]> git.saurik.com Git - apple/libdispatch.git/blob - src/io.c
projects / apple / libdispatch.git / blob
? search:


b306054be2e36b0f3039b4b8c5832229fa420d0c
[apple/libdispatch.git] / src / io.c
1 /*
2 * Copyright (c) 2009-2011 Apple Inc. All rights reserved.
3 *
4 * @APPLE_APACHE_LICENSE_HEADER_START@
5 *
6 * Licensed under the Apache License, Version 2.0 (the "License");
7 * you may not use this file except in compliance with the License.
8 * You may obtain a copy of the License at
9 *
10 * http://www.apache.org/licenses/LICENSE-2.0
11 *
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an "AS IS" BASIS,
14 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
17 *
18 * @APPLE_APACHE_LICENSE_HEADER_END@
19 */
20
21 #include "internal.h"
22
23 typedef void (^dispatch_fd_entry_init_callback_t)(dispatch_fd_entry_t fd_entry);
24
25 DISPATCH_EXPORT DISPATCH_NOTHROW
26 void _dispatch_iocntl(uint32_t param, uint64_t value);
27
28 static void _dispatch_io_dispose(dispatch_io_t channel);
29 static dispatch_operation_t _dispatch_operation_create(
30 dispatch_op_direction_t direction, dispatch_io_t channel, off_t offset,
31 size_t length, dispatch_data_t data, dispatch_queue_t queue,
32 dispatch_io_handler_t handler);
33 static void _dispatch_operation_dispose(dispatch_operation_t operation);
34 static void _dispatch_operation_enqueue(dispatch_operation_t op,
35 dispatch_op_direction_t direction, dispatch_data_t data);
36 static dispatch_source_t _dispatch_operation_timer(dispatch_queue_t tq,
37 dispatch_operation_t op);
38 static inline void _dispatch_fd_entry_retain(dispatch_fd_entry_t fd_entry);
39 static inline void _dispatch_fd_entry_release(dispatch_fd_entry_t fd_entry);
40 static void _dispatch_fd_entry_init_async(dispatch_fd_t fd,
41 dispatch_fd_entry_init_callback_t completion_callback);
42 static dispatch_fd_entry_t _dispatch_fd_entry_create_with_fd(dispatch_fd_t fd,
43 uintptr_t hash);
44 static dispatch_fd_entry_t _dispatch_fd_entry_create_with_path(
45 dispatch_io_path_data_t path_data, dev_t dev, mode_t mode);
46 static int _dispatch_fd_entry_open(dispatch_fd_entry_t fd_entry,
47 dispatch_io_t channel);
48 static void _dispatch_fd_entry_cleanup_operations(dispatch_fd_entry_t fd_entry,
49 dispatch_io_t channel);
50 static void _dispatch_stream_init(dispatch_fd_entry_t fd_entry,
51 dispatch_queue_t tq);
52 static void _dispatch_stream_dispose(dispatch_fd_entry_t fd_entry,
53 dispatch_op_direction_t direction);
54 static void _dispatch_disk_init(dispatch_fd_entry_t fd_entry, dev_t dev);
55 static void _dispatch_disk_dispose(dispatch_disk_t disk);
56 static void _dispatch_stream_enqueue_operation(dispatch_stream_t stream,
57 dispatch_operation_t operation, dispatch_data_t data);
58 static void _dispatch_disk_enqueue_operation(dispatch_disk_t dsk,
59 dispatch_operation_t operation, dispatch_data_t data);
60 static void _dispatch_stream_cleanup_operations(dispatch_stream_t stream,
61 dispatch_io_t channel);
62 static void _dispatch_disk_cleanup_operations(dispatch_disk_t disk,
63 dispatch_io_t channel);
64 static void _dispatch_stream_source_handler(void *ctx);
65 static void _dispatch_stream_handler(void *ctx);
66 static void _dispatch_disk_handler(void *ctx);
67 static void _dispatch_disk_perform(void *ctxt);
68 static void _dispatch_operation_advise(dispatch_operation_t op,
69 size_t chunk_size);
70 static int _dispatch_operation_perform(dispatch_operation_t op);
71 static void _dispatch_operation_deliver_data(dispatch_operation_t op,
72 dispatch_op_flags_t flags);
73
74 // Macros to wrap syscalls which return -1 on error, and retry on EINTR
75 #define _dispatch_io_syscall_switch_noerr(_err, _syscall, ...) do { \
76 switch (((_err) = (((_syscall) == -1) ? errno : 0))) { \
77 case EINTR: continue; \
78 __VA_ARGS__ \
79 } \
80 } while (0)
81 #define _dispatch_io_syscall_switch(__err, __syscall, ...) do { \
82 _dispatch_io_syscall_switch_noerr(__err, __syscall, \
83 case 0: break; \
84 __VA_ARGS__ \
85 ); \
86 } while (0)
87 #define _dispatch_io_syscall(__syscall) do { int __err; \
88 _dispatch_io_syscall_switch(__err, __syscall); \
89 } while (0)
90
91 enum {
92 DISPATCH_OP_COMPLETE = 1,
93 DISPATCH_OP_DELIVER,
94 DISPATCH_OP_DELIVER_AND_COMPLETE,
95 DISPATCH_OP_COMPLETE_RESUME,
96 DISPATCH_OP_RESUME,
97 DISPATCH_OP_ERR,
98 DISPATCH_OP_FD_ERR,
99 };
100
101 #pragma mark -
102 #pragma mark dispatch_io_vtable
103
104 static const struct dispatch_io_vtable_s _dispatch_io_vtable = {
105 .do_type = DISPATCH_IO_TYPE,
106 .do_kind = "channel",
107 .do_dispose = _dispatch_io_dispose,
108 .do_invoke = NULL,
109 .do_probe = (void *)dummy_function_r0,
110 .do_debug = (void *)dummy_function_r0,
111 };
112
113 static const struct dispatch_operation_vtable_s _dispatch_operation_vtable = {
114 .do_type = DISPATCH_OPERATION_TYPE,
115 .do_kind = "operation",
116 .do_dispose = _dispatch_operation_dispose,
117 .do_invoke = NULL,
118 .do_probe = (void *)dummy_function_r0,
119 .do_debug = (void *)dummy_function_r0,
120 };
121
122 static const struct dispatch_disk_vtable_s _dispatch_disk_vtable = {
123 .do_type = DISPATCH_DISK_TYPE,
124 .do_kind = "disk",
125 .do_dispose = _dispatch_disk_dispose,
126 .do_invoke = NULL,
127 .do_probe = (void *)dummy_function_r0,
128 .do_debug = (void *)dummy_function_r0,
129 };
130
131 #pragma mark -
132 #pragma mark dispatch_io_hashtables
133
134 #if TARGET_OS_EMBEDDED
135 #define DIO_HASH_SIZE 64u // must be a power of two
136 #else
137 #define DIO_HASH_SIZE 256u // must be a power of two
138 #endif
139 #define DIO_HASH(x) ((uintptr_t)((x) & (DIO_HASH_SIZE - 1)))
140
141 // Global hashtable of dev_t -> disk_s mappings
142 DISPATCH_CACHELINE_ALIGN
143 static TAILQ_HEAD(, dispatch_disk_s) _dispatch_io_devs[DIO_HASH_SIZE];
144 // Global hashtable of fd -> fd_entry_s mappings
145 DISPATCH_CACHELINE_ALIGN
146 static TAILQ_HEAD(, dispatch_fd_entry_s) _dispatch_io_fds[DIO_HASH_SIZE];
147
148 static dispatch_once_t _dispatch_io_devs_lockq_pred;
149 static dispatch_queue_t _dispatch_io_devs_lockq;
150 static dispatch_queue_t _dispatch_io_fds_lockq;
151
152 static void
153 _dispatch_io_fds_lockq_init(void *context DISPATCH_UNUSED)
154 {
155 _dispatch_io_fds_lockq = dispatch_queue_create(
156 "com.apple.libdispatch-io.fd_lockq", NULL);
157 unsigned int i;
158 for (i = 0; i < DIO_HASH_SIZE; i++) {
159 TAILQ_INIT(&_dispatch_io_fds[i]);
160 }
161 }
162
163 static void
164 _dispatch_io_devs_lockq_init(void *context DISPATCH_UNUSED)
165 {
166 _dispatch_io_devs_lockq = dispatch_queue_create(
167 "com.apple.libdispatch-io.dev_lockq", NULL);
168 unsigned int i;
169 for (i = 0; i < DIO_HASH_SIZE; i++) {
170 TAILQ_INIT(&_dispatch_io_devs[i]);
171 }
172 }
173
174 #pragma mark -
175 #pragma mark dispatch_io_defaults
176
177 enum {
178 DISPATCH_IOCNTL_CHUNK_PAGES = 1,
179 DISPATCH_IOCNTL_LOW_WATER_CHUNKS,
180 DISPATCH_IOCNTL_INITIAL_DELIVERY,
181 DISPATCH_IOCNTL_MAX_PENDING_IO_REQS,
182 };
183
184 static struct dispatch_io_defaults_s {
185 size_t chunk_pages, low_water_chunks, max_pending_io_reqs;
186 bool initial_delivery;
187 } dispatch_io_defaults = {
188 .chunk_pages = DIO_MAX_CHUNK_PAGES,
189 .low_water_chunks = DIO_DEFAULT_LOW_WATER_CHUNKS,
190 .max_pending_io_reqs = DIO_MAX_PENDING_IO_REQS,
191 };
192
193 #define _dispatch_iocntl_set_default(p, v) do { \
194 dispatch_io_defaults.p = (typeof(dispatch_io_defaults.p))(v); \
195 } while (0)
196
197 void
198 _dispatch_iocntl(uint32_t param, uint64_t value)
199 {
200 switch (param) {
201 case DISPATCH_IOCNTL_CHUNK_PAGES:
202 _dispatch_iocntl_set_default(chunk_pages, value);
203 break;
204 case DISPATCH_IOCNTL_LOW_WATER_CHUNKS:
205 _dispatch_iocntl_set_default(low_water_chunks, value);
206 break;
207 case DISPATCH_IOCNTL_INITIAL_DELIVERY:
208 _dispatch_iocntl_set_default(initial_delivery, value);
209 case DISPATCH_IOCNTL_MAX_PENDING_IO_REQS:
210 _dispatch_iocntl_set_default(max_pending_io_reqs, value);
211 break;
212 }
213 }
214
215 #pragma mark -
216 #pragma mark dispatch_io_t
217
218 static dispatch_io_t
219 _dispatch_io_create(dispatch_io_type_t type)
