]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
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> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
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 | */ | |
1da177e4 LT |
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> | |
320ae51f | 19 | #include <linux/blk-mq.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
25 | #include <linux/completion.h> |
26 | #include <linux/slab.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/writeback.h> | |
faccbd4b | 29 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
73c10101 | 31 | #include <linux/list_sort.h> |
e3c78ca5 | 32 | #include <linux/delay.h> |
aaf7c680 | 33 | #include <linux/ratelimit.h> |
6c954667 | 34 | #include <linux/pm_runtime.h> |
eea8f41c | 35 | #include <linux/blk-cgroup.h> |
18fbda91 | 36 | #include <linux/debugfs.h> |
55782138 LZ |
37 | |
38 | #define CREATE_TRACE_POINTS | |
39 | #include <trace/events/block.h> | |
1da177e4 | 40 | |
8324aa91 | 41 | #include "blk.h" |
43a5e4e2 | 42 | #include "blk-mq.h" |
bd166ef1 | 43 | #include "blk-mq-sched.h" |
87760e5e | 44 | #include "blk-wbt.h" |
8324aa91 | 45 | |
18fbda91 OS |
46 | #ifdef CONFIG_DEBUG_FS |
47 | struct dentry *blk_debugfs_root; | |
48 | #endif | |
49 | ||
d07335e5 | 50 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 51 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 52 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 55 | |
a73f730d TH |
56 | DEFINE_IDA(blk_queue_ida); |
57 | ||
1da177e4 LT |
58 | /* |
59 | * For the allocated request tables | |
60 | */ | |
d674d414 | 61 | struct kmem_cache *request_cachep; |
1da177e4 LT |
62 | |
63 | /* | |
64 | * For queue allocation | |
65 | */ | |
6728cb0e | 66 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 67 | |
1da177e4 LT |
68 | /* |
69 | * Controlling structure to kblockd | |
70 | */ | |
ff856bad | 71 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 72 | |
d40f75a0 TH |
73 | static void blk_clear_congested(struct request_list *rl, int sync) |
74 | { | |
d40f75a0 TH |
75 | #ifdef CONFIG_CGROUP_WRITEBACK |
76 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
77 | #else | |
482cf79c TH |
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) | |
dc3b17cc | 83 | clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
84 | #endif |
85 | } | |
86 | ||
87 | static void blk_set_congested(struct request_list *rl, int sync) | |
88 | { | |
d40f75a0 TH |
89 | #ifdef CONFIG_CGROUP_WRITEBACK |
90 | set_wb_congested(rl->blkg->wb_congested, sync); | |
91 | #else | |
482cf79c TH |
92 | /* see blk_clear_congested() */ |
93 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 94 | set_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
95 | #endif |
96 | } | |
97 | ||
8324aa91 | 98 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
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 | ||
2a4aa30c | 113 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 114 | { |
1afb20f3 FT |
115 | memset(rq, 0, sizeof(*rq)); |
116 | ||
1da177e4 | 117 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 118 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 119 | rq->cpu = -1; |
63a71386 | 120 | rq->q = q; |
a2dec7b3 | 121 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
122 | INIT_HLIST_NODE(&rq->hash); |
123 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 124 | rq->tag = -1; |
bd166ef1 | 125 | rq->internal_tag = -1; |
b243ddcb | 126 | rq->start_time = jiffies; |
9195291e | 127 | set_start_time_ns(rq); |
09e099d4 | 128 | rq->part = NULL; |
1da177e4 | 129 | } |
2a4aa30c | 130 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 131 | |
5bb23a68 N |
132 | static void req_bio_endio(struct request *rq, struct bio *bio, |
133 | unsigned int nbytes, int error) | |
1da177e4 | 134 | { |
78d8e58a | 135 | if (error) |
4246a0b6 | 136 | bio->bi_error = error; |
797e7dbb | 137 | |
e8064021 | 138 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 139 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 140 | |
f79ea416 | 141 | bio_advance(bio, nbytes); |
7ba1ba12 | 142 | |
143a87f4 | 143 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 144 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 145 | bio_endio(bio); |
1da177e4 | 146 | } |
1da177e4 | 147 | |
1da177e4 LT |
148 | void blk_dump_rq_flags(struct request *rq, char *msg) |
149 | { | |
aebf526b CH |
150 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
151 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 152 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 153 | |
83096ebf TH |
154 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
155 | (unsigned long long)blk_rq_pos(rq), | |
156 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
157 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
158 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 159 | } |
1da177e4 LT |
160 | EXPORT_SYMBOL(blk_dump_rq_flags); |
161 | ||
3cca6dc1 | 162 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 163 | { |
3cca6dc1 | 164 | struct request_queue *q; |
1da177e4 | 165 | |
3cca6dc1 JA |
166 | q = container_of(work, struct request_queue, delay_work.work); |
167 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 168 | __blk_run_queue(q); |
3cca6dc1 | 169 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 170 | } |
1da177e4 LT |
171 | |
172 | /** | |
3cca6dc1 JA |
173 | * blk_delay_queue - restart queueing after defined interval |
174 | * @q: The &struct request_queue in question | |
175 | * @msecs: Delay in msecs | |
1da177e4 LT |
176 | * |
177 | * Description: | |
3cca6dc1 JA |
178 | * Sometimes queueing needs to be postponed for a little while, to allow |
179 | * resources to come back. This function will make sure that queueing is | |
70460571 | 180 | * restarted around the specified time. Queue lock must be held. |
3cca6dc1 JA |
181 | */ |
182 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 183 | { |
70460571 BVA |
184 | if (likely(!blk_queue_dead(q))) |
185 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
186 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 187 | } |
3cca6dc1 | 188 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 189 | |
21491412 JA |
190 | /** |
191 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
192 | * @q: The &struct request_queue in question | |
193 | * | |
194 | * Description: | |
195 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
196 | * ensure that the request_fn for the queue is run from an async | |
197 | * context. | |
198 | **/ | |
199 | void blk_start_queue_async(struct request_queue *q) | |
200 | { | |
201 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
202 | blk_run_queue_async(q); | |
203 | } | |
204 | EXPORT_SYMBOL(blk_start_queue_async); | |
205 | ||
1da177e4 LT |
206 | /** |
207 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 208 | * @q: The &struct request_queue in question |
1da177e4 LT |
209 | * |
210 | * Description: | |
211 | * blk_start_queue() will clear the stop flag on the queue, and call | |
212 | * the request_fn for the queue if it was in a stopped state when | |
213 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
214 | **/ | |
165125e1 | 215 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 216 | { |
a038e253 PBG |
217 | WARN_ON(!irqs_disabled()); |
218 | ||
75ad23bc | 219 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 220 | __blk_run_queue(q); |
1da177e4 | 221 | } |
1da177e4 LT |
222 | EXPORT_SYMBOL(blk_start_queue); |
223 | ||
224 | /** | |
225 | * blk_stop_queue - stop a queue | |
165125e1 | 226 | * @q: The &struct request_queue in question |
1da177e4 LT |
227 | * |
228 | * Description: | |
229 | * The Linux block layer assumes that a block driver will consume all | |
230 | * entries on the request queue when the request_fn strategy is called. | |
231 | * Often this will not happen, because of hardware limitations (queue | |
232 | * depth settings). If a device driver gets a 'queue full' response, | |
233 | * or if it simply chooses not to queue more I/O at one point, it can | |
234 | * call this function to prevent the request_fn from being called until | |
235 | * the driver has signalled it's ready to go again. This happens by calling | |
236 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
237 | **/ | |
165125e1 | 238 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 239 | { |
136b5721 | 240 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 241 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
242 | } |
243 | EXPORT_SYMBOL(blk_stop_queue); | |
244 | ||
245 | /** | |
246 | * blk_sync_queue - cancel any pending callbacks on a queue | |
247 | * @q: the queue | |
248 | * | |
249 | * Description: | |
250 | * The block layer may perform asynchronous callback activity | |
251 | * on a queue, such as calling the unplug function after a timeout. | |
252 | * A block device may call blk_sync_queue to ensure that any | |
253 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 254 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
255 | * that its ->make_request_fn will not re-add plugging prior to calling |
256 | * this function. | |
257 | * | |
da527770 | 258 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 259 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 260 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 261 | * |
1da177e4 LT |
262 | */ |
263 | void blk_sync_queue(struct request_queue *q) | |
264 | { | |
70ed28b9 | 265 | del_timer_sync(&q->timeout); |
f04c1fe7 ML |
266 | |
267 | if (q->mq_ops) { | |
268 | struct blk_mq_hw_ctx *hctx; | |
269 | int i; | |
270 | ||
70f4db63 | 271 | queue_for_each_hw_ctx(q, hctx, i) { |
27489a3c | 272 | cancel_work_sync(&hctx->run_work); |
70f4db63 CH |
273 | cancel_delayed_work_sync(&hctx->delay_work); |
274 | } | |
f04c1fe7 ML |
275 | } else { |
276 | cancel_delayed_work_sync(&q->delay_work); | |
277 | } | |
1da177e4 LT |
278 | } |
279 | EXPORT_SYMBOL(blk_sync_queue); | |
280 | ||
c246e80d BVA |
281 | /** |
282 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
283 | * @q: The queue to run | |
284 | * | |
285 | * Description: | |
286 | * Invoke request handling on a queue if there are any pending requests. | |
287 | * May be used to restart request handling after a request has completed. | |
288 | * This variant runs the queue whether or not the queue has been | |
289 | * stopped. Must be called with the queue lock held and interrupts | |
290 | * disabled. See also @blk_run_queue. | |
291 | */ | |
292 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
293 | { | |
294 | if (unlikely(blk_queue_dead(q))) | |
295 | return; | |
296 | ||
24faf6f6 BVA |
297 | /* |
298 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
299 | * the queue lock internally. As a result multiple threads may be | |
300 | * running such a request function concurrently. Keep track of the | |
301 | * number of active request_fn invocations such that blk_drain_queue() | |
302 | * can wait until all these request_fn calls have finished. | |
303 | */ | |
304 | q->request_fn_active++; | |
c246e80d | 305 | q->request_fn(q); |
24faf6f6 | 306 | q->request_fn_active--; |
c246e80d | 307 | } |
a7928c15 | 308 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 309 | |
1da177e4 | 310 | /** |
80a4b58e | 311 | * __blk_run_queue - run a single device queue |
1da177e4 | 312 | * @q: The queue to run |
80a4b58e JA |
313 | * |
314 | * Description: | |
315 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 316 | * held and interrupts disabled. |
1da177e4 | 317 | */ |
24ecfbe2 | 318 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 319 | { |
a538cd03 TH |
320 | if (unlikely(blk_queue_stopped(q))) |
321 | return; | |
322 | ||
c246e80d | 323 | __blk_run_queue_uncond(q); |
75ad23bc NP |
324 | } |
325 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 326 | |
24ecfbe2 CH |
327 | /** |
328 | * blk_run_queue_async - run a single device queue in workqueue context | |
329 | * @q: The queue to run | |
330 | * | |
331 | * Description: | |
332 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
70460571 | 333 | * of us. The caller must hold the queue lock. |
24ecfbe2 CH |
334 | */ |
335 | void blk_run_queue_async(struct request_queue *q) | |
336 | { | |
70460571 | 337 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 338 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 339 | } |
c21e6beb | 340 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 341 | |
75ad23bc NP |
342 | /** |
343 | * blk_run_queue - run a single device queue | |
344 | * @q: The queue to run | |
80a4b58e JA |
345 | * |
346 | * Description: | |
347 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 348 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
349 | */ |
350 | void blk_run_queue(struct request_queue *q) | |
351 | { | |
352 | unsigned long flags; | |
353 | ||
354 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 355 | __blk_run_queue(q); |
1da177e4 LT |
356 | spin_unlock_irqrestore(q->queue_lock, flags); |
357 | } | |
358 | EXPORT_SYMBOL(blk_run_queue); | |
359 | ||
165125e1 | 360 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
361 | { |
362 | kobject_put(&q->kobj); | |
363 | } | |
d86e0e83 | 364 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 365 | |
e3c78ca5 | 366 | /** |
807592a4 | 367 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 368 | * @q: queue to drain |
c9a929dd | 369 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 370 | * |
c9a929dd TH |
371 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
372 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
373 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 374 | */ |
807592a4 BVA |
375 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
376 | __releases(q->queue_lock) | |
377 | __acquires(q->queue_lock) | |
e3c78ca5 | 378 | { |
458f27a9 AH |
379 | int i; |
380 | ||
807592a4 BVA |
381 | lockdep_assert_held(q->queue_lock); |
382 | ||
e3c78ca5 | 383 | while (true) { |
481a7d64 | 384 | bool drain = false; |
e3c78ca5 | 385 | |
b855b04a TH |
386 | /* |
387 | * The caller might be trying to drain @q before its | |
388 | * elevator is initialized. | |
389 | */ | |
390 | if (q->elevator) | |
391 | elv_drain_elevator(q); | |
392 | ||
5efd6113 | 393 | blkcg_drain_queue(q); |
e3c78ca5 | 394 | |
4eabc941 TH |
395 | /* |
396 | * This function might be called on a queue which failed | |
b855b04a TH |
397 | * driver init after queue creation or is not yet fully |
398 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
399 | * in such cases. Kick queue iff dispatch queue has | |
400 | * something on it and @q has request_fn set. | |
4eabc941 | 401 | */ |
b855b04a | 402 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 403 | __blk_run_queue(q); |
c9a929dd | 404 | |
8a5ecdd4 | 405 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 406 | drain |= q->request_fn_active; |
481a7d64 TH |
407 | |
408 | /* | |
409 | * Unfortunately, requests are queued at and tracked from | |
410 | * multiple places and there's no single counter which can | |
411 | * be drained. Check all the queues and counters. | |
412 | */ | |
413 | if (drain_all) { | |
e97c293c | 414 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
415 | drain |= !list_empty(&q->queue_head); |
416 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 417 | drain |= q->nr_rqs[i]; |
481a7d64 | 418 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
419 | if (fq) |
420 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
421 | } |
422 | } | |
e3c78ca5 | 423 | |
481a7d64 | 424 | if (!drain) |
e3c78ca5 | 425 | break; |
807592a4 BVA |
426 | |
427 | spin_unlock_irq(q->queue_lock); | |
428 | ||
e3c78ca5 | 429 | msleep(10); |
807592a4 BVA |
430 | |
431 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 432 | } |
458f27a9 AH |
433 | |
434 | /* | |
435 | * With queue marked dead, any woken up waiter will fail the | |
436 | * allocation path, so the wakeup chaining is lost and we're | |
437 | * left with hung waiters. We need to wake up those waiters. | |
438 | */ | |
439 | if (q->request_fn) { | |
a051661c TH |
440 | struct request_list *rl; |
441 | ||
a051661c TH |
442 | blk_queue_for_each_rl(rl, q) |
443 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
444 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 445 | } |
e3c78ca5 TH |
446 | } |
447 | ||
d732580b TH |
448 | /** |
449 | * blk_queue_bypass_start - enter queue bypass mode | |
450 | * @q: queue of interest | |
451 | * | |
452 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
453 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 454 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
455 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
456 | * inside queue or RCU read lock. | |
