]>
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> |
55782138 LZ |
36 | |
37 | #define CREATE_TRACE_POINTS | |
38 | #include <trace/events/block.h> | |
1da177e4 | 39 | |
8324aa91 | 40 | #include "blk.h" |
43a5e4e2 | 41 | #include "blk-mq.h" |
8324aa91 | 42 | |
d07335e5 | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 44 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 45 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 46 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 47 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 48 | |
a73f730d TH |
49 | DEFINE_IDA(blk_queue_ida); |
50 | ||
1da177e4 LT |
51 | /* |
52 | * For the allocated request tables | |
53 | */ | |
320ae51f | 54 | struct kmem_cache *request_cachep = NULL; |
1da177e4 LT |
55 | |
56 | /* | |
57 | * For queue allocation | |
58 | */ | |
6728cb0e | 59 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 60 | |
1da177e4 LT |
61 | /* |
62 | * Controlling structure to kblockd | |
63 | */ | |
ff856bad | 64 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 65 | |
8324aa91 | 66 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
67 | { |
68 | int nr; | |
69 | ||
70 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
71 | if (nr > q->nr_requests) | |
72 | nr = q->nr_requests; | |
73 | q->nr_congestion_on = nr; | |
74 | ||
75 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
76 | if (nr < 1) | |
77 | nr = 1; | |
78 | q->nr_congestion_off = nr; | |
79 | } | |
80 | ||
1da177e4 LT |
81 | /** |
82 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
83 | * @bdev: device | |
84 | * | |
85 | * Locates the passed device's request queue and returns the address of its | |
ff9ea323 TH |
86 | * backing_dev_info. This function can only be called if @bdev is opened |
87 | * and the return value is never NULL. | |
1da177e4 LT |
88 | */ |
89 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
90 | { | |
165125e1 | 91 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 | 92 | |
ff9ea323 | 93 | return &q->backing_dev_info; |
1da177e4 | 94 | } |
1da177e4 LT |
95 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
96 | ||
2a4aa30c | 97 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 98 | { |
1afb20f3 FT |
99 | memset(rq, 0, sizeof(*rq)); |
100 | ||
1da177e4 | 101 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 102 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 103 | rq->cpu = -1; |
63a71386 | 104 | rq->q = q; |
a2dec7b3 | 105 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
106 | INIT_HLIST_NODE(&rq->hash); |
107 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 108 | rq->cmd = rq->__cmd; |
e2494e1b | 109 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 110 | rq->tag = -1; |
b243ddcb | 111 | rq->start_time = jiffies; |
9195291e | 112 | set_start_time_ns(rq); |
09e099d4 | 113 | rq->part = NULL; |
1da177e4 | 114 | } |
2a4aa30c | 115 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 116 | |
5bb23a68 N |
117 | static void req_bio_endio(struct request *rq, struct bio *bio, |
118 | unsigned int nbytes, int error) | |
1da177e4 | 119 | { |
5f1b670d | 120 | if (error && !(rq->cmd_flags & REQ_CLONE)) |
143a87f4 TH |
121 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
122 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
123 | error = -EIO; | |
797e7dbb | 124 | |
143a87f4 TH |
125 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
126 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 127 | |
f79ea416 | 128 | bio_advance(bio, nbytes); |
7ba1ba12 | 129 | |
143a87f4 | 130 | /* don't actually finish bio if it's part of flush sequence */ |
5f1b670d CH |
131 | if (bio->bi_iter.bi_size == 0 && |
132 | !(rq->cmd_flags & (REQ_FLUSH_SEQ|REQ_CLONE))) | |
143a87f4 | 133 | bio_endio(bio, error); |
1da177e4 | 134 | } |
1da177e4 | 135 | |
1da177e4 LT |
136 | void blk_dump_rq_flags(struct request *rq, char *msg) |
137 | { | |
138 | int bit; | |
139 | ||
5953316d | 140 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg, |
4aff5e23 | 141 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
5953316d | 142 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 143 | |
83096ebf TH |
144 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
145 | (unsigned long long)blk_rq_pos(rq), | |
146 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
147 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
148 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 149 | |
33659ebb | 150 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 151 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 152 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
153 | printk("%02x ", rq->cmd[bit]); |
154 | printk("\n"); | |
155 | } | |
156 | } | |
1da177e4 LT |
157 | EXPORT_SYMBOL(blk_dump_rq_flags); |
158 | ||
3cca6dc1 | 159 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 160 | { |
3cca6dc1 | 161 | struct request_queue *q; |
1da177e4 | 162 | |
3cca6dc1 JA |
163 | q = container_of(work, struct request_queue, delay_work.work); |
164 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 165 | __blk_run_queue(q); |
3cca6dc1 | 166 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 167 | } |
1da177e4 LT |
168 | |
169 | /** | |
3cca6dc1 JA |
170 | * blk_delay_queue - restart queueing after defined interval |
171 | * @q: The &struct request_queue in question | |
172 | * @msecs: Delay in msecs | |
1da177e4 LT |
173 | * |
174 | * Description: | |
3cca6dc1 JA |
175 | * Sometimes queueing needs to be postponed for a little while, to allow |
176 | * resources to come back. This function will make sure that queueing is | |
70460571 | 177 | * restarted around the specified time. Queue lock must be held. |
3cca6dc1 JA |
178 | */ |
179 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 180 | { |
70460571 BVA |
181 | if (likely(!blk_queue_dead(q))) |
182 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
183 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 184 | } |
3cca6dc1 | 185 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 186 | |
1da177e4 LT |
187 | /** |
188 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 189 | * @q: The &struct request_queue in question |
1da177e4 LT |
190 | * |
191 | * Description: | |
192 | * blk_start_queue() will clear the stop flag on the queue, and call | |
193 | * the request_fn for the queue if it was in a stopped state when | |
194 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
195 | **/ | |
165125e1 | 196 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 197 | { |
a038e253 PBG |
198 | WARN_ON(!irqs_disabled()); |
199 | ||
75ad23bc | 200 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 201 | __blk_run_queue(q); |
1da177e4 | 202 | } |
1da177e4 LT |
203 | EXPORT_SYMBOL(blk_start_queue); |
204 | ||
205 | /** | |
206 | * blk_stop_queue - stop a queue | |
165125e1 | 207 | * @q: The &struct request_queue in question |
1da177e4 LT |
208 | * |
209 | * Description: | |
210 | * The Linux block layer assumes that a block driver will consume all | |
211 | * entries on the request queue when the request_fn strategy is called. | |
212 | * Often this will not happen, because of hardware limitations (queue | |
213 | * depth settings). If a device driver gets a 'queue full' response, | |
214 | * or if it simply chooses not to queue more I/O at one point, it can | |
215 | * call this function to prevent the request_fn from being called until | |
216 | * the driver has signalled it's ready to go again. This happens by calling | |
217 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
218 | **/ | |
165125e1 | 219 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 220 | { |
136b5721 | 221 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 222 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
223 | } |
224 | EXPORT_SYMBOL(blk_stop_queue); | |
225 | ||
226 | /** | |
227 | * blk_sync_queue - cancel any pending callbacks on a queue | |
228 | * @q: the queue | |
229 | * | |
230 | * Description: | |
231 | * The block layer may perform asynchronous callback activity | |
232 | * on a queue, such as calling the unplug function after a timeout. | |
233 | * A block device may call blk_sync_queue to ensure that any | |
234 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 235 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
236 | * that its ->make_request_fn will not re-add plugging prior to calling |
237 | * this function. | |
238 | * | |
da527770 | 239 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 240 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 241 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 242 | * |
1da177e4 LT |
243 | */ |
244 | void blk_sync_queue(struct request_queue *q) | |
245 | { | |
70ed28b9 | 246 | del_timer_sync(&q->timeout); |
f04c1fe7 ML |
247 | |
248 | if (q->mq_ops) { | |
249 | struct blk_mq_hw_ctx *hctx; | |
250 | int i; | |
251 | ||
70f4db63 CH |
252 | queue_for_each_hw_ctx(q, hctx, i) { |
253 | cancel_delayed_work_sync(&hctx->run_work); | |
254 | cancel_delayed_work_sync(&hctx->delay_work); | |
255 | } | |
f04c1fe7 ML |
256 | } else { |
257 | cancel_delayed_work_sync(&q->delay_work); | |
258 | } | |
1da177e4 LT |
259 | } |
260 | EXPORT_SYMBOL(blk_sync_queue); | |
261 | ||
c246e80d BVA |
262 | /** |
263 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
264 | * @q: The queue to run | |
265 | * | |
266 | * Description: | |
267 | * Invoke request handling on a queue if there are any pending requests. | |
268 | * May be used to restart request handling after a request has completed. | |
269 | * This variant runs the queue whether or not the queue has been | |
270 | * stopped. Must be called with the queue lock held and interrupts | |
271 | * disabled. See also @blk_run_queue. | |
272 | */ | |
273 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
274 | { | |
275 | if (unlikely(blk_queue_dead(q))) | |
276 | return; | |
277 | ||
24faf6f6 BVA |
278 | /* |
279 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
280 | * the queue lock internally. As a result multiple threads may be | |
281 | * running such a request function concurrently. Keep track of the | |
282 | * number of active request_fn invocations such that blk_drain_queue() | |
283 | * can wait until all these request_fn calls have finished. | |
284 | */ | |
285 | q->request_fn_active++; | |
c246e80d | 286 | q->request_fn(q); |
24faf6f6 | 287 | q->request_fn_active--; |
c246e80d | 288 | } |
a7928c15 | 289 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 290 | |
1da177e4 | 291 | /** |
80a4b58e | 292 | * __blk_run_queue - run a single device queue |
1da177e4 | 293 | * @q: The queue to run |
80a4b58e JA |
294 | * |
295 | * Description: | |
296 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 297 | * held and interrupts disabled. |
1da177e4 | 298 | */ |
24ecfbe2 | 299 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 300 | { |
a538cd03 TH |
301 | if (unlikely(blk_queue_stopped(q))) |
302 | return; | |
303 | ||
c246e80d | 304 | __blk_run_queue_uncond(q); |
75ad23bc NP |
305 | } |
306 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 307 | |
24ecfbe2 CH |
308 | /** |
309 | * blk_run_queue_async - run a single device queue in workqueue context | |
310 | * @q: The queue to run | |
311 | * | |
312 | * Description: | |
313 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
70460571 | 314 | * of us. The caller must hold the queue lock. |
24ecfbe2 CH |
315 | */ |
316 | void blk_run_queue_async(struct request_queue *q) | |
317 | { | |
70460571 | 318 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 319 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 320 | } |
c21e6beb | 321 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 322 | |
75ad23bc NP |
323 | /** |
324 | * blk_run_queue - run a single device queue | |
325 | * @q: The queue to run | |
80a4b58e JA |
326 | * |
327 | * Description: | |
328 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 329 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
330 | */ |
331 | void blk_run_queue(struct request_queue *q) | |
332 | { | |
333 | unsigned long flags; | |
334 | ||
335 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 336 | __blk_run_queue(q); |
1da177e4 LT |
337 | spin_unlock_irqrestore(q->queue_lock, flags); |
338 | } | |
339 | EXPORT_SYMBOL(blk_run_queue); | |
340 | ||
165125e1 | 341 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
342 | { |
343 | kobject_put(&q->kobj); | |
344 | } | |
d86e0e83 | 345 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 346 | |
e3c78ca5 | 347 | /** |
807592a4 | 348 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 349 | * @q: queue to drain |
c9a929dd | 350 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 351 | * |
c9a929dd TH |
352 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
353 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
354 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 355 | */ |
807592a4 BVA |
356 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
357 | __releases(q->queue_lock) | |
358 | __acquires(q->queue_lock) | |
e3c78ca5 | 359 | { |
458f27a9 AH |
360 | int i; |
361 | ||
807592a4 BVA |
362 | lockdep_assert_held(q->queue_lock); |
363 | ||
e3c78ca5 | 364 | while (true) { |
481a7d64 | 365 | bool drain = false; |
e3c78ca5 | 366 | |
b855b04a TH |
367 | /* |
368 | * The caller might be trying to drain @q before its | |
369 | * elevator is initialized. | |
370 | */ | |
371 | if (q->elevator) | |
372 | elv_drain_elevator(q); | |
373 | ||
5efd6113 | 374 | blkcg_drain_queue(q); |
e3c78ca5 | 375 | |
4eabc941 TH |
376 | /* |
377 | * This function might be called on a queue which failed | |
b855b04a TH |
378 | * driver init after queue creation or is not yet fully |
379 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
380 | * in such cases. Kick queue iff dispatch queue has | |
381 | * something on it and @q has request_fn set. | |
4eabc941 | 382 | */ |
b855b04a | 383 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 384 | __blk_run_queue(q); |
c9a929dd | 385 | |
8a5ecdd4 | 386 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 387 | drain |= q->request_fn_active; |
481a7d64 TH |
388 | |
389 | /* | |
390 | * Unfortunately, requests are queued at and tracked from | |
391 | * multiple places and there's no single counter which can | |
392 | * be drained. Check all the queues and counters. | |
393 | */ | |
394 | if (drain_all) { | |
e97c293c | 395 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
396 | drain |= !list_empty(&q->queue_head); |
397 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 398 | drain |= q->nr_rqs[i]; |
481a7d64 | 399 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
400 | if (fq) |
401 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
402 | } |
403 | } | |
e3c78ca5 | 404 | |
481a7d64 | 405 | if (!drain) |
e3c78ca5 | 406 | break; |
807592a4 BVA |
407 | |
408 | spin_unlock_irq(q->queue_lock); | |
409 | ||
e3c78ca5 | 410 | msleep(10); |
807592a4 BVA |
411 | |
412 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 413 | } |
458f27a9 AH |
414 | |
415 | /* | |
416 | * With queue marked dead, any woken up waiter will fail the | |
417 | * allocation path, so the wakeup chaining is lost and we're | |
418 | * left with hung waiters. We need to wake up those waiters. | |
419 | */ | |
420 | if (q->request_fn) { | |
a051661c TH |
421 | struct request_list *rl; |
422 | ||
a051661c TH |
423 | blk_queue_for_each_rl(rl, q) |
424 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
425 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 426 | } |
e3c78ca5 TH |
427 | } |
428 | ||
d732580b TH |
429 | /** |
430 | * blk_queue_bypass_start - enter queue bypass mode | |
431 | * @q: queue of interest | |
432 | * | |
433 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
