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" |
87760e5e | 42 | #include "blk-wbt.h" |
8324aa91 | 43 | |
d07335e5 | 44 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 45 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 46 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 47 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 48 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 49 | |
a73f730d TH |
50 | DEFINE_IDA(blk_queue_ida); |
51 | ||
1da177e4 LT |
52 | /* |
53 | * For the allocated request tables | |
54 | */ | |
d674d414 | 55 | struct kmem_cache *request_cachep; |
1da177e4 LT |
56 | |
57 | /* | |
58 | * For queue allocation | |
59 | */ | |
6728cb0e | 60 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 61 | |
1da177e4 LT |
62 | /* |
63 | * Controlling structure to kblockd | |
64 | */ | |
ff856bad | 65 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 66 | |
d40f75a0 TH |
67 | static void blk_clear_congested(struct request_list *rl, int sync) |
68 | { | |
d40f75a0 TH |
69 | #ifdef CONFIG_CGROUP_WRITEBACK |
70 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
71 | #else | |
482cf79c TH |
72 | /* |
73 | * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't | |
74 | * flip its congestion state for events on other blkcgs. | |
75 | */ | |
76 | if (rl == &rl->q->root_rl) | |
77 | clear_wb_congested(rl->q->backing_dev_info.wb.congested, sync); | |
d40f75a0 TH |
78 | #endif |
79 | } | |
80 | ||
81 | static void blk_set_congested(struct request_list *rl, int sync) | |
82 | { | |
d40f75a0 TH |
83 | #ifdef CONFIG_CGROUP_WRITEBACK |
84 | set_wb_congested(rl->blkg->wb_congested, sync); | |
85 | #else | |
482cf79c TH |
86 | /* see blk_clear_congested() */ |
87 | if (rl == &rl->q->root_rl) | |
88 | set_wb_congested(rl->q->backing_dev_info.wb.congested, sync); | |
d40f75a0 TH |
89 | #endif |
90 | } | |
91 | ||
8324aa91 | 92 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
93 | { |
94 | int nr; | |
95 | ||
96 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
97 | if (nr > q->nr_requests) | |
98 | nr = q->nr_requests; | |
99 | q->nr_congestion_on = nr; | |
100 | ||
101 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
102 | if (nr < 1) | |
103 | nr = 1; | |
104 | q->nr_congestion_off = nr; | |
105 | } | |
106 | ||
1da177e4 LT |
107 | /** |
108 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
109 | * @bdev: device | |
110 | * | |
111 | * Locates the passed device's request queue and returns the address of its | |
ff9ea323 TH |
112 | * backing_dev_info. This function can only be called if @bdev is opened |
113 | * and the return value is never NULL. | |
1da177e4 LT |
114 | */ |
115 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
116 | { | |
165125e1 | 117 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 | 118 | |
ff9ea323 | 119 | return &q->backing_dev_info; |
1da177e4 | 120 | } |
1da177e4 LT |
121 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
122 | ||
2a4aa30c | 123 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 124 | { |
1afb20f3 FT |
125 | memset(rq, 0, sizeof(*rq)); |
126 | ||
1da177e4 | 127 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 128 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 129 | rq->cpu = -1; |
63a71386 | 130 | rq->q = q; |
a2dec7b3 | 131 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
132 | INIT_HLIST_NODE(&rq->hash); |
133 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 134 | rq->cmd = rq->__cmd; |
e2494e1b | 135 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 136 | rq->tag = -1; |
b243ddcb | 137 | rq->start_time = jiffies; |
9195291e | 138 | set_start_time_ns(rq); |
09e099d4 | 139 | rq->part = NULL; |
1da177e4 | 140 | } |
2a4aa30c | 141 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 142 | |
5bb23a68 N |
143 | static void req_bio_endio(struct request *rq, struct bio *bio, |
144 | unsigned int nbytes, int error) | |
1da177e4 | 145 | { |
78d8e58a | 146 | if (error) |
4246a0b6 | 147 | bio->bi_error = error; |
797e7dbb | 148 | |
e8064021 | 149 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 150 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 151 | |
f79ea416 | 152 | bio_advance(bio, nbytes); |
7ba1ba12 | 153 | |
143a87f4 | 154 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 155 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 156 | bio_endio(bio); |
1da177e4 | 157 | } |
1da177e4 | 158 | |
1da177e4 LT |
159 | void blk_dump_rq_flags(struct request *rq, char *msg) |
160 | { | |
161 | int bit; | |
162 | ||
5953316d | 163 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%llx\n", msg, |
4aff5e23 | 164 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
5953316d | 165 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 166 | |
83096ebf TH |
167 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
168 | (unsigned long long)blk_rq_pos(rq), | |
169 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
170 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
171 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 172 | |
33659ebb | 173 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 174 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 175 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
176 | printk("%02x ", rq->cmd[bit]); |
177 | printk("\n"); | |
178 | } | |