d732580b TH |
457 | */ |
458 | void blk_queue_bypass_start(struct request_queue *q) | |
459 | { | |
460 | spin_lock_irq(q->queue_lock); | |
776687bc | 461 | q->bypass_depth++; |
d732580b TH |
462 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
463 | spin_unlock_irq(q->queue_lock); | |
464 | ||
776687bc TH |
465 | /* |
466 | * Queues start drained. Skip actual draining till init is | |
467 | * complete. This avoids lenghty delays during queue init which | |
468 | * can happen many times during boot. | |
469 | */ | |
470 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
471 | spin_lock_irq(q->queue_lock); |
472 | __blk_drain_queue(q, false); | |
473 | spin_unlock_irq(q->queue_lock); | |
474 | ||
b82d4b19 TH |
475 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
476 | synchronize_rcu(); | |
477 | } | |
d732580b TH |
478 | } |
479 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
480 | ||
481 | /** | |
482 | * blk_queue_bypass_end - leave queue bypass mode | |
483 | * @q: queue of interest | |
484 | * | |
485 | * Leave bypass mode and restore the normal queueing behavior. | |
486 | */ | |
487 | void blk_queue_bypass_end(struct request_queue *q) | |
488 | { | |
489 | spin_lock_irq(q->queue_lock); | |
490 | if (!--q->bypass_depth) | |
491 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
492 | WARN_ON_ONCE(q->bypass_depth < 0); | |
493 | spin_unlock_irq(q->queue_lock); | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
496 | ||
aed3ea94 JA |
497 | void blk_set_queue_dying(struct request_queue *q) |
498 | { | |
1b856086 BVA |
499 | spin_lock_irq(q->queue_lock); |
500 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
501 | spin_unlock_irq(q->queue_lock); | |
aed3ea94 JA |
502 | |
503 | if (q->mq_ops) | |
504 | blk_mq_wake_waiters(q); | |
505 | else { | |
506 | struct request_list *rl; | |
507 | ||
bbfc3c5d | 508 | spin_lock_irq(q->queue_lock); |
aed3ea94 JA |
509 | blk_queue_for_each_rl(rl, q) { |
510 | if (rl->rq_pool) { | |
511 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
512 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
513 | } | |
514 | } | |
bbfc3c5d | 515 | spin_unlock_irq(q->queue_lock); |
aed3ea94 JA |
516 | } |
517 | } | |
518 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
519 | ||
c9a929dd TH |
520 | /** |
521 | * blk_cleanup_queue - shutdown a request queue | |
522 | * @q: request queue to shutdown | |
523 | * | |
c246e80d BVA |
524 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
525 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 526 | */ |
6728cb0e | 527 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 528 | { |
c9a929dd | 529 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 530 | |
3f3299d5 | 531 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 532 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 533 | blk_set_queue_dying(q); |
c9a929dd | 534 | spin_lock_irq(lock); |
6ecf23af | 535 | |
80fd9979 | 536 | /* |
3f3299d5 | 537 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
538 | * that, unlike blk_queue_bypass_start(), we aren't performing |
539 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
540 | * as some drivers create and destroy a lot of queues while | |
541 | * probing. This is still safe because blk_release_queue() will be | |
542 | * called only after the queue refcnt drops to zero and nothing, | |
543 | * RCU or not, would be traversing the queue by then. | |
544 | */ | |
6ecf23af TH |
545 | q->bypass_depth++; |
546 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
547 | ||
c9a929dd TH |
548 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
549 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 550 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
551 | spin_unlock_irq(lock); |
552 | mutex_unlock(&q->sysfs_lock); | |
553 | ||
c246e80d BVA |
554 | /* |
555 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
556 | * prevent that q->request_fn() gets invoked after draining finished. | |
557 | */ | |
3ef28e83 DW |
558 | blk_freeze_queue(q); |
559 | spin_lock_irq(lock); | |
560 | if (!q->mq_ops) | |
43a5e4e2 | 561 | __blk_drain_queue(q, true); |
c246e80d | 562 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 563 | spin_unlock_irq(lock); |
c9a929dd | 564 | |
5a48fc14 DW |
565 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
566 | blk_flush_integrity(); | |
567 | ||
c9a929dd | 568 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 569 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
570 | blk_sync_queue(q); |
571 | ||
45a9c9d9 BVA |
572 | if (q->mq_ops) |
573 | blk_mq_free_queue(q); | |
3ef28e83 | 574 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 575 | |
5e5cfac0 AH |
576 | spin_lock_irq(lock); |
577 | if (q->queue_lock != &q->__queue_lock) | |
578 | q->queue_lock = &q->__queue_lock; | |
579 | spin_unlock_irq(lock); | |
580 | ||
c9a929dd | 581 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
582 | blk_put_queue(q); |
583 | } | |
1da177e4 LT |
584 | EXPORT_SYMBOL(blk_cleanup_queue); |
585 | ||
271508db | 586 | /* Allocate memory local to the request queue */ |
6d247d7f | 587 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) |
271508db | 588 | { |
6d247d7f CH |
589 | struct request_queue *q = data; |
590 | ||
591 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
271508db DR |
592 | } |
593 | ||
6d247d7f | 594 | static void free_request_simple(void *element, void *data) |
271508db DR |
595 | { |
596 | kmem_cache_free(request_cachep, element); | |
597 | } | |
598 | ||
6d247d7f CH |
599 | static void *alloc_request_size(gfp_t gfp_mask, void *data) |
600 | { | |
601 | struct request_queue *q = data; | |
602 | struct request *rq; | |
603 | ||
604 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
605 | q->node); | |
606 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
607 | kfree(rq); | |
608 | rq = NULL; | |
609 | } | |
610 | return rq; | |
611 | } | |
612 | ||
613 | static void free_request_size(void *element, void *data) | |
614 | { | |
615 | struct request_queue *q = data; | |
616 | ||
617 | if (q->exit_rq_fn) | |
618 | q->exit_rq_fn(q, element); | |
619 | kfree(element); | |
620 | } | |
621 | ||
5b788ce3 TH |
622 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
623 | gfp_t gfp_mask) | |
1da177e4 | 624 | { |
1abec4fd MS |
625 | if (unlikely(rl->rq_pool)) |
626 | return 0; | |
627 | ||
5b788ce3 | 628 | rl->q = q; |
1faa16d2 JA |
629 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
630 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
631 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
632 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 633 | |
6d247d7f CH |
634 | if (q->cmd_size) { |
635 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
636 | alloc_request_size, free_request_size, | |
637 | q, gfp_mask, q->node); | |
638 | } else { | |
639 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
640 | alloc_request_simple, free_request_simple, | |
641 | q, gfp_mask, q->node); | |
642 | } | |
1da177e4 LT |
643 | if (!rl->rq_pool) |
644 | return -ENOMEM; | |
645 | ||
646 | return 0; | |
647 | } | |
648 | ||
5b788ce3 TH |
649 | void blk_exit_rl(struct request_list *rl) |
650 | { | |
651 | if (rl->rq_pool) | |
652 | mempool_destroy(rl->rq_pool); | |
653 | } | |
654 | ||
165125e1 | 655 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 656 | { |
c304a51b | 657 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
658 | } |
659 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 660 | |
6f3b0e8b | 661 | int blk_queue_enter(struct request_queue *q, bool nowait) |
3ef28e83 DW |
662 | { |
663 | while (true) { | |
664 | int ret; | |
665 | ||
666 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
667 | return 0; | |
668 | ||
6f3b0e8b | 669 | if (nowait) |
3ef28e83 DW |
670 | return -EBUSY; |
671 | ||
672 | ret = wait_event_interruptible(q->mq_freeze_wq, | |
673 | !atomic_read(&q->mq_freeze_depth) || | |
674 | blk_queue_dying(q)); | |
675 | if (blk_queue_dying(q)) | |
676 | return -ENODEV; | |
677 | if (ret) | |
678 | return ret; | |
679 | } | |
680 | } | |
681 | ||
682 | void blk_queue_exit(struct request_queue *q) | |
683 | { | |
684 | percpu_ref_put(&q->q_usage_counter); | |
685 | } | |
686 | ||
687 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
688 | { | |
689 | struct request_queue *q = | |
690 | container_of(ref, struct request_queue, q_usage_counter); | |
691 | ||
692 | wake_up_all(&q->mq_freeze_wq); | |
693 | } | |
694 | ||
287922eb CH |
695 | static void blk_rq_timed_out_timer(unsigned long data) |
696 | { | |
697 | struct request_queue *q = (struct request_queue *)data; | |
698 | ||
699 | kblockd_schedule_work(&q->timeout_work); | |
700 | } | |
701 | ||
165125e1 | 702 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 703 | { |
165125e1 | 704 | struct request_queue *q; |
1946089a | 705 | |
8324aa91 | 706 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 707 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
708 | if (!q) |
709 | return NULL; | |
710 | ||
00380a40 | 711 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 712 | if (q->id < 0) |
3d2936f4 | 713 | goto fail_q; |
a73f730d | 714 | |
54efd50b KO |
715 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0); |
716 | if (!q->bio_split) | |
717 | goto fail_id; | |
718 | ||
d03f6cdc JK |
719 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
720 | if (!q->backing_dev_info) | |
721 | goto fail_split; | |
722 | ||
dc3b17cc | 723 | q->backing_dev_info->ra_pages = |
09cbfeaf | 724 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
dc3b17cc JK |
725 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
726 | q->backing_dev_info->name = "block"; | |
5151412d | 727 | q->node = node_id; |
0989a025 | 728 | |
dc3b17cc | 729 | setup_timer(&q->backing_dev_info->laptop_mode_wb_timer, |
31373d09 | 730 | laptop_mode_timer_fn, (unsigned long) q); |
242f9dcb | 731 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 732 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 733 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 734 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 735 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 736 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 737 | #endif |
3cca6dc1 | 738 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 739 | |
8324aa91 | 740 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 741 | |
483f4afc | 742 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 743 | spin_lock_init(&q->__queue_lock); |
483f4afc | 744 | |
c94a96ac VG |
745 | /* |
746 | * By default initialize queue_lock to internal lock and driver can | |
747 | * override it later if need be. | |
748 | */ | |
749 | q->queue_lock = &q->__queue_lock; | |
750 | ||
b82d4b19 TH |
751 | /* |
752 | * A queue starts its life with bypass turned on to avoid | |
753 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
754 | * init. The initial bypass will be finished when the queue is |
755 | * registered by blk_register_queue(). | |
b82d4b19 TH |
756 | */ |
757 | q->bypass_depth = 1; | |
758 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
759 | ||
320ae51f JA |
760 | init_waitqueue_head(&q->mq_freeze_wq); |
761 | ||
3ef28e83 DW |
762 | /* |
763 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
764 | * See blk_register_queue() for details. | |
765 | */ | |
766 | if (percpu_ref_init(&q->q_usage_counter, | |
767 | blk_queue_usage_counter_release, | |
768 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 769 | goto fail_bdi; |
f51b802c | 770 | |
3ef28e83 DW |
771 | if (blkcg_init_queue(q)) |
772 | goto fail_ref; | |
773 | ||
1da177e4 | 774 | return q; |
a73f730d | 775 | |
3ef28e83 DW |
776 | fail_ref: |
777 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 778 | fail_bdi: |
d03f6cdc | 779 | bdi_put(q->backing_dev_info); |
54efd50b KO |
780 | fail_split: |
781 | bioset_free(q->bio_split); | |
a73f730d TH |
782 | fail_id: |
783 | ida_simple_remove(&blk_queue_ida, q->id); | |
784 | fail_q: | |
785 | kmem_cache_free(blk_requestq_cachep, q); | |
786 | return NULL; | |
1da177e4 | 787 | } |
1946089a | 788 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
789 | |
790 | /** | |
791 | * blk_init_queue - prepare a request queue for use with a block device | |
792 | * @rfn: The function to be called to process requests that have been | |
793 | * placed on the queue. | |
794 | * @lock: Request queue spin lock | |
795 | * | |
796 | * Description: | |
797 | * If a block device wishes to use the standard request handling procedures, | |
798 | * which sorts requests and coalesces adjacent requests, then it must | |
799 | * call blk_init_queue(). The function @rfn will be called when there | |
800 | * are requests on the queue that need to be processed. If the device | |
801 | * supports plugging, then @rfn may not be called immediately when requests | |
802 | * are available on the queue, but may be called at some time later instead. | |
803 | * Plugged queues are generally unplugged when a buffer belonging to one | |
804 | * of the requests on the queue is needed, or due to memory pressure. | |
805 | * | |
806 | * @rfn is not required, or even expected, to remove all requests off the | |
807 | * queue, but only as many as it can handle at a time. If it does leave | |
808 | * requests on the queue, it is responsible for arranging that the requests | |
809 | * get dealt with eventually. | |
810 | * | |
811 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
812 | * request queue; this lock will be taken also from interrupt context, so irq |
813 | * disabling is needed for it. | |
1da177e4 | 814 | * |
710027a4 | 815 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
816 | * it didn't succeed. |
817 | * | |
818 | * Note: | |
819 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
820 | * when the block device is deactivated (such as at module unload). | |
821 | **/ | |
1946089a | 822 | |
165125e1 | 823 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 824 | { |
c304a51b | 825 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
826 | } |
827 | EXPORT_SYMBOL(blk_init_queue); | |
828 | ||
165125e1 | 829 | struct request_queue * |
1946089a CL |
830 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
831 | { | |
5ea708d1 | 832 | struct request_queue *q; |
1da177e4 | 833 | |
5ea708d1 CH |
834 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
835 | if (!q) | |
c86d1b8a MS |
836 | return NULL; |
837 | ||
5ea708d1 CH |
838 | q->request_fn = rfn; |
839 | if (lock) | |
840 | q->queue_lock = lock; | |
841 | if (blk_init_allocated_queue(q) < 0) { | |
842 | blk_cleanup_queue(q); | |
843 | return NULL; | |
844 | } | |
18741986 | 845 | |
7982e90c | 846 | return q; |
01effb0d MS |
847 | } |
848 | EXPORT_SYMBOL(blk_init_queue_node); | |
849 | ||
dece1635 | 850 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 851 | |
1da177e4 | 852 | |
5ea708d1 CH |
853 | int blk_init_allocated_queue(struct request_queue *q) |
854 | { | |
6d247d7f | 855 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); |
ba483388 | 856 | if (!q->fq) |
5ea708d1 | 857 | return -ENOMEM; |
7982e90c | 858 | |
6d247d7f CH |
859 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) |
860 | goto out_free_flush_queue; | |
7982e90c | 861 | |
a051661c | 862 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
6d247d7f | 863 | goto out_exit_flush_rq; |
1da177e4 | 864 | |
287922eb | 865 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
60ea8226 | 866 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac | 867 | |
f3b144aa JA |
868 | /* |
869 | * This also sets hw/phys segments, boundary and size | |
870 | */ | |
c20e8de2 | 871 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 872 | |
44ec9542 AS |
873 | q->sg_reserved_size = INT_MAX; |
874 | ||
eb1c160b TS |
875 | /* Protect q->elevator from elevator_change */ |
876 | mutex_lock(&q->sysfs_lock); | |
877 | ||
b82d4b19 | 878 | /* init elevator */ |
eb1c160b TS |
879 | if (elevator_init(q, NULL)) { |
880 | mutex_unlock(&q->sysfs_lock); | |
6d247d7f | 881 | goto out_exit_flush_rq; |
eb1c160b TS |
882 | } |
883 | ||
884 | mutex_unlock(&q->sysfs_lock); | |
5ea708d1 | 885 | return 0; |
eb1c160b | 886 | |
6d247d7f CH |
887 | out_exit_flush_rq: |
888 | if (q->exit_rq_fn) | |
889 | q->exit_rq_fn(q, q->fq->flush_rq); | |
890 | out_free_flush_queue: | |
ba483388 | 891 | blk_free_flush_queue(q->fq); |
87760e5e | 892 | wbt_exit(q); |
5ea708d1 | 893 | return -ENOMEM; |
1da177e4 | 894 | } |
5151412d | 895 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 896 | |
09ac46c4 | 897 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 898 | { |
3f3299d5 | 899 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