220 {
221 dispatch_io_t channel = calloc(1ul, sizeof(struct dispatch_io_s));
222 channel->do_vtable = &_dispatch_io_vtable;
223 channel->do_next = DISPATCH_OBJECT_LISTLESS;
224 channel->do_ref_cnt = 1;
225 channel->do_xref_cnt = 1;
226 channel->do_targetq = _dispatch_get_root_queue(0, true);
227 channel->params.type = type;
228 channel->params.high = SIZE_MAX;
229 channel->params.low = dispatch_io_defaults.low_water_chunks *
230 dispatch_io_defaults.chunk_pages * PAGE_SIZE;
231 channel->queue = dispatch_queue_create("com.apple.libdispatch-io.channelq",
232 NULL);
233 return channel;
234 }
235
236 static void
237 _dispatch_io_init(dispatch_io_t channel, dispatch_fd_entry_t fd_entry,
238 dispatch_queue_t queue, int err, void (^cleanup_handler)(int))
239 {
240 // Enqueue the cleanup handler on the suspended close queue
241 if (cleanup_handler) {
242 _dispatch_retain(queue);
243 dispatch_async(!err ? fd_entry->close_queue : channel->queue, ^{
244 dispatch_async(queue, ^{
245 _dispatch_io_debug("cleanup handler invoke", -1);
246 cleanup_handler(err);
247 });
248 _dispatch_release(queue);
249 });
250 }
251 if (fd_entry) {
252 channel->fd_entry = fd_entry;
253 dispatch_retain(fd_entry->barrier_queue);
254 dispatch_retain(fd_entry->barrier_group);
255 channel->barrier_queue = fd_entry->barrier_queue;
256 channel->barrier_group = fd_entry->barrier_group;
257 } else {
258 // Still need to create a barrier queue, since all operations go
259 // through it
260 channel->barrier_queue = dispatch_queue_create(
261 "com.apple.libdispatch-io.barrierq", NULL);
262 channel->barrier_group = dispatch_group_create();
263 }
264 }
265
266 static void
267 _dispatch_io_dispose(dispatch_io_t channel)
268 {
269 if (channel->fd_entry) {
270 if (channel->fd_entry->path_data) {
271 // This modification is safe since path_data->channel is checked
272 // only on close_queue (which is still suspended at this point)
273 channel->fd_entry->path_data->channel = NULL;
274 }
275 // Cleanup handlers will only run when all channels related to this
276 // fd are complete
277 _dispatch_fd_entry_release(channel->fd_entry);
278 }
279 if (channel->queue) {
280 dispatch_release(channel->queue);
281 }
282 if (channel->barrier_queue) {
283 dispatch_release(channel->barrier_queue);
284 }
285 if (channel->barrier_group) {
286 dispatch_release(channel->barrier_group);
287 }
288 _dispatch_dispose(channel);
289 }
290
291 static int
292 _dispatch_io_validate_type(dispatch_io_t channel, mode_t mode)
293 {
294 int err = 0;
295 if (S_ISDIR(mode)) {
296 err = EISDIR;
297 } else if (channel->params.type == DISPATCH_IO_RANDOM &&
298 (S_ISFIFO(mode) || S_ISSOCK(mode))) {
299 err = ESPIPE;
300 }
301 return err;
302 }
303
304 static int
305 _dispatch_io_get_error(dispatch_operation_t op, dispatch_io_t channel,
306 bool ignore_closed)
307 {
308 // On _any_ queue
309 int err;
310 if (op) {
311 channel = op->channel;
312 }
313 if (channel->atomic_flags & (DIO_CLOSED|DIO_STOPPED)) {
314 if (!ignore_closed || channel->atomic_flags & DIO_STOPPED) {
315 err = ECANCELED;
316 } else {
317 err = 0;
318 }
319 } else {
320 err = op ? op->fd_entry->err : channel->err;
321 }
322 return err;
323 }
324
325 #pragma mark -
326 #pragma mark dispatch_io_channels
327
328 dispatch_io_t
329 dispatch_io_create(dispatch_io_type_t type, dispatch_fd_t fd,
330 dispatch_queue_t queue, void (^cleanup_handler)(int))
331 {
332 if (type != DISPATCH_IO_STREAM && type != DISPATCH_IO_RANDOM) {
333 return NULL;
334 }
335 _dispatch_io_debug("io create", fd);
336 dispatch_io_t channel = _dispatch_io_create(type);
337 channel->fd = fd;
338 channel->fd_actual = fd;
339 dispatch_suspend(channel->queue);
340 _dispatch_retain(queue);
341 _dispatch_retain(channel);
342 _dispatch_fd_entry_init_async(fd, ^(dispatch_fd_entry_t fd_entry) {
343 // On barrier queue
344 int err = fd_entry->err;
345 if (!err) {
346 err = _dispatch_io_validate_type(channel, fd_entry->stat.mode);
347 }
348 if (!err && type == DISPATCH_IO_RANDOM) {
349 off_t f_ptr;
350 _dispatch_io_syscall_switch_noerr(err,
351 f_ptr = lseek(fd_entry->fd, 0, SEEK_CUR),
352 case 0: channel->f_ptr = f_ptr; break;
353 default: (void)dispatch_assume_zero(err); break;
354 );
355 }
356 channel->err = err;
357 _dispatch_fd_entry_retain(fd_entry);
358 _dispatch_io_init(channel, fd_entry, queue, err, cleanup_handler);
359 dispatch_resume(channel->queue);
360 _dispatch_release(channel);
361 _dispatch_release(queue);
362 });
363 return channel;
364 }
365
366 dispatch_io_t
367 dispatch_io_create_with_path(dispatch_io_type_t type, const char *path,
368 int oflag, mode_t mode, dispatch_queue_t queue,
369 void (^cleanup_handler)(int error))
370 {
371 if ((type != DISPATCH_IO_STREAM && type != DISPATCH_IO_RANDOM) ||
372 !(path && *path == '/')) {
373 return NULL;
374 }
375 size_t pathlen = strlen(path);
376 dispatch_io_path_data_t path_data = malloc(sizeof(*path_data) + pathlen+1);
377 if (!path_data) {
378 return NULL;
379 }
380 _dispatch_io_debug("io create with path %s", -1, path);
381 dispatch_io_t channel = _dispatch_io_create(type);
382 channel->fd = -1;
383 channel->fd_actual = -1;
384 path_data->channel = channel;
385 path_data->oflag = oflag;
386 path_data->mode = mode;
387 path_data->pathlen = pathlen;
388 memcpy(path_data->path, path, pathlen + 1);
389 _dispatch_retain(queue);
390 _dispatch_retain(channel);
391 dispatch_async(channel->queue, ^{
392 int err = 0;
393 struct stat st;
394 _dispatch_io_syscall_switch_noerr(err,
395 (path_data->oflag & O_NOFOLLOW) == O_NOFOLLOW ||
396 (path_data->oflag & O_SYMLINK) == O_SYMLINK ?
397 lstat(path_data->path, &st) : stat(path_data->path, &st),
398 case 0:
399 err = _dispatch_io_validate_type(channel, st.st_mode);
400 break;
401 default:
402 if ((path_data->oflag & O_CREAT) &&
403 (*(path_data->path + path_data->pathlen - 1) != '/')) {
404 // Check parent directory
405 char *c = strrchr(path_data->path, '/');
406 dispatch_assert(c);
407 *c = 0;
408 int perr;
409 _dispatch_io_syscall_switch_noerr(perr,
410 stat(path_data->path, &st),
411 case 0:
412 // Since the parent directory exists, open() will
413 // create a regular file after the fd_entry has
414 // been filled in
415 st.st_mode = S_IFREG;
416 err = 0;
417 break;
418 );
419 *c = '/';
420 }
421 break;
422 );
423 channel->err = err;
424 if (err) {
425 free(path_data);
426 _dispatch_io_init(channel, NULL, queue, err, cleanup_handler);
427 _dispatch_release(channel);
428 _dispatch_release(queue);
429 return;
430 }
431 dispatch_suspend(channel->queue);
432 dispatch_once_f(&_dispatch_io_devs_lockq_pred, NULL,
433 _dispatch_io_devs_lockq_init);
434 dispatch_async(_dispatch_io_devs_lockq, ^{
435 dispatch_fd_entry_t fd_entry = _dispatch_fd_entry_create_with_path(
436 path_data, st.st_dev, st.st_mode);
437 _dispatch_io_init(channel, fd_entry, queue, 0, cleanup_handler);
438 dispatch_resume(channel->queue);
439 _dispatch_release(channel);
440 _dispatch_release(queue);
441 });
442 });
443 return channel;
444 }
445
446 dispatch_io_t
447 dispatch_io_create_with_io(dispatch_io_type_t type, dispatch_io_t in_channel,
448 dispatch_queue_t queue, void (^cleanup_handler)(int error))
449 {
450 if (type != DISPATCH_IO_STREAM && type != DISPATCH_IO_RANDOM) {
451 return NULL;
452 }
453 _dispatch_io_debug("io create with io %p", -1, in_channel);
454 dispatch_io_t channel = _dispatch_io_create(type);
455 dispatch_suspend(channel->queue);
456 _dispatch_retain(queue);
457 _dispatch_retain(channel);
458 _dispatch_retain(in_channel);
459 dispatch_async(in_channel->queue, ^{
460 int err0 = _dispatch_io_get_error(NULL, in_channel, false);
461 if (err0) {
462 channel->err = err0;
463 _dispatch_io_init(channel, NULL, queue, err0, cleanup_handler);
464 dispatch_resume(channel->queue);
465 _dispatch_release(channel);
466 _dispatch_release(in_channel);
467 _dispatch_release(queue);
468 return;
469 }
470 dispatch_async(in_channel->barrier_queue, ^{
471 int err = _dispatch_io_get_error(NULL, in_channel, false);
472 // If there is no error, the fd_entry for the in_channel is valid.
