]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 1991, 1992 Linus Torvalds | |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> | |
7 | * - July2000 | |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 | |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/blk-mq.h> | |
20 | #include <linux/highmem.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/kernel_stat.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/completion.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/writeback.h> | |
29 | #include <linux/task_io_accounting_ops.h> | |
30 | #include <linux/fault-inject.h> | |
31 | #include <linux/list_sort.h> | |
32 | #include <linux/delay.h> | |
33 | #include <linux/ratelimit.h> | |
34 | #include <linux/pm_runtime.h> | |
35 | #include <linux/blk-cgroup.h> | |
36 | #include <linux/debugfs.h> | |
37 | ||
38 | #define CREATE_TRACE_POINTS | |
39 | #include <trace/events/block.h> | |
40 | ||
41 | #include "blk.h" | |
42 | #include "blk-mq.h" | |
43 | #include "blk-mq-sched.h" | |
44 | #include "blk-wbt.h" | |
45 | ||
46 | #ifdef CONFIG_DEBUG_FS | |
47 | struct dentry *blk_debugfs_root; | |
48 | #endif | |
49 | ||
50 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); | |
51 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); | |
52 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); | |
53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); | |
54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); | |
55 | ||
56 | DEFINE_IDA(blk_queue_ida); | |
57 | ||
58 | /* | |
59 | * For the allocated request tables | |
60 | */ | |
61 | struct kmem_cache *request_cachep; | |
62 | ||
63 | /* | |
64 | * For queue allocation | |
65 | */ | |
66 | struct kmem_cache *blk_requestq_cachep; | |
67 | ||
68 | /* | |
69 | * Controlling structure to kblockd | |
70 | */ | |
71 | static struct workqueue_struct *kblockd_workqueue; | |
72 | ||
73 | static void blk_clear_congested(struct request_list *rl, int sync) | |
74 | { | |
75 | #ifdef CONFIG_CGROUP_WRITEBACK | |
76 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
77 | #else | |
78 | /* | |
79 | * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't | |
80 | * flip its congestion state for events on other blkcgs. | |
81 | */ | |
82 | if (rl == &rl->q->root_rl) | |
83 | clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync); | |
84 | #endif | |
85 | } | |
86 | ||
87 | static void blk_set_congested(struct request_list *rl, int sync) | |
88 | { | |
89 | #ifdef CONFIG_CGROUP_WRITEBACK | |
90 | set_wb_congested(rl->blkg->wb_congested, sync); | |
91 | #else | |
92 | /* see blk_clear_congested() */ | |
93 | if (rl == &rl->q->root_rl) | |
94 | set_wb_congested(rl->q->backing_dev_info->wb.congested, sync); | |
95 | #endif | |
96 | } | |
97 | ||
98 | void blk_queue_congestion_threshold(struct request_queue *q) | |
99 | { | |
100 | int nr; | |
101 | ||
102 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
103 | if (nr > q->nr_requests) | |
104 | nr = q->nr_requests; | |
105 | q->nr_congestion_on = nr; | |
106 | ||
107 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
108 | if (nr < 1) | |
109 | nr = 1; | |
110 | q->nr_congestion_off = nr; | |
111 | } | |
112 | ||
113 | void blk_rq_init(struct request_queue *q, struct request *rq) | |
114 | { | |
115 | memset(rq, 0, sizeof(*rq)); | |
116 | ||
117 | INIT_LIST_HEAD(&rq->queuelist); | |
118 | INIT_LIST_HEAD(&rq->timeout_list); | |
119 | rq->cpu = -1; | |
120 | rq->q = q; | |
121 | rq->__sector = (sector_t) -1; | |
122 | INIT_HLIST_NODE(&rq->hash); | |
123 | RB_CLEAR_NODE(&rq->rb_node); | |
124 | rq->tag = -1; | |
125 | rq->internal_tag = -1; | |
126 | rq->start_time = jiffies; | |
127 | set_start_time_ns(rq); | |
128 | rq->part = NULL; | |
129 | } | |
130 | EXPORT_SYMBOL(blk_rq_init); | |
131 | ||
132 | static const struct { | |
133 | int errno; | |
134 | const char *name; | |
135 | } blk_errors[] = { | |
136 | [BLK_STS_OK] = { 0, "" }, | |
137 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
138 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
139 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
140 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
141 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
142 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
143 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
144 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
145 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
146 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, | |
147 | ||
148 | /* device mapper special case, should not leak out: */ | |
149 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
150 | ||
151 | /* everything else not covered above: */ | |
152 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
153 | }; | |
154 | ||
155 | blk_status_t errno_to_blk_status(int errno) | |
156 | { | |
157 | int i; | |
158 | ||
159 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
160 | if (blk_errors[i].errno == errno) | |
161 | return (__force blk_status_t)i; | |
162 | } | |
163 | ||
164 | return BLK_STS_IOERR; | |
165 | } | |
166 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
167 | ||
168 | int blk_status_to_errno(blk_status_t status) | |
169 | { | |
170 | int idx = (__force int)status; | |
171 | ||
172 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) | |
173 | return -EIO; | |
174 | return blk_errors[idx].errno; | |
175 | } | |
176 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
177 | ||
178 | static void print_req_error(struct request *req, blk_status_t status) | |
179 | { | |
180 | int idx = (__force int)status; | |
181 | ||
182 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) | |
183 | return; | |
184 | ||
185 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", | |
186 | __func__, blk_errors[idx].name, req->rq_disk ? | |
187 | req->rq_disk->disk_name : "?", | |
188 | (unsigned long long)blk_rq_pos(req)); | |
189 | } | |
190 | ||
191 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
192 | unsigned int nbytes, blk_status_t error) | |
193 | { | |
194 | if (error) | |
195 | bio->bi_status = error; | |
196 | ||
197 | if (unlikely(rq->rq_flags & RQF_QUIET)) | |
198 | bio_set_flag(bio, BIO_QUIET); | |
199 | ||
200 | bio_advance(bio, nbytes); | |
201 | ||
202 | /* don't actually finish bio if it's part of flush sequence */ | |
203 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) | |
204 | bio_endio(bio); | |
205 | } | |
206 | ||
207 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
208 | { | |
209 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, | |
210 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
211 | (unsigned long long) rq->cmd_flags); | |
212 | ||
213 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", | |
214 | (unsigned long long)blk_rq_pos(rq), | |
215 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
216 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", | |
217 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
218 | } | |
219 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
220 | ||
221 | static void blk_delay_work(struct work_struct *work) | |
222 | { | |
223 | struct request_queue *q; | |
224 | ||
225 | q = container_of(work, struct request_queue, delay_work.work); | |
226 | spin_lock_irq(q->queue_lock); | |
227 | __blk_run_queue(q); | |
228 | spin_unlock_irq(q->queue_lock); | |
229 | } | |
230 | ||
231 | /** | |
232 | * blk_delay_queue - restart queueing after defined interval | |
233 | * @q: The &struct request_queue in question | |
234 | * @msecs: Delay in msecs | |
235 | * | |
236 | * Description: | |
237 | * Sometimes queueing needs to be postponed for a little while, to allow | |
238 | * resources to come back. This function will make sure that queueing is | |
239 | * restarted around the specified time. | |
240 | */ | |
241 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
242 | { | |
243 | lockdep_assert_held(q->queue_lock); | |
244 | WARN_ON_ONCE(q->mq_ops); | |
245 | ||
246 | if (likely(!blk_queue_dead(q))) | |
247 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
248 | msecs_to_jiffies(msecs)); | |
249 | } | |
250 | EXPORT_SYMBOL(blk_delay_queue); | |
251 | ||
252 | /** | |
253 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
254 | * @q: The &struct request_queue in question | |
255 | * | |
256 | * Description: | |
257 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
258 | * ensure that the request_fn for the queue is run from an async | |
259 | * context. | |
260 | **/ | |
261 | void blk_start_queue_async(struct request_queue *q) | |
262 | { | |
263 | lockdep_assert_held(q->queue_lock); | |
264 | WARN_ON_ONCE(q->mq_ops); | |
265 | ||
266 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
267 | blk_run_queue_async(q); | |
268 | } | |
269 | EXPORT_SYMBOL(blk_start_queue_async); | |
270 | ||
271 | /** | |
272 | * blk_start_queue - restart a previously stopped queue | |
273 | * @q: The &struct request_queue in question | |
274 | * | |
275 | * Description: | |
276 | * blk_start_queue() will clear the stop flag on the queue, and call | |
277 | * the request_fn for the queue if it was in a stopped state when | |
278 | * entered. Also see blk_stop_queue(). | |
279 | **/ | |
280 | void blk_start_queue(struct request_queue *q) | |
281 | { | |
282 | lockdep_assert_held(q->queue_lock); | |
283 | WARN_ON(!in_interrupt() && !irqs_disabled()); | |
284 | WARN_ON_ONCE(q->mq_ops); | |
285 | ||
286 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
287 | __blk_run_queue(q); | |
288 | } | |
289 | EXPORT_SYMBOL(blk_start_queue); | |
290 | ||
291 | /** | |
292 | * blk_stop_queue - stop a queue | |
293 | * @q: The &struct request_queue in question | |
294 | * | |
295 | * Description: | |
296 | * The Linux block layer assumes that a block driver will consume all | |
297 | * entries on the request queue when the request_fn strategy is called. | |
298 | * Often this will not happen, because of hardware limitations (queue | |
299 | * depth settings). If a device driver gets a 'queue full' response, | |
300 | * or if it simply chooses not to queue more I/O at one point, it can | |
301 | * call this function to prevent the request_fn from being called until | |
302 | * the driver has signalled it's ready to go again. This happens by calling | |
303 | * blk_start_queue() to restart queue operations. | |
304 | **/ | |
305 | void blk_stop_queue(struct request_queue *q) | |
306 | { | |
307 | lockdep_assert_held(q->queue_lock); | |
308 | WARN_ON_ONCE(q->mq_ops); | |
309 | ||
310 | cancel_delayed_work(&q->delay_work); | |
311 | queue_flag_set(QUEUE_FLAG_STOPPED, q); | |
312 | } | |
313 | EXPORT_SYMBOL(blk_stop_queue); | |
314 | ||
315 | /** | |
316 | * blk_sync_queue - cancel any pending callbacks on a queue | |
317 | * @q: the queue | |
318 | * | |
319 | * Description: | |
320 | * The block layer may perform asynchronous callback activity | |
321 | * on a queue, such as calling the unplug function after a timeout. | |
322 | * A block device may call blk_sync_queue to ensure that any | |
323 | * such activity is cancelled, thus allowing it to release resources | |
324 | * that the callbacks might use. The caller must already have made sure | |
325 | * that its ->make_request_fn will not re-add plugging prior to calling | |
326 | * this function. | |
327 | * | |
328 | * This function does not cancel any asynchronous activity arising | |
329 | * out of elevator or throttling code. That would require elevator_exit() | |
330 | * and blkcg_exit_queue() to be called with queue lock initialized. | |
331 | * | |
332 | */ | |
333 | void blk_sync_queue(struct request_queue *q) | |
334 | { | |
335 | del_timer_sync(&q->timeout); | |
336 | cancel_work_sync(&q->timeout_work); | |
337 | ||
338 | if (q->mq_ops) { | |
339 | struct blk_mq_hw_ctx *hctx; | |
340 | int i; | |
341 | ||
342 | queue_for_each_hw_ctx(q, hctx, i) | |
343 | cancel_delayed_work_sync(&hctx->run_work); | |
344 | } else { | |
345 | cancel_delayed_work_sync(&q->delay_work); | |
346 | } | |
347 | } | |
348 | EXPORT_SYMBOL(blk_sync_queue); | |
349 | ||
350 | /** | |
351 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
352 | * @q: The queue to run | |
353 | * | |
354 | * Description: | |
355 | * Invoke request handling on a queue if there are any pending requests. | |
356 | * May be used to restart request handling after a request has completed. | |
357 | * This variant runs the queue whether or not the queue has been | |
358 | * stopped. Must be called with the queue lock held and interrupts | |
359 | * disabled. See also @blk_run_queue. | |
360 | */ | |
361 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
362 | { | |
363 | lockdep_assert_held(q->queue_lock); | |
364 | WARN_ON_ONCE(q->mq_ops); | |
365 | ||
366 | if (unlikely(blk_queue_dead(q))) | |
367 | return; | |
368 | ||
369 | /* | |
370 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
371 | * the queue lock internally. As a result multiple threads may be | |
372 | * running such a request function concurrently. Keep track of the | |
373 | * number of active request_fn invocations such that blk_drain_queue() | |
374 | * can wait until all these request_fn calls have finished. | |
375 | */ | |
376 | q->request_fn_active++; | |
377 | q->request_fn(q); | |
378 | q->request_fn_active--; | |
379 | } | |
380 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); | |
381 | ||
382 | /** | |
383 | * __blk_run_queue - run a single device queue | |
384 | * @q: The queue to run | |
385 | * | |
386 | * Description: | |
387 | * See @blk_run_queue. | |
388 | */ | |
389 | void __blk_run_queue(struct request_queue *q) | |
390 | { | |
391 | lockdep_assert_held(q->queue_lock); | |
392 | WARN_ON_ONCE(q->mq_ops); | |
393 | ||
394 | if (unlikely(blk_queue_stopped(q))) | |
395 | return; | |
396 | ||
397 | __blk_run_queue_uncond(q); | |
398 | } | |
399 | EXPORT_SYMBOL(__blk_run_queue); | |
400 | ||
401 | /** | |
402 | * blk_run_queue_async - run a single device queue in workqueue context | |
403 | * @q: The queue to run | |
404 | * | |
405 | * Description: | |
406 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
407 | * of us. | |
408 | * | |
409 | * Note: | |
410 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
411 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
412 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
413 | */ | |
414 | void blk_run_queue_async(struct request_queue *q) | |
415 | { | |
416 | lockdep_assert_held(q->queue_lock); | |
417 | WARN_ON_ONCE(q->mq_ops); | |
418 | ||
419 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) | |
420 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); | |
421 | } | |
422 | EXPORT_SYMBOL(blk_run_queue_async); | |
423 | ||
424 | /** | |
425 | * blk_run_queue - run a single device queue | |
426 | * @q: The queue to run | |
427 | * | |
428 | * Description: | |
429 | * Invoke request handling on this queue, if it has pending work to do. | |
430 | * May be used to restart queueing when a request has completed. | |
431 | */ | |
432 | void blk_run_queue(struct request_queue *q) | |
433 | { | |
434 | unsigned long flags; | |
435 | ||
436 | WARN_ON_ONCE(q->mq_ops); | |
437 | ||
438 | spin_lock_irqsave(q->queue_lock, flags); | |
439 | __blk_run_queue(q); | |
440 | spin_unlock_irqrestore(q->queue_lock, flags); | |
441 | } | |
442 | EXPORT_SYMBOL(blk_run_queue); | |
443 | ||
444 | void blk_put_queue(struct request_queue *q) | |
445 | { | |
446 | kobject_put(&q->kobj); | |
447 | } | |
448 | EXPORT_SYMBOL(blk_put_queue); | |
449 | ||
450 | /** | |
451 | * __blk_drain_queue - drain requests from request_queue | |
452 | * @q: queue to drain | |
453 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV | |
454 | * | |
455 | * Drain requests from @q. If @drain_all is set, all requests are drained. | |
456 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
457 | * for ensuring that no new requests which need to be drained are queued. | |
458 | */ | |
459 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) | |
460 | __releases(q->queue_lock) | |
461 | __acquires(q->queue_lock) | |
462 | { | |
463 | int i; | |
464 | ||
465 | lockdep_assert_held(q->queue_lock); | |
466 | WARN_ON_ONCE(q->mq_ops); | |
467 | ||
468 | while (true) { | |
469 | bool drain = false; | |
470 | ||
471 | /* | |
472 | * The caller might be trying to drain @q before its | |