900 | __blk_get_queue(q); |
901 | return true; | |
1da177e4 LT |
902 | } |
903 | ||
09ac46c4 | 904 | return false; |
1da177e4 | 905 | } |
d86e0e83 | 906 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 907 | |
5b788ce3 | 908 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 909 | { |
e8064021 | 910 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 911 | elv_put_request(rl->q, rq); |
f1f8cc94 | 912 | if (rq->elv.icq) |
11a3122f | 913 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
914 | } |
915 | ||
5b788ce3 | 916 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
917 | } |
918 | ||
1da177e4 LT |
919 | /* |
920 | * ioc_batching returns true if the ioc is a valid batching request and | |
921 | * should be given priority access to a request. | |
922 | */ | |
165125e1 | 923 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
924 | { |
925 | if (!ioc) | |
926 | return 0; | |
927 | ||
928 | /* | |
929 | * Make sure the process is able to allocate at least 1 request | |
930 | * even if the batch times out, otherwise we could theoretically | |
931 | * lose wakeups. | |
932 | */ | |
933 | return ioc->nr_batch_requests == q->nr_batching || | |
934 | (ioc->nr_batch_requests > 0 | |
935 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
936 | } | |
937 | ||
938 | /* | |
939 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
940 | * will cause the process to be a "batcher" on all queues in the system. This | |
941 | * is the behaviour we want though - once it gets a wakeup it should be given | |
942 | * a nice run. | |
943 | */ | |
165125e1 | 944 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
945 | { |
946 | if (!ioc || ioc_batching(q, ioc)) | |
947 | return; | |
948 | ||
949 | ioc->nr_batch_requests = q->nr_batching; | |
950 | ioc->last_waited = jiffies; | |
951 | } | |
952 | ||
5b788ce3 | 953 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 954 | { |
5b788ce3 | 955 | struct request_queue *q = rl->q; |
1da177e4 | 956 | |
d40f75a0 TH |
957 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
958 | blk_clear_congested(rl, sync); | |
1da177e4 | 959 | |
1faa16d2 JA |
960 | if (rl->count[sync] + 1 <= q->nr_requests) { |
961 | if (waitqueue_active(&rl->wait[sync])) | |
962 | wake_up(&rl->wait[sync]); | |
1da177e4 | 963 | |
5b788ce3 | 964 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
965 | } |
966 | } | |
967 | ||
968 | /* | |
969 | * A request has just been released. Account for it, update the full and | |
970 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
971 | */ | |
e8064021 CH |
972 | static void freed_request(struct request_list *rl, bool sync, |
973 | req_flags_t rq_flags) | |
1da177e4 | 974 | { |
5b788ce3 | 975 | struct request_queue *q = rl->q; |
1da177e4 | 976 | |
8a5ecdd4 | 977 | q->nr_rqs[sync]--; |
1faa16d2 | 978 | rl->count[sync]--; |
e8064021 | 979 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 980 | q->nr_rqs_elvpriv--; |
1da177e4 | 981 | |
5b788ce3 | 982 | __freed_request(rl, sync); |
1da177e4 | 983 | |
1faa16d2 | 984 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 985 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
986 | } |
987 | ||
e3a2b3f9 JA |
988 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
989 | { | |
990 | struct request_list *rl; | |
d40f75a0 | 991 | int on_thresh, off_thresh; |
e3a2b3f9 JA |
992 | |
993 | spin_lock_irq(q->queue_lock); | |
994 | q->nr_requests = nr; | |
995 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
996 | on_thresh = queue_congestion_on_threshold(q); |
997 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 998 | |
d40f75a0 TH |
999 | blk_queue_for_each_rl(rl, q) { |
1000 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1001 | blk_set_congested(rl, BLK_RW_SYNC); | |
1002 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1003 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 1004 | |
d40f75a0 TH |
1005 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1006 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1007 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1008 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1009 | |
e3a2b3f9 JA |
1010 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1011 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1012 | } else { | |
1013 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1014 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1015 | } | |
1016 | ||
1017 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1018 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1019 | } else { | |
1020 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1021 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1022 | } | |
1023 | } | |
1024 | ||
1025 | spin_unlock_irq(q->queue_lock); | |
1026 | return 0; | |
1027 | } | |
1028 | ||
da8303c6 | 1029 | /** |
a06e05e6 | 1030 | * __get_request - get a free request |
5b788ce3 | 1031 | * @rl: request list to allocate from |
ef295ecf | 1032 | * @op: operation and flags |
da8303c6 TH |
1033 | * @bio: bio to allocate request for (can be %NULL) |
1034 | * @gfp_mask: allocation mask | |
1035 | * | |
1036 | * Get a free request from @q. This function may fail under memory | |
1037 | * pressure or if @q is dead. | |
1038 | * | |
da3dae54 | 1039 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1040 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1041 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1042 | */ |
ef295ecf CH |
1043 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
1044 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1045 | { |
5b788ce3 | 1046 | struct request_queue *q = rl->q; |
b679281a | 1047 | struct request *rq; |
7f4b35d1 TH |
1048 | struct elevator_type *et = q->elevator->type; |
1049 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1050 | struct io_cq *icq = NULL; |
ef295ecf | 1051 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1052 | int may_queue; |
e8064021 | 1053 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1054 | |
3f3299d5 | 1055 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1056 | return ERR_PTR(-ENODEV); |
da8303c6 | 1057 | |
ef295ecf | 1058 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1059 | if (may_queue == ELV_MQUEUE_NO) |
1060 | goto rq_starved; | |
1061 | ||
1faa16d2 JA |
1062 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1063 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1064 | /* |
1065 | * The queue will fill after this allocation, so set | |
1066 | * it as full, and mark this process as "batching". | |
1067 | * This process will be allowed to complete a batch of | |
1068 | * requests, others will be blocked. | |
1069 | */ | |
5b788ce3 | 1070 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1071 | ioc_set_batching(q, ioc); |
5b788ce3 | 1072 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1073 | } else { |
1074 | if (may_queue != ELV_MQUEUE_MUST | |
1075 | && !ioc_batching(q, ioc)) { | |
1076 | /* | |
1077 | * The queue is full and the allocating | |
1078 | * process is not a "batcher", and not | |
1079 | * exempted by the IO scheduler | |
1080 | */ | |
a492f075 | 1081 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1082 | } |
1083 | } | |
1da177e4 | 1084 | } |
d40f75a0 | 1085 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1086 | } |
1087 | ||
082cf69e JA |
1088 | /* |
1089 | * Only allow batching queuers to allocate up to 50% over the defined | |
1090 | * limit of requests, otherwise we could have thousands of requests | |
1091 | * allocated with any setting of ->nr_requests | |
1092 | */ | |
1faa16d2 | 1093 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1094 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1095 | |
8a5ecdd4 | 1096 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1097 | rl->count[is_sync]++; |
1098 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1099 | |
f1f8cc94 TH |
1100 | /* |
1101 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1102 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1103 | * prevent the current elevator from being destroyed until the new |
1104 | * request is freed. This guarantees icq's won't be destroyed and | |
1105 | * makes creating new ones safe. | |
1106 | * | |
e6f7f93d CH |
1107 | * Flush requests do not use the elevator so skip initialization. |
1108 | * This allows a request to share the flush and elevator data. | |
1109 | * | |
f1f8cc94 TH |
1110 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
1111 | * it will be created after releasing queue_lock. | |
1112 | */ | |
e6f7f93d | 1113 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { |
e8064021 | 1114 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1115 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1116 | if (et->icq_cache && ioc) |
1117 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1118 | } |
cb98fc8b | 1119 | |
f253b86b | 1120 | if (blk_queue_io_stat(q)) |
e8064021 | 1121 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1122 | spin_unlock_irq(q->queue_lock); |
1123 | ||
29e2b09a | 1124 | /* allocate and init request */ |
5b788ce3 | 1125 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1126 | if (!rq) |
b679281a | 1127 | goto fail_alloc; |
1da177e4 | 1128 | |
29e2b09a | 1129 | blk_rq_init(q, rq); |
a051661c | 1130 | blk_rq_set_rl(rq, rl); |
5dc8b362 | 1131 | blk_rq_set_prio(rq, ioc); |
ef295ecf | 1132 | rq->cmd_flags = op; |
e8064021 | 1133 | rq->rq_flags = rq_flags; |
29e2b09a | 1134 | |
aaf7c680 | 1135 | /* init elvpriv */ |
e8064021 | 1136 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1137 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1138 | if (ioc) |
1139 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1140 | if (!icq) |
1141 | goto fail_elvpriv; | |
29e2b09a | 1142 | } |
aaf7c680 TH |
1143 | |
1144 | rq->elv.icq = icq; | |
1145 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1146 | goto fail_elvpriv; | |
1147 | ||
1148 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1149 | if (icq) |
1150 | get_io_context(icq->ioc); | |
1151 | } | |
aaf7c680 | 1152 | out: |
88ee5ef1 JA |
1153 | /* |
1154 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1155 | * OK, if the queue is under the request limit then requests need | |
1156 | * not count toward the nr_batch_requests limit. There will always | |
1157 | * be some limit enforced by BLK_BATCH_TIME. | |
1158 | */ | |
1da177e4 LT |
1159 | if (ioc_batching(q, ioc)) |
1160 | ioc->nr_batch_requests--; | |
6728cb0e | 1161 | |
e6a40b09 | 1162 | trace_block_getrq(q, bio, op); |
1da177e4 | 1163 | return rq; |
b679281a | 1164 | |
aaf7c680 TH |
1165 | fail_elvpriv: |
1166 | /* | |
1167 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1168 | * and may fail indefinitely under memory pressure and thus | |
1169 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1170 | * disturb iosched and blkcg but weird is bettern than dead. | |
1171 | */ | |
7b2b10e0 | 1172 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
dc3b17cc | 1173 | __func__, dev_name(q->backing_dev_info->dev)); |
aaf7c680 | 1174 | |
e8064021 | 1175 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1176 | rq->elv.icq = NULL; |
1177 | ||
1178 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1179 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1180 | spin_unlock_irq(q->queue_lock); |
1181 | goto out; | |
1182 | ||
b679281a TH |
1183 | fail_alloc: |
1184 | /* | |
1185 | * Allocation failed presumably due to memory. Undo anything we | |
1186 | * might have messed up. | |
1187 | * | |
1188 | * Allocating task should really be put onto the front of the wait | |
1189 | * queue, but this is pretty rare. | |
1190 | */ | |
1191 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1192 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1193 | |
1194 | /* | |
1195 | * in the very unlikely event that allocation failed and no | |
1196 | * requests for this direction was pending, mark us starved so that | |
1197 | * freeing of a request in the other direction will notice | |
1198 | * us. another possible fix would be to split the rq mempool into | |
1199 | * READ and WRITE | |
1200 | */ | |
1201 | rq_starved: | |
1202 | if (unlikely(rl->count[is_sync] == 0)) | |
1203 | rl->starved[is_sync] = 1; | |
a492f075 | 1204 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1205 | } |
1206 | ||
da8303c6 | 1207 | /** |
a06e05e6 | 1208 | * get_request - get a free request |
da8303c6 | 1209 | * @q: request_queue to allocate request from |
ef295ecf | 1210 | * @op: operation and flags |
da8303c6 | 1211 | * @bio: bio to allocate request for (can be %NULL) |
a06e05e6 | 1212 | * @gfp_mask: allocation mask |
da8303c6 | 1213 | * |
d0164adc MG |
1214 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, |
1215 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1216 | * |
da3dae54 | 1217 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1218 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1219 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1220 | */ |
ef295ecf CH |
1221 | static struct request *get_request(struct request_queue *q, unsigned int op, |
1222 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1223 | { |
ef295ecf | 1224 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1225 | DEFINE_WAIT(wait); |
a051661c | 1226 | struct request_list *rl; |
1da177e4 | 1227 | struct request *rq; |
a051661c TH |
1228 | |
1229 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
a06e05e6 | 1230 | retry: |
ef295ecf | 1231 | rq = __get_request(rl, op, bio, gfp_mask); |
a492f075 | 1232 | if (!IS_ERR(rq)) |
a06e05e6 | 1233 | return rq; |
1da177e4 | 1234 | |
d0164adc | 1235 | if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) { |
a051661c | 1236 | blk_put_rl(rl); |
a492f075 | 1237 | return rq; |
a051661c | 1238 | } |
1da177e4 | 1239 | |
a06e05e6 TH |
1240 | /* wait on @rl and retry */ |
1241 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1242 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1243 | |
e6a40b09 | 1244 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1245 | |
a06e05e6 TH |
1246 | spin_unlock_irq(q->queue_lock); |
1247 | io_schedule(); | |
d6344532 | 1248 | |
a06e05e6 TH |
1249 | /* |
1250 | * After sleeping, we become a "batching" process and will be able | |
1251 | * to allocate at least one request, and up to a big batch of them | |
1252 | * for a small period time. See ioc_batching, ioc_set_batching | |
1253 | */ | |
a06e05e6 | 1254 | ioc_set_batching(q, current->io_context); |
05caf8db | 1255 | |
a06e05e6 TH |
1256 | spin_lock_irq(q->queue_lock); |
1257 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1258 | |
a06e05e6 | 1259 | goto retry; |
1da177e4 LT |
1260 | } |
1261 | ||
320ae51f JA |
1262 | static struct request *blk_old_get_request(struct request_queue *q, int rw, |
1263 | gfp_t gfp_mask) | |
1da177e4 LT |
1264 | { |
1265 | struct request *rq; | |
1266 | ||
7f4b35d1 TH |
1267 | /* create ioc upfront */ |
1268 | create_io_context(gfp_mask, q->node); | |
1269 | ||
d6344532 | 1270 | spin_lock_irq(q->queue_lock); |
ef295ecf | 1271 | rq = get_request(q, rw, NULL, gfp_mask); |
0c4de0f3 | 1272 | if (IS_ERR(rq)) { |
da8303c6 | 1273 | spin_unlock_irq(q->queue_lock); |
0c4de0f3 CH |
1274 | return rq; |
1275 | } | |
1da177e4 | 1276 | |
0c4de0f3 CH |
1277 | /* q->queue_lock is unlocked at this point */ |
1278 | rq->__data_len = 0; | |
1279 | rq->__sector = (sector_t) -1; | |
1280 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1281 | return rq; |
1282 | } | |
320ae51f JA |
1283 | |
1284 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
1285 | { | |
1286 | if (q->mq_ops) | |
6f3b0e8b CH |
1287 | return blk_mq_alloc_request(q, rw, |
1288 | (gfp_mask & __GFP_DIRECT_RECLAIM) ? | |
1289 | 0 : BLK_MQ_REQ_NOWAIT); | |
320ae51f JA |
1290 | else |
1291 | return blk_old_get_request(q, rw, gfp_mask); | |
1292 | } | |
1da177e4 LT |
1293 | EXPORT_SYMBOL(blk_get_request); |
1294 | ||
1295 | /** | |
1296 | * blk_requeue_request - put a request back on queue | |
1297 | * @q: request queue where request should be inserted | |
1298 | * @rq: request to be inserted | |
1299 | * | |
1300 | * Description: | |
1301 | * Drivers often keep queueing requests until the hardware cannot accept | |
1302 | * more, when that condition happens we need to put the request back | |
1303 | * on the queue. Must be called with queue lock held. | |
1304 | */ | |
165125e1 | 1305 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1306 | { |
242f9dcb JA |
1307 | blk_delete_timer(rq); |
1308 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1309 | trace_block_rq_requeue(q, rq); |
87760e5e | 1310 | wbt_requeue(q->rq_wb, &rq->issue_stat); |
2056a782 | 1311 | |