434 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 435 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
436 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
437 | * inside queue or RCU read lock. | |
d732580b TH |
438 | */ |
439 | void blk_queue_bypass_start(struct request_queue *q) | |
440 | { | |
441 | spin_lock_irq(q->queue_lock); | |
776687bc | 442 | q->bypass_depth++; |
d732580b TH |
443 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
444 | spin_unlock_irq(q->queue_lock); | |
445 | ||
776687bc TH |
446 | /* |
447 | * Queues start drained. Skip actual draining till init is | |
448 | * complete. This avoids lenghty delays during queue init which | |
449 | * can happen many times during boot. | |
450 | */ | |
451 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
452 | spin_lock_irq(q->queue_lock); |
453 | __blk_drain_queue(q, false); | |
454 | spin_unlock_irq(q->queue_lock); | |
455 | ||
b82d4b19 TH |
456 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
457 | synchronize_rcu(); | |
458 | } | |
d732580b TH |
459 | } |
460 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
461 | ||
462 | /** | |
463 | * blk_queue_bypass_end - leave queue bypass mode | |
464 | * @q: queue of interest | |
465 | * | |
466 | * Leave bypass mode and restore the normal queueing behavior. | |
467 | */ | |
468 | void blk_queue_bypass_end(struct request_queue *q) | |
469 | { | |
470 | spin_lock_irq(q->queue_lock); | |
471 | if (!--q->bypass_depth) | |
472 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
473 | WARN_ON_ONCE(q->bypass_depth < 0); | |
474 | spin_unlock_irq(q->queue_lock); | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
477 | ||
aed3ea94 JA |
478 | void blk_set_queue_dying(struct request_queue *q) |
479 | { | |
480 | queue_flag_set_unlocked(QUEUE_FLAG_DYING, q); | |
481 | ||
482 | if (q->mq_ops) | |
483 | blk_mq_wake_waiters(q); | |
484 | else { | |
485 | struct request_list *rl; | |
486 | ||
487 | blk_queue_for_each_rl(rl, q) { | |
488 | if (rl->rq_pool) { | |
489 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
490 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
491 | } | |
492 | } | |
493 | } | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
496 | ||
c9a929dd TH |
497 | /** |
498 | * blk_cleanup_queue - shutdown a request queue | |
499 | * @q: request queue to shutdown | |
500 | * | |
c246e80d BVA |
501 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
502 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 503 | */ |
6728cb0e | 504 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 505 | { |
c9a929dd | 506 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 507 | |
3f3299d5 | 508 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 509 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 510 | blk_set_queue_dying(q); |
c9a929dd | 511 | spin_lock_irq(lock); |
6ecf23af | 512 | |
80fd9979 | 513 | /* |
3f3299d5 | 514 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
515 | * that, unlike blk_queue_bypass_start(), we aren't performing |
516 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
517 | * as some drivers create and destroy a lot of queues while | |
518 | * probing. This is still safe because blk_release_queue() will be | |
519 | * called only after the queue refcnt drops to zero and nothing, | |
520 | * RCU or not, would be traversing the queue by then. | |
521 | */ | |
6ecf23af TH |
522 | q->bypass_depth++; |
523 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
524 | ||
c9a929dd TH |
525 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
526 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 527 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
528 | spin_unlock_irq(lock); |
529 | mutex_unlock(&q->sysfs_lock); | |
530 | ||
c246e80d BVA |
531 | /* |
532 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
533 | * prevent that q->request_fn() gets invoked after draining finished. | |
534 | */ | |
43a5e4e2 | 535 | if (q->mq_ops) { |
780db207 | 536 | blk_mq_freeze_queue(q); |
43a5e4e2 ML |
537 | spin_lock_irq(lock); |
538 | } else { | |
539 | spin_lock_irq(lock); | |
540 | __blk_drain_queue(q, true); | |
541 | } | |
c246e80d | 542 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 543 | spin_unlock_irq(lock); |
c9a929dd TH |
544 | |
545 | /* @q won't process any more request, flush async actions */ | |
546 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
547 | blk_sync_queue(q); | |
548 | ||
45a9c9d9 BVA |
549 | if (q->mq_ops) |
550 | blk_mq_free_queue(q); | |
551 | ||
5e5cfac0 AH |
552 | spin_lock_irq(lock); |
553 | if (q->queue_lock != &q->__queue_lock) | |
554 | q->queue_lock = &q->__queue_lock; | |
555 | spin_unlock_irq(lock); | |
556 | ||
c9a929dd | 557 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
558 | blk_put_queue(q); |
559 | } | |
1da177e4 LT |
560 | EXPORT_SYMBOL(blk_cleanup_queue); |
561 | ||
271508db DR |
562 | /* Allocate memory local to the request queue */ |
563 | static void *alloc_request_struct(gfp_t gfp_mask, void *data) | |
564 | { | |
565 | int nid = (int)(long)data; | |
566 | return kmem_cache_alloc_node(request_cachep, gfp_mask, nid); | |
567 | } | |
568 | ||
569 | static void free_request_struct(void *element, void *unused) | |
570 | { | |
571 | kmem_cache_free(request_cachep, element); | |
572 | } | |
573 | ||
5b788ce3 TH |
574 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
575 | gfp_t gfp_mask) | |
1da177e4 | 576 | { |
1abec4fd MS |
577 | if (unlikely(rl->rq_pool)) |
578 | return 0; | |
579 | ||
5b788ce3 | 580 | rl->q = q; |
1faa16d2 JA |
581 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
582 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
583 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
584 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 585 | |
271508db DR |
586 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct, |
587 | free_request_struct, | |
588 | (void *)(long)q->node, gfp_mask, | |
589 | q->node); | |
1da177e4 LT |
590 | if (!rl->rq_pool) |
591 | return -ENOMEM; | |
592 | ||
593 | return 0; | |
594 | } | |
595 | ||
5b788ce3 TH |
596 | void blk_exit_rl(struct request_list *rl) |
597 | { | |
598 | if (rl->rq_pool) | |
599 | mempool_destroy(rl->rq_pool); | |
600 | } | |
601 | ||
165125e1 | 602 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 603 | { |
c304a51b | 604 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
605 | } |
606 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 607 | |
165125e1 | 608 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 609 | { |
165125e1 | 610 | struct request_queue *q; |
e0bf68dd | 611 | int err; |
1946089a | 612 | |
8324aa91 | 613 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 614 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
615 | if (!q) |
616 | return NULL; | |
617 | ||
00380a40 | 618 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 619 | if (q->id < 0) |
3d2936f4 | 620 | goto fail_q; |
a73f730d | 621 | |
0989a025 JA |
622 | q->backing_dev_info.ra_pages = |
623 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
b4caecd4 | 624 | q->backing_dev_info.capabilities = 0; |
d993831f | 625 | q->backing_dev_info.name = "block"; |
5151412d | 626 | q->node = node_id; |
0989a025 | 627 | |
e0bf68dd | 628 | err = bdi_init(&q->backing_dev_info); |
a73f730d TH |
629 | if (err) |
630 | goto fail_id; | |
e0bf68dd | 631 | |
31373d09 MG |
632 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
633 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 634 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 635 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 636 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 637 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 638 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 639 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 640 | #endif |
3cca6dc1 | 641 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 642 | |
8324aa91 | 643 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 644 | |
483f4afc | 645 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 646 | spin_lock_init(&q->__queue_lock); |
483f4afc | 647 | |
c94a96ac VG |
648 | /* |
649 | * By default initialize queue_lock to internal lock and driver can | |
650 | * override it later if need be. | |
651 | */ | |
652 | q->queue_lock = &q->__queue_lock; | |
653 | ||
b82d4b19 TH |
654 | /* |
655 | * A queue starts its life with bypass turned on to avoid | |
656 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
657 | * init. The initial bypass will be finished when the queue is |
658 | * registered by blk_register_queue(). | |
b82d4b19 TH |
659 | */ |
660 | q->bypass_depth = 1; | |
661 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
662 | ||
320ae51f JA |
663 | init_waitqueue_head(&q->mq_freeze_wq); |
664 | ||
5efd6113 | 665 | if (blkcg_init_queue(q)) |
fff4996b | 666 | goto fail_bdi; |
f51b802c | 667 | |
1da177e4 | 668 | return q; |
a73f730d | 669 | |
fff4996b MP |
670 | fail_bdi: |
671 | bdi_destroy(&q->backing_dev_info); | |
a73f730d TH |
672 | fail_id: |
673 | ida_simple_remove(&blk_queue_ida, q->id); | |
674 | fail_q: | |
675 | kmem_cache_free(blk_requestq_cachep, q); | |
676 | return NULL; | |
1da177e4 | 677 | } |
1946089a | 678 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
679 | |
680 | /** | |
681 | * blk_init_queue - prepare a request queue for use with a block device | |
682 | * @rfn: The function to be called to process requests that have been | |
683 | * placed on the queue. | |
684 | * @lock: Request queue spin lock | |
685 | * | |
686 | * Description: | |
687 | * If a block device wishes to use the standard request handling procedures, | |
688 | * which sorts requests and coalesces adjacent requests, then it must | |
689 | * call blk_init_queue(). The function @rfn will be called when there | |
690 | * are requests on the queue that need to be processed. If the device | |
691 | * supports plugging, then @rfn may not be called immediately when requests | |
692 | * are available on the queue, but may be called at some time later instead. | |
693 | * Plugged queues are generally unplugged when a buffer belonging to one | |
694 | * of the requests on the queue is needed, or due to memory pressure. | |
695 | * | |
696 | * @rfn is not required, or even expected, to remove all requests off the | |
697 | * queue, but only as many as it can handle at a time. If it does leave | |
698 | * requests on the queue, it is responsible for arranging that the requests | |
699 | * get dealt with eventually. | |
700 | * | |
701 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
702 | * request queue; this lock will be taken also from interrupt context, so irq |
703 | * disabling is needed for it. | |
1da177e4 | 704 | * |
710027a4 | 705 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
706 | * it didn't succeed. |
707 | * | |
708 | * Note: | |
709 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
710 | * when the block device is deactivated (such as at module unload). | |
711 | **/ | |
1946089a | 712 | |
165125e1 | 713 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 714 | { |
c304a51b | 715 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
716 | } |
717 | EXPORT_SYMBOL(blk_init_queue); | |
718 | ||
165125e1 | 719 | struct request_queue * |
1946089a CL |
720 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
721 | { | |
c86d1b8a | 722 | struct request_queue *uninit_q, *q; |
1da177e4 | 723 | |
c86d1b8a MS |
724 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
725 | if (!uninit_q) | |
726 | return NULL; | |
727 | ||
5151412d | 728 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a | 729 | if (!q) |
7982e90c | 730 | blk_cleanup_queue(uninit_q); |
18741986 | 731 | |
7982e90c | 732 | return q; |
01effb0d MS |
733 | } |
734 | EXPORT_SYMBOL(blk_init_queue_node); | |
735 | ||
736 | struct request_queue * | |
737 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
738 | spinlock_t *lock) | |
01effb0d | 739 | { |
1da177e4 LT |
740 | if (!q) |
741 | return NULL; | |
742 | ||
f70ced09 | 743 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0); |
ba483388 | 744 | if (!q->fq) |
7982e90c MS |
745 | return NULL; |
746 | ||
a051661c | 747 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
708f04d2 | 748 | goto fail; |
1da177e4 LT |
749 | |
750 | q->request_fn = rfn; | |
1da177e4 | 751 | q->prep_rq_fn = NULL; |
28018c24 | 752 | q->unprep_rq_fn = NULL; |
60ea8226 | 753 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
754 | |
755 | /* Override internal queue lock with supplied lock pointer */ | |
756 | if (lock) | |
757 | q->queue_lock = lock; | |
1da177e4 | 758 | |
f3b144aa JA |
759 | /* |
760 | * This also sets hw/phys segments, boundary and size | |
761 | */ | |
c20e8de2 | 762 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 763 | |
44ec9542 AS |
764 | q->sg_reserved_size = INT_MAX; |
765 | ||
eb1c160b TS |
766 | /* Protect q->elevator from elevator_change */ |
767 | mutex_lock(&q->sysfs_lock); | |
768 | ||
b82d4b19 | 769 | /* init elevator */ |
eb1c160b TS |
770 | if (elevator_init(q, NULL)) { |
771 | mutex_unlock(&q->sysfs_lock); | |
708f04d2 | 772 | goto fail; |
eb1c160b TS |
773 | } |
774 | ||
775 | mutex_unlock(&q->sysfs_lock); | |
776 | ||
b82d4b19 | 777 | return q; |
708f04d2 DJ |
778 | |
779 | fail: | |
ba483388 | 780 | blk_free_flush_queue(q->fq); |
708f04d2 | 781 | return NULL; |
1da177e4 | 782 | } |
5151412d | 783 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 784 | |
09ac46c4 | 785 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 786 | { |
3f3299d5 | 787 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
788 | __blk_get_queue(q); |
789 | return true; | |
1da177e4 LT |
790 | } |
791 | ||
09ac46c4 | 792 | return false; |
1da177e4 | 793 | } |
d86e0e83 | 794 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 795 | |
5b788ce3 | 796 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 797 | { |
f1f8cc94 | 798 | if (rq->cmd_flags & REQ_ELVPRIV) { |
5b788ce3 | 799 | elv_put_request(rl->q, rq); |
f1f8cc94 | 800 | if (rq->elv.icq) |
11a3122f | 801 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
802 | } |
803 | ||
5b788ce3 | 804 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
805 | } |
806 | ||
1da177e4 LT |
807 | /* |
808 | * ioc_batching returns true if the ioc is a valid batching request and | |
809 | * should be given priority access to a request. | |
810 | */ | |
165125e1 | 811 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
812 | { |
813 | if (!ioc) | |
814 | return 0; | |
815 | ||
816 | /* | |
817 | * Make sure the process is able to allocate at least 1 request | |
818 | * even if the batch times out, otherwise we could theoretically | |
819 | * lose wakeups. | |
820 | */ | |
821 | return ioc->nr_batch_requests == q->nr_batching || | |
822 | (ioc->nr_batch_requests > 0 | |
823 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
824 | } | |
825 | ||
826 | /* | |
827 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
828 | * will cause the process to be a "batcher" on all queues in the system. This | |
829 | * is the behaviour we want though - once it gets a wakeup it should be given | |
830 | * a nice run. | |
831 | */ | |
165125e1 | 832 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
833 | { |
834 | if (!ioc || ioc_batching(q, ioc)) | |
835 | return; | |
836 | ||
837 | ioc->nr_batch_requests = q->nr_batching; | |
838 | ioc->last_waited = jiffies; | |
839 | } | |
840 | ||
5b788ce3 | 841 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 842 | { |
5b788ce3 | 843 | struct request_queue *q = rl->q; |
1da177e4 | 844 | |
a051661c TH |
845 | /* |
846 | * bdi isn't aware of blkcg yet. As all async IOs end up root | |
