>(kuint32max)) 274 if (len > static_cast<int64>(kuint32max))
156 return kuint32max; 275 return kuint32max;
157 276
158 return static_cast<size_t>(len); 277 return static_cast<size_t>(len);
159 } 278 }
160 279
161 // Static. 280 // Static.
162 void File::WaitForPendingIO(int* num_pending_io) { 281 void File::WaitForPendingIO(int* num_pending_io) {
163 // We are running unit tests so we should wait for all callbacks. Sadly, the 282 // We may be running unit tests so we should allow be able to reset the
164 // worker pool only waits for tasks on the worker pool, not the "Reply" tasks 283 // message loop.
165 // so we have to let the current message loop to run. 284 GetFileInFlightIO()->WaitForPendingIO();
166 s_worker_pool.Get().FlushForTesting(); 285 DeleteFileInFlightIO();
167 base::RunLoop().RunUntilIdle(); 286 }
287
288 // Static.
289 void File::DropPendingIO() {
290 GetFileInFlightIO()->DropPendingIO();
291 DeleteFileInFlightIO();
168 } 292 }
169 293
170 File::~File() { 294 File::~File() {
171 if (IsValid()) 295 if (IsValid())
172 base::ClosePlatformFile(platform_file_); 296 base::ClosePlatformFile(platform_file_);
173 } 297 }
174 298
175 // Runs on a worker thread. 299 bool File::AsyncWrite(const void* buffer, size_t buffer_len, size_t offset,
176 int File::DoRead(void* buffer, size_t buffer_len, size_t offset) { 300 FileIOCallback* callback, bool* completed) {
177 if (Read(const_cast<void*>(buffer), buffer_len, offset)) 301 DCHECK(init_);
178 return static_cast<int>(buffer_len); 302 if (buffer_len > ULONG_MAX || offset > ULONG_MAX)
303 return false;
179 304
180 return net::ERR_CACHE_READ_FAILURE; 305 GetFileInFlightIO()->PostWrite(this, buffer, buffer_len, offset, callback);
181 }
182 306
183 // Runs on a worker thread. 307 if (completed)
184 int File::DoWrite(const void* buffer, size_t buffer_len, size_t offset) { 308 *completed = false;
185 if (Write(const_cast<void*>(buffer), buffer_len, offset)) 309 return true;
186 return static_cast<int>(buffer_len);
187
188 return net::ERR_CACHE_WRITE_FAILURE;
189 }
190
191 // This method actually makes sure that the last reference to the file doesn't
192 // go away on the worker pool.
193 void File::OnOperationComplete(FileIOCallback* callback, int result) {
194 callback->OnFileIOComplete(result);
195 } 310 }
196 311
197 } // namespace disk_cache 312 } // namespace disk_cache
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698