e8064021 | 1312 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1313 | blk_queue_end_tag(q, rq); |
1314 | ||
ba396a6c JB |
1315 | BUG_ON(blk_queued_rq(rq)); |
1316 | ||
1da177e4 LT |
1317 | elv_requeue_request(q, rq); |
1318 | } | |
1da177e4 LT |
1319 | EXPORT_SYMBOL(blk_requeue_request); |
1320 | ||
73c10101 JA |
1321 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1322 | int where) | |
1323 | { | |
320ae51f | 1324 | blk_account_io_start(rq, true); |
7eaceacc | 1325 | __elv_add_request(q, rq, where); |
73c10101 JA |
1326 | } |
1327 | ||
074a7aca TH |
1328 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1329 | unsigned long now) | |
1330 | { | |
7276d02e JA |
1331 | int inflight; |
1332 | ||
074a7aca TH |
1333 | if (now == part->stamp) |
1334 | return; | |
1335 | ||
7276d02e JA |
1336 | inflight = part_in_flight(part); |
1337 | if (inflight) { | |
074a7aca | 1338 | __part_stat_add(cpu, part, time_in_queue, |
7276d02e | 1339 | inflight * (now - part->stamp)); |
074a7aca TH |
1340 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1341 | } | |
1342 | part->stamp = now; | |
1343 | } | |
1344 | ||
1345 | /** | |
496aa8a9 RD |
1346 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1347 | * @cpu: cpu number for stats access | |
1348 | * @part: target partition | |
1da177e4 LT |
1349 | * |
1350 | * The average IO queue length and utilisation statistics are maintained | |
1351 | * by observing the current state of the queue length and the amount of | |
1352 | * time it has been in this state for. | |
1353 | * | |
1354 | * Normally, that accounting is done on IO completion, but that can result | |
1355 | * in more than a second's worth of IO being accounted for within any one | |
1356 | * second, leading to >100% utilisation. To deal with that, we call this | |
1357 | * function to do a round-off before returning the results when reading | |
1358 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1359 | * the current jiffies and restarts the counters again. | |
1360 | */ | |
c9959059 | 1361 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1362 | { |
1363 | unsigned long now = jiffies; | |
1364 | ||
074a7aca TH |
1365 | if (part->partno) |
1366 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1367 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1368 | } |
074a7aca | 1369 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1370 | |
47fafbc7 | 1371 | #ifdef CONFIG_PM |
c8158819 LM |
1372 | static void blk_pm_put_request(struct request *rq) |
1373 | { | |
e8064021 | 1374 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1375 | pm_runtime_mark_last_busy(rq->q->dev); |
1376 | } | |
1377 | #else | |
1378 | static inline void blk_pm_put_request(struct request *rq) {} | |
1379 | #endif | |
1380 | ||
1da177e4 LT |
1381 | /* |
1382 | * queue lock must be held | |
1383 | */ | |
165125e1 | 1384 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1385 | { |
e8064021 CH |
1386 | req_flags_t rq_flags = req->rq_flags; |
1387 | ||
1da177e4 LT |
1388 | if (unlikely(!q)) |
1389 | return; | |
1da177e4 | 1390 | |
6f5ba581 CH |
1391 | if (q->mq_ops) { |
1392 | blk_mq_free_request(req); | |
1393 | return; | |
1394 | } | |
1395 | ||
c8158819 LM |
1396 | blk_pm_put_request(req); |
1397 | ||
8922e16c TH |
1398 | elv_completed_request(q, req); |
1399 | ||
1cd96c24 BH |
1400 | /* this is a bio leak */ |
1401 | WARN_ON(req->bio != NULL); | |
1402 | ||
87760e5e JA |
1403 | wbt_done(q->rq_wb, &req->issue_stat); |
1404 | ||
1da177e4 LT |
1405 | /* |
1406 | * Request may not have originated from ll_rw_blk. if not, | |
1407 | * it didn't come out of our reserved rq pools | |
1408 | */ | |
e8064021 | 1409 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1410 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1411 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1412 | |
1da177e4 | 1413 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1414 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1415 | |
a051661c | 1416 | blk_free_request(rl, req); |
e8064021 | 1417 | freed_request(rl, sync, rq_flags); |
a051661c | 1418 | blk_put_rl(rl); |
1da177e4 LT |
1419 | } |
1420 | } | |
6e39b69e MC |
1421 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1422 | ||
1da177e4 LT |
1423 | void blk_put_request(struct request *req) |
1424 | { | |
165125e1 | 1425 | struct request_queue *q = req->q; |
8922e16c | 1426 | |
320ae51f JA |
1427 | if (q->mq_ops) |
1428 | blk_mq_free_request(req); | |
1429 | else { | |
1430 | unsigned long flags; | |
1431 | ||
1432 | spin_lock_irqsave(q->queue_lock, flags); | |
1433 | __blk_put_request(q, req); | |
1434 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1435 | } | |
1da177e4 | 1436 | } |
1da177e4 LT |
1437 | EXPORT_SYMBOL(blk_put_request); |
1438 | ||
320ae51f JA |
1439 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1440 | struct bio *bio) | |
73c10101 | 1441 | { |
1eff9d32 | 1442 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1443 | |
73c10101 JA |
1444 | if (!ll_back_merge_fn(q, req, bio)) |
1445 | return false; | |
1446 | ||
8c1cf6bb | 1447 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1448 | |
1449 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1450 | blk_rq_set_mixed_merge(req); | |
1451 | ||
1452 | req->biotail->bi_next = bio; | |
1453 | req->biotail = bio; | |
4f024f37 | 1454 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1455 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1456 | ||
320ae51f | 1457 | blk_account_io_start(req, false); |
73c10101 JA |
1458 | return true; |
1459 | } | |
1460 | ||
320ae51f JA |
1461 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1462 | struct bio *bio) | |
73c10101 | 1463 | { |
1eff9d32 | 1464 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1465 | |
73c10101 JA |
1466 | if (!ll_front_merge_fn(q, req, bio)) |
1467 | return false; | |
1468 | ||
8c1cf6bb | 1469 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1470 | |
1471 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1472 | blk_rq_set_mixed_merge(req); | |
1473 | ||
73c10101 JA |
1474 | bio->bi_next = req->bio; |
1475 | req->bio = bio; | |
1476 | ||
4f024f37 KO |
1477 | req->__sector = bio->bi_iter.bi_sector; |
1478 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1479 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1480 | ||
320ae51f | 1481 | blk_account_io_start(req, false); |
73c10101 JA |
1482 | return true; |
1483 | } | |
1484 | ||
1e739730 CH |
1485 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
1486 | struct bio *bio) | |
1487 | { | |
1488 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1489 | ||
1490 | if (segments >= queue_max_discard_segments(q)) | |
1491 | goto no_merge; | |
1492 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1493 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1494 | goto no_merge; | |
1495 | ||
1496 | req->biotail->bi_next = bio; | |
1497 | req->biotail = bio; | |
1498 | req->__data_len += bio->bi_iter.bi_size; | |
1499 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1500 | req->nr_phys_segments = segments + 1; | |
1501 | ||
1502 | blk_account_io_start(req, false); | |
1503 | return true; | |
1504 | no_merge: | |
1505 | req_set_nomerge(q, req); | |
1506 | return false; | |
1507 | } | |
1508 | ||
bd87b589 | 1509 | /** |
320ae51f | 1510 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1511 | * @q: request_queue new bio is being queued at |
1512 | * @bio: new bio being queued | |
1513 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1514 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1515 | * another request associated with @q is found on the plug list | |
1516 | * (optional, may be %NULL) | |
bd87b589 TH |
1517 | * |
1518 | * Determine whether @bio being queued on @q can be merged with a request | |
1519 | * on %current's plugged list. Returns %true if merge was successful, | |
1520 | * otherwise %false. | |
1521 | * | |
07c2bd37 TH |
1522 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1523 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1524 | * than scheduling, and the request, while may have elvpriv data, is not | |
1525 | * added on the elevator at this point. In addition, we don't have | |
1526 | * reliable access to the elevator outside queue lock. Only check basic | |
1527 | * merging parameters without querying the elevator. | |
da41a589 RE |
1528 | * |
1529 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1530 | */ |
320ae51f | 1531 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1532 | unsigned int *request_count, |
1533 | struct request **same_queue_rq) | |
73c10101 JA |
1534 | { |
1535 | struct blk_plug *plug; | |
1536 | struct request *rq; | |
92f399c7 | 1537 | struct list_head *plug_list; |
73c10101 | 1538 | |
bd87b589 | 1539 | plug = current->plug; |
73c10101 | 1540 | if (!plug) |
34fe7c05 | 1541 | return false; |
56ebdaf2 | 1542 | *request_count = 0; |
73c10101 | 1543 | |
92f399c7 SL |
1544 | if (q->mq_ops) |
1545 | plug_list = &plug->mq_list; | |
1546 | else | |
1547 | plug_list = &plug->list; | |
1548 | ||
1549 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 1550 | bool merged = false; |
73c10101 | 1551 | |
5b3f341f | 1552 | if (rq->q == q) { |
1b2e19f1 | 1553 | (*request_count)++; |
5b3f341f SL |
1554 | /* |
1555 | * Only blk-mq multiple hardware queues case checks the | |
1556 | * rq in the same queue, there should be only one such | |
1557 | * rq in a queue | |
1558 | **/ | |
1559 | if (same_queue_rq) | |
1560 | *same_queue_rq = rq; | |
1561 | } | |
56ebdaf2 | 1562 | |
07c2bd37 | 1563 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1564 | continue; |
1565 | ||
34fe7c05 CH |
1566 | switch (blk_try_merge(rq, bio)) { |
1567 | case ELEVATOR_BACK_MERGE: | |
1568 | merged = bio_attempt_back_merge(q, rq, bio); | |
1569 | break; | |
1570 | case ELEVATOR_FRONT_MERGE: | |
1571 | merged = bio_attempt_front_merge(q, rq, bio); | |
1572 | break; | |
1e739730 CH |
1573 | case ELEVATOR_DISCARD_MERGE: |
1574 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1575 | break; | |
34fe7c05 CH |
1576 | default: |
1577 | break; | |
73c10101 | 1578 | } |
34fe7c05 CH |
1579 | |
1580 | if (merged) | |
1581 | return true; | |
73c10101 | 1582 | } |
34fe7c05 CH |
1583 | |
1584 | return false; | |
73c10101 JA |
1585 | } |
1586 | ||
0809e3ac JM |
1587 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1588 | { | |
1589 | struct blk_plug *plug; | |
1590 | struct request *rq; | |
1591 | struct list_head *plug_list; | |
1592 | unsigned int ret = 0; | |
1593 | ||
1594 | plug = current->plug; | |
1595 | if (!plug) | |
1596 | goto out; | |
1597 | ||
1598 | if (q->mq_ops) | |
1599 | plug_list = &plug->mq_list; | |
1600 | else | |
1601 | plug_list = &plug->list; | |
1602 | ||
1603 | list_for_each_entry(rq, plug_list, queuelist) { | |
1604 | if (rq->q == q) | |
1605 | ret++; | |
1606 | } | |
1607 | out: | |
1608 | return ret; | |
1609 | } | |
1610 | ||
86db1e29 | 1611 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1612 | { |
1eff9d32 | 1613 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1614 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1615 | |
52d9e675 | 1616 | req->errors = 0; |
4f024f37 | 1617 | req->__sector = bio->bi_iter.bi_sector; |
5dc8b362 AM |
1618 | if (ioprio_valid(bio_prio(bio))) |
1619 | req->ioprio = bio_prio(bio); | |
bc1c56fd | 1620 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1621 | } |
1622 | ||
dece1635 | 1623 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1624 | { |
73c10101 | 1625 | struct blk_plug *plug; |
34fe7c05 | 1626 | int where = ELEVATOR_INSERT_SORT; |
e4d750c9 | 1627 | struct request *req, *free; |
56ebdaf2 | 1628 | unsigned int request_count = 0; |
87760e5e | 1629 | unsigned int wb_acct; |
1da177e4 | 1630 | |
1da177e4 LT |
1631 | /* |
1632 | * low level driver can indicate that it wants pages above a | |
1633 | * certain limit bounced to low memory (ie for highmem, or even | |
1634 | * ISA dma in theory) | |
1635 | */ | |
1636 | blk_queue_bounce(q, &bio); | |
1637 | ||
23688bf4 JN |
1638 | blk_queue_split(q, &bio, q->bio_split); |
1639 | ||
ffecfd1a | 1640 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) { |
4246a0b6 CH |
1641 | bio->bi_error = -EIO; |
1642 | bio_endio(bio); | |
dece1635 | 1643 | return BLK_QC_T_NONE; |
ffecfd1a DW |
1644 | } |
1645 | ||
f73f44eb | 1646 | if (op_is_flush(bio->bi_opf)) { |
73c10101 | 1647 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1648 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1649 | goto get_rq; |
1650 | } | |
1651 | ||
73c10101 JA |
1652 | /* |
1653 | * Check if we can merge with the plugged list before grabbing | |
1654 | * any locks. | |
1655 | */ | |
0809e3ac JM |
1656 | if (!blk_queue_nomerges(q)) { |
1657 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 1658 | return BLK_QC_T_NONE; |
0809e3ac JM |
1659 | } else |
1660 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 1661 | |
73c10101 | 1662 | spin_lock_irq(q->queue_lock); |
2056a782 | 1663 | |
34fe7c05 CH |
1664 | switch (elv_merge(q, &req, bio)) { |
1665 | case ELEVATOR_BACK_MERGE: | |
1666 | if (!bio_attempt_back_merge(q, req, bio)) | |
1667 | break; | |
1668 | elv_bio_merged(q, req, bio); | |
1669 | free = attempt_back_merge(q, req); | |
1670 | if (free) | |
1671 | __blk_put_request(q, free); | |
1672 | else | |
1673 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1674 | goto out_unlock; | |
1675 | case ELEVATOR_FRONT_MERGE: | |
1676 | if (!bio_attempt_front_merge(q, req, bio)) | |
1677 | break; | |
1678 | elv_bio_merged(q, req, bio); | |
1679 | free = attempt_front_merge(q, req); | |
1680 | if (free) | |
1681 | __blk_put_request(q, free); | |
1682 | else | |
1683 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1684 | goto out_unlock; | |
1685 | default: | |
1686 | break; | |
1da177e4 LT |
1687 | } |
1688 | ||
450991bc | 1689 | get_rq: |
87760e5e JA |
1690 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); |
1691 | ||
1da177e4 | 1692 | /* |
450991bc | 1693 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1694 | * Returns with the queue unlocked. |
450991bc | 1695 | */ |
ef295ecf | 1696 | req = get_request(q, bio->bi_opf, bio, GFP_NOIO); |
a492f075 | 1697 | if (IS_ERR(req)) { |
87760e5e | 1698 | __wbt_done(q->rq_wb, wb_acct); |
4246a0b6 CH |
1699 | bio->bi_error = PTR_ERR(req); |
1700 | bio_endio(bio); | |
da8303c6 TH |
1701 | goto out_unlock; |
1702 | } | |
d6344532 | 1703 | |
87760e5e JA |
1704 | wbt_track(&req->issue_stat, wb_acct); |
1705 | ||
450991bc NP |
1706 | /* |
1707 | * After dropping the lock and possibly sleeping here, our request | |
1708 | * may now be mergeable after it had proven unmergeable (above). | |
1709 | * We don't worry about that case for efficiency. It won't happen | |
1710 | * often, and the elevators are able to handle it. | |
1da177e4 | 1711 | */ |
52d9e675 | 1712 | init_request_from_bio(req, bio); |
1da177e4 | 1713 | |
9562ad9a | 1714 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1715 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1716 | |
1717 | plug = current->plug; | |
721a9602 | 1718 | if (plug) { |
dc6d36c9 JA |
1719 | /* |
1720 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1721 | * of a plug trace. |
0a6219a9 ML |
1722 | * |
1723 | * @request_count may become stale because of schedule | |
1724 | * out, so check plug list again. | |
dc6d36c9 | 1725 | */ |
0a6219a9 | 1726 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 1727 | trace_block_plug(q); |
3540d5e8 | 1728 | else { |
50d24c34 SL |
1729 | struct request *last = list_entry_rq(plug->list.prev); |
1730 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1731 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 1732 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1733 | trace_block_plug(q); |
1734 | } | |
73c10101 | 1735 | } |
73c10101 | 1736 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 1737 | blk_account_io_start(req, true); |
73c10101 JA |
1738 | } else { |
1739 | spin_lock_irq(q->queue_lock); | |
1740 | add_acct_request(q, req, where); | |
24ecfbe2 | 1741 | __blk_run_queue(q); |
73c10101 JA |
1742 | out_unlock: |
1743 | spin_unlock_irq(q->queue_lock); | |
1744 | } | |
dece1635 JA |
1745 | |
1746 | return BLK_QC_T_NONE; | |
1da177e4 LT |
1747 | } |
1748 | ||
1749 | /* | |
1750 | * If bio->bi_dev is a partition, remap the location | |
1751 | */ | |
1752 | static inline void blk_partition_remap(struct bio *bio) | |
1753 | { | |
1754 | struct block_device *bdev = bio->bi_bdev; | |
1755 | ||
778889d8 ST |
1756 | /* |
1757 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
1758 | * Include a test for the reset op code and perform the remap if needed. | |
1759 | */ | |
1760 | if (bdev != bdev->bd_contains && | |
1761 | (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) { | |