473 | * elevator is initialized. | |
474 | */ | |
475 | if (q->elevator) | |
476 | elv_drain_elevator(q); | |
477 | ||
478 | blkcg_drain_queue(q); | |
479 | ||
480 | /* | |
481 | * This function might be called on a queue which failed | |
482 | * driver init after queue creation or is not yet fully | |
483 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
484 | * in such cases. Kick queue iff dispatch queue has | |
485 | * something on it and @q has request_fn set. | |
486 | */ | |
487 | if (!list_empty(&q->queue_head) && q->request_fn) | |
488 | __blk_run_queue(q); | |
489 | ||
490 | drain |= q->nr_rqs_elvpriv; | |
491 | drain |= q->request_fn_active; | |
492 | ||
493 | /* | |
494 | * Unfortunately, requests are queued at and tracked from | |
495 | * multiple places and there's no single counter which can | |
496 | * be drained. Check all the queues and counters. | |
497 | */ | |
498 | if (drain_all) { | |
499 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
500 | drain |= !list_empty(&q->queue_head); | |
501 | for (i = 0; i < 2; i++) { | |
502 | drain |= q->nr_rqs[i]; | |
503 | drain |= q->in_flight[i]; | |
504 | if (fq) | |
505 | drain |= !list_empty(&fq->flush_queue[i]); | |
506 | } | |
507 | } | |
508 | ||
509 | if (!drain) | |
510 | break; | |
511 | ||
512 | spin_unlock_irq(q->queue_lock); | |
513 | ||
514 | msleep(10); | |
515 | ||
516 | spin_lock_irq(q->queue_lock); | |
517 | } | |
518 | ||
519 | /* | |
520 | * With queue marked dead, any woken up waiter will fail the | |
521 | * allocation path, so the wakeup chaining is lost and we're | |
522 | * left with hung waiters. We need to wake up those waiters. | |
523 | */ | |
524 | if (q->request_fn) { | |
525 | struct request_list *rl; | |
526 | ||
527 | blk_queue_for_each_rl(rl, q) | |
528 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
529 | wake_up_all(&rl->wait[i]); | |
530 | } | |
531 | } | |
532 | ||
533 | /** | |
534 | * blk_queue_bypass_start - enter queue bypass mode | |
535 | * @q: queue of interest | |
536 | * | |
537 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
538 | * function makes @q enter bypass mode and drains all requests which were | |
539 | * throttled or issued before. On return, it's guaranteed that no request | |
540 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true | |
541 | * inside queue or RCU read lock. | |
542 | */ | |
543 | void blk_queue_bypass_start(struct request_queue *q) | |
544 | { | |
545 | WARN_ON_ONCE(q->mq_ops); | |
546 | ||
547 | spin_lock_irq(q->queue_lock); | |
548 | q->bypass_depth++; | |
549 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
550 | spin_unlock_irq(q->queue_lock); | |
551 | ||
552 | /* | |
553 | * Queues start drained. Skip actual draining till init is | |
554 | * complete. This avoids lenghty delays during queue init which | |
555 | * can happen many times during boot. | |
556 | */ | |
557 | if (blk_queue_init_done(q)) { | |
558 | spin_lock_irq(q->queue_lock); | |
559 | __blk_drain_queue(q, false); | |
560 | spin_unlock_irq(q->queue_lock); | |
561 | ||
562 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ | |
563 | synchronize_rcu(); | |
564 | } | |
565 | } | |
566 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
567 | ||
568 | /** | |
569 | * blk_queue_bypass_end - leave queue bypass mode | |
570 | * @q: queue of interest | |
571 | * | |
572 | * Leave bypass mode and restore the normal queueing behavior. | |
573 | * | |
574 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
575 | * this function is called for both blk-sq and blk-mq queues. | |
576 | */ | |
577 | void blk_queue_bypass_end(struct request_queue *q) | |
578 | { | |
579 | spin_lock_irq(q->queue_lock); | |
580 | if (!--q->bypass_depth) | |
581 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
582 | WARN_ON_ONCE(q->bypass_depth < 0); | |
583 | spin_unlock_irq(q->queue_lock); | |
584 | } | |
585 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
586 | ||
587 | void blk_set_queue_dying(struct request_queue *q) | |
588 | { | |
589 | spin_lock_irq(q->queue_lock); | |
590 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
591 | spin_unlock_irq(q->queue_lock); | |
592 | ||
593 | /* | |
594 | * When queue DYING flag is set, we need to block new req | |
595 | * entering queue, so we call blk_freeze_queue_start() to | |
596 | * prevent I/O from crossing blk_queue_enter(). | |
597 | */ | |
598 | blk_freeze_queue_start(q); | |
599 | ||
600 | if (q->mq_ops) | |
601 | blk_mq_wake_waiters(q); | |
602 | else { | |
603 | struct request_list *rl; | |
604 | ||
605 | spin_lock_irq(q->queue_lock); | |
606 | blk_queue_for_each_rl(rl, q) { | |
607 | if (rl->rq_pool) { | |
608 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
609 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
610 | } | |
611 | } | |
612 | spin_unlock_irq(q->queue_lock); | |
613 | } | |
614 | } | |
615 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
616 | ||
617 | /** | |
618 | * blk_cleanup_queue - shutdown a request queue | |
619 | * @q: request queue to shutdown | |
620 | * | |
621 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and | |
622 | * put it. All future requests will be failed immediately with -ENODEV. | |
623 | */ | |
624 | void blk_cleanup_queue(struct request_queue *q) | |
625 | { | |
626 | spinlock_t *lock = q->queue_lock; | |
627 | ||
628 | /* mark @q DYING, no new request or merges will be allowed afterwards */ | |
629 | mutex_lock(&q->sysfs_lock); | |
630 | blk_set_queue_dying(q); | |
631 | spin_lock_irq(lock); | |
632 | ||
633 | /* | |
634 | * A dying queue is permanently in bypass mode till released. Note | |
635 | * that, unlike blk_queue_bypass_start(), we aren't performing | |
636 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
637 | * as some drivers create and destroy a lot of queues while | |
638 | * probing. This is still safe because blk_release_queue() will be | |
639 | * called only after the queue refcnt drops to zero and nothing, | |
640 | * RCU or not, would be traversing the queue by then. | |
641 | */ | |
642 | q->bypass_depth++; | |
643 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
644 | ||
645 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); | |
646 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
647 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
648 | spin_unlock_irq(lock); | |
649 | mutex_unlock(&q->sysfs_lock); | |
650 | ||
651 | /* | |
652 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
653 | * prevent that q->request_fn() gets invoked after draining finished. | |
654 | */ | |
655 | blk_freeze_queue(q); | |
656 | spin_lock_irq(lock); | |
657 | if (!q->mq_ops) | |
658 | __blk_drain_queue(q, true); | |
659 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
660 | spin_unlock_irq(lock); | |
661 | ||
662 | /* for synchronous bio-based driver finish in-flight integrity i/o */ | |
663 | blk_flush_integrity(); | |
664 | ||
665 | /* @q won't process any more request, flush async actions */ | |
666 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); | |
667 | blk_sync_queue(q); | |
668 | ||
669 | if (q->mq_ops) | |
670 | blk_mq_free_queue(q); | |
671 | percpu_ref_exit(&q->q_usage_counter); | |
672 | ||
673 | spin_lock_irq(lock); | |
674 | if (q->queue_lock != &q->__queue_lock) | |
675 | q->queue_lock = &q->__queue_lock; | |
676 | spin_unlock_irq(lock); | |
677 | ||
678 | /* @q is and will stay empty, shutdown and put */ | |
679 | blk_put_queue(q); | |
680 | } | |
681 | EXPORT_SYMBOL(blk_cleanup_queue); | |
682 | ||
683 | /* Allocate memory local to the request queue */ | |
684 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) | |
685 | { | |
686 | struct request_queue *q = data; | |
687 | ||
688 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
689 | } | |
690 | ||
691 | static void free_request_simple(void *element, void *data) | |
692 | { | |
693 | kmem_cache_free(request_cachep, element); | |
694 | } | |
695 | ||
696 | static void *alloc_request_size(gfp_t gfp_mask, void *data) | |
697 | { | |
698 | struct request_queue *q = data; | |
699 | struct request *rq; | |
700 | ||
701 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
702 | q->node); | |
703 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
704 | kfree(rq); | |
705 | rq = NULL; | |
706 | } | |
707 | return rq; | |
708 | } | |
709 | ||
710 | static void free_request_size(void *element, void *data) | |
711 | { | |
712 | struct request_queue *q = data; | |
713 | ||
714 | if (q->exit_rq_fn) | |
715 | q->exit_rq_fn(q, element); | |
716 | kfree(element); | |
717 | } | |
718 | ||
719 | int blk_init_rl(struct request_list *rl, struct request_queue *q, | |
720 | gfp_t gfp_mask) | |
721 | { | |
722 | if (unlikely(rl->rq_pool) || q->mq_ops) | |
723 | return 0; | |
724 | ||
725 | rl->q = q; | |
726 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; | |
727 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
728 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); | |
729 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
730 | ||
731 | if (q->cmd_size) { | |
732 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
733 | alloc_request_size, free_request_size, | |
734 | q, gfp_mask, q->node); | |
735 | } else { | |
736 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
737 | alloc_request_simple, free_request_simple, | |
738 | q, gfp_mask, q->node); | |
739 | } | |
740 | if (!rl->rq_pool) | |
741 | return -ENOMEM; | |
742 | ||
743 | if (rl != &q->root_rl) | |
744 | WARN_ON_ONCE(!blk_get_queue(q)); | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
749 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) | |
750 | { | |
751 | if (rl->rq_pool) { | |
752 | mempool_destroy(rl->rq_pool); | |
753 | if (rl != &q->root_rl) | |
754 | blk_put_queue(q); | |
755 | } | |
756 | } | |
757 | ||
758 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
759 | { | |
760 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); | |
761 | } | |
762 | EXPORT_SYMBOL(blk_alloc_queue); | |
763 | ||
764 | int blk_queue_enter(struct request_queue *q, bool nowait) | |
765 | { | |
766 | while (true) { | |
767 | int ret; | |
768 | ||
769 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
770 | return 0; | |
771 | ||
772 | if (nowait) | |
773 | return -EBUSY; | |
774 | ||
775 | /* | |
776 | * read pair of barrier in blk_freeze_queue_start(), | |
777 | * we need to order reading __PERCPU_REF_DEAD flag of | |
778 | * .q_usage_counter and reading .mq_freeze_depth or | |
779 | * queue dying flag, otherwise the following wait may | |
780 | * never return if the two reads are reordered. | |
781 | */ | |
782 | smp_rmb(); | |
783 | ||
784 | ret = wait_event_interruptible(q->mq_freeze_wq, | |
785 | !atomic_read(&q->mq_freeze_depth) || | |
786 | blk_queue_dying(q)); | |
787 | if (blk_queue_dying(q)) | |
788 | return -ENODEV; | |
789 | if (ret) | |
790 | return ret; | |
791 | } | |
792 | } | |
793 | ||
794 | void blk_queue_exit(struct request_queue *q) | |
795 | { | |
796 | percpu_ref_put(&q->q_usage_counter); | |
797 | } | |
798 | ||
799 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
800 | { | |
801 | struct request_queue *q = | |
802 | container_of(ref, struct request_queue, q_usage_counter); | |
803 | ||
804 | wake_up_all(&q->mq_freeze_wq); | |
805 | } | |
806 | ||
807 | static void blk_rq_timed_out_timer(unsigned long data) | |
808 | { | |
809 | struct request_queue *q = (struct request_queue *)data; | |
810 | ||
811 | kblockd_schedule_work(&q->timeout_work); | |
812 | } | |
813 | ||
814 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
815 | { | |
816 | struct request_queue *q; | |
817 | ||
818 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
819 | gfp_mask | __GFP_ZERO, node_id); | |
820 | if (!q) | |
821 | return NULL; | |
822 | ||
823 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); | |
824 | if (q->id < 0) | |
825 | goto fail_q; | |
826 | ||
827 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
828 | if (!q->bio_split) | |
829 | goto fail_id; | |
830 | ||
831 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); | |
832 | if (!q->backing_dev_info) | |
833 | goto fail_split; | |
834 | ||
835 | q->stats = blk_alloc_queue_stats(); | |
836 | if (!q->stats) | |
837 | goto fail_stats; | |
838 | ||
839 | q->backing_dev_info->ra_pages = | |
840 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; | |
841 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; | |
842 | q->backing_dev_info->name = "block"; | |
843 | q->node = node_id; | |
844 | ||
845 | setup_timer(&q->backing_dev_info->laptop_mode_wb_timer, | |
846 | laptop_mode_timer_fn, (unsigned long) q); | |
847 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); | |
848 | INIT_WORK(&q->timeout_work, NULL); | |
849 | INIT_LIST_HEAD(&q->queue_head); | |
850 | INIT_LIST_HEAD(&q->timeout_list); | |
851 | INIT_LIST_HEAD(&q->icq_list); | |
852 | #ifdef CONFIG_BLK_CGROUP | |
853 | INIT_LIST_HEAD(&q->blkg_list); | |
854 | #endif | |
855 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); | |
856 | ||
857 | kobject_init(&q->kobj, &blk_queue_ktype); | |
858 | ||
859 | #ifdef CONFIG_BLK_DEV_IO_TRACE | |
860 | mutex_init(&q->blk_trace_mutex); | |
861 | #endif | |
862 | mutex_init(&q->sysfs_lock); | |
863 | spin_lock_init(&q->__queue_lock); | |
864 | ||
865 | /* | |
866 | * By default initialize queue_lock to internal lock and driver can | |
867 | * override it later if need be. | |
868 | */ | |
869 | q->queue_lock = &q->__queue_lock; | |
870 | ||
871 | /* | |
872 | * A queue starts its life with bypass turned on to avoid | |
873 | * unnecessary bypass on/off overhead and nasty surprises during | |
874 | * init. The initial bypass will be finished when the queue is | |
875 | * registered by blk_register_queue(). | |
876 | */ | |
877 | q->bypass_depth = 1; | |
878 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
879 | ||
880 | init_waitqueue_head(&q->mq_freeze_wq); | |
881 | ||
882 | /* | |
883 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
884 | * See blk_register_queue() for details. | |
885 | */ | |
886 | if (percpu_ref_init(&q->q_usage_counter, | |
887 | blk_queue_usage_counter_release, | |
888 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
889 | goto fail_bdi; | |
890 | ||
891 | if (blkcg_init_queue(q)) | |
892 | goto fail_ref; | |
893 | ||
894 | return q; | |
895 | ||
896 | fail_ref: | |
897 | percpu_ref_exit(&q->q_usage_counter); | |
898 | fail_bdi: | |
899 | blk_free_queue_stats(q->stats); | |
900 | fail_stats: | |
901 | bdi_put(q->backing_dev_info); | |
902 | fail_split: | |
903 | bioset_free(q->bio_split); | |
904 | fail_id: | |
905 | ida_simple_remove(&blk_queue_ida, q->id); | |
906 | fail_q: | |
907 | kmem_cache_free(blk_requestq_cachep, q); | |
908 | return NULL; | |
909 | } | |
910 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
911 | ||
912 | /** | |
913 | * blk_init_queue - prepare a request queue for use with a block device | |
914 | * @rfn: The function to be called to process requests that have been | |
915 | * placed on the queue. | |
916 | * @lock: Request queue spin lock | |
917 | * | |
918 | * Description: | |
919 | * If a block device wishes to use the standard request handling procedures, | |
920 | * which sorts requests and coalesces adjacent requests, then it must | |
921 | * call blk_init_queue(). The function @rfn will be called when there | |
922 | * are requests on the queue that need to be processed. If the device | |
923 | * supports plugging, then @rfn may not be called immediately when requests | |
924 | * are available on the queue, but may be called at some time later instead. | |
925 | * Plugged queues are generally unplugged when a buffer belonging to one | |
926 | * of the requests on the queue is needed, or due to memory pressure. | |
927 | * | |
928 | * @rfn is not required, or even expected, to remove all requests off the | |
929 | * queue, but only as many as it can handle at a time. If it does leave | |
930 | * requests on the queue, it is responsible for arranging that the requests | |
931 | * get dealt with eventually. | |
932 | * | |
933 | * The queue spin lock must be held while manipulating the requests on the | |
934 | * request queue; this lock will be taken also from interrupt context, so irq | |
935 | * disabling is needed for it. | |
936 | * | |