473 // Since we are running on in_channel's queue, the fd_entry has been
474 // fully resolved and will stay valid for the duration of this block
475 if (!err) {
476 err = in_channel->err;
477 if (!err) {
478 err = in_channel->fd_entry->err;
479 }
480 }
481 if (!err) {
482 err = _dispatch_io_validate_type(channel,
483 in_channel->fd_entry->stat.mode);
484 }
485 if (!err && type == DISPATCH_IO_RANDOM && in_channel->fd != -1) {
486 off_t f_ptr;
487 _dispatch_io_syscall_switch_noerr(err,
488 f_ptr = lseek(in_channel->fd_entry->fd, 0, SEEK_CUR),
489 case 0: channel->f_ptr = f_ptr; break;
490 default: (void)dispatch_assume_zero(err); break;
491 );
492 }
493 channel->err = err;
494 if (err) {
495 _dispatch_io_init(channel, NULL, queue, err, cleanup_handler);
496 dispatch_resume(channel->queue);
497 _dispatch_release(channel);
498 _dispatch_release(in_channel);
499 _dispatch_release(queue);
500 return;
501 }
502 if (in_channel->fd == -1) {
503 // in_channel was created from path
504 channel->fd = -1;
505 channel->fd_actual = -1;
506 mode_t mode = in_channel->fd_entry->stat.mode;
507 dev_t dev = in_channel->fd_entry->stat.dev;
508 size_t path_data_len = sizeof(struct dispatch_io_path_data_s) +
509 in_channel->fd_entry->path_data->pathlen + 1;
510 dispatch_io_path_data_t path_data = malloc(path_data_len);
511 memcpy(path_data, in_channel->fd_entry->path_data,
512 path_data_len);
513 path_data->channel = channel;
514 // lockq_io_devs is known to already exist
515 dispatch_async(_dispatch_io_devs_lockq, ^{
516 dispatch_fd_entry_t fd_entry;
517 fd_entry = _dispatch_fd_entry_create_with_path(path_data,
518 dev, mode);
519 _dispatch_io_init(channel, fd_entry, queue, 0,
520 cleanup_handler);
521 dispatch_resume(channel->queue);
522 _dispatch_release(channel);
523 _dispatch_release(queue);
524 });
525 } else {
526 dispatch_fd_entry_t fd_entry = in_channel->fd_entry;
527 channel->fd = in_channel->fd;
528 channel->fd_actual = in_channel->fd_actual;
529 _dispatch_fd_entry_retain(fd_entry);
530 _dispatch_io_init(channel, fd_entry, queue, 0, cleanup_handler);
531 dispatch_resume(channel->queue);
532 _dispatch_release(channel);
533 _dispatch_release(queue);
534 }
535 _dispatch_release(in_channel);
536 });
537 });
538 return channel;
539 }
540
541 #pragma mark -
542 #pragma mark dispatch_io_accessors
543
544 void
545 dispatch_io_set_high_water(dispatch_io_t channel, size_t high_water)
546 {
547 _dispatch_retain(channel);
548 dispatch_async(channel->queue, ^{
549 _dispatch_io_debug("io set high water", channel->fd);
550 if (channel->params.low > high_water) {
551 channel->params.low = high_water;
552 }
553 channel->params.high = high_water ? high_water : 1;
554 _dispatch_release(channel);
555 });
556 }
557
558 void
559 dispatch_io_set_low_water(dispatch_io_t channel, size_t low_water)
560 {
561 _dispatch_retain(channel);
562 dispatch_async(channel->queue, ^{
563 _dispatch_io_debug("io set low water", channel->fd);
564 if (channel->params.high < low_water) {
565 channel->params.high = low_water ? low_water : 1;
566 }
567 channel->params.low = low_water;
568 _dispatch_release(channel);
569 });
570 }
571
572 void
573 dispatch_io_set_interval(dispatch_io_t channel, uint64_t interval,
574 unsigned long flags)
575 {
576 _dispatch_retain(channel);
577 dispatch_async(channel->queue, ^{
578 _dispatch_io_debug("io set interval", channel->fd);
579 channel->params.interval = interval;
580 channel->params.interval_flags = flags;
581 _dispatch_release(channel);
582 });
583 }
584
585 void
586 _dispatch_io_set_target_queue(dispatch_io_t channel, dispatch_queue_t dq)
587 {
588 _dispatch_retain(dq);
589 _dispatch_retain(channel);
590 dispatch_async(channel->queue, ^{
591 dispatch_queue_t prev_dq = channel->do_targetq;
592 channel->do_targetq = dq;
593 _dispatch_release(prev_dq);
594 _dispatch_release(channel);
595 });
596 }
597
598 dispatch_fd_t
599 dispatch_io_get_descriptor(dispatch_io_t channel)
600 {
601 if (channel->atomic_flags & (DIO_CLOSED|DIO_STOPPED)) {
602 return -1;
603 }
604 dispatch_fd_t fd = channel->fd_actual;
605 if (fd == -1 &&
606 _dispatch_thread_getspecific(dispatch_io_key) == channel) {
607 dispatch_fd_entry_t fd_entry = channel->fd_entry;
608 (void)_dispatch_fd_entry_open(fd_entry, channel);
609 }
610 return channel->fd_actual;
611 }
612
613 #pragma mark -
614 #pragma mark dispatch_io_operations
615
616 static void
617 _dispatch_io_stop(dispatch_io_t channel)
618 {
619 _dispatch_io_debug("io stop", channel->fd);
620 (void)dispatch_atomic_or2o(channel, atomic_flags, DIO_STOPPED);
621 _dispatch_retain(channel);
622 dispatch_async(channel->queue, ^{
623 dispatch_async(channel->barrier_queue, ^{
624 dispatch_fd_entry_t fd_entry = channel->fd_entry;
625 if (fd_entry) {
626 _dispatch_io_debug("io stop cleanup", channel->fd);
627 _dispatch_fd_entry_cleanup_operations(fd_entry, channel);
628 channel->fd_entry = NULL;
629 _dispatch_fd_entry_release(fd_entry);
630 } else if (channel->fd != -1) {
631 // Stop after close, need to check if fd_entry still exists
632 _dispatch_retain(channel);
633 dispatch_async(_dispatch_io_fds_lockq, ^{
634 _dispatch_io_debug("io stop after close cleanup",
635 channel->fd);
636 dispatch_fd_entry_t fdi;
637 uintptr_t hash = DIO_HASH(channel->fd);
638 TAILQ_FOREACH(fdi, &_dispatch_io_fds[hash], fd_list) {
639 if (fdi->fd == channel->fd) {
640 _dispatch_fd_entry_cleanup_operations(fdi, channel);
641 break;
642 }
643 }
644 _dispatch_release(channel);
645 });
646 }
647 _dispatch_release(channel);
648 });
649 });
650 }
651
652 void
653 dispatch_io_close(dispatch_io_t channel, unsigned long flags)
654 {
655 if (flags & DISPATCH_IO_STOP) {
656 // Don't stop an already stopped channel
657 if (channel->atomic_flags & DIO_STOPPED) {
658 return;
659 }
660 return _dispatch_io_stop(channel);
661 }
662 // Don't close an already closed or stopped channel
663 if (channel->atomic_flags & (DIO_CLOSED|DIO_STOPPED)) {
664 return;
665 }
666 _dispatch_retain(channel);
667 dispatch_async(channel->queue, ^{
668 dispatch_async(channel->barrier_queue, ^{
669 _dispatch_io_debug("io close", channel->fd);
670 (void)dispatch_atomic_or2o(channel, atomic_flags, DIO_CLOSED);
671 dispatch_fd_entry_t fd_entry = channel->fd_entry;
672 if (fd_entry) {
673 if (!fd_entry->path_data) {
674 channel->fd_entry = NULL;
675 }
676 _dispatch_fd_entry_release(fd_entry);
677 }
678 _dispatch_release(channel);
679 });
680 });
681 }
682
683 void
684 dispatch_io_barrier(dispatch_io_t channel, dispatch_block_t barrier)
685 {
686 _dispatch_retain(channel);
687 dispatch_async(channel->queue, ^{
688 dispatch_queue_t io_q = channel->do_targetq;
689 dispatch_queue_t barrier_queue = channel->barrier_queue;
690 dispatch_group_t barrier_group = channel->barrier_group;
691 dispatch_async(barrier_queue, ^{
692 dispatch_suspend(barrier_queue);
693 dispatch_group_notify(barrier_group, io_q, ^{
694 _dispatch_thread_setspecific(dispatch_io_key, channel);
695 barrier();
696 _dispatch_thread_setspecific(dispatch_io_key, NULL);
697 dispatch_resume(barrier_queue);
698 _dispatch_release(channel);
699 });
700 });
701 });
702 }
703
704 void
705 dispatch_io_read(dispatch_io_t channel, off_t offset, size_t length,
706 dispatch_queue_t queue, dispatch_io_handler_t handler)
707 {
708 _dispatch_retain(channel);
709 _dispatch_retain(queue);
710 dispatch_async(channel->queue, ^{
711 dispatch_operation_t op;
712 op = _dispatch_operation_create(DOP_DIR_READ, channel, offset,
713 length, dispatch_data_empty, queue, handler);
714 if (op) {
715 dispatch_queue_t barrier_q = channel->barrier_queue;
716 dispatch_async(barrier_q, ^{
717 _dispatch_operation_enqueue(op, DOP_DIR_READ,
718 dispatch_data_empty);
719 });
720 }
721 _dispatch_release(channel);
722 _dispatch_release(queue);
723 });
724 }
725
726 void
727 dispatch_io_write(dispatch_io_t channel, off_t offset, dispatch_data_t data,
728 dispatch_queue_t queue, dispatch_io_handler_t handler)
729 {
730 _dispatch_io_data_retain(data);
731 _dispatch_retain(channel);
732 _dispatch_retain(queue);
733 dispatch_async(channel->queue, ^{
734 dispatch_operation_t op;
735 op = _dispatch_operation_create(DOP_DIR_WRITE, channel, offset,
736 dispatch_data_get_size(data), data, queue, handler);
737 if (op) {
738 dispatch_queue_t barrier_q = channel->barrier_queue;
739 dispatch_async(barrier_q, ^{
740 _dispatch_operation_enqueue(op, DOP_DIR_WRITE, data);
741 _dispatch_io_data_release(data);
742 });
743 } else {
744 _dispatch_io_data_release(data);
745 }
746 _dispatch_release(channel);
747 _dispatch_release(queue);
748 });
749 }
750
751 void
752 dispatch_read(dispatch_fd_t fd, size_t length, dispatch_queue_t queue,
753 void (^handler)(dispatch_data_t, int))
754 {
755 _dispatch_retain(queue);
756 _dispatch_fd_entry_init_async(fd, ^(dispatch_fd_entry_t fd_entry) {
757 // On barrier queue
758 if (fd_entry->err) {
759 int err = fd_entry->err;
760 dispatch_async(queue, ^{
761 _dispatch_io_debug("convenience handler invoke", fd);