847 | * blkcg anyway, just use root blkcg state. | |
848 | */ | |
849 | if (rl == &q->root_rl && | |
850 | rl->count[sync] < queue_congestion_off_threshold(q)) | |
1faa16d2 | 851 | blk_clear_queue_congested(q, sync); |
1da177e4 | 852 | |
1faa16d2 JA |
853 | if (rl->count[sync] + 1 <= q->nr_requests) { |
854 | if (waitqueue_active(&rl->wait[sync])) | |
855 | wake_up(&rl->wait[sync]); | |
1da177e4 | 856 | |
5b788ce3 | 857 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
858 | } |
859 | } | |
860 | ||
861 | /* | |
862 | * A request has just been released. Account for it, update the full and | |
863 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
864 | */ | |
5b788ce3 | 865 | static void freed_request(struct request_list *rl, unsigned int flags) |
1da177e4 | 866 | { |
5b788ce3 | 867 | struct request_queue *q = rl->q; |
75eb6c37 | 868 | int sync = rw_is_sync(flags); |
1da177e4 | 869 | |
8a5ecdd4 | 870 | q->nr_rqs[sync]--; |
1faa16d2 | 871 | rl->count[sync]--; |
75eb6c37 | 872 | if (flags & REQ_ELVPRIV) |
8a5ecdd4 | 873 | q->nr_rqs_elvpriv--; |
1da177e4 | 874 | |
5b788ce3 | 875 | __freed_request(rl, sync); |
1da177e4 | 876 | |
1faa16d2 | 877 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 878 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
879 | } |
880 | ||
e3a2b3f9 JA |
881 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
882 | { | |
883 | struct request_list *rl; | |
884 | ||
885 | spin_lock_irq(q->queue_lock); | |
886 | q->nr_requests = nr; | |
887 | blk_queue_congestion_threshold(q); | |
888 | ||
889 | /* congestion isn't cgroup aware and follows root blkcg for now */ | |
890 | rl = &q->root_rl; | |
891 | ||
892 | if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q)) | |
893 | blk_set_queue_congested(q, BLK_RW_SYNC); | |
894 | else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q)) | |
895 | blk_clear_queue_congested(q, BLK_RW_SYNC); | |
896 | ||
897 | if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q)) | |
898 | blk_set_queue_congested(q, BLK_RW_ASYNC); | |
899 | else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q)) | |
900 | blk_clear_queue_congested(q, BLK_RW_ASYNC); | |
901 | ||
902 | blk_queue_for_each_rl(rl, q) { | |
903 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { | |
904 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
905 | } else { | |
906 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
907 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
908 | } | |
909 | ||
910 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
911 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
912 | } else { | |
913 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
914 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
915 | } | |
916 | } | |
917 | ||
918 | spin_unlock_irq(q->queue_lock); | |
919 | return 0; | |
920 | } | |
921 | ||
9d5a4e94 MS |
922 | /* |
923 | * Determine if elevator data should be initialized when allocating the | |
924 | * request associated with @bio. | |
925 | */ | |
926 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
927 | { | |
928 | if (!bio) | |
929 | return true; | |
930 | ||
931 | /* | |
932 | * Flush requests do not use the elevator so skip initialization. | |
933 | * This allows a request to share the flush and elevator data. | |
934 | */ | |
935 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
936 | return false; | |
937 | ||
938 | return true; | |
939 | } | |
940 | ||
852c788f TH |
941 | /** |
942 | * rq_ioc - determine io_context for request allocation | |
943 | * @bio: request being allocated is for this bio (can be %NULL) | |
944 | * | |
945 | * Determine io_context to use for request allocation for @bio. May return | |
946 | * %NULL if %current->io_context doesn't exist. | |
947 | */ | |
948 | static struct io_context *rq_ioc(struct bio *bio) | |
949 | { | |
950 | #ifdef CONFIG_BLK_CGROUP | |
951 | if (bio && bio->bi_ioc) | |
952 | return bio->bi_ioc; | |
953 | #endif | |
954 | return current->io_context; | |
955 | } | |
956 | ||
da8303c6 | 957 | /** |
a06e05e6 | 958 | * __get_request - get a free request |
5b788ce3 | 959 | * @rl: request list to allocate from |
da8303c6 TH |
960 | * @rw_flags: RW and SYNC flags |
961 | * @bio: bio to allocate request for (can be %NULL) | |
962 | * @gfp_mask: allocation mask | |
963 | * | |
964 | * Get a free request from @q. This function may fail under memory | |
965 | * pressure or if @q is dead. | |
966 | * | |
da3dae54 | 967 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
968 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
969 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 970 | */ |
5b788ce3 | 971 | static struct request *__get_request(struct request_list *rl, int rw_flags, |
a06e05e6 | 972 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 973 | { |
5b788ce3 | 974 | struct request_queue *q = rl->q; |
b679281a | 975 | struct request *rq; |
7f4b35d1 TH |
976 | struct elevator_type *et = q->elevator->type; |
977 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 978 | struct io_cq *icq = NULL; |
1faa16d2 | 979 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
75eb6c37 | 980 | int may_queue; |
88ee5ef1 | 981 | |
3f3299d5 | 982 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 983 | return ERR_PTR(-ENODEV); |
da8303c6 | 984 | |
7749a8d4 | 985 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
986 | if (may_queue == ELV_MQUEUE_NO) |
987 | goto rq_starved; | |
988 | ||
1faa16d2 JA |
989 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
990 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
991 | /* |
992 | * The queue will fill after this allocation, so set | |
993 | * it as full, and mark this process as "batching". | |
994 | * This process will be allowed to complete a batch of | |
995 | * requests, others will be blocked. | |
996 | */ | |
5b788ce3 | 997 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 998 | ioc_set_batching(q, ioc); |
5b788ce3 | 999 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1000 | } else { |
1001 | if (may_queue != ELV_MQUEUE_MUST | |
1002 | && !ioc_batching(q, ioc)) { | |
1003 | /* | |
1004 | * The queue is full and the allocating | |
1005 | * process is not a "batcher", and not | |
1006 | * exempted by the IO scheduler | |
1007 | */ | |
a492f075 | 1008 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1009 | } |
1010 | } | |
1da177e4 | 1011 | } |
a051661c TH |
1012 | /* |
1013 | * bdi isn't aware of blkcg yet. As all async IOs end up | |
1014 | * root blkcg anyway, just use root blkcg state. | |
1015 | */ | |
1016 | if (rl == &q->root_rl) | |
1017 | blk_set_queue_congested(q, is_sync); | |
1da177e4 LT |
1018 | } |
1019 | ||
082cf69e JA |
1020 | /* |
1021 | * Only allow batching queuers to allocate up to 50% over the defined | |
1022 | * limit of requests, otherwise we could have thousands of requests | |
1023 | * allocated with any setting of ->nr_requests | |
1024 | */ | |
1faa16d2 | 1025 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1026 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1027 | |
8a5ecdd4 | 1028 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1029 | rl->count[is_sync]++; |
1030 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1031 | |
f1f8cc94 TH |
1032 | /* |
1033 | * Decide whether the new request will be managed by elevator. If | |
1034 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will | |
1035 | * prevent the current elevator from being destroyed until the new | |
1036 | * request is freed. This guarantees icq's won't be destroyed and | |
1037 | * makes creating new ones safe. | |
1038 | * | |
1039 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
1040 | * it will be created after releasing queue_lock. | |
1041 | */ | |
d732580b | 1042 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
75eb6c37 | 1043 | rw_flags |= REQ_ELVPRIV; |
8a5ecdd4 | 1044 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1045 | if (et->icq_cache && ioc) |
1046 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1047 | } |
cb98fc8b | 1048 | |
f253b86b JA |
1049 | if (blk_queue_io_stat(q)) |
1050 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
1051 | spin_unlock_irq(q->queue_lock); |
1052 | ||
29e2b09a | 1053 | /* allocate and init request */ |
5b788ce3 | 1054 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1055 | if (!rq) |
b679281a | 1056 | goto fail_alloc; |
1da177e4 | 1057 | |
29e2b09a | 1058 | blk_rq_init(q, rq); |
a051661c | 1059 | blk_rq_set_rl(rq, rl); |
29e2b09a TH |
1060 | rq->cmd_flags = rw_flags | REQ_ALLOCED; |
1061 | ||
aaf7c680 | 1062 | /* init elvpriv */ |
29e2b09a | 1063 | if (rw_flags & REQ_ELVPRIV) { |
aaf7c680 | 1064 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1065 | if (ioc) |
1066 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1067 | if (!icq) |
1068 | goto fail_elvpriv; | |
29e2b09a | 1069 | } |
aaf7c680 TH |
1070 | |
1071 | rq->elv.icq = icq; | |
1072 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1073 | goto fail_elvpriv; | |
1074 | ||
1075 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1076 | if (icq) |
1077 | get_io_context(icq->ioc); | |
1078 | } | |
aaf7c680 | 1079 | out: |
88ee5ef1 JA |
1080 | /* |
1081 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1082 | * OK, if the queue is under the request limit then requests need | |
1083 | * not count toward the nr_batch_requests limit. There will always | |
1084 | * be some limit enforced by BLK_BATCH_TIME. | |
1085 | */ | |
1da177e4 LT |
1086 | if (ioc_batching(q, ioc)) |
1087 | ioc->nr_batch_requests--; | |
6728cb0e | 1088 | |
1faa16d2 | 1089 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 1090 | return rq; |
b679281a | 1091 | |
aaf7c680 TH |
1092 | fail_elvpriv: |
1093 | /* | |
1094 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1095 | * and may fail indefinitely under memory pressure and thus | |
1096 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1097 | * disturb iosched and blkcg but weird is bettern than dead. | |
1098 | */ | |
7b2b10e0 RE |
1099 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
1100 | __func__, dev_name(q->backing_dev_info.dev)); | |
aaf7c680 TH |
1101 | |
1102 | rq->cmd_flags &= ~REQ_ELVPRIV; | |
1103 | rq->elv.icq = NULL; | |
1104 | ||
1105 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1106 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1107 | spin_unlock_irq(q->queue_lock); |
1108 | goto out; | |
1109 | ||
b679281a TH |
1110 | fail_alloc: |
1111 | /* | |
1112 | * Allocation failed presumably due to memory. Undo anything we | |
1113 | * might have messed up. | |
1114 | * | |
1115 | * Allocating task should really be put onto the front of the wait | |
1116 | * queue, but this is pretty rare. | |
1117 | */ | |
1118 | spin_lock_irq(q->queue_lock); | |
5b788ce3 | 1119 | freed_request(rl, rw_flags); |
b679281a TH |
1120 | |
1121 | /* | |
1122 | * in the very unlikely event that allocation failed and no | |
1123 | * requests for this direction was pending, mark us starved so that | |
1124 | * freeing of a request in the other direction will notice | |
1125 | * us. another possible fix would be to split the rq mempool into | |
1126 | * READ and WRITE | |
1127 | */ | |
1128 | rq_starved: | |
1129 | if (unlikely(rl->count[is_sync] == 0)) | |
1130 | rl->starved[is_sync] = 1; | |
a492f075 | 1131 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1132 | } |
1133 | ||
da8303c6 | 1134 | /** |
a06e05e6 | 1135 | * get_request - get a free request |
da8303c6 TH |
1136 | * @q: request_queue to allocate request from |
1137 | * @rw_flags: RW and SYNC flags | |
1138 | * @bio: bio to allocate request for (can be %NULL) | |
a06e05e6 | 1139 | * @gfp_mask: allocation mask |
da8303c6 | 1140 | * |
a06e05e6 TH |
1141 | * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this |
1142 | * function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1143 | * |
da3dae54 | 1144 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1145 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1146 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1147 | */ |
a06e05e6 TH |
1148 | static struct request *get_request(struct request_queue *q, int rw_flags, |
1149 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1150 | { |
1faa16d2 | 1151 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
a06e05e6 | 1152 | DEFINE_WAIT(wait); |
a051661c | 1153 | struct request_list *rl; |
1da177e4 | 1154 | struct request *rq; |
a051661c TH |
1155 | |
1156 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
a06e05e6 | 1157 | retry: |
a051661c | 1158 | rq = __get_request(rl, rw_flags, bio, gfp_mask); |
a492f075 | 1159 | if (!IS_ERR(rq)) |
a06e05e6 | 1160 | return rq; |
1da177e4 | 1161 | |
3f3299d5 | 1162 | if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) { |
a051661c | 1163 | blk_put_rl(rl); |
a492f075 | 1164 | return rq; |
a051661c | 1165 | } |
1da177e4 | 1166 | |
a06e05e6 TH |
1167 | /* wait on @rl and retry */ |
1168 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1169 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1170 | |
a06e05e6 | 1171 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 1172 | |
a06e05e6 TH |
1173 | spin_unlock_irq(q->queue_lock); |
1174 | io_schedule(); | |
d6344532 | 1175 | |
a06e05e6 TH |
1176 | /* |
1177 | * After sleeping, we become a "batching" process and will be able | |
1178 | * to allocate at least one request, and up to a big batch of them | |
1179 | * for a small period time. See ioc_batching, ioc_set_batching | |
1180 | */ | |
a06e05e6 | 1181 | ioc_set_batching(q, current->io_context); |
05caf8db | 1182 | |
a06e05e6 TH |
1183 | spin_lock_irq(q->queue_lock); |
1184 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1185 | |
a06e05e6 | 1186 | goto retry; |
1da177e4 LT |
1187 | } |
1188 | ||
320ae51f JA |
1189 | static struct request *blk_old_get_request(struct request_queue *q, int rw, |
1190 | gfp_t gfp_mask) | |
1da177e4 LT |
1191 | { |
1192 | struct request *rq; | |
1193 | ||
1194 | BUG_ON(rw != READ && rw != WRITE); | |
1195 | ||
7f4b35d1 TH |
1196 | /* create ioc upfront */ |
1197 | create_io_context(gfp_mask, q->node); | |
1198 | ||
d6344532 | 1199 | spin_lock_irq(q->queue_lock); |
a06e05e6 | 1200 | rq = get_request(q, rw, NULL, gfp_mask); |
a492f075 | 1201 | if (IS_ERR(rq)) |
da8303c6 | 1202 | spin_unlock_irq(q->queue_lock); |
d6344532 | 1203 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
1204 | |
1205 | return rq; | |
1206 | } | |
320ae51f JA |
1207 | |
1208 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
1209 | { | |
1210 | if (q->mq_ops) | |
4ce01dd1 | 1211 | return blk_mq_alloc_request(q, rw, gfp_mask, false); |
320ae51f JA |
1212 | else |
1213 | return blk_old_get_request(q, rw, gfp_mask); | |
1214 | } | |
1da177e4 LT |
1215 | EXPORT_SYMBOL(blk_get_request); |
1216 | ||
dc72ef4a | 1217 | /** |
79eb63e9 | 1218 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 1219 | * @q: target request queue |
79eb63e9 BH |
1220 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
1221 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 1222 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 1223 | * |
79eb63e9 BH |
1224 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
1225 | * type commands. Where the struct request needs to be farther initialized by | |
1226 | * the caller. It is passed a &struct bio, which describes the memory info of | |
1227 | * the I/O transfer. | |
dc72ef4a | 1228 | * |
79eb63e9 BH |
1229 | * The caller of blk_make_request must make sure that bi_io_vec |
1230 | * are set to describe the memory buffers. That bio_data_dir() will return | |
1231 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
1232 | * are properly set accordingly) | |
1233 | * | |
1234 | * If called under none-sleepable conditions, mapped bio buffers must not | |
