]>
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 | cancel_delayed_work_sync(&q->requeue_work); | |
343 | queue_for_each_hw_ctx(q, hctx, i) | |
344 | cancel_delayed_work_sync(&hctx->run_work); | |
345 | } else { | |
346 | cancel_delayed_work_sync(&q->delay_work); | |
347 | } | |
348 | } | |
349 | EXPORT_SYMBOL(blk_sync_queue); | |
350 | ||
351 | /** | |
352 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
353 | * @q: The queue to run | |
354 | * | |
355 | * Description: | |
356 | * Invoke request handling on a queue if there are any pending requests. | |
357 | * May be used to restart request handling after a request has completed. | |
358 | * This variant runs the queue whether or not the queue has been | |
359 | * stopped. Must be called with the queue lock held and interrupts | |
360 | * disabled. See also @blk_run_queue. | |
361 | */ | |
362 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
363 | { | |
364 | lockdep_assert_held(q->queue_lock); | |
365 | WARN_ON_ONCE(q->mq_ops); | |
366 | ||
367 | if (unlikely(blk_queue_dead(q))) | |
368 | return; | |
369 | ||
370 | /* | |
371 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
372 | * the queue lock internally. As a result multiple threads may be | |
373 | * running such a request function concurrently. Keep track of the | |
374 | * number of active request_fn invocations such that blk_drain_queue() | |
375 | * can wait until all these request_fn calls have finished. | |
376 | */ | |
377 | q->request_fn_active++; | |
378 | q->request_fn(q); | |
379 | q->request_fn_active--; | |
380 | } | |
381 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); | |
382 | ||
383 | /** | |
384 | * __blk_run_queue - run a single device queue | |
385 | * @q: The queue to run | |
386 | * | |
387 | * Description: | |
388 | * See @blk_run_queue. | |
389 | */ | |
390 | void __blk_run_queue(struct request_queue *q) | |
391 | { | |
392 | lockdep_assert_held(q->queue_lock); | |
393 | WARN_ON_ONCE(q->mq_ops); | |
394 | ||
395 | if (unlikely(blk_queue_stopped(q))) | |
396 | return; | |
397 | ||
398 | __blk_run_queue_uncond(q); | |
399 | } | |
400 | EXPORT_SYMBOL(__blk_run_queue); | |
401 | ||
402 | /** | |
403 | * blk_run_queue_async - run a single device queue in workqueue context | |
404 | * @q: The queue to run | |
405 | * | |
406 | * Description: | |
407 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
408 | * of us. | |
409 | * | |
410 | * Note: | |
411 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
412 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
413 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
414 | */ | |
415 | void blk_run_queue_async(struct request_queue *q) | |
416 | { | |
417 | lockdep_assert_held(q->queue_lock); | |
418 | WARN_ON_ONCE(q->mq_ops); | |
419 | ||
420 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) | |
421 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); | |
422 | } | |
423 | EXPORT_SYMBOL(blk_run_queue_async); | |
424 | ||
425 | /** | |
426 | * blk_run_queue - run a single device queue | |
427 | * @q: The queue to run | |
428 | * | |
429 | * Description: | |
430 | * Invoke request handling on this queue, if it has pending work to do. | |
431 | * May be used to restart queueing when a request has completed. | |
432 | */ | |
433 | void blk_run_queue(struct request_queue *q) | |
434 | { | |
435 | unsigned long flags; | |
436 | ||
437 | WARN_ON_ONCE(q->mq_ops); | |
438 | ||
439 | spin_lock_irqsave(q->queue_lock, flags); | |
440 | __blk_run_queue(q); | |
441 | spin_unlock_irqrestore(q->queue_lock, flags); | |
442 | } | |
443 | EXPORT_SYMBOL(blk_run_queue); | |
444 | ||
445 | void blk_put_queue(struct request_queue *q) | |
446 | { | |
447 | kobject_put(&q->kobj); | |
448 | } | |
449 | EXPORT_SYMBOL(blk_put_queue); | |
450 | ||
451 | /** | |
452 | * __blk_drain_queue - drain requests from request_queue | |
453 | * @q: queue to drain | |
454 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV | |
455 | * | |
456 | * Drain requests from @q. If @drain_all is set, all requests are drained. | |
457 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
458 | * for ensuring that no new requests which need to be drained are queued. | |
459 | */ | |
460 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) | |
461 | __releases(q->queue_lock) | |
462 | __acquires(q->queue_lock) | |
463 | { | |
464 | int i; | |
465 | ||
466 | lockdep_assert_held(q->queue_lock); | |
467 | WARN_ON_ONCE(q->mq_ops); | |
468 | ||
469 | while (true) { | |
470 | bool drain = false; | |
471 | ||
472 | /* | |
473 | * The caller might be trying to drain @q before its | |
474 | * elevator is initialized. | |
475 | */ | |
476 | if (q->elevator) | |
477 | elv_drain_elevator(q); | |
478 | ||
479 | blkcg_drain_queue(q); | |
480 | ||
481 | /* | |
482 | * This function might be called on a queue which failed | |
483 | * driver init after queue creation or is not yet fully | |
484 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
485 | * in such cases. Kick queue iff dispatch queue has | |
486 | * something on it and @q has request_fn set. | |
487 | */ | |
488 | if (!list_empty(&q->queue_head) && q->request_fn) | |
489 | __blk_run_queue(q); | |
490 | ||
491 | drain |= q->nr_rqs_elvpriv; | |
492 | drain |= q->request_fn_active; | |
493 | ||
494 | /* | |
495 | * Unfortunately, requests are queued at and tracked from | |
496 | * multiple places and there's no single counter which can | |
497 | * be drained. Check all the queues and counters. | |
498 | */ | |
499 | if (drain_all) { | |
500 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
501 | drain |= !list_empty(&q->queue_head); | |
502 | for (i = 0; i < 2; i++) { | |
503 | drain |= q->nr_rqs[i]; | |
504 | drain |= q->in_flight[i]; | |
505 | if (fq) | |
506 | drain |= !list_empty(&fq->flush_queue[i]); | |
507 | } | |
508 | } | |
509 | ||
510 | if (!drain) | |
511 | break; | |
512 | ||
513 | spin_unlock_irq(q->queue_lock); | |
514 | ||
515 | msleep(10); | |
516 | ||
517 | spin_lock_irq(q->queue_lock); | |
518 | } | |
519 | ||
520 | /* | |
521 | * With queue marked dead, any woken up waiter will fail the | |
522 | * allocation path, so the wakeup chaining is lost and we're | |
523 | * left with hung waiters. We need to wake up those waiters. | |
524 | */ | |
525 | if (q->request_fn) { | |
526 | struct request_list *rl; | |
527 | ||
528 | blk_queue_for_each_rl(rl, q) | |
529 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
530 | wake_up_all(&rl->wait[i]); | |
531 | } | |
532 | } | |
533 | ||
534 | /** | |
535 | * blk_queue_bypass_start - enter queue bypass mode | |
536 | * @q: queue of interest | |
537 | * | |
538 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
539 | * function makes @q enter bypass mode and drains all requests which were | |
540 | * throttled or issued before. On return, it's guaranteed that no request | |
541 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true | |
542 | * inside queue or RCU read lock. | |
543 | */ | |
544 | void blk_queue_bypass_start(struct request_queue *q) | |
545 | { | |
546 | WARN_ON_ONCE(q->mq_ops); | |
547 | ||
548 | spin_lock_irq(q->queue_lock); | |
549 | q->bypass_depth++; | |
550 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
551 | spin_unlock_irq(q->queue_lock); | |
552 | ||
553 | /* | |
554 | * Queues start drained. Skip actual draining till init is | |
555 | * complete. This avoids lenghty delays during queue init which | |
556 | * can happen many times during boot. | |
557 | */ | |
558 | if (blk_queue_init_done(q)) { | |
559 | spin_lock_irq(q->queue_lock); | |
560 | __blk_drain_queue(q, false); | |
561 | spin_unlock_irq(q->queue_lock); | |
562 | ||
563 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ | |
564 | synchronize_rcu(); | |
565 | } | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
568 | ||
569 | /** | |
570 | * blk_queue_bypass_end - leave queue bypass mode | |
571 | * @q: queue of interest | |
572 | * | |
573 | * Leave bypass mode and restore the normal queueing behavior. | |
574 | * | |
575 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
576 | * this function is called for both blk-sq and blk-mq queues. | |
577 | */ | |
578 | void blk_queue_bypass_end(struct request_queue *q) | |
579 | { | |
580 | spin_lock_irq(q->queue_lock); | |
581 | if (!--q->bypass_depth) | |
582 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
583 | WARN_ON_ONCE(q->bypass_depth < 0); | |
584 | spin_unlock_irq(q->queue_lock); | |
585 | } | |
586 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
587 | ||
588 | void blk_set_queue_dying(struct request_queue *q) | |
589 | { | |
590 | spin_lock_irq(q->queue_lock); | |
591 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
592 | spin_unlock_irq(q->queue_lock); | |
593 | ||
594 | /* | |
595 | * When queue DYING flag is set, we need to block new req | |
596 | * entering queue, so we call blk_freeze_queue_start() to | |
597 | * prevent I/O from crossing blk_queue_enter(). | |
598 | */ | |
599 | blk_freeze_queue_start(q); | |
600 | ||
601 | if (q->mq_ops) | |
602 | blk_mq_wake_waiters(q); | |
603 | else { | |
604 | struct request_list *rl; | |
605 | ||
606 | spin_lock_irq(q->queue_lock); | |
607 | blk_queue_for_each_rl(rl, q) { | |
608 | if (rl->rq_pool) { | |
609 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
610 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
611 | } | |
612 | } | |
613 | spin_unlock_irq(q->queue_lock); | |
614 | } | |
615 | ||
616 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
617 | wake_up_all(&q->mq_freeze_wq); | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
620 | ||
621 | /** | |
622 | * blk_cleanup_queue - shutdown a request queue | |
623 | * @q: request queue to shutdown | |
624 | * | |
625 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and | |
626 | * put it. All future requests will be failed immediately with -ENODEV. | |
627 | */ | |
628 | void blk_cleanup_queue(struct request_queue *q) | |
629 | { | |
630 | spinlock_t *lock = q->queue_lock; | |
631 | ||
632 | /* mark @q DYING, no new request or merges will be allowed afterwards */ | |
633 | mutex_lock(&q->sysfs_lock); | |
634 | blk_set_queue_dying(q); | |
635 | spin_lock_irq(lock); | |
636 | ||
637 | /* | |
638 | * A dying queue is permanently in bypass mode till released. Note | |
639 | * that, unlike blk_queue_bypass_start(), we aren't performing | |
640 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
641 | * as some drivers create and destroy a lot of queues while | |
642 | * probing. This is still safe because blk_release_queue() will be | |
643 | * called only after the queue refcnt drops to zero and nothing, | |
644 | * RCU or not, would be traversing the queue by then. | |
645 | */ | |
646 | q->bypass_depth++; | |
647 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
648 | ||
649 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); | |
650 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
651 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
652 | spin_unlock_irq(lock); | |
653 | mutex_unlock(&q->sysfs_lock); | |
654 | ||
655 | /* | |
656 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
657 | * prevent that q->request_fn() gets invoked after draining finished. | |
658 | */ | |
659 | blk_freeze_queue(q); | |
660 | spin_lock_irq(lock); | |
661 | if (!q->mq_ops) | |
662 | __blk_drain_queue(q, true); | |
663 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
664 | spin_unlock_irq(lock); | |
665 | ||
666 | /* for synchronous bio-based driver finish in-flight integrity i/o */ | |
667 | blk_flush_integrity(); | |
668 | ||
669 | /* @q won't process any more request, flush async actions */ | |
670 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); | |
671 | blk_sync_queue(q); | |
672 | ||
673 | if (q->mq_ops) | |
674 | blk_mq_free_queue(q); | |
675 | percpu_ref_exit(&q->q_usage_counter); | |
676 | ||
677 | spin_lock_irq(lock); | |
678 | if (q->queue_lock != &q->__queue_lock) | |
679 | q->queue_lock = &q->__queue_lock; | |
680 | spin_unlock_irq(lock); | |
681 | ||
682 | /* @q is and will stay empty, shutdown and put */ | |
683 | blk_put_queue(q); | |
684 | } | |
685 | EXPORT_SYMBOL(blk_cleanup_queue); | |
686 | ||
687 | /* Allocate memory local to the request queue */ | |
688 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) | |
689 | { | |
690 | struct request_queue *q = data; | |
691 | ||
692 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
693 | } | |
694 | ||
695 | static void free_request_simple(void *element, void *data) | |
696 | { | |
697 | kmem_cache_free(request_cachep, element); | |
698 | } | |
699 | ||
700 | static void *alloc_request_size(gfp_t gfp_mask, void *data) | |
701 | { | |
702 | struct request_queue *q = data; | |
703 | struct request *rq; | |
704 | ||
705 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
706 | q->node); | |
707 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
708 | kfree(rq); | |
709 | rq = NULL; | |
710 | } | |
711 | return rq; | |
712 | } | |
713 | ||
714 | static void free_request_size(void *element, void *data) | |
715 | { | |
716 | struct request_queue *q = data; | |
717 | ||
718 | if (q->exit_rq_fn) | |
719 | q->exit_rq_fn(q, element); | |
720 | kfree(element); | |
721 | } | |
722 | ||
723 | int blk_init_rl(struct request_list *rl, struct request_queue *q, | |
724 | gfp_t gfp_mask) | |
725 | { | |
726 | if (unlikely(rl->rq_pool) || q->mq_ops) | |
727 | return 0; | |
728 | ||
729 | rl->q = q; | |
730 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; | |
731 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
732 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); | |
733 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
734 | ||
735 | if (q->cmd_size) { | |
736 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
737 | alloc_request_size, free_request_size, | |
738 | q, gfp_mask, q->node); | |
739 | } else { | |
740 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
741 | alloc_request_simple, free_request_simple, | |
742 | q, gfp_mask, q->node); | |
743 | } | |
744 | if (!rl->rq_pool) | |
745 | return -ENOMEM; | |
746 | ||
747 | if (rl != &q->root_rl) | |
748 | WARN_ON_ONCE(!blk_get_queue(q)); | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) | |
754 | { | |
755 | if (rl->rq_pool) { | |
756 | mempool_destroy(rl->rq_pool); | |
757 | if (rl != &q->root_rl) | |
758 | blk_put_queue(q); | |
759 | } | |
760 | } | |
761 | ||
762 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
763 | { | |
764 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); | |
765 | } | |
766 | EXPORT_SYMBOL(blk_alloc_queue); | |
767 | ||
768 | int blk_queue_enter(struct request_queue *q, bool nowait) | |
769 | { | |
770 | while (true) { | |
771 | int ret; | |
772 | ||
773 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
774 | return 0; | |
775 | ||
776 | if (nowait) | |
777 | return -EBUSY; | |
778 | ||
779 | /* | |
780 | * read pair of barrier in blk_freeze_queue_start(), | |