1da177e4 LT |
1762 | struct hd_struct *p = bdev->bd_part; |
1763 | ||
4f024f37 | 1764 | bio->bi_iter.bi_sector += p->start_sect; |
1da177e4 | 1765 | bio->bi_bdev = bdev->bd_contains; |
c7149d6b | 1766 | |
d07335e5 MS |
1767 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1768 | bdev->bd_dev, | |
4f024f37 | 1769 | bio->bi_iter.bi_sector - p->start_sect); |
1da177e4 LT |
1770 | } |
1771 | } | |
1772 | ||
1da177e4 LT |
1773 | static void handle_bad_sector(struct bio *bio) |
1774 | { | |
1775 | char b[BDEVNAME_SIZE]; | |
1776 | ||
1777 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 1778 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
1da177e4 | 1779 | bdevname(bio->bi_bdev, b), |
1eff9d32 | 1780 | bio->bi_opf, |
f73a1c7d | 1781 | (unsigned long long)bio_end_sector(bio), |
77304d2a | 1782 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1783 | } |
1784 | ||
c17bb495 AM |
1785 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1786 | ||
1787 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1788 | ||
1789 | static int __init setup_fail_make_request(char *str) | |
1790 | { | |
1791 | return setup_fault_attr(&fail_make_request, str); | |
1792 | } | |
1793 | __setup("fail_make_request=", setup_fail_make_request); | |
1794 | ||
b2c9cd37 | 1795 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1796 | { |
b2c9cd37 | 1797 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1798 | } |
1799 | ||
1800 | static int __init fail_make_request_debugfs(void) | |
1801 | { | |
dd48c085 AM |
1802 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1803 | NULL, &fail_make_request); | |
1804 | ||
21f9fcd8 | 1805 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
1806 | } |
1807 | ||
1808 | late_initcall(fail_make_request_debugfs); | |
1809 | ||
1810 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1811 | ||
b2c9cd37 AM |
1812 | static inline bool should_fail_request(struct hd_struct *part, |
1813 | unsigned int bytes) | |
c17bb495 | 1814 | { |
b2c9cd37 | 1815 | return false; |
c17bb495 AM |
1816 | } |
1817 | ||
1818 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1819 | ||
c07e2b41 JA |
1820 | /* |
1821 | * Check whether this bio extends beyond the end of the device. | |
1822 | */ | |
1823 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1824 | { | |
1825 | sector_t maxsector; | |
1826 | ||
1827 | if (!nr_sectors) | |
1828 | return 0; | |
1829 | ||
1830 | /* Test device or partition size, when known. */ | |
77304d2a | 1831 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 | 1832 | if (maxsector) { |
4f024f37 | 1833 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
1834 | |
1835 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1836 | /* | |
1837 | * This may well happen - the kernel calls bread() | |
1838 | * without checking the size of the device, e.g., when | |
1839 | * mounting a device. | |
1840 | */ | |
1841 | handle_bad_sector(bio); | |
1842 | return 1; | |
1843 | } | |
1844 | } | |
1845 | ||
1846 | return 0; | |
1847 | } | |
1848 | ||
27a84d54 CH |
1849 | static noinline_for_stack bool |
1850 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1851 | { |
165125e1 | 1852 | struct request_queue *q; |
5a7bbad2 | 1853 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1854 | int err = -EIO; |
5a7bbad2 CH |
1855 | char b[BDEVNAME_SIZE]; |
1856 | struct hd_struct *part; | |
1da177e4 LT |
1857 | |
1858 | might_sleep(); | |
1da177e4 | 1859 | |
c07e2b41 JA |
1860 | if (bio_check_eod(bio, nr_sectors)) |
1861 | goto end_io; | |
1da177e4 | 1862 | |
5a7bbad2 CH |
1863 | q = bdev_get_queue(bio->bi_bdev); |
1864 | if (unlikely(!q)) { | |
1865 | printk(KERN_ERR | |
1866 | "generic_make_request: Trying to access " | |
1867 | "nonexistent block-device %s (%Lu)\n", | |
1868 | bdevname(bio->bi_bdev, b), | |
4f024f37 | 1869 | (long long) bio->bi_iter.bi_sector); |
5a7bbad2 CH |
1870 | goto end_io; |
1871 | } | |
c17bb495 | 1872 | |
5a7bbad2 | 1873 | part = bio->bi_bdev->bd_part; |
4f024f37 | 1874 | if (should_fail_request(part, bio->bi_iter.bi_size) || |
5a7bbad2 | 1875 | should_fail_request(&part_to_disk(part)->part0, |
4f024f37 | 1876 | bio->bi_iter.bi_size)) |
5a7bbad2 | 1877 | goto end_io; |
2056a782 | 1878 | |
5a7bbad2 CH |
1879 | /* |
1880 | * If this device has partitions, remap block n | |
1881 | * of partition p to block n+start(p) of the disk. | |
1882 | */ | |
1883 | blk_partition_remap(bio); | |
2056a782 | 1884 | |
5a7bbad2 CH |
1885 | if (bio_check_eod(bio, nr_sectors)) |
1886 | goto end_io; | |
1e87901e | 1887 | |
5a7bbad2 CH |
1888 | /* |
1889 | * Filter flush bio's early so that make_request based | |
1890 | * drivers without flush support don't have to worry | |
1891 | * about them. | |
1892 | */ | |
f3a8ab7d | 1893 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 1894 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 1895 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 CH |
1896 | if (!nr_sectors) { |
1897 | err = 0; | |
51fd77bd JA |
1898 | goto end_io; |
1899 | } | |
5a7bbad2 | 1900 | } |
5ddfe969 | 1901 | |
288dab8a CH |
1902 | switch (bio_op(bio)) { |
1903 | case REQ_OP_DISCARD: | |
1904 | if (!blk_queue_discard(q)) | |
1905 | goto not_supported; | |
1906 | break; | |
1907 | case REQ_OP_SECURE_ERASE: | |
1908 | if (!blk_queue_secure_erase(q)) | |
1909 | goto not_supported; | |
1910 | break; | |
1911 | case REQ_OP_WRITE_SAME: | |
1912 | if (!bdev_write_same(bio->bi_bdev)) | |
1913 | goto not_supported; | |
58886785 | 1914 | break; |
2d253440 ST |
1915 | case REQ_OP_ZONE_REPORT: |
1916 | case REQ_OP_ZONE_RESET: | |
1917 | if (!bdev_is_zoned(bio->bi_bdev)) | |
1918 | goto not_supported; | |
288dab8a | 1919 | break; |
a6f0788e CK |
1920 | case REQ_OP_WRITE_ZEROES: |
1921 | if (!bdev_write_zeroes_sectors(bio->bi_bdev)) | |
1922 | goto not_supported; | |
1923 | break; | |
288dab8a CH |
1924 | default: |
1925 | break; | |
5a7bbad2 | 1926 | } |
01edede4 | 1927 | |
7f4b35d1 TH |
1928 | /* |
1929 | * Various block parts want %current->io_context and lazy ioc | |
1930 | * allocation ends up trading a lot of pain for a small amount of | |
1931 | * memory. Just allocate it upfront. This may fail and block | |
1932 | * layer knows how to live with it. | |
1933 | */ | |
1934 | create_io_context(GFP_ATOMIC, q->node); | |
1935 | ||
ae118896 TH |
1936 | if (!blkcg_bio_issue_check(q, bio)) |
1937 | return false; | |
27a84d54 | 1938 | |
5a7bbad2 | 1939 | trace_block_bio_queue(q, bio); |
27a84d54 | 1940 | return true; |
a7384677 | 1941 | |
288dab8a CH |
1942 | not_supported: |
1943 | err = -EOPNOTSUPP; | |
a7384677 | 1944 | end_io: |
4246a0b6 CH |
1945 | bio->bi_error = err; |
1946 | bio_endio(bio); | |
27a84d54 | 1947 | return false; |
1da177e4 LT |
1948 | } |
1949 | ||
27a84d54 CH |
1950 | /** |
1951 | * generic_make_request - hand a buffer to its device driver for I/O | |
1952 | * @bio: The bio describing the location in memory and on the device. | |
1953 | * | |
1954 | * generic_make_request() is used to make I/O requests of block | |
1955 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1956 | * to be done. | |
1957 | * | |
1958 | * generic_make_request() does not return any status. The | |
1959 | * success/failure status of the request, along with notification of | |
1960 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1961 | * function described (one day) else where. | |
1962 | * | |
1963 | * The caller of generic_make_request must make sure that bi_io_vec | |
1964 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1965 | * set to describe the device address, and the | |
1966 | * bi_end_io and optionally bi_private are set to describe how | |
1967 | * completion notification should be signaled. | |
1968 | * | |
1969 | * generic_make_request and the drivers it calls may use bi_next if this | |
1970 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1971 | * a lower device by calling into generic_make_request recursively, which | |
1972 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 1973 | */ |
dece1635 | 1974 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 1975 | { |
f5fe1b51 N |
1976 | /* |
1977 | * bio_list_on_stack[0] contains bios submitted by the current | |
1978 | * make_request_fn. | |
1979 | * bio_list_on_stack[1] contains bios that were submitted before | |
1980 | * the current make_request_fn, but that haven't been processed | |
1981 | * yet. | |
1982 | */ | |
1983 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 1984 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 1985 | |
27a84d54 | 1986 | if (!generic_make_request_checks(bio)) |
dece1635 | 1987 | goto out; |
27a84d54 CH |
1988 | |
1989 | /* | |
1990 | * We only want one ->make_request_fn to be active at a time, else | |
1991 | * stack usage with stacked devices could be a problem. So use | |
1992 | * current->bio_list to keep a list of requests submited by a | |
1993 | * make_request_fn function. current->bio_list is also used as a | |
1994 | * flag to say if generic_make_request is currently active in this | |
1995 | * task or not. If it is NULL, then no make_request is active. If | |
1996 | * it is non-NULL, then a make_request is active, and new requests | |
1997 | * should be added at the tail | |
1998 | */ | |
bddd87c7 | 1999 | if (current->bio_list) { |
f5fe1b51 | 2000 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 2001 | goto out; |
d89d8796 | 2002 | } |
27a84d54 | 2003 | |
d89d8796 NB |
2004 | /* following loop may be a bit non-obvious, and so deserves some |
2005 | * explanation. | |
2006 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2007 | * ensure that) so we have a list with a single bio. | |
2008 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2009 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2010 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2011 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2012 | * through a recursive call to generic_make_request. If it |
2013 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2014 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2015 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2016 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2017 | */ |
2018 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
2019 | bio_list_init(&bio_list_on_stack[0]); |
2020 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 2021 | do { |
27a84d54 CH |
2022 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
2023 | ||
6f3b0e8b | 2024 | if (likely(blk_queue_enter(q, false) == 0)) { |
79bd9959 N |
2025 | struct bio_list lower, same; |
2026 | ||
2027 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
2028 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
2029 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 2030 | ret = q->make_request_fn(q, bio); |
3ef28e83 DW |
2031 | |
2032 | blk_queue_exit(q); | |
27a84d54 | 2033 | |
79bd9959 N |
2034 | /* sort new bios into those for a lower level |
2035 | * and those for the same level | |
2036 | */ | |
2037 | bio_list_init(&lower); | |
2038 | bio_list_init(&same); | |
f5fe1b51 | 2039 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
79bd9959 N |
2040 | if (q == bdev_get_queue(bio->bi_bdev)) |
2041 | bio_list_add(&same, bio); | |
2042 | else | |
2043 | bio_list_add(&lower, bio); | |
2044 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
2045 | bio_list_merge(&bio_list_on_stack[0], &lower); |
2046 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2047 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 2048 | } else { |
3ef28e83 | 2049 | bio_io_error(bio); |
3ef28e83 | 2050 | } |
f5fe1b51 | 2051 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 2052 | } while (bio); |
bddd87c7 | 2053 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2054 | |
2055 | out: | |
2056 | return ret; | |
d89d8796 | 2057 | } |
1da177e4 LT |
2058 | EXPORT_SYMBOL(generic_make_request); |
2059 | ||
2060 | /** | |
710027a4 | 2061 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2062 | * @bio: The &struct bio which describes the I/O |
2063 | * | |
2064 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2065 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2066 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2067 | * |
2068 | */ | |
4e49ea4a | 2069 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2070 | { |
bf2de6f5 JA |
2071 | /* |
2072 | * If it's a regular read/write or a barrier with data attached, | |
2073 | * go through the normal accounting stuff before submission. | |
2074 | */ | |
e2a60da7 | 2075 | if (bio_has_data(bio)) { |
4363ac7c MP |
2076 | unsigned int count; |
2077 | ||
95fe6c1a | 2078 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
4363ac7c MP |
2079 | count = bdev_logical_block_size(bio->bi_bdev) >> 9; |
2080 | else | |
2081 | count = bio_sectors(bio); | |
2082 | ||
a8ebb056 | 2083 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2084 | count_vm_events(PGPGOUT, count); |
2085 | } else { | |
4f024f37 | 2086 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2087 | count_vm_events(PGPGIN, count); |
2088 | } | |
2089 | ||
2090 | if (unlikely(block_dump)) { | |
2091 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2092 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2093 | current->comm, task_pid_nr(current), |
a8ebb056 | 2094 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2095 | (unsigned long long)bio->bi_iter.bi_sector, |
8dcbdc74 SM |
2096 | bdevname(bio->bi_bdev, b), |
2097 | count); | |
bf2de6f5 | 2098 | } |
1da177e4 LT |
2099 | } |
2100 | ||
dece1635 | 2101 | return generic_make_request(bio); |
1da177e4 | 2102 | } |
1da177e4 LT |
2103 | EXPORT_SYMBOL(submit_bio); |
2104 | ||
82124d60 | 2105 | /** |
bf4e6b4e HR |
2106 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2107 | * for new the queue limits | |
82124d60 KU |
2108 | * @q: the queue |
2109 | * @rq: the request being checked | |
2110 | * | |
2111 | * Description: | |
2112 | * @rq may have been made based on weaker limitations of upper-level queues | |
2113 | * in request stacking drivers, and it may violate the limitation of @q. | |
2114 | * Since the block layer and the underlying device driver trust @rq | |
2115 | * after it is inserted to @q, it should be checked against @q before | |
2116 | * the insertion using this generic function. | |
2117 | * | |
82124d60 | 2118 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2119 | * limits when retrying requests on other queues. Those requests need |
2120 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2121 | */ |
bf4e6b4e HR |
2122 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2123 | struct request *rq) | |
82124d60 | 2124 | { |
8fe0d473 | 2125 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2126 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2127 | return -EIO; | |
2128 | } | |
2129 | ||
2130 | /* | |
2131 | * queue's settings related to segment counting like q->bounce_pfn | |
2132 | * may differ from that of other stacking queues. | |
2133 | * Recalculate it to check the request correctly on this queue's | |
2134 | * limitation. | |
2135 | */ | |
2136 | blk_recalc_rq_segments(rq); | |
8a78362c | 2137 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2138 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2139 | return -EIO; | |
2140 | } | |
2141 | ||
2142 | return 0; | |
2143 | } | |
82124d60 KU |
2144 | |
2145 | /** | |
2146 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2147 | * @q: the queue to submit the request | |
2148 | * @rq: the request being queued | |
2149 | */ | |
2150 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
2151 | { | |
2152 | unsigned long flags; | |
4853abaa | 2153 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2154 | |
bf4e6b4e | 2155 | if (blk_cloned_rq_check_limits(q, rq)) |
82124d60 KU |
2156 | return -EIO; |
2157 | ||
b2c9cd37 AM |
2158 | if (rq->rq_disk && |
2159 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 2160 | return -EIO; |
82124d60 | 2161 | |
7fb4898e KB |
2162 | if (q->mq_ops) { |
2163 | if (blk_queue_io_stat(q)) | |
2164 | blk_account_io_start(rq, true); | |
bd6737f1 | 2165 | blk_mq_sched_insert_request(rq, false, true, false, false); |
7fb4898e KB |
2166 | return 0; |
2167 | } | |
2168 | ||
82124d60 | 2169 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2170 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
2171 | spin_unlock_irqrestore(q->queue_lock, flags); |
2172 | return -ENODEV; | |
2173 | } | |
82124d60 KU |
2174 | |
2175 | /* | |
2176 | * Submitting request must be dequeued before calling this function | |
2177 | * because it will be linked to another request_queue | |
2178 | */ | |
2179 | BUG_ON(blk_queued_rq(rq)); | |
2180 | ||