937 | * Function returns a pointer to the initialized request queue, or %NULL if | |
938 | * it didn't succeed. | |
939 | * | |
940 | * Note: | |
941 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
942 | * when the block device is deactivated (such as at module unload). | |
943 | **/ | |
944 | ||
945 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
946 | { | |
947 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); | |
948 | } | |
949 | EXPORT_SYMBOL(blk_init_queue); | |
950 | ||
951 | struct request_queue * | |
952 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
953 | { | |
954 | struct request_queue *q; | |
955 | ||
956 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
957 | if (!q) | |
958 | return NULL; | |
959 | ||
960 | q->request_fn = rfn; | |
961 | if (lock) | |
962 | q->queue_lock = lock; | |
963 | if (blk_init_allocated_queue(q) < 0) { | |
964 | blk_cleanup_queue(q); | |
965 | return NULL; | |
966 | } | |
967 | ||
968 | return q; | |
969 | } | |
970 | EXPORT_SYMBOL(blk_init_queue_node); | |
971 | ||
972 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); | |
973 | ||
974 | ||
975 | int blk_init_allocated_queue(struct request_queue *q) | |
976 | { | |
977 | WARN_ON_ONCE(q->mq_ops); | |
978 | ||
979 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); | |
980 | if (!q->fq) | |
981 | return -ENOMEM; | |
982 | ||
983 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) | |
984 | goto out_free_flush_queue; | |
985 | ||
986 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) | |
987 | goto out_exit_flush_rq; | |
988 | ||
989 | INIT_WORK(&q->timeout_work, blk_timeout_work); | |
990 | q->queue_flags |= QUEUE_FLAG_DEFAULT; | |
991 | ||
992 | /* | |
993 | * This also sets hw/phys segments, boundary and size | |
994 | */ | |
995 | blk_queue_make_request(q, blk_queue_bio); | |
996 | ||
997 | q->sg_reserved_size = INT_MAX; | |
998 | ||
999 | /* Protect q->elevator from elevator_change */ | |
1000 | mutex_lock(&q->sysfs_lock); | |
1001 | ||
1002 | /* init elevator */ | |
1003 | if (elevator_init(q, NULL)) { | |
1004 | mutex_unlock(&q->sysfs_lock); | |
1005 | goto out_exit_flush_rq; | |
1006 | } | |
1007 | ||
1008 | mutex_unlock(&q->sysfs_lock); | |
1009 | return 0; | |
1010 | ||
1011 | out_exit_flush_rq: | |
1012 | if (q->exit_rq_fn) | |
1013 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1014 | out_free_flush_queue: | |
1015 | blk_free_flush_queue(q->fq); | |
1016 | return -ENOMEM; | |
1017 | } | |
1018 | EXPORT_SYMBOL(blk_init_allocated_queue); | |
1019 | ||
1020 | bool blk_get_queue(struct request_queue *q) | |
1021 | { | |
1022 | if (likely(!blk_queue_dying(q))) { | |
1023 | __blk_get_queue(q); | |
1024 | return true; | |
1025 | } | |
1026 | ||
1027 | return false; | |
1028 | } | |
1029 | EXPORT_SYMBOL(blk_get_queue); | |
1030 | ||
1031 | static inline void blk_free_request(struct request_list *rl, struct request *rq) | |
1032 | { | |
1033 | if (rq->rq_flags & RQF_ELVPRIV) { | |
1034 | elv_put_request(rl->q, rq); | |
1035 | if (rq->elv.icq) | |
1036 | put_io_context(rq->elv.icq->ioc); | |
1037 | } | |
1038 | ||
1039 | mempool_free(rq, rl->rq_pool); | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * ioc_batching returns true if the ioc is a valid batching request and | |
1044 | * should be given priority access to a request. | |
1045 | */ | |
1046 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
1047 | { | |
1048 | if (!ioc) | |
1049 | return 0; | |
1050 | ||
1051 | /* | |
1052 | * Make sure the process is able to allocate at least 1 request | |
1053 | * even if the batch times out, otherwise we could theoretically | |
1054 | * lose wakeups. | |
1055 | */ | |
1056 | return ioc->nr_batch_requests == q->nr_batching || | |
1057 | (ioc->nr_batch_requests > 0 | |
1058 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1063 | * will cause the process to be a "batcher" on all queues in the system. This | |
1064 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1065 | * a nice run. | |
1066 | */ | |
1067 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
1068 | { | |
1069 | if (!ioc || ioc_batching(q, ioc)) | |
1070 | return; | |
1071 | ||
1072 | ioc->nr_batch_requests = q->nr_batching; | |
1073 | ioc->last_waited = jiffies; | |
1074 | } | |
1075 | ||
1076 | static void __freed_request(struct request_list *rl, int sync) | |
1077 | { | |
1078 | struct request_queue *q = rl->q; | |
1079 | ||
1080 | if (rl->count[sync] < queue_congestion_off_threshold(q)) | |
1081 | blk_clear_congested(rl, sync); | |
1082 | ||
1083 | if (rl->count[sync] + 1 <= q->nr_requests) { | |
1084 | if (waitqueue_active(&rl->wait[sync])) | |
1085 | wake_up(&rl->wait[sync]); | |
1086 | ||
1087 | blk_clear_rl_full(rl, sync); | |
1088 | } | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * A request has just been released. Account for it, update the full and | |
1093 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1094 | */ | |
1095 | static void freed_request(struct request_list *rl, bool sync, | |
1096 | req_flags_t rq_flags) | |
1097 | { | |
1098 | struct request_queue *q = rl->q; | |
1099 | ||
1100 | q->nr_rqs[sync]--; | |
1101 | rl->count[sync]--; | |
1102 | if (rq_flags & RQF_ELVPRIV) | |
1103 | q->nr_rqs_elvpriv--; | |
1104 | ||
1105 | __freed_request(rl, sync); | |
1106 | ||
1107 | if (unlikely(rl->starved[sync ^ 1])) | |
1108 | __freed_request(rl, sync ^ 1); | |
1109 | } | |
1110 | ||
1111 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) | |
1112 | { | |
1113 | struct request_list *rl; | |
1114 | int on_thresh, off_thresh; | |
1115 | ||
1116 | WARN_ON_ONCE(q->mq_ops); | |
1117 | ||
1118 | spin_lock_irq(q->queue_lock); | |
1119 | q->nr_requests = nr; | |
1120 | blk_queue_congestion_threshold(q); | |
1121 | on_thresh = queue_congestion_on_threshold(q); | |
1122 | off_thresh = queue_congestion_off_threshold(q); | |
1123 | ||
1124 | blk_queue_for_each_rl(rl, q) { | |
1125 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1126 | blk_set_congested(rl, BLK_RW_SYNC); | |
1127 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1128 | blk_clear_congested(rl, BLK_RW_SYNC); | |
1129 | ||
1130 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) | |
1131 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1132 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1133 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
1134 | ||
1135 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { | |
1136 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1137 | } else { | |
1138 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1139 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1140 | } | |
1141 | ||
1142 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1143 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1144 | } else { | |
1145 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1146 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | spin_unlock_irq(q->queue_lock); | |
1151 | return 0; | |
1152 | } | |
1153 | ||
1154 | /** | |
1155 | * __get_request - get a free request | |
1156 | * @rl: request list to allocate from | |
1157 | * @op: operation and flags | |
1158 | * @bio: bio to allocate request for (can be %NULL) | |
1159 | * @gfp_mask: allocation mask | |
1160 | * | |
1161 | * Get a free request from @q. This function may fail under memory | |
1162 | * pressure or if @q is dead. | |
1163 | * | |
1164 | * Must be called with @q->queue_lock held and, | |
1165 | * Returns ERR_PTR on failure, with @q->queue_lock held. | |
1166 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1167 | */ | |
1168 | static struct request *__get_request(struct request_list *rl, unsigned int op, | |
1169 | struct bio *bio, gfp_t gfp_mask) | |
1170 | { | |
1171 | struct request_queue *q = rl->q; | |
1172 | struct request *rq; | |
1173 | struct elevator_type *et = q->elevator->type; | |
1174 | struct io_context *ioc = rq_ioc(bio); | |
1175 | struct io_cq *icq = NULL; | |
1176 | const bool is_sync = op_is_sync(op); | |
1177 | int may_queue; | |
1178 | req_flags_t rq_flags = RQF_ALLOCED; | |
1179 | ||
1180 | lockdep_assert_held(q->queue_lock); | |
1181 | ||
1182 | if (unlikely(blk_queue_dying(q))) | |
1183 | return ERR_PTR(-ENODEV); | |
1184 | ||
1185 | may_queue = elv_may_queue(q, op); | |
1186 | if (may_queue == ELV_MQUEUE_NO) | |
1187 | goto rq_starved; | |
1188 | ||
1189 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { | |
1190 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
1191 | /* | |
1192 | * The queue will fill after this allocation, so set | |
1193 | * it as full, and mark this process as "batching". | |
1194 | * This process will be allowed to complete a batch of | |
1195 | * requests, others will be blocked. | |
1196 | */ | |
1197 | if (!blk_rl_full(rl, is_sync)) { | |
1198 | ioc_set_batching(q, ioc); | |
1199 | blk_set_rl_full(rl, is_sync); | |
1200 | } else { | |
1201 | if (may_queue != ELV_MQUEUE_MUST | |
1202 | && !ioc_batching(q, ioc)) { | |
1203 | /* | |
1204 | * The queue is full and the allocating | |
1205 | * process is not a "batcher", and not | |
1206 | * exempted by the IO scheduler | |
1207 | */ | |
1208 | return ERR_PTR(-ENOMEM); | |
1209 | } | |
1210 | } | |
1211 | } | |
1212 | blk_set_congested(rl, is_sync); | |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * Only allow batching queuers to allocate up to 50% over the defined | |
1217 | * limit of requests, otherwise we could have thousands of requests | |
1218 | * allocated with any setting of ->nr_requests | |
1219 | */ | |
1220 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) | |
1221 | return ERR_PTR(-ENOMEM); | |
1222 | ||
1223 | q->nr_rqs[is_sync]++; | |
1224 | rl->count[is_sync]++; | |
1225 | rl->starved[is_sync] = 0; | |
1226 | ||
1227 | /* | |
1228 | * Decide whether the new request will be managed by elevator. If | |
1229 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will | |
1230 | * prevent the current elevator from being destroyed until the new | |
1231 | * request is freed. This guarantees icq's won't be destroyed and | |
1232 | * makes creating new ones safe. | |
1233 | * | |
1234 | * Flush requests do not use the elevator so skip initialization. | |
1235 | * This allows a request to share the flush and elevator data. | |
1236 | * | |
1237 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
1238 | * it will be created after releasing queue_lock. | |
1239 | */ | |
1240 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { | |
1241 | rq_flags |= RQF_ELVPRIV; | |
1242 | q->nr_rqs_elvpriv++; | |
1243 | if (et->icq_cache && ioc) | |
1244 | icq = ioc_lookup_icq(ioc, q); | |
1245 | } | |
1246 | ||
1247 | if (blk_queue_io_stat(q)) | |
1248 | rq_flags |= RQF_IO_STAT; | |
1249 | spin_unlock_irq(q->queue_lock); | |
1250 | ||
1251 | /* allocate and init request */ | |
1252 | rq = mempool_alloc(rl->rq_pool, gfp_mask); | |
1253 | if (!rq) | |
1254 | goto fail_alloc; | |
1255 | ||
1256 | blk_rq_init(q, rq); | |
1257 | blk_rq_set_rl(rq, rl); | |
1258 | rq->cmd_flags = op; | |
1259 | rq->rq_flags = rq_flags; | |
1260 | ||
1261 | /* init elvpriv */ | |
1262 | if (rq_flags & RQF_ELVPRIV) { | |
1263 | if (unlikely(et->icq_cache && !icq)) { | |
1264 | if (ioc) | |
1265 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
1266 | if (!icq) | |
1267 | goto fail_elvpriv; | |
1268 | } | |
1269 | ||
1270 | rq->elv.icq = icq; | |
1271 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1272 | goto fail_elvpriv; | |
1273 | ||
1274 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
1275 | if (icq) | |
1276 | get_io_context(icq->ioc); | |
1277 | } | |
1278 | out: | |
1279 | /* | |
1280 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1281 | * OK, if the queue is under the request limit then requests need | |
1282 | * not count toward the nr_batch_requests limit. There will always | |
1283 | * be some limit enforced by BLK_BATCH_TIME. | |
1284 | */ | |
1285 | if (ioc_batching(q, ioc)) | |
1286 | ioc->nr_batch_requests--; | |
1287 | ||
1288 | trace_block_getrq(q, bio, op); | |
1289 | return rq; | |
1290 | ||
1291 | fail_elvpriv: | |
1292 | /* | |
1293 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1294 | * and may fail indefinitely under memory pressure and thus | |
1295 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1296 | * disturb iosched and blkcg but weird is bettern than dead. | |
1297 | */ | |
1298 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", | |
1299 | __func__, dev_name(q->backing_dev_info->dev)); | |
1300 | ||
1301 | rq->rq_flags &= ~RQF_ELVPRIV; | |
1302 | rq->elv.icq = NULL; | |
1303 | ||
1304 | spin_lock_irq(q->queue_lock); | |
1305 | q->nr_rqs_elvpriv--; | |
1306 | spin_unlock_irq(q->queue_lock); | |
1307 | goto out; | |
1308 | ||
1309 | fail_alloc: | |
1310 | /* | |
1311 | * Allocation failed presumably due to memory. Undo anything we | |
1312 | * might have messed up. | |
1313 | * | |
1314 | * Allocating task should really be put onto the front of the wait | |
1315 | * queue, but this is pretty rare. | |
1316 | */ | |
1317 | spin_lock_irq(q->queue_lock); | |
1318 | freed_request(rl, is_sync, rq_flags); | |
1319 | ||
1320 | /* | |
1321 | * in the very unlikely event that allocation failed and no | |
1322 | * requests for this direction was pending, mark us starved so that | |
1323 | * freeing of a request in the other direction will notice | |
1324 | * us. another possible fix would be to split the rq mempool into | |
1325 | * READ and WRITE | |
1326 | */ | |
1327 | rq_starved: | |
1328 | if (unlikely(rl->count[is_sync] == 0)) | |
1329 | rl->starved[is_sync] = 1; | |
1330 | return ERR_PTR(-ENOMEM); | |
1331 | } | |
1332 | ||
1333 | /** | |
1334 | * get_request - get a free request | |
1335 | * @q: request_queue to allocate request from | |
1336 | * @op: operation and flags | |
1337 | * @bio: bio to allocate request for (can be %NULL) | |
1338 | * @gfp_mask: allocation mask | |
1339 | * | |
1340 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, | |
1341 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
1342 | * | |
1343 | * Must be called with @q->queue_lock held and, | |
1344 | * Returns ERR_PTR on failure, with @q->queue_lock held. | |
1345 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1346 | */ | |
1347 | static struct request *get_request(struct request_queue *q, unsigned int op, | |
1348 | struct bio *bio, gfp_t gfp_mask) | |
1349 | { | |
1350 | const bool is_sync = op_is_sync(op); | |
1351 | DEFINE_WAIT(wait); | |
1352 | struct request_list *rl; | |
1353 | struct request *rq; | |
1354 | ||
1355 | lockdep_assert_held(q->queue_lock); | |
1356 | WARN_ON_ONCE(q->mq_ops); | |
1357 | ||
1358 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
1359 | retry: | |
1360 | rq = __get_request(rl, op, bio, gfp_mask); | |
1361 | if (!IS_ERR(rq)) | |
1362 | return rq; | |
1363 | ||
1364 | if (op & REQ_NOWAIT) { | |
1365 | blk_put_rl(rl); | |
1366 | return ERR_PTR(-EAGAIN); | |
1367 | } | |
1368 | ||
1369 | if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) { | |
1370 | blk_put_rl(rl); | |
1371 | return rq; | |
1372 | } | |
1373 | ||
1374 | /* wait on @rl and retry */ | |
1375 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1376 | TASK_UNINTERRUPTIBLE); | |
1377 | ||
1378 | trace_block_sleeprq(q, bio, op); | |
1379 | ||
1380 | spin_unlock_irq(q->queue_lock); | |
1381 | io_schedule(); | |
1382 | ||
1383 | /* | |
1384 | * After sleeping, we become a "batching" process and will be able | |
1385 | * to allocate at least one request, and up to a big batch of them | |
1386 | * for a small period time. See ioc_batching, ioc_set_batching | |
1387 | */ | |
1388 | ioc_set_batching(q, current->io_context); | |
1389 | ||
1390 | spin_lock_irq(q->queue_lock); | |
1391 | finish_wait(&rl->wait[is_sync], &wait); | |
1392 | ||
1393 | goto retry; | |
1394 | } | |
1395 | ||
1396 | static struct request *blk_old_get_request(struct request_queue *q, | |
1397 | unsigned int op, gfp_t gfp_mask) | |
1398 | { | |
1399 | struct request *rq; | |
1400 | ||
1401 | WARN_ON_ONCE(q->mq_ops); | |
1402 | ||
1403 | /* create ioc upfront */ | |
1404 | create_io_context(gfp_mask, q->node); | |
1405 | ||
1406 | spin_lock_irq(q->queue_lock); | |
1407 | rq = get_request(q, op, NULL, gfp_mask); | |
1408 | if (IS_ERR(rq)) { | |
1409 | spin_unlock_irq(q->queue_lock); | |
1410 | return rq; | |
1411 | } | |
1412 | ||