781 | * we need to order reading __PERCPU_REF_DEAD flag of | |
782 | * .q_usage_counter and reading .mq_freeze_depth or | |
783 | * queue dying flag, otherwise the following wait may | |
784 | * never return if the two reads are reordered. | |
785 | */ | |
786 | smp_rmb(); | |
787 | ||
788 | ret = wait_event_interruptible(q->mq_freeze_wq, | |
789 | !atomic_read(&q->mq_freeze_depth) || | |
790 | blk_queue_dying(q)); | |
791 | if (blk_queue_dying(q)) | |
792 | return -ENODEV; | |
793 | if (ret) | |
794 | return ret; | |
795 | } | |
796 | } | |
797 | ||
798 | void blk_queue_exit(struct request_queue *q) | |
799 | { | |
800 | percpu_ref_put(&q->q_usage_counter); | |
801 | } | |
802 | ||
803 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
804 | { | |
805 | struct request_queue *q = | |
806 | container_of(ref, struct request_queue, q_usage_counter); | |
807 | ||
808 | wake_up_all(&q->mq_freeze_wq); | |
809 | } | |
810 | ||
811 | static void blk_rq_timed_out_timer(unsigned long data) | |
812 | { | |
813 | struct request_queue *q = (struct request_queue *)data; | |
814 | ||
815 | kblockd_schedule_work(&q->timeout_work); | |
816 | } | |
817 | ||
818 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
819 | { | |
820 | struct request_queue *q; | |
821 | ||
822 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
823 | gfp_mask | __GFP_ZERO, node_id); | |
824 | if (!q) | |
825 | return NULL; | |
826 | ||
827 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); | |
828 | if (q->id < 0) | |
829 | goto fail_q; | |
830 | ||
831 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
832 | if (!q->bio_split) | |
833 | goto fail_id; | |
834 | ||
835 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); | |
836 | if (!q->backing_dev_info) | |
837 | goto fail_split; | |
838 | ||
839 | q->stats = blk_alloc_queue_stats(); | |
840 | if (!q->stats) | |
841 | goto fail_stats; | |
842 | ||
843 | q->backing_dev_info->ra_pages = | |
844 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; | |
845 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; | |
846 | q->backing_dev_info->name = "block"; | |
847 | q->node = node_id; | |
848 | ||
849 | setup_timer(&q->backing_dev_info->laptop_mode_wb_timer, | |
850 | laptop_mode_timer_fn, (unsigned long) q); | |
851 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); | |
852 | INIT_WORK(&q->timeout_work, NULL); | |
853 | INIT_LIST_HEAD(&q->queue_head); | |
854 | INIT_LIST_HEAD(&q->timeout_list); | |
855 | INIT_LIST_HEAD(&q->icq_list); | |
856 | #ifdef CONFIG_BLK_CGROUP | |
857 | INIT_LIST_HEAD(&q->blkg_list); | |
858 | #endif | |
859 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); | |
860 | ||
861 | kobject_init(&q->kobj, &blk_queue_ktype); | |
862 | ||
863 | #ifdef CONFIG_BLK_DEV_IO_TRACE | |
864 | mutex_init(&q->blk_trace_mutex); | |
865 | #endif | |
866 | mutex_init(&q->sysfs_lock); | |
867 | spin_lock_init(&q->__queue_lock); | |
868 | ||
869 | /* | |
870 | * By default initialize queue_lock to internal lock and driver can | |
871 | * override it later if need be. | |
872 | */ | |
873 | q->queue_lock = &q->__queue_lock; | |
874 | ||
875 | /* | |
876 | * A queue starts its life with bypass turned on to avoid | |
877 | * unnecessary bypass on/off overhead and nasty surprises during | |
878 | * init. The initial bypass will be finished when the queue is | |
879 | * registered by blk_register_queue(). | |
880 | */ | |
881 | q->bypass_depth = 1; | |
882 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
883 | ||
884 | init_waitqueue_head(&q->mq_freeze_wq); | |
885 | ||
886 | /* | |
887 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
888 | * See blk_register_queue() for details. | |
889 | */ | |
890 | if (percpu_ref_init(&q->q_usage_counter, | |
891 | blk_queue_usage_counter_release, | |
892 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
893 | goto fail_bdi; | |
894 | ||
895 | if (blkcg_init_queue(q)) | |
896 | goto fail_ref; | |
897 | ||
898 | return q; | |
899 | ||
900 | fail_ref: | |
901 | percpu_ref_exit(&q->q_usage_counter); | |
902 | fail_bdi: | |
903 | blk_free_queue_stats(q->stats); | |
904 | fail_stats: | |
905 | bdi_put(q->backing_dev_info); | |
906 | fail_split: | |
907 | bioset_free(q->bio_split); | |
908 | fail_id: | |
909 | ida_simple_remove(&blk_queue_ida, q->id); | |
910 | fail_q: | |
911 | kmem_cache_free(blk_requestq_cachep, q); | |
912 | return NULL; | |
913 | } | |
914 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
915 | ||
916 | /** | |
917 | * blk_init_queue - prepare a request queue for use with a block device | |
918 | * @rfn: The function to be called to process requests that have been | |
919 | * placed on the queue. | |
920 | * @lock: Request queue spin lock | |
921 | * | |
922 | * Description: | |
923 | * If a block device wishes to use the standard request handling procedures, | |
924 | * which sorts requests and coalesces adjacent requests, then it must | |
925 | * call blk_init_queue(). The function @rfn will be called when there | |
926 | * are requests on the queue that need to be processed. If the device | |
927 | * supports plugging, then @rfn may not be called immediately when requests | |
928 | * are available on the queue, but may be called at some time later instead. | |
929 | * Plugged queues are generally unplugged when a buffer belonging to one | |
930 | * of the requests on the queue is needed, or due to memory pressure. | |
931 | * | |
932 | * @rfn is not required, or even expected, to remove all requests off the | |
933 | * queue, but only as many as it can handle at a time. If it does leave | |
934 | * requests on the queue, it is responsible for arranging that the requests | |
935 | * get dealt with eventually. | |
936 | * | |
937 | * The queue spin lock must be held while manipulating the requests on the | |
938 | * request queue; this lock will be taken also from interrupt context, so irq | |
939 | * disabling is needed for it. | |
940 | * | |
941 | * Function returns a pointer to the initialized request queue, or %NULL if | |
942 | * it didn't succeed. | |
943 | * | |
944 | * Note: | |
945 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
946 | * when the block device is deactivated (such as at module unload). | |
947 | **/ | |
948 | ||
949 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
950 | { | |
951 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); | |
952 | } | |
953 | EXPORT_SYMBOL(blk_init_queue); | |
954 | ||
955 | struct request_queue * | |
956 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
957 | { | |
958 | struct request_queue *q; | |
959 | ||
960 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
961 | if (!q) | |
962 | return NULL; | |
963 | ||
964 | q->request_fn = rfn; | |
965 | if (lock) | |
966 | q->queue_lock = lock; | |
967 | if (blk_init_allocated_queue(q) < 0) { | |
968 | blk_cleanup_queue(q); | |
969 | return NULL; | |
970 | } | |
971 | ||
972 | return q; | |
973 | } | |
974 | EXPORT_SYMBOL(blk_init_queue_node); | |
975 | ||
976 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); | |
977 | ||
978 | ||
979 | int blk_init_allocated_queue(struct request_queue *q) | |
980 | { | |
981 | WARN_ON_ONCE(q->mq_ops); | |
982 | ||
983 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); | |
984 | if (!q->fq) | |
985 | return -ENOMEM; | |
986 | ||
987 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) | |
988 | goto out_free_flush_queue; | |
989 | ||
990 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) | |
991 | goto out_exit_flush_rq; | |
992 | ||
993 | INIT_WORK(&q->timeout_work, blk_timeout_work); | |
994 | q->queue_flags |= QUEUE_FLAG_DEFAULT; | |
995 | ||
996 | /* | |
997 | * This also sets hw/phys segments, boundary and size | |
998 | */ | |
999 | blk_queue_make_request(q, blk_queue_bio); | |
1000 | ||
1001 | q->sg_reserved_size = INT_MAX; | |
1002 | ||
1003 | /* Protect q->elevator from elevator_change */ | |
1004 | mutex_lock(&q->sysfs_lock); | |
1005 | ||
1006 | /* init elevator */ | |
1007 | if (elevator_init(q, NULL)) { | |
1008 | mutex_unlock(&q->sysfs_lock); | |
1009 | goto out_exit_flush_rq; | |
1010 | } | |
1011 | ||
1012 | mutex_unlock(&q->sysfs_lock); | |
1013 | return 0; | |
1014 | ||
1015 | out_exit_flush_rq: | |
1016 | if (q->exit_rq_fn) | |
1017 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1018 | out_free_flush_queue: | |
1019 | blk_free_flush_queue(q->fq); | |
1020 | return -ENOMEM; | |
1021 | } | |
1022 | EXPORT_SYMBOL(blk_init_allocated_queue); | |
1023 | ||
1024 | bool blk_get_queue(struct request_queue *q) | |
1025 | { | |
1026 | if (likely(!blk_queue_dying(q))) { | |
1027 | __blk_get_queue(q); | |
1028 | return true; | |
1029 | } | |
1030 | ||
1031 | return false; | |
1032 | } | |
1033 | EXPORT_SYMBOL(blk_get_queue); | |
1034 | ||
1035 | static inline void blk_free_request(struct request_list *rl, struct request *rq) | |
1036 | { | |
1037 | if (rq->rq_flags & RQF_ELVPRIV) { | |
1038 | elv_put_request(rl->q, rq); | |
1039 | if (rq->elv.icq) | |
1040 | put_io_context(rq->elv.icq->ioc); | |
1041 | } | |
1042 | ||
1043 | mempool_free(rq, rl->rq_pool); | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * ioc_batching returns true if the ioc is a valid batching request and | |
1048 | * should be given priority access to a request. | |
1049 | */ | |
1050 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
1051 | { | |
1052 | if (!ioc) | |
1053 | return 0; | |
1054 | ||
1055 | /* | |
1056 | * Make sure the process is able to allocate at least 1 request | |
1057 | * even if the batch times out, otherwise we could theoretically | |
1058 | * lose wakeups. | |
1059 | */ | |
1060 | return ioc->nr_batch_requests == q->nr_batching || | |
1061 | (ioc->nr_batch_requests > 0 | |
1062 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1067 | * will cause the process to be a "batcher" on all queues in the system. This | |
1068 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1069 | * a nice run. | |
1070 | */ | |
1071 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
1072 | { | |
1073 | if (!ioc || ioc_batching(q, ioc)) | |
1074 | return; | |
1075 | ||
1076 | ioc->nr_batch_requests = q->nr_batching; | |
1077 | ioc->last_waited = jiffies; | |
1078 | } | |
1079 | ||
1080 | static void __freed_request(struct request_list *rl, int sync) | |
1081 | { | |
1082 | struct request_queue *q = rl->q; | |
1083 | ||
1084 | if (rl->count[sync] < queue_congestion_off_threshold(q)) | |
1085 | blk_clear_congested(rl, sync); | |
1086 | ||
1087 | if (rl->count[sync] + 1 <= q->nr_requests) { | |
1088 | if (waitqueue_active(&rl->wait[sync])) | |
1089 | wake_up(&rl->wait[sync]); | |
1090 | ||
1091 | blk_clear_rl_full(rl, sync); | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * A request has just been released. Account for it, update the full and | |
1097 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1098 | */ | |
1099 | static void freed_request(struct request_list *rl, bool sync, | |
1100 | req_flags_t rq_flags) | |
1101 | { | |
1102 | struct request_queue *q = rl->q; | |
1103 | ||
1104 | q->nr_rqs[sync]--; | |
1105 | rl->count[sync]--; | |
1106 | if (rq_flags & RQF_ELVPRIV) | |
1107 | q->nr_rqs_elvpriv--; | |
1108 | ||
1109 | __freed_request(rl, sync); | |
1110 | ||
1111 | if (unlikely(rl->starved[sync ^ 1])) | |
1112 | __freed_request(rl, sync ^ 1); | |
1113 | } | |
1114 | ||
1115 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) | |
1116 | { | |
1117 | struct request_list *rl; | |
1118 | int on_thresh, off_thresh; | |
1119 | ||
1120 | WARN_ON_ONCE(q->mq_ops); | |
1121 | ||
1122 | spin_lock_irq(q->queue_lock); | |
1123 | q->nr_requests = nr; | |
1124 | blk_queue_congestion_threshold(q); | |
1125 | on_thresh = queue_congestion_on_threshold(q); | |
1126 | off_thresh = queue_congestion_off_threshold(q); | |
1127 | ||
1128 | blk_queue_for_each_rl(rl, q) { | |
1129 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1130 | blk_set_congested(rl, BLK_RW_SYNC); | |
1131 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1132 | blk_clear_congested(rl, BLK_RW_SYNC); | |
1133 | ||
1134 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) | |
1135 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1136 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1137 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
1138 | ||
1139 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { | |
1140 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1141 | } else { | |
1142 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1143 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1144 | } | |
1145 | ||
1146 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1147 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1148 | } else { | |
1149 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1150 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1151 | } | |
1152 | } | |
1153 | ||
1154 | spin_unlock_irq(q->queue_lock); | |
1155 | return 0; | |
1156 | } | |
1157 | ||
1158 | /** | |
1159 | * __get_request - get a free request | |
1160 | * @rl: request list to allocate from | |
1161 | * @op: operation and flags | |
1162 | * @bio: bio to allocate request for (can be %NULL) | |
1163 | * @flags: BLQ_MQ_REQ_* flags | |
1164 | * | |
1165 | * Get a free request from @q. This function may fail under memory | |
1166 | * pressure or if @q is dead. | |
1167 | * | |
1168 | * Must be called with @q->queue_lock held and, | |
1169 | * Returns ERR_PTR on failure, with @q->queue_lock held. | |
1170 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1171 | */ | |
1172 | static struct request *__get_request(struct request_list *rl, unsigned int op, | |
1173 | struct bio *bio, unsigned int flags) | |
1174 | { | |
1175 | struct request_queue *q = rl->q; | |
1176 | struct request *rq; | |
1177 | struct elevator_type *et = q->elevator->type; | |
1178 | struct io_context *ioc = rq_ioc(bio); | |
1179 | struct io_cq *icq = NULL; | |
1180 | const bool is_sync = op_is_sync(op); | |
1181 | int may_queue; | |
1182 | gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : | |
1183 | __GFP_DIRECT_RECLAIM; | |
1184 | req_flags_t rq_flags = RQF_ALLOCED; | |
1185 | ||
1186 | lockdep_assert_held(q->queue_lock); | |
1187 | ||
1188 | if (unlikely(blk_queue_dying(q))) | |
1189 | return ERR_PTR(-ENODEV); | |
1190 | ||
1191 | may_queue = elv_may_queue(q, op); | |
1192 | if (may_queue == ELV_MQUEUE_NO) | |
1193 | goto rq_starved; | |
1194 | ||
1195 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { | |
1196 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
1197 | /* | |
1198 | * The queue will fill after this allocation, so set | |
1199 | * it as full, and mark this process as "batching". | |
1200 | * This process will be allowed to complete a batch of | |
1201 | * requests, others will be blocked. | |
1202 | */ | |
1203 | if (!blk_rl_full(rl, is_sync)) { | |
1204 | ioc_set_batching(q, ioc); | |
1205 | blk_set_rl_full(rl, is_sync); | |
1206 | } else { | |
1207 | if (may_queue != ELV_MQUEUE_MUST | |