179 | } | |
1da177e4 LT |
180 | EXPORT_SYMBOL(blk_dump_rq_flags); |
181 | ||
3cca6dc1 | 182 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 183 | { |
3cca6dc1 | 184 | struct request_queue *q; |
1da177e4 | 185 | |
3cca6dc1 JA |
186 | q = container_of(work, struct request_queue, delay_work.work); |
187 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 188 | __blk_run_queue(q); |
3cca6dc1 | 189 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 190 | } |
1da177e4 LT |
191 | |
192 | /** | |
3cca6dc1 JA |
193 | * blk_delay_queue - restart queueing after defined interval |
194 | * @q: The &struct request_queue in question | |
195 | * @msecs: Delay in msecs | |
1da177e4 LT |
196 | * |
197 | * Description: | |
3cca6dc1 JA |
198 | * Sometimes queueing needs to be postponed for a little while, to allow |
199 | * resources to come back. This function will make sure that queueing is | |
70460571 | 200 | * restarted around the specified time. Queue lock must be held. |
3cca6dc1 JA |
201 | */ |
202 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 203 | { |
70460571 BVA |
204 | if (likely(!blk_queue_dead(q))) |
205 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
206 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 207 | } |
3cca6dc1 | 208 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 209 | |
21491412 JA |
210 | /** |
211 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
212 | * @q: The &struct request_queue in question | |
213 | * | |
214 | * Description: | |
215 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
216 | * ensure that the request_fn for the queue is run from an async | |
217 | * context. | |
218 | **/ | |
219 | void blk_start_queue_async(struct request_queue *q) | |
220 | { | |
221 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
222 | blk_run_queue_async(q); | |
223 | } | |
224 | EXPORT_SYMBOL(blk_start_queue_async); | |
225 | ||
1da177e4 LT |
226 | /** |
227 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 228 | * @q: The &struct request_queue in question |
1da177e4 LT |
229 | * |
230 | * Description: | |
231 | * blk_start_queue() will clear the stop flag on the queue, and call | |
232 | * the request_fn for the queue if it was in a stopped state when | |
233 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
234 | **/ | |
165125e1 | 235 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 236 | { |
a038e253 PBG |
237 | WARN_ON(!irqs_disabled()); |
238 | ||
75ad23bc | 239 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 240 | __blk_run_queue(q); |
1da177e4 | 241 | } |
1da177e4 LT |
242 | EXPORT_SYMBOL(blk_start_queue); |
243 | ||
244 | /** | |
245 | * blk_stop_queue - stop a queue | |
165125e1 | 246 | * @q: The &struct request_queue in question |
1da177e4 LT |
247 | * |
248 | * Description: | |
249 | * The Linux block layer assumes that a block driver will consume all | |
250 | * entries on the request queue when the request_fn strategy is called. | |
251 | * Often this will not happen, because of hardware limitations (queue | |
252 | * depth settings). If a device driver gets a 'queue full' response, | |
253 | * or if it simply chooses not to queue more I/O at one point, it can | |
254 | * call this function to prevent the request_fn from being called until | |
255 | * the driver has signalled it's ready to go again. This happens by calling | |
256 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
257 | **/ | |
165125e1 | 258 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 259 | { |
136b5721 | 260 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 261 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
262 | } |
263 | EXPORT_SYMBOL(blk_stop_queue); | |
264 | ||
265 | /** | |
266 | * blk_sync_queue - cancel any pending callbacks on a queue | |
267 | * @q: the queue | |
268 | * | |
269 | * Description: | |
270 | * The block layer may perform asynchronous callback activity | |
271 | * on a queue, such as calling the unplug function after a timeout. | |
272 | * A block device may call blk_sync_queue to ensure that any | |
273 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 274 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
275 | * that its ->make_request_fn will not re-add plugging prior to calling |
276 | * this function. | |
277 | * | |
da527770 | 278 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 279 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 280 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 281 | * |
1da177e4 LT |
282 | */ |
283 | void blk_sync_queue(struct request_queue *q) | |
284 | { | |
70ed28b9 | 285 | del_timer_sync(&q->timeout); |
f04c1fe7 ML |
286 | |
287 | if (q->mq_ops) { | |
288 | struct blk_mq_hw_ctx *hctx; | |
289 | int i; | |
290 | ||
70f4db63 | 291 | queue_for_each_hw_ctx(q, hctx, i) { |
27489a3c | 292 | cancel_work_sync(&hctx->run_work); |
70f4db63 CH |
293 | cancel_delayed_work_sync(&hctx->delay_work); |
294 | } | |
f04c1fe7 ML |
295 | } else { |
296 | cancel_delayed_work_sync(&q->delay_work); | |
297 | } | |
1da177e4 LT |
298 | } |
299 | EXPORT_SYMBOL(blk_sync_queue); | |
300 | ||
c246e80d BVA |
301 | /** |
302 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
303 | * @q: The queue to run | |
304 | * | |
305 | * Description: | |
306 | * Invoke request handling on a queue if there are any pending requests. | |