1413 | /* q->queue_lock is unlocked at this point */ | |
1414 | rq->__data_len = 0; | |
1415 | rq->__sector = (sector_t) -1; | |
1416 | rq->bio = rq->biotail = NULL; | |
1417 | return rq; | |
1418 | } | |
1419 | ||
1420 | struct request *blk_get_request(struct request_queue *q, unsigned int op, | |
1421 | gfp_t gfp_mask) | |
1422 | { | |
1423 | struct request *req; | |
1424 | ||
1425 | if (q->mq_ops) { | |
1426 | req = blk_mq_alloc_request(q, op, | |
1427 | (gfp_mask & __GFP_DIRECT_RECLAIM) ? | |
1428 | 0 : BLK_MQ_REQ_NOWAIT); | |
1429 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
1430 | q->mq_ops->initialize_rq_fn(req); | |
1431 | } else { | |
1432 | req = blk_old_get_request(q, op, gfp_mask); | |
1433 | if (!IS_ERR(req) && q->initialize_rq_fn) | |
1434 | q->initialize_rq_fn(req); | |
1435 | } | |
1436 | ||
1437 | return req; | |
1438 | } | |
1439 | EXPORT_SYMBOL(blk_get_request); | |
1440 | ||
1441 | /** | |
1442 | * blk_requeue_request - put a request back on queue | |
1443 | * @q: request queue where request should be inserted | |
1444 | * @rq: request to be inserted | |
1445 | * | |
1446 | * Description: | |
1447 | * Drivers often keep queueing requests until the hardware cannot accept | |
1448 | * more, when that condition happens we need to put the request back | |
1449 | * on the queue. Must be called with queue lock held. | |
1450 | */ | |
1451 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
1452 | { | |
1453 | lockdep_assert_held(q->queue_lock); | |
1454 | WARN_ON_ONCE(q->mq_ops); | |
1455 | ||
1456 | blk_delete_timer(rq); | |
1457 | blk_clear_rq_complete(rq); | |
1458 | trace_block_rq_requeue(q, rq); | |
1459 | wbt_requeue(q->rq_wb, &rq->issue_stat); | |
1460 | ||
1461 | if (rq->rq_flags & RQF_QUEUED) | |
1462 | blk_queue_end_tag(q, rq); | |
1463 | ||
1464 | BUG_ON(blk_queued_rq(rq)); | |
1465 | ||
1466 | elv_requeue_request(q, rq); | |
1467 | } | |
1468 | EXPORT_SYMBOL(blk_requeue_request); | |
1469 | ||
1470 | static void add_acct_request(struct request_queue *q, struct request *rq, | |
1471 | int where) | |
1472 | { | |
1473 | blk_account_io_start(rq, true); | |
1474 | __elv_add_request(q, rq, where); | |
1475 | } | |
1476 | ||
1477 | static void part_round_stats_single(struct request_queue *q, int cpu, | |
1478 | struct hd_struct *part, unsigned long now, | |
1479 | unsigned int inflight) | |
1480 | { | |
1481 | if (inflight) { | |
1482 | __part_stat_add(cpu, part, time_in_queue, | |
1483 | inflight * (now - part->stamp)); | |
1484 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1485 | } | |
1486 | part->stamp = now; | |
1487 | } | |
1488 | ||
1489 | /** | |
1490 | * part_round_stats() - Round off the performance stats on a struct disk_stats. | |
1491 | * @q: target block queue | |
1492 | * @cpu: cpu number for stats access | |
1493 | * @part: target partition | |
1494 | * | |
1495 | * The average IO queue length and utilisation statistics are maintained | |
1496 | * by observing the current state of the queue length and the amount of | |
1497 | * time it has been in this state for. | |
1498 | * | |
1499 | * Normally, that accounting is done on IO completion, but that can result | |
1500 | * in more than a second's worth of IO being accounted for within any one | |
1501 | * second, leading to >100% utilisation. To deal with that, we call this | |
1502 | * function to do a round-off before returning the results when reading | |
1503 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1504 | * the current jiffies and restarts the counters again. | |
1505 | */ | |
1506 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) | |
1507 | { | |
1508 | struct hd_struct *part2 = NULL; | |
1509 | unsigned long now = jiffies; | |
1510 | unsigned int inflight[2]; | |
1511 | int stats = 0; | |
1512 | ||
1513 | if (part->stamp != now) | |
1514 | stats |= 1; | |
1515 | ||
1516 | if (part->partno) { | |
1517 | part2 = &part_to_disk(part)->part0; | |
1518 | if (part2->stamp != now) | |
1519 | stats |= 2; | |
1520 | } | |
1521 | ||
1522 | if (!stats) | |
1523 | return; | |
1524 | ||
1525 | part_in_flight(q, part, inflight); | |
1526 | ||
1527 | if (stats & 2) | |
1528 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1529 | if (stats & 1) | |
1530 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
1531 | } | |
1532 | EXPORT_SYMBOL_GPL(part_round_stats); | |
1533 | ||
1534 | #ifdef CONFIG_PM | |
1535 | static void blk_pm_put_request(struct request *rq) | |
1536 | { | |
1537 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) | |
1538 | pm_runtime_mark_last_busy(rq->q->dev); | |
1539 | } | |
1540 | #else | |
1541 | static inline void blk_pm_put_request(struct request *rq) {} | |
1542 | #endif | |
1543 | ||
1544 | void __blk_put_request(struct request_queue *q, struct request *req) | |
1545 | { | |
1546 | req_flags_t rq_flags = req->rq_flags; | |
1547 | ||
1548 | if (unlikely(!q)) | |
1549 | return; | |
1550 | ||
1551 | if (q->mq_ops) { | |
1552 | blk_mq_free_request(req); | |
1553 | return; | |
1554 | } | |
1555 | ||
1556 | lockdep_assert_held(q->queue_lock); | |
1557 | ||
1558 | blk_pm_put_request(req); | |
1559 | ||
1560 | elv_completed_request(q, req); | |
1561 | ||
1562 | /* this is a bio leak */ | |
1563 | WARN_ON(req->bio != NULL); | |
1564 | ||
1565 | wbt_done(q->rq_wb, &req->issue_stat); | |
1566 | ||
1567 | /* | |
1568 | * Request may not have originated from ll_rw_blk. if not, | |
1569 | * it didn't come out of our reserved rq pools | |
1570 | */ | |
1571 | if (rq_flags & RQF_ALLOCED) { | |
1572 | struct request_list *rl = blk_rq_rl(req); | |
1573 | bool sync = op_is_sync(req->cmd_flags); | |
1574 | ||
1575 | BUG_ON(!list_empty(&req->queuelist)); | |
1576 | BUG_ON(ELV_ON_HASH(req)); | |
1577 | ||
1578 | blk_free_request(rl, req); | |
1579 | freed_request(rl, sync, rq_flags); | |
1580 | blk_put_rl(rl); | |
1581 | } | |
1582 | } | |
1583 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
1584 | ||
1585 | void blk_put_request(struct request *req) | |
1586 | { | |
1587 | struct request_queue *q = req->q; | |
1588 | ||
1589 | if (q->mq_ops) | |
1590 | blk_mq_free_request(req); | |
1591 | else { | |
1592 | unsigned long flags; | |
1593 | ||
1594 | spin_lock_irqsave(q->queue_lock, flags); | |
1595 | __blk_put_request(q, req); | |
1596 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1597 | } | |
1598 | } | |
1599 | EXPORT_SYMBOL(blk_put_request); | |
1600 | ||
1601 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, | |
1602 | struct bio *bio) | |
1603 | { | |
1604 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
1605 | ||
1606 | if (!ll_back_merge_fn(q, req, bio)) | |
1607 | return false; | |
1608 | ||
1609 | trace_block_bio_backmerge(q, req, bio); | |
1610 | ||
1611 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1612 | blk_rq_set_mixed_merge(req); | |
1613 | ||
1614 | req->biotail->bi_next = bio; | |
1615 | req->biotail = bio; | |
1616 | req->__data_len += bio->bi_iter.bi_size; | |
1617 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1618 | ||
1619 | blk_account_io_start(req, false); | |
1620 | return true; | |
1621 | } | |
1622 | ||
1623 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, | |
1624 | struct bio *bio) | |
1625 | { | |
1626 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
1627 | ||
1628 | if (!ll_front_merge_fn(q, req, bio)) | |
1629 | return false; | |
1630 | ||
1631 | trace_block_bio_frontmerge(q, req, bio); | |
1632 | ||
1633 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1634 | blk_rq_set_mixed_merge(req); | |
1635 | ||
1636 | bio->bi_next = req->bio; | |
1637 | req->bio = bio; | |
1638 | ||
1639 | req->__sector = bio->bi_iter.bi_sector; | |
1640 | req->__data_len += bio->bi_iter.bi_size; | |
1641 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1642 | ||
1643 | blk_account_io_start(req, false); | |
1644 | return true; | |
1645 | } | |
1646 | ||
1647 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, | |
1648 | struct bio *bio) | |
1649 | { | |
1650 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1651 | ||
1652 | if (segments >= queue_max_discard_segments(q)) | |
1653 | goto no_merge; | |
1654 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1655 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1656 | goto no_merge; | |
1657 | ||
1658 | req->biotail->bi_next = bio; | |
1659 | req->biotail = bio; | |
1660 | req->__data_len += bio->bi_iter.bi_size; | |
1661 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1662 | req->nr_phys_segments = segments + 1; | |
1663 | ||
1664 | blk_account_io_start(req, false); | |
1665 | return true; | |
1666 | no_merge: | |
1667 | req_set_nomerge(q, req); | |
1668 | return false; | |
1669 | } | |
1670 | ||
1671 | /** | |
1672 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1673 | * @q: request_queue new bio is being queued at | |
1674 | * @bio: new bio being queued | |
1675 | * @request_count: out parameter for number of traversed plugged requests | |
1676 | * @same_queue_rq: pointer to &struct request that gets filled in when | |
1677 | * another request associated with @q is found on the plug list | |
1678 | * (optional, may be %NULL) | |
1679 | * | |
1680 | * Determine whether @bio being queued on @q can be merged with a request | |
1681 | * on %current's plugged list. Returns %true if merge was successful, | |
1682 | * otherwise %false. | |
1683 | * | |
1684 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1685 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1686 | * than scheduling, and the request, while may have elvpriv data, is not | |
1687 | * added on the elevator at this point. In addition, we don't have | |
1688 | * reliable access to the elevator outside queue lock. Only check basic | |
1689 | * merging parameters without querying the elevator. | |
1690 | * | |
1691 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1692 | */ | |
1693 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
1694 | unsigned int *request_count, | |
1695 | struct request **same_queue_rq) | |
1696 | { | |
1697 | struct blk_plug *plug; | |
1698 | struct request *rq; | |
1699 | struct list_head *plug_list; | |
1700 | ||
1701 | plug = current->plug; | |
1702 | if (!plug) | |
1703 | return false; | |
1704 | *request_count = 0; | |
1705 | ||
1706 | if (q->mq_ops) | |
1707 | plug_list = &plug->mq_list; | |
1708 | else | |
1709 | plug_list = &plug->list; | |
1710 | ||
1711 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
1712 | bool merged = false; | |
1713 | ||
1714 | if (rq->q == q) { | |
1715 | (*request_count)++; | |
1716 | /* | |
1717 | * Only blk-mq multiple hardware queues case checks the | |
1718 | * rq in the same queue, there should be only one such | |
1719 | * rq in a queue | |
1720 | **/ | |
1721 | if (same_queue_rq) | |
1722 | *same_queue_rq = rq; | |
1723 | } | |
1724 | ||
1725 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) | |
1726 | continue; | |
1727 | ||
1728 | switch (blk_try_merge(rq, bio)) { | |
1729 | case ELEVATOR_BACK_MERGE: | |
1730 | merged = bio_attempt_back_merge(q, rq, bio); | |
1731 | break; | |
1732 | case ELEVATOR_FRONT_MERGE: | |
1733 | merged = bio_attempt_front_merge(q, rq, bio); | |
1734 | break; | |
1735 | case ELEVATOR_DISCARD_MERGE: | |
1736 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1737 | break; | |
1738 | default: | |
1739 | break; | |
1740 | } | |
1741 | ||
1742 | if (merged) | |
1743 | return true; | |
1744 | } | |
1745 | ||
1746 | return false; | |
1747 | } | |
1748 | ||
1749 | unsigned int blk_plug_queued_count(struct request_queue *q) | |
1750 | { | |
1751 | struct blk_plug *plug; | |
1752 | struct request *rq; | |
1753 | struct list_head *plug_list; | |
1754 | unsigned int ret = 0; | |
1755 | ||
1756 | plug = current->plug; | |
1757 | if (!plug) | |
1758 | goto out; | |
1759 | ||
1760 | if (q->mq_ops) | |
1761 | plug_list = &plug->mq_list; | |
1762 | else | |
1763 | plug_list = &plug->list; | |
1764 | ||
1765 | list_for_each_entry(rq, plug_list, queuelist) { | |
1766 | if (rq->q == q) | |
1767 | ret++; | |
1768 | } | |
1769 | out: | |
1770 | return ret; | |
1771 | } | |
1772 | ||
1773 | void blk_init_request_from_bio(struct request *req, struct bio *bio) | |
1774 | { | |
1775 | struct io_context *ioc = rq_ioc(bio); | |
1776 | ||
1777 | if (bio->bi_opf & REQ_RAHEAD) | |
1778 | req->cmd_flags |= REQ_FAILFAST_MASK; | |
1779 | ||
1780 | req->__sector = bio->bi_iter.bi_sector; | |
1781 | if (ioprio_valid(bio_prio(bio))) | |
1782 | req->ioprio = bio_prio(bio); | |
1783 | else if (ioc) | |
1784 | req->ioprio = ioc->ioprio; | |
1785 | else | |
1786 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
1787 | req->write_hint = bio->bi_write_hint; | |
1788 | blk_rq_bio_prep(req->q, req, bio); | |
1789 | } | |
1790 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); | |
1791 | ||
1792 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) | |
1793 | { | |
1794 | struct blk_plug *plug; | |
1795 | int where = ELEVATOR_INSERT_SORT; | |
1796 | struct request *req, *free; | |
1797 | unsigned int request_count = 0; | |
1798 | unsigned int wb_acct; | |
1799 | ||
1800 | /* | |
1801 | * low level driver can indicate that it wants pages above a | |
1802 | * certain limit bounced to low memory (ie for highmem, or even | |
1803 | * ISA dma in theory) | |
1804 | */ | |
1805 | blk_queue_bounce(q, &bio); | |
1806 | ||
1807 | blk_queue_split(q, &bio); | |
1808 | ||
1809 | if (!bio_integrity_prep(bio)) | |
1810 | return BLK_QC_T_NONE; | |
1811 | ||
1812 | if (op_is_flush(bio->bi_opf)) { | |
1813 | spin_lock_irq(q->queue_lock); | |
1814 | where = ELEVATOR_INSERT_FLUSH; | |
1815 | goto get_rq; | |
1816 | } | |
1817 | ||
1818 | /* | |
1819 | * Check if we can merge with the plugged list before grabbing | |
1820 | * any locks. | |
1821 | */ | |
1822 | if (!blk_queue_nomerges(q)) { | |
1823 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
1824 | return BLK_QC_T_NONE; | |
1825 | } else | |
1826 | request_count = blk_plug_queued_count(q); | |
1827 | ||
1828 | spin_lock_irq(q->queue_lock); | |
1829 | ||
1830 | switch (elv_merge(q, &req, bio)) { | |
1831 | case ELEVATOR_BACK_MERGE: | |
1832 | if (!bio_attempt_back_merge(q, req, bio)) | |
1833 | break; | |
1834 | elv_bio_merged(q, req, bio); | |
1835 | free = attempt_back_merge(q, req); | |
1836 | if (free) | |
1837 | __blk_put_request(q, free); | |
1838 | else | |
1839 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1840 | goto out_unlock; | |
1841 | case ELEVATOR_FRONT_MERGE: | |
1842 | if (!bio_attempt_front_merge(q, req, bio)) | |
1843 | break; | |
1844 | elv_bio_merged(q, req, bio); | |
1845 | free = attempt_front_merge(q, req); | |
1846 | if (free) | |
1847 | __blk_put_request(q, free); | |
1848 | else | |
1849 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1850 | goto out_unlock; | |
1851 | default: | |
1852 | break; | |
1853 | } | |
1854 | ||
1855 | get_rq: | |
1856 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); | |
1857 | ||
1858 | /* | |
1859 | * Grab a free request. This is might sleep but can not fail. | |
1860 | * Returns with the queue unlocked. | |
1861 | */ | |
1862 | req = get_request(q, bio->bi_opf, bio, GFP_NOIO); | |
1863 | if (IS_ERR(req)) { | |
1864 | __wbt_done(q->rq_wb, wb_acct); | |
1865 | if (PTR_ERR(req) == -ENOMEM) | |
1866 | bio->bi_status = BLK_STS_RESOURCE; | |
1867 | else | |
1868 | bio->bi_status = BLK_STS_IOERR; | |
1869 | bio_endio(bio); | |
1870 | goto out_unlock; | |
1871 | } | |
1872 | ||
1873 | wbt_track(&req->issue_stat, wb_acct); | |
1874 | ||
1875 | /* | |
1876 | * After dropping the lock and possibly sleeping here, our request | |
1877 | * may now be mergeable after it had proven unmergeable (above). | |
1878 | * We don't worry about that case for efficiency. It won't happen | |
1879 | * often, and the elevators are able to handle it. | |
1880 | */ | |
1881 | blk_init_request_from_bio(req, bio); | |
1882 | ||
1883 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) | |
1884 | req->cpu = raw_smp_processor_id(); | |
1885 | ||
1886 | plug = current->plug; | |
1887 | if (plug) { | |
1888 | /* | |
1889 | * If this is the first request added after a plug, fire | |
1890 | * of a plug trace. | |
1891 | * | |
1892 | * @request_count may become stale because of schedule | |
1893 | * out, so check plug list again. | |
1894 | */ | |
1895 | if (!request_count || list_empty(&plug->list)) | |