762 handler(dispatch_data_empty, err);
763 });
764 _dispatch_release(queue);
765 return;
766 }
767 // Safe to access fd_entry on barrier queue
768 dispatch_io_t channel = fd_entry->convenience_channel;
769 if (!channel) {
770 channel = _dispatch_io_create(DISPATCH_IO_STREAM);
771 channel->fd = fd;
772 channel->fd_actual = fd;
773 channel->fd_entry = fd_entry;
774 dispatch_retain(fd_entry->barrier_queue);
775 dispatch_retain(fd_entry->barrier_group);
776 channel->barrier_queue = fd_entry->barrier_queue;
777 channel->barrier_group = fd_entry->barrier_group;
778 fd_entry->convenience_channel = channel;
779 }
780 __block dispatch_data_t deliver_data = dispatch_data_empty;
781 __block int err = 0;
782 dispatch_async(fd_entry->close_queue, ^{
783 dispatch_async(queue, ^{
784 _dispatch_io_debug("convenience handler invoke", fd);
785 handler(deliver_data, err);
786 _dispatch_io_data_release(deliver_data);
787 });
788 _dispatch_release(queue);
789 });
790 dispatch_operation_t op =
791 _dispatch_operation_create(DOP_DIR_READ, channel, 0,
792 length, dispatch_data_empty,
793 _dispatch_get_root_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT,
794 false), ^(bool done, dispatch_data_t data, int error) {
795 if (data) {
796 data = dispatch_data_create_concat(deliver_data, data);
797 _dispatch_io_data_release(deliver_data);
798 deliver_data = data;
799 }
800 if (done) {
801 err = error;
802 }
803 });
804 if (op) {
805 _dispatch_operation_enqueue(op, DOP_DIR_READ, dispatch_data_empty);
806 }
807 });
808 }
809
810 void
811 dispatch_write(dispatch_fd_t fd, dispatch_data_t data, dispatch_queue_t queue,
812 void (^handler)(dispatch_data_t, int))
813 {
814 _dispatch_io_data_retain(data);
815 _dispatch_retain(queue);
816 _dispatch_fd_entry_init_async(fd, ^(dispatch_fd_entry_t fd_entry) {
817 // On barrier queue
818 if (fd_entry->err) {
819 int err = fd_entry->err;
820 dispatch_async(queue, ^{
821 _dispatch_io_debug("convenience handler invoke", fd);
822 handler(NULL, err);
823 });
824 _dispatch_release(queue);
825 return;
826 }
827 // Safe to access fd_entry on barrier queue
828 dispatch_io_t channel = fd_entry->convenience_channel;
829 if (!channel) {
830 channel = _dispatch_io_create(DISPATCH_IO_STREAM);
831 channel->fd = fd;
832 channel->fd_actual = fd;
833 channel->fd_entry = fd_entry;
834 dispatch_retain(fd_entry->barrier_queue);
835 dispatch_retain(fd_entry->barrier_group);
836 channel->barrier_queue = fd_entry->barrier_queue;
837 channel->barrier_group = fd_entry->barrier_group;
838 fd_entry->convenience_channel = channel;
839 }
840 __block dispatch_data_t deliver_data = NULL;
841 __block int err = 0;
842 dispatch_async(fd_entry->close_queue, ^{
843 dispatch_async(queue, ^{
844 _dispatch_io_debug("convenience handler invoke", fd);
845 handler(deliver_data, err);
846 if (deliver_data) {
847 _dispatch_io_data_release(deliver_data);
848 }
849 });
850 _dispatch_release(queue);
851 });
852 dispatch_operation_t op =
853 _dispatch_operation_create(DOP_DIR_WRITE, channel, 0,
854 dispatch_data_get_size(data), data,
855 _dispatch_get_root_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT,
856 false), ^(bool done, dispatch_data_t d, int error) {
857 if (done) {
858 if (d) {
859 _dispatch_io_data_retain(d);
860 deliver_data = d;
861 }
862 err = error;
863 }
864 });
865 if (op) {
866 _dispatch_operation_enqueue(op, DOP_DIR_WRITE, data);
867 }
868 _dispatch_io_data_release(data);
869 });
870 }
871
872 #pragma mark -
873 #pragma mark dispatch_operation_t
874
875 static dispatch_operation_t
876 _dispatch_operation_create(dispatch_op_direction_t direction,
877 dispatch_io_t channel, off_t offset, size_t length,
878 dispatch_data_t data, dispatch_queue_t queue,
879 dispatch_io_handler_t handler)
880 {
881 // On channel queue
882 dispatch_assert(direction < DOP_DIR_MAX);
883 _dispatch_io_debug("operation create", channel->fd);
884 #if DISPATCH_IO_DEBUG
885 int fd = channel->fd;
886 #endif
887 // Safe to call _dispatch_io_get_error() with channel->fd_entry since
888 // that can only be NULL if atomic_flags are set rdar://problem/8362514
889 int err = _dispatch_io_get_error(NULL, channel, false);
890 if (err || !length) {
891 _dispatch_io_data_retain(data);
892 _dispatch_retain(queue);
893 dispatch_async(channel->barrier_queue, ^{
894 dispatch_async(queue, ^{
895 dispatch_data_t d = data;
896 if (direction == DOP_DIR_READ && err) {
897 d = NULL;
898 } else if (direction == DOP_DIR_WRITE && !err) {
899 d = NULL;
900 }
901 _dispatch_io_debug("IO handler invoke", fd);
902 handler(true, d, err);
903 _dispatch_io_data_release(data);
904 });
905 _dispatch_release(queue);
906 });
907 return NULL;
908 }
909 dispatch_operation_t op;
910 op = calloc(1ul, sizeof(struct dispatch_operation_s));
911 op->do_vtable = &_dispatch_operation_vtable;
912 op->do_next = DISPATCH_OBJECT_LISTLESS;
913 op->do_ref_cnt = 1;
914 op->do_xref_cnt = 0; // operation object is not exposed externally
915 op->op_q = dispatch_queue_create("com.apple.libdispatch-io.opq", NULL);
916 op->op_q->do_targetq = queue;
917 _dispatch_retain(queue);
918 op->active = false;
919 op->direction = direction;
920 op->offset = offset + channel->f_ptr;
921 op->length = length;
922 op->handler = Block_copy(handler);
923 _dispatch_retain(channel);
924 op->channel = channel;
925 op->params = channel->params;
926 // Take a snapshot of the priority of the channel queue. The actual I/O
927 // for this operation will be performed at this priority
928 dispatch_queue_t targetq = op->channel->do_targetq;
929 while (fastpath(targetq->do_targetq)) {
930 targetq = targetq->do_targetq;
931 }
932 op->do_targetq = targetq;
933 return op;
934 }
935
936 static void
937 _dispatch_operation_dispose(dispatch_operation_t op)
938 {
939 // Deliver the data if there's any
940 if (op->fd_entry) {
941 _dispatch_operation_deliver_data(op, DOP_DONE);
942 dispatch_group_leave(op->fd_entry->barrier_group);
943 _dispatch_fd_entry_release(op->fd_entry);
944 }
945 if (op->channel) {
946 _dispatch_release(op->channel);
947 }
948 if (op->timer) {
949 dispatch_release(op->timer);
950 }
951 // For write operations, op->buf is owned by op->buf_data
952 if (op->buf && op->direction == DOP_DIR_READ) {
953 free(op->buf);
954 }
955 if (op->buf_data) {
956 _dispatch_io_data_release(op->buf_data);
957 }
958 if (op->data) {
959 _dispatch_io_data_release(op->data);
960 }
961 if (op->op_q) {
962 dispatch_release(op->op_q);
963 }
964 Block_release(op->handler);
965 _dispatch_dispose(op);
966 }
967
968 static void
969 _dispatch_operation_enqueue(dispatch_operation_t op,
970 dispatch_op_direction_t direction, dispatch_data_t data)
971 {
972 // Called from the barrier queue
973 _dispatch_io_data_retain(data);
974 // If channel is closed or stopped, then call the handler immediately
975 int err = _dispatch_io_get_error(NULL, op->channel, false);
976 if (err) {
977 dispatch_io_handler_t handler = op->handler;
978 dispatch_async(op->op_q, ^{
979 dispatch_data_t d = data;
980 if (direction == DOP_DIR_READ && err) {
981 d = NULL;
982 } else if (direction == DOP_DIR_WRITE && !err) {
983 d = NULL;
984 }
985 handler(true, d, err);
986 _dispatch_io_data_release(data);
987 });
988 _dispatch_release(op);
989 return;
990 }
991 // Finish operation init
992 op->fd_entry = op->channel->fd_entry;
993 _dispatch_fd_entry_retain(op->fd_entry);
994 dispatch_group_enter(op->fd_entry->barrier_group);
995 dispatch_disk_t disk = op->fd_entry->disk;
996 if (!disk) {
997 dispatch_stream_t stream = op->fd_entry->streams[direction];
998 dispatch_async(stream->dq, ^{
999 _dispatch_stream_enqueue_operation(stream, op, data);
1000 _dispatch_io_data_release(data);
1001 });
1002 } else {
1003 dispatch_async(disk->pick_queue, ^{
1004 _dispatch_disk_enqueue_operation(disk, op, data);
1005 _dispatch_io_data_release(data);
1006 });
1007 }
1008 }
1009
1010 static bool
1011 _dispatch_operation_should_enqueue(dispatch_operation_t op,
1012 dispatch_queue_t tq, dispatch_data_t data)
1013 {
1014 // On stream queue or disk queue
1015 _dispatch_io_debug("enqueue operation", op->fd_entry->fd);
1016 _dispatch_io_data_retain(data);
1017 op->data = data;
1018 int err = _dispatch_io_get_error(op, NULL, true);
1019 if (err) {
1020 op->err = err;
1021 // Final release
1022 _dispatch_release(op);
1023 return false;
1024 }
1025 if (op->params.interval) {
1026 dispatch_resume(_dispatch_operation_timer(tq, op));
1027 }
1028 return true;
1029 }
1030
1031 static dispatch_source_t
1032 _dispatch_operation_timer(dispatch_queue_t tq, dispatch_operation_t op)
1033 {
1034 // On stream queue or pick queue
1035 if (op->timer) {
1036 return op->timer;
1037 }
1038 dispatch_source_t timer = dispatch_source_create(
1039 DISPATCH_SOURCE_TYPE_TIMER, 0, 0, tq);
1040 dispatch_source_set_timer(timer, dispatch_time(DISPATCH_TIME_NOW,
1041 op->params.interval), op->params.interval, 0);
1042 dispatch_source_set_event_handler(timer, ^{