307 | * May be used to restart request handling after a request has completed. | |
308 | * This variant runs the queue whether or not the queue has been | |
309 | * stopped. Must be called with the queue lock held and interrupts | |
310 | * disabled. See also @blk_run_queue. | |
311 | */ | |
312 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
313 | { | |
314 | if (unlikely(blk_queue_dead(q))) | |
315 | return; | |
316 | ||
24faf6f6 BVA |
317 | /* |
318 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
319 | * the queue lock internally. As a result multiple threads may be | |
320 | * running such a request function concurrently. Keep track of the | |
321 | * number of active request_fn invocations such that blk_drain_queue() | |
322 | * can wait until all these request_fn calls have finished. | |
323 | */ | |
324 | q->request_fn_active++; | |
c246e80d | 325 | q->request_fn(q); |
24faf6f6 | 326 | q->request_fn_active--; |
c246e80d | 327 | } |
a7928c15 | 328 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 329 | |
1da177e4 | 330 | /** |
80a4b58e | 331 | * __blk_run_queue - run a single device queue |
1da177e4 | 332 | * @q: The queue to run |
80a4b58e JA |
333 | * |
334 | * Description: | |
335 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 336 | * held and interrupts disabled. |
1da177e4 | 337 | */ |
24ecfbe2 | 338 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 339 | { |
a538cd03 TH |
340 | if (unlikely(blk_queue_stopped(q))) |
341 | return; | |
342 | ||
c246e80d | 343 | __blk_run_queue_uncond(q); |
75ad23bc NP |
344 | } |
345 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 346 | |
24ecfbe2 CH |
347 | /** |
348 | * blk_run_queue_async - run a single device queue in workqueue context | |
349 | * @q: The queue to run | |
350 | * | |
351 | * Description: | |
352 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
70460571 | 353 | * of us. The caller must hold the queue lock. |
24ecfbe2 CH |
354 | */ |
355 | void blk_run_queue_async(struct request_queue *q) | |
356 | { | |
70460571 | 357 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 358 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 359 | } |
c21e6beb | 360 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 361 | |
75ad23bc NP |
362 | /** |
363 | * blk_run_queue - run a single device queue | |
364 | * @q: The queue to run | |
80a4b58e JA |
365 | * |
366 | * Description: | |
367 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 368 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
369 | */ |
370 | void blk_run_queue(struct request_queue *q) | |
371 | { | |
372 | unsigned long flags; | |
373 | ||
374 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 375 | __blk_run_queue(q); |
1da177e4 LT |
376 | spin_unlock_irqrestore(q->queue_lock, flags); |
377 | } | |
378 | EXPORT_SYMBOL(blk_run_queue); | |
379 | ||
165125e1 | 380 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
381 | { |
382 | kobject_put(&q->kobj); | |
383 | } | |
d86e0e83 | 384 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 385 | |
e3c78ca5 | 386 | /** |
807592a4 | 387 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 388 | * @q: queue to drain |
c9a929dd | 389 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 390 | * |
c9a929dd TH |
391 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
392 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
393 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 394 | */ |
807592a4 BVA |
395 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
396 | __releases(q->queue_lock) | |
397 | __acquires(q->queue_lock) | |
e3c78ca5 | 398 | { |
458f27a9 AH |
399 | int i; |
400 | ||
807592a4 BVA |
401 | lockdep_assert_held(q->queue_lock); |
402 | ||
e3c78ca5 | 403 | while (true) { |
481a7d64 | 404 | bool drain = false; |
e3c78ca5 | 405 | |
b855b04a TH |
406 | /* |
407 | * The caller might be trying to drain @q before its | |
408 | * elevator is initialized. | |
409 | */ | |
410 | if (q->elevator) | |
411 | elv_drain_elevator(q); | |
412 | ||
5efd6113 | 413 | blkcg_drain_queue(q); |
e3c78ca5 | 414 | |
4eabc941 TH |
415 | /* |
416 | * This function might be called on a queue which failed | |
b855b04a TH |
417 | * driver init after queue creation or is not yet fully |
418 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
419 | * in such cases. Kick queue iff dispatch queue has | |
420 | * something on it and @q has request_fn set. | |
4eabc941 | 421 | */ |
b855b04a | 422 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 423 | __blk_run_queue(q); |
c9a929dd | 424 | |
8a5ecdd4 | 425 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 426 | drain |= q->request_fn_active; |
481a7d64 TH |
427 | |
428 | /* | |
429 | * Unfortunately, requests are queued at and tracked from | |
430 | * multiple places and there's no single counter which can | |
431 | * be drained. Check all the queues and counters. | |
432 | */ | |
433 | if (drain_all) { | |
e97c293c | 434 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
435 | drain |= !list_empty(&q->queue_head); |
436 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 437 | drain |= q->nr_rqs[i]; |
481a7d64 | 438 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
439 | if (fq) |
440 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