1235 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
1236 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
1237 | * BUG. | |
53674ac5 JA |
1238 | * |
1239 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
1240 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
1241 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
1242 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
1243 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
1244 | * of bio_alloc(), as that avoids the mempool deadlock. | |
1245 | * If possible a big IO should be split into smaller parts when allocation | |
1246 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 1247 | */ |
79eb63e9 BH |
1248 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
1249 | gfp_t gfp_mask) | |
dc72ef4a | 1250 | { |
79eb63e9 BH |
1251 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1252 | ||
a492f075 JL |
1253 | if (IS_ERR(rq)) |
1254 | return rq; | |
79eb63e9 | 1255 | |
f27b087b JA |
1256 | blk_rq_set_block_pc(rq); |
1257 | ||
79eb63e9 BH |
1258 | for_each_bio(bio) { |
1259 | struct bio *bounce_bio = bio; | |
1260 | int ret; | |
1261 | ||
1262 | blk_queue_bounce(q, &bounce_bio); | |
1263 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1264 | if (unlikely(ret)) { | |
1265 | blk_put_request(rq); | |
1266 | return ERR_PTR(ret); | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | return rq; | |
dc72ef4a | 1271 | } |
79eb63e9 | 1272 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1273 | |
f27b087b | 1274 | /** |
da3dae54 | 1275 | * blk_rq_set_block_pc - initialize a request to type BLOCK_PC |
f27b087b JA |
1276 | * @rq: request to be initialized |
1277 | * | |
1278 | */ | |
1279 | void blk_rq_set_block_pc(struct request *rq) | |
1280 | { | |
1281 | rq->cmd_type = REQ_TYPE_BLOCK_PC; | |
1282 | rq->__data_len = 0; | |
1283 | rq->__sector = (sector_t) -1; | |
1284 | rq->bio = rq->biotail = NULL; | |
1285 | memset(rq->__cmd, 0, sizeof(rq->__cmd)); | |
f27b087b JA |
1286 | } |
1287 | EXPORT_SYMBOL(blk_rq_set_block_pc); | |
1288 | ||
1da177e4 LT |
1289 | /** |
1290 | * blk_requeue_request - put a request back on queue | |
1291 | * @q: request queue where request should be inserted | |
1292 | * @rq: request to be inserted | |
1293 | * | |
1294 | * Description: | |
1295 | * Drivers often keep queueing requests until the hardware cannot accept | |
1296 | * more, when that condition happens we need to put the request back | |
1297 | * on the queue. Must be called with queue lock held. | |
1298 | */ | |
165125e1 | 1299 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1300 | { |
242f9dcb JA |
1301 | blk_delete_timer(rq); |
1302 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1303 | trace_block_rq_requeue(q, rq); |
2056a782 | 1304 | |
125c99bc | 1305 | if (rq->cmd_flags & REQ_QUEUED) |
1da177e4 LT |
1306 | blk_queue_end_tag(q, rq); |
1307 | ||
ba396a6c JB |
1308 | BUG_ON(blk_queued_rq(rq)); |
1309 | ||
1da177e4 LT |
1310 | elv_requeue_request(q, rq); |
1311 | } | |
1da177e4 LT |
1312 | EXPORT_SYMBOL(blk_requeue_request); |
1313 | ||
73c10101 JA |
1314 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1315 | int where) | |
1316 | { | |
320ae51f | 1317 | blk_account_io_start(rq, true); |
7eaceacc | 1318 | __elv_add_request(q, rq, where); |
73c10101 JA |
1319 | } |
1320 | ||
074a7aca TH |
1321 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1322 | unsigned long now) | |
1323 | { | |
7276d02e JA |
1324 | int inflight; |
1325 | ||
074a7aca TH |
1326 | if (now == part->stamp) |
1327 | return; | |
1328 | ||
7276d02e JA |
1329 | inflight = part_in_flight(part); |
1330 | if (inflight) { | |
074a7aca | 1331 | __part_stat_add(cpu, part, time_in_queue, |
7276d02e | 1332 | inflight * (now - part->stamp)); |
074a7aca TH |
1333 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1334 | } | |
1335 | part->stamp = now; | |
1336 | } | |
1337 | ||
1338 | /** | |
496aa8a9 RD |
1339 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1340 | * @cpu: cpu number for stats access | |
1341 | * @part: target partition | |
1da177e4 LT |
1342 | * |
1343 | * The average IO queue length and utilisation statistics are maintained | |
1344 | * by observing the current state of the queue length and the amount of | |
1345 | * time it has been in this state for. | |
1346 | * | |
1347 | * Normally, that accounting is done on IO completion, but that can result | |
1348 | * in more than a second's worth of IO being accounted for within any one | |
1349 | * second, leading to >100% utilisation. To deal with that, we call this | |
1350 | * function to do a round-off before returning the results when reading | |
1351 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1352 | * the current jiffies and restarts the counters again. | |
1353 | */ | |
c9959059 | 1354 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1355 | { |
1356 | unsigned long now = jiffies; | |
1357 | ||
074a7aca TH |
1358 | if (part->partno) |
1359 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1360 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1361 | } |
074a7aca | 1362 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1363 | |
47fafbc7 | 1364 | #ifdef CONFIG_PM |
c8158819 LM |
1365 | static void blk_pm_put_request(struct request *rq) |
1366 | { | |
1367 | if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending) | |
1368 | pm_runtime_mark_last_busy(rq->q->dev); | |
1369 | } | |
1370 | #else | |
1371 | static inline void blk_pm_put_request(struct request *rq) {} | |
1372 | #endif | |
1373 | ||
1da177e4 LT |
1374 | /* |
1375 | * queue lock must be held | |
1376 | */ | |
165125e1 | 1377 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1378 | { |
1da177e4 LT |
1379 | if (unlikely(!q)) |
1380 | return; | |
1da177e4 | 1381 | |
6f5ba581 CH |
1382 | if (q->mq_ops) { |
1383 | blk_mq_free_request(req); | |
1384 | return; | |
1385 | } | |
1386 | ||
c8158819 LM |
1387 | blk_pm_put_request(req); |
1388 | ||
8922e16c TH |
1389 | elv_completed_request(q, req); |
1390 | ||
1cd96c24 BH |
1391 | /* this is a bio leak */ |
1392 | WARN_ON(req->bio != NULL); | |
1393 | ||
1da177e4 LT |
1394 | /* |
1395 | * Request may not have originated from ll_rw_blk. if not, | |
1396 | * it didn't come out of our reserved rq pools | |
1397 | */ | |
49171e5c | 1398 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1399 | unsigned int flags = req->cmd_flags; |
a051661c | 1400 | struct request_list *rl = blk_rq_rl(req); |
1da177e4 | 1401 | |
1da177e4 | 1402 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1403 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1404 | |
a051661c TH |
1405 | blk_free_request(rl, req); |
1406 | freed_request(rl, flags); | |
1407 | blk_put_rl(rl); | |
1da177e4 LT |
1408 | } |
1409 | } | |
6e39b69e MC |
1410 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1411 | ||
1da177e4 LT |
1412 | void blk_put_request(struct request *req) |
1413 | { | |
165125e1 | 1414 | struct request_queue *q = req->q; |
8922e16c | 1415 | |
320ae51f JA |
1416 | if (q->mq_ops) |
1417 | blk_mq_free_request(req); | |
1418 | else { | |
1419 | unsigned long flags; | |
1420 | ||
1421 | spin_lock_irqsave(q->queue_lock, flags); | |
1422 | __blk_put_request(q, req); | |
1423 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1424 | } | |
1da177e4 | 1425 | } |
1da177e4 LT |
1426 | EXPORT_SYMBOL(blk_put_request); |
1427 | ||
66ac0280 CH |
1428 | /** |
1429 | * blk_add_request_payload - add a payload to a request | |
1430 | * @rq: request to update | |
1431 | * @page: page backing the payload | |
1432 | * @len: length of the payload. | |
1433 | * | |
1434 | * This allows to later add a payload to an already submitted request by | |
1435 | * a block driver. The driver needs to take care of freeing the payload | |
1436 | * itself. | |
1437 | * | |
1438 | * Note that this is a quite horrible hack and nothing but handling of | |
1439 | * discard requests should ever use it. | |
1440 | */ | |
1441 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1442 | unsigned int len) | |
1443 | { | |
1444 | struct bio *bio = rq->bio; | |
1445 | ||
1446 | bio->bi_io_vec->bv_page = page; | |
1447 | bio->bi_io_vec->bv_offset = 0; | |
1448 | bio->bi_io_vec->bv_len = len; | |
1449 | ||
4f024f37 | 1450 | bio->bi_iter.bi_size = len; |
66ac0280 CH |
1451 | bio->bi_vcnt = 1; |
1452 | bio->bi_phys_segments = 1; | |
1453 | ||
1454 | rq->__data_len = rq->resid_len = len; | |
1455 | rq->nr_phys_segments = 1; | |
66ac0280 CH |
1456 | } |
1457 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1458 | ||
320ae51f JA |
1459 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1460 | struct bio *bio) | |
73c10101 JA |
1461 | { |
1462 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1463 | ||
73c10101 JA |
1464 | if (!ll_back_merge_fn(q, req, bio)) |
1465 | return false; | |
1466 | ||
8c1cf6bb | 1467 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1468 | |
1469 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1470 | blk_rq_set_mixed_merge(req); | |
1471 | ||
1472 | req->biotail->bi_next = bio; | |
1473 | req->biotail = bio; | |
4f024f37 | 1474 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1475 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1476 | ||
320ae51f | 1477 | blk_account_io_start(req, false); |
73c10101 JA |
1478 | return true; |
1479 | } | |
1480 | ||
320ae51f JA |
1481 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1482 | struct bio *bio) | |
73c10101 JA |
1483 | { |
1484 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1485 | |
73c10101 JA |
1486 | if (!ll_front_merge_fn(q, req, bio)) |
1487 | return false; | |
1488 | ||
8c1cf6bb | 1489 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1490 | |
1491 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1492 | blk_rq_set_mixed_merge(req); | |
1493 | ||
73c10101 JA |
1494 | bio->bi_next = req->bio; |
1495 | req->bio = bio; | |
1496 | ||
4f024f37 KO |
1497 | req->__sector = bio->bi_iter.bi_sector; |
1498 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1499 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1500 | ||
320ae51f | 1501 | blk_account_io_start(req, false); |
73c10101 JA |
1502 | return true; |
1503 | } | |
1504 | ||
bd87b589 | 1505 | /** |
320ae51f | 1506 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1507 | * @q: request_queue new bio is being queued at |
1508 | * @bio: new bio being queued | |
1509 | * @request_count: out parameter for number of traversed plugged requests | |
1510 | * | |
1511 | * Determine whether @bio being queued on @q can be merged with a request | |
1512 | * on %current's plugged list. Returns %true if merge was successful, | |
1513 | * otherwise %false. | |
1514 | * | |
07c2bd37 TH |
1515 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1516 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1517 | * than scheduling, and the request, while may have elvpriv data, is not | |
1518 | * added on the elevator at this point. In addition, we don't have | |
1519 | * reliable access to the elevator outside queue lock. Only check basic | |
1520 | * merging parameters without querying the elevator. | |
da41a589 RE |
1521 | * |
1522 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1523 | */ |
320ae51f | 1524 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1525 | unsigned int *request_count, |
1526 | struct request **same_queue_rq) | |
73c10101 JA |
1527 | { |
1528 | struct blk_plug *plug; | |
1529 | struct request *rq; | |
1530 | bool ret = false; | |
92f399c7 | 1531 | struct list_head *plug_list; |
73c10101 | 1532 | |
bd87b589 | 1533 | plug = current->plug; |
73c10101 JA |
1534 | if (!plug) |
1535 | goto out; | |
56ebdaf2 | 1536 | *request_count = 0; |
73c10101 | 1537 | |
92f399c7 SL |
1538 | if (q->mq_ops) |
1539 | plug_list = &plug->mq_list; | |
1540 | else | |
1541 | plug_list = &plug->list; | |
1542 | ||
1543 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
73c10101 JA |
1544 | int el_ret; |
1545 | ||
5b3f341f | 1546 | if (rq->q == q) { |
1b2e19f1 | 1547 | (*request_count)++; |
5b3f341f SL |
1548 | /* |
1549 | * Only blk-mq multiple hardware queues case checks the | |
1550 | * rq in the same queue, there should be only one such | |
1551 | * rq in a queue | |
1552 | **/ | |
1553 | if (same_queue_rq) | |
1554 | *same_queue_rq = rq; | |
1555 | } | |
56ebdaf2 | 1556 | |
07c2bd37 | 1557 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1558 | continue; |
1559 | ||
050c8ea8 | 1560 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1561 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1562 | ret = bio_attempt_back_merge(q, rq, bio); | |
1563 | if (ret) | |
1564 | break; | |
1565 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1566 | ret = bio_attempt_front_merge(q, rq, bio); | |
1567 | if (ret) | |
1568 | break; | |
1569 | } | |
1570 | } | |
1571 | out: | |
1572 | return ret; | |
1573 | } | |
1574 | ||
86db1e29 | 1575 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1576 | { |
4aff5e23 | 1577 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1578 | |
7b6d91da CH |
1579 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1580 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1581 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1582 | |
52d9e675 | 1583 | req->errors = 0; |
4f024f37 | 1584 | req->__sector = bio->bi_iter.bi_sector; |
52d9e675 | 1585 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1586 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1587 | } |
1588 | ||
5a7bbad2 | 1589 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1590 | { |
5e00d1b5 | 1591 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1592 | struct blk_plug *plug; |
1593 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1594 | struct request *req; | |
56ebdaf2 | 1595 | unsigned int request_count = 0; |
1da177e4 | 1596 | |
1da177e4 LT |
1597 | /* |
1598 | * low level driver can indicate that it wants pages above a | |
1599 | * certain limit bounced to low memory (ie for highmem, or even | |
1600 | * ISA dma in theory) | |
1601 | */ | |
1602 | blk_queue_bounce(q, &bio); | |
1603 | ||
ffecfd1a DW |
1604 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) { |
1605 | bio_endio(bio, -EIO); | |
1606 | return; | |
1607 | } | |
1608 | ||
4fed947c | 1609 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1610 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1611 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1612 | goto get_rq; |
1613 | } | |
1614 | ||
73c10101 JA |
1615 | /* |
1616 | * Check if we can merge with the plugged list before grabbing | |
1617 | * any locks. | |
1618 | */ | |
da41a589 | 1619 | if (!blk_queue_nomerges(q) && |
5b3f341f | 1620 | blk_attempt_plug_merge(q, bio, &request_count, NULL)) |
5a7bbad2 | 1621 | return; |
1da177e4 | 1622 | |
73c10101 | 1623 | spin_lock_irq(q->queue_lock); |
2056a782 | 1624 | |
73c10101 JA |
1625 | el_ret = elv_merge(q, &req, bio); |
1626 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1627 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1628 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1629 | if (!attempt_back_merge(q, req)) |
1630 | elv_merged_request(q, req, el_ret); | |
1631 | goto out_unlock; | |
1632 | } | |
1633 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1634 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1635 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1636 | if (!attempt_front_merge(q, req)) |
1637 | elv_merged_request(q, req, el_ret); | |
1638 | goto out_unlock; | |
80a761fd | 1639 | } |
1da177e4 LT |
1640 | } |
1641 | ||
450991bc | 1642 | get_rq: |
7749a8d4 JA |
1643 | /* |
1644 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1645 | * but we need to set it earlier to expose the sync flag to the | |
1646 | * rq allocator and io schedulers. | |