1896 | trace_block_plug(q); | |
1897 | else { | |
1898 | struct request *last = list_entry_rq(plug->list.prev); | |
1899 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1900 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
1901 | blk_flush_plug_list(plug, false); | |
1902 | trace_block_plug(q); | |
1903 | } | |
1904 | } | |
1905 | list_add_tail(&req->queuelist, &plug->list); | |
1906 | blk_account_io_start(req, true); | |
1907 | } else { | |
1908 | spin_lock_irq(q->queue_lock); | |
1909 | add_acct_request(q, req, where); | |
1910 | __blk_run_queue(q); | |
1911 | out_unlock: | |
1912 | spin_unlock_irq(q->queue_lock); | |
1913 | } | |
1914 | ||
1915 | return BLK_QC_T_NONE; | |
1916 | } | |
1917 | ||
1918 | static void handle_bad_sector(struct bio *bio) | |
1919 | { | |
1920 | char b[BDEVNAME_SIZE]; | |
1921 | ||
1922 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1923 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", | |
1924 | bio_devname(bio, b), bio->bi_opf, | |
1925 | (unsigned long long)bio_end_sector(bio), | |
1926 | (long long)get_capacity(bio->bi_disk)); | |
1927 | } | |
1928 | ||
1929 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1930 | ||
1931 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1932 | ||
1933 | static int __init setup_fail_make_request(char *str) | |
1934 | { | |
1935 | return setup_fault_attr(&fail_make_request, str); | |
1936 | } | |
1937 | __setup("fail_make_request=", setup_fail_make_request); | |
1938 | ||
1939 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) | |
1940 | { | |
1941 | return part->make_it_fail && should_fail(&fail_make_request, bytes); | |
1942 | } | |
1943 | ||
1944 | static int __init fail_make_request_debugfs(void) | |
1945 | { | |
1946 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", | |
1947 | NULL, &fail_make_request); | |
1948 | ||
1949 | return PTR_ERR_OR_ZERO(dir); | |
1950 | } | |
1951 | ||
1952 | late_initcall(fail_make_request_debugfs); | |
1953 | ||
1954 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1955 | ||
1956 | static inline bool should_fail_request(struct hd_struct *part, | |
1957 | unsigned int bytes) | |
1958 | { | |
1959 | return false; | |
1960 | } | |
1961 | ||
1962 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1963 | ||
1964 | /* | |
1965 | * Remap block n of partition p to block n+start(p) of the disk. | |
1966 | */ | |
1967 | static inline int blk_partition_remap(struct bio *bio) | |
1968 | { | |
1969 | struct hd_struct *p; | |
1970 | int ret = 0; | |
1971 | ||
1972 | /* | |
1973 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
1974 | * Include a test for the reset op code and perform the remap if needed. | |
1975 | */ | |
1976 | if (!bio->bi_partno || | |
1977 | (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET)) | |
1978 | return 0; | |
1979 | ||
1980 | rcu_read_lock(); | |
1981 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
1982 | if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) { | |
1983 | bio->bi_iter.bi_sector += p->start_sect; | |
1984 | bio->bi_partno = 0; | |
1985 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), | |
1986 | bio->bi_iter.bi_sector - p->start_sect); | |
1987 | } else { | |
1988 | printk("%s: fail for partition %d\n", __func__, bio->bi_partno); | |
1989 | ret = -EIO; | |
1990 | } | |
1991 | rcu_read_unlock(); | |
1992 | ||
1993 | return ret; | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * Check whether this bio extends beyond the end of the device. | |
1998 | */ | |
1999 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
2000 | { | |
2001 | sector_t maxsector; | |
2002 | ||
2003 | if (!nr_sectors) | |
2004 | return 0; | |
2005 | ||
2006 | /* Test device or partition size, when known. */ | |
2007 | maxsector = get_capacity(bio->bi_disk); | |
2008 | if (maxsector) { | |
2009 | sector_t sector = bio->bi_iter.bi_sector; | |
2010 | ||
2011 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2012 | /* | |
2013 | * This may well happen - the kernel calls bread() | |
2014 | * without checking the size of the device, e.g., when | |
2015 | * mounting a device. | |
2016 | */ | |
2017 | handle_bad_sector(bio); | |
2018 | return 1; | |
2019 | } | |
2020 | } | |
2021 | ||
2022 | return 0; | |
2023 | } | |
2024 | ||
2025 | static noinline_for_stack bool | |
2026 | generic_make_request_checks(struct bio *bio) | |
2027 | { | |
2028 | struct request_queue *q; | |
2029 | int nr_sectors = bio_sectors(bio); | |
2030 | blk_status_t status = BLK_STS_IOERR; | |
2031 | char b[BDEVNAME_SIZE]; | |
2032 | ||
2033 | might_sleep(); | |
2034 | ||
2035 | if (bio_check_eod(bio, nr_sectors)) | |
2036 | goto end_io; | |
2037 | ||
2038 | q = bio->bi_disk->queue; | |
2039 | if (unlikely(!q)) { | |
2040 | printk(KERN_ERR | |
2041 | "generic_make_request: Trying to access " | |
2042 | "nonexistent block-device %s (%Lu)\n", | |
2043 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); | |
2044 | goto end_io; | |
2045 | } | |
2046 | ||
2047 | /* | |
2048 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2049 | * if queue is not a request based queue. | |
2050 | */ | |
2051 | ||
2052 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) | |
2053 | goto not_supported; | |
2054 | ||
2055 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
2056 | goto end_io; | |
2057 | ||
2058 | if (blk_partition_remap(bio)) | |
2059 | goto end_io; | |
2060 | ||
2061 | if (bio_check_eod(bio, nr_sectors)) | |
2062 | goto end_io; | |
2063 | ||
2064 | /* | |
2065 | * Filter flush bio's early so that make_request based | |
2066 | * drivers without flush support don't have to worry | |
2067 | * about them. | |
2068 | */ | |
2069 | if (op_is_flush(bio->bi_opf) && | |
2070 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { | |
2071 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); | |
2072 | if (!nr_sectors) { | |
2073 | status = BLK_STS_OK; | |
2074 | goto end_io; | |
2075 | } | |
2076 | } | |
2077 | ||
2078 | switch (bio_op(bio)) { | |
2079 | case REQ_OP_DISCARD: | |
2080 | if (!blk_queue_discard(q)) | |
2081 | goto not_supported; | |
2082 | break; | |
2083 | case REQ_OP_SECURE_ERASE: | |
2084 | if (!blk_queue_secure_erase(q)) | |
2085 | goto not_supported; | |
2086 | break; | |
2087 | case REQ_OP_WRITE_SAME: | |
2088 | if (!q->limits.max_write_same_sectors) | |
2089 | goto not_supported; | |
2090 | break; | |
2091 | case REQ_OP_ZONE_REPORT: | |
2092 | case REQ_OP_ZONE_RESET: | |
2093 | if (!blk_queue_is_zoned(q)) | |
2094 | goto not_supported; | |
2095 | break; | |
2096 | case REQ_OP_WRITE_ZEROES: | |
2097 | if (!q->limits.max_write_zeroes_sectors) | |
2098 | goto not_supported; | |
2099 | break; | |
2100 | default: | |
2101 | break; | |
2102 | } | |
2103 | ||
2104 | /* | |
2105 | * Various block parts want %current->io_context and lazy ioc | |
2106 | * allocation ends up trading a lot of pain for a small amount of | |
2107 | * memory. Just allocate it upfront. This may fail and block | |
2108 | * layer knows how to live with it. | |
2109 | */ | |
2110 | create_io_context(GFP_ATOMIC, q->node); | |
2111 | ||
2112 | if (!blkcg_bio_issue_check(q, bio)) | |
2113 | return false; | |
2114 | ||
2115 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { | |
2116 | trace_block_bio_queue(q, bio); | |
2117 | /* Now that enqueuing has been traced, we need to trace | |
2118 | * completion as well. | |
2119 | */ | |
2120 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2121 | } | |
2122 | return true; | |
2123 | ||
2124 | not_supported: | |
2125 | status = BLK_STS_NOTSUPP; | |
2126 | end_io: | |
2127 | bio->bi_status = status; | |
2128 | bio_endio(bio); | |
2129 | return false; | |
2130 | } | |
2131 | ||
2132 | /** | |
2133 | * generic_make_request - hand a buffer to its device driver for I/O | |
2134 | * @bio: The bio describing the location in memory and on the device. | |
2135 | * | |
2136 | * generic_make_request() is used to make I/O requests of block | |
2137 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2138 | * to be done. | |
2139 | * | |
2140 | * generic_make_request() does not return any status. The | |
2141 | * success/failure status of the request, along with notification of | |
2142 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2143 | * function described (one day) else where. | |
2144 | * | |
2145 | * The caller of generic_make_request must make sure that bi_io_vec | |
2146 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2147 | * set to describe the device address, and the | |
2148 | * bi_end_io and optionally bi_private are set to describe how | |
2149 | * completion notification should be signaled. | |
2150 | * | |
2151 | * generic_make_request and the drivers it calls may use bi_next if this | |
2152 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2153 | * a lower device by calling into generic_make_request recursively, which | |
2154 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
2155 | */ | |
2156 | blk_qc_t generic_make_request(struct bio *bio) | |
2157 | { | |
2158 | /* | |
2159 | * bio_list_on_stack[0] contains bios submitted by the current | |
2160 | * make_request_fn. | |
2161 | * bio_list_on_stack[1] contains bios that were submitted before | |
2162 | * the current make_request_fn, but that haven't been processed | |
2163 | * yet. | |
2164 | */ | |
2165 | struct bio_list bio_list_on_stack[2]; | |
2166 | blk_qc_t ret = BLK_QC_T_NONE; | |
2167 | ||
2168 | if (!generic_make_request_checks(bio)) | |
2169 | goto out; | |
2170 | ||
2171 | /* | |
2172 | * We only want one ->make_request_fn to be active at a time, else | |
2173 | * stack usage with stacked devices could be a problem. So use | |
2174 | * current->bio_list to keep a list of requests submited by a | |
2175 | * make_request_fn function. current->bio_list is also used as a | |
2176 | * flag to say if generic_make_request is currently active in this | |
2177 | * task or not. If it is NULL, then no make_request is active. If | |
2178 | * it is non-NULL, then a make_request is active, and new requests | |
2179 | * should be added at the tail | |
2180 | */ | |
2181 | if (current->bio_list) { | |
2182 | bio_list_add(¤t->bio_list[0], bio); | |
2183 | goto out; | |
2184 | } | |
2185 | ||
2186 | /* following loop may be a bit non-obvious, and so deserves some | |
2187 | * explanation. | |
2188 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2189 | * ensure that) so we have a list with a single bio. | |
2190 | * We pretend that we have just taken it off a longer list, so | |
2191 | * we assign bio_list to a pointer to the bio_list_on_stack, | |
2192 | * thus initialising the bio_list of new bios to be | |
2193 | * added. ->make_request() may indeed add some more bios | |
2194 | * through a recursive call to generic_make_request. If it | |
2195 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2196 | * from the top. In this case we really did just take the bio | |
2197 | * of the top of the list (no pretending) and so remove it from | |
2198 | * bio_list, and call into ->make_request() again. | |
2199 | */ | |
2200 | BUG_ON(bio->bi_next); | |
2201 | bio_list_init(&bio_list_on_stack[0]); | |
2202 | current->bio_list = bio_list_on_stack; | |
2203 | do { | |
2204 | struct request_queue *q = bio->bi_disk->queue; | |
2205 | ||
2206 | if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) { | |
2207 | struct bio_list lower, same; | |
2208 | ||
2209 | /* Create a fresh bio_list for all subordinate requests */ | |
2210 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
2211 | bio_list_init(&bio_list_on_stack[0]); | |
2212 | ret = q->make_request_fn(q, bio); | |
2213 | ||
2214 | blk_queue_exit(q); | |
2215 | ||
2216 | /* sort new bios into those for a lower level | |
2217 | * and those for the same level | |
2218 | */ | |
2219 | bio_list_init(&lower); | |
2220 | bio_list_init(&same); | |
2221 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
2222 | if (q == bio->bi_disk->queue) | |
2223 | bio_list_add(&same, bio); | |
2224 | else | |
2225 | bio_list_add(&lower, bio); | |
2226 | /* now assemble so we handle the lowest level first */ | |
2227 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
2228 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2229 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
2230 | } else { | |
2231 | if (unlikely(!blk_queue_dying(q) && | |
2232 | (bio->bi_opf & REQ_NOWAIT))) | |
2233 | bio_wouldblock_error(bio); | |
2234 | else | |
2235 | bio_io_error(bio); | |
2236 | } | |
2237 | bio = bio_list_pop(&bio_list_on_stack[0]); | |
2238 | } while (bio); | |
2239 | current->bio_list = NULL; /* deactivate */ | |
2240 | ||
2241 | out: | |
2242 | return ret; | |
2243 | } | |
2244 | EXPORT_SYMBOL(generic_make_request); | |
2245 | ||
2246 | /** | |
2247 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
2248 | * @bio: The bio describing the location in memory and on the device. | |
2249 | * | |
2250 | * This function behaves like generic_make_request(), but does not protect | |
2251 | * against recursion. Must only be used if the called driver is known | |
2252 | * to not call generic_make_request (or direct_make_request) again from | |
2253 | * its make_request function. (Calling direct_make_request again from | |
2254 | * a workqueue is perfectly fine as that doesn't recurse). | |
2255 | */ | |
2256 | blk_qc_t direct_make_request(struct bio *bio) | |
2257 | { | |
2258 | struct request_queue *q = bio->bi_disk->queue; | |
2259 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
2260 | blk_qc_t ret; | |
2261 | ||
2262 | if (!generic_make_request_checks(bio)) | |
2263 | return BLK_QC_T_NONE; | |
2264 | ||
2265 | if (unlikely(blk_queue_enter(q, nowait))) { | |
2266 | if (nowait && !blk_queue_dying(q)) | |
2267 | bio->bi_status = BLK_STS_AGAIN; | |
2268 | else | |
2269 | bio->bi_status = BLK_STS_IOERR; | |
2270 | bio_endio(bio); | |
2271 | return BLK_QC_T_NONE; | |
2272 | } | |
2273 | ||
2274 | ret = q->make_request_fn(q, bio); | |
2275 | blk_queue_exit(q); | |
2276 | return ret; | |
2277 | } | |
2278 | EXPORT_SYMBOL_GPL(direct_make_request); | |
2279 | ||
2280 | /** | |
2281 | * submit_bio - submit a bio to the block device layer for I/O | |
2282 | * @bio: The &struct bio which describes the I/O | |
2283 | * | |
2284 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2285 | * uses that function to do most of the work. Both are fairly rough | |
2286 | * interfaces; @bio must be presetup and ready for I/O. | |
2287 | * | |
2288 | */ | |
2289 | blk_qc_t submit_bio(struct bio *bio) | |
2290 | { | |
2291 | /* | |
2292 | * If it's a regular read/write or a barrier with data attached, | |
2293 | * go through the normal accounting stuff before submission. | |
2294 | */ | |
2295 | if (bio_has_data(bio)) { | |
2296 | unsigned int count; | |
2297 | ||
2298 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) | |
2299 | count = queue_logical_block_size(bio->bi_disk->queue); | |
2300 | else | |
2301 | count = bio_sectors(bio); | |
2302 | ||
2303 | if (op_is_write(bio_op(bio))) { | |
2304 | count_vm_events(PGPGOUT, count); | |
2305 | } else { | |
2306 | task_io_account_read(bio->bi_iter.bi_size); | |
2307 | count_vm_events(PGPGIN, count); | |
2308 | } | |
2309 | ||
2310 | if (unlikely(block_dump)) { | |
2311 | char b[BDEVNAME_SIZE]; | |
2312 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", | |
2313 | current->comm, task_pid_nr(current), | |
2314 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", | |
2315 | (unsigned long long)bio->bi_iter.bi_sector, | |
2316 | bio_devname(bio, b), count); | |
2317 | } | |
2318 | } | |
2319 | ||
2320 | return generic_make_request(bio); | |
2321 | } | |
2322 | EXPORT_SYMBOL(submit_bio); | |
2323 | ||
2324 | bool blk_poll(struct request_queue *q, blk_qc_t cookie) | |
2325 | { | |
2326 | if (!q->poll_fn || !blk_qc_t_valid(cookie)) | |
2327 | return false; | |
2328 | ||
2329 | if (current->plug) | |
2330 | blk_flush_plug_list(current->plug, false); | |
2331 | return q->poll_fn(q, cookie); | |
2332 | } | |
2333 | EXPORT_SYMBOL_GPL(blk_poll); | |
2334 | ||
2335 | /** | |
2336 | * blk_cloned_rq_check_limits - Helper function to check a cloned request | |
2337 | * for new the queue limits | |
2338 | * @q: the queue | |
2339 | * @rq: the request being checked | |
2340 | * | |
2341 | * Description: | |
2342 | * @rq may have been made based on weaker limitations of upper-level queues | |