441 | } |
442 | } | |
e3c78ca5 | 443 | |
481a7d64 | 444 | if (!drain) |
e3c78ca5 | 445 | break; |
807592a4 BVA |
446 | |
447 | spin_unlock_irq(q->queue_lock); | |
448 | ||
e3c78ca5 | 449 | msleep(10); |
807592a4 BVA |
450 | |
451 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 452 | } |
458f27a9 AH |
453 | |
454 | /* | |
455 | * With queue marked dead, any woken up waiter will fail the | |
456 | * allocation path, so the wakeup chaining is lost and we're | |
457 | * left with hung waiters. We need to wake up those waiters. | |
458 | */ | |
459 | if (q->request_fn) { | |
a051661c TH |
460 | struct request_list *rl; |
461 | ||
a051661c TH |
462 | blk_queue_for_each_rl(rl, q) |
463 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
464 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 465 | } |
e3c78ca5 TH |
466 | } |
467 | ||
d732580b TH |
468 | /** |
469 | * blk_queue_bypass_start - enter queue bypass mode | |
470 | * @q: queue of interest | |
471 | * | |
472 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
473 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 474 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
475 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
476 | * inside queue or RCU read lock. | |
d732580b TH |
477 | */ |
478 | void blk_queue_bypass_start(struct request_queue *q) | |
479 | { | |
480 | spin_lock_irq(q->queue_lock); | |
776687bc | 481 | q->bypass_depth++; |
d732580b TH |
482 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
483 | spin_unlock_irq(q->queue_lock); | |
484 | ||
776687bc TH |
485 | /* |
486 | * Queues start drained. Skip actual draining till init is | |
487 | * complete. This avoids lenghty delays during queue init which | |
488 | * can happen many times during boot. | |
489 | */ | |
490 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
491 | spin_lock_irq(q->queue_lock); |
492 | __blk_drain_queue(q, false); | |
493 | spin_unlock_irq(q->queue_lock); | |
494 | ||
b82d4b19 TH |
495 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
496 | synchronize_rcu(); | |
497 | } | |
d732580b TH |
498 | } |
499 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
500 | ||
501 | /** | |
502 | * blk_queue_bypass_end - leave queue bypass mode | |
503 | * @q: queue of interest | |
504 | * | |
505 | * Leave bypass mode and restore the normal queueing behavior. | |
506 | */ | |
507 | void blk_queue_bypass_end(struct request_queue *q) | |
508 | { | |
509 | spin_lock_irq(q->queue_lock); | |
510 | if (!--q->bypass_depth) | |
511 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
512 | WARN_ON_ONCE(q->bypass_depth < 0); | |
513 | spin_unlock_irq(q->queue_lock); | |
514 | } | |
515 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
516 | ||
aed3ea94 JA |
517 | void blk_set_queue_dying(struct request_queue *q) |
518 | { | |
1b856086 BVA |
519 | spin_lock_irq(q->queue_lock); |
520 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
521 | spin_unlock_irq(q->queue_lock); | |
aed3ea94 JA |
522 | |
523 | if (q->mq_ops) | |
524 | blk_mq_wake_waiters(q); | |
525 | else { | |
526 | struct request_list *rl; | |
527 | ||
528 | blk_queue_for_each_rl(rl, q) { | |
529 | if (rl->rq_pool) { | |
530 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
531 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
532 | } | |
533 | } | |
534 | } | |
535 | } | |
536 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
537 | ||
c9a929dd TH |
538 | /** |
539 | * blk_cleanup_queue - shutdown a request queue | |
540 | * @q: request queue to shutdown | |
541 | * | |
c246e80d BVA |
542 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
543 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 544 | */ |
6728cb0e | 545 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 546 | { |
c9a929dd | 547 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 548 | |
3f3299d5 | 549 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 550 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 551 | blk_set_queue_dying(q); |
c9a929dd | 552 | spin_lock_irq(lock); |
6ecf23af | 553 | |
80fd9979 | 554 | /* |
3f3299d5 | 555 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
556 | * that, unlike blk_queue_bypass_start(), we aren't performing |
557 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
558 | * as some drivers create and destroy a lot of queues while | |
559 | * probing. This is still safe because blk_release_queue() will be | |
560 | * called only after the queue refcnt drops to zero and nothing, | |
561 | * RCU or not, would be traversing the queue by then. | |
562 | */ | |
6ecf23af TH |
563 | q->bypass_depth++; |
564 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
565 | ||
c9a929dd TH |
566 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
567 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 568 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
569 | spin_unlock_irq(lock); |
570 | mutex_unlock(&q->sysfs_lock); | |
571 | ||
c246e80d BVA |
572 | /* |
573 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
574 | * prevent that q->request_fn() gets invoked after draining finished. | |
575 | */ | |
3ef28e83 DW |
576 | blk_freeze_queue(q); |
577 | spin_lock_irq(lock); | |
578 | if (!q->mq_ops) | |
43a5e4e2 | 579 | __blk_drain_queue(q, true); |