1043 // On stream queue or pick queue
1044 if (dispatch_source_testcancel(timer)) {
1045 // Do nothing. The operation has already completed
1046 return;
1047 }
1048 dispatch_op_flags_t flags = DOP_DEFAULT;
1049 if (op->params.interval_flags & DISPATCH_IO_STRICT_INTERVAL) {
1050 // Deliver even if there is less data than the low-water mark
1051 flags |= DOP_DELIVER;
1052 }
1053 // If the operation is active, dont deliver data
1054 if ((op->active) && (flags & DOP_DELIVER)) {
1055 op->flags = flags;
1056 } else {
1057 _dispatch_operation_deliver_data(op, flags);
1058 }
1059 });
1060 op->timer = timer;
1061 return op->timer;
1062 }
1063
1064 #pragma mark -
1065 #pragma mark dispatch_fd_entry_t
1066
1067 static inline void
1068 _dispatch_fd_entry_retain(dispatch_fd_entry_t fd_entry) {
1069 dispatch_suspend(fd_entry->close_queue);
1070 }
1071
1072 static inline void
1073 _dispatch_fd_entry_release(dispatch_fd_entry_t fd_entry) {
1074 dispatch_resume(fd_entry->close_queue);
1075 }
1076
1077 static void
1078 _dispatch_fd_entry_init_async(dispatch_fd_t fd,
1079 dispatch_fd_entry_init_callback_t completion_callback)
1080 {
1081 static dispatch_once_t _dispatch_io_fds_lockq_pred;
1082 dispatch_once_f(&_dispatch_io_fds_lockq_pred, NULL,
1083 _dispatch_io_fds_lockq_init);
1084 dispatch_async(_dispatch_io_fds_lockq, ^{
1085 _dispatch_io_debug("fd entry init", fd);
1086 dispatch_fd_entry_t fd_entry = NULL;
1087 // Check to see if there is an existing entry for the given fd
1088 uintptr_t hash = DIO_HASH(fd);
1089 TAILQ_FOREACH(fd_entry, &_dispatch_io_fds[hash], fd_list) {
1090 if (fd_entry->fd == fd) {
1091 // Retain the fd_entry to ensure it cannot go away until the
1092 // stat() has completed
1093 _dispatch_fd_entry_retain(fd_entry);
1094 break;
1095 }
1096 }
1097 if (!fd_entry) {
1098 // If we did not find an existing entry, create one
1099 fd_entry = _dispatch_fd_entry_create_with_fd(fd, hash);
1100 }
1101 dispatch_async(fd_entry->barrier_queue, ^{
1102 _dispatch_io_debug("fd entry init completion", fd);
1103 completion_callback(fd_entry);
1104 // stat() is complete, release reference to fd_entry
1105 _dispatch_fd_entry_release(fd_entry);
1106 });
1107 });
1108 }
1109
1110 static dispatch_fd_entry_t
1111 _dispatch_fd_entry_create(dispatch_queue_t q)
1112 {
1113 dispatch_fd_entry_t fd_entry;
1114 fd_entry = calloc(1ul, sizeof(struct dispatch_fd_entry_s));
1115 fd_entry->close_queue = dispatch_queue_create(
1116 "com.apple.libdispatch-io.closeq", NULL);
1117 // Use target queue to ensure that no concurrent lookups are going on when
1118 // the close queue is running
1119 fd_entry->close_queue->do_targetq = q;
1120 _dispatch_retain(q);
1121 // Suspend the cleanup queue until closing
1122 _dispatch_fd_entry_retain(fd_entry);
1123 return fd_entry;
1124 }
1125
1126 static dispatch_fd_entry_t
1127 _dispatch_fd_entry_create_with_fd(dispatch_fd_t fd, uintptr_t hash)
1128 {
1129 // On fds lock queue
1130 _dispatch_io_debug("fd entry create", fd);
1131 dispatch_fd_entry_t fd_entry = _dispatch_fd_entry_create(
1132 _dispatch_io_fds_lockq);
1133 fd_entry->fd = fd;
1134 TAILQ_INSERT_TAIL(&_dispatch_io_fds[hash], fd_entry, fd_list);
1135 fd_entry->barrier_queue = dispatch_queue_create(
1136 "com.apple.libdispatch-io.barrierq", NULL);
1137 fd_entry->barrier_group = dispatch_group_create();
1138 dispatch_async(fd_entry->barrier_queue, ^{
1139 _dispatch_io_debug("fd entry stat", fd);
1140 int err, orig_flags, orig_nosigpipe = -1;
1141 struct stat st;
1142 _dispatch_io_syscall_switch(err,
1143 fstat(fd, &st),
1144 default: fd_entry->err = err; return;
1145 );
1146 fd_entry->stat.dev = st.st_dev;
1147 fd_entry->stat.mode = st.st_mode;
1148 _dispatch_io_syscall_switch(err,
1149 orig_flags = fcntl(fd, F_GETFL),
1150 default: (void)dispatch_assume_zero(err); break;
1151 );
1152 #if DISPATCH_USE_SETNOSIGPIPE // rdar://problem/4121123
1153 if (S_ISFIFO(st.st_mode)) {
1154 _dispatch_io_syscall_switch(err,
1155 orig_nosigpipe = fcntl(fd, F_GETNOSIGPIPE),
1156 default: (void)dispatch_assume_zero(err); break;
1157 );
1158 if (orig_nosigpipe != -1) {
1159 _dispatch_io_syscall_switch(err,
1160 orig_nosigpipe = fcntl(fd, F_SETNOSIGPIPE, 1),
1161 default:
1162 orig_nosigpipe = -1;
1163 (void)dispatch_assume_zero(err);
1164 break;
1165 );
1166 }
1167 }
1168 #endif
1169 if (S_ISREG(st.st_mode)) {
1170 if (orig_flags != -1) {
1171 _dispatch_io_syscall_switch(err,
1172 fcntl(fd, F_SETFL, orig_flags & ~O_NONBLOCK),
1173 default:
1174 orig_flags = -1;
1175 (void)dispatch_assume_zero(err);
1176 break;
1177 );
1178 }
1179 int32_t dev = major(st.st_dev);
1180 // We have to get the disk on the global dev queue. The
1181 // barrier queue cannot continue until that is complete
1182 dispatch_suspend(fd_entry->barrier_queue);
1183 dispatch_once_f(&_dispatch_io_devs_lockq_pred, NULL,
1184 _dispatch_io_devs_lockq_init);
1185 dispatch_async(_dispatch_io_devs_lockq, ^{
1186 _dispatch_disk_init(fd_entry, dev);
1187 dispatch_resume(fd_entry->barrier_queue);
1188 });
1189 } else {
1190 if (orig_flags != -1) {
1191 _dispatch_io_syscall_switch(err,
1192 fcntl(fd, F_SETFL, orig_flags | O_NONBLOCK),
1193 default:
1194 orig_flags = -1;
1195 (void)dispatch_assume_zero(err);
1196 break;
1197 );
1198 }
1199 _dispatch_stream_init(fd_entry, _dispatch_get_root_queue(
1200 DISPATCH_QUEUE_PRIORITY_DEFAULT, false));
1201 }
1202 fd_entry->orig_flags = orig_flags;
1203 fd_entry->orig_nosigpipe = orig_nosigpipe;
1204 });
1205 // This is the first item run when the close queue is resumed, indicating
1206 // that all channels associated with this entry have been closed and that
1207 // all operations associated with this entry have been freed
1208 dispatch_async(fd_entry->close_queue, ^{
1209 if (!fd_entry->disk) {
1210 _dispatch_io_debug("close queue fd_entry cleanup", fd);
1211 dispatch_op_direction_t dir;
1212 for (dir = 0; dir < DOP_DIR_MAX; dir++) {
1213 _dispatch_stream_dispose(fd_entry, dir);
1214 }
1215 } else {
1216 dispatch_disk_t disk = fd_entry->disk;
1217 dispatch_async(_dispatch_io_devs_lockq, ^{
1218 _dispatch_release(disk);
1219 });
1220 }
1221 // Remove this entry from the global fd list
1222 TAILQ_REMOVE(&_dispatch_io_fds[hash], fd_entry, fd_list);
1223 });
1224 // If there was a source associated with this stream, disposing of the
1225 // source cancels it and suspends the close queue. Freeing the fd_entry
1226 // structure must happen after the source cancel handler has finished
1227 dispatch_async(fd_entry->close_queue, ^{
1228 _dispatch_io_debug("close queue release", fd);
1229 dispatch_release(fd_entry->close_queue);
1230 _dispatch_io_debug("barrier queue release", fd);
1231 dispatch_release(fd_entry->barrier_queue);
1232 _dispatch_io_debug("barrier group release", fd);
1233 dispatch_release(fd_entry->barrier_group);
1234 if (fd_entry->orig_flags != -1) {
1235 _dispatch_io_syscall(
1236 fcntl(fd, F_SETFL, fd_entry->orig_flags)
1237 );
1238 }
1239 #if DISPATCH_USE_SETNOSIGPIPE // rdar://problem/4121123
1240 if (fd_entry->orig_nosigpipe != -1) {
1241 _dispatch_io_syscall(
1242 fcntl(fd, F_SETNOSIGPIPE, fd_entry->orig_nosigpipe)
1243 );
1244 }
1245 #endif
1246 if (fd_entry->convenience_channel) {
1247 fd_entry->convenience_channel->fd_entry = NULL;
1248 dispatch_release(fd_entry->convenience_channel);
1249 }
1250 free(fd_entry);
1251 });
1252 return fd_entry;
1253 }
1254
1255 static dispatch_fd_entry_t
1256 _dispatch_fd_entry_create_with_path(dispatch_io_path_data_t path_data,
1257 dev_t dev, mode_t mode)
1258 {
1259 // On devs lock queue
1260 _dispatch_io_debug("fd entry create with path %s", -1, path_data->path);
1261 dispatch_fd_entry_t fd_entry = _dispatch_fd_entry_create(
1262 path_data->channel->queue);
1263 if (S_ISREG(mode)) {
1264 _dispatch_disk_init(fd_entry, major(dev));
1265 } else {
1266 _dispatch_stream_init(fd_entry, _dispatch_get_root_queue(
1267 DISPATCH_QUEUE_PRIORITY_DEFAULT, false));
1268 }
1269 fd_entry->fd = -1;
1270 fd_entry->orig_flags = -1;
1271 fd_entry->path_data = path_data;
1272 fd_entry->stat.dev = dev;
1273 fd_entry->stat.mode = mode;
1274 fd_entry->barrier_queue = dispatch_queue_create(
1275 "com.apple.libdispatch-io.barrierq", NULL);
1276 fd_entry->barrier_group = dispatch_group_create();
1277 // This is the first item run when the close queue is resumed, indicating
1278 // that the channel associated with this entry has been closed and that
1279 // all operations associated with this entry have been freed
1280 dispatch_async(fd_entry->close_queue, ^{
1281 _dispatch_io_debug("close queue fd_entry cleanup", -1);
1282 if (!fd_entry->disk) {
1283 dispatch_op_direction_t dir;
1284 for (dir = 0; dir < DOP_DIR_MAX; dir++) {
1285 _dispatch_stream_dispose(fd_entry, dir);
1286 }
1287 }
1288 if (fd_entry->fd != -1) {
1289 close(fd_entry->fd);
1290 }
1291 if (fd_entry->path_data->channel) {
1292 // If associated channel has not been released yet, mark it as
1293 // no longer having an fd_entry (for stop after close).