f73f44eb | 2181 | if (op_is_flush(rq->cmd_flags)) |
4853abaa JM |
2182 | where = ELEVATOR_INSERT_FLUSH; |
2183 | ||
2184 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2185 | if (where == ELEVATOR_INSERT_FLUSH) |
2186 | __blk_run_queue(q); | |
82124d60 KU |
2187 | spin_unlock_irqrestore(q->queue_lock, flags); |
2188 | ||
2189 | return 0; | |
2190 | } | |
2191 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2192 | ||
80a761fd TH |
2193 | /** |
2194 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2195 | * @rq: request to examine | |
2196 | * | |
2197 | * Description: | |
2198 | * A request could be merge of IOs which require different failure | |
2199 | * handling. This function determines the number of bytes which | |
2200 | * can be failed from the beginning of the request without | |
2201 | * crossing into area which need to be retried further. | |
2202 | * | |
2203 | * Return: | |
2204 | * The number of bytes to fail. | |
2205 | * | |
2206 | * Context: | |
2207 | * queue_lock must be held. | |
2208 | */ | |
2209 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2210 | { | |
2211 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2212 | unsigned int bytes = 0; | |
2213 | struct bio *bio; | |
2214 | ||
e8064021 | 2215 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2216 | return blk_rq_bytes(rq); |
2217 | ||
2218 | /* | |
2219 | * Currently the only 'mixing' which can happen is between | |
2220 | * different fastfail types. We can safely fail portions | |
2221 | * which have all the failfast bits that the first one has - | |
2222 | * the ones which are at least as eager to fail as the first | |
2223 | * one. | |
2224 | */ | |
2225 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2226 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2227 | break; |
4f024f37 | 2228 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2229 | } |
2230 | ||
2231 | /* this could lead to infinite loop */ | |
2232 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2233 | return bytes; | |
2234 | } | |
2235 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2236 | ||
320ae51f | 2237 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2238 | { |
c2553b58 | 2239 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2240 | const int rw = rq_data_dir(req); |
2241 | struct hd_struct *part; | |
2242 | int cpu; | |
2243 | ||
2244 | cpu = part_stat_lock(); | |
09e099d4 | 2245 | part = req->part; |
bc58ba94 JA |
2246 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2247 | part_stat_unlock(); | |
2248 | } | |
2249 | } | |
2250 | ||
320ae51f | 2251 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2252 | { |
bc58ba94 | 2253 | /* |
dd4c133f TH |
2254 | * Account IO completion. flush_rq isn't accounted as a |
2255 | * normal IO on queueing nor completion. Accounting the | |
2256 | * containing request is enough. | |
bc58ba94 | 2257 | */ |
e8064021 | 2258 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
bc58ba94 JA |
2259 | unsigned long duration = jiffies - req->start_time; |
2260 | const int rw = rq_data_dir(req); | |
2261 | struct hd_struct *part; | |
2262 | int cpu; | |
2263 | ||
2264 | cpu = part_stat_lock(); | |
09e099d4 | 2265 | part = req->part; |
bc58ba94 JA |
2266 | |
2267 | part_stat_inc(cpu, part, ios[rw]); | |
2268 | part_stat_add(cpu, part, ticks[rw], duration); | |
2269 | part_round_stats(cpu, part); | |
316d315b | 2270 | part_dec_in_flight(part, rw); |
bc58ba94 | 2271 | |
6c23a968 | 2272 | hd_struct_put(part); |
bc58ba94 JA |
2273 | part_stat_unlock(); |
2274 | } | |
2275 | } | |
2276 | ||
47fafbc7 | 2277 | #ifdef CONFIG_PM |
c8158819 LM |
2278 | /* |
2279 | * Don't process normal requests when queue is suspended | |
2280 | * or in the process of suspending/resuming | |
2281 | */ | |
2282 | static struct request *blk_pm_peek_request(struct request_queue *q, | |
2283 | struct request *rq) | |
2284 | { | |
2285 | if (q->dev && (q->rpm_status == RPM_SUSPENDED || | |
e8064021 | 2286 | (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM)))) |
c8158819 LM |
2287 | return NULL; |
2288 | else | |
2289 | return rq; | |
2290 | } | |
2291 | #else | |
2292 | static inline struct request *blk_pm_peek_request(struct request_queue *q, | |
2293 | struct request *rq) | |
2294 | { | |
2295 | return rq; | |
2296 | } | |
2297 | #endif | |
2298 | ||
320ae51f JA |
2299 | void blk_account_io_start(struct request *rq, bool new_io) |
2300 | { | |
2301 | struct hd_struct *part; | |
2302 | int rw = rq_data_dir(rq); | |
2303 | int cpu; | |
2304 | ||
2305 | if (!blk_do_io_stat(rq)) | |
2306 | return; | |
2307 | ||
2308 | cpu = part_stat_lock(); | |
2309 | ||
2310 | if (!new_io) { | |
2311 | part = rq->part; | |
2312 | part_stat_inc(cpu, part, merges[rw]); | |
2313 | } else { | |
2314 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2315 | if (!hd_struct_try_get(part)) { | |
2316 | /* | |
2317 | * The partition is already being removed, | |
2318 | * the request will be accounted on the disk only | |
2319 | * | |
2320 | * We take a reference on disk->part0 although that | |
2321 | * partition will never be deleted, so we can treat | |
2322 | * it as any other partition. | |
2323 | */ | |
2324 | part = &rq->rq_disk->part0; | |
2325 | hd_struct_get(part); | |
2326 | } | |
2327 | part_round_stats(cpu, part); | |
2328 | part_inc_in_flight(part, rw); | |
2329 | rq->part = part; | |
2330 | } | |
2331 | ||
2332 | part_stat_unlock(); | |
2333 | } | |
2334 | ||
3bcddeac | 2335 | /** |
9934c8c0 TH |
2336 | * blk_peek_request - peek at the top of a request queue |
2337 | * @q: request queue to peek at | |
2338 | * | |
2339 | * Description: | |
2340 | * Return the request at the top of @q. The returned request | |
2341 | * should be started using blk_start_request() before LLD starts | |
2342 | * processing it. | |
2343 | * | |
2344 | * Return: | |
2345 | * Pointer to the request at the top of @q if available. Null | |
2346 | * otherwise. | |
2347 | * | |
2348 | * Context: | |
2349 | * queue_lock must be held. | |
2350 | */ | |
2351 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2352 | { |
2353 | struct request *rq; | |
2354 | int ret; | |
2355 | ||
2356 | while ((rq = __elv_next_request(q)) != NULL) { | |
c8158819 LM |
2357 | |
2358 | rq = blk_pm_peek_request(q, rq); | |
2359 | if (!rq) | |
2360 | break; | |
2361 | ||
e8064021 | 2362 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2363 | /* |
2364 | * This is the first time the device driver | |
2365 | * sees this request (possibly after | |
2366 | * requeueing). Notify IO scheduler. | |
2367 | */ | |
e8064021 | 2368 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2369 | elv_activate_rq(q, rq); |
2370 | ||
2371 | /* | |
2372 | * just mark as started even if we don't start | |
2373 | * it, a request that has been delayed should | |
2374 | * not be passed by new incoming requests | |
2375 | */ | |
e8064021 | 2376 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2377 | trace_block_rq_issue(q, rq); |
2378 | } | |
2379 | ||
2380 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2381 | q->end_sector = rq_end_sector(rq); | |
2382 | q->boundary_rq = NULL; | |
2383 | } | |
2384 | ||
e8064021 | 2385 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2386 | break; |
2387 | ||
2e46e8b2 | 2388 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2389 | /* |
2390 | * make sure space for the drain appears we | |
2391 | * know we can do this because max_hw_segments | |
2392 | * has been adjusted to be one fewer than the | |
2393 | * device can handle | |
2394 | */ | |
2395 | rq->nr_phys_segments++; | |
2396 | } | |
2397 | ||
2398 | if (!q->prep_rq_fn) | |
2399 | break; | |
2400 | ||
2401 | ret = q->prep_rq_fn(q, rq); | |
2402 | if (ret == BLKPREP_OK) { | |
2403 | break; | |
2404 | } else if (ret == BLKPREP_DEFER) { | |
2405 | /* | |
2406 | * the request may have been (partially) prepped. | |
2407 | * we need to keep this request in the front to | |
e8064021 | 2408 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2409 | * prevent other fs requests from passing this one. |
2410 | */ | |
2e46e8b2 | 2411 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2412 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2413 | /* |
2414 | * remove the space for the drain we added | |
2415 | * so that we don't add it again | |
2416 | */ | |
2417 | --rq->nr_phys_segments; | |
2418 | } | |
2419 | ||
2420 | rq = NULL; | |
2421 | break; | |
0fb5b1fb MP |
2422 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
2423 | int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO; | |
2424 | ||
e8064021 | 2425 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2426 | /* |
2427 | * Mark this request as started so we don't trigger | |
2428 | * any debug logic in the end I/O path. | |
2429 | */ | |
2430 | blk_start_request(rq); | |
0fb5b1fb | 2431 | __blk_end_request_all(rq, err); |
158dbda0 TH |
2432 | } else { |
2433 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2434 | break; | |
2435 | } | |
2436 | } | |
2437 | ||
2438 | return rq; | |
2439 | } | |
9934c8c0 | 2440 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2441 | |
9934c8c0 | 2442 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2443 | { |
9934c8c0 TH |
2444 | struct request_queue *q = rq->q; |
2445 | ||
158dbda0 TH |
2446 | BUG_ON(list_empty(&rq->queuelist)); |
2447 | BUG_ON(ELV_ON_HASH(rq)); | |
2448 | ||
2449 | list_del_init(&rq->queuelist); | |
2450 | ||
2451 | /* | |
2452 | * the time frame between a request being removed from the lists | |
2453 | * and to it is freed is accounted as io that is in progress at | |
2454 | * the driver side. | |
2455 | */ | |
9195291e | 2456 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2457 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2458 | set_io_start_time_ns(rq); |
2459 | } | |
158dbda0 TH |
2460 | } |
2461 | ||
9934c8c0 TH |
2462 | /** |
2463 | * blk_start_request - start request processing on the driver | |
2464 | * @req: request to dequeue | |
2465 | * | |
2466 | * Description: | |
2467 | * Dequeue @req and start timeout timer on it. This hands off the | |
2468 | * request to the driver. | |
2469 | * | |
2470 | * Block internal functions which don't want to start timer should | |
2471 | * call blk_dequeue_request(). | |
2472 | * | |
2473 | * Context: | |
2474 | * queue_lock must be held. | |
2475 | */ | |
2476 | void blk_start_request(struct request *req) | |
2477 | { | |
2478 | blk_dequeue_request(req); | |
2479 | ||
cf43e6be JA |
2480 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
2481 | blk_stat_set_issue_time(&req->issue_stat); | |
2482 | req->rq_flags |= RQF_STATS; | |
87760e5e | 2483 | wbt_issue(req->q->rq_wb, &req->issue_stat); |
cf43e6be JA |
2484 | } |
2485 | ||
4912aa6c | 2486 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); |
9934c8c0 TH |
2487 | blk_add_timer(req); |
2488 | } | |
2489 | EXPORT_SYMBOL(blk_start_request); | |
2490 | ||
2491 | /** | |
2492 | * blk_fetch_request - fetch a request from a request queue | |
2493 | * @q: request queue to fetch a request from | |
2494 | * | |
2495 | * Description: | |
2496 | * Return the request at the top of @q. The request is started on | |
2497 | * return and LLD can start processing it immediately. | |
2498 | * | |
2499 | * Return: | |
2500 | * Pointer to the request at the top of @q if available. Null | |
2501 | * otherwise. | |
2502 | * | |
2503 | * Context: | |
2504 | * queue_lock must be held. | |
2505 | */ | |
2506 | struct request *blk_fetch_request(struct request_queue *q) | |
2507 | { | |
2508 | struct request *rq; | |
2509 | ||
2510 | rq = blk_peek_request(q); | |
2511 | if (rq) | |
2512 | blk_start_request(rq); | |
2513 | return rq; | |
2514 | } | |
2515 | EXPORT_SYMBOL(blk_fetch_request); | |
2516 | ||
3bcddeac | 2517 | /** |
2e60e022 | 2518 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2519 | * @req: the request being processed |
710027a4 | 2520 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2521 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2522 | * |
2523 | * Description: | |
8ebf9756 RD |
2524 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2525 | * the request structure even if @req doesn't have leftover. | |
2526 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2527 | * |
2528 | * This special helper function is only for request stacking drivers | |
2529 | * (e.g. request-based dm) so that they can handle partial completion. | |
2530 | * Actual device drivers should use blk_end_request instead. | |
2531 | * | |
2532 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2533 | * %false return from this function. | |
3bcddeac KU |
2534 | * |
2535 | * Return: | |
2e60e022 TH |
2536 | * %false - this request doesn't have any more data |
2537 | * %true - this request has more data | |
3bcddeac | 2538 | **/ |
2e60e022 | 2539 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2540 | { |
f79ea416 | 2541 | int total_bytes; |
1da177e4 | 2542 | |
4a0efdc9 HR |
2543 | trace_block_rq_complete(req->q, req, nr_bytes); |
2544 | ||
2e60e022 TH |
2545 | if (!req->bio) |
2546 | return false; | |
2547 | ||
1da177e4 | 2548 | /* |
6f41469c TH |
2549 | * For fs requests, rq is just carrier of independent bio's |
2550 | * and each partial completion should be handled separately. | |
2551 | * Reset per-request error on each partial completion. | |
2552 | * | |
2553 | * TODO: tj: This is too subtle. It would be better to let | |
2554 | * low level drivers do what they see fit. | |
1da177e4 | 2555 | */ |
57292b58 | 2556 | if (!blk_rq_is_passthrough(req)) |
1da177e4 LT |
2557 | req->errors = 0; |
2558 | ||
57292b58 | 2559 | if (error && !blk_rq_is_passthrough(req) && |
e8064021 | 2560 | !(req->rq_flags & RQF_QUIET)) { |
79775567 HR |
2561 | char *error_type; |
2562 | ||
2563 | switch (error) { | |
2564 | case -ENOLINK: | |
2565 | error_type = "recoverable transport"; | |
2566 | break; | |
2567 | case -EREMOTEIO: | |
2568 | error_type = "critical target"; | |
2569 | break; | |
2570 | case -EBADE: | |
2571 | error_type = "critical nexus"; | |
2572 | break; | |
d1ffc1f8 HR |
2573 | case -ETIMEDOUT: |
2574 | error_type = "timeout"; | |
2575 | break; | |
a9d6ceb8 HR |
2576 | case -ENOSPC: |
2577 | error_type = "critical space allocation"; | |
2578 | break; | |
7e782af5 HR |
2579 | case -ENODATA: |
2580 | error_type = "critical medium"; | |
2581 | break; | |
79775567 HR |
2582 | case -EIO: |
2583 | default: | |
2584 | error_type = "I/O"; | |
2585 | break; | |
2586 | } | |
ef3ecb66 RE |
2587 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", |
2588 | __func__, error_type, req->rq_disk ? | |
37d7b34f YZ |
2589 | req->rq_disk->disk_name : "?", |
2590 | (unsigned long long)blk_rq_pos(req)); | |
2591 | ||
1da177e4 LT |
2592 | } |
2593 | ||
bc58ba94 | 2594 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2595 | |
f79ea416 KO |
2596 | total_bytes = 0; |
2597 | while (req->bio) { | |
2598 | struct bio *bio = req->bio; | |
4f024f37 | 2599 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2600 | |
4f024f37 | 2601 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2602 | req->bio = bio->bi_next; |
1da177e4 | 2603 | |
f79ea416 | 2604 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2605 | |
f79ea416 KO |
2606 | total_bytes += bio_bytes; |
2607 | nr_bytes -= bio_bytes; | |
1da177e4 | 2608 | |
f79ea416 KO |
2609 | if (!nr_bytes) |
2610 | break; | |
1da177e4 LT |
2611 | } |
2612 | ||
2613 | /* | |
2614 | * completely done | |
2615 | */ | |
2e60e022 TH |
2616 | if (!req->bio) { |
2617 | /* | |
2618 | * Reset counters so that the request stacking driver | |
2619 | * can find how many bytes remain in the request | |
2620 | * later. | |
2621 | */ | |
a2dec7b3 | 2622 | req->__data_len = 0; |
2e60e022 TH |
2623 | return false; |
2624 | } | |
1da177e4 | 2625 | |
f9d03f96 CH |
2626 | WARN_ON_ONCE(req->rq_flags & RQF_SPECIAL_PAYLOAD); |
2627 | ||
a2dec7b3 | 2628 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2629 | |
2630 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 2631 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 2632 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2633 | |
80a761fd | 2634 | /* mixed attributes always follow the first bio */ |
e8064021 | 2635 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 2636 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 2637 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
2638 | } |
2639 | ||
2e46e8b2 TH |
2640 | /* |
2641 | * If total number of sectors is less than the first segment | |
2642 | * size, something has gone terribly wrong. | |
2643 | */ | |
2644 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2645 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2646 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2647 | } |