c246e80d | 580 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 581 | spin_unlock_irq(lock); |
c9a929dd | 582 | |
5a48fc14 DW |
583 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
584 | blk_flush_integrity(); | |
585 | ||
c9a929dd TH |
586 | /* @q won't process any more request, flush async actions */ |
587 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
588 | blk_sync_queue(q); | |
589 | ||
45a9c9d9 BVA |
590 | if (q->mq_ops) |
591 | blk_mq_free_queue(q); | |
3ef28e83 | 592 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 593 | |
5e5cfac0 AH |
594 | spin_lock_irq(lock); |
595 | if (q->queue_lock != &q->__queue_lock) | |
596 | q->queue_lock = &q->__queue_lock; | |
597 | spin_unlock_irq(lock); | |
598 | ||
b02176f3 | 599 | bdi_unregister(&q->backing_dev_info); |
6cd18e71 | 600 | |
c9a929dd | 601 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
602 | blk_put_queue(q); |
603 | } | |
1da177e4 LT |
604 | EXPORT_SYMBOL(blk_cleanup_queue); |
605 | ||
271508db DR |
606 | /* Allocate memory local to the request queue */ |
607 | static void *alloc_request_struct(gfp_t gfp_mask, void *data) | |
608 | { | |
609 | int nid = (int)(long)data; | |
610 | return kmem_cache_alloc_node(request_cachep, gfp_mask, nid); | |
611 | } | |
612 | ||
613 | static void free_request_struct(void *element, void *unused) | |
614 | { | |
615 | kmem_cache_free(request_cachep, element); | |
616 | } | |
617 | ||
5b788ce3 TH |
618 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
619 | gfp_t gfp_mask) | |
1da177e4 | 620 | { |
1abec4fd MS |
621 | if (unlikely(rl->rq_pool)) |
622 | return 0; | |
623 | ||
5b788ce3 | 624 | rl->q = q; |
1faa16d2 JA |
625 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
626 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
627 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
628 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 629 | |
271508db DR |
630 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, alloc_request_struct, |
631 | free_request_struct, | |
632 | (void *)(long)q->node, gfp_mask, | |
633 | q->node); | |
1da177e4 LT |
634 | if (!rl->rq_pool) |
635 | return -ENOMEM; | |
636 | ||
637 | return 0; | |
638 | } | |
639 | ||
5b788ce3 TH |
640 | void blk_exit_rl(struct request_list *rl) |
641 | { | |
642 | if (rl->rq_pool) | |
643 | mempool_destroy(rl->rq_pool); | |
644 | } | |
645 | ||
165125e1 | 646 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 647 | { |
c304a51b | 648 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
649 | } |
650 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 651 | |
6f3b0e8b | 652 | int blk_queue_enter(struct request_queue *q, bool nowait) |
3ef28e83 DW |
653 | { |
654 | while (true) { | |
655 | int ret; | |
656 | ||
657 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
658 | return 0; | |
659 | ||
6f3b0e8b | 660 | if (nowait) |
3ef28e83 DW |
661 | return -EBUSY; |
662 | ||
663 | ret = wait_event_interruptible(q->mq_freeze_wq, | |
664 | !atomic_read(&q->mq_freeze_depth) || | |
665 | blk_queue_dying(q)); | |
666 | if (blk_queue_dying(q)) | |
667 | return -ENODEV; | |
668 | if (ret) | |
669 | return ret; | |
670 | } | |
671 | } | |
672 | ||
673 | void blk_queue_exit(struct request_queue *q) | |
674 | { | |
675 | percpu_ref_put(&q->q_usage_counter); | |
676 | } | |
677 | ||
678 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
679 | { | |
680 | struct request_queue *q = | |
681 | container_of(ref, struct request_queue, q_usage_counter); | |
682 | ||
683 | wake_up_all(&q->mq_freeze_wq); | |
684 | } | |
685 | ||
287922eb CH |
686 | static void blk_rq_timed_out_timer(unsigned long data) |
687 | { | |
688 | struct request_queue *q = (struct request_queue *)data; | |
689 | ||
690 | kblockd_schedule_work(&q->timeout_work); | |
691 | } | |
692 | ||
165125e1 | 693 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 694 | { |
165125e1 | 695 | struct request_queue *q; |
e0bf68dd | 696 | int err; |
1946089a | 697 | |
8324aa91 | 698 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 699 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
700 | if (!q) |
701 | return NULL; | |
702 | ||
00380a40 | 703 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 704 | if (q->id < 0) |
3d2936f4 | 705 | goto fail_q; |
a73f730d | 706 | |
54efd50b KO |
707 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0); |
708 | if (!q->bio_split) | |
709 | goto fail_id; | |
710 | ||
0989a025 | 711 | q->backing_dev_info.ra_pages = |
09cbfeaf | 712 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
89e9b9e0 | 713 | q->backing_dev_info.capabilities = BDI_CAP_CGROUP_WRITEBACK; |
d993831f | 714 | q->backing_dev_info.name = "block"; |
5151412d | 715 | q->node = node_id; |
0989a025 | 716 | |
e0bf68dd | 717 | err = bdi_init(&q->backing_dev_info); |
a73f730d | 718 | if (err) |
54efd50b | 719 | goto fail_split; |
e0bf68dd | 720 | |
31373d09 MG |
721 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
722 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 723 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 724 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 725 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 726 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 727 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 728 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 729 | #endif |