1208 | && !ioc_batching(q, ioc)) { | |
1209 | /* | |
1210 | * The queue is full and the allocating | |
1211 | * process is not a "batcher", and not | |
1212 | * exempted by the IO scheduler | |
1213 | */ | |
1214 | return ERR_PTR(-ENOMEM); | |
1215 | } | |
1216 | } | |
1217 | } | |
1218 | blk_set_congested(rl, is_sync); | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Only allow batching queuers to allocate up to 50% over the defined | |
1223 | * limit of requests, otherwise we could have thousands of requests | |
1224 | * allocated with any setting of ->nr_requests | |
1225 | */ | |
1226 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) | |
1227 | return ERR_PTR(-ENOMEM); | |
1228 | ||
1229 | q->nr_rqs[is_sync]++; | |
1230 | rl->count[is_sync]++; | |
1231 | rl->starved[is_sync] = 0; | |
1232 | ||
1233 | /* | |
1234 | * Decide whether the new request will be managed by elevator. If | |
1235 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will | |
1236 | * prevent the current elevator from being destroyed until the new | |
1237 | * request is freed. This guarantees icq's won't be destroyed and | |
1238 | * makes creating new ones safe. | |
1239 | * | |
1240 | * Flush requests do not use the elevator so skip initialization. | |
1241 | * This allows a request to share the flush and elevator data. | |
1242 | * | |
1243 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
1244 | * it will be created after releasing queue_lock. | |
1245 | */ | |
1246 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { | |
1247 | rq_flags |= RQF_ELVPRIV; | |
1248 | q->nr_rqs_elvpriv++; | |
1249 | if (et->icq_cache && ioc) | |
1250 | icq = ioc_lookup_icq(ioc, q); | |
1251 | } | |
1252 | ||
1253 | if (blk_queue_io_stat(q)) | |
1254 | rq_flags |= RQF_IO_STAT; | |
1255 | spin_unlock_irq(q->queue_lock); | |
1256 | ||
1257 | /* allocate and init request */ | |
1258 | rq = mempool_alloc(rl->rq_pool, gfp_mask); | |
1259 | if (!rq) | |
1260 | goto fail_alloc; | |
1261 | ||
1262 | blk_rq_init(q, rq); | |
1263 | blk_rq_set_rl(rq, rl); | |
1264 | rq->cmd_flags = op; | |
1265 | rq->rq_flags = rq_flags; | |
1266 | ||
1267 | /* init elvpriv */ | |
1268 | if (rq_flags & RQF_ELVPRIV) { | |
1269 | if (unlikely(et->icq_cache && !icq)) { | |
1270 | if (ioc) | |
1271 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
1272 | if (!icq) | |
1273 | goto fail_elvpriv; | |
1274 | } | |
1275 | ||
1276 | rq->elv.icq = icq; | |
1277 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1278 | goto fail_elvpriv; | |
1279 | ||
1280 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
1281 | if (icq) | |
1282 | get_io_context(icq->ioc); | |
1283 | } | |
1284 | out: | |
1285 | /* | |
1286 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1287 | * OK, if the queue is under the request limit then requests need | |
1288 | * not count toward the nr_batch_requests limit. There will always | |
1289 | * be some limit enforced by BLK_BATCH_TIME. | |
1290 | */ | |
1291 | if (ioc_batching(q, ioc)) | |
1292 | ioc->nr_batch_requests--; | |
1293 | ||
1294 | trace_block_getrq(q, bio, op); | |
1295 | return rq; | |
1296 | ||
1297 | fail_elvpriv: | |
1298 | /* | |
1299 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1300 | * and may fail indefinitely under memory pressure and thus | |
1301 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1302 | * disturb iosched and blkcg but weird is bettern than dead. | |
1303 | */ | |
1304 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", | |
1305 | __func__, dev_name(q->backing_dev_info->dev)); | |
1306 | ||
1307 | rq->rq_flags &= ~RQF_ELVPRIV; | |
1308 | rq->elv.icq = NULL; | |
1309 | ||
1310 | spin_lock_irq(q->queue_lock); | |
1311 | q->nr_rqs_elvpriv--; | |
1312 | spin_unlock_irq(q->queue_lock); | |
1313 | goto out; | |
1314 | ||
1315 | fail_alloc: | |
1316 | /* | |
1317 | * Allocation failed presumably due to memory. Undo anything we | |
1318 | * might have messed up. | |
1319 | * | |
1320 | * Allocating task should really be put onto the front of the wait | |
1321 | * queue, but this is pretty rare. | |
1322 | */ | |
1323 | spin_lock_irq(q->queue_lock); | |
1324 | freed_request(rl, is_sync, rq_flags); | |
1325 | ||
1326 | /* | |
1327 | * in the very unlikely event that allocation failed and no | |
1328 | * requests for this direction was pending, mark us starved so that | |
1329 | * freeing of a request in the other direction will notice | |
1330 | * us. another possible fix would be to split the rq mempool into | |
1331 | * READ and WRITE | |
1332 | */ | |
1333 | rq_starved: | |
1334 | if (unlikely(rl->count[is_sync] == 0)) | |
1335 | rl->starved[is_sync] = 1; | |
1336 | return ERR_PTR(-ENOMEM); | |
1337 | } | |
1338 | ||
1339 | /** | |
1340 | * get_request - get a free request | |
1341 | * @q: request_queue to allocate request from | |
1342 | * @op: operation and flags | |
1343 | * @bio: bio to allocate request for (can be %NULL) | |
1344 | * @flags: BLK_MQ_REQ_* flags. | |
1345 | * | |
1346 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, | |
1347 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
1348 | * | |
1349 | * Must be called with @q->queue_lock held and, | |
1350 | * Returns ERR_PTR on failure, with @q->queue_lock held. | |
1351 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1352 | */ | |
1353 | static struct request *get_request(struct request_queue *q, unsigned int op, | |
1354 | struct bio *bio, unsigned int flags) | |
1355 | { | |
1356 | const bool is_sync = op_is_sync(op); | |
1357 | DEFINE_WAIT(wait); | |
1358 | struct request_list *rl; | |
1359 | struct request *rq; | |
1360 | ||
1361 | lockdep_assert_held(q->queue_lock); | |
1362 | WARN_ON_ONCE(q->mq_ops); | |
1363 | ||
1364 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
1365 | retry: | |
1366 | rq = __get_request(rl, op, bio, flags); | |
1367 | if (!IS_ERR(rq)) | |
1368 | return rq; | |
1369 | ||
1370 | if (op & REQ_NOWAIT) { | |
1371 | blk_put_rl(rl); | |
1372 | return ERR_PTR(-EAGAIN); | |
1373 | } | |
1374 | ||
1375 | if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) { | |
1376 | blk_put_rl(rl); | |
1377 | return rq; | |
1378 | } | |
1379 | ||
1380 | /* wait on @rl and retry */ | |
1381 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1382 | TASK_UNINTERRUPTIBLE); | |
1383 | ||
1384 | trace_block_sleeprq(q, bio, op); | |
1385 | ||
1386 | spin_unlock_irq(q->queue_lock); | |
1387 | io_schedule(); | |
1388 | ||
1389 | /* | |
1390 | * After sleeping, we become a "batching" process and will be able | |
1391 | * to allocate at least one request, and up to a big batch of them | |
1392 | * for a small period time. See ioc_batching, ioc_set_batching | |
1393 | */ | |
1394 | ioc_set_batching(q, current->io_context); | |
1395 | ||
1396 | spin_lock_irq(q->queue_lock); | |
1397 | finish_wait(&rl->wait[is_sync], &wait); | |
1398 | ||
1399 | goto retry; | |
1400 | } | |
1401 | ||
1402 | /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */ | |
1403 | static struct request *blk_old_get_request(struct request_queue *q, | |
1404 | unsigned int op, unsigned int flags) | |
1405 | { | |
1406 | struct request *rq; | |
1407 | gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : | |
1408 | __GFP_DIRECT_RECLAIM; | |
1409 | int ret = 0; | |
1410 | ||
1411 | WARN_ON_ONCE(q->mq_ops); | |
1412 | ||
1413 | /* create ioc upfront */ | |
1414 | create_io_context(gfp_mask, q->node); | |
1415 | ||
1416 | ret = blk_queue_enter(q, !(gfp_mask & __GFP_DIRECT_RECLAIM) || | |
1417 | (op & REQ_NOWAIT)); | |
1418 | if (ret) | |
1419 | return ERR_PTR(ret); | |
1420 | spin_lock_irq(q->queue_lock); | |
1421 | rq = get_request(q, op, NULL, flags); | |
1422 | if (IS_ERR(rq)) { | |
1423 | spin_unlock_irq(q->queue_lock); | |
1424 | blk_queue_exit(q); | |
1425 | return rq; | |
1426 | } | |
1427 | ||
1428 | /* q->queue_lock is unlocked at this point */ | |
1429 | rq->__data_len = 0; | |
1430 | rq->__sector = (sector_t) -1; | |
1431 | rq->bio = rq->biotail = NULL; | |
1432 | return rq; | |
1433 | } | |
1434 | ||
1435 | /** | |
1436 | * blk_get_request_flags - allocate a request | |
1437 | * @q: request queue to allocate a request for | |
1438 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
1439 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1440 | */ | |
1441 | struct request *blk_get_request_flags(struct request_queue *q, unsigned int op, | |
1442 | unsigned int flags) | |
1443 | { | |
1444 | struct request *req; | |
1445 | ||
1446 | WARN_ON_ONCE(op & REQ_NOWAIT); | |
1447 | WARN_ON_ONCE(flags & ~BLK_MQ_REQ_NOWAIT); | |
1448 | ||
1449 | if (q->mq_ops) { | |
1450 | req = blk_mq_alloc_request(q, op, flags); | |
1451 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) | |
1452 | q->mq_ops->initialize_rq_fn(req); | |
1453 | } else { | |
1454 | req = blk_old_get_request(q, op, flags); | |
1455 | if (!IS_ERR(req) && q->initialize_rq_fn) | |
1456 | q->initialize_rq_fn(req); | |
1457 | } | |
1458 | ||
1459 | return req; | |
1460 | } | |
1461 | EXPORT_SYMBOL(blk_get_request_flags); | |
1462 | ||
1463 | struct request *blk_get_request(struct request_queue *q, unsigned int op, | |
1464 | gfp_t gfp_mask) | |
1465 | { | |
1466 | return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ? | |
1467 | 0 : BLK_MQ_REQ_NOWAIT); | |
1468 | } | |
1469 | EXPORT_SYMBOL(blk_get_request); | |
1470 | ||
1471 | /** | |
1472 | * blk_requeue_request - put a request back on queue | |
1473 | * @q: request queue where request should be inserted | |
1474 | * @rq: request to be inserted | |
1475 | * | |
1476 | * Description: | |
1477 | * Drivers often keep queueing requests until the hardware cannot accept | |
1478 | * more, when that condition happens we need to put the request back | |
1479 | * on the queue. Must be called with queue lock held. | |
1480 | */ | |
1481 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
1482 | { | |
1483 | lockdep_assert_held(q->queue_lock); | |
1484 | WARN_ON_ONCE(q->mq_ops); | |
1485 | ||
1486 | blk_delete_timer(rq); | |
1487 | blk_clear_rq_complete(rq); | |
1488 | trace_block_rq_requeue(q, rq); | |
1489 | wbt_requeue(q->rq_wb, &rq->issue_stat); | |
1490 | ||
1491 | if (rq->rq_flags & RQF_QUEUED) | |
1492 | blk_queue_end_tag(q, rq); | |
1493 | ||
1494 | BUG_ON(blk_queued_rq(rq)); | |
1495 | ||
1496 | elv_requeue_request(q, rq); | |
1497 | } | |
1498 | EXPORT_SYMBOL(blk_requeue_request); | |
1499 | ||
1500 | static void add_acct_request(struct request_queue *q, struct request *rq, | |
1501 | int where) | |
1502 | { | |
1503 | blk_account_io_start(rq, true); | |
1504 | __elv_add_request(q, rq, where); | |
1505 | } | |
1506 | ||
1507 | static void part_round_stats_single(struct request_queue *q, int cpu, | |
1508 | struct hd_struct *part, unsigned long now, | |
1509 | unsigned int inflight) | |
1510 | { | |
1511 | if (inflight) { | |
1512 | __part_stat_add(cpu, part, time_in_queue, | |
1513 | inflight * (now - part->stamp)); | |
1514 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1515 | } | |
1516 | part->stamp = now; | |
1517 | } | |
1518 | ||
1519 | /** | |
1520 | * part_round_stats() - Round off the performance stats on a struct disk_stats. | |
1521 | * @q: target block queue | |
1522 | * @cpu: cpu number for stats access | |
1523 | * @part: target partition | |
1524 | * | |
1525 | * The average IO queue length and utilisation statistics are maintained | |
1526 | * by observing the current state of the queue length and the amount of | |
1527 | * time it has been in this state for. | |
1528 | * | |
1529 | * Normally, that accounting is done on IO completion, but that can result | |
1530 | * in more than a second's worth of IO being accounted for within any one | |
1531 | * second, leading to >100% utilisation. To deal with that, we call this | |
1532 | * function to do a round-off before returning the results when reading | |
1533 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1534 | * the current jiffies and restarts the counters again. | |
1535 | */ | |
1536 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) | |
1537 | { | |
1538 | struct hd_struct *part2 = NULL; | |
1539 | unsigned long now = jiffies; | |
1540 | unsigned int inflight[2]; | |
1541 | int stats = 0; | |
1542 | ||
1543 | if (part->stamp != now) | |
1544 | stats |= 1; | |
1545 | ||
1546 | if (part->partno) { | |
1547 | part2 = &part_to_disk(part)->part0; | |
1548 | if (part2->stamp != now) | |
1549 | stats |= 2; | |
1550 | } | |
1551 | ||
1552 | if (!stats) | |
1553 | return; | |
1554 | ||
1555 | part_in_flight(q, part, inflight); | |
1556 | ||
1557 | if (stats & 2) | |
1558 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1559 | if (stats & 1) | |
1560 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
1561 | } | |
1562 | EXPORT_SYMBOL_GPL(part_round_stats); | |
1563 | ||
1564 | #ifdef CONFIG_PM | |
1565 | static void blk_pm_put_request(struct request *rq) | |
1566 | { | |
1567 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) | |
1568 | pm_runtime_mark_last_busy(rq->q->dev); | |
1569 | } | |
1570 | #else | |
1571 | static inline void blk_pm_put_request(struct request *rq) {} | |
1572 | #endif | |
1573 | ||
1574 | void __blk_put_request(struct request_queue *q, struct request *req) | |
1575 | { | |
1576 | req_flags_t rq_flags = req->rq_flags; | |
1577 | ||
1578 | if (unlikely(!q)) | |
1579 | return; | |
1580 | ||
1581 | if (q->mq_ops) { | |
1582 | blk_mq_free_request(req); | |
1583 | return; | |
1584 | } | |
1585 | ||
1586 | lockdep_assert_held(q->queue_lock); | |
1587 | ||
1588 | blk_pm_put_request(req); | |
1589 | ||
1590 | elv_completed_request(q, req); | |
1591 | ||
1592 | /* this is a bio leak */ | |
1593 | WARN_ON(req->bio != NULL); | |
1594 | ||
1595 | wbt_done(q->rq_wb, &req->issue_stat); | |
1596 | ||
1597 | /* | |
1598 | * Request may not have originated from ll_rw_blk. if not, | |
1599 | * it didn't come out of our reserved rq pools | |
1600 | */ | |
1601 | if (rq_flags & RQF_ALLOCED) { | |
1602 | struct request_list *rl = blk_rq_rl(req); | |
1603 | bool sync = op_is_sync(req->cmd_flags); | |
1604 | ||
1605 | BUG_ON(!list_empty(&req->queuelist)); | |
1606 | BUG_ON(ELV_ON_HASH(req)); | |
1607 | ||
1608 | blk_free_request(rl, req); | |
1609 | freed_request(rl, sync, rq_flags); | |
1610 | blk_put_rl(rl); | |
1611 | blk_queue_exit(q); | |
1612 | } | |
1613 | } | |
1614 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
1615 | ||
1616 | void blk_put_request(struct request *req) | |
1617 | { | |
1618 | struct request_queue *q = req->q; | |
1619 | ||
1620 | if (q->mq_ops) | |
1621 | blk_mq_free_request(req); | |
1622 | else { | |
1623 | unsigned long flags; | |
1624 | ||
1625 | spin_lock_irqsave(q->queue_lock, flags); | |
1626 | __blk_put_request(q, req); | |
1627 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1628 | } | |
1629 | } | |
1630 | EXPORT_SYMBOL(blk_put_request); | |
1631 | ||
1632 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, | |
1633 | struct bio *bio) | |
1634 | { | |
1635 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