1647 | */ | |
1648 | rw_flags = bio_data_dir(bio); | |
1649 | if (sync) | |
7b6d91da | 1650 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1651 | |
1da177e4 | 1652 | /* |
450991bc | 1653 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1654 | * Returns with the queue unlocked. |
450991bc | 1655 | */ |
a06e05e6 | 1656 | req = get_request(q, rw_flags, bio, GFP_NOIO); |
a492f075 JL |
1657 | if (IS_ERR(req)) { |
1658 | bio_endio(bio, PTR_ERR(req)); /* @q is dead */ | |
da8303c6 TH |
1659 | goto out_unlock; |
1660 | } | |
d6344532 | 1661 | |
450991bc NP |
1662 | /* |
1663 | * After dropping the lock and possibly sleeping here, our request | |
1664 | * may now be mergeable after it had proven unmergeable (above). | |
1665 | * We don't worry about that case for efficiency. It won't happen | |
1666 | * often, and the elevators are able to handle it. | |
1da177e4 | 1667 | */ |
52d9e675 | 1668 | init_request_from_bio(req, bio); |
1da177e4 | 1669 | |
9562ad9a | 1670 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1671 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1672 | |
1673 | plug = current->plug; | |
721a9602 | 1674 | if (plug) { |
dc6d36c9 JA |
1675 | /* |
1676 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1677 | * of a plug trace. |
dc6d36c9 | 1678 | */ |
7aef2e78 | 1679 | if (!request_count) |
dc6d36c9 | 1680 | trace_block_plug(q); |
3540d5e8 | 1681 | else { |
019ceb7d | 1682 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
3540d5e8 | 1683 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1684 | trace_block_plug(q); |
1685 | } | |
73c10101 | 1686 | } |
73c10101 | 1687 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 1688 | blk_account_io_start(req, true); |
73c10101 JA |
1689 | } else { |
1690 | spin_lock_irq(q->queue_lock); | |
1691 | add_acct_request(q, req, where); | |
24ecfbe2 | 1692 | __blk_run_queue(q); |
73c10101 JA |
1693 | out_unlock: |
1694 | spin_unlock_irq(q->queue_lock); | |
1695 | } | |
1da177e4 | 1696 | } |
c20e8de2 | 1697 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1698 | |
1699 | /* | |
1700 | * If bio->bi_dev is a partition, remap the location | |
1701 | */ | |
1702 | static inline void blk_partition_remap(struct bio *bio) | |
1703 | { | |
1704 | struct block_device *bdev = bio->bi_bdev; | |
1705 | ||
bf2de6f5 | 1706 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1707 | struct hd_struct *p = bdev->bd_part; |
1708 | ||
4f024f37 | 1709 | bio->bi_iter.bi_sector += p->start_sect; |
1da177e4 | 1710 | bio->bi_bdev = bdev->bd_contains; |
c7149d6b | 1711 | |
d07335e5 MS |
1712 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1713 | bdev->bd_dev, | |
4f024f37 | 1714 | bio->bi_iter.bi_sector - p->start_sect); |
1da177e4 LT |
1715 | } |
1716 | } | |
1717 | ||
1da177e4 LT |
1718 | static void handle_bad_sector(struct bio *bio) |
1719 | { | |
1720 | char b[BDEVNAME_SIZE]; | |
1721 | ||
1722 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1723 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1724 | bdevname(bio->bi_bdev, b), | |
1725 | bio->bi_rw, | |
f73a1c7d | 1726 | (unsigned long long)bio_end_sector(bio), |
77304d2a | 1727 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1728 | } |
1729 | ||
c17bb495 AM |
1730 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1731 | ||
1732 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1733 | ||
1734 | static int __init setup_fail_make_request(char *str) | |
1735 | { | |
1736 | return setup_fault_attr(&fail_make_request, str); | |
1737 | } | |
1738 | __setup("fail_make_request=", setup_fail_make_request); | |
1739 | ||
b2c9cd37 | 1740 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1741 | { |
b2c9cd37 | 1742 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1743 | } |
1744 | ||
1745 | static int __init fail_make_request_debugfs(void) | |
1746 | { | |
dd48c085 AM |
1747 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1748 | NULL, &fail_make_request); | |
1749 | ||
21f9fcd8 | 1750 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
1751 | } |
1752 | ||
1753 | late_initcall(fail_make_request_debugfs); | |
1754 | ||
1755 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1756 | ||
b2c9cd37 AM |
1757 | static inline bool should_fail_request(struct hd_struct *part, |
1758 | unsigned int bytes) | |
c17bb495 | 1759 | { |
b2c9cd37 | 1760 | return false; |
c17bb495 AM |
1761 | } |
1762 | ||
1763 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1764 | ||
c07e2b41 JA |
1765 | /* |
1766 | * Check whether this bio extends beyond the end of the device. | |
1767 | */ | |
1768 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1769 | { | |
1770 | sector_t maxsector; | |
1771 | ||
1772 | if (!nr_sectors) | |
1773 | return 0; | |
1774 | ||
1775 | /* Test device or partition size, when known. */ | |
77304d2a | 1776 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 | 1777 | if (maxsector) { |
4f024f37 | 1778 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
1779 | |
1780 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1781 | /* | |
1782 | * This may well happen - the kernel calls bread() | |
1783 | * without checking the size of the device, e.g., when | |
1784 | * mounting a device. | |
1785 | */ | |
1786 | handle_bad_sector(bio); | |
1787 | return 1; | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | return 0; | |
1792 | } | |
1793 | ||
27a84d54 CH |
1794 | static noinline_for_stack bool |
1795 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1796 | { |
165125e1 | 1797 | struct request_queue *q; |
5a7bbad2 | 1798 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1799 | int err = -EIO; |
5a7bbad2 CH |
1800 | char b[BDEVNAME_SIZE]; |
1801 | struct hd_struct *part; | |
1da177e4 LT |
1802 | |
1803 | might_sleep(); | |
1da177e4 | 1804 | |
c07e2b41 JA |
1805 | if (bio_check_eod(bio, nr_sectors)) |
1806 | goto end_io; | |
1da177e4 | 1807 | |
5a7bbad2 CH |
1808 | q = bdev_get_queue(bio->bi_bdev); |
1809 | if (unlikely(!q)) { | |
1810 | printk(KERN_ERR | |
1811 | "generic_make_request: Trying to access " | |
1812 | "nonexistent block-device %s (%Lu)\n", | |
1813 | bdevname(bio->bi_bdev, b), | |
4f024f37 | 1814 | (long long) bio->bi_iter.bi_sector); |
5a7bbad2 CH |
1815 | goto end_io; |
1816 | } | |
c17bb495 | 1817 | |
e2a60da7 MP |
1818 | if (likely(bio_is_rw(bio) && |
1819 | nr_sectors > queue_max_hw_sectors(q))) { | |
5a7bbad2 CH |
1820 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1821 | bdevname(bio->bi_bdev, b), | |
1822 | bio_sectors(bio), | |
1823 | queue_max_hw_sectors(q)); | |
1824 | goto end_io; | |
1825 | } | |
1da177e4 | 1826 | |
5a7bbad2 | 1827 | part = bio->bi_bdev->bd_part; |
4f024f37 | 1828 | if (should_fail_request(part, bio->bi_iter.bi_size) || |
5a7bbad2 | 1829 | should_fail_request(&part_to_disk(part)->part0, |
4f024f37 | 1830 | bio->bi_iter.bi_size)) |
5a7bbad2 | 1831 | goto end_io; |
2056a782 | 1832 | |
5a7bbad2 CH |
1833 | /* |
1834 | * If this device has partitions, remap block n | |
1835 | * of partition p to block n+start(p) of the disk. | |
1836 | */ | |
1837 | blk_partition_remap(bio); | |
2056a782 | 1838 | |
5a7bbad2 CH |
1839 | if (bio_check_eod(bio, nr_sectors)) |
1840 | goto end_io; | |
1e87901e | 1841 | |
5a7bbad2 CH |
1842 | /* |
1843 | * Filter flush bio's early so that make_request based | |
1844 | * drivers without flush support don't have to worry | |
1845 | * about them. | |
1846 | */ | |
1847 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1848 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1849 | if (!nr_sectors) { | |
1850 | err = 0; | |
51fd77bd JA |
1851 | goto end_io; |
1852 | } | |
5a7bbad2 | 1853 | } |
5ddfe969 | 1854 | |
5a7bbad2 CH |
1855 | if ((bio->bi_rw & REQ_DISCARD) && |
1856 | (!blk_queue_discard(q) || | |
e2a60da7 | 1857 | ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) { |
5a7bbad2 CH |
1858 | err = -EOPNOTSUPP; |
1859 | goto end_io; | |
1860 | } | |
01edede4 | 1861 | |
4363ac7c | 1862 | if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) { |
5a7bbad2 CH |
1863 | err = -EOPNOTSUPP; |
1864 | goto end_io; | |
1865 | } | |
01edede4 | 1866 | |
7f4b35d1 TH |
1867 | /* |
1868 | * Various block parts want %current->io_context and lazy ioc | |
1869 | * allocation ends up trading a lot of pain for a small amount of | |
1870 | * memory. Just allocate it upfront. This may fail and block | |
1871 | * layer knows how to live with it. | |
1872 | */ | |
1873 | create_io_context(GFP_ATOMIC, q->node); | |
1874 | ||
bc16a4f9 TH |
1875 | if (blk_throtl_bio(q, bio)) |
1876 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1877 | |
5a7bbad2 | 1878 | trace_block_bio_queue(q, bio); |
27a84d54 | 1879 | return true; |
a7384677 TH |
1880 | |
1881 | end_io: | |
1882 | bio_endio(bio, err); | |
27a84d54 | 1883 | return false; |
1da177e4 LT |
1884 | } |
1885 | ||
27a84d54 CH |
1886 | /** |
1887 | * generic_make_request - hand a buffer to its device driver for I/O | |
1888 | * @bio: The bio describing the location in memory and on the device. | |
1889 | * | |
1890 | * generic_make_request() is used to make I/O requests of block | |
1891 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1892 | * to be done. | |
1893 | * | |
1894 | * generic_make_request() does not return any status. The | |
1895 | * success/failure status of the request, along with notification of | |
1896 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1897 | * function described (one day) else where. | |
1898 | * | |
1899 | * The caller of generic_make_request must make sure that bi_io_vec | |
1900 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1901 | * set to describe the device address, and the | |
1902 | * bi_end_io and optionally bi_private are set to describe how | |
1903 | * completion notification should be signaled. | |
1904 | * | |
1905 | * generic_make_request and the drivers it calls may use bi_next if this | |
1906 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1907 | * a lower device by calling into generic_make_request recursively, which | |
1908 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1909 | */ |
1910 | void generic_make_request(struct bio *bio) | |
1911 | { | |
bddd87c7 AM |
1912 | struct bio_list bio_list_on_stack; |
1913 | ||
27a84d54 CH |
1914 | if (!generic_make_request_checks(bio)) |
1915 | return; | |
1916 | ||
1917 | /* | |
1918 | * We only want one ->make_request_fn to be active at a time, else | |
1919 | * stack usage with stacked devices could be a problem. So use | |
1920 | * current->bio_list to keep a list of requests submited by a | |
1921 | * make_request_fn function. current->bio_list is also used as a | |
1922 | * flag to say if generic_make_request is currently active in this | |
1923 | * task or not. If it is NULL, then no make_request is active. If | |
1924 | * it is non-NULL, then a make_request is active, and new requests | |
1925 | * should be added at the tail | |
1926 | */ | |
bddd87c7 | 1927 | if (current->bio_list) { |
bddd87c7 | 1928 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1929 | return; |
1930 | } | |
27a84d54 | 1931 | |
d89d8796 NB |
1932 | /* following loop may be a bit non-obvious, and so deserves some |
1933 | * explanation. | |
1934 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1935 | * ensure that) so we have a list with a single bio. | |
1936 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1937 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1938 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1939 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1940 | * through a recursive call to generic_make_request. If it |
1941 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1942 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1943 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1944 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1945 | */ |
1946 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1947 | bio_list_init(&bio_list_on_stack); |
1948 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1949 | do { |
27a84d54 CH |
1950 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1951 | ||
1952 | q->make_request_fn(q, bio); | |
1953 | ||
bddd87c7 | 1954 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1955 | } while (bio); |
bddd87c7 | 1956 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1957 | } |
1da177e4 LT |
1958 | EXPORT_SYMBOL(generic_make_request); |
1959 | ||
1960 | /** | |
710027a4 | 1961 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1962 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1963 | * @bio: The &struct bio which describes the I/O | |
1964 | * | |
1965 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1966 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1967 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1968 | * |
1969 | */ | |
1970 | void submit_bio(int rw, struct bio *bio) | |
1971 | { | |
22e2c507 | 1972 | bio->bi_rw |= rw; |
1da177e4 | 1973 | |
bf2de6f5 JA |
1974 | /* |
1975 | * If it's a regular read/write or a barrier with data attached, | |
1976 | * go through the normal accounting stuff before submission. | |
1977 | */ | |
e2a60da7 | 1978 | if (bio_has_data(bio)) { |
4363ac7c MP |
1979 | unsigned int count; |
1980 | ||
1981 | if (unlikely(rw & REQ_WRITE_SAME)) | |
1982 | count = bdev_logical_block_size(bio->bi_bdev) >> 9; | |
1983 | else | |
1984 | count = bio_sectors(bio); | |
1985 | ||
bf2de6f5 JA |
1986 | if (rw & WRITE) { |
1987 | count_vm_events(PGPGOUT, count); | |
1988 | } else { | |
4f024f37 | 1989 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
1990 | count_vm_events(PGPGIN, count); |
1991 | } | |
1992 | ||
1993 | if (unlikely(block_dump)) { | |
1994 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1995 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1996 | current->comm, task_pid_nr(current), |
bf2de6f5 | 1997 | (rw & WRITE) ? "WRITE" : "READ", |
4f024f37 | 1998 | (unsigned long long)bio->bi_iter.bi_sector, |
8dcbdc74 SM |
1999 | bdevname(bio->bi_bdev, b), |
2000 | count); | |
bf2de6f5 | 2001 | } |
1da177e4 LT |
2002 | } |
2003 | ||
2004 | generic_make_request(bio); | |
2005 | } | |
1da177e4 LT |
2006 | EXPORT_SYMBOL(submit_bio); |
2007 | ||
82124d60 KU |
2008 | /** |
2009 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
2010 | * @q: the queue | |
2011 | * @rq: the request being checked | |
2012 | * | |
2013 | * Description: | |
2014 | * @rq may have been made based on weaker limitations of upper-level queues | |
2015 | * in request stacking drivers, and it may violate the limitation of @q. | |
2016 | * Since the block layer and the underlying device driver trust @rq | |
2017 | * after it is inserted to @q, it should be checked against @q before | |
2018 | * the insertion using this generic function. | |
2019 | * | |
2020 | * This function should also be useful for request stacking drivers | |
eef35c2d | 2021 | * in some cases below, so export this function. |
82124d60 KU |
2022 | * Request stacking drivers like request-based dm may change the queue |
2023 | * limits while requests are in the queue (e.g. dm's table swapping). | |
e227867f | 2024 | * Such request stacking drivers should check those requests against |
82124d60 KU |
2025 | * the new queue limits again when they dispatch those requests, |
2026 | * although such checkings are also done against the old queue limits | |
2027 | * when submitting requests. | |
2028 | */ | |
2029 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
2030 | { | |
e2a60da7 | 2031 | if (!rq_mergeable(rq)) |
3383977f S |
2032 | return 0; |
2033 | ||
f31dc1cd | 2034 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) { |
82124d60 KU |
2035 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2036 | return -EIO; | |
2037 | } | |
2038 | ||
2039 | /* | |