3cca6dc1 | 730 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 731 | |
8324aa91 | 732 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 733 | |
483f4afc | 734 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 735 | spin_lock_init(&q->__queue_lock); |
483f4afc | 736 | |
c94a96ac VG |
737 | /* |
738 | * By default initialize queue_lock to internal lock and driver can | |
739 | * override it later if need be. | |
740 | */ | |
741 | q->queue_lock = &q->__queue_lock; | |
742 | ||
b82d4b19 TH |
743 | /* |
744 | * A queue starts its life with bypass turned on to avoid | |
745 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
746 | * init. The initial bypass will be finished when the queue is |
747 | * registered by blk_register_queue(). | |
b82d4b19 TH |
748 | */ |
749 | q->bypass_depth = 1; | |
750 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
751 | ||
320ae51f JA |
752 | init_waitqueue_head(&q->mq_freeze_wq); |
753 | ||
3ef28e83 DW |
754 | /* |
755 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
756 | * See blk_register_queue() for details. | |
757 | */ | |
758 | if (percpu_ref_init(&q->q_usage_counter, | |
759 | blk_queue_usage_counter_release, | |
760 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 761 | goto fail_bdi; |
f51b802c | 762 | |
3ef28e83 DW |
763 | if (blkcg_init_queue(q)) |
764 | goto fail_ref; | |
765 | ||
1da177e4 | 766 | return q; |
a73f730d | 767 | |
3ef28e83 DW |
768 | fail_ref: |
769 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b MP |
770 | fail_bdi: |
771 | bdi_destroy(&q->backing_dev_info); | |
54efd50b KO |
772 | fail_split: |
773 | bioset_free(q->bio_split); | |
a73f730d TH |
774 | fail_id: |
775 | ida_simple_remove(&blk_queue_ida, q->id); | |
776 | fail_q: | |
777 | kmem_cache_free(blk_requestq_cachep, q); | |
778 | return NULL; | |
1da177e4 | 779 | } |
1946089a | 780 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
781 | |
782 | /** | |
783 | * blk_init_queue - prepare a request queue for use with a block device | |
784 | * @rfn: The function to be called to process requests that have been | |
785 | * placed on the queue. | |
786 | * @lock: Request queue spin lock | |
787 | * | |
788 | * Description: | |
789 | * If a block device wishes to use the standard request handling procedures, | |
790 | * which sorts requests and coalesces adjacent requests, then it must | |
791 | * call blk_init_queue(). The function @rfn will be called when there | |
792 | * are requests on the queue that need to be processed. If the device | |
793 | * supports plugging, then @rfn may not be called immediately when requests | |
794 | * are available on the queue, but may be called at some time later instead. | |
795 | * Plugged queues are generally unplugged when a buffer belonging to one | |
796 | * of the requests on the queue is needed, or due to memory pressure. | |
797 | * | |
798 | * @rfn is not required, or even expected, to remove all requests off the | |
799 | * queue, but only as many as it can handle at a time. If it does leave | |
800 | * requests on the queue, it is responsible for arranging that the requests | |
801 | * get dealt with eventually. | |
802 | * | |
803 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
804 | * request queue; this lock will be taken also from interrupt context, so irq |
805 | * disabling is needed for it. | |
1da177e4 | 806 | * |
710027a4 | 807 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
808 | * it didn't succeed. |
809 | * | |
810 | * Note: | |
811 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
812 | * when the block device is deactivated (such as at module unload). | |
813 | **/ | |
1946089a | 814 | |
165125e1 | 815 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 816 | { |
c304a51b | 817 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
818 | } |
819 | EXPORT_SYMBOL(blk_init_queue); | |
820 | ||
165125e1 | 821 | struct request_queue * |
1946089a CL |
822 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
823 | { | |
c86d1b8a | 824 | struct request_queue *uninit_q, *q; |
1da177e4 | 825 | |
c86d1b8a MS |
826 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
827 | if (!uninit_q) | |
828 | return NULL; | |
829 | ||
5151412d | 830 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a | 831 | if (!q) |
7982e90c | 832 | blk_cleanup_queue(uninit_q); |
18741986 | 833 | |
7982e90c | 834 | return q; |
01effb0d MS |
835 | } |
836 | EXPORT_SYMBOL(blk_init_queue_node); | |
837 | ||
dece1635 | 838 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 839 | |
01effb0d MS |
840 | struct request_queue * |
841 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
842 | spinlock_t *lock) | |
01effb0d | 843 | { |
1da177e4 LT |
844 | if (!q) |
845 | return NULL; | |
846 | ||
f70ced09 | 847 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, 0); |
ba483388 | 848 | if (!q->fq) |
7982e90c MS |
849 | return NULL; |
850 | ||
a051661c | 851 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
708f04d2 | 852 | goto fail; |
1da177e4 | 853 | |
287922eb | 854 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
1da177e4 | 855 | q->request_fn = rfn; |
1da177e4 | 856 | q->prep_rq_fn = NULL; |
28018c24 | 857 | q->unprep_rq_fn = NULL; |
60ea8226 | 858 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