1636 | ||
1637 | if (!ll_back_merge_fn(q, req, bio)) | |
1638 | return false; | |
1639 | ||
1640 | trace_block_bio_backmerge(q, req, bio); | |
1641 | ||
1642 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1643 | blk_rq_set_mixed_merge(req); | |
1644 | ||
1645 | req->biotail->bi_next = bio; | |
1646 | req->biotail = bio; | |
1647 | req->__data_len += bio->bi_iter.bi_size; | |
1648 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1649 | ||
1650 | blk_account_io_start(req, false); | |
1651 | return true; | |
1652 | } | |
1653 | ||
1654 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, | |
1655 | struct bio *bio) | |
1656 | { | |
1657 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
1658 | ||
1659 | if (!ll_front_merge_fn(q, req, bio)) | |
1660 | return false; | |
1661 | ||
1662 | trace_block_bio_frontmerge(q, req, bio); | |
1663 | ||
1664 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1665 | blk_rq_set_mixed_merge(req); | |
1666 | ||
1667 | bio->bi_next = req->bio; | |
1668 | req->bio = bio; | |
1669 | ||
1670 | req->__sector = bio->bi_iter.bi_sector; | |
1671 | req->__data_len += bio->bi_iter.bi_size; | |
1672 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1673 | ||
1674 | blk_account_io_start(req, false); | |
1675 | return true; | |
1676 | } | |
1677 | ||
1678 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, | |
1679 | struct bio *bio) | |
1680 | { | |
1681 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1682 | ||
1683 | if (segments >= queue_max_discard_segments(q)) | |
1684 | goto no_merge; | |
1685 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1686 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1687 | goto no_merge; | |
1688 | ||
1689 | req->biotail->bi_next = bio; | |
1690 | req->biotail = bio; | |
1691 | req->__data_len += bio->bi_iter.bi_size; | |
1692 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1693 | req->nr_phys_segments = segments + 1; | |
1694 | ||
1695 | blk_account_io_start(req, false); | |
1696 | return true; | |
1697 | no_merge: | |
1698 | req_set_nomerge(q, req); | |
1699 | return false; | |
1700 | } | |
1701 | ||
1702 | /** | |
1703 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1704 | * @q: request_queue new bio is being queued at | |
1705 | * @bio: new bio being queued | |
1706 | * @request_count: out parameter for number of traversed plugged requests | |
1707 | * @same_queue_rq: pointer to &struct request that gets filled in when | |
1708 | * another request associated with @q is found on the plug list | |
1709 | * (optional, may be %NULL) | |
1710 | * | |
1711 | * Determine whether @bio being queued on @q can be merged with a request | |
1712 | * on %current's plugged list. Returns %true if merge was successful, | |
1713 | * otherwise %false. | |
1714 | * | |
1715 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1716 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1717 | * than scheduling, and the request, while may have elvpriv data, is not | |
1718 | * added on the elevator at this point. In addition, we don't have | |
1719 | * reliable access to the elevator outside queue lock. Only check basic | |
1720 | * merging parameters without querying the elevator. | |
1721 | * | |
1722 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1723 | */ | |
1724 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
1725 | unsigned int *request_count, | |
1726 | struct request **same_queue_rq) | |
1727 | { | |
1728 | struct blk_plug *plug; | |
1729 | struct request *rq; | |
1730 | struct list_head *plug_list; | |
1731 | ||
1732 | plug = current->plug; | |
1733 | if (!plug) | |
1734 | return false; | |
1735 | *request_count = 0; | |
1736 | ||
1737 | if (q->mq_ops) | |
1738 | plug_list = &plug->mq_list; | |
1739 | else | |
1740 | plug_list = &plug->list; | |
1741 | ||
1742 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
1743 | bool merged = false; | |
1744 | ||
1745 | if (rq->q == q) { | |
1746 | (*request_count)++; | |
1747 | /* | |
1748 | * Only blk-mq multiple hardware queues case checks the | |
1749 | * rq in the same queue, there should be only one such | |
1750 | * rq in a queue | |
1751 | **/ | |
1752 | if (same_queue_rq) | |
1753 | *same_queue_rq = rq; | |
1754 | } | |
1755 | ||
1756 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) | |
1757 | continue; | |
1758 | ||
1759 | switch (blk_try_merge(rq, bio)) { | |
1760 | case ELEVATOR_BACK_MERGE: | |
1761 | merged = bio_attempt_back_merge(q, rq, bio); | |
1762 | break; | |
1763 | case ELEVATOR_FRONT_MERGE: | |
1764 | merged = bio_attempt_front_merge(q, rq, bio); | |
1765 | break; | |
1766 | case ELEVATOR_DISCARD_MERGE: | |
1767 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1768 | break; | |
1769 | default: | |
1770 | break; | |
1771 | } | |
1772 | ||
1773 | if (merged) | |
1774 | return true; | |
1775 | } | |
1776 | ||
1777 | return false; | |
1778 | } | |
1779 | ||
1780 | unsigned int blk_plug_queued_count(struct request_queue *q) | |
1781 | { | |
1782 | struct blk_plug *plug; | |
1783 | struct request *rq; | |
1784 | struct list_head *plug_list; | |
1785 | unsigned int ret = 0; | |
1786 | ||
1787 | plug = current->plug; | |
1788 | if (!plug) | |
1789 | goto out; | |
1790 | ||
1791 | if (q->mq_ops) | |
1792 | plug_list = &plug->mq_list; | |
1793 | else | |
1794 | plug_list = &plug->list; | |
1795 | ||
1796 | list_for_each_entry(rq, plug_list, queuelist) { | |
1797 | if (rq->q == q) | |
1798 | ret++; | |
1799 | } | |
1800 | out: | |
1801 | return ret; | |
1802 | } | |
1803 | ||
1804 | void blk_init_request_from_bio(struct request *req, struct bio *bio) | |
1805 | { | |
1806 | struct io_context *ioc = rq_ioc(bio); | |
1807 | ||
1808 | if (bio->bi_opf & REQ_RAHEAD) | |
1809 | req->cmd_flags |= REQ_FAILFAST_MASK; | |
1810 | ||
1811 | req->__sector = bio->bi_iter.bi_sector; | |
1812 | if (ioprio_valid(bio_prio(bio))) | |
1813 | req->ioprio = bio_prio(bio); | |
1814 | else if (ioc) | |
1815 | req->ioprio = ioc->ioprio; | |
1816 | else | |
1817 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
1818 | req->write_hint = bio->bi_write_hint; | |
1819 | blk_rq_bio_prep(req->q, req, bio); | |
1820 | } | |
1821 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); | |
1822 | ||
1823 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) | |
1824 | { | |
1825 | struct blk_plug *plug; | |
1826 | int where = ELEVATOR_INSERT_SORT; | |
1827 | struct request *req, *free; | |
1828 | unsigned int request_count = 0; | |
1829 | unsigned int wb_acct; | |
1830 | ||
1831 | /* | |
1832 | * low level driver can indicate that it wants pages above a | |
1833 | * certain limit bounced to low memory (ie for highmem, or even | |
1834 | * ISA dma in theory) | |
1835 | */ | |
1836 | blk_queue_bounce(q, &bio); | |
1837 | ||
1838 | blk_queue_split(q, &bio); | |
1839 | ||
1840 | if (!bio_integrity_prep(bio)) | |
1841 | return BLK_QC_T_NONE; | |
1842 | ||
1843 | if (op_is_flush(bio->bi_opf)) { | |
1844 | spin_lock_irq(q->queue_lock); | |
1845 | where = ELEVATOR_INSERT_FLUSH; | |
1846 | goto get_rq; | |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * Check if we can merge with the plugged list before grabbing | |
1851 | * any locks. | |
1852 | */ | |
1853 | if (!blk_queue_nomerges(q)) { | |
1854 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
1855 | return BLK_QC_T_NONE; | |
1856 | } else | |
1857 | request_count = blk_plug_queued_count(q); | |
1858 | ||
1859 | spin_lock_irq(q->queue_lock); | |
1860 | ||
1861 | switch (elv_merge(q, &req, bio)) { | |
1862 | case ELEVATOR_BACK_MERGE: | |
1863 | if (!bio_attempt_back_merge(q, req, bio)) | |
1864 | break; | |
1865 | elv_bio_merged(q, req, bio); | |
1866 | free = attempt_back_merge(q, req); | |
1867 | if (free) | |
1868 | __blk_put_request(q, free); | |
1869 | else | |
1870 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1871 | goto out_unlock; | |
1872 | case ELEVATOR_FRONT_MERGE: | |
1873 | if (!bio_attempt_front_merge(q, req, bio)) | |
1874 | break; | |
1875 | elv_bio_merged(q, req, bio); | |
1876 | free = attempt_front_merge(q, req); | |
1877 | if (free) | |
1878 | __blk_put_request(q, free); | |
1879 | else | |
1880 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1881 | goto out_unlock; | |
1882 | default: | |
1883 | break; | |
1884 | } | |
1885 | ||
1886 | get_rq: | |
1887 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); | |
1888 | ||
1889 | /* | |
1890 | * Grab a free request. This is might sleep but can not fail. | |
1891 | * Returns with the queue unlocked. | |
1892 | */ | |
1893 | blk_queue_enter_live(q); | |
1894 | req = get_request(q, bio->bi_opf, bio, 0); | |
1895 | if (IS_ERR(req)) { | |
1896 | blk_queue_exit(q); | |
1897 | __wbt_done(q->rq_wb, wb_acct); | |
1898 | if (PTR_ERR(req) == -ENOMEM) | |
1899 | bio->bi_status = BLK_STS_RESOURCE; | |
1900 | else | |
1901 | bio->bi_status = BLK_STS_IOERR; | |
1902 | bio_endio(bio); | |
1903 | goto out_unlock; | |
1904 | } | |
1905 | ||
1906 | wbt_track(&req->issue_stat, wb_acct); | |
1907 | ||
1908 | /* | |
1909 | * After dropping the lock and possibly sleeping here, our request | |
1910 | * may now be mergeable after it had proven unmergeable (above). | |
1911 | * We don't worry about that case for efficiency. It won't happen | |
1912 | * often, and the elevators are able to handle it. | |
1913 | */ | |
1914 | blk_init_request_from_bio(req, bio); | |
1915 | ||
1916 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) | |
1917 | req->cpu = raw_smp_processor_id(); | |
1918 | ||
1919 | plug = current->plug; | |
1920 | if (plug) { | |
1921 | /* | |
1922 | * If this is the first request added after a plug, fire | |
1923 | * of a plug trace. | |
1924 | * | |
1925 | * @request_count may become stale because of schedule | |
1926 | * out, so check plug list again. | |
1927 | */ | |
1928 | if (!request_count || list_empty(&plug->list)) | |
1929 | trace_block_plug(q); | |
1930 | else { | |
1931 | struct request *last = list_entry_rq(plug->list.prev); | |
1932 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1933 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
1934 | blk_flush_plug_list(plug, false); | |
1935 | trace_block_plug(q); | |
1936 | } | |
1937 | } | |
1938 | list_add_tail(&req->queuelist, &plug->list); | |
1939 | blk_account_io_start(req, true); | |
1940 | } else { | |
1941 | spin_lock_irq(q->queue_lock); | |
1942 | add_acct_request(q, req, where); | |
1943 | __blk_run_queue(q); | |
1944 | out_unlock: | |
1945 | spin_unlock_irq(q->queue_lock); | |
1946 | } | |
1947 | ||
1948 | return BLK_QC_T_NONE; | |
1949 | } | |
1950 | ||
1951 | static void handle_bad_sector(struct bio *bio) | |
1952 | { | |
1953 | char b[BDEVNAME_SIZE]; | |
1954 | ||
1955 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1956 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", | |
1957 | bio_devname(bio, b), bio->bi_opf, | |
1958 | (unsigned long long)bio_end_sector(bio), | |
1959 | (long long)get_capacity(bio->bi_disk)); | |
1960 | } | |
1961 | ||
1962 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1963 | ||
1964 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1965 | ||
1966 | static int __init setup_fail_make_request(char *str) | |
1967 | { | |
1968 | return setup_fault_attr(&fail_make_request, str); | |
1969 | } | |
1970 | __setup("fail_make_request=", setup_fail_make_request); | |
1971 | ||
1972 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) | |
1973 | { | |
1974 | return part->make_it_fail && should_fail(&fail_make_request, bytes); | |
1975 | } | |
1976 | ||
1977 | static int __init fail_make_request_debugfs(void) | |
1978 | { | |
1979 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", | |
1980 | NULL, &fail_make_request); | |
1981 | ||
1982 | return PTR_ERR_OR_ZERO(dir); | |
1983 | } | |
1984 | ||
1985 | late_initcall(fail_make_request_debugfs); | |
1986 | ||
1987 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1988 | ||
1989 | static inline bool should_fail_request(struct hd_struct *part, | |
1990 | unsigned int bytes) | |
1991 | { | |
1992 | return false; | |
1993 | } | |
1994 | ||
1995 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1996 | ||
1997 | /* | |
1998 | * Remap block n of partition p to block n+start(p) of the disk. | |
1999 | */ | |
2000 | static inline int blk_partition_remap(struct bio *bio) | |
2001 | { | |
2002 | struct hd_struct *p; | |
2003 | int ret = 0; | |
2004 | ||
2005 | /* | |
2006 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
2007 | * Include a test for the reset op code and perform the remap if needed. | |
2008 | */ | |
2009 | if (!bio->bi_partno || | |
2010 | (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET)) | |
2011 | return 0; | |
2012 | ||
2013 | rcu_read_lock(); | |
2014 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
2015 | if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) { | |
2016 | bio->bi_iter.bi_sector += p->start_sect; | |
2017 | bio->bi_partno = 0; | |
2018 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), | |
2019 | bio->bi_iter.bi_sector - p->start_sect); | |
2020 | } else { | |
2021 | printk("%s: fail for partition %d\n", __func__, bio->bi_partno); | |
2022 | ret = -EIO; | |
2023 | } | |
2024 | rcu_read_unlock(); | |
2025 | ||
2026 | return ret; | |
2027 | } | |
2028 | ||
2029 | /* | |
2030 | * Check whether this bio extends beyond the end of the device. | |
2031 | */ | |
2032 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
2033 | { | |
2034 | sector_t maxsector; | |
2035 | ||
2036 | if (!nr_sectors) | |
2037 | return 0; | |
2038 | ||
2039 | /* Test device or partition size, when known. */ | |
2040 | maxsector = get_capacity(bio->bi_disk); | |
2041 | if (maxsector) { | |
2042 | sector_t sector = bio->bi_iter.bi_sector; | |
2043 | ||
2044 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2045 | /* | |
2046 | * This may well happen - the kernel calls bread() | |
2047 | * without checking the size of the device, e.g., when | |
2048 | * mounting a device. | |
2049 | */ | |
2050 | handle_bad_sector(bio); | |
2051 | return 1; | |
2052 | } | |
2053 | } | |
2054 | ||
2055 | return 0; | |
2056 | } | |
2057 | ||
2058 | static noinline_for_stack bool | |
2059 | generic_make_request_checks(struct bio *bio) | |
2060 | { | |
2061 | struct request_queue *q; | |
2062 | int nr_sectors = bio_sectors(bio); | |
2063 | blk_status_t status = BLK_STS_IOERR; | |
2064 | char b[BDEVNAME_SIZE]; | |
2065 | ||
2066 | might_sleep(); | |
2067 | ||
2068 | if (bio_check_eod(bio, nr_sectors)) | |
2069 | goto end_io; | |
2070 | ||
2071 | q = bio->bi_disk->queue; | |
2072 | if (unlikely(!q)) { | |
2073 | printk(KERN_ERR | |
2074 | "generic_make_request: Trying to access " | |
2075 | "nonexistent block-device %s (%Lu)\n", | |
2076 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); | |
2077 | goto end_io; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2082 | * if queue is not a request based queue. | |
2083 | */ | |
2084 | ||
2085 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) | |