2648 | ||
2649 | /* recalculate the number of segments */ | |
1da177e4 | 2650 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2651 | |
2e60e022 | 2652 | return true; |
1da177e4 | 2653 | } |
2e60e022 | 2654 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2655 | |
2e60e022 TH |
2656 | static bool blk_update_bidi_request(struct request *rq, int error, |
2657 | unsigned int nr_bytes, | |
2658 | unsigned int bidi_bytes) | |
5efccd17 | 2659 | { |
2e60e022 TH |
2660 | if (blk_update_request(rq, error, nr_bytes)) |
2661 | return true; | |
5efccd17 | 2662 | |
2e60e022 TH |
2663 | /* Bidi request must be completed as a whole */ |
2664 | if (unlikely(blk_bidi_rq(rq)) && | |
2665 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2666 | return true; | |
5efccd17 | 2667 | |
e2e1a148 JA |
2668 | if (blk_queue_add_random(rq->q)) |
2669 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2670 | |
2671 | return false; | |
1da177e4 LT |
2672 | } |
2673 | ||
28018c24 JB |
2674 | /** |
2675 | * blk_unprep_request - unprepare a request | |
2676 | * @req: the request | |
2677 | * | |
2678 | * This function makes a request ready for complete resubmission (or | |
2679 | * completion). It happens only after all error handling is complete, | |
2680 | * so represents the appropriate moment to deallocate any resources | |
2681 | * that were allocated to the request in the prep_rq_fn. The queue | |
2682 | * lock is held when calling this. | |
2683 | */ | |
2684 | void blk_unprep_request(struct request *req) | |
2685 | { | |
2686 | struct request_queue *q = req->q; | |
2687 | ||
e8064021 | 2688 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
2689 | if (q->unprep_rq_fn) |
2690 | q->unprep_rq_fn(q, req); | |
2691 | } | |
2692 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2693 | ||
1da177e4 LT |
2694 | /* |
2695 | * queue lock must be held | |
2696 | */ | |
12120077 | 2697 | void blk_finish_request(struct request *req, int error) |
1da177e4 | 2698 | { |
cf43e6be JA |
2699 | struct request_queue *q = req->q; |
2700 | ||
2701 | if (req->rq_flags & RQF_STATS) | |
2702 | blk_stat_add(&q->rq_stats[rq_data_dir(req)], req); | |
2703 | ||
e8064021 | 2704 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 2705 | blk_queue_end_tag(q, req); |
b8286239 | 2706 | |
ba396a6c | 2707 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2708 | |
57292b58 | 2709 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) |
dc3b17cc | 2710 | laptop_io_completion(req->q->backing_dev_info); |
1da177e4 | 2711 | |
e78042e5 MA |
2712 | blk_delete_timer(req); |
2713 | ||
e8064021 | 2714 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
2715 | blk_unprep_request(req); |
2716 | ||
bc58ba94 | 2717 | blk_account_io_done(req); |
b8286239 | 2718 | |
87760e5e JA |
2719 | if (req->end_io) { |
2720 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
8ffdc655 | 2721 | req->end_io(req, error); |
87760e5e | 2722 | } else { |
b8286239 KU |
2723 | if (blk_bidi_rq(req)) |
2724 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2725 | ||
cf43e6be | 2726 | __blk_put_request(q, req); |
b8286239 | 2727 | } |
1da177e4 | 2728 | } |
12120077 | 2729 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 2730 | |
3b11313a | 2731 | /** |
2e60e022 TH |
2732 | * blk_end_bidi_request - Complete a bidi request |
2733 | * @rq: the request to complete | |
2734 | * @error: %0 for success, < %0 for error | |
2735 | * @nr_bytes: number of bytes to complete @rq | |
2736 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2737 | * |
2738 | * Description: | |
e3a04fe3 | 2739 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2740 | * Drivers that supports bidi can safely call this member for any |
2741 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2742 | * just ignored. | |
336cdb40 KU |
2743 | * |
2744 | * Return: | |
2e60e022 TH |
2745 | * %false - we are done with this request |
2746 | * %true - still buffers pending for this request | |
a0cd1285 | 2747 | **/ |
b1f74493 | 2748 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2749 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2750 | { | |
336cdb40 | 2751 | struct request_queue *q = rq->q; |
2e60e022 | 2752 | unsigned long flags; |
32fab448 | 2753 | |
2e60e022 TH |
2754 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2755 | return true; | |
32fab448 | 2756 | |
336cdb40 | 2757 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2758 | blk_finish_request(rq, error); |
336cdb40 KU |
2759 | spin_unlock_irqrestore(q->queue_lock, flags); |
2760 | ||
2e60e022 | 2761 | return false; |
32fab448 KU |
2762 | } |
2763 | ||
336cdb40 | 2764 | /** |
2e60e022 TH |
2765 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2766 | * @rq: the request to complete | |
710027a4 | 2767 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2768 | * @nr_bytes: number of bytes to complete @rq |
2769 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2770 | * |
2771 | * Description: | |
2e60e022 TH |
2772 | * Identical to blk_end_bidi_request() except that queue lock is |
2773 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2774 | * |
2775 | * Return: | |
2e60e022 TH |
2776 | * %false - we are done with this request |
2777 | * %true - still buffers pending for this request | |
336cdb40 | 2778 | **/ |
4853abaa | 2779 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2780 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2781 | { |
2e60e022 TH |
2782 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2783 | return true; | |
336cdb40 | 2784 | |
2e60e022 | 2785 | blk_finish_request(rq, error); |
336cdb40 | 2786 | |
2e60e022 | 2787 | return false; |
336cdb40 | 2788 | } |
e19a3ab0 KU |
2789 | |
2790 | /** | |
2791 | * blk_end_request - Helper function for drivers to complete the request. | |
2792 | * @rq: the request being processed | |
710027a4 | 2793 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2794 | * @nr_bytes: number of bytes to complete |
2795 | * | |
2796 | * Description: | |
2797 | * Ends I/O on a number of bytes attached to @rq. | |
2798 | * If @rq has leftover, sets it up for the next range of segments. | |
2799 | * | |
2800 | * Return: | |
b1f74493 FT |
2801 | * %false - we are done with this request |
2802 | * %true - still buffers pending for this request | |
e19a3ab0 | 2803 | **/ |
b1f74493 | 2804 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2805 | { |
b1f74493 | 2806 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2807 | } |
56ad1740 | 2808 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2809 | |
2810 | /** | |
b1f74493 FT |
2811 | * blk_end_request_all - Helper function for drives to finish the request. |
2812 | * @rq: the request to finish | |
8ebf9756 | 2813 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2814 | * |
2815 | * Description: | |
b1f74493 FT |
2816 | * Completely finish @rq. |
2817 | */ | |
2818 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2819 | { |
b1f74493 FT |
2820 | bool pending; |
2821 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2822 | |
b1f74493 FT |
2823 | if (unlikely(blk_bidi_rq(rq))) |
2824 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2825 | |
b1f74493 FT |
2826 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2827 | BUG_ON(pending); | |
2828 | } | |
56ad1740 | 2829 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2830 | |
b1f74493 FT |
2831 | /** |
2832 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2833 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2834 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2835 | * |
2836 | * Description: | |
2837 | * Complete the current consecutively mapped chunk from @rq. | |
2838 | * | |
2839 | * Return: | |
2840 | * %false - we are done with this request | |
2841 | * %true - still buffers pending for this request | |
2842 | */ | |
2843 | bool blk_end_request_cur(struct request *rq, int error) | |
2844 | { | |
2845 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2846 | } |
56ad1740 | 2847 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2848 | |
80a761fd TH |
2849 | /** |
2850 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2851 | * @rq: the request to finish till the next failure boundary for | |
2852 | * @error: must be negative errno | |
2853 | * | |
2854 | * Description: | |
2855 | * Complete @rq till the next failure boundary. | |
2856 | * | |
2857 | * Return: | |
2858 | * %false - we are done with this request | |
2859 | * %true - still buffers pending for this request | |
2860 | */ | |
2861 | bool blk_end_request_err(struct request *rq, int error) | |
2862 | { | |
2863 | WARN_ON(error >= 0); | |
2864 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2865 | } | |
2866 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2867 | ||
e3a04fe3 | 2868 | /** |
b1f74493 FT |
2869 | * __blk_end_request - Helper function for drivers to complete the request. |
2870 | * @rq: the request being processed | |
2871 | * @error: %0 for success, < %0 for error | |
2872 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2873 | * |
2874 | * Description: | |
b1f74493 | 2875 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2876 | * |
2877 | * Return: | |
b1f74493 FT |
2878 | * %false - we are done with this request |
2879 | * %true - still buffers pending for this request | |
e3a04fe3 | 2880 | **/ |
b1f74493 | 2881 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2882 | { |
b1f74493 | 2883 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2884 | } |
56ad1740 | 2885 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2886 | |
32fab448 | 2887 | /** |
b1f74493 FT |
2888 | * __blk_end_request_all - Helper function for drives to finish the request. |
2889 | * @rq: the request to finish | |
8ebf9756 | 2890 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2891 | * |
2892 | * Description: | |
b1f74493 | 2893 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2894 | */ |
b1f74493 | 2895 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2896 | { |
b1f74493 FT |
2897 | bool pending; |
2898 | unsigned int bidi_bytes = 0; | |
2899 | ||
2900 | if (unlikely(blk_bidi_rq(rq))) | |
2901 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2902 | ||
2903 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2904 | BUG_ON(pending); | |
32fab448 | 2905 | } |
56ad1740 | 2906 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2907 | |
e19a3ab0 | 2908 | /** |
b1f74493 FT |
2909 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2910 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2911 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2912 | * |
2913 | * Description: | |
b1f74493 FT |
2914 | * Complete the current consecutively mapped chunk from @rq. Must |
2915 | * be called with queue lock held. | |
e19a3ab0 KU |
2916 | * |
2917 | * Return: | |
b1f74493 FT |
2918 | * %false - we are done with this request |
2919 | * %true - still buffers pending for this request | |
2920 | */ | |
2921 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2922 | { |
b1f74493 | 2923 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2924 | } |
56ad1740 | 2925 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2926 | |
80a761fd TH |
2927 | /** |
2928 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2929 | * @rq: the request to finish till the next failure boundary for | |
2930 | * @error: must be negative errno | |
2931 | * | |
2932 | * Description: | |
2933 | * Complete @rq till the next failure boundary. Must be called | |
2934 | * with queue lock held. | |
2935 | * | |
2936 | * Return: | |
2937 | * %false - we are done with this request | |
2938 | * %true - still buffers pending for this request | |
2939 | */ | |
2940 | bool __blk_end_request_err(struct request *rq, int error) | |
2941 | { | |
2942 | WARN_ON(error >= 0); | |
2943 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2944 | } | |
2945 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2946 | ||
86db1e29 JA |
2947 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2948 | struct bio *bio) | |
1da177e4 | 2949 | { |
b4f42e28 | 2950 | if (bio_has_data(bio)) |
fb2dce86 | 2951 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
b4f42e28 | 2952 | |
4f024f37 | 2953 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 2954 | rq->bio = rq->biotail = bio; |
1da177e4 | 2955 | |
66846572 N |
2956 | if (bio->bi_bdev) |
2957 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2958 | } | |
1da177e4 | 2959 | |
2d4dc890 IL |
2960 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2961 | /** | |
2962 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2963 | * @rq: the request to be flushed | |
2964 | * | |
2965 | * Description: | |
2966 | * Flush all pages in @rq. | |
2967 | */ | |
2968 | void rq_flush_dcache_pages(struct request *rq) | |
2969 | { | |
2970 | struct req_iterator iter; | |
7988613b | 2971 | struct bio_vec bvec; |
2d4dc890 IL |
2972 | |
2973 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 2974 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
2975 | } |
2976 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2977 | #endif | |
2978 | ||
ef9e3fac KU |
2979 | /** |
2980 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2981 | * @q : the queue of the device being checked | |
2982 | * | |
2983 | * Description: | |
2984 | * Check if underlying low-level drivers of a device are busy. | |
2985 | * If the drivers want to export their busy state, they must set own | |
2986 | * exporting function using blk_queue_lld_busy() first. | |
2987 | * | |
2988 | * Basically, this function is used only by request stacking drivers | |
2989 | * to stop dispatching requests to underlying devices when underlying | |
2990 | * devices are busy. This behavior helps more I/O merging on the queue | |
2991 | * of the request stacking driver and prevents I/O throughput regression | |
2992 | * on burst I/O load. | |
2993 | * | |
2994 | * Return: | |
2995 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2996 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2997 | */ | |
2998 | int blk_lld_busy(struct request_queue *q) | |
2999 | { | |
3000 | if (q->lld_busy_fn) | |
3001 | return q->lld_busy_fn(q); | |
3002 | ||
3003 | return 0; | |
3004 | } | |
3005 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3006 | ||
78d8e58a MS |
3007 | /** |
3008 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3009 | * @rq: the clone request to be cleaned up | |
3010 | * | |
3011 | * Description: | |
3012 | * Free all bios in @rq for a cloned request. | |
3013 | */ | |
3014 | void blk_rq_unprep_clone(struct request *rq) | |
3015 | { | |
3016 | struct bio *bio; | |
3017 | ||
3018 | while ((bio = rq->bio) != NULL) { | |
3019 | rq->bio = bio->bi_next; | |
3020 | ||
3021 | bio_put(bio); | |
3022 | } | |
3023 | } | |
3024 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3025 | ||
3026 | /* | |
3027 | * Copy attributes of the original request to the clone request. | |
3028 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3029 | */ | |
3030 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3031 | { |
3032 | dst->cpu = src->cpu; | |
b0fd271d KU |
3033 | dst->__sector = blk_rq_pos(src); |
3034 | dst->__data_len = blk_rq_bytes(src); | |
3035 | dst->nr_phys_segments = src->nr_phys_segments; | |
3036 | dst->ioprio = src->ioprio; | |
3037 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3038 | } |
3039 | ||
3040 | /** | |
3041 | * blk_rq_prep_clone - Helper function to setup clone request | |
3042 | * @rq: the request to be setup | |
3043 | * @rq_src: original request to be cloned | |
3044 | * @bs: bio_set that bios for clone are allocated from | |
3045 | * @gfp_mask: memory allocation mask for bio | |
3046 | * @bio_ctr: setup function to be called for each clone bio. | |
3047 | * Returns %0 for success, non %0 for failure. | |
3048 | * @data: private data to be passed to @bio_ctr | |
3049 | * | |
3050 | * Description: | |
3051 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3052 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3053 | * are not copied, and copying such parts is the caller's responsibility. | |
3054 | * Also, pages which the original bios are pointing to are not copied | |
3055 | * and the cloned bios just point same pages. | |
3056 | * So cloned bios must be completed before original bios, which means | |
3057 | * the caller must complete @rq before @rq_src. | |
3058 | */ | |
3059 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3060 | struct bio_set *bs, gfp_t gfp_mask, | |
3061 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3062 | void *data) | |
3063 | { | |
3064 | struct bio *bio, *bio_src; | |
3065 | ||
3066 | if (!bs) | |
3067 | bs = fs_bio_set; | |
3068 | ||