2040 | * queue's settings related to segment counting like q->bounce_pfn | |
2041 | * may differ from that of other stacking queues. | |
2042 | * Recalculate it to check the request correctly on this queue's | |
2043 | * limitation. | |
2044 | */ | |
2045 | blk_recalc_rq_segments(rq); | |
8a78362c | 2046 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2047 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2048 | return -EIO; | |
2049 | } | |
2050 | ||
2051 | return 0; | |
2052 | } | |
2053 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
2054 | ||
2055 | /** | |
2056 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2057 | * @q: the queue to submit the request | |
2058 | * @rq: the request being queued | |
2059 | */ | |
2060 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
2061 | { | |
2062 | unsigned long flags; | |
4853abaa | 2063 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
2064 | |
2065 | if (blk_rq_check_limits(q, rq)) | |
2066 | return -EIO; | |
2067 | ||
b2c9cd37 AM |
2068 | if (rq->rq_disk && |
2069 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 2070 | return -EIO; |
82124d60 | 2071 | |
7fb4898e KB |
2072 | if (q->mq_ops) { |
2073 | if (blk_queue_io_stat(q)) | |
2074 | blk_account_io_start(rq, true); | |
2075 | blk_mq_insert_request(rq, false, true, true); | |
2076 | return 0; | |
2077 | } | |
2078 | ||
82124d60 | 2079 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2080 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
2081 | spin_unlock_irqrestore(q->queue_lock, flags); |
2082 | return -ENODEV; | |
2083 | } | |
82124d60 KU |
2084 | |
2085 | /* | |
2086 | * Submitting request must be dequeued before calling this function | |
2087 | * because it will be linked to another request_queue | |
2088 | */ | |
2089 | BUG_ON(blk_queued_rq(rq)); | |
2090 | ||
4853abaa JM |
2091 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
2092 | where = ELEVATOR_INSERT_FLUSH; | |
2093 | ||
2094 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2095 | if (where == ELEVATOR_INSERT_FLUSH) |
2096 | __blk_run_queue(q); | |
82124d60 KU |
2097 | spin_unlock_irqrestore(q->queue_lock, flags); |
2098 | ||
2099 | return 0; | |
2100 | } | |
2101 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2102 | ||
80a761fd TH |
2103 | /** |
2104 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2105 | * @rq: request to examine | |
2106 | * | |
2107 | * Description: | |
2108 | * A request could be merge of IOs which require different failure | |
2109 | * handling. This function determines the number of bytes which | |
2110 | * can be failed from the beginning of the request without | |
2111 | * crossing into area which need to be retried further. | |
2112 | * | |
2113 | * Return: | |
2114 | * The number of bytes to fail. | |
2115 | * | |
2116 | * Context: | |
2117 | * queue_lock must be held. | |
2118 | */ | |
2119 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2120 | { | |
2121 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2122 | unsigned int bytes = 0; | |
2123 | struct bio *bio; | |
2124 | ||
2125 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
2126 | return blk_rq_bytes(rq); | |
2127 | ||
2128 | /* | |
2129 | * Currently the only 'mixing' which can happen is between | |
2130 | * different fastfail types. We can safely fail portions | |
2131 | * which have all the failfast bits that the first one has - | |
2132 | * the ones which are at least as eager to fail as the first | |
2133 | * one. | |
2134 | */ | |
2135 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
2136 | if ((bio->bi_rw & ff) != ff) | |
2137 | break; | |
4f024f37 | 2138 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2139 | } |
2140 | ||
2141 | /* this could lead to infinite loop */ | |
2142 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2143 | return bytes; | |
2144 | } | |
2145 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2146 | ||
320ae51f | 2147 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2148 | { |
c2553b58 | 2149 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2150 | const int rw = rq_data_dir(req); |
2151 | struct hd_struct *part; | |
2152 | int cpu; | |
2153 | ||
2154 | cpu = part_stat_lock(); | |
09e099d4 | 2155 | part = req->part; |
bc58ba94 JA |
2156 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2157 | part_stat_unlock(); | |
2158 | } | |
2159 | } | |
2160 | ||
320ae51f | 2161 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2162 | { |
bc58ba94 | 2163 | /* |
dd4c133f TH |
2164 | * Account IO completion. flush_rq isn't accounted as a |
2165 | * normal IO on queueing nor completion. Accounting the | |
2166 | * containing request is enough. | |
bc58ba94 | 2167 | */ |
414b4ff5 | 2168 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
2169 | unsigned long duration = jiffies - req->start_time; |
2170 | const int rw = rq_data_dir(req); | |
2171 | struct hd_struct *part; | |
2172 | int cpu; | |
2173 | ||
2174 | cpu = part_stat_lock(); | |
09e099d4 | 2175 | part = req->part; |
bc58ba94 JA |
2176 | |
2177 | part_stat_inc(cpu, part, ios[rw]); | |
2178 | part_stat_add(cpu, part, ticks[rw], duration); | |
2179 | part_round_stats(cpu, part); | |
316d315b | 2180 | part_dec_in_flight(part, rw); |
bc58ba94 | 2181 | |
6c23a968 | 2182 | hd_struct_put(part); |
bc58ba94 JA |
2183 | part_stat_unlock(); |
2184 | } | |
2185 | } | |
2186 | ||
47fafbc7 | 2187 | #ifdef CONFIG_PM |
c8158819 LM |
2188 | /* |
2189 | * Don't process normal requests when queue is suspended | |
2190 | * or in the process of suspending/resuming | |
2191 | */ | |
2192 | static struct request *blk_pm_peek_request(struct request_queue *q, | |
2193 | struct request *rq) | |
2194 | { | |
2195 | if (q->dev && (q->rpm_status == RPM_SUSPENDED || | |
2196 | (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM)))) | |
2197 | return NULL; | |
2198 | else | |
2199 | return rq; | |
2200 | } | |
2201 | #else | |
2202 | static inline struct request *blk_pm_peek_request(struct request_queue *q, | |
2203 | struct request *rq) | |
2204 | { | |
2205 | return rq; | |
2206 | } | |
2207 | #endif | |
2208 | ||
320ae51f JA |
2209 | void blk_account_io_start(struct request *rq, bool new_io) |
2210 | { | |
2211 | struct hd_struct *part; | |
2212 | int rw = rq_data_dir(rq); | |
2213 | int cpu; | |
2214 | ||
2215 | if (!blk_do_io_stat(rq)) | |
2216 | return; | |
2217 | ||
2218 | cpu = part_stat_lock(); | |
2219 | ||
2220 | if (!new_io) { | |
2221 | part = rq->part; | |
2222 | part_stat_inc(cpu, part, merges[rw]); | |
2223 | } else { | |
2224 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2225 | if (!hd_struct_try_get(part)) { | |
2226 | /* | |
2227 | * The partition is already being removed, | |
2228 | * the request will be accounted on the disk only | |
2229 | * | |
2230 | * We take a reference on disk->part0 although that | |
2231 | * partition will never be deleted, so we can treat | |
2232 | * it as any other partition. | |
2233 | */ | |
2234 | part = &rq->rq_disk->part0; | |
2235 | hd_struct_get(part); | |
2236 | } | |
2237 | part_round_stats(cpu, part); | |
2238 | part_inc_in_flight(part, rw); | |
2239 | rq->part = part; | |
2240 | } | |
2241 | ||
2242 | part_stat_unlock(); | |
2243 | } | |
2244 | ||
3bcddeac | 2245 | /** |
9934c8c0 TH |
2246 | * blk_peek_request - peek at the top of a request queue |
2247 | * @q: request queue to peek at | |
2248 | * | |
2249 | * Description: | |
2250 | * Return the request at the top of @q. The returned request | |
2251 | * should be started using blk_start_request() before LLD starts | |
2252 | * processing it. | |
2253 | * | |
2254 | * Return: | |
2255 | * Pointer to the request at the top of @q if available. Null | |
2256 | * otherwise. | |
2257 | * | |
2258 | * Context: | |
2259 | * queue_lock must be held. | |
2260 | */ | |
2261 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2262 | { |
2263 | struct request *rq; | |
2264 | int ret; | |
2265 | ||
2266 | while ((rq = __elv_next_request(q)) != NULL) { | |
c8158819 LM |
2267 | |
2268 | rq = blk_pm_peek_request(q, rq); | |
2269 | if (!rq) | |
2270 | break; | |
2271 | ||
158dbda0 TH |
2272 | if (!(rq->cmd_flags & REQ_STARTED)) { |
2273 | /* | |
2274 | * This is the first time the device driver | |
2275 | * sees this request (possibly after | |
2276 | * requeueing). Notify IO scheduler. | |
2277 | */ | |
33659ebb | 2278 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
2279 | elv_activate_rq(q, rq); |
2280 | ||
2281 | /* | |
2282 | * just mark as started even if we don't start | |
2283 | * it, a request that has been delayed should | |
2284 | * not be passed by new incoming requests | |
2285 | */ | |
2286 | rq->cmd_flags |= REQ_STARTED; | |
2287 | trace_block_rq_issue(q, rq); | |
2288 | } | |
2289 | ||
2290 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2291 | q->end_sector = rq_end_sector(rq); | |
2292 | q->boundary_rq = NULL; | |
2293 | } | |
2294 | ||
2295 | if (rq->cmd_flags & REQ_DONTPREP) | |
2296 | break; | |
2297 | ||
2e46e8b2 | 2298 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2299 | /* |
2300 | * make sure space for the drain appears we | |
2301 | * know we can do this because max_hw_segments | |
2302 | * has been adjusted to be one fewer than the | |
2303 | * device can handle | |
2304 | */ | |
2305 | rq->nr_phys_segments++; | |
2306 | } | |
2307 | ||
2308 | if (!q->prep_rq_fn) | |
2309 | break; | |
2310 | ||
2311 | ret = q->prep_rq_fn(q, rq); | |
2312 | if (ret == BLKPREP_OK) { | |
2313 | break; | |
2314 | } else if (ret == BLKPREP_DEFER) { | |
2315 | /* | |
2316 | * the request may have been (partially) prepped. | |
2317 | * we need to keep this request in the front to | |
2318 | * avoid resource deadlock. REQ_STARTED will | |
2319 | * prevent other fs requests from passing this one. | |
2320 | */ | |
2e46e8b2 | 2321 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
2322 | !(rq->cmd_flags & REQ_DONTPREP)) { |
2323 | /* | |
2324 | * remove the space for the drain we added | |
2325 | * so that we don't add it again | |
2326 | */ | |
2327 | --rq->nr_phys_segments; | |
2328 | } | |
2329 | ||
2330 | rq = NULL; | |
2331 | break; | |
2332 | } else if (ret == BLKPREP_KILL) { | |
2333 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
2334 | /* |
2335 | * Mark this request as started so we don't trigger | |
2336 | * any debug logic in the end I/O path. | |
2337 | */ | |
2338 | blk_start_request(rq); | |
40cbbb78 | 2339 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
2340 | } else { |
2341 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2342 | break; | |
2343 | } | |
2344 | } | |
2345 | ||
2346 | return rq; | |
2347 | } | |
9934c8c0 | 2348 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2349 | |
9934c8c0 | 2350 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2351 | { |
9934c8c0 TH |
2352 | struct request_queue *q = rq->q; |
2353 | ||
158dbda0 TH |
2354 | BUG_ON(list_empty(&rq->queuelist)); |
2355 | BUG_ON(ELV_ON_HASH(rq)); | |
2356 | ||
2357 | list_del_init(&rq->queuelist); | |
2358 | ||
2359 | /* | |
2360 | * the time frame between a request being removed from the lists | |
2361 | * and to it is freed is accounted as io that is in progress at | |
2362 | * the driver side. | |
2363 | */ | |
9195291e | 2364 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2365 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2366 | set_io_start_time_ns(rq); |
2367 | } | |
158dbda0 TH |
2368 | } |
2369 | ||
9934c8c0 TH |
2370 | /** |
2371 | * blk_start_request - start request processing on the driver | |
2372 | * @req: request to dequeue | |
2373 | * | |
2374 | * Description: | |
2375 | * Dequeue @req and start timeout timer on it. This hands off the | |
2376 | * request to the driver. | |
2377 | * | |
2378 | * Block internal functions which don't want to start timer should | |
2379 | * call blk_dequeue_request(). | |
2380 | * | |
2381 | * Context: | |
2382 | * queue_lock must be held. | |
2383 | */ | |
2384 | void blk_start_request(struct request *req) | |
2385 | { | |
2386 | blk_dequeue_request(req); | |
2387 | ||
2388 | /* | |
5f49f631 TH |
2389 | * We are now handing the request to the hardware, initialize |
2390 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2391 | */ |
5f49f631 | 2392 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2393 | if (unlikely(blk_bidi_rq(req))) |
2394 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2395 | ||
4912aa6c | 2396 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); |
9934c8c0 TH |
2397 | blk_add_timer(req); |
2398 | } | |
2399 | EXPORT_SYMBOL(blk_start_request); | |
2400 | ||
2401 | /** | |
2402 | * blk_fetch_request - fetch a request from a request queue | |
2403 | * @q: request queue to fetch a request from | |
2404 | * | |
2405 | * Description: | |
2406 | * Return the request at the top of @q. The request is started on | |
2407 | * return and LLD can start processing it immediately. | |
2408 | * | |
2409 | * Return: | |
2410 | * Pointer to the request at the top of @q if available. Null | |
2411 | * otherwise. | |
2412 | * | |
2413 | * Context: | |
2414 | * queue_lock must be held. | |
2415 | */ | |
2416 | struct request *blk_fetch_request(struct request_queue *q) | |
2417 | { | |
2418 | struct request *rq; | |
2419 | ||
2420 | rq = blk_peek_request(q); | |
2421 | if (rq) | |
2422 | blk_start_request(rq); | |
2423 | return rq; | |
2424 | } | |
2425 | EXPORT_SYMBOL(blk_fetch_request); | |
2426 | ||
3bcddeac | 2427 | /** |
2e60e022 | 2428 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2429 | * @req: the request being processed |
710027a4 | 2430 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2431 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2432 | * |
2433 | * Description: | |
8ebf9756 RD |
2434 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2435 | * the request structure even if @req doesn't have leftover. | |
2436 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2437 | * |
2438 | * This special helper function is only for request stacking drivers | |
2439 | * (e.g. request-based dm) so that they can handle partial completion. | |
2440 | * Actual device drivers should use blk_end_request instead. | |
2441 | * | |
2442 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2443 | * %false return from this function. | |
3bcddeac KU |
2444 | * |
2445 | * Return: | |
2e60e022 TH |
2446 | * %false - this request doesn't have any more data |
2447 | * %true - this request has more data | |
3bcddeac | 2448 | **/ |
2e60e022 | 2449 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2450 | { |
f79ea416 | 2451 | int total_bytes; |
1da177e4 | 2452 | |
4a0efdc9 HR |
2453 | trace_block_rq_complete(req->q, req, nr_bytes); |
2454 | ||
2e60e022 TH |
2455 | if (!req->bio) |
2456 | return false; | |
2457 | ||
1da177e4 | 2458 | /* |
6f41469c TH |
2459 | * For fs requests, rq is just carrier of independent bio's |
2460 | * and each partial completion should be handled separately. | |
2461 | * Reset per-request error on each partial completion. | |
2462 | * | |
2463 | * TODO: tj: This is too subtle. It would be better to let | |
2464 | * low level drivers do what they see fit. | |
1da177e4 | 2465 | */ |
33659ebb | 2466 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2467 | req->errors = 0; |
2468 | ||
33659ebb CH |
2469 | if (error && req->cmd_type == REQ_TYPE_FS && |
2470 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2471 | char *error_type; |
2472 | ||
2473 | switch (error) { | |
2474 | case -ENOLINK: | |
2475 | error_type = "recoverable transport"; | |
2476 | break; | |
2477 | case -EREMOTEIO: | |
2478 | error_type = "critical target"; | |
2479 | break; | |
2480 | case -EBADE: | |
2481 | error_type = "critical nexus"; | |
2482 | break; | |
d1ffc1f8 HR |
2483 | case -ETIMEDOUT: |