2086 | goto not_supported; | |
2087 | ||
2088 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
2089 | goto end_io; | |
2090 | ||
2091 | if (blk_partition_remap(bio)) | |
2092 | goto end_io; | |
2093 | ||
2094 | if (bio_check_eod(bio, nr_sectors)) | |
2095 | goto end_io; | |
2096 | ||
2097 | /* | |
2098 | * Filter flush bio's early so that make_request based | |
2099 | * drivers without flush support don't have to worry | |
2100 | * about them. | |
2101 | */ | |
2102 | if (op_is_flush(bio->bi_opf) && | |
2103 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { | |
2104 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); | |
2105 | if (!nr_sectors) { | |
2106 | status = BLK_STS_OK; | |
2107 | goto end_io; | |
2108 | } | |
2109 | } | |
2110 | ||
2111 | switch (bio_op(bio)) { | |
2112 | case REQ_OP_DISCARD: | |
2113 | if (!blk_queue_discard(q)) | |
2114 | goto not_supported; | |
2115 | break; | |
2116 | case REQ_OP_SECURE_ERASE: | |
2117 | if (!blk_queue_secure_erase(q)) | |
2118 | goto not_supported; | |
2119 | break; | |
2120 | case REQ_OP_WRITE_SAME: | |
2121 | if (!q->limits.max_write_same_sectors) | |
2122 | goto not_supported; | |
2123 | break; | |
2124 | case REQ_OP_ZONE_REPORT: | |
2125 | case REQ_OP_ZONE_RESET: | |
2126 | if (!blk_queue_is_zoned(q)) | |
2127 | goto not_supported; | |
2128 | break; | |
2129 | case REQ_OP_WRITE_ZEROES: | |
2130 | if (!q->limits.max_write_zeroes_sectors) | |
2131 | goto not_supported; | |
2132 | break; | |
2133 | default: | |
2134 | break; | |
2135 | } | |
2136 | ||
2137 | /* | |
2138 | * Various block parts want %current->io_context and lazy ioc | |
2139 | * allocation ends up trading a lot of pain for a small amount of | |
2140 | * memory. Just allocate it upfront. This may fail and block | |
2141 | * layer knows how to live with it. | |
2142 | */ | |
2143 | create_io_context(GFP_ATOMIC, q->node); | |
2144 | ||
2145 | if (!blkcg_bio_issue_check(q, bio)) | |
2146 | return false; | |
2147 | ||
2148 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { | |
2149 | trace_block_bio_queue(q, bio); | |
2150 | /* Now that enqueuing has been traced, we need to trace | |
2151 | * completion as well. | |
2152 | */ | |
2153 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2154 | } | |
2155 | return true; | |
2156 | ||
2157 | not_supported: | |
2158 | status = BLK_STS_NOTSUPP; | |
2159 | end_io: | |
2160 | bio->bi_status = status; | |
2161 | bio_endio(bio); | |
2162 | return false; | |
2163 | } | |
2164 | ||
2165 | /** | |
2166 | * generic_make_request - hand a buffer to its device driver for I/O | |
2167 | * @bio: The bio describing the location in memory and on the device. | |
2168 | * | |
2169 | * generic_make_request() is used to make I/O requests of block | |
2170 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2171 | * to be done. | |
2172 | * | |
2173 | * generic_make_request() does not return any status. The | |
2174 | * success/failure status of the request, along with notification of | |
2175 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2176 | * function described (one day) else where. | |
2177 | * | |
2178 | * The caller of generic_make_request must make sure that bi_io_vec | |
2179 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2180 | * set to describe the device address, and the | |
2181 | * bi_end_io and optionally bi_private are set to describe how | |
2182 | * completion notification should be signaled. | |
2183 | * | |
2184 | * generic_make_request and the drivers it calls may use bi_next if this | |
2185 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2186 | * a lower device by calling into generic_make_request recursively, which | |
2187 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
2188 | */ | |
2189 | blk_qc_t generic_make_request(struct bio *bio) | |
2190 | { | |
2191 | /* | |
2192 | * bio_list_on_stack[0] contains bios submitted by the current | |
2193 | * make_request_fn. | |
2194 | * bio_list_on_stack[1] contains bios that were submitted before | |
2195 | * the current make_request_fn, but that haven't been processed | |
2196 | * yet. | |
2197 | */ | |
2198 | struct bio_list bio_list_on_stack[2]; | |
2199 | blk_qc_t ret = BLK_QC_T_NONE; | |
2200 | ||
2201 | if (!generic_make_request_checks(bio)) | |
2202 | goto out; | |
2203 | ||
2204 | /* | |
2205 | * We only want one ->make_request_fn to be active at a time, else | |
2206 | * stack usage with stacked devices could be a problem. So use | |
2207 | * current->bio_list to keep a list of requests submited by a | |
2208 | * make_request_fn function. current->bio_list is also used as a | |
2209 | * flag to say if generic_make_request is currently active in this | |
2210 | * task or not. If it is NULL, then no make_request is active. If | |
2211 | * it is non-NULL, then a make_request is active, and new requests | |
2212 | * should be added at the tail | |
2213 | */ | |
2214 | if (current->bio_list) { | |
2215 | bio_list_add(¤t->bio_list[0], bio); | |
2216 | goto out; | |
2217 | } | |
2218 | ||
2219 | /* following loop may be a bit non-obvious, and so deserves some | |
2220 | * explanation. | |
2221 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2222 | * ensure that) so we have a list with a single bio. | |
2223 | * We pretend that we have just taken it off a longer list, so | |
2224 | * we assign bio_list to a pointer to the bio_list_on_stack, | |
2225 | * thus initialising the bio_list of new bios to be | |
2226 | * added. ->make_request() may indeed add some more bios | |
2227 | * through a recursive call to generic_make_request. If it | |
2228 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2229 | * from the top. In this case we really did just take the bio | |
2230 | * of the top of the list (no pretending) and so remove it from | |
2231 | * bio_list, and call into ->make_request() again. | |
2232 | */ | |
2233 | BUG_ON(bio->bi_next); | |
2234 | bio_list_init(&bio_list_on_stack[0]); | |
2235 | current->bio_list = bio_list_on_stack; | |
2236 | do { | |
2237 | struct request_queue *q = bio->bi_disk->queue; | |
2238 | ||
2239 | if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) { | |
2240 | struct bio_list lower, same; | |
2241 | ||
2242 | /* Create a fresh bio_list for all subordinate requests */ | |
2243 | bio_list_on_stack[1] = bio_list_on_stack[0]; | |
2244 | bio_list_init(&bio_list_on_stack[0]); | |
2245 | ret = q->make_request_fn(q, bio); | |
2246 | ||
2247 | blk_queue_exit(q); | |
2248 | ||
2249 | /* sort new bios into those for a lower level | |
2250 | * and those for the same level | |
2251 | */ | |
2252 | bio_list_init(&lower); | |
2253 | bio_list_init(&same); | |
2254 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) | |
2255 | if (q == bio->bi_disk->queue) | |
2256 | bio_list_add(&same, bio); | |
2257 | else | |
2258 | bio_list_add(&lower, bio); | |
2259 | /* now assemble so we handle the lowest level first */ | |
2260 | bio_list_merge(&bio_list_on_stack[0], &lower); | |
2261 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2262 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
2263 | } else { | |
2264 | if (unlikely(!blk_queue_dying(q) && | |
2265 | (bio->bi_opf & REQ_NOWAIT))) | |
2266 | bio_wouldblock_error(bio); | |
2267 | else | |
2268 | bio_io_error(bio); | |
2269 | } | |
2270 | bio = bio_list_pop(&bio_list_on_stack[0]); | |
2271 | } while (bio); | |
2272 | current->bio_list = NULL; /* deactivate */ | |
2273 | ||
2274 | out: | |
2275 | return ret; | |
2276 | } | |
2277 | EXPORT_SYMBOL(generic_make_request); | |
2278 | ||
2279 | /** | |
2280 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
2281 | * @bio: The bio describing the location in memory and on the device. | |
2282 | * | |
2283 | * This function behaves like generic_make_request(), but does not protect | |
2284 | * against recursion. Must only be used if the called driver is known | |
2285 | * to not call generic_make_request (or direct_make_request) again from | |
2286 | * its make_request function. (Calling direct_make_request again from | |
2287 | * a workqueue is perfectly fine as that doesn't recurse). | |
2288 | */ | |
2289 | blk_qc_t direct_make_request(struct bio *bio) | |
2290 | { | |
2291 | struct request_queue *q = bio->bi_disk->queue; | |
2292 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
2293 | blk_qc_t ret; | |
2294 | ||
2295 | if (!generic_make_request_checks(bio)) | |
2296 | return BLK_QC_T_NONE; | |
2297 | ||
2298 | if (unlikely(blk_queue_enter(q, nowait))) { | |
2299 | if (nowait && !blk_queue_dying(q)) | |
2300 | bio->bi_status = BLK_STS_AGAIN; | |
2301 | else | |
2302 | bio->bi_status = BLK_STS_IOERR; | |
2303 | bio_endio(bio); | |
2304 | return BLK_QC_T_NONE; | |
2305 | } | |
2306 | ||
2307 | ret = q->make_request_fn(q, bio); | |
2308 | blk_queue_exit(q); | |
2309 | return ret; | |
2310 | } | |
2311 | EXPORT_SYMBOL_GPL(direct_make_request); | |
2312 | ||
2313 | /** | |
2314 | * submit_bio - submit a bio to the block device layer for I/O | |
2315 | * @bio: The &struct bio which describes the I/O | |
2316 | * | |
2317 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2318 | * uses that function to do most of the work. Both are fairly rough | |
2319 | * interfaces; @bio must be presetup and ready for I/O. | |
2320 | * | |
2321 | */ | |
2322 | blk_qc_t submit_bio(struct bio *bio) | |
2323 | { | |
2324 | /* | |
2325 | * If it's a regular read/write or a barrier with data attached, | |
2326 | * go through the normal accounting stuff before submission. | |
2327 | */ | |
2328 | if (bio_has_data(bio)) { | |
2329 | unsigned int count; | |
2330 | ||
2331 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) | |
2332 | count = queue_logical_block_size(bio->bi_disk->queue); | |
2333 | else | |
2334 | count = bio_sectors(bio); | |
2335 | ||
2336 | if (op_is_write(bio_op(bio))) { | |
2337 | count_vm_events(PGPGOUT, count); | |
2338 | } else { | |
2339 | task_io_account_read(bio->bi_iter.bi_size); | |
2340 | count_vm_events(PGPGIN, count); | |
2341 | } | |
2342 | ||
2343 | if (unlikely(block_dump)) { | |
2344 | char b[BDEVNAME_SIZE]; | |
2345 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", | |
2346 | current->comm, task_pid_nr(current), | |
2347 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", | |
2348 | (unsigned long long)bio->bi_iter.bi_sector, | |
2349 | bio_devname(bio, b), count); | |
2350 | } | |
2351 | } | |
2352 | ||
2353 | return generic_make_request(bio); | |
2354 | } | |
2355 | EXPORT_SYMBOL(submit_bio); | |
2356 | ||
2357 | bool blk_poll(struct request_queue *q, blk_qc_t cookie) | |
2358 | { | |
2359 | if (!q->poll_fn || !blk_qc_t_valid(cookie)) | |
2360 | return false; | |
2361 | ||
2362 | if (current->plug) | |
2363 | blk_flush_plug_list(current->plug, false); | |
2364 | return q->poll_fn(q, cookie); | |
2365 | } | |
2366 | EXPORT_SYMBOL_GPL(blk_poll); | |
2367 | ||
2368 | /** | |
2369 | * blk_cloned_rq_check_limits - Helper function to check a cloned request | |
2370 | * for new the queue limits | |
2371 | * @q: the queue | |
2372 | * @rq: the request being checked | |
2373 | * | |
2374 | * Description: | |
2375 | * @rq may have been made based on weaker limitations of upper-level queues | |
2376 | * in request stacking drivers, and it may violate the limitation of @q. | |
2377 | * Since the block layer and the underlying device driver trust @rq | |
2378 | * after it is inserted to @q, it should be checked against @q before | |
2379 | * the insertion using this generic function. | |
2380 | * | |
2381 | * Request stacking drivers like request-based dm may change the queue | |
2382 | * limits when retrying requests on other queues. Those requests need | |
2383 | * to be checked against the new queue limits again during dispatch. | |
2384 | */ | |
2385 | static int blk_cloned_rq_check_limits(struct request_queue *q, | |
2386 | struct request *rq) | |
2387 | { | |
2388 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { | |
2389 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
2390 | return -EIO; | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * queue's settings related to segment counting like q->bounce_pfn | |
2395 | * may differ from that of other stacking queues. | |
2396 | * Recalculate it to check the request correctly on this queue's | |
2397 | * limitation. | |
2398 | */ | |
2399 | blk_recalc_rq_segments(rq); | |
2400 | if (rq->nr_phys_segments > queue_max_segments(q)) { | |
2401 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
2402 | return -EIO; | |
2403 | } | |
2404 | ||
2405 | return 0; | |
2406 | } | |
2407 | ||
2408 | /** | |
2409 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2410 | * @q: the queue to submit the request | |
2411 | * @rq: the request being queued | |
2412 | */ | |
2413 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
2414 | { | |
2415 | unsigned long flags; | |
2416 | int where = ELEVATOR_INSERT_BACK; | |
2417 | ||
2418 | if (blk_cloned_rq_check_limits(q, rq)) | |
2419 | return BLK_STS_IOERR; | |
2420 | ||
2421 | if (rq->rq_disk && | |
2422 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2423 | return BLK_STS_IOERR; | |
2424 | ||
2425 | if (q->mq_ops) { | |
2426 | if (blk_queue_io_stat(q)) | |
2427 | blk_account_io_start(rq, true); | |
2428 | /* | |
2429 | * Since we have a scheduler attached on the top device, | |
2430 | * bypass a potential scheduler on the bottom device for | |
2431 | * insert. | |
2432 | */ | |
2433 | blk_mq_request_bypass_insert(rq, true); | |
2434 | return BLK_STS_OK; | |
2435 | } | |
2436 | ||
2437 | spin_lock_irqsave(q->queue_lock, flags); | |
2438 | if (unlikely(blk_queue_dying(q))) { | |
2439 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2440 | return BLK_STS_IOERR; | |
2441 | } | |
2442 | ||
2443 | /* | |
2444 | * Submitting request must be dequeued before calling this function | |
2445 | * because it will be linked to another request_queue | |
2446 | */ | |
2447 | BUG_ON(blk_queued_rq(rq)); | |
2448 | ||
2449 | if (op_is_flush(rq->cmd_flags)) | |
2450 | where = ELEVATOR_INSERT_FLUSH; | |
2451 | ||
2452 | add_acct_request(q, rq, where); | |
2453 | if (where == ELEVATOR_INSERT_FLUSH) | |
2454 | __blk_run_queue(q); | |
2455 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2456 | ||
2457 | return BLK_STS_OK; | |
2458 | } | |
2459 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2460 | ||
2461 | /** | |
2462 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2463 | * @rq: request to examine | |
2464 | * | |
2465 | * Description: | |
2466 | * A request could be merge of IOs which require different failure | |
2467 | * handling. This function determines the number of bytes which | |
2468 | * can be failed from the beginning of the request without | |
2469 | * crossing into area which need to be retried further. | |
2470 | * | |
2471 | * Return: | |
2472 | * The number of bytes to fail. | |
2473 | */ | |
2474 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2475 | { | |
2476 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2477 | unsigned int bytes = 0; | |