3069 | __rq_for_each_bio(bio_src, rq_src) { | |
3070 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3071 | if (!bio) | |
3072 | goto free_and_out; | |
3073 | ||
3074 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3075 | goto free_and_out; | |
3076 | ||
3077 | if (rq->bio) { | |
3078 | rq->biotail->bi_next = bio; | |
3079 | rq->biotail = bio; | |
3080 | } else | |
3081 | rq->bio = rq->biotail = bio; | |
3082 | } | |
3083 | ||
3084 | __blk_rq_prep_clone(rq, rq_src); | |
3085 | ||
3086 | return 0; | |
3087 | ||
3088 | free_and_out: | |
3089 | if (bio) | |
3090 | bio_put(bio); | |
3091 | blk_rq_unprep_clone(rq); | |
3092 | ||
3093 | return -ENOMEM; | |
b0fd271d KU |
3094 | } |
3095 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3096 | ||
59c3d45e | 3097 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3098 | { |
3099 | return queue_work(kblockd_workqueue, work); | |
3100 | } | |
1da177e4 LT |
3101 | EXPORT_SYMBOL(kblockd_schedule_work); |
3102 | ||
ee63cfa7 JA |
3103 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3104 | { | |
3105 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3106 | } | |
3107 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3108 | ||
59c3d45e JA |
3109 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, |
3110 | unsigned long delay) | |
e43473b7 VG |
3111 | { |
3112 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3113 | } | |
3114 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3115 | ||
8ab14595 JA |
3116 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3117 | unsigned long delay) | |
3118 | { | |
3119 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3120 | } | |
3121 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3122 | ||
75df7136 SJ |
3123 | /** |
3124 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3125 | * @plug: The &struct blk_plug that needs to be initialized | |
3126 | * | |
3127 | * Description: | |
3128 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3129 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3130 | * blk_finish_plug(). This is important from a performance perspective, but | |
3131 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3132 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3133 | * page belonging to that request that is currently residing in our private | |
3134 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3135 | * this kind of deadlock. | |
3136 | */ | |
73c10101 JA |
3137 | void blk_start_plug(struct blk_plug *plug) |
3138 | { | |
3139 | struct task_struct *tsk = current; | |
3140 | ||
dd6cf3e1 SL |
3141 | /* |
3142 | * If this is a nested plug, don't actually assign it. | |
3143 | */ | |
3144 | if (tsk->plug) | |
3145 | return; | |
3146 | ||
73c10101 | 3147 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3148 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3149 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3150 | /* |
dd6cf3e1 SL |
3151 | * Store ordering should not be needed here, since a potential |
3152 | * preempt will imply a full memory barrier | |
73c10101 | 3153 | */ |
dd6cf3e1 | 3154 | tsk->plug = plug; |
73c10101 JA |
3155 | } |
3156 | EXPORT_SYMBOL(blk_start_plug); | |
3157 | ||
3158 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3159 | { | |
3160 | struct request *rqa = container_of(a, struct request, queuelist); | |
3161 | struct request *rqb = container_of(b, struct request, queuelist); | |
3162 | ||
975927b9 JM |
3163 | return !(rqa->q < rqb->q || |
3164 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3165 | } |
3166 | ||
49cac01e JA |
3167 | /* |
3168 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3169 | * until a safe kblockd context. We due this to avoid accidental big | |
3170 | * additional stack usage in driver dispatch, in places where the originally | |
3171 | * plugger did not intend it. | |
3172 | */ | |
f6603783 | 3173 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3174 | bool from_schedule) |
99e22598 | 3175 | __releases(q->queue_lock) |
94b5eb28 | 3176 | { |
49cac01e | 3177 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3178 | |
70460571 | 3179 | if (from_schedule) |
24ecfbe2 | 3180 | blk_run_queue_async(q); |
70460571 | 3181 | else |
24ecfbe2 | 3182 | __blk_run_queue(q); |
70460571 | 3183 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3184 | } |
3185 | ||
74018dc3 | 3186 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3187 | { |
3188 | LIST_HEAD(callbacks); | |
3189 | ||
2a7d5559 SL |
3190 | while (!list_empty(&plug->cb_list)) { |
3191 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3192 | |
2a7d5559 SL |
3193 | while (!list_empty(&callbacks)) { |
3194 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3195 | struct blk_plug_cb, |
3196 | list); | |
2a7d5559 | 3197 | list_del(&cb->list); |
74018dc3 | 3198 | cb->callback(cb, from_schedule); |
2a7d5559 | 3199 | } |
048c9374 N |
3200 | } |
3201 | } | |
3202 | ||
9cbb1750 N |
3203 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3204 | int size) | |
3205 | { | |
3206 | struct blk_plug *plug = current->plug; | |
3207 | struct blk_plug_cb *cb; | |
3208 | ||
3209 | if (!plug) | |
3210 | return NULL; | |
3211 | ||
3212 | list_for_each_entry(cb, &plug->cb_list, list) | |
3213 | if (cb->callback == unplug && cb->data == data) | |
3214 | return cb; | |
3215 | ||
3216 | /* Not currently on the callback list */ | |
3217 | BUG_ON(size < sizeof(*cb)); | |
3218 | cb = kzalloc(size, GFP_ATOMIC); | |
3219 | if (cb) { | |
3220 | cb->data = data; | |
3221 | cb->callback = unplug; | |
3222 | list_add(&cb->list, &plug->cb_list); | |
3223 | } | |
3224 | return cb; | |
3225 | } | |
3226 | EXPORT_SYMBOL(blk_check_plugged); | |
3227 | ||
49cac01e | 3228 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3229 | { |
3230 | struct request_queue *q; | |
3231 | unsigned long flags; | |
3232 | struct request *rq; | |
109b8129 | 3233 | LIST_HEAD(list); |
94b5eb28 | 3234 | unsigned int depth; |
73c10101 | 3235 | |
74018dc3 | 3236 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3237 | |
3238 | if (!list_empty(&plug->mq_list)) | |
3239 | blk_mq_flush_plug_list(plug, from_schedule); | |
3240 | ||
73c10101 JA |
3241 | if (list_empty(&plug->list)) |
3242 | return; | |
3243 | ||
109b8129 N |
3244 | list_splice_init(&plug->list, &list); |
3245 | ||
422765c2 | 3246 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3247 | |
3248 | q = NULL; | |
94b5eb28 | 3249 | depth = 0; |
18811272 JA |
3250 | |
3251 | /* | |
3252 | * Save and disable interrupts here, to avoid doing it for every | |
3253 | * queue lock we have to take. | |
3254 | */ | |
73c10101 | 3255 | local_irq_save(flags); |
109b8129 N |
3256 | while (!list_empty(&list)) { |
3257 | rq = list_entry_rq(list.next); | |
73c10101 | 3258 | list_del_init(&rq->queuelist); |
73c10101 JA |
3259 | BUG_ON(!rq->q); |
3260 | if (rq->q != q) { | |
99e22598 JA |
3261 | /* |
3262 | * This drops the queue lock | |
3263 | */ | |
3264 | if (q) | |
49cac01e | 3265 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3266 | q = rq->q; |
94b5eb28 | 3267 | depth = 0; |
73c10101 JA |
3268 | spin_lock(q->queue_lock); |
3269 | } | |
8ba61435 TH |
3270 | |
3271 | /* | |
3272 | * Short-circuit if @q is dead | |
3273 | */ | |
3f3299d5 | 3274 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
3275 | __blk_end_request_all(rq, -ENODEV); |
3276 | continue; | |
3277 | } | |
3278 | ||
73c10101 JA |
3279 | /* |
3280 | * rq is already accounted, so use raw insert | |
3281 | */ | |
f73f44eb | 3282 | if (op_is_flush(rq->cmd_flags)) |
401a18e9 JA |
3283 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3284 | else | |
3285 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3286 | |
3287 | depth++; | |
73c10101 JA |
3288 | } |
3289 | ||
99e22598 JA |
3290 | /* |
3291 | * This drops the queue lock | |
3292 | */ | |
3293 | if (q) | |
49cac01e | 3294 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3295 | |
73c10101 JA |
3296 | local_irq_restore(flags); |
3297 | } | |
73c10101 JA |
3298 | |
3299 | void blk_finish_plug(struct blk_plug *plug) | |
3300 | { | |
dd6cf3e1 SL |
3301 | if (plug != current->plug) |
3302 | return; | |
f6603783 | 3303 | blk_flush_plug_list(plug, false); |
73c10101 | 3304 | |
dd6cf3e1 | 3305 | current->plug = NULL; |
73c10101 | 3306 | } |
88b996cd | 3307 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3308 | |
47fafbc7 | 3309 | #ifdef CONFIG_PM |
6c954667 LM |
3310 | /** |
3311 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3312 | * @q: the queue of the device | |
3313 | * @dev: the device the queue belongs to | |
3314 | * | |
3315 | * Description: | |
3316 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3317 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3318 | * should call this function after @dev has been initialized, and its | |
3319 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3320 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3321 | * cases, driver should call this function before any I/O has taken place. | |
3322 | * | |
3323 | * This function takes care of setting up using auto suspend for the device, | |
3324 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3325 | * until an updated value is either set by user or by driver. Drivers do | |
3326 | * not need to touch other autosuspend settings. | |
3327 | * | |
3328 | * The block layer runtime PM is request based, so only works for drivers | |
3329 | * that use request as their IO unit instead of those directly use bio's. | |
3330 | */ | |
3331 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3332 | { | |
3333 | q->dev = dev; | |
3334 | q->rpm_status = RPM_ACTIVE; | |
3335 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3336 | pm_runtime_use_autosuspend(q->dev); | |
3337 | } | |
3338 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3339 | ||
3340 | /** | |
3341 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3342 | * @q: the queue of the device | |
3343 | * | |
3344 | * Description: | |
3345 | * This function will check if runtime suspend is allowed for the device | |
3346 | * by examining if there are any requests pending in the queue. If there | |
3347 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3348 | * the queue's status will be updated to SUSPENDING and the driver can | |
3349 | * proceed to suspend the device. | |
3350 | * | |
3351 | * For the not allowed case, we mark last busy for the device so that | |
3352 | * runtime PM core will try to autosuspend it some time later. | |
3353 | * | |
3354 | * This function should be called near the start of the device's | |
3355 | * runtime_suspend callback. | |
3356 | * | |
3357 | * Return: | |
3358 | * 0 - OK to runtime suspend the device | |
3359 | * -EBUSY - Device should not be runtime suspended | |
3360 | */ | |
3361 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3362 | { | |
3363 | int ret = 0; | |
3364 | ||
4fd41a85 KX |
3365 | if (!q->dev) |
3366 | return ret; | |
3367 | ||
6c954667 LM |
3368 | spin_lock_irq(q->queue_lock); |
3369 | if (q->nr_pending) { | |
3370 | ret = -EBUSY; | |
3371 | pm_runtime_mark_last_busy(q->dev); | |
3372 | } else { | |
3373 | q->rpm_status = RPM_SUSPENDING; | |
3374 | } | |
3375 | spin_unlock_irq(q->queue_lock); | |
3376 | return ret; | |
3377 | } | |
3378 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3379 | ||
3380 | /** | |
3381 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3382 | * @q: the queue of the device | |
3383 | * @err: return value of the device's runtime_suspend function | |
3384 | * | |
3385 | * Description: | |
3386 | * Update the queue's runtime status according to the return value of the | |
3387 | * device's runtime suspend function and mark last busy for the device so | |
3388 | * that PM core will try to auto suspend the device at a later time. | |
3389 | * | |
3390 | * This function should be called near the end of the device's | |
3391 | * runtime_suspend callback. | |
3392 | */ | |
3393 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3394 | { | |
4fd41a85 KX |
3395 | if (!q->dev) |
3396 | return; | |
3397 | ||
6c954667 LM |
3398 | spin_lock_irq(q->queue_lock); |
3399 | if (!err) { | |
3400 | q->rpm_status = RPM_SUSPENDED; | |
3401 | } else { | |
3402 | q->rpm_status = RPM_ACTIVE; | |
3403 | pm_runtime_mark_last_busy(q->dev); | |
3404 | } | |
3405 | spin_unlock_irq(q->queue_lock); | |
3406 | } | |
3407 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3408 | ||
3409 | /** | |
3410 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3411 | * @q: the queue of the device | |
3412 | * | |
3413 | * Description: | |
3414 | * Update the queue's runtime status to RESUMING in preparation for the | |
3415 | * runtime resume of the device. | |
3416 | * | |
3417 | * This function should be called near the start of the device's | |
3418 | * runtime_resume callback. | |
3419 | */ | |
3420 | void blk_pre_runtime_resume(struct request_queue *q) | |
3421 | { | |
4fd41a85 KX |
3422 | if (!q->dev) |
3423 | return; | |
3424 | ||
6c954667 LM |
3425 | spin_lock_irq(q->queue_lock); |
3426 | q->rpm_status = RPM_RESUMING; | |
3427 | spin_unlock_irq(q->queue_lock); | |
3428 | } | |
3429 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3430 | ||
3431 | /** | |
3432 | * blk_post_runtime_resume - Post runtime resume processing | |
3433 | * @q: the queue of the device | |
3434 | * @err: return value of the device's runtime_resume function | |
3435 | * | |
3436 | * Description: | |
3437 | * Update the queue's runtime status according to the return value of the | |
3438 | * device's runtime_resume function. If it is successfully resumed, process | |
3439 | * the requests that are queued into the device's queue when it is resuming | |
3440 | * and then mark last busy and initiate autosuspend for it. | |
3441 | * | |
3442 | * This function should be called near the end of the device's | |
3443 | * runtime_resume callback. | |
3444 | */ | |
3445 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3446 | { | |
4fd41a85 KX |
3447 | if (!q->dev) |
3448 | return; | |
3449 | ||
6c954667 LM |
3450 | spin_lock_irq(q->queue_lock); |
3451 | if (!err) { | |
3452 | q->rpm_status = RPM_ACTIVE; | |
3453 | __blk_run_queue(q); | |
3454 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3455 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3456 | } else { |
3457 | q->rpm_status = RPM_SUSPENDED; | |
3458 | } | |
3459 | spin_unlock_irq(q->queue_lock); | |
3460 | } | |
3461 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3462 | |
3463 | /** | |
3464 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3465 | * @q: the queue of the device | |
3466 | * | |
3467 | * If the device is left runtime suspended during system suspend the resume | |
3468 | * hook typically resumes the device and corrects runtime status | |
3469 | * accordingly. However, that does not affect the queue runtime PM status | |
3470 | * which is still "suspended". This prevents processing requests from the | |
3471 | * queue. | |
3472 | * | |
3473 | * This function can be used in driver's resume hook to correct queue | |
3474 | * runtime PM status and re-enable peeking requests from the queue. It | |
3475 | * should be called before first request is added to the queue. | |
3476 | */ | |
3477 | void blk_set_runtime_active(struct request_queue *q) | |
3478 | { | |
3479 | spin_lock_irq(q->queue_lock); | |
3480 | q->rpm_status = RPM_ACTIVE; | |
3481 | pm_runtime_mark_last_busy(q->dev); | |
3482 | pm_request_autosuspend(q->dev); | |
3483 | spin_unlock_irq(q->queue_lock); | |
3484 | } | |
3485 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3486 | #endif |
3487 | ||
1da177e4 LT |
3488 | int __init blk_dev_init(void) |
3489 | { | |
ef295ecf CH |
3490 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3491 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3492 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3493 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3494 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3495 | |
89b90be2 TH |
3496 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3497 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3498 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3499 | if (!kblockd_workqueue) |
3500 | panic("Failed to create kblockd\n"); | |
3501 | ||
3502 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3503 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3504 | |
c2789bd4 | 3505 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3506 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3507 | |
18fbda91 OS |
3508 | #ifdef CONFIG_DEBUG_FS |
3509 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3510 | #endif | |
3511 | ||
d38ecf93 | 3512 | return 0; |
1da177e4 | 3513 | } |