859 | |
860 | /* Override internal queue lock with supplied lock pointer */ | |
861 | if (lock) | |
862 | q->queue_lock = lock; | |
1da177e4 | 863 | |
f3b144aa JA |
864 | /* |
865 | * This also sets hw/phys segments, boundary and size | |
866 | */ | |
c20e8de2 | 867 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 868 | |
44ec9542 AS |
869 | q->sg_reserved_size = INT_MAX; |
870 | ||
eb1c160b TS |
871 | /* Protect q->elevator from elevator_change */ |
872 | mutex_lock(&q->sysfs_lock); | |
873 | ||
b82d4b19 | 874 | /* init elevator */ |
eb1c160b TS |
875 | if (elevator_init(q, NULL)) { |
876 | mutex_unlock(&q->sysfs_lock); | |
708f04d2 | 877 | goto fail; |
eb1c160b TS |
878 | } |
879 | ||
880 | mutex_unlock(&q->sysfs_lock); | |
881 | ||
b82d4b19 | 882 | return q; |
708f04d2 DJ |
883 | |
884 | fail: | |
ba483388 | 885 | blk_free_flush_queue(q->fq); |
87760e5e | 886 | wbt_exit(q); |
708f04d2 | 887 | return NULL; |
1da177e4 | 888 | } |
5151412d | 889 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 890 | |
09ac46c4 | 891 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 892 | { |
3f3299d5 | 893 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
894 | __blk_get_queue(q); |
895 | return true; | |
1da177e4 LT |
896 | } |
897 | ||
09ac46c4 | 898 | return false; |
1da177e4 | 899 | } |
d86e0e83 | 900 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 901 | |
5b788ce3 | 902 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 903 | { |
e8064021 | 904 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 905 | elv_put_request(rl->q, rq); |
f1f8cc94 | 906 | if (rq->elv.icq) |
11a3122f | 907 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
908 | } |
909 | ||
5b788ce3 | 910 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
911 | } |
912 | ||
1da177e4 LT |
913 | /* |
914 | * ioc_batching returns true if the ioc is a valid batching request and | |
915 | * should be given priority access to a request. | |
916 | */ | |
165125e1 | 917 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
918 | { |
919 | if (!ioc) | |
920 | return 0; | |
921 | ||
922 | /* | |
923 | * Make sure the process is able to allocate at least 1 request | |
924 | * even if the batch times out, otherwise we could theoretically | |
925 | * lose wakeups. | |
926 | */ | |
927 | return ioc->nr_batch_requests == q->nr_batching || | |
928 | (ioc->nr_batch_requests > 0 | |
929 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
930 | } | |
931 | ||
932 | /* | |
933 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
934 | * will cause the process to be a "batcher" on all queues in the system. This | |
935 | * is the behaviour we want though - once it gets a wakeup it should be given | |
936 | * a nice run. | |
937 | */ | |
165125e1 | 938 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
939 | { |
940 | if (!ioc || ioc_batching(q, ioc)) | |
941 | return; | |
942 | ||
943 | ioc->nr_batch_requests = q->nr_batching; | |
944 | ioc->last_waited = jiffies; | |
945 | } | |
946 | ||
5b788ce3 | 947 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 948 | { |
5b788ce3 | 949 | struct request_queue *q = rl->q; |
1da177e4 | 950 | |
d40f75a0 TH |
951 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
952 | blk_clear_congested(rl, sync); | |
1da177e4 | 953 | |
1faa16d2 JA |
954 | if (rl->count[sync] + 1 <= q->nr_requests) { |
955 | if (waitqueue_active(&rl->wait[sync])) | |
956 | wake_up(&rl->wait[sync]); | |
1da177e4 | 957 | |
5b788ce3 | 958 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
959 | } |
960 | } | |
961 | ||
962 | /* | |
963 | * A request has just been released. Account for it, update the full and | |
964 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
965 | */ | |
e8064021 CH |
966 | static void freed_request(struct request_list *rl, bool sync, |
967 | req_flags_t rq_flags) | |
1da177e4 | 968 | { |
5b788ce3 | 969 | struct request_queue *q = rl->q; |
1da177e4 | 970 | |
8a5ecdd4 | 971 | q->nr_rqs[sync]--; |
1faa16d2 | 972 | rl->count[sync]--; |
e8064021 | 973 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 974 | q->nr_rqs_elvpriv--; |
1da177e4 | 975 | |
5b788ce3 | 976 | __freed_request(rl, sync); |
1da177e4 | 977 | |
1faa16d2 | 978 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 979 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
980 | } |
981 | ||
e3a2b3f9 JA |
982 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
983 | { | |
984 | struct request_list *rl; | |
d40f75a0 | 985 | int on_thresh, off_thresh; |
e3a2b3f9 JA |
986 | |
987 | spin_lock_irq(q->queue_lock); | |
988 | q->nr_requests = nr; | |
989 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
990 | on_thresh = queue_congestion_on_threshold(q); |
991 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 992 | |
d40f75a0 TH |
993 | blk_queue_for_each_rl(rl, q) { |
994 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
995 | blk_set_congested(rl, BLK_RW_SYNC); | |
996 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
997 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 998 | |
d40f75a0 TH |
999 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1000 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1001 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1002 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1003 | |