2478 | struct bio *bio; | |
2479 | ||
2480 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) | |
2481 | return blk_rq_bytes(rq); | |
2482 | ||
2483 | /* | |
2484 | * Currently the only 'mixing' which can happen is between | |
2485 | * different fastfail types. We can safely fail portions | |
2486 | * which have all the failfast bits that the first one has - | |
2487 | * the ones which are at least as eager to fail as the first | |
2488 | * one. | |
2489 | */ | |
2490 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
2491 | if ((bio->bi_opf & ff) != ff) | |
2492 | break; | |
2493 | bytes += bio->bi_iter.bi_size; | |
2494 | } | |
2495 | ||
2496 | /* this could lead to infinite loop */ | |
2497 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2498 | return bytes; | |
2499 | } | |
2500 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2501 | ||
2502 | void blk_account_io_completion(struct request *req, unsigned int bytes) | |
2503 | { | |
2504 | if (blk_do_io_stat(req)) { | |
2505 | const int rw = rq_data_dir(req); | |
2506 | struct hd_struct *part; | |
2507 | int cpu; | |
2508 | ||
2509 | cpu = part_stat_lock(); | |
2510 | part = req->part; | |
2511 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
2512 | part_stat_unlock(); | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | void blk_account_io_done(struct request *req) | |
2517 | { | |
2518 | /* | |
2519 | * Account IO completion. flush_rq isn't accounted as a | |
2520 | * normal IO on queueing nor completion. Accounting the | |
2521 | * containing request is enough. | |
2522 | */ | |
2523 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { | |
2524 | unsigned long duration = jiffies - req->start_time; | |
2525 | const int rw = rq_data_dir(req); | |
2526 | struct hd_struct *part; | |
2527 | int cpu; | |
2528 | ||
2529 | cpu = part_stat_lock(); | |
2530 | part = req->part; | |
2531 | ||
2532 | part_stat_inc(cpu, part, ios[rw]); | |
2533 | part_stat_add(cpu, part, ticks[rw], duration); | |
2534 | part_round_stats(req->q, cpu, part); | |
2535 | part_dec_in_flight(req->q, part, rw); | |
2536 | ||
2537 | hd_struct_put(part); | |
2538 | part_stat_unlock(); | |
2539 | } | |
2540 | } | |
2541 | ||
2542 | #ifdef CONFIG_PM | |
2543 | /* | |
2544 | * Don't process normal requests when queue is suspended | |
2545 | * or in the process of suspending/resuming | |
2546 | */ | |
2547 | static bool blk_pm_allow_request(struct request *rq) | |
2548 | { | |
2549 | switch (rq->q->rpm_status) { | |
2550 | case RPM_RESUMING: | |
2551 | case RPM_SUSPENDING: | |
2552 | return rq->rq_flags & RQF_PM; | |
2553 | case RPM_SUSPENDED: | |
2554 | return false; | |
2555 | } | |
2556 | ||
2557 | return true; | |
2558 | } | |
2559 | #else | |
2560 | static bool blk_pm_allow_request(struct request *rq) | |
2561 | { | |
2562 | return true; | |
2563 | } | |
2564 | #endif | |
2565 | ||
2566 | void blk_account_io_start(struct request *rq, bool new_io) | |
2567 | { | |
2568 | struct hd_struct *part; | |
2569 | int rw = rq_data_dir(rq); | |
2570 | int cpu; | |
2571 | ||
2572 | if (!blk_do_io_stat(rq)) | |
2573 | return; | |
2574 | ||
2575 | cpu = part_stat_lock(); | |
2576 | ||
2577 | if (!new_io) { | |
2578 | part = rq->part; | |
2579 | part_stat_inc(cpu, part, merges[rw]); | |
2580 | } else { | |
2581 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2582 | if (!hd_struct_try_get(part)) { | |
2583 | /* | |
2584 | * The partition is already being removed, | |
2585 | * the request will be accounted on the disk only | |
2586 | * | |
2587 | * We take a reference on disk->part0 although that | |
2588 | * partition will never be deleted, so we can treat | |
2589 | * it as any other partition. | |
2590 | */ | |
2591 | part = &rq->rq_disk->part0; | |
2592 | hd_struct_get(part); | |
2593 | } | |
2594 | part_round_stats(rq->q, cpu, part); | |
2595 | part_inc_in_flight(rq->q, part, rw); | |
2596 | rq->part = part; | |
2597 | } | |
2598 | ||
2599 | part_stat_unlock(); | |
2600 | } | |
2601 | ||
2602 | static struct request *elv_next_request(struct request_queue *q) | |
2603 | { | |
2604 | struct request *rq; | |
2605 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
2606 | ||
2607 | WARN_ON_ONCE(q->mq_ops); | |
2608 | ||
2609 | while (1) { | |
2610 | list_for_each_entry(rq, &q->queue_head, queuelist) { | |
2611 | if (blk_pm_allow_request(rq)) | |
2612 | return rq; | |
2613 | ||
2614 | if (rq->rq_flags & RQF_SOFTBARRIER) | |
2615 | break; | |
2616 | } | |
2617 | ||
2618 | /* | |
2619 | * Flush request is running and flush request isn't queueable | |
2620 | * in the drive, we can hold the queue till flush request is | |
2621 | * finished. Even we don't do this, driver can't dispatch next | |
2622 | * requests and will requeue them. And this can improve | |
2623 | * throughput too. For example, we have request flush1, write1, | |
2624 | * flush 2. flush1 is dispatched, then queue is hold, write1 | |
2625 | * isn't inserted to queue. After flush1 is finished, flush2 | |
2626 | * will be dispatched. Since disk cache is already clean, | |
2627 | * flush2 will be finished very soon, so looks like flush2 is | |
2628 | * folded to flush1. | |
2629 | * Since the queue is hold, a flag is set to indicate the queue | |
2630 | * should be restarted later. Please see flush_end_io() for | |
2631 | * details. | |
2632 | */ | |
2633 | if (fq->flush_pending_idx != fq->flush_running_idx && | |
2634 | !queue_flush_queueable(q)) { | |
2635 | fq->flush_queue_delayed = 1; | |
2636 | return NULL; | |
2637 | } | |
2638 | if (unlikely(blk_queue_bypass(q)) || | |
2639 | !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0)) | |
2640 | return NULL; | |
2641 | } | |
2642 | } | |
2643 | ||
2644 | /** | |
2645 | * blk_peek_request - peek at the top of a request queue | |
2646 | * @q: request queue to peek at | |
2647 | * | |
2648 | * Description: | |
2649 | * Return the request at the top of @q. The returned request | |
2650 | * should be started using blk_start_request() before LLD starts | |
2651 | * processing it. | |
2652 | * | |
2653 | * Return: | |
2654 | * Pointer to the request at the top of @q if available. Null | |
2655 | * otherwise. | |
2656 | */ | |
2657 | struct request *blk_peek_request(struct request_queue *q) | |
2658 | { | |
2659 | struct request *rq; | |
2660 | int ret; | |
2661 | ||
2662 | lockdep_assert_held(q->queue_lock); | |
2663 | WARN_ON_ONCE(q->mq_ops); | |
2664 | ||
2665 | while ((rq = elv_next_request(q)) != NULL) { | |
2666 | if (!(rq->rq_flags & RQF_STARTED)) { | |
2667 | /* | |
2668 | * This is the first time the device driver | |
2669 | * sees this request (possibly after | |
2670 | * requeueing). Notify IO scheduler. | |
2671 | */ | |
2672 | if (rq->rq_flags & RQF_SORTED) | |
2673 | elv_activate_rq(q, rq); | |
2674 | ||
2675 | /* | |
2676 | * just mark as started even if we don't start | |
2677 | * it, a request that has been delayed should | |
2678 | * not be passed by new incoming requests | |
2679 | */ | |
2680 | rq->rq_flags |= RQF_STARTED; | |
2681 | trace_block_rq_issue(q, rq); | |
2682 | } | |
2683 | ||
2684 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2685 | q->end_sector = rq_end_sector(rq); | |
2686 | q->boundary_rq = NULL; | |
2687 | } | |
2688 | ||
2689 | if (rq->rq_flags & RQF_DONTPREP) | |
2690 | break; | |
2691 | ||
2692 | if (q->dma_drain_size && blk_rq_bytes(rq)) { | |
2693 | /* | |
2694 | * make sure space for the drain appears we | |
2695 | * know we can do this because max_hw_segments | |
2696 | * has been adjusted to be one fewer than the | |
2697 | * device can handle | |
2698 | */ | |
2699 | rq->nr_phys_segments++; | |
2700 | } | |
2701 | ||
2702 | if (!q->prep_rq_fn) | |
2703 | break; | |
2704 | ||
2705 | ret = q->prep_rq_fn(q, rq); | |
2706 | if (ret == BLKPREP_OK) { | |
2707 | break; | |
2708 | } else if (ret == BLKPREP_DEFER) { | |
2709 | /* | |
2710 | * the request may have been (partially) prepped. | |
2711 | * we need to keep this request in the front to | |
2712 | * avoid resource deadlock. RQF_STARTED will | |
2713 | * prevent other fs requests from passing this one. | |
2714 | */ | |
2715 | if (q->dma_drain_size && blk_rq_bytes(rq) && | |
2716 | !(rq->rq_flags & RQF_DONTPREP)) { | |
2717 | /* | |
2718 | * remove the space for the drain we added | |
2719 | * so that we don't add it again | |
2720 | */ | |
2721 | --rq->nr_phys_segments; | |
2722 | } | |
2723 | ||
2724 | rq = NULL; | |
2725 | break; | |
2726 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { | |
2727 | rq->rq_flags |= RQF_QUIET; | |
2728 | /* | |
2729 | * Mark this request as started so we don't trigger | |
2730 | * any debug logic in the end I/O path. | |
2731 | */ | |
2732 | blk_start_request(rq); | |
2733 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? | |
2734 | BLK_STS_TARGET : BLK_STS_IOERR); | |
2735 | } else { | |
2736 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2737 | break; | |
2738 | } | |
2739 | } | |
2740 | ||
2741 | return rq; | |
2742 | } | |
2743 | EXPORT_SYMBOL(blk_peek_request); | |
2744 | ||
2745 | static void blk_dequeue_request(struct request *rq) | |
2746 | { | |
2747 | struct request_queue *q = rq->q; | |
2748 | ||
2749 | BUG_ON(list_empty(&rq->queuelist)); | |
2750 | BUG_ON(ELV_ON_HASH(rq)); | |
2751 | ||
2752 | list_del_init(&rq->queuelist); | |
2753 | ||
2754 | /* | |
2755 | * the time frame between a request being removed from the lists | |
2756 | * and to it is freed is accounted as io that is in progress at | |
2757 | * the driver side. | |
2758 | */ | |
2759 | if (blk_account_rq(rq)) { | |
2760 | q->in_flight[rq_is_sync(rq)]++; | |
2761 | set_io_start_time_ns(rq); | |
2762 | } | |
2763 | } | |
2764 | ||
2765 | /** | |
2766 | * blk_start_request - start request processing on the driver | |
2767 | * @req: request to dequeue | |
2768 | * | |
2769 | * Description: | |
2770 | * Dequeue @req and start timeout timer on it. This hands off the | |
2771 | * request to the driver. | |
2772 | */ | |
2773 | void blk_start_request(struct request *req) | |
2774 | { | |
2775 | lockdep_assert_held(req->q->queue_lock); | |
2776 | WARN_ON_ONCE(req->q->mq_ops); | |
2777 | ||
2778 | blk_dequeue_request(req); | |
2779 | ||
2780 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { | |
2781 | blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req)); | |
2782 | req->rq_flags |= RQF_STATS; | |
2783 | wbt_issue(req->q->rq_wb, &req->issue_stat); | |
2784 | } | |
2785 | ||
2786 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); | |
2787 | blk_add_timer(req); | |
2788 | } | |
2789 | EXPORT_SYMBOL(blk_start_request); | |
2790 | ||
2791 | /** | |
2792 | * blk_fetch_request - fetch a request from a request queue | |
2793 | * @q: request queue to fetch a request from | |
2794 | * | |
2795 | * Description: | |
2796 | * Return the request at the top of @q. The request is started on | |
2797 | * return and LLD can start processing it immediately. | |
2798 | * | |
2799 | * Return: | |
2800 | * Pointer to the request at the top of @q if available. Null | |
2801 | * otherwise. | |
2802 | */ | |
2803 | struct request *blk_fetch_request(struct request_queue *q) | |
2804 | { | |
2805 | struct request *rq; | |
2806 | ||
2807 | lockdep_assert_held(q->queue_lock); | |
2808 | WARN_ON_ONCE(q->mq_ops); | |
2809 | ||
2810 | rq = blk_peek_request(q); | |
2811 | if (rq) | |
2812 | blk_start_request(rq); | |
2813 | return rq; | |
2814 | } | |
2815 | EXPORT_SYMBOL(blk_fetch_request); | |
2816 | ||
2817 | /* | |
2818 | * Steal bios from a request and add them to a bio list. | |
2819 | * The request must not have been partially completed before. | |
2820 | */ | |
2821 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
2822 | { | |
2823 | if (rq->bio) { | |
2824 | if (list->tail) | |
2825 | list->tail->bi_next = rq->bio; | |
2826 | else | |
2827 | list->head = rq->bio; | |
2828 | list->tail = rq->biotail; | |
2829 | ||
2830 | rq->bio = NULL; | |
2831 | rq->biotail = NULL; | |
2832 | } | |
2833 | ||
2834 | rq->__data_len = 0; | |
2835 | } | |
2836 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
2837 | ||
2838 | /** | |
2839 | * blk_update_request - Special helper function for request stacking drivers | |
2840 | * @req: the request being processed | |
2841 | * @error: block status code | |
2842 | * @nr_bytes: number of bytes to complete @req | |
2843 | * | |
2844 | * Description: | |
2845 | * Ends I/O on a number of bytes attached to @req, but doesn't complete | |
2846 | * the request structure even if @req doesn't have leftover. | |
2847 | * If @req has leftover, sets it up for the next range of segments. | |
2848 | * | |
2849 | * This special helper function is only for request stacking drivers | |
2850 | * (e.g. request-based dm) so that they can handle partial completion. | |
2851 | * Actual device drivers should use blk_end_request instead. | |
2852 | * | |
2853 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2854 | * %false return from this function. | |
2855 | * | |
2856 | * Return: | |
2857 | * %false - this request doesn't have any more data | |
2858 | * %true - this request has more data | |
2859 | **/ | |
2860 | bool blk_update_request(struct request *req, blk_status_t error, | |
2861 | unsigned int nr_bytes) | |
2862 | { | |
2863 | int total_bytes; | |
2864 | ||
2865 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); | |
2866 | ||
2867 | if (!req->bio) | |
2868 | return false; | |
2869 | ||
2870 | if (unlikely(error && !blk_rq_is_passthrough(req) && | |
2871 | !(req->rq_flags & RQF_QUIET))) | |
2872 | print_req_error(req, error); | |
2873 | ||
2874 | blk_account_io_completion(req, nr_bytes); | |
2875 | ||
2876 | total_bytes = 0; | |
2877 | while (req->bio) { | |
2878 | struct bio *bio = req->bio; | |
2879 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); | |
2880 | ||
2881 | if (bio_bytes == bio->bi_iter.bi_size) | |
2882 | req->bio = bio->bi_next; | |
2883 | ||
2884 | /* Completion has already been traced */ | |
2885 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
2886 | req_bio_endio(req, bio, bio_bytes, error); | |
2887 | ||
2888 | total_bytes += bio_bytes; | |
2889 | nr_bytes -= bio_bytes; | |
2890 | ||
2891 | if (!nr_bytes) | |
2892 | break; | |
2893 | } | |
2894 | ||
2895 | /* | |
2896 | * completely done | |
2897 | */ | |
2898 | if (!req->bio) { | |
2899 | /* | |
2900 | * Reset counters so that the request stacking driver | |
2901 | * can find how many bytes remain in the request | |
2902 | * later. | |
2903 | */ | |
2904 | req->__data_len = 0; | |
2905 | return false; | |
2906 | } | |
2907 | ||
2908 | req->__data_len -= total_bytes; | |
2909 | ||
2910 | /* update sector only for requests with clear definition of sector */ | |
2911 | if (!blk_rq_is_passthrough(req)) | |
2912 | req->__sector += total_bytes >> 9; | |
2913 | ||
2914 | /* mixed attributes always follow the first bio */ | |
2915 | if (req->rq_flags & RQF_MIXED_MERGE) { | |
2916 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2917 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; | |