2343 | * in request stacking drivers, and it may violate the limitation of @q. | |
2344 | * Since the block layer and the underlying device driver trust @rq | |
2345 | * after it is inserted to @q, it should be checked against @q before | |
2346 | * the insertion using this generic function. | |
2347 | * | |
2348 | * Request stacking drivers like request-based dm may change the queue | |
2349 | * limits when retrying requests on other queues. Those requests need | |
2350 | * to be checked against the new queue limits again during dispatch. | |
2351 | */ | |
2352 | static int blk_cloned_rq_check_limits(struct request_queue *q, | |
2353 | struct request *rq) | |
2354 | { | |
2355 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { | |
2356 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
2357 | return -EIO; | |
2358 | } | |
2359 | ||
2360 | /* | |
2361 | * queue's settings related to segment counting like q->bounce_pfn | |
2362 | * may differ from that of other stacking queues. | |
2363 | * Recalculate it to check the request correctly on this queue's | |
2364 | * limitation. | |
2365 | */ | |
2366 | blk_recalc_rq_segments(rq); | |
2367 | if (rq->nr_phys_segments > queue_max_segments(q)) { | |
2368 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
2369 | return -EIO; | |
2370 | } | |
2371 | ||
2372 | return 0; | |
2373 | } | |
2374 | ||
2375 | /** | |
2376 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2377 | * @q: the queue to submit the request | |
2378 | * @rq: the request being queued | |
2379 | */ | |
2380 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
2381 | { | |
2382 | unsigned long flags; | |
2383 | int where = ELEVATOR_INSERT_BACK; | |
2384 | ||
2385 | if (blk_cloned_rq_check_limits(q, rq)) | |
2386 | return BLK_STS_IOERR; | |
2387 | ||
2388 | if (rq->rq_disk && | |
2389 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2390 | return BLK_STS_IOERR; | |
2391 | ||
2392 | if (q->mq_ops) { | |
2393 | if (blk_queue_io_stat(q)) | |
2394 | blk_account_io_start(rq, true); | |
2395 | /* | |
2396 | * Since we have a scheduler attached on the top device, | |
2397 | * bypass a potential scheduler on the bottom device for | |
2398 | * insert. | |
2399 | */ | |
2400 | blk_mq_request_bypass_insert(rq, true); | |
2401 | return BLK_STS_OK; | |
2402 | } | |
2403 | ||
2404 | spin_lock_irqsave(q->queue_lock, flags); | |
2405 | if (unlikely(blk_queue_dying(q))) { | |
2406 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2407 | return BLK_STS_IOERR; | |
2408 | } | |
2409 | ||
2410 | /* | |
2411 | * Submitting request must be dequeued before calling this function | |
2412 | * because it will be linked to another request_queue | |
2413 | */ | |
2414 | BUG_ON(blk_queued_rq(rq)); | |
2415 | ||
2416 | if (op_is_flush(rq->cmd_flags)) | |
2417 | where = ELEVATOR_INSERT_FLUSH; | |
2418 | ||
2419 | add_acct_request(q, rq, where); | |
2420 | if (where == ELEVATOR_INSERT_FLUSH) | |
2421 | __blk_run_queue(q); | |
2422 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2423 | ||
2424 | return BLK_STS_OK; | |
2425 | } | |
2426 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2427 | ||
2428 | /** | |
2429 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2430 | * @rq: request to examine | |
2431 | * | |
2432 | * Description: | |
2433 | * A request could be merge of IOs which require different failure | |
2434 | * handling. This function determines the number of bytes which | |
2435 | * can be failed from the beginning of the request without | |
2436 | * crossing into area which need to be retried further. | |
2437 | * | |
2438 | * Return: | |
2439 | * The number of bytes to fail. | |
2440 | */ | |
2441 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2442 | { | |
2443 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2444 | unsigned int bytes = 0; | |
2445 | struct bio *bio; | |
2446 | ||
2447 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) | |
2448 | return blk_rq_bytes(rq); | |
2449 | ||
2450 | /* | |
2451 | * Currently the only 'mixing' which can happen is between | |
2452 | * different fastfail types. We can safely fail portions | |
2453 | * which have all the failfast bits that the first one has - | |
2454 | * the ones which are at least as eager to fail as the first | |
2455 | * one. | |
2456 | */ | |
2457 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
2458 | if ((bio->bi_opf & ff) != ff) | |
2459 | break; | |
2460 | bytes += bio->bi_iter.bi_size; | |
2461 | } | |
2462 | ||
2463 | /* this could lead to infinite loop */ | |
2464 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2465 | return bytes; | |
2466 | } | |
2467 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2468 | ||
2469 | void blk_account_io_completion(struct request *req, unsigned int bytes) | |
2470 | { | |
2471 | if (blk_do_io_stat(req)) { | |
2472 | const int rw = rq_data_dir(req); | |
2473 | struct hd_struct *part; | |
2474 | int cpu; | |
2475 | ||
2476 | cpu = part_stat_lock(); | |
2477 | part = req->part; | |
2478 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
2479 | part_stat_unlock(); | |
2480 | } | |
2481 | } | |
2482 | ||
2483 | void blk_account_io_done(struct request *req) | |
2484 | { | |
2485 | /* | |
2486 | * Account IO completion. flush_rq isn't accounted as a | |
2487 | * normal IO on queueing nor completion. Accounting the | |
2488 | * containing request is enough. | |
2489 | */ | |
2490 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { | |
2491 | unsigned long duration = jiffies - req->start_time; | |
2492 | const int rw = rq_data_dir(req); | |
2493 | struct hd_struct *part; | |
2494 | int cpu; | |
2495 | ||
2496 | cpu = part_stat_lock(); | |
2497 | part = req->part; | |
2498 | ||
2499 | part_stat_inc(cpu, part, ios[rw]); | |
2500 | part_stat_add(cpu, part, ticks[rw], duration); | |
2501 | part_round_stats(req->q, cpu, part); | |
2502 | part_dec_in_flight(req->q, part, rw); | |
2503 | ||
2504 | hd_struct_put(part); | |
2505 | part_stat_unlock(); | |
2506 | } | |
2507 | } | |
2508 | ||
2509 | #ifdef CONFIG_PM | |
2510 | /* | |
2511 | * Don't process normal requests when queue is suspended | |
2512 | * or in the process of suspending/resuming | |
2513 | */ | |
2514 | static bool blk_pm_allow_request(struct request *rq) | |
2515 | { | |
2516 | switch (rq->q->rpm_status) { | |
2517 | case RPM_RESUMING: | |
2518 | case RPM_SUSPENDING: | |
2519 | return rq->rq_flags & RQF_PM; | |
2520 | case RPM_SUSPENDED: | |
2521 | return false; | |
2522 | } | |
2523 | ||
2524 | return true; | |
2525 | } | |
2526 | #else | |
2527 | static bool blk_pm_allow_request(struct request *rq) | |
2528 | { | |
2529 | return true; | |
2530 | } | |
2531 | #endif | |
2532 | ||
2533 | void blk_account_io_start(struct request *rq, bool new_io) | |
2534 | { | |
2535 | struct hd_struct *part; | |
2536 | int rw = rq_data_dir(rq); | |
2537 | int cpu; | |
2538 | ||
2539 | if (!blk_do_io_stat(rq)) | |
2540 | return; | |
2541 | ||
2542 | cpu = part_stat_lock(); | |
2543 | ||
2544 | if (!new_io) { | |
2545 | part = rq->part; | |
2546 | part_stat_inc(cpu, part, merges[rw]); | |
2547 | } else { | |
2548 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2549 | if (!hd_struct_try_get(part)) { | |
2550 | /* | |
2551 | * The partition is already being removed, | |
2552 | * the request will be accounted on the disk only | |
2553 | * | |
2554 | * We take a reference on disk->part0 although that | |
2555 | * partition will never be deleted, so we can treat | |
2556 | * it as any other partition. | |
2557 | */ | |
2558 | part = &rq->rq_disk->part0; | |
2559 | hd_struct_get(part); | |
2560 | } | |
2561 | part_round_stats(rq->q, cpu, part); | |
2562 | part_inc_in_flight(rq->q, part, rw); | |
2563 | rq->part = part; | |
2564 | } | |
2565 | ||
2566 | part_stat_unlock(); | |
2567 | } | |
2568 | ||
2569 | static struct request *elv_next_request(struct request_queue *q) | |
2570 | { | |
2571 | struct request *rq; | |
2572 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
2573 | ||
2574 | WARN_ON_ONCE(q->mq_ops); | |
2575 | ||
2576 | while (1) { | |
2577 | list_for_each_entry(rq, &q->queue_head, queuelist) { | |
2578 | if (blk_pm_allow_request(rq)) | |
2579 | return rq; | |
2580 | ||
2581 | if (rq->rq_flags & RQF_SOFTBARRIER) | |
2582 | break; | |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * Flush request is running and flush request isn't queueable | |
2587 | * in the drive, we can hold the queue till flush request is | |
2588 | * finished. Even we don't do this, driver can't dispatch next | |
2589 | * requests and will requeue them. And this can improve | |
2590 | * throughput too. For example, we have request flush1, write1, | |
2591 | * flush 2. flush1 is dispatched, then queue is hold, write1 | |
2592 | * isn't inserted to queue. After flush1 is finished, flush2 | |
2593 | * will be dispatched. Since disk cache is already clean, | |
2594 | * flush2 will be finished very soon, so looks like flush2 is | |
2595 | * folded to flush1. | |
2596 | * Since the queue is hold, a flag is set to indicate the queue | |
2597 | * should be restarted later. Please see flush_end_io() for | |
2598 | * details. | |
2599 | */ | |
2600 | if (fq->flush_pending_idx != fq->flush_running_idx && | |
2601 | !queue_flush_queueable(q)) { | |
2602 | fq->flush_queue_delayed = 1; | |
2603 | return NULL; | |
2604 | } | |
2605 | if (unlikely(blk_queue_bypass(q)) || | |
2606 | !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0)) | |
2607 | return NULL; | |
2608 | } | |
2609 | } | |
2610 | ||
2611 | /** | |
2612 | * blk_peek_request - peek at the top of a request queue | |
2613 | * @q: request queue to peek at | |
2614 | * | |
2615 | * Description: | |
2616 | * Return the request at the top of @q. The returned request | |
2617 | * should be started using blk_start_request() before LLD starts | |
2618 | * processing it. | |
2619 | * | |
2620 | * Return: | |
2621 | * Pointer to the request at the top of @q if available. Null | |
2622 | * otherwise. | |
2623 | */ | |
2624 | struct request *blk_peek_request(struct request_queue *q) | |
2625 | { | |
2626 | struct request *rq; | |
2627 | int ret; | |
2628 | ||
2629 | lockdep_assert_held(q->queue_lock); | |
2630 | WARN_ON_ONCE(q->mq_ops); | |
2631 | ||
2632 | while ((rq = elv_next_request(q)) != NULL) { | |
2633 | if (!(rq->rq_flags & RQF_STARTED)) { | |
2634 | /* | |
2635 | * This is the first time the device driver | |
2636 | * sees this request (possibly after | |
2637 | * requeueing). Notify IO scheduler. | |
2638 | */ | |
2639 | if (rq->rq_flags & RQF_SORTED) | |
2640 | elv_activate_rq(q, rq); | |
2641 | ||
2642 | /* | |
2643 | * just mark as started even if we don't start | |
2644 | * it, a request that has been delayed should | |
2645 | * not be passed by new incoming requests | |
2646 | */ | |
2647 | rq->rq_flags |= RQF_STARTED; | |
2648 | trace_block_rq_issue(q, rq); | |
2649 | } | |
2650 | ||
2651 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2652 | q->end_sector = rq_end_sector(rq); | |
2653 | q->boundary_rq = NULL; | |
2654 | } | |
2655 | ||
2656 | if (rq->rq_flags & RQF_DONTPREP) | |
2657 | break; | |
2658 | ||
2659 | if (q->dma_drain_size && blk_rq_bytes(rq)) { | |
2660 | /* | |
2661 | * make sure space for the drain appears we | |
2662 | * know we can do this because max_hw_segments | |
2663 | * has been adjusted to be one fewer than the | |
2664 | * device can handle | |
2665 | */ | |
2666 | rq->nr_phys_segments++; | |
2667 | } | |
2668 | ||
2669 | if (!q->prep_rq_fn) | |
2670 | break; | |
2671 | ||
2672 | ret = q->prep_rq_fn(q, rq); | |
2673 | if (ret == BLKPREP_OK) { | |
2674 | break; | |
2675 | } else if (ret == BLKPREP_DEFER) { | |
2676 | /* | |
2677 | * the request may have been (partially) prepped. | |
2678 | * we need to keep this request in the front to | |
2679 | * avoid resource deadlock. RQF_STARTED will | |
2680 | * prevent other fs requests from passing this one. | |
2681 | */ | |
2682 | if (q->dma_drain_size && blk_rq_bytes(rq) && | |
2683 | !(rq->rq_flags & RQF_DONTPREP)) { | |
2684 | /* | |
2685 | * remove the space for the drain we added | |
2686 | * so that we don't add it again | |
2687 | */ | |
2688 | --rq->nr_phys_segments; | |
2689 | } | |
2690 | ||
2691 | rq = NULL; | |
2692 | break; | |
2693 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { | |
2694 | rq->rq_flags |= RQF_QUIET; | |
2695 | /* | |
2696 | * Mark this request as started so we don't trigger | |
2697 | * any debug logic in the end I/O path. | |
2698 | */ | |
2699 | blk_start_request(rq); | |
2700 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? | |
2701 | BLK_STS_TARGET : BLK_STS_IOERR); | |
2702 | } else { | |
2703 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2704 | break; | |
2705 | } | |
2706 | } | |
2707 | ||
2708 | return rq; | |
2709 | } | |
2710 | EXPORT_SYMBOL(blk_peek_request); | |
2711 | ||
2712 | static void blk_dequeue_request(struct request *rq) | |
2713 | { | |
2714 | struct request_queue *q = rq->q; | |
2715 | ||
2716 | BUG_ON(list_empty(&rq->queuelist)); | |
2717 | BUG_ON(ELV_ON_HASH(rq)); | |
2718 | ||
2719 | list_del_init(&rq->queuelist); | |
2720 | ||
2721 | /* | |
2722 | * the time frame between a request being removed from the lists | |
2723 | * and to it is freed is accounted as io that is in progress at | |
2724 | * the driver side. | |
2725 | */ | |
2726 | if (blk_account_rq(rq)) { | |
2727 | q->in_flight[rq_is_sync(rq)]++; | |
2728 | set_io_start_time_ns(rq); | |
2729 | } | |
2730 | } | |
2731 | ||
2732 | /** | |
2733 | * blk_start_request - start request processing on the driver | |
2734 | * @req: request to dequeue | |
2735 | * | |
2736 | * Description: | |
2737 | * Dequeue @req and start timeout timer on it. This hands off the | |
2738 | * request to the driver. | |
2739 | */ | |
2740 | void blk_start_request(struct request *req) | |
2741 | { | |
2742 | lockdep_assert_held(req->q->queue_lock); | |
2743 | WARN_ON_ONCE(req->q->mq_ops); | |
2744 | ||
2745 | blk_dequeue_request(req); | |
2746 | ||
2747 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { | |
2748 | blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req)); | |
2749 | req->rq_flags |= RQF_STATS; | |
2750 | wbt_issue(req->q->rq_wb, &req->issue_stat); | |
2751 | } | |
2752 | ||
2753 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); | |
2754 | blk_add_timer(req); | |
2755 | } | |
2756 | EXPORT_SYMBOL(blk_start_request); | |
2757 | ||
2758 | /** | |
2759 | * blk_fetch_request - fetch a request from a request queue | |
2760 | * @q: request queue to fetch a request from | |
2761 | * | |
2762 | * Description: | |
2763 | * Return the request at the top of @q. The request is started on | |
2764 | * return and LLD can start processing it immediately. | |
2765 | * | |
2766 | * Return: | |
2767 | * Pointer to the request at the top of @q if available. Null | |
2768 | * otherwise. | |
2769 | */ | |
2770 | struct request *blk_fetch_request(struct request_queue *q) | |
2771 | { | |
2772 | struct request *rq; | |
2773 | ||
2774 | lockdep_assert_held(q->queue_lock); | |
2775 | WARN_ON_ONCE(q->mq_ops); | |
2776 | ||
2777 | rq = blk_peek_request(q); | |
2778 | if (rq) | |
2779 | blk_start_request(rq); | |
2780 | return rq; | |
2781 | } | |
2782 | EXPORT_SYMBOL(blk_fetch_request); | |
2783 | ||
2784 | /* | |
2785 | * Steal bios from a request and add them to a bio list. | |
2786 | * The request must not have been partially completed before. | |
2787 | */ | |
2788 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
2789 | { | |
2790 | if (rq->bio) { | |
2791 | if (list->tail) | |
2792 | list->tail->bi_next = rq->bio; | |
2793 | else | |
2794 | list->head = rq->bio; | |
2795 | list->tail = rq->biotail; | |
2796 | ||
2797 | rq->bio = NULL; | |
2798 | rq->biotail = NULL; | |
2799 | } | |
2800 | ||
2801 | rq->__data_len = 0; | |
2802 | } | |
2803 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
2804 | ||
2805 | /** | |
2806 | * blk_update_request - Special helper function for request stacking drivers | |
2807 | * @req: the request being processed | |
2808 | * @error: block status code | |
2809 | * @nr_bytes: number of bytes to complete @req | |
2810 | * | |
2811 | * Description: | |
2812 | * Ends I/O on a number of bytes attached to @req, but doesn't complete | |
2813 | * the request structure even if @req doesn't have leftover. | |