2484 | error_type = "timeout"; | |
2485 | break; | |
a9d6ceb8 HR |
2486 | case -ENOSPC: |
2487 | error_type = "critical space allocation"; | |
2488 | break; | |
7e782af5 HR |
2489 | case -ENODATA: |
2490 | error_type = "critical medium"; | |
2491 | break; | |
79775567 HR |
2492 | case -EIO: |
2493 | default: | |
2494 | error_type = "I/O"; | |
2495 | break; | |
2496 | } | |
ef3ecb66 RE |
2497 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", |
2498 | __func__, error_type, req->rq_disk ? | |
37d7b34f YZ |
2499 | req->rq_disk->disk_name : "?", |
2500 | (unsigned long long)blk_rq_pos(req)); | |
2501 | ||
1da177e4 LT |
2502 | } |
2503 | ||
bc58ba94 | 2504 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2505 | |
f79ea416 KO |
2506 | total_bytes = 0; |
2507 | while (req->bio) { | |
2508 | struct bio *bio = req->bio; | |
4f024f37 | 2509 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2510 | |
4f024f37 | 2511 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2512 | req->bio = bio->bi_next; |
1da177e4 | 2513 | |
f79ea416 | 2514 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2515 | |
f79ea416 KO |
2516 | total_bytes += bio_bytes; |
2517 | nr_bytes -= bio_bytes; | |
1da177e4 | 2518 | |
f79ea416 KO |
2519 | if (!nr_bytes) |
2520 | break; | |
1da177e4 LT |
2521 | } |
2522 | ||
2523 | /* | |
2524 | * completely done | |
2525 | */ | |
2e60e022 TH |
2526 | if (!req->bio) { |
2527 | /* | |
2528 | * Reset counters so that the request stacking driver | |
2529 | * can find how many bytes remain in the request | |
2530 | * later. | |
2531 | */ | |
a2dec7b3 | 2532 | req->__data_len = 0; |
2e60e022 TH |
2533 | return false; |
2534 | } | |
1da177e4 | 2535 | |
a2dec7b3 | 2536 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2537 | |
2538 | /* update sector only for requests with clear definition of sector */ | |
e2a60da7 | 2539 | if (req->cmd_type == REQ_TYPE_FS) |
a2dec7b3 | 2540 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2541 | |
80a761fd TH |
2542 | /* mixed attributes always follow the first bio */ |
2543 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2544 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2545 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2546 | } | |
2547 | ||
2e46e8b2 TH |
2548 | /* |
2549 | * If total number of sectors is less than the first segment | |
2550 | * size, something has gone terribly wrong. | |
2551 | */ | |
2552 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2553 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2554 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2555 | } |
2556 | ||
2557 | /* recalculate the number of segments */ | |
1da177e4 | 2558 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2559 | |
2e60e022 | 2560 | return true; |
1da177e4 | 2561 | } |
2e60e022 | 2562 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2563 | |
2e60e022 TH |
2564 | static bool blk_update_bidi_request(struct request *rq, int error, |
2565 | unsigned int nr_bytes, | |
2566 | unsigned int bidi_bytes) | |
5efccd17 | 2567 | { |
2e60e022 TH |
2568 | if (blk_update_request(rq, error, nr_bytes)) |
2569 | return true; | |
5efccd17 | 2570 | |
2e60e022 TH |
2571 | /* Bidi request must be completed as a whole */ |
2572 | if (unlikely(blk_bidi_rq(rq)) && | |
2573 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2574 | return true; | |
5efccd17 | 2575 | |
e2e1a148 JA |
2576 | if (blk_queue_add_random(rq->q)) |
2577 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2578 | |
2579 | return false; | |
1da177e4 LT |
2580 | } |
2581 | ||
28018c24 JB |
2582 | /** |
2583 | * blk_unprep_request - unprepare a request | |
2584 | * @req: the request | |
2585 | * | |
2586 | * This function makes a request ready for complete resubmission (or | |
2587 | * completion). It happens only after all error handling is complete, | |
2588 | * so represents the appropriate moment to deallocate any resources | |
2589 | * that were allocated to the request in the prep_rq_fn. The queue | |
2590 | * lock is held when calling this. | |
2591 | */ | |
2592 | void blk_unprep_request(struct request *req) | |
2593 | { | |
2594 | struct request_queue *q = req->q; | |
2595 | ||
2596 | req->cmd_flags &= ~REQ_DONTPREP; | |
2597 | if (q->unprep_rq_fn) | |
2598 | q->unprep_rq_fn(q, req); | |
2599 | } | |
2600 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2601 | ||
1da177e4 LT |
2602 | /* |
2603 | * queue lock must be held | |
2604 | */ | |
12120077 | 2605 | void blk_finish_request(struct request *req, int error) |
1da177e4 | 2606 | { |
125c99bc | 2607 | if (req->cmd_flags & REQ_QUEUED) |
b8286239 KU |
2608 | blk_queue_end_tag(req->q, req); |
2609 | ||
ba396a6c | 2610 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2611 | |
33659ebb | 2612 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2613 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2614 | |
e78042e5 MA |
2615 | blk_delete_timer(req); |
2616 | ||
28018c24 JB |
2617 | if (req->cmd_flags & REQ_DONTPREP) |
2618 | blk_unprep_request(req); | |
2619 | ||
bc58ba94 | 2620 | blk_account_io_done(req); |
b8286239 | 2621 | |
1da177e4 | 2622 | if (req->end_io) |
8ffdc655 | 2623 | req->end_io(req, error); |
b8286239 KU |
2624 | else { |
2625 | if (blk_bidi_rq(req)) | |
2626 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2627 | ||
1da177e4 | 2628 | __blk_put_request(req->q, req); |
b8286239 | 2629 | } |
1da177e4 | 2630 | } |
12120077 | 2631 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 2632 | |
3b11313a | 2633 | /** |
2e60e022 TH |
2634 | * blk_end_bidi_request - Complete a bidi request |
2635 | * @rq: the request to complete | |
2636 | * @error: %0 for success, < %0 for error | |
2637 | * @nr_bytes: number of bytes to complete @rq | |
2638 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2639 | * |
2640 | * Description: | |
e3a04fe3 | 2641 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2642 | * Drivers that supports bidi can safely call this member for any |
2643 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2644 | * just ignored. | |
336cdb40 KU |
2645 | * |
2646 | * Return: | |
2e60e022 TH |
2647 | * %false - we are done with this request |
2648 | * %true - still buffers pending for this request | |
a0cd1285 | 2649 | **/ |
b1f74493 | 2650 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2651 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2652 | { | |
336cdb40 | 2653 | struct request_queue *q = rq->q; |
2e60e022 | 2654 | unsigned long flags; |
32fab448 | 2655 | |
2e60e022 TH |
2656 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2657 | return true; | |
32fab448 | 2658 | |
336cdb40 | 2659 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2660 | blk_finish_request(rq, error); |
336cdb40 KU |
2661 | spin_unlock_irqrestore(q->queue_lock, flags); |
2662 | ||
2e60e022 | 2663 | return false; |
32fab448 KU |
2664 | } |
2665 | ||
336cdb40 | 2666 | /** |
2e60e022 TH |
2667 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2668 | * @rq: the request to complete | |
710027a4 | 2669 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2670 | * @nr_bytes: number of bytes to complete @rq |
2671 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2672 | * |
2673 | * Description: | |
2e60e022 TH |
2674 | * Identical to blk_end_bidi_request() except that queue lock is |
2675 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2676 | * |
2677 | * Return: | |
2e60e022 TH |
2678 | * %false - we are done with this request |
2679 | * %true - still buffers pending for this request | |
336cdb40 | 2680 | **/ |
4853abaa | 2681 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2682 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2683 | { |
2e60e022 TH |
2684 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2685 | return true; | |
336cdb40 | 2686 | |
2e60e022 | 2687 | blk_finish_request(rq, error); |
336cdb40 | 2688 | |
2e60e022 | 2689 | return false; |
336cdb40 | 2690 | } |
e19a3ab0 KU |
2691 | |
2692 | /** | |
2693 | * blk_end_request - Helper function for drivers to complete the request. | |
2694 | * @rq: the request being processed | |
710027a4 | 2695 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2696 | * @nr_bytes: number of bytes to complete |
2697 | * | |
2698 | * Description: | |
2699 | * Ends I/O on a number of bytes attached to @rq. | |
2700 | * If @rq has leftover, sets it up for the next range of segments. | |
2701 | * | |
2702 | * Return: | |
b1f74493 FT |
2703 | * %false - we are done with this request |
2704 | * %true - still buffers pending for this request | |
e19a3ab0 | 2705 | **/ |
b1f74493 | 2706 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2707 | { |
b1f74493 | 2708 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2709 | } |
56ad1740 | 2710 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2711 | |
2712 | /** | |
b1f74493 FT |
2713 | * blk_end_request_all - Helper function for drives to finish the request. |
2714 | * @rq: the request to finish | |
8ebf9756 | 2715 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2716 | * |
2717 | * Description: | |
b1f74493 FT |
2718 | * Completely finish @rq. |
2719 | */ | |
2720 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2721 | { |
b1f74493 FT |
2722 | bool pending; |
2723 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2724 | |
b1f74493 FT |
2725 | if (unlikely(blk_bidi_rq(rq))) |
2726 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2727 | |
b1f74493 FT |
2728 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2729 | BUG_ON(pending); | |
2730 | } | |
56ad1740 | 2731 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2732 | |
b1f74493 FT |
2733 | /** |
2734 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2735 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2736 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2737 | * |
2738 | * Description: | |
2739 | * Complete the current consecutively mapped chunk from @rq. | |
2740 | * | |
2741 | * Return: | |
2742 | * %false - we are done with this request | |
2743 | * %true - still buffers pending for this request | |
2744 | */ | |
2745 | bool blk_end_request_cur(struct request *rq, int error) | |
2746 | { | |
2747 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2748 | } |
56ad1740 | 2749 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2750 | |
80a761fd TH |
2751 | /** |
2752 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2753 | * @rq: the request to finish till the next failure boundary for | |
2754 | * @error: must be negative errno | |
2755 | * | |
2756 | * Description: | |
2757 | * Complete @rq till the next failure boundary. | |
2758 | * | |
2759 | * Return: | |
2760 | * %false - we are done with this request | |
2761 | * %true - still buffers pending for this request | |
2762 | */ | |
2763 | bool blk_end_request_err(struct request *rq, int error) | |
2764 | { | |
2765 | WARN_ON(error >= 0); | |
2766 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2767 | } | |
2768 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2769 | ||
e3a04fe3 | 2770 | /** |
b1f74493 FT |
2771 | * __blk_end_request - Helper function for drivers to complete the request. |
2772 | * @rq: the request being processed | |
2773 | * @error: %0 for success, < %0 for error | |
2774 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2775 | * |
2776 | * Description: | |
b1f74493 | 2777 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2778 | * |
2779 | * Return: | |
b1f74493 FT |
2780 | * %false - we are done with this request |
2781 | * %true - still buffers pending for this request | |
e3a04fe3 | 2782 | **/ |
b1f74493 | 2783 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2784 | { |
b1f74493 | 2785 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2786 | } |
56ad1740 | 2787 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2788 | |
32fab448 | 2789 | /** |
b1f74493 FT |
2790 | * __blk_end_request_all - Helper function for drives to finish the request. |
2791 | * @rq: the request to finish | |
8ebf9756 | 2792 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2793 | * |
2794 | * Description: | |
b1f74493 | 2795 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2796 | */ |
b1f74493 | 2797 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2798 | { |
b1f74493 FT |
2799 | bool pending; |
2800 | unsigned int bidi_bytes = 0; | |
2801 | ||
2802 | if (unlikely(blk_bidi_rq(rq))) | |
2803 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2804 | ||
2805 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2806 | BUG_ON(pending); | |
32fab448 | 2807 | } |
56ad1740 | 2808 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2809 | |
e19a3ab0 | 2810 | /** |
b1f74493 FT |
2811 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2812 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2813 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2814 | * |
2815 | * Description: | |
b1f74493 FT |
2816 | * Complete the current consecutively mapped chunk from @rq. Must |
2817 | * be called with queue lock held. | |
e19a3ab0 KU |
2818 | * |
2819 | * Return: | |
b1f74493 FT |
2820 | * %false - we are done with this request |
2821 | * %true - still buffers pending for this request | |
2822 | */ | |
2823 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2824 | { |
b1f74493 | 2825 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2826 | } |
56ad1740 | 2827 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2828 | |
80a761fd TH |
2829 | /** |
2830 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2831 | * @rq: the request to finish till the next failure boundary for | |
2832 | * @error: must be negative errno | |
2833 | * | |
2834 | * Description: | |
2835 | * Complete @rq till the next failure boundary. Must be called | |
2836 | * with queue lock held. | |
2837 | * | |
2838 | * Return: | |
2839 | * %false - we are done with this request | |
2840 | * %true - still buffers pending for this request | |
2841 | */ | |
2842 | bool __blk_end_request_err(struct request *rq, int error) | |
2843 | { | |
2844 | WARN_ON(error >= 0); | |
2845 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2846 | } | |
2847 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2848 | ||
86db1e29 JA |
2849 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2850 | struct bio *bio) | |
1da177e4 | 2851 | { |
a82afdfc | 2852 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2853 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2854 | |
b4f42e28 | 2855 | if (bio_has_data(bio)) |
fb2dce86 | 2856 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
b4f42e28 | 2857 | |
4f024f37 | 2858 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 2859 | rq->bio = rq->biotail = bio; |
1da177e4 | 2860 | |
66846572 N |
2861 | if (bio->bi_bdev) |
2862 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2863 | } | |
1da177e4 | 2864 | |
2d4dc890 IL |
2865 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2866 | /** | |
2867 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2868 | * @rq: the request to be flushed | |
2869 | * | |
2870 | * Description: | |
2871 | * Flush all pages in @rq. | |
2872 | */ | |
2873 | void rq_flush_dcache_pages(struct request *rq) | |
2874 | { | |
2875 | struct req_iterator iter; | |
7988613b | 2876 | struct bio_vec bvec; |
2d4dc890 IL |
2877 | |
2878 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 2879 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
2880 | } |
2881 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2882 | #endif | |
2883 | ||
ef9e3fac KU |
2884 | /** |
2885 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2886 | * @q : the queue of the device being checked | |
2887 | * | |
2888 | * Description: | |
2889 | * Check if underlying low-level drivers of a device are busy. | |