2918 | } | |
2919 | ||
2920 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { | |
2921 | /* | |
2922 | * If total number of sectors is less than the first segment | |
2923 | * size, something has gone terribly wrong. | |
2924 | */ | |
2925 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2926 | blk_dump_rq_flags(req, "request botched"); | |
2927 | req->__data_len = blk_rq_cur_bytes(req); | |
2928 | } | |
2929 | ||
2930 | /* recalculate the number of segments */ | |
2931 | blk_recalc_rq_segments(req); | |
2932 | } | |
2933 | ||
2934 | return true; | |
2935 | } | |
2936 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2937 | ||
2938 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, | |
2939 | unsigned int nr_bytes, | |
2940 | unsigned int bidi_bytes) | |
2941 | { | |
2942 | if (blk_update_request(rq, error, nr_bytes)) | |
2943 | return true; | |
2944 | ||
2945 | /* Bidi request must be completed as a whole */ | |
2946 | if (unlikely(blk_bidi_rq(rq)) && | |
2947 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2948 | return true; | |
2949 | ||
2950 | if (blk_queue_add_random(rq->q)) | |
2951 | add_disk_randomness(rq->rq_disk); | |
2952 | ||
2953 | return false; | |
2954 | } | |
2955 | ||
2956 | /** | |
2957 | * blk_unprep_request - unprepare a request | |
2958 | * @req: the request | |
2959 | * | |
2960 | * This function makes a request ready for complete resubmission (or | |
2961 | * completion). It happens only after all error handling is complete, | |
2962 | * so represents the appropriate moment to deallocate any resources | |
2963 | * that were allocated to the request in the prep_rq_fn. The queue | |
2964 | * lock is held when calling this. | |
2965 | */ | |
2966 | void blk_unprep_request(struct request *req) | |
2967 | { | |
2968 | struct request_queue *q = req->q; | |
2969 | ||
2970 | req->rq_flags &= ~RQF_DONTPREP; | |
2971 | if (q->unprep_rq_fn) | |
2972 | q->unprep_rq_fn(q, req); | |
2973 | } | |
2974 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2975 | ||
2976 | void blk_finish_request(struct request *req, blk_status_t error) | |
2977 | { | |
2978 | struct request_queue *q = req->q; | |
2979 | ||
2980 | lockdep_assert_held(req->q->queue_lock); | |
2981 | WARN_ON_ONCE(q->mq_ops); | |
2982 | ||
2983 | if (req->rq_flags & RQF_STATS) | |
2984 | blk_stat_add(req); | |
2985 | ||
2986 | if (req->rq_flags & RQF_QUEUED) | |
2987 | blk_queue_end_tag(q, req); | |
2988 | ||
2989 | BUG_ON(blk_queued_rq(req)); | |
2990 | ||
2991 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) | |
2992 | laptop_io_completion(req->q->backing_dev_info); | |
2993 | ||
2994 | blk_delete_timer(req); | |
2995 | ||
2996 | if (req->rq_flags & RQF_DONTPREP) | |
2997 | blk_unprep_request(req); | |
2998 | ||
2999 | blk_account_io_done(req); | |
3000 | ||
3001 | if (req->end_io) { | |
3002 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
3003 | req->end_io(req, error); | |
3004 | } else { | |
3005 | if (blk_bidi_rq(req)) | |
3006 | __blk_put_request(req->next_rq->q, req->next_rq); | |
3007 | ||
3008 | __blk_put_request(q, req); | |
3009 | } | |
3010 | } | |
3011 | EXPORT_SYMBOL(blk_finish_request); | |
3012 | ||
3013 | /** | |
3014 | * blk_end_bidi_request - Complete a bidi request | |
3015 | * @rq: the request to complete | |
3016 | * @error: block status code | |
3017 | * @nr_bytes: number of bytes to complete @rq | |
3018 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
3019 | * | |
3020 | * Description: | |
3021 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
3022 | * Drivers that supports bidi can safely call this member for any | |
3023 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
3024 | * just ignored. | |
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 | struct request_queue *q = rq->q; | |
3034 | unsigned long flags; | |
3035 | ||
3036 | WARN_ON_ONCE(q->mq_ops); | |
3037 | ||
3038 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
3039 | return true; | |
3040 | ||
3041 | spin_lock_irqsave(q->queue_lock, flags); | |
3042 | blk_finish_request(rq, error); | |
3043 | spin_unlock_irqrestore(q->queue_lock, flags); | |
3044 | ||
3045 | return false; | |
3046 | } | |
3047 | ||
3048 | /** | |
3049 | * __blk_end_bidi_request - Complete a bidi request with queue lock held | |
3050 | * @rq: the request to complete | |
3051 | * @error: block status code | |
3052 | * @nr_bytes: number of bytes to complete @rq | |
3053 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
3054 | * | |
3055 | * Description: | |
3056 | * Identical to blk_end_bidi_request() except that queue lock is | |
3057 | * assumed to be locked on entry and remains so on return. | |
3058 | * | |
3059 | * Return: | |
3060 | * %false - we are done with this request | |
3061 | * %true - still buffers pending for this request | |
3062 | **/ | |
3063 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, | |
3064 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
3065 | { | |
3066 | lockdep_assert_held(rq->q->queue_lock); | |
3067 | WARN_ON_ONCE(rq->q->mq_ops); | |
3068 | ||
3069 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
3070 | return true; | |
3071 | ||
3072 | blk_finish_request(rq, error); | |
3073 | ||
3074 | return false; | |
3075 | } | |
3076 | ||
3077 | /** | |
3078 | * blk_end_request - Helper function for drivers to complete the request. | |
3079 | * @rq: the request being processed | |
3080 | * @error: block status code | |
3081 | * @nr_bytes: number of bytes to complete | |
3082 | * | |
3083 | * Description: | |
3084 | * Ends I/O on a number of bytes attached to @rq. | |
3085 | * If @rq has leftover, sets it up for the next range of segments. | |
3086 | * | |
3087 | * Return: | |
3088 | * %false - we are done with this request | |
3089 | * %true - still buffers pending for this request | |
3090 | **/ | |
3091 | bool blk_end_request(struct request *rq, blk_status_t error, | |
3092 | unsigned int nr_bytes) | |
3093 | { | |
3094 | WARN_ON_ONCE(rq->q->mq_ops); | |
3095 | return blk_end_bidi_request(rq, error, nr_bytes, 0); | |
3096 | } | |
3097 | EXPORT_SYMBOL(blk_end_request); | |
3098 | ||
3099 | /** | |
3100 | * blk_end_request_all - Helper function for drives to finish the request. | |
3101 | * @rq: the request to finish | |
3102 | * @error: block status code | |
3103 | * | |
3104 | * Description: | |
3105 | * Completely finish @rq. | |
3106 | */ | |
3107 | void blk_end_request_all(struct request *rq, blk_status_t error) | |
3108 | { | |
3109 | bool pending; | |
3110 | unsigned int bidi_bytes = 0; | |
3111 | ||
3112 | if (unlikely(blk_bidi_rq(rq))) | |
3113 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3114 | ||
3115 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3116 | BUG_ON(pending); | |
3117 | } | |
3118 | EXPORT_SYMBOL(blk_end_request_all); | |
3119 | ||
3120 | /** | |
3121 | * __blk_end_request - Helper function for drivers to complete the request. | |
3122 | * @rq: the request being processed | |
3123 | * @error: block status code | |
3124 | * @nr_bytes: number of bytes to complete | |
3125 | * | |
3126 | * Description: | |
3127 | * Must be called with queue lock held unlike blk_end_request(). | |
3128 | * | |
3129 | * Return: | |
3130 | * %false - we are done with this request | |
3131 | * %true - still buffers pending for this request | |
3132 | **/ | |
3133 | bool __blk_end_request(struct request *rq, blk_status_t error, | |
3134 | unsigned int nr_bytes) | |
3135 | { | |
3136 | lockdep_assert_held(rq->q->queue_lock); | |
3137 | WARN_ON_ONCE(rq->q->mq_ops); | |
3138 | ||
3139 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); | |
3140 | } | |
3141 | EXPORT_SYMBOL(__blk_end_request); | |
3142 | ||
3143 | /** | |
3144 | * __blk_end_request_all - Helper function for drives to finish the request. | |
3145 | * @rq: the request to finish | |
3146 | * @error: block status code | |
3147 | * | |
3148 | * Description: | |
3149 | * Completely finish @rq. Must be called with queue lock held. | |
3150 | */ | |
3151 | void __blk_end_request_all(struct request *rq, blk_status_t error) | |
3152 | { | |
3153 | bool pending; | |
3154 | unsigned int bidi_bytes = 0; | |
3155 | ||
3156 | lockdep_assert_held(rq->q->queue_lock); | |
3157 | WARN_ON_ONCE(rq->q->mq_ops); | |
3158 | ||
3159 | if (unlikely(blk_bidi_rq(rq))) | |
3160 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3161 | ||
3162 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3163 | BUG_ON(pending); | |
3164 | } | |
3165 | EXPORT_SYMBOL(__blk_end_request_all); | |
3166 | ||
3167 | /** | |
3168 | * __blk_end_request_cur - Helper function to finish the current request chunk. | |
3169 | * @rq: the request to finish the current chunk for | |
3170 | * @error: block status code | |
3171 | * | |
3172 | * Description: | |
3173 | * Complete the current consecutively mapped chunk from @rq. Must | |
3174 | * be called with queue lock held. | |
3175 | * | |
3176 | * Return: | |
3177 | * %false - we are done with this request | |
3178 | * %true - still buffers pending for this request | |
3179 | */ | |
3180 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) | |
3181 | { | |
3182 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
3183 | } | |
3184 | EXPORT_SYMBOL(__blk_end_request_cur); | |
3185 | ||
3186 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
3187 | struct bio *bio) | |
3188 | { | |
3189 | if (bio_has_data(bio)) | |
3190 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3191 | ||
3192 | rq->__data_len = bio->bi_iter.bi_size; | |
3193 | rq->bio = rq->biotail = bio; | |
3194 | ||
3195 | if (bio->bi_disk) | |
3196 | rq->rq_disk = bio->bi_disk; | |
3197 | } | |
3198 | ||
3199 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
3200 | /** | |
3201 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3202 | * @rq: the request to be flushed | |
3203 | * | |
3204 | * Description: | |
3205 | * Flush all pages in @rq. | |
3206 | */ | |
3207 | void rq_flush_dcache_pages(struct request *rq) | |
3208 | { | |
3209 | struct req_iterator iter; | |
3210 | struct bio_vec bvec; | |
3211 | ||
3212 | rq_for_each_segment(bvec, rq, iter) | |
3213 | flush_dcache_page(bvec.bv_page); | |
3214 | } | |
3215 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3216 | #endif | |
3217 | ||
3218 | /** | |
3219 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3220 | * @q : the queue of the device being checked | |
3221 | * | |
3222 | * Description: | |
3223 | * Check if underlying low-level drivers of a device are busy. | |
3224 | * If the drivers want to export their busy state, they must set own | |
3225 | * exporting function using blk_queue_lld_busy() first. | |
3226 | * | |
3227 | * Basically, this function is used only by request stacking drivers | |
3228 | * to stop dispatching requests to underlying devices when underlying | |
3229 | * devices are busy. This behavior helps more I/O merging on the queue | |
3230 | * of the request stacking driver and prevents I/O throughput regression | |
3231 | * on burst I/O load. | |
3232 | * | |
3233 | * Return: | |
3234 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3235 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3236 | */ | |
3237 | int blk_lld_busy(struct request_queue *q) | |
3238 | { | |
3239 | if (q->lld_busy_fn) | |
3240 | return q->lld_busy_fn(q); | |
3241 | ||
3242 | return 0; | |
3243 | } | |
3244 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3245 | ||
3246 | /** | |
3247 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3248 | * @rq: the clone request to be cleaned up | |
3249 | * | |
3250 | * Description: | |
3251 | * Free all bios in @rq for a cloned request. | |
3252 | */ | |
3253 | void blk_rq_unprep_clone(struct request *rq) | |
3254 | { | |
3255 | struct bio *bio; | |
3256 | ||
3257 | while ((bio = rq->bio) != NULL) { | |
3258 | rq->bio = bio->bi_next; | |
3259 | ||
3260 | bio_put(bio); | |
3261 | } | |
3262 | } | |
3263 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3264 | ||
3265 | /* | |
3266 | * Copy attributes of the original request to the clone request. | |
3267 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3268 | */ | |
3269 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
3270 | { | |
3271 | dst->cpu = src->cpu; | |
3272 | dst->__sector = blk_rq_pos(src); | |
3273 | dst->__data_len = blk_rq_bytes(src); | |
3274 | dst->nr_phys_segments = src->nr_phys_segments; | |
3275 | dst->ioprio = src->ioprio; | |
3276 | dst->extra_len = src->extra_len; | |
3277 | } | |
3278 | ||
3279 | /** | |
3280 | * blk_rq_prep_clone - Helper function to setup clone request | |
3281 | * @rq: the request to be setup | |
3282 | * @rq_src: original request to be cloned | |
3283 | * @bs: bio_set that bios for clone are allocated from | |
3284 | * @gfp_mask: memory allocation mask for bio | |
3285 | * @bio_ctr: setup function to be called for each clone bio. | |
3286 | * Returns %0 for success, non %0 for failure. | |
3287 | * @data: private data to be passed to @bio_ctr | |
3288 | * | |
3289 | * Description: | |
3290 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3291 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3292 | * are not copied, and copying such parts is the caller's responsibility. | |
3293 | * Also, pages which the original bios are pointing to are not copied | |
3294 | * and the cloned bios just point same pages. | |
3295 | * So cloned bios must be completed before original bios, which means | |
3296 | * the caller must complete @rq before @rq_src. | |
3297 | */ | |
3298 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3299 | struct bio_set *bs, gfp_t gfp_mask, | |
3300 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3301 | void *data) | |
3302 | { | |
3303 | struct bio *bio, *bio_src; | |
3304 | ||
3305 | if (!bs) | |
3306 | bs = fs_bio_set; | |
3307 | ||
3308 | __rq_for_each_bio(bio_src, rq_src) { | |
3309 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3310 | if (!bio) | |
3311 | goto free_and_out; | |
3312 | ||
3313 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3314 | goto free_and_out; | |
3315 | ||
3316 | if (rq->bio) { | |
3317 | rq->biotail->bi_next = bio; | |
3318 | rq->biotail = bio; | |
3319 | } else | |
3320 | rq->bio = rq->biotail = bio; | |
3321 | } | |
3322 | ||
3323 | __blk_rq_prep_clone(rq, rq_src); | |
3324 | ||
3325 | return 0; | |
3326 | ||
3327 | free_and_out: | |
3328 | if (bio) | |
3329 | bio_put(bio); | |
3330 | blk_rq_unprep_clone(rq); | |
3331 | ||
3332 | return -ENOMEM; | |
3333 | } | |
3334 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3335 | ||
3336 | int kblockd_schedule_work(struct work_struct *work) | |
3337 | { | |
3338 | return queue_work(kblockd_workqueue, work); | |
3339 | } | |
3340 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3341 | ||
3342 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) | |
3343 | { | |
3344 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3345 | } | |
3346 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3347 | ||