2814 | * If @req has leftover, sets it up for the next range of segments. | |
2815 | * | |
2816 | * This special helper function is only for request stacking drivers | |
2817 | * (e.g. request-based dm) so that they can handle partial completion. | |
2818 | * Actual device drivers should use blk_end_request instead. | |
2819 | * | |
2820 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2821 | * %false return from this function. | |
2822 | * | |
2823 | * Return: | |
2824 | * %false - this request doesn't have any more data | |
2825 | * %true - this request has more data | |
2826 | **/ | |
2827 | bool blk_update_request(struct request *req, blk_status_t error, | |
2828 | unsigned int nr_bytes) | |
2829 | { | |
2830 | int total_bytes; | |
2831 | ||
2832 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); | |
2833 | ||
2834 | if (!req->bio) | |
2835 | return false; | |
2836 | ||
2837 | if (unlikely(error && !blk_rq_is_passthrough(req) && | |
2838 | !(req->rq_flags & RQF_QUIET))) | |
2839 | print_req_error(req, error); | |
2840 | ||
2841 | blk_account_io_completion(req, nr_bytes); | |
2842 | ||
2843 | total_bytes = 0; | |
2844 | while (req->bio) { | |
2845 | struct bio *bio = req->bio; | |
2846 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); | |
2847 | ||
2848 | if (bio_bytes == bio->bi_iter.bi_size) | |
2849 | req->bio = bio->bi_next; | |
2850 | ||
2851 | /* Completion has already been traced */ | |
2852 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
2853 | req_bio_endio(req, bio, bio_bytes, error); | |
2854 | ||
2855 | total_bytes += bio_bytes; | |
2856 | nr_bytes -= bio_bytes; | |
2857 | ||
2858 | if (!nr_bytes) | |
2859 | break; | |
2860 | } | |
2861 | ||
2862 | /* | |
2863 | * completely done | |
2864 | */ | |
2865 | if (!req->bio) { | |
2866 | /* | |
2867 | * Reset counters so that the request stacking driver | |
2868 | * can find how many bytes remain in the request | |
2869 | * later. | |
2870 | */ | |
2871 | req->__data_len = 0; | |
2872 | return false; | |
2873 | } | |
2874 | ||
2875 | req->__data_len -= total_bytes; | |
2876 | ||
2877 | /* update sector only for requests with clear definition of sector */ | |
2878 | if (!blk_rq_is_passthrough(req)) | |
2879 | req->__sector += total_bytes >> 9; | |
2880 | ||
2881 | /* mixed attributes always follow the first bio */ | |
2882 | if (req->rq_flags & RQF_MIXED_MERGE) { | |
2883 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2884 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; | |
2885 | } | |
2886 | ||
2887 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { | |
2888 | /* | |
2889 | * If total number of sectors is less than the first segment | |
2890 | * size, something has gone terribly wrong. | |
2891 | */ | |
2892 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2893 | blk_dump_rq_flags(req, "request botched"); | |
2894 | req->__data_len = blk_rq_cur_bytes(req); | |
2895 | } | |
2896 | ||
2897 | /* recalculate the number of segments */ | |
2898 | blk_recalc_rq_segments(req); | |
2899 | } | |
2900 | ||
2901 | return true; | |
2902 | } | |
2903 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2904 | ||
2905 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, | |
2906 | unsigned int nr_bytes, | |
2907 | unsigned int bidi_bytes) | |
2908 | { | |
2909 | if (blk_update_request(rq, error, nr_bytes)) | |
2910 | return true; | |
2911 | ||
2912 | /* Bidi request must be completed as a whole */ | |
2913 | if (unlikely(blk_bidi_rq(rq)) && | |
2914 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2915 | return true; | |
2916 | ||
2917 | if (blk_queue_add_random(rq->q)) | |
2918 | add_disk_randomness(rq->rq_disk); | |
2919 | ||
2920 | return false; | |
2921 | } | |
2922 | ||
2923 | /** | |
2924 | * blk_unprep_request - unprepare a request | |
2925 | * @req: the request | |
2926 | * | |
2927 | * This function makes a request ready for complete resubmission (or | |
2928 | * completion). It happens only after all error handling is complete, | |
2929 | * so represents the appropriate moment to deallocate any resources | |
2930 | * that were allocated to the request in the prep_rq_fn. The queue | |
2931 | * lock is held when calling this. | |
2932 | */ | |
2933 | void blk_unprep_request(struct request *req) | |
2934 | { | |
2935 | struct request_queue *q = req->q; | |
2936 | ||
2937 | req->rq_flags &= ~RQF_DONTPREP; | |
2938 | if (q->unprep_rq_fn) | |
2939 | q->unprep_rq_fn(q, req); | |
2940 | } | |
2941 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2942 | ||
2943 | void blk_finish_request(struct request *req, blk_status_t error) | |
2944 | { | |
2945 | struct request_queue *q = req->q; | |
2946 | ||
2947 | lockdep_assert_held(req->q->queue_lock); | |
2948 | WARN_ON_ONCE(q->mq_ops); | |
2949 | ||
2950 | if (req->rq_flags & RQF_STATS) | |
2951 | blk_stat_add(req); | |
2952 | ||
2953 | if (req->rq_flags & RQF_QUEUED) | |
2954 | blk_queue_end_tag(q, req); | |
2955 | ||
2956 | BUG_ON(blk_queued_rq(req)); | |
2957 | ||
2958 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) | |
2959 | laptop_io_completion(req->q->backing_dev_info); | |
2960 | ||
2961 | blk_delete_timer(req); | |
2962 | ||
2963 | if (req->rq_flags & RQF_DONTPREP) | |
2964 | blk_unprep_request(req); | |
2965 | ||
2966 | blk_account_io_done(req); | |
2967 | ||
2968 | if (req->end_io) { | |
2969 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
2970 | req->end_io(req, error); | |
2971 | } else { | |
2972 | if (blk_bidi_rq(req)) | |
2973 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2974 | ||
2975 | __blk_put_request(q, req); | |
2976 | } | |
2977 | } | |
2978 | EXPORT_SYMBOL(blk_finish_request); | |
2979 | ||
2980 | /** | |
2981 | * blk_end_bidi_request - Complete a bidi request | |
2982 | * @rq: the request to complete | |
2983 | * @error: block status code | |
2984 | * @nr_bytes: number of bytes to complete @rq | |
2985 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2986 | * | |
2987 | * Description: | |
2988 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2989 | * Drivers that supports bidi can safely call this member for any | |
2990 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2991 | * just ignored. | |
2992 | * | |
2993 | * Return: | |
2994 | * %false - we are done with this request | |
2995 | * %true - still buffers pending for this request | |
2996 | **/ | |
2997 | static bool blk_end_bidi_request(struct request *rq, blk_status_t error, | |
2998 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
2999 | { | |
3000 | struct request_queue *q = rq->q; | |
3001 | unsigned long flags; | |
3002 | ||
3003 | WARN_ON_ONCE(q->mq_ops); | |
3004 | ||
3005 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
3006 | return true; | |
3007 | ||
3008 | spin_lock_irqsave(q->queue_lock, flags); | |
3009 | blk_finish_request(rq, error); | |
3010 | spin_unlock_irqrestore(q->queue_lock, flags); | |
3011 | ||
3012 | return false; | |
3013 | } | |
3014 | ||
3015 | /** | |
3016 | * __blk_end_bidi_request - Complete a bidi request with queue lock held | |
3017 | * @rq: the request to complete | |
3018 | * @error: block status code | |
3019 | * @nr_bytes: number of bytes to complete @rq | |
3020 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
3021 | * | |
3022 | * Description: | |
3023 | * Identical to blk_end_bidi_request() except that queue lock is | |
3024 | * assumed to be locked on entry and remains so on return. | |
3025 | * | |
3026 | * Return: | |
3027 | * %false - we are done with this request | |
3028 | * %true - still buffers pending for this request | |
3029 | **/ | |
3030 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, | |
3031 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
3032 | { | |
3033 | lockdep_assert_held(rq->q->queue_lock); | |
3034 | WARN_ON_ONCE(rq->q->mq_ops); | |
3035 | ||
3036 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
3037 | return true; | |
3038 | ||
3039 | blk_finish_request(rq, error); | |
3040 | ||
3041 | return false; | |
3042 | } | |
3043 | ||
3044 | /** | |
3045 | * blk_end_request - Helper function for drivers to complete the request. | |
3046 | * @rq: the request being processed | |
3047 | * @error: block status code | |
3048 | * @nr_bytes: number of bytes to complete | |
3049 | * | |
3050 | * Description: | |
3051 | * Ends I/O on a number of bytes attached to @rq. | |
3052 | * If @rq has leftover, sets it up for the next range of segments. | |
3053 | * | |
3054 | * Return: | |
3055 | * %false - we are done with this request | |
3056 | * %true - still buffers pending for this request | |
3057 | **/ | |
3058 | bool blk_end_request(struct request *rq, blk_status_t error, | |
3059 | unsigned int nr_bytes) | |
3060 | { | |
3061 | WARN_ON_ONCE(rq->q->mq_ops); | |
3062 | return blk_end_bidi_request(rq, error, nr_bytes, 0); | |
3063 | } | |
3064 | EXPORT_SYMBOL(blk_end_request); | |
3065 | ||
3066 | /** | |
3067 | * blk_end_request_all - Helper function for drives to finish the request. | |
3068 | * @rq: the request to finish | |
3069 | * @error: block status code | |
3070 | * | |
3071 | * Description: | |
3072 | * Completely finish @rq. | |
3073 | */ | |
3074 | void blk_end_request_all(struct request *rq, blk_status_t error) | |
3075 | { | |
3076 | bool pending; | |
3077 | unsigned int bidi_bytes = 0; | |
3078 | ||
3079 | if (unlikely(blk_bidi_rq(rq))) | |
3080 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3081 | ||
3082 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3083 | BUG_ON(pending); | |
3084 | } | |
3085 | EXPORT_SYMBOL(blk_end_request_all); | |
3086 | ||
3087 | /** | |
3088 | * __blk_end_request - Helper function for drivers to complete the request. | |
3089 | * @rq: the request being processed | |
3090 | * @error: block status code | |
3091 | * @nr_bytes: number of bytes to complete | |
3092 | * | |
3093 | * Description: | |
3094 | * Must be called with queue lock held unlike blk_end_request(). | |
3095 | * | |
3096 | * Return: | |
3097 | * %false - we are done with this request | |
3098 | * %true - still buffers pending for this request | |
3099 | **/ | |
3100 | bool __blk_end_request(struct request *rq, blk_status_t error, | |
3101 | unsigned int nr_bytes) | |
3102 | { | |
3103 | lockdep_assert_held(rq->q->queue_lock); | |
3104 | WARN_ON_ONCE(rq->q->mq_ops); | |
3105 | ||
3106 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); | |
3107 | } | |
3108 | EXPORT_SYMBOL(__blk_end_request); | |
3109 | ||
3110 | /** | |
3111 | * __blk_end_request_all - Helper function for drives to finish the request. | |
3112 | * @rq: the request to finish | |
3113 | * @error: block status code | |
3114 | * | |
3115 | * Description: | |
3116 | * Completely finish @rq. Must be called with queue lock held. | |
3117 | */ | |
3118 | void __blk_end_request_all(struct request *rq, blk_status_t error) | |
3119 | { | |
3120 | bool pending; | |
3121 | unsigned int bidi_bytes = 0; | |
3122 | ||
3123 | lockdep_assert_held(rq->q->queue_lock); | |
3124 | WARN_ON_ONCE(rq->q->mq_ops); | |
3125 | ||
3126 | if (unlikely(blk_bidi_rq(rq))) | |
3127 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3128 | ||
3129 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3130 | BUG_ON(pending); | |
3131 | } | |
3132 | EXPORT_SYMBOL(__blk_end_request_all); | |
3133 | ||
3134 | /** | |
3135 | * __blk_end_request_cur - Helper function to finish the current request chunk. | |
3136 | * @rq: the request to finish the current chunk for | |
3137 | * @error: block status code | |
3138 | * | |
3139 | * Description: | |
3140 | * Complete the current consecutively mapped chunk from @rq. Must | |
3141 | * be called with queue lock held. | |
3142 | * | |
3143 | * Return: | |
3144 | * %false - we are done with this request | |
3145 | * %true - still buffers pending for this request | |
3146 | */ | |
3147 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) | |
3148 | { | |
3149 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
3150 | } | |
3151 | EXPORT_SYMBOL(__blk_end_request_cur); | |
3152 | ||
3153 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
3154 | struct bio *bio) | |
3155 | { | |
3156 | if (bio_has_data(bio)) | |
3157 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3158 | ||
3159 | rq->__data_len = bio->bi_iter.bi_size; | |
3160 | rq->bio = rq->biotail = bio; | |
3161 | ||
3162 | if (bio->bi_disk) | |
3163 | rq->rq_disk = bio->bi_disk; | |
3164 | } | |
3165 | ||
3166 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
3167 | /** | |
3168 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3169 | * @rq: the request to be flushed | |
3170 | * | |
3171 | * Description: | |
3172 | * Flush all pages in @rq. | |
3173 | */ | |
3174 | void rq_flush_dcache_pages(struct request *rq) | |
3175 | { | |
3176 | struct req_iterator iter; | |
3177 | struct bio_vec bvec; | |
3178 | ||
3179 | rq_for_each_segment(bvec, rq, iter) | |
3180 | flush_dcache_page(bvec.bv_page); | |
3181 | } | |
3182 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3183 | #endif | |
3184 | ||
3185 | /** | |
3186 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3187 | * @q : the queue of the device being checked | |
3188 | * | |
3189 | * Description: | |
3190 | * Check if underlying low-level drivers of a device are busy. | |
3191 | * If the drivers want to export their busy state, they must set own | |
3192 | * exporting function using blk_queue_lld_busy() first. | |
3193 | * | |
3194 | * Basically, this function is used only by request stacking drivers | |
3195 | * to stop dispatching requests to underlying devices when underlying | |
3196 | * devices are busy. This behavior helps more I/O merging on the queue | |
3197 | * of the request stacking driver and prevents I/O throughput regression | |
3198 | * on burst I/O load. | |
3199 | * | |
3200 | * Return: | |
3201 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3202 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3203 | */ | |
3204 | int blk_lld_busy(struct request_queue *q) | |
3205 | { | |
3206 | if (q->lld_busy_fn) | |
3207 | return q->lld_busy_fn(q); | |
3208 | ||
3209 | return 0; | |
3210 | } | |
3211 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3212 | ||
3213 | /** | |
3214 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3215 | * @rq: the clone request to be cleaned up | |
3216 | * | |
3217 | * Description: | |
3218 | * Free all bios in @rq for a cloned request. | |
3219 | */ | |
3220 | void blk_rq_unprep_clone(struct request *rq) | |
3221 | { | |
3222 | struct bio *bio; | |
3223 | ||
3224 | while ((bio = rq->bio) != NULL) { | |
3225 | rq->bio = bio->bi_next; | |
3226 | ||
3227 | bio_put(bio); | |
3228 | } | |
3229 | } | |
3230 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3231 | ||
3232 | /* | |
3233 | * Copy attributes of the original request to the clone request. | |
3234 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3235 | */ | |
3236 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
3237 | { | |
3238 | dst->cpu = src->cpu; | |
3239 | dst->__sector = blk_rq_pos(src); | |
3240 | dst->__data_len = blk_rq_bytes(src); | |
3241 | dst->nr_phys_segments = src->nr_phys_segments; | |
3242 | dst->ioprio = src->ioprio; | |
3243 | dst->extra_len = src->extra_len; | |
3244 | } | |
3245 | ||
3246 | /** | |
3247 | * blk_rq_prep_clone - Helper function to setup clone request | |
3248 | * @rq: the request to be setup | |
3249 | * @rq_src: original request to be cloned | |
3250 | * @bs: bio_set that bios for clone are allocated from | |
3251 | * @gfp_mask: memory allocation mask for bio | |
3252 | * @bio_ctr: setup function to be called for each clone bio. | |
3253 | * Returns %0 for success, non %0 for failure. | |
3254 | * @data: private data to be passed to @bio_ctr | |
3255 | * | |
3256 | * Description: | |
3257 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3258 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3259 | * are not copied, and copying such parts is the caller's responsibility. | |
3260 | * Also, pages which the original bios are pointing to are not copied | |
3261 | * and the cloned bios just point same pages. | |
3262 | * So cloned bios must be completed before original bios, which means | |
3263 | * the caller must complete @rq before @rq_src. | |
3264 | */ | |
3265 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3266 | struct bio_set *bs, gfp_t gfp_mask, | |