e3a2b3f9 JA |
1004 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1005 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1006 | } else { | |
1007 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1008 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1009 | } | |
1010 | ||
1011 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1012 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1013 | } else { | |
1014 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1015 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1016 | } | |
1017 | } | |
1018 | ||
1019 | spin_unlock_irq(q->queue_lock); | |
1020 | return 0; | |
1021 | } | |
1022 | ||
9d5a4e94 MS |
1023 | /* |
1024 | * Determine if elevator data should be initialized when allocating the | |
1025 | * request associated with @bio. | |
1026 | */ | |
1027 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
1028 | { | |
1029 | if (!bio) | |
1030 | return true; | |
1031 | ||
1032 | /* | |
1033 | * Flush requests do not use the elevator so skip initialization. | |
1034 | * This allows a request to share the flush and elevator data. | |
1035 | */ | |
1eff9d32 | 1036 | if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) |
9d5a4e94 MS |
1037 | return false; |
1038 | ||
1039 | return true; | |
1040 | } | |
1041 | ||
852c788f TH |
1042 | /** |
1043 | * rq_ioc - determine io_context for request allocation | |
1044 | * @bio: request being allocated is for this bio (can be %NULL) | |
1045 | * | |
1046 | * Determine io_context to use for request allocation for @bio. May return | |
1047 | * %NULL if %current->io_context doesn't exist. | |
1048 | */ | |
1049 | static struct io_context *rq_ioc(struct bio *bio) | |
1050 | { | |
1051 | #ifdef CONFIG_BLK_CGROUP | |
1052 | if (bio && bio->bi_ioc) | |
1053 | return bio->bi_ioc; | |
1054 | #endif | |
1055 | return current->io_context; | |
1056 | } | |
1057 | ||
da8303c6 | 1058 | /** |
a06e05e6 | 1059 | * __get_request - get a free request |
5b788ce3 | 1060 | * @rl: request list to allocate from |
ef295ecf | 1061 | * @op: operation and flags |
da8303c6 TH |
1062 | * @bio: bio to allocate request for (can be %NULL) |
1063 | * @gfp_mask: allocation mask | |
1064 | * | |
1065 | * Get a free request from @q. This function may fail under memory | |
1066 | * pressure or if @q is dead. | |
1067 | * | |
da3dae54 | 1068 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1069 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1070 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1071 | */ |
ef295ecf CH |
1072 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
1073 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1074 | { |
5b788ce3 | 1075 | struct request_queue *q = rl->q; |
b679281a | 1076 | struct request *rq; |
7f4b35d1 TH |
1077 | struct elevator_type *et = q->elevator->type; |
1078 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1079 | struct io_cq *icq = NULL; |
ef295ecf | 1080 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1081 | int may_queue; |
e8064021 | 1082 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1083 | |
3f3299d5 | 1084 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1085 | return ERR_PTR(-ENODEV); |
da8303c6 | 1086 | |
ef295ecf | 1087 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1088 | if (may_queue == ELV_MQUEUE_NO) |
1089 | goto rq_starved; | |
1090 | ||
1faa16d2 JA |
1091 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1092 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1093 | /* |
1094 | * The queue will fill after this allocation, so set | |
1095 | * it as full, and mark this process as "batching". | |
1096 | * This process will be allowed to complete a batch of | |
1097 | * requests, others will be blocked. | |
1098 | */ | |
5b788ce3 | 1099 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1100 | ioc_set_batching(q, ioc); |
5b788ce3 | 1101 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1102 | } else { |
1103 | if (may_queue != ELV_MQUEUE_MUST | |
1104 | && !ioc_batching(q, ioc)) { | |
1105 | /* | |
1106 | * The queue is full and the allocating | |
1107 | * process is not a "batcher", and not | |
1108 | * exempted by the IO scheduler | |
1109 | */ | |
a492f075 | 1110 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1111 | } |
1112 | } | |
1da177e4 | 1113 | } |
d40f75a0 | 1114 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1115 | } |
1116 | ||
082cf69e JA |
1117 | /* |
1118 | * Only allow batching queuers to allocate up to 50% over the defined | |
1119 | * limit of requests, otherwise we could have thousands of requests | |
1120 | * allocated with any setting of ->nr_requests | |
1121 | */ | |
1faa16d2 | 1122 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1123 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1124 | |
8a5ecdd4 | 1125 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1126 | rl->count[is_sync]++; |
1127 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1128 | |
f1f8cc94 TH |
1129 | /* |
1130 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1131 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1132 | * prevent the current elevator from being destroyed until the new |
1133 | * request is freed. This guarantees icq's won't be destroyed and | |
1134 | * makes creating new ones safe. | |