3348 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, | |
3349 | unsigned long delay) | |
3350 | { | |
3351 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3352 | } | |
3353 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3354 | ||
3355 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, | |
3356 | unsigned long delay) | |
3357 | { | |
3358 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3359 | } | |
3360 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3361 | ||
3362 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, | |
3363 | unsigned long delay) | |
3364 | { | |
3365 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3366 | } | |
3367 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3368 | ||
3369 | /** | |
3370 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3371 | * @plug: The &struct blk_plug that needs to be initialized | |
3372 | * | |
3373 | * Description: | |
3374 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3375 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3376 | * blk_finish_plug(). This is important from a performance perspective, but | |
3377 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3378 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3379 | * page belonging to that request that is currently residing in our private | |
3380 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3381 | * this kind of deadlock. | |
3382 | */ | |
3383 | void blk_start_plug(struct blk_plug *plug) | |
3384 | { | |
3385 | struct task_struct *tsk = current; | |
3386 | ||
3387 | /* | |
3388 | * If this is a nested plug, don't actually assign it. | |
3389 | */ | |
3390 | if (tsk->plug) | |
3391 | return; | |
3392 | ||
3393 | INIT_LIST_HEAD(&plug->list); | |
3394 | INIT_LIST_HEAD(&plug->mq_list); | |
3395 | INIT_LIST_HEAD(&plug->cb_list); | |
3396 | /* | |
3397 | * Store ordering should not be needed here, since a potential | |
3398 | * preempt will imply a full memory barrier | |
3399 | */ | |
3400 | tsk->plug = plug; | |
3401 | } | |
3402 | EXPORT_SYMBOL(blk_start_plug); | |
3403 | ||
3404 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3405 | { | |
3406 | struct request *rqa = container_of(a, struct request, queuelist); | |
3407 | struct request *rqb = container_of(b, struct request, queuelist); | |
3408 | ||
3409 | return !(rqa->q < rqb->q || | |
3410 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
3411 | } | |
3412 | ||
3413 | /* | |
3414 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3415 | * until a safe kblockd context. We due this to avoid accidental big | |
3416 | * additional stack usage in driver dispatch, in places where the originally | |
3417 | * plugger did not intend it. | |
3418 | */ | |
3419 | static void queue_unplugged(struct request_queue *q, unsigned int depth, | |
3420 | bool from_schedule) | |
3421 | __releases(q->queue_lock) | |
3422 | { | |
3423 | lockdep_assert_held(q->queue_lock); | |
3424 | ||
3425 | trace_block_unplug(q, depth, !from_schedule); | |
3426 | ||
3427 | if (from_schedule) | |
3428 | blk_run_queue_async(q); | |
3429 | else | |
3430 | __blk_run_queue(q); | |
3431 | spin_unlock(q->queue_lock); | |
3432 | } | |
3433 | ||
3434 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) | |
3435 | { | |
3436 | LIST_HEAD(callbacks); | |
3437 | ||
3438 | while (!list_empty(&plug->cb_list)) { | |
3439 | list_splice_init(&plug->cb_list, &callbacks); | |
3440 | ||
3441 | while (!list_empty(&callbacks)) { | |
3442 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
3443 | struct blk_plug_cb, | |
3444 | list); | |
3445 | list_del(&cb->list); | |
3446 | cb->callback(cb, from_schedule); | |
3447 | } | |
3448 | } | |
3449 | } | |
3450 | ||
3451 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, | |
3452 | int size) | |
3453 | { | |
3454 | struct blk_plug *plug = current->plug; | |
3455 | struct blk_plug_cb *cb; | |
3456 | ||
3457 | if (!plug) | |
3458 | return NULL; | |
3459 | ||
3460 | list_for_each_entry(cb, &plug->cb_list, list) | |
3461 | if (cb->callback == unplug && cb->data == data) | |
3462 | return cb; | |
3463 | ||
3464 | /* Not currently on the callback list */ | |
3465 | BUG_ON(size < sizeof(*cb)); | |
3466 | cb = kzalloc(size, GFP_ATOMIC); | |
3467 | if (cb) { | |
3468 | cb->data = data; | |
3469 | cb->callback = unplug; | |
3470 | list_add(&cb->list, &plug->cb_list); | |
3471 | } | |
3472 | return cb; | |
3473 | } | |
3474 | EXPORT_SYMBOL(blk_check_plugged); | |
3475 | ||
3476 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) | |
3477 | { | |
3478 | struct request_queue *q; | |
3479 | unsigned long flags; | |
3480 | struct request *rq; | |
3481 | LIST_HEAD(list); | |
3482 | unsigned int depth; | |
3483 | ||
3484 | flush_plug_callbacks(plug, from_schedule); | |
3485 | ||
3486 | if (!list_empty(&plug->mq_list)) | |
3487 | blk_mq_flush_plug_list(plug, from_schedule); | |
3488 | ||
3489 | if (list_empty(&plug->list)) | |
3490 | return; | |
3491 | ||
3492 | list_splice_init(&plug->list, &list); | |
3493 | ||
3494 | list_sort(NULL, &list, plug_rq_cmp); | |
3495 | ||
3496 | q = NULL; | |
3497 | depth = 0; | |
3498 | ||
3499 | /* | |
3500 | * Save and disable interrupts here, to avoid doing it for every | |
3501 | * queue lock we have to take. | |
3502 | */ | |
3503 | local_irq_save(flags); | |
3504 | while (!list_empty(&list)) { | |
3505 | rq = list_entry_rq(list.next); | |
3506 | list_del_init(&rq->queuelist); | |
3507 | BUG_ON(!rq->q); | |
3508 | if (rq->q != q) { | |
3509 | /* | |
3510 | * This drops the queue lock | |
3511 | */ | |
3512 | if (q) | |
3513 | queue_unplugged(q, depth, from_schedule); | |
3514 | q = rq->q; | |
3515 | depth = 0; | |
3516 | spin_lock(q->queue_lock); | |
3517 | } | |
3518 | ||
3519 | /* | |
3520 | * Short-circuit if @q is dead | |
3521 | */ | |
3522 | if (unlikely(blk_queue_dying(q))) { | |
3523 | __blk_end_request_all(rq, BLK_STS_IOERR); | |
3524 | continue; | |
3525 | } | |
3526 | ||
3527 | /* | |
3528 | * rq is already accounted, so use raw insert | |
3529 | */ | |
3530 | if (op_is_flush(rq->cmd_flags)) | |
3531 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
3532 | else | |
3533 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
3534 | ||
3535 | depth++; | |
3536 | } | |
3537 | ||
3538 | /* | |
3539 | * This drops the queue lock | |
3540 | */ | |
3541 | if (q) | |
3542 | queue_unplugged(q, depth, from_schedule); | |
3543 | ||
3544 | local_irq_restore(flags); | |
3545 | } | |
3546 | ||
3547 | void blk_finish_plug(struct blk_plug *plug) | |
3548 | { | |
3549 | if (plug != current->plug) | |
3550 | return; | |
3551 | blk_flush_plug_list(plug, false); | |
3552 | ||
3553 | current->plug = NULL; | |
3554 | } | |
3555 | EXPORT_SYMBOL(blk_finish_plug); | |
3556 | ||
3557 | #ifdef CONFIG_PM | |
3558 | /** | |
3559 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3560 | * @q: the queue of the device | |
3561 | * @dev: the device the queue belongs to | |
3562 | * | |
3563 | * Description: | |
3564 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3565 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3566 | * should call this function after @dev has been initialized, and its | |
3567 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3568 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3569 | * cases, driver should call this function before any I/O has taken place. | |
3570 | * | |
3571 | * This function takes care of setting up using auto suspend for the device, | |
3572 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3573 | * until an updated value is either set by user or by driver. Drivers do | |
3574 | * not need to touch other autosuspend settings. | |
3575 | * | |
3576 | * The block layer runtime PM is request based, so only works for drivers | |
3577 | * that use request as their IO unit instead of those directly use bio's. | |
3578 | */ | |
3579 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3580 | { | |
3581 | /* not support for RQF_PM and ->rpm_status in blk-mq yet */ | |
3582 | if (q->mq_ops) | |
3583 | return; | |
3584 | ||
3585 | q->dev = dev; | |
3586 | q->rpm_status = RPM_ACTIVE; | |
3587 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3588 | pm_runtime_use_autosuspend(q->dev); | |
3589 | } | |
3590 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3591 | ||
3592 | /** | |
3593 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3594 | * @q: the queue of the device | |
3595 | * | |
3596 | * Description: | |
3597 | * This function will check if runtime suspend is allowed for the device | |
3598 | * by examining if there are any requests pending in the queue. If there | |
3599 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3600 | * the queue's status will be updated to SUSPENDING and the driver can | |
3601 | * proceed to suspend the device. | |
3602 | * | |
3603 | * For the not allowed case, we mark last busy for the device so that | |
3604 | * runtime PM core will try to autosuspend it some time later. | |
3605 | * | |
3606 | * This function should be called near the start of the device's | |
3607 | * runtime_suspend callback. | |
3608 | * | |
3609 | * Return: | |
3610 | * 0 - OK to runtime suspend the device | |
3611 | * -EBUSY - Device should not be runtime suspended | |
3612 | */ | |
3613 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3614 | { | |
3615 | int ret = 0; | |
3616 | ||
3617 | if (!q->dev) | |
3618 | return ret; | |
3619 | ||
3620 | spin_lock_irq(q->queue_lock); | |
3621 | if (q->nr_pending) { | |
3622 | ret = -EBUSY; | |
3623 | pm_runtime_mark_last_busy(q->dev); | |
3624 | } else { | |
3625 | q->rpm_status = RPM_SUSPENDING; | |
3626 | } | |
3627 | spin_unlock_irq(q->queue_lock); | |
3628 | return ret; | |
3629 | } | |
3630 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3631 | ||
3632 | /** | |
3633 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3634 | * @q: the queue of the device | |
3635 | * @err: return value of the device's runtime_suspend function | |
3636 | * | |
3637 | * Description: | |
3638 | * Update the queue's runtime status according to the return value of the | |
3639 | * device's runtime suspend function and mark last busy for the device so | |
3640 | * that PM core will try to auto suspend the device at a later time. | |
3641 | * | |
3642 | * This function should be called near the end of the device's | |
3643 | * runtime_suspend callback. | |
3644 | */ | |
3645 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3646 | { | |
3647 | if (!q->dev) | |
3648 | return; | |
3649 | ||
3650 | spin_lock_irq(q->queue_lock); | |
3651 | if (!err) { | |
3652 | q->rpm_status = RPM_SUSPENDED; | |
3653 | } else { | |
3654 | q->rpm_status = RPM_ACTIVE; | |
3655 | pm_runtime_mark_last_busy(q->dev); | |
3656 | } | |
3657 | spin_unlock_irq(q->queue_lock); | |
3658 | } | |
3659 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3660 | ||
3661 | /** | |
3662 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3663 | * @q: the queue of the device | |
3664 | * | |
3665 | * Description: | |
3666 | * Update the queue's runtime status to RESUMING in preparation for the | |
3667 | * runtime resume of the device. | |
3668 | * | |
3669 | * This function should be called near the start of the device's | |
3670 | * runtime_resume callback. | |
3671 | */ | |
3672 | void blk_pre_runtime_resume(struct request_queue *q) | |
3673 | { | |
3674 | if (!q->dev) | |
3675 | return; | |
3676 | ||
3677 | spin_lock_irq(q->queue_lock); | |
3678 | q->rpm_status = RPM_RESUMING; | |
3679 | spin_unlock_irq(q->queue_lock); | |
3680 | } | |
3681 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3682 | ||
3683 | /** | |
3684 | * blk_post_runtime_resume - Post runtime resume processing | |
3685 | * @q: the queue of the device | |
3686 | * @err: return value of the device's runtime_resume function | |
3687 | * | |
3688 | * Description: | |
3689 | * Update the queue's runtime status according to the return value of the | |
3690 | * device's runtime_resume function. If it is successfully resumed, process | |
3691 | * the requests that are queued into the device's queue when it is resuming | |
3692 | * and then mark last busy and initiate autosuspend for it. | |
3693 | * | |
3694 | * This function should be called near the end of the device's | |
3695 | * runtime_resume callback. | |
3696 | */ | |
3697 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3698 | { | |
3699 | if (!q->dev) | |
3700 | return; | |
3701 | ||
3702 | spin_lock_irq(q->queue_lock); | |
3703 | if (!err) { | |
3704 | q->rpm_status = RPM_ACTIVE; | |
3705 | __blk_run_queue(q); | |
3706 | pm_runtime_mark_last_busy(q->dev); | |
3707 | pm_request_autosuspend(q->dev); | |
3708 | } else { | |
3709 | q->rpm_status = RPM_SUSPENDED; | |
3710 | } | |
3711 | spin_unlock_irq(q->queue_lock); | |
3712 | } | |
3713 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
3714 | ||
3715 | /** | |
3716 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3717 | * @q: the queue of the device | |
3718 | * | |
3719 | * If the device is left runtime suspended during system suspend the resume | |
3720 | * hook typically resumes the device and corrects runtime status | |
3721 | * accordingly. However, that does not affect the queue runtime PM status | |
3722 | * which is still "suspended". This prevents processing requests from the | |
3723 | * queue. | |
3724 | * | |
3725 | * This function can be used in driver's resume hook to correct queue | |
3726 | * runtime PM status and re-enable peeking requests from the queue. It | |
3727 | * should be called before first request is added to the queue. | |
3728 | */ | |
3729 | void blk_set_runtime_active(struct request_queue *q) | |
3730 | { | |
3731 | spin_lock_irq(q->queue_lock); | |
3732 | q->rpm_status = RPM_ACTIVE; | |
3733 | pm_runtime_mark_last_busy(q->dev); | |
3734 | pm_request_autosuspend(q->dev); | |
3735 | spin_unlock_irq(q->queue_lock); | |
3736 | } | |
3737 | EXPORT_SYMBOL(blk_set_runtime_active); | |
3738 | #endif | |
3739 | ||
3740 | int __init blk_dev_init(void) | |
3741 | { | |
3742 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); | |
3743 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
3744 | FIELD_SIZEOF(struct request, cmd_flags)); | |
3745 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
3746 | FIELD_SIZEOF(struct bio, bi_opf)); | |
3747 | ||
3748 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ | |
3749 | kblockd_workqueue = alloc_workqueue("kblockd", | |
3750 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
3751 | if (!kblockd_workqueue) | |
3752 | panic("Failed to create kblockd\n"); | |
3753 | ||
3754 | request_cachep = kmem_cache_create("blkdev_requests", | |
3755 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
3756 | ||
3757 | blk_requestq_cachep = kmem_cache_create("request_queue", | |
3758 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
3759 | ||
3760 | #ifdef CONFIG_DEBUG_FS | |
3761 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3762 | #endif | |
3763 | ||
3764 | return 0; | |
3765 | } |