3267 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3268 | void *data) | |
3269 | { | |
3270 | struct bio *bio, *bio_src; | |
3271 | ||
3272 | if (!bs) | |
3273 | bs = fs_bio_set; | |
3274 | ||
3275 | __rq_for_each_bio(bio_src, rq_src) { | |
3276 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3277 | if (!bio) | |
3278 | goto free_and_out; | |
3279 | ||
3280 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3281 | goto free_and_out; | |
3282 | ||
3283 | if (rq->bio) { | |
3284 | rq->biotail->bi_next = bio; | |
3285 | rq->biotail = bio; | |
3286 | } else | |
3287 | rq->bio = rq->biotail = bio; | |
3288 | } | |
3289 | ||
3290 | __blk_rq_prep_clone(rq, rq_src); | |
3291 | ||
3292 | return 0; | |
3293 | ||
3294 | free_and_out: | |
3295 | if (bio) | |
3296 | bio_put(bio); | |
3297 | blk_rq_unprep_clone(rq); | |
3298 | ||
3299 | return -ENOMEM; | |
3300 | } | |
3301 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3302 | ||
3303 | int kblockd_schedule_work(struct work_struct *work) | |
3304 | { | |
3305 | return queue_work(kblockd_workqueue, work); | |
3306 | } | |
3307 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3308 | ||
3309 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) | |
3310 | { | |
3311 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3312 | } | |
3313 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3314 | ||
3315 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, | |
3316 | unsigned long delay) | |
3317 | { | |
3318 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3319 | } | |
3320 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3321 | ||
3322 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, | |
3323 | unsigned long delay) | |
3324 | { | |
3325 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3326 | } | |
3327 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3328 | ||
3329 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, | |
3330 | unsigned long delay) | |
3331 | { | |
3332 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3333 | } | |
3334 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3335 | ||
3336 | /** | |
3337 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3338 | * @plug: The &struct blk_plug that needs to be initialized | |
3339 | * | |
3340 | * Description: | |
3341 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3342 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3343 | * blk_finish_plug(). This is important from a performance perspective, but | |
3344 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3345 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3346 | * page belonging to that request that is currently residing in our private | |
3347 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3348 | * this kind of deadlock. | |
3349 | */ | |
3350 | void blk_start_plug(struct blk_plug *plug) | |
3351 | { | |
3352 | struct task_struct *tsk = current; | |
3353 | ||
3354 | /* | |
3355 | * If this is a nested plug, don't actually assign it. | |
3356 | */ | |
3357 | if (tsk->plug) | |
3358 | return; | |
3359 | ||
3360 | INIT_LIST_HEAD(&plug->list); | |
3361 | INIT_LIST_HEAD(&plug->mq_list); | |
3362 | INIT_LIST_HEAD(&plug->cb_list); | |
3363 | /* | |
3364 | * Store ordering should not be needed here, since a potential | |
3365 | * preempt will imply a full memory barrier | |
3366 | */ | |
3367 | tsk->plug = plug; | |
3368 | } | |
3369 | EXPORT_SYMBOL(blk_start_plug); | |
3370 | ||
3371 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3372 | { | |
3373 | struct request *rqa = container_of(a, struct request, queuelist); | |
3374 | struct request *rqb = container_of(b, struct request, queuelist); | |
3375 | ||
3376 | return !(rqa->q < rqb->q || | |
3377 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3382 | * until a safe kblockd context. We due this to avoid accidental big | |
3383 | * additional stack usage in driver dispatch, in places where the originally | |
3384 | * plugger did not intend it. | |
3385 | */ | |
3386 | static void queue_unplugged(struct request_queue *q, unsigned int depth, | |
3387 | bool from_schedule) | |
3388 | __releases(q->queue_lock) | |
3389 | { | |
3390 | lockdep_assert_held(q->queue_lock); | |
3391 | ||
3392 | trace_block_unplug(q, depth, !from_schedule); | |
3393 | ||
3394 | if (from_schedule) | |
3395 | blk_run_queue_async(q); | |
3396 | else | |
3397 | __blk_run_queue(q); | |
3398 | spin_unlock(q->queue_lock); | |
3399 | } | |
3400 | ||
3401 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) | |
3402 | { | |
3403 | LIST_HEAD(callbacks); | |
3404 | ||
3405 | while (!list_empty(&plug->cb_list)) { | |
3406 | list_splice_init(&plug->cb_list, &callbacks); | |
3407 | ||
3408 | while (!list_empty(&callbacks)) { | |
3409 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
3410 | struct blk_plug_cb, | |
3411 | list); | |
3412 | list_del(&cb->list); | |
3413 | cb->callback(cb, from_schedule); | |
3414 | } | |
3415 | } | |
3416 | } | |
3417 | ||
3418 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, | |
3419 | int size) | |
3420 | { | |
3421 | struct blk_plug *plug = current->plug; | |
3422 | struct blk_plug_cb *cb; | |
3423 | ||
3424 | if (!plug) | |
3425 | return NULL; | |
3426 | ||
3427 | list_for_each_entry(cb, &plug->cb_list, list) | |
3428 | if (cb->callback == unplug && cb->data == data) | |
3429 | return cb; | |
3430 | ||
3431 | /* Not currently on the callback list */ | |
3432 | BUG_ON(size < sizeof(*cb)); | |
3433 | cb = kzalloc(size, GFP_ATOMIC); | |
3434 | if (cb) { | |
3435 | cb->data = data; | |
3436 | cb->callback = unplug; | |
3437 | list_add(&cb->list, &plug->cb_list); | |
3438 | } | |
3439 | return cb; | |
3440 | } | |
3441 | EXPORT_SYMBOL(blk_check_plugged); | |
3442 | ||
3443 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) | |
3444 | { | |
3445 | struct request_queue *q; | |
3446 | unsigned long flags; | |
3447 | struct request *rq; | |
3448 | LIST_HEAD(list); | |
3449 | unsigned int depth; | |
3450 | ||
3451 | flush_plug_callbacks(plug, from_schedule); | |
3452 | ||
3453 | if (!list_empty(&plug->mq_list)) | |
3454 | blk_mq_flush_plug_list(plug, from_schedule); | |
3455 | ||
3456 | if (list_empty(&plug->list)) | |
3457 | return; | |
3458 | ||
3459 | list_splice_init(&plug->list, &list); | |
3460 | ||
3461 | list_sort(NULL, &list, plug_rq_cmp); | |
3462 | ||
3463 | q = NULL; | |
3464 | depth = 0; | |
3465 | ||
3466 | /* | |
3467 | * Save and disable interrupts here, to avoid doing it for every | |
3468 | * queue lock we have to take. | |
3469 | */ | |
3470 | local_irq_save(flags); | |
3471 | while (!list_empty(&list)) { | |
3472 | rq = list_entry_rq(list.next); | |
3473 | list_del_init(&rq->queuelist); | |
3474 | BUG_ON(!rq->q); | |
3475 | if (rq->q != q) { | |
3476 | /* | |
3477 | * This drops the queue lock | |
3478 | */ | |
3479 | if (q) | |
3480 | queue_unplugged(q, depth, from_schedule); | |
3481 | q = rq->q; | |
3482 | depth = 0; | |
3483 | spin_lock(q->queue_lock); | |
3484 | } | |
3485 | ||
3486 | /* | |
3487 | * Short-circuit if @q is dead | |
3488 | */ | |
3489 | if (unlikely(blk_queue_dying(q))) { | |
3490 | __blk_end_request_all(rq, BLK_STS_IOERR); | |
3491 | continue; | |
3492 | } | |
3493 | ||
3494 | /* | |
3495 | * rq is already accounted, so use raw insert | |
3496 | */ | |
3497 | if (op_is_flush(rq->cmd_flags)) | |
3498 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
3499 | else | |
3500 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
3501 | ||
3502 | depth++; | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * This drops the queue lock | |
3507 | */ | |
3508 | if (q) | |
3509 | queue_unplugged(q, depth, from_schedule); | |
3510 | ||
3511 | local_irq_restore(flags); | |
3512 | } | |
3513 | ||
3514 | void blk_finish_plug(struct blk_plug *plug) | |
3515 | { | |
3516 | if (plug != current->plug) | |
3517 | return; | |
3518 | blk_flush_plug_list(plug, false); | |
3519 | ||
3520 | current->plug = NULL; | |
3521 | } | |
3522 | EXPORT_SYMBOL(blk_finish_plug); | |
3523 | ||
3524 | #ifdef CONFIG_PM | |
3525 | /** | |
3526 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3527 | * @q: the queue of the device | |
3528 | * @dev: the device the queue belongs to | |
3529 | * | |
3530 | * Description: | |
3531 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3532 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3533 | * should call this function after @dev has been initialized, and its | |
3534 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3535 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3536 | * cases, driver should call this function before any I/O has taken place. | |
3537 | * | |
3538 | * This function takes care of setting up using auto suspend for the device, | |
3539 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3540 | * until an updated value is either set by user or by driver. Drivers do | |
3541 | * not need to touch other autosuspend settings. | |
3542 | * | |
3543 | * The block layer runtime PM is request based, so only works for drivers | |
3544 | * that use request as their IO unit instead of those directly use bio's. | |
3545 | */ | |
3546 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3547 | { | |
3548 | /* not support for RQF_PM and ->rpm_status in blk-mq yet */ | |
3549 | if (q->mq_ops) | |
3550 | return; | |
3551 | ||
3552 | q->dev = dev; | |
3553 | q->rpm_status = RPM_ACTIVE; | |
3554 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3555 | pm_runtime_use_autosuspend(q->dev); | |
3556 | } | |
3557 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3558 | ||
3559 | /** | |
3560 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3561 | * @q: the queue of the device | |
3562 | * | |
3563 | * Description: | |
3564 | * This function will check if runtime suspend is allowed for the device | |
3565 | * by examining if there are any requests pending in the queue. If there | |
3566 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3567 | * the queue's status will be updated to SUSPENDING and the driver can | |
3568 | * proceed to suspend the device. | |
3569 | * | |
3570 | * For the not allowed case, we mark last busy for the device so that | |
3571 | * runtime PM core will try to autosuspend it some time later. | |
3572 | * | |
3573 | * This function should be called near the start of the device's | |
3574 | * runtime_suspend callback. | |
3575 | * | |
3576 | * Return: | |
3577 | * 0 - OK to runtime suspend the device | |
3578 | * -EBUSY - Device should not be runtime suspended | |
3579 | */ | |
3580 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3581 | { | |
3582 | int ret = 0; | |
3583 | ||
3584 | if (!q->dev) | |
3585 | return ret; | |
3586 | ||
3587 | spin_lock_irq(q->queue_lock); | |
3588 | if (q->nr_pending) { | |
3589 | ret = -EBUSY; | |
3590 | pm_runtime_mark_last_busy(q->dev); | |
3591 | } else { | |
3592 | q->rpm_status = RPM_SUSPENDING; | |
3593 | } | |
3594 | spin_unlock_irq(q->queue_lock); | |
3595 | return ret; | |
3596 | } | |
3597 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3598 | ||
3599 | /** | |
3600 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3601 | * @q: the queue of the device | |
3602 | * @err: return value of the device's runtime_suspend function | |
3603 | * | |
3604 | * Description: | |
3605 | * Update the queue's runtime status according to the return value of the | |
3606 | * device's runtime suspend function and mark last busy for the device so | |
3607 | * that PM core will try to auto suspend the device at a later time. | |
3608 | * | |
3609 | * This function should be called near the end of the device's | |
3610 | * runtime_suspend callback. | |
3611 | */ | |
3612 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3613 | { | |
3614 | if (!q->dev) | |
3615 | return; | |
3616 | ||
3617 | spin_lock_irq(q->queue_lock); | |
3618 | if (!err) { | |
3619 | q->rpm_status = RPM_SUSPENDED; | |
3620 | } else { | |
3621 | q->rpm_status = RPM_ACTIVE; | |
3622 | pm_runtime_mark_last_busy(q->dev); | |
3623 | } | |
3624 | spin_unlock_irq(q->queue_lock); | |
3625 | } | |
3626 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3627 | ||
3628 | /** | |
3629 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3630 | * @q: the queue of the device | |
3631 | * | |
3632 | * Description: | |
3633 | * Update the queue's runtime status to RESUMING in preparation for the | |
3634 | * runtime resume of the device. | |
3635 | * | |
3636 | * This function should be called near the start of the device's | |
3637 | * runtime_resume callback. | |
3638 | */ | |
3639 | void blk_pre_runtime_resume(struct request_queue *q) | |
3640 | { | |
3641 | if (!q->dev) | |
3642 | return; | |
3643 | ||
3644 | spin_lock_irq(q->queue_lock); | |
3645 | q->rpm_status = RPM_RESUMING; | |
3646 | spin_unlock_irq(q->queue_lock); | |
3647 | } | |
3648 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3649 | ||
3650 | /** | |
3651 | * blk_post_runtime_resume - Post runtime resume processing | |
3652 | * @q: the queue of the device | |
3653 | * @err: return value of the device's runtime_resume function | |
3654 | * | |
3655 | * Description: | |
3656 | * Update the queue's runtime status according to the return value of the | |
3657 | * device's runtime_resume function. If it is successfully resumed, process | |
3658 | * the requests that are queued into the device's queue when it is resuming | |
3659 | * and then mark last busy and initiate autosuspend for it. | |
3660 | * | |
3661 | * This function should be called near the end of the device's | |
3662 | * runtime_resume callback. | |
3663 | */ | |
3664 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3665 | { | |
3666 | if (!q->dev) | |
3667 | return; | |
3668 | ||
3669 | spin_lock_irq(q->queue_lock); | |
3670 | if (!err) { | |
3671 | q->rpm_status = RPM_ACTIVE; | |
3672 | __blk_run_queue(q); | |
3673 | pm_runtime_mark_last_busy(q->dev); | |
3674 | pm_request_autosuspend(q->dev); | |
3675 | } else { | |
3676 | q->rpm_status = RPM_SUSPENDED; | |
3677 | } | |
3678 | spin_unlock_irq(q->queue_lock); | |
3679 | } | |
3680 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
3681 | ||
3682 | /** | |
3683 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3684 | * @q: the queue of the device | |
3685 | * | |
3686 | * If the device is left runtime suspended during system suspend the resume | |
3687 | * hook typically resumes the device and corrects runtime status | |
3688 | * accordingly. However, that does not affect the queue runtime PM status | |
3689 | * which is still "suspended". This prevents processing requests from the | |
3690 | * queue. | |
3691 | * | |
3692 | * This function can be used in driver's resume hook to correct queue | |
3693 | * runtime PM status and re-enable peeking requests from the queue. It | |
3694 | * should be called before first request is added to the queue. | |
3695 | */ | |
3696 | void blk_set_runtime_active(struct request_queue *q) | |
3697 | { | |
3698 | spin_lock_irq(q->queue_lock); | |
3699 | q->rpm_status = RPM_ACTIVE; | |
3700 | pm_runtime_mark_last_busy(q->dev); | |
3701 | pm_request_autosuspend(q->dev); | |
3702 | spin_unlock_irq(q->queue_lock); | |
3703 | } | |
3704 | EXPORT_SYMBOL(blk_set_runtime_active); | |
3705 | #endif | |
3706 | ||
3707 | int __init blk_dev_init(void) | |
3708 | { | |
3709 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); | |
3710 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
3711 | FIELD_SIZEOF(struct request, cmd_flags)); | |
3712 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
3713 | FIELD_SIZEOF(struct bio, bi_opf)); | |
3714 | ||
3715 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ | |
3716 | kblockd_workqueue = alloc_workqueue("kblockd", | |
3717 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
3718 | if (!kblockd_workqueue) | |
3719 | panic("Failed to create kblockd\n"); | |
3720 | ||
3721 | request_cachep = kmem_cache_create("blkdev_requests", | |
3722 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
3723 | ||
3724 | blk_requestq_cachep = kmem_cache_create("request_queue", | |
3725 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
3726 | ||
3727 | #ifdef CONFIG_DEBUG_FS | |
3728 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3729 | #endif | |
3730 | ||
3731 | return 0; | |
3732 | } |