2890 | * If the drivers want to export their busy state, they must set own | |
2891 | * exporting function using blk_queue_lld_busy() first. | |
2892 | * | |
2893 | * Basically, this function is used only by request stacking drivers | |
2894 | * to stop dispatching requests to underlying devices when underlying | |
2895 | * devices are busy. This behavior helps more I/O merging on the queue | |
2896 | * of the request stacking driver and prevents I/O throughput regression | |
2897 | * on burst I/O load. | |
2898 | * | |
2899 | * Return: | |
2900 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2901 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2902 | */ | |
2903 | int blk_lld_busy(struct request_queue *q) | |
2904 | { | |
2905 | if (q->lld_busy_fn) | |
2906 | return q->lld_busy_fn(q); | |
2907 | ||
2908 | return 0; | |
2909 | } | |
2910 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2911 | ||
5f1b670d | 2912 | void blk_rq_prep_clone(struct request *dst, struct request *src) |
b0fd271d KU |
2913 | { |
2914 | dst->cpu = src->cpu; | |
5f1b670d CH |
2915 | dst->cmd_flags |= (src->cmd_flags & REQ_CLONE_MASK); |
2916 | dst->cmd_flags |= REQ_NOMERGE | REQ_CLONE; | |
b0fd271d KU |
2917 | dst->cmd_type = src->cmd_type; |
2918 | dst->__sector = blk_rq_pos(src); | |
2919 | dst->__data_len = blk_rq_bytes(src); | |
2920 | dst->nr_phys_segments = src->nr_phys_segments; | |
2921 | dst->ioprio = src->ioprio; | |
2922 | dst->extra_len = src->extra_len; | |
5f1b670d CH |
2923 | dst->bio = src->bio; |
2924 | dst->biotail = src->biotail; | |
2925 | dst->cmd = src->cmd; | |
2926 | dst->cmd_len = src->cmd_len; | |
2927 | dst->sense = src->sense; | |
b0fd271d KU |
2928 | } |
2929 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2930 | ||
59c3d45e | 2931 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
2932 | { |
2933 | return queue_work(kblockd_workqueue, work); | |
2934 | } | |
1da177e4 LT |
2935 | EXPORT_SYMBOL(kblockd_schedule_work); |
2936 | ||
59c3d45e JA |
2937 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, |
2938 | unsigned long delay) | |
e43473b7 VG |
2939 | { |
2940 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2941 | } | |
2942 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2943 | ||
8ab14595 JA |
2944 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
2945 | unsigned long delay) | |
2946 | { | |
2947 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
2948 | } | |
2949 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
2950 | ||
75df7136 SJ |
2951 | /** |
2952 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2953 | * @plug: The &struct blk_plug that needs to be initialized | |
2954 | * | |
2955 | * Description: | |
2956 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2957 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2958 | * blk_finish_plug(). This is important from a performance perspective, but | |
2959 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2960 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2961 | * page belonging to that request that is currently residing in our private | |
2962 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2963 | * this kind of deadlock. | |
2964 | */ | |
73c10101 JA |
2965 | void blk_start_plug(struct blk_plug *plug) |
2966 | { | |
2967 | struct task_struct *tsk = current; | |
2968 | ||
dd6cf3e1 SL |
2969 | /* |
2970 | * If this is a nested plug, don't actually assign it. | |
2971 | */ | |
2972 | if (tsk->plug) | |
2973 | return; | |
2974 | ||
73c10101 | 2975 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 2976 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 2977 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 2978 | /* |
dd6cf3e1 SL |
2979 | * Store ordering should not be needed here, since a potential |
2980 | * preempt will imply a full memory barrier | |
73c10101 | 2981 | */ |
dd6cf3e1 | 2982 | tsk->plug = plug; |
73c10101 JA |
2983 | } |
2984 | EXPORT_SYMBOL(blk_start_plug); | |
2985 | ||
2986 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2987 | { | |
2988 | struct request *rqa = container_of(a, struct request, queuelist); | |
2989 | struct request *rqb = container_of(b, struct request, queuelist); | |
2990 | ||
975927b9 JM |
2991 | return !(rqa->q < rqb->q || |
2992 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
2993 | } |
2994 | ||
49cac01e JA |
2995 | /* |
2996 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2997 | * until a safe kblockd context. We due this to avoid accidental big | |
2998 | * additional stack usage in driver dispatch, in places where the originally | |
2999 | * plugger did not intend it. | |
3000 | */ | |
f6603783 | 3001 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3002 | bool from_schedule) |
99e22598 | 3003 | __releases(q->queue_lock) |
94b5eb28 | 3004 | { |
49cac01e | 3005 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3006 | |
70460571 | 3007 | if (from_schedule) |
24ecfbe2 | 3008 | blk_run_queue_async(q); |
70460571 | 3009 | else |
24ecfbe2 | 3010 | __blk_run_queue(q); |
70460571 | 3011 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3012 | } |
3013 | ||
74018dc3 | 3014 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3015 | { |
3016 | LIST_HEAD(callbacks); | |
3017 | ||
2a7d5559 SL |
3018 | while (!list_empty(&plug->cb_list)) { |
3019 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3020 | |
2a7d5559 SL |
3021 | while (!list_empty(&callbacks)) { |
3022 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3023 | struct blk_plug_cb, |
3024 | list); | |
2a7d5559 | 3025 | list_del(&cb->list); |
74018dc3 | 3026 | cb->callback(cb, from_schedule); |
2a7d5559 | 3027 | } |
048c9374 N |
3028 | } |
3029 | } | |
3030 | ||
9cbb1750 N |
3031 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3032 | int size) | |
3033 | { | |
3034 | struct blk_plug *plug = current->plug; | |
3035 | struct blk_plug_cb *cb; | |
3036 | ||
3037 | if (!plug) | |
3038 | return NULL; | |
3039 | ||
3040 | list_for_each_entry(cb, &plug->cb_list, list) | |
3041 | if (cb->callback == unplug && cb->data == data) | |
3042 | return cb; | |
3043 | ||
3044 | /* Not currently on the callback list */ | |
3045 | BUG_ON(size < sizeof(*cb)); | |
3046 | cb = kzalloc(size, GFP_ATOMIC); | |
3047 | if (cb) { | |
3048 | cb->data = data; | |
3049 | cb->callback = unplug; | |
3050 | list_add(&cb->list, &plug->cb_list); | |
3051 | } | |
3052 | return cb; | |
3053 | } | |
3054 | EXPORT_SYMBOL(blk_check_plugged); | |
3055 | ||
49cac01e | 3056 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3057 | { |
3058 | struct request_queue *q; | |
3059 | unsigned long flags; | |
3060 | struct request *rq; | |
109b8129 | 3061 | LIST_HEAD(list); |
94b5eb28 | 3062 | unsigned int depth; |
73c10101 | 3063 | |
74018dc3 | 3064 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3065 | |
3066 | if (!list_empty(&plug->mq_list)) | |
3067 | blk_mq_flush_plug_list(plug, from_schedule); | |
3068 | ||
73c10101 JA |
3069 | if (list_empty(&plug->list)) |
3070 | return; | |
3071 | ||
109b8129 N |
3072 | list_splice_init(&plug->list, &list); |
3073 | ||
422765c2 | 3074 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3075 | |
3076 | q = NULL; | |
94b5eb28 | 3077 | depth = 0; |
18811272 JA |
3078 | |
3079 | /* | |
3080 | * Save and disable interrupts here, to avoid doing it for every | |
3081 | * queue lock we have to take. | |
3082 | */ | |
73c10101 | 3083 | local_irq_save(flags); |
109b8129 N |
3084 | while (!list_empty(&list)) { |
3085 | rq = list_entry_rq(list.next); | |
73c10101 | 3086 | list_del_init(&rq->queuelist); |
73c10101 JA |
3087 | BUG_ON(!rq->q); |
3088 | if (rq->q != q) { | |
99e22598 JA |
3089 | /* |
3090 | * This drops the queue lock | |
3091 | */ | |
3092 | if (q) | |
49cac01e | 3093 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3094 | q = rq->q; |
94b5eb28 | 3095 | depth = 0; |
73c10101 JA |
3096 | spin_lock(q->queue_lock); |
3097 | } | |
8ba61435 TH |
3098 | |
3099 | /* | |
3100 | * Short-circuit if @q is dead | |
3101 | */ | |
3f3299d5 | 3102 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
3103 | __blk_end_request_all(rq, -ENODEV); |
3104 | continue; | |
3105 | } | |
3106 | ||
73c10101 JA |
3107 | /* |
3108 | * rq is already accounted, so use raw insert | |
3109 | */ | |
401a18e9 JA |
3110 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
3111 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
3112 | else | |
3113 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3114 | |
3115 | depth++; | |
73c10101 JA |
3116 | } |
3117 | ||
99e22598 JA |
3118 | /* |
3119 | * This drops the queue lock | |
3120 | */ | |
3121 | if (q) | |
49cac01e | 3122 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3123 | |
73c10101 JA |
3124 | local_irq_restore(flags); |
3125 | } | |
73c10101 JA |
3126 | |
3127 | void blk_finish_plug(struct blk_plug *plug) | |
3128 | { | |
dd6cf3e1 SL |
3129 | if (plug != current->plug) |
3130 | return; | |
f6603783 | 3131 | blk_flush_plug_list(plug, false); |
73c10101 | 3132 | |
dd6cf3e1 | 3133 | current->plug = NULL; |
73c10101 | 3134 | } |
88b996cd | 3135 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3136 | |
47fafbc7 | 3137 | #ifdef CONFIG_PM |
6c954667 LM |
3138 | /** |
3139 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3140 | * @q: the queue of the device | |
3141 | * @dev: the device the queue belongs to | |
3142 | * | |
3143 | * Description: | |
3144 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3145 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3146 | * should call this function after @dev has been initialized, and its | |
3147 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3148 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3149 | * cases, driver should call this function before any I/O has taken place. | |
3150 | * | |
3151 | * This function takes care of setting up using auto suspend for the device, | |
3152 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3153 | * until an updated value is either set by user or by driver. Drivers do | |
3154 | * not need to touch other autosuspend settings. | |
3155 | * | |
3156 | * The block layer runtime PM is request based, so only works for drivers | |
3157 | * that use request as their IO unit instead of those directly use bio's. | |
3158 | */ | |
3159 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3160 | { | |
3161 | q->dev = dev; | |
3162 | q->rpm_status = RPM_ACTIVE; | |
3163 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3164 | pm_runtime_use_autosuspend(q->dev); | |
3165 | } | |
3166 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3167 | ||
3168 | /** | |
3169 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3170 | * @q: the queue of the device | |
3171 | * | |
3172 | * Description: | |
3173 | * This function will check if runtime suspend is allowed for the device | |
3174 | * by examining if there are any requests pending in the queue. If there | |
3175 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3176 | * the queue's status will be updated to SUSPENDING and the driver can | |
3177 | * proceed to suspend the device. | |
3178 | * | |
3179 | * For the not allowed case, we mark last busy for the device so that | |
3180 | * runtime PM core will try to autosuspend it some time later. | |
3181 | * | |
3182 | * This function should be called near the start of the device's | |
3183 | * runtime_suspend callback. | |
3184 | * | |
3185 | * Return: | |
3186 | * 0 - OK to runtime suspend the device | |
3187 | * -EBUSY - Device should not be runtime suspended | |
3188 | */ | |
3189 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3190 | { | |
3191 | int ret = 0; | |
3192 | ||
3193 | spin_lock_irq(q->queue_lock); | |
3194 | if (q->nr_pending) { | |
3195 | ret = -EBUSY; | |
3196 | pm_runtime_mark_last_busy(q->dev); | |
3197 | } else { | |
3198 | q->rpm_status = RPM_SUSPENDING; | |
3199 | } | |
3200 | spin_unlock_irq(q->queue_lock); | |
3201 | return ret; | |
3202 | } | |
3203 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3204 | ||
3205 | /** | |
3206 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3207 | * @q: the queue of the device | |
3208 | * @err: return value of the device's runtime_suspend function | |
3209 | * | |
3210 | * Description: | |
3211 | * Update the queue's runtime status according to the return value of the | |
3212 | * device's runtime suspend function and mark last busy for the device so | |
3213 | * that PM core will try to auto suspend the device at a later time. | |
3214 | * | |
3215 | * This function should be called near the end of the device's | |
3216 | * runtime_suspend callback. | |
3217 | */ | |
3218 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3219 | { | |
3220 | spin_lock_irq(q->queue_lock); | |
3221 | if (!err) { | |
3222 | q->rpm_status = RPM_SUSPENDED; | |
3223 | } else { | |
3224 | q->rpm_status = RPM_ACTIVE; | |
3225 | pm_runtime_mark_last_busy(q->dev); | |
3226 | } | |
3227 | spin_unlock_irq(q->queue_lock); | |
3228 | } | |
3229 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3230 | ||
3231 | /** | |
3232 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3233 | * @q: the queue of the device | |
3234 | * | |
3235 | * Description: | |
3236 | * Update the queue's runtime status to RESUMING in preparation for the | |
3237 | * runtime resume of the device. | |
3238 | * | |
3239 | * This function should be called near the start of the device's | |
3240 | * runtime_resume callback. | |
3241 | */ | |
3242 | void blk_pre_runtime_resume(struct request_queue *q) | |
3243 | { | |
3244 | spin_lock_irq(q->queue_lock); | |
3245 | q->rpm_status = RPM_RESUMING; | |
3246 | spin_unlock_irq(q->queue_lock); | |
3247 | } | |
3248 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3249 | ||
3250 | /** | |
3251 | * blk_post_runtime_resume - Post runtime resume processing | |
3252 | * @q: the queue of the device | |
3253 | * @err: return value of the device's runtime_resume function | |
3254 | * | |
3255 | * Description: | |
3256 | * Update the queue's runtime status according to the return value of the | |
3257 | * device's runtime_resume function. If it is successfully resumed, process | |
3258 | * the requests that are queued into the device's queue when it is resuming | |
3259 | * and then mark last busy and initiate autosuspend for it. | |
3260 | * | |
3261 | * This function should be called near the end of the device's | |
3262 | * runtime_resume callback. | |
3263 | */ | |
3264 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3265 | { | |
3266 | spin_lock_irq(q->queue_lock); | |
3267 | if (!err) { | |
3268 | q->rpm_status = RPM_ACTIVE; | |
3269 | __blk_run_queue(q); | |
3270 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3271 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3272 | } else { |
3273 | q->rpm_status = RPM_SUSPENDED; | |
3274 | } | |
3275 | spin_unlock_irq(q->queue_lock); | |
3276 | } | |
3277 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
3278 | #endif | |
3279 | ||
1da177e4 LT |
3280 | int __init blk_dev_init(void) |
3281 | { | |
9eb55b03 NK |
3282 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
3283 | sizeof(((struct request *)0)->cmd_flags)); | |
3284 | ||
89b90be2 TH |
3285 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3286 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3287 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3288 | if (!kblockd_workqueue) |
3289 | panic("Failed to create kblockd\n"); | |
3290 | ||
3291 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3292 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3293 | |
8324aa91 | 3294 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 3295 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3296 | |
d38ecf93 | 3297 | return 0; |
1da177e4 | 3298 | } |