1294 // It is safe to modify channel since we are on close_queue with
1295 // target queue the channel queue
1296 fd_entry->path_data->channel->fd_entry = NULL;
1297 }
1298 });
1299 dispatch_async(fd_entry->close_queue, ^{
1300 _dispatch_io_debug("close queue release", -1);
1301 dispatch_release(fd_entry->close_queue);
1302 dispatch_release(fd_entry->barrier_queue);
1303 dispatch_release(fd_entry->barrier_group);
1304 free(fd_entry->path_data);
1305 free(fd_entry);
1306 });
1307 return fd_entry;
1308 }
1309
1310 static int
1311 _dispatch_fd_entry_open(dispatch_fd_entry_t fd_entry, dispatch_io_t channel)
1312 {
1313 if (!(fd_entry->fd == -1 && fd_entry->path_data)) {
1314 return 0;
1315 }
1316 if (fd_entry->err) {
1317 return fd_entry->err;
1318 }
1319 int fd = -1;
1320 int oflag = fd_entry->disk ? fd_entry->path_data->oflag & ~O_NONBLOCK :
1321 fd_entry->path_data->oflag | O_NONBLOCK;
1322 open:
1323 fd = open(fd_entry->path_data->path, oflag, fd_entry->path_data->mode);
1324 if (fd == -1) {
1325 int err = errno;
1326 if (err == EINTR) {
1327 goto open;
1328 }
1329 (void)dispatch_atomic_cmpxchg2o(fd_entry, err, 0, err);
1330 return err;
1331 }
1332 if (!dispatch_atomic_cmpxchg2o(fd_entry, fd, -1, fd)) {
1333 // Lost the race with another open
1334 close(fd);
1335 } else {
1336 channel->fd_actual = fd;
1337 }
1338 return 0;
1339 }
1340
1341 static void
1342 _dispatch_fd_entry_cleanup_operations(dispatch_fd_entry_t fd_entry,
1343 dispatch_io_t channel)
1344 {
1345 if (fd_entry->disk) {
1346 if (channel) {
1347 _dispatch_retain(channel);
1348 }
1349 _dispatch_fd_entry_retain(fd_entry);
1350 dispatch_async(fd_entry->disk->pick_queue, ^{
1351 _dispatch_disk_cleanup_operations(fd_entry->disk, channel);
1352 _dispatch_fd_entry_release(fd_entry);
1353 if (channel) {
1354 _dispatch_release(channel);
1355 }
1356 });
1357 } else {
1358 dispatch_op_direction_t direction;
1359 for (direction = 0; direction < DOP_DIR_MAX; direction++) {
1360 dispatch_stream_t stream = fd_entry->streams[direction];
1361 if (!stream) {
1362 continue;
1363 }
1364 if (channel) {
1365 _dispatch_retain(channel);
1366 }
1367 _dispatch_fd_entry_retain(fd_entry);
1368 dispatch_async(stream->dq, ^{
1369 _dispatch_stream_cleanup_operations(stream, channel);
1370 _dispatch_fd_entry_release(fd_entry);
1371 if (channel) {
1372 _dispatch_release(channel);
1373 }
1374 });
1375 }
1376 }
1377 }
1378
1379 #pragma mark -
1380 #pragma mark dispatch_stream_t/dispatch_disk_t
1381
1382 static void
1383 _dispatch_stream_init(dispatch_fd_entry_t fd_entry, dispatch_queue_t tq)
1384 {
1385 dispatch_op_direction_t direction;
1386 for (direction = 0; direction < DOP_DIR_MAX; direction++) {
1387 dispatch_stream_t stream;
1388 stream = calloc(1ul, sizeof(struct dispatch_stream_s));
1389 stream->dq = dispatch_queue_create("com.apple.libdispatch-io.streamq",
1390 NULL);
1391 _dispatch_retain(tq);
1392 stream->dq->do_targetq = tq;
1393 TAILQ_INIT(&stream->operations[DISPATCH_IO_RANDOM]);
1394 TAILQ_INIT(&stream->operations[DISPATCH_IO_STREAM]);
1395 fd_entry->streams[direction] = stream;
1396 }
1397 }
1398
1399 static void
1400 _dispatch_stream_dispose(dispatch_fd_entry_t fd_entry,
1401 dispatch_op_direction_t direction)
1402 {
1403 // On close queue
1404 dispatch_stream_t stream = fd_entry->streams[direction];
1405 if (!stream) {
1406 return;
1407 }
1408 dispatch_assert(TAILQ_EMPTY(&stream->operations[DISPATCH_IO_STREAM]));
1409 dispatch_assert(TAILQ_EMPTY(&stream->operations[DISPATCH_IO_RANDOM]));
1410 if (stream->source) {
1411 // Balanced by source cancel handler:
1412 _dispatch_fd_entry_retain(fd_entry);
1413 dispatch_source_cancel(stream->source);
1414 dispatch_resume(stream->source);
1415 dispatch_release(stream->source);
1416 }
1417 dispatch_release(stream->dq);
1418 free(stream);
1419 }
1420
1421 static void
1422 _dispatch_disk_init(dispatch_fd_entry_t fd_entry, dev_t dev)
1423 {
1424 // On devs lock queue
1425 dispatch_disk_t disk;
1426 char label_name[256];
1427 // Check to see if there is an existing entry for the given device
1428 uintptr_t hash = DIO_HASH(dev);
1429 TAILQ_FOREACH(disk, &_dispatch_io_devs[hash], disk_list) {
1430 if (disk->dev == dev) {
1431 _dispatch_retain(disk);
1432 goto out;
1433 }
1434 }
1435 // Otherwise create a new entry
1436 size_t pending_reqs_depth = dispatch_io_defaults.max_pending_io_reqs;
1437 disk = calloc(1ul, sizeof(struct dispatch_disk_s) + (pending_reqs_depth *
1438 sizeof(dispatch_operation_t)));
1439 disk->do_vtable = &_dispatch_disk_vtable;
1440 disk->do_next = DISPATCH_OBJECT_LISTLESS;
1441 disk->do_ref_cnt = 1;
1442 disk->do_xref_cnt = 0;
1443 disk->advise_list_depth = pending_reqs_depth;
1444 disk->do_targetq = _dispatch_get_root_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT,
1445 false);
1446 disk->dev = dev;
1447 TAILQ_INIT(&disk->operations);
1448 disk->cur_rq = TAILQ_FIRST(&disk->operations);
1449 sprintf(label_name, "com.apple.libdispatch-io.deviceq.%d", dev);
1450 disk->pick_queue = dispatch_queue_create(label_name, NULL);
1451 TAILQ_INSERT_TAIL(&_dispatch_io_devs[hash], disk, disk_list);
1452 out:
1453 fd_entry->disk = disk;
1454 TAILQ_INIT(&fd_entry->stream_ops);
1455 }
1456
1457 static void
1458 _dispatch_disk_dispose(dispatch_disk_t disk)
1459 {
1460 uintptr_t hash = DIO_HASH(disk->dev);
1461 TAILQ_REMOVE(&_dispatch_io_devs[hash], disk, disk_list);
1462 dispatch_assert(TAILQ_EMPTY(&disk->operations));
1463 size_t i;
1464 for (i=0; i<disk->advise_list_depth; ++i) {
1465 dispatch_assert(!disk->advise_list[i]);
1466 }
1467 dispatch_release(disk->pick_queue);
1468 free(disk);
1469 }
1470
1471 #pragma mark -
1472 #pragma mark dispatch_stream_operations/dispatch_disk_operations
1473
1474 static inline bool
1475 _dispatch_stream_operation_avail(dispatch_stream_t stream)
1476 {
1477 return !(TAILQ_EMPTY(&stream->operations[DISPATCH_IO_RANDOM])) ||
1478 !(TAILQ_EMPTY(&stream->operations[DISPATCH_IO_STREAM]));
1479 }
1480
1481 static void
1482 _dispatch_stream_enqueue_operation(dispatch_stream_t stream,
1483 dispatch_operation_t op, dispatch_data_t data)
1484 {
1485 if (!_dispatch_operation_should_enqueue(op, stream->dq, data)) {
1486 return;
1487 }
1488 bool no_ops = !_dispatch_stream_operation_avail(stream);
1489 TAILQ_INSERT_TAIL(&stream->operations[op->params.type], op, operation_list);
1490 if (no_ops) {
1491 dispatch_async_f(stream->dq, stream, _dispatch_stream_handler);
1492 }
1493 }
1494
1495 static void
1496 _dispatch_disk_enqueue_operation(dispatch_disk_t disk, dispatch_operation_t op,
1497 dispatch_data_t data)
1498 {
1499 if (!_dispatch_operation_should_enqueue(op, disk->pick_queue, data)) {
1500 return;
1501 }
1502 if (op->params.type == DISPATCH_IO_STREAM) {
1503 if (TAILQ_EMPTY(&op->fd_entry->stream_ops)) {
1504 TAILQ_INSERT_TAIL(&disk->operations, op, operation_list);
1505 }
1506 TAILQ_INSERT_TAIL(&op->fd_entry->stream_ops, op, stream_list);
1507 } else {
1508 TAILQ_INSERT_TAIL(&disk->operations, op, operation_list);
1509 }
1510 _dispatch_disk_handler(disk);
1511 }
1512
1513 static void
1514 _dispatch_stream_complete_operation(dispatch_stream_t stream,
1515 dispatch_operation_t op)
1516 {
1517 // On stream queue
1518 _dispatch_io_debug("complete operation", op->fd_entry->fd);
1519 TAILQ_REMOVE(&stream->operations[op->params.type], op, operation_list);
1520 if (op == stream->op) {
1521 stream->op = NULL;
1522 }
1523 if (op->timer) {
1524 dispatch_source_cancel(op->timer);
1525 }
1526 // Final release will deliver any pending data
1527 _dispatch_release(op);
1528 }
1529
1530 static void
1531 _dispatch_disk_complete_operation(dispatch_disk_t disk, dispatch_operation_t op)
1532 {
1533 // On pick queue
1534 _dispatch_io_debug("complete operation", op->fd_entry->fd);
1535 // Current request is always the last op returned
1536 if (disk->cur_rq == op) {
1537 disk->cur_rq = TAILQ_PREV(op, dispatch_disk_operations_s,
1538 operation_list);
1539 }
1540 if (op->params.type == DISPATCH_IO_STREAM) {
1541 // Check if there are other pending stream operations behind it
1542 dispatch_operation_t op_next = TAILQ_NEXT(op, stream_list);
1543 TAILQ_REMOVE(&op->fd_entry->stream_ops, op, stream_list);
1544 if (op_next) {
1545 TAILQ_INSERT_TAIL(&disk->operations, op_next, operation_list);
1546 }
1547 }
1548 TAILQ_REMOVE(&disk->operations, op, operation_list);
1549 if (op->timer) {
1550 dispatch_source_cancel(op->timer);
1551 }
1552 // Final release will deliver any pending data
1553 _dispatch_release(op);
1554 }
1555
1556 static dispatch_operation_t
1557 _dispatch_stream_pick_next_operation(dispatch_stream_t stream,
1558 dispatch_operation_t op)
1559 {
1560 // On stream queue
1561 if (!op) {
1562 // On the first run through, pick the first operation
1563 if (!_dispatch_stream_operation_avail(stream)) {
1564 return op;
1565 }
1566 if (!TAILQ_EMPTY(&stream->operations[DISPATCH_IO_STREAM])) {
1567 op = TAILQ_FIRST(&stream->operations[DISPATCH_IO_STREAM]);
1568 } else if (!TAILQ_EMPTY(&stream->operations[DISPATCH_IO_RANDOM])) {
1569 op = TAILQ_FIRST(&stream->operations[DISPATCH_IO_RANDOM]);
1570 }
1571 return op;
1572 }
1573 if (op->params.type == DISPATCH_IO_STREAM) {
1574 // Stream operations need to be serialized so continue the current
1575 // operation until it is finished
1576 return op;
1577 }
1578 // Get the next random operation (round-robin)
1579 if (op->params.type == DISPATCH_IO_RANDOM) {
1580 op = TAILQ_NEXT(op, operation_list);
1581 if (!op) {
1582 op = TAILQ_FIRST(&stream->operations[DISPATCH_IO_RANDOM]);
1583 }
1584 return op;
1585 }
1586 return NULL;
1587 }
1588
1589 static dispatch_operation_t
1590 _dispatch_disk_pick_next_operation(dispatch_disk_t disk)
1591 {
1592 // On pick queue
1593 dispatch_operation_t op;
1594 if (!TAILQ_EMPTY(&disk->operations)) {
1595 if (disk->cur_rq == NULL) {
1596 op = TAILQ_FIRST(&disk->operations);
1597 } else {
1598 op = disk->cur_rq;
1599 do {
1600 op = TAILQ_NEXT(op, operation_list);
1601 if (!op) {
1602 op = TAILQ_FIRST(&disk->operations);
1603 }
1604 // TODO: more involved picking algorithm rdar://problem/8780312
1605 } while (op->active && op != disk->cur_rq);
1606 }
1607 if (!op->active) {
1608 disk->cur_rq = op;
1609 return op;
1610 }
1611 }
1612 return NULL;
1613 }
1614
1615 static void
1616 _dispatch_stream_cleanup_operations(dispatch_stream_t stream,
1617 dispatch_io_t channel)
1618 {
1619 // On stream queue
1620 dispatch_operation_t op, tmp;
1621 typeof(*stream->operations) *operations;
1622 operations = &stream->operations[DISPATCH_IO_RANDOM];
1623 TAILQ_FOREACH_SAFE(op, operations, operation_list, tmp) {
1624 if (!channel || op->channel == channel) {
1625 _dispatch_stream_complete_operation(stream, op);
1626 }
1627 }
1628 operations = &stream->operations[DISPATCH_IO_STREAM];
1629 TAILQ_FOREACH_SAFE(op, operations, operation_list, tmp) {
1630 if (!channel || op->channel == channel) {
1631 _dispatch_stream_complete_operation(stream, op);
1632 }
1633 }
1634 if (stream->source_running && !_dispatch_stream_operation_avail(stream)) {
1635 dispatch_suspend(stream->source);
1636 stream->source_running = false;
1637 }
1638 }
1639
1640 static void
1641 _dispatch_disk_cleanup_operations(dispatch_disk_t disk, dispatch_io_t channel)
1642 {
1643 // On pick queue
1644 dispatch_operation_t op, tmp;
1645 TAILQ_FOREACH_SAFE(op, &disk->operations, operation_list, tmp) {
1646 if (!channel || op->channel == channel) {
1647 _dispatch_disk_complete_operation(disk, op);
1648 }
1649 }
1650 }
1651
1652 #pragma mark -
1653 #pragma mark dispatch_stream_handler/dispatch_disk_handler
1654
1655 static dispatch_source_t
1656 _dispatch_stream_source(dispatch_stream_t stream, dispatch_operation_t op)
1657 {
1658 // On stream queue
1659 if (stream->source) {
1660 return stream->source;
1661 }
1662 dispatch_fd_t fd = op->fd_entry->fd;
1663 _dispatch_io_debug("stream source create", fd);
1664 dispatch_source_t source = NULL;
1665 if (op->direction == DOP_DIR_READ) {
1666 source = dispatch_source_create(DISPATCH_SOURCE_TYPE_READ, fd, 0,
1667 stream->dq);
1668 } else if (op->direction == DOP_DIR_WRITE) {
1669 source = dispatch_source_create(DISPATCH_SOURCE_TYPE_WRITE, fd, 0,
1670 stream->dq);
1671 } else {
1672 dispatch_assert(op->direction < DOP_DIR_MAX);
1673 return NULL;
1674 }
1675 dispatch_set_context(source, stream);
1676 dispatch_source_set_event_handler_f(source,
1677 _dispatch_stream_source_handler);
1678 // Close queue must not run user cleanup handlers until sources are fully
1679 // unregistered
1680 dispatch_queue_t close_queue = op->fd_entry->close_queue;
1681 dispatch_source_set_cancel_handler(source, ^{
1682 _dispatch_io_debug("stream source cancel", fd);
1683 dispatch_resume(close_queue);
1684 });
1685 stream->source = source;
1686 return stream->source;
1687 }
1688
1689 static void
1690 _dispatch_stream_source_handler(void *ctx)
1691 {
1692 // On stream queue
1693 dispatch_stream_t stream = (dispatch_stream_t)ctx;
1694 dispatch_suspend(stream->source);
1695 stream->source_running = false;
1696 return _dispatch_stream_handler(stream);
1697 }
1698
1699 static void
1700 _dispatch_stream_handler(void *ctx)
1701 {
1702 // On stream queue
1703 dispatch_stream_t stream = (dispatch_stream_t)ctx;
1704 dispatch_operation_t op;
1705 pick:
1706 op = _dispatch_stream_pick_next_operation(stream, stream->op);
1707 if (!op) {
1708 _dispatch_debug("no operation found: stream %p", stream);
1709 return;
1710 }
1711 int err = _dispatch_io_get_error(op, NULL, true);
1712 if (err) {
1713 op->err = err;
1714 _dispatch_stream_complete_operation(stream, op);
1715 goto pick;
1716 }
1717 stream->op = op;
1718 _dispatch_io_debug("stream handler", op->fd_entry->fd);
1719 dispatch_fd_entry_t fd_entry = op->fd_entry;
1720 _dispatch_fd_entry_retain(fd_entry);
1721 // For performance analysis
1722 if (!op->total && dispatch_io_defaults.initial_delivery) {
1723 // Empty delivery to signal the start of the operation
1724 _dispatch_io_debug("initial delivery", op->fd_entry->fd);
1725 _dispatch_operation_deliver_data(op, DOP_DELIVER);
1726 }
1727 // TODO: perform on the operation target queue to get correct priority
1728 int result = _dispatch_operation_perform(op), flags = -1;
1729 switch (result) {
1730 case DISPATCH_OP_DELIVER:
1731 flags = DOP_DEFAULT;
1732 // Fall through
1733 case DISPATCH_OP_DELIVER_AND_COMPLETE:
1734 flags = (flags != DOP_DEFAULT) ? DOP_DELIVER | DOP_NO_EMPTY :
1735 DOP_DEFAULT;
1736 _dispatch_operation_deliver_data(op, flags);
1737 // Fall through
1738 case DISPATCH_OP_COMPLETE:
1739 if (flags != DOP_DEFAULT) {
1740 _dispatch_stream_complete_operation(stream, op);
1741 }
1742 if (_dispatch_stream_operation_avail(stream)) {
1743 dispatch_async_f(stream->dq, stream, _dispatch_stream_handler);
1744 }
1745 break;
1746 case DISPATCH_OP_COMPLETE_RESUME:
1747 _dispatch_stream_complete_operation(stream, op);
1748 // Fall through
1749 case DISPATCH_OP_RESUME:
1750 if (_dispatch_stream_operation_avail(stream)) {
1751 stream->source_running = true;
1752 dispatch_resume(_dispatch_stream_source(stream, op));
1753 }
1754 break;
1755 case DISPATCH_OP_ERR:
1756 _dispatch_stream_cleanup_operations(stream, op->channel);
1757 break;
1758 case DISPATCH_OP_FD_ERR:
1759 _dispatch_fd_entry_retain(fd_entry);
1760 dispatch_async(fd_entry->barrier_queue, ^{
1761 _dispatch_fd_entry_cleanup_operations(fd_entry, NULL);
1762 _dispatch_fd_entry_release(fd_entry);
1763 });
1764 break;
1765 default:
1766 break;
1767 }
1768 _dispatch_fd_entry_release(fd_entry);
1769 return;
1770 }
1771
1772 static void
1773 _dispatch_disk_handler(void *ctx)
1774 {
1775 // On pick queue
1776 dispatch_disk_t disk = (dispatch_disk_t)ctx;
1777 if (disk->io_active) {
1778 return;
1779 }
1780 _dispatch_io_debug("disk handler", -1);
1781 dispatch_operation_t op;
1782 size_t i = disk->free_idx, j = disk->req_idx;
1783 if (j <= i) {
1784 j += disk->advise_list_depth;
1785 }
1786 while (i <= j) {
1787 if ((!disk->advise_list[i%disk->advise_list_depth]) &&
1788 (op = _dispatch_disk_pick_next_operation(disk))) {
1789 int err = _dispatch_io_get_error(op, NULL, true);
1790 if (err) {
1791 op->err = err;
1792 _dispatch_disk_complete_operation(disk, op);
1793 continue;
1794 }
1795 _dispatch_retain(op);
1796 disk->advise_list[i%disk->advise_list_depth] = op;
1797 op->active = true;
1798 } else {
1799 // No more operations to get
1800 break;
1801 }
1802 i++;
1803 }
1804 disk->free_idx = (i%disk->advise_list_depth);
1805 op = disk->advise_list[disk->req_idx];
1806 if (op) {
1807 disk->io_active = true;
1808 dispatch_async_f(op->do_targetq, disk, _dispatch_disk_perform);
1809 }
1810 }
1811
1812 static void
1813 _dispatch_disk_perform(void *ctxt)
1814 {
1815 dispatch_disk_t disk = ctxt;
1816 size_t chunk_size = dispatch_io_defaults.chunk_pages * PAGE_SIZE;
1817 _dispatch_io_debug("disk perform", -1);
1818 dispatch_operation_t op;
1819 size_t i = disk->advise_idx, j = disk->free_idx;
1820 if (j <= i) {
1821 j += disk->advise_list_depth;
1822 }
1823 do {
1824 op = disk->advise_list[i%disk->advise_list_depth];
1825 if (!op) {
1826 // Nothing more to advise, must be at free_idx
1827 dispatch_assert(i%disk->advise_list_depth == disk->free_idx);
1828 break;
1829 }
1830 if (op->direction == DOP_DIR_WRITE) {
1831 // TODO: preallocate writes ? rdar://problem/9032172
1832 continue;
1833 }
1834 if (op->fd_entry->fd == -1 && _dispatch_fd_entry_open(op->fd_entry,
1835 op->channel)) {
1836 continue;
1837 }
1838 // For performance analysis
1839 if (!op->total && dispatch_io_defaults.initial_delivery) {
1840 // Empty delivery to signal the start of the operation
1841 _dispatch_io_debug("initial delivery", op->fd_entry->fd);
1842 _dispatch_operation_deliver_data(op, DOP_DELIVER);
1843 }
1844 // Advise two chunks if the list only has one element and this is the
1845 // first advise on the operation
1846 if ((j-i) == 1 && !disk->advise_list[disk->free_idx] &&
1847 !op->advise_offset) {
1848 chunk_size *= 2;
1849 }
1850 _dispatch_operation_advise(op, chunk_size);
1851 } while (++i < j);
1852 disk->advise_idx = i%disk->advise_list_depth;
1853 op = disk->advise_list[disk->req_idx];
1854 int result = _dispatch_operation_perform(op);
1855 disk->advise_list[disk->req_idx] = NULL;
1856 disk->req_idx = (++disk->req_idx)%disk->advise_list_depth;
1857 dispatch_async(disk->pick_queue, ^{
1858 switch (result) {
1859 case DISPATCH_OP_DELIVER:
1860 _dispatch_operation_deliver_data(op, DOP_DELIVER);
1861 break;
1862 case DISPATCH_OP_COMPLETE:
1863 _dispatch_disk_complete_operation(disk, op);
1864 break;
1865 case DISPATCH_OP_DELIVER_AND_COMPLETE:
1866 _dispatch_operation_deliver_data(op, DOP_DELIVER);
1867 _dispatch_disk_complete_operation(disk, op);
1868 break;
1869 case DISPATCH_OP_ERR:
1870 _dispatch_disk_cleanup_operations(disk, op->channel);
1871 break;
1872 case DISPATCH_OP_FD_ERR:
1873 _dispatch_disk_cleanup_operations(disk, NULL);
1874 break;
1875 default:
1876 dispatch_assert(result);
1877 break;
1878 }
1879 op->active = false;
1880 disk->io_active = false;
1881 _dispatch_disk_handler(disk);
1882 // Balancing the retain in _dispatch_disk_handler. Note that op must be
1883 // released at the very end, since it might hold the last reference to
1884 // the disk
1885 _dispatch_release(op);
1886 });
1887 }
1888
1889 #pragma mark -
1890 #pragma mark dispatch_operation_perform
1891
1892 static void
1893 _dispatch_operation_advise(dispatch_operation_t op, size_t chunk_size)
1894 {
1895 int err;
1896 struct radvisory advise;
1897 // No point in issuing a read advise for the next chunk if we are already
1898 // a chunk ahead from reading the bytes
1899 if (op->advise_offset > (off_t)((op->offset+op->total) + chunk_size +
1900 PAGE_SIZE)) {
1901 return;
1902 }
1903 advise.ra_count = (int)chunk_size;
1904 if (!op->advise_offset) {
1905 op->advise_offset = op->offset;
1906 // If this is the first time through, align the advised range to a
1907 // page boundary
1908 size_t pg_fraction = (size_t)((op->offset + chunk_size) % PAGE_SIZE);
1909 advise.ra_count += (int)(pg_fraction ? PAGE_SIZE - pg_fraction : 0);
1910 }
1911 advise.ra_offset = op->advise_offset;
1912 op->advise_offset += advise.ra_count;
1913 _dispatch_io_syscall_switch(err,
1914 fcntl(op->fd_entry->fd, F_RDADVISE, &advise),
1915 // TODO: set disk status on error
1916 default: (void)dispatch_assume_zero(err); break;
1917 );
1918 }
1919
1920 static int
1921 _dispatch_operation_perform(dispatch_operation_t op)
1922 {
1923 int err = _dispatch_io_get_error(op, NULL, true);
1924 if (err) {
1925 goto error;
1926 }
1927 if (!op->buf) {
1928 size_t max_buf_siz = op->params.high;
1929 size_t chunk_siz = dispatch_io_defaults.chunk_pages * PAGE_SIZE;
1930 if (op->direction == DOP_DIR_READ) {
1931 // If necessary, create a buffer for the ongoing operation, large
1932 // enough to fit chunk_pages but at most high-water
1933 size_t data_siz = dispatch_data_get_size(op->data);
1934 if (data_siz) {
1935 dispatch_assert(data_siz < max_buf_siz);
1936 max_buf_siz -= data_siz;
1937 }
1938 if (max_buf_siz > chunk_siz) {
1939 max_buf_siz = chunk_siz;
1940 }
1941 if (op->length < SIZE_MAX) {
1942 op->buf_siz = op->length - op->total;
1943 if (op->buf_siz > max_buf_siz) {
1944 op->buf_siz = max_buf_siz;
1945 }
1946 } else {
1947 op->buf_siz = max_buf_siz;
1948 }
1949 op->buf = valloc(op->buf_siz);
1950 _dispatch_io_debug("buffer allocated", op->fd_entry->fd);
1951 } else if (op->direction == DOP_DIR_WRITE) {
1952 // Always write the first data piece, if that is smaller than a
1953 // chunk, accumulate further data pieces until chunk size is reached
1954 if (chunk_siz > max_buf_siz) {
1955 chunk_siz = max_buf_siz;
1956 }
1957 op->buf_siz = 0;
1958 dispatch_data_apply(op->data,
1959 ^(dispatch_data_t region DISPATCH_UNUSED,
1960 size_t offset DISPATCH_UNUSED,
1961 const void* buf DISPATCH_UNUSED, size_t len) {
1962 size_t siz = op->buf_siz + len;
1963 if (!op->buf_siz || siz <= chunk_siz) {
1964 op->buf_siz = siz;
1965 }
1966 return (bool)(siz < chunk_siz);
1967 });
1968 if (op->buf_siz > max_buf_siz) {
1969 op->buf_siz = max_buf_siz;
1970 }
1971 dispatch_data_t d;
1972 d = dispatch_data_create_subrange(op->data, 0, op->buf_siz);
1973 op->buf_data = dispatch_data_create_map(d, (const void**)&op->buf,
1974 NULL);
1975 _dispatch_io_data_release(d);
1976 _dispatch_io_debug("buffer mapped", op->fd_entry->fd);
1977 }
1978 }
1979 if (op->fd_entry->fd == -1) {
1980 err = _dispatch_fd_entry_open(op->fd_entry, op->channel);
1981 if (err) {
1982 goto error;
1983 }
1984 }
1985 void *buf = op->buf + op->buf_len;
1986 size_t len = op->buf_siz - op->buf_len;
1987 off_t off = op->offset + op->total;
1988 ssize_t processed = -1;
1989 syscall:
1990 if (op->direction == DOP_DIR_READ) {
1991 if (op->params.type == DISPATCH_IO_STREAM) {
1992 processed = read(op->fd_entry->fd, buf, len);
1993 } else if (op->params.type == DISPATCH_IO_RANDOM) {
1994 processed = pread(op->fd_entry->fd, buf, len, off);
1995 }
1996 } else if (op->direction == DOP_DIR_WRITE) {
1997 if (op->params.type == DISPATCH_IO_STREAM) {
1998 processed = write(op->fd_entry->fd, buf, len);
1999 } else if (op->params.type == DISPATCH_IO_RANDOM) {
2000 processed = pwrite(op->fd_entry->fd, buf, len, off);
2001 }
2002 }
2003 // Encountered an error on the file descriptor
2004 if (processed == -1) {
2005 err = errno;
2006 if (err == EINTR) {
2007 goto syscall;
2008 }
2009 goto error;
2010 }
2011 // EOF is indicated by two handler invocations
2012 if (processed == 0) {
2013 _dispatch_io_debug("EOF", op->fd_entry->fd);
2014 return DISPATCH_OP_DELIVER_AND_COMPLETE;
2015 }
2016 op->buf_len += processed;
2017 op->total += processed;
2018 if (op->total == op->length) {
2019 // Finished processing all the bytes requested by the operation
2020 return DISPATCH_OP_COMPLETE;
2021 } else {
2022 // Deliver data only if we satisfy the filters
2023 return DISPATCH_OP_DELIVER;
2024 }
2025 error:
2026 if (err == EAGAIN) {
2027 // For disk based files with blocking I/O we should never get EAGAIN
2028 dispatch_assert(!op->fd_entry->disk);
2029 _dispatch_io_debug("EAGAIN %d", op->fd_entry->fd, err);
2030 if (op->direction == DOP_DIR_READ && op->total &&
2031 op->channel == op->fd_entry->convenience_channel) {
2032 // Convenience read with available data completes on EAGAIN
2033 return DISPATCH_OP_COMPLETE_RESUME;
2034 }
2035 return DISPATCH_OP_RESUME;
2036 }
2037 op->err = err;
2038 switch (err) {
2039 case ECANCELED:
2040 return DISPATCH_OP_ERR;
2041 case EBADF:
2042 (void)dispatch_atomic_cmpxchg2o(op->fd_entry, err, 0, err);
2043 return DISPATCH_OP_FD_ERR;
2044 default:
2045 return DISPATCH_OP_COMPLETE;
2046 }
2047 }
2048
2049 static void
2050 _dispatch_operation_deliver_data(dispatch_operation_t op,
2051 dispatch_op_flags_t flags)
2052 {
2053 // Either called from stream resp. pick queue or when op is finalized
2054 dispatch_data_t data = NULL;
2055 int err = 0;
2056 size_t undelivered = op->undelivered + op->buf_len;
2057 bool deliver = (flags & (DOP_DELIVER|DOP_DONE)) ||
2058 (op->flags & DOP_DELIVER);
2059 op->flags = DOP_DEFAULT;
2060 if (!deliver) {
2061 // Don't deliver data until low water mark has been reached
2062 if (undelivered >= op->params.low) {
2063 deliver = true;
2064 } else if (op->buf_len < op->buf_siz) {
2065 // Request buffer is not yet used up
2066 _dispatch_io_debug("buffer data", op->fd_entry->fd);
2067 return;
2068 }
2069 } else {
2070 err = op->err;
2071 if (!err && (op->channel->atomic_flags & DIO_STOPPED)) {
2072 err = ECANCELED;
2073 op->err = err;
2074 }
2075 }
2076 // Deliver data or buffer used up
2077 if (op->direction == DOP_DIR_READ) {
2078 if (op->buf_len) {
2079 void *buf = op->buf;
2080 data = dispatch_data_create(buf, op->buf_len, NULL,
2081 DISPATCH_DATA_DESTRUCTOR_FREE);
2082 op->buf = NULL;
2083 op->buf_len = 0;
2084 dispatch_data_t d = dispatch_data_create_concat(op->data, data);
2085 _dispatch_io_data_release(op->data);
2086 _dispatch_io_data_release(data);
2087 data = d;
2088 } else {
2089 data = op->data;
2090 }
2091 op->data = deliver ? dispatch_data_empty : data;
2092 } else if (op->direction == DOP_DIR_WRITE) {
2093 if (deliver) {
2094 data = dispatch_data_create_subrange(op->data, op->buf_len,
2095 op->length);
2096 }
2097 if (op->buf_len == op->buf_siz) {
2098 _dispatch_io_data_release(op->buf_data);
2099 op->buf_data = NULL;
2100 op->buf = NULL;
2101 op->buf_len = 0;
2102 // Trim newly written buffer from head of unwritten data
2103 dispatch_data_t d;
2104 if (deliver) {
2105 _dispatch_io_data_retain(data);
2106 d = data;
2107 } else {
2108 d = dispatch_data_create_subrange(op->data, op->buf_len,
2109 op->length);
2110 }
2111 _dispatch_io_data_release(op->data);
2112 op->data = d;
2113 }
2114 } else {
2115 dispatch_assert(op->direction < DOP_DIR_MAX);
2116 return;
2117 }
2118 if (!deliver || ((flags & DOP_NO_EMPTY) && !dispatch_data_get_size(data))) {
2119 op->undelivered = undelivered;
2120 _dispatch_io_debug("buffer data", op->fd_entry->fd);
2121 return;
2122 }
2123 op->undelivered = 0;
2124 _dispatch_io_debug("deliver data", op->fd_entry->fd);
2125 dispatch_op_direction_t direction = op->direction;
2126 __block dispatch_data_t d = data;
2127 dispatch_io_handler_t handler = op->handler;
2128 #if DISPATCH_IO_DEBUG
2129 int fd = op->fd_entry->fd;
2130 #endif
2131 dispatch_fd_entry_t fd_entry = op->fd_entry;
2132 _dispatch_fd_entry_retain(fd_entry);
2133 dispatch_io_t channel = op->channel;
2134 _dispatch_retain(channel);
2135 // Note that data delivery may occur after the operation is freed
2136 dispatch_async(op->op_q, ^{
2137 bool done = (flags & DOP_DONE);
2138 if (done) {
2139 if (direction == DOP_DIR_READ && err) {
2140 if (dispatch_data_get_size(d)) {
2141 _dispatch_io_debug("IO handler invoke", fd);
2142 handler(false, d, 0);
2143 }
2144 d = NULL;
2145 } else if (direction == DOP_DIR_WRITE && !err) {
2146 d = NULL;
2147 }
2148 }
2149 _dispatch_io_debug("IO handler invoke", fd);
2150 handler(done, d, err);
2151 _dispatch_release(channel);
2152 _dispatch_fd_entry_release(fd_entry);
2153 _dispatch_io_data_release(data);
2154 });
2155 }
mirror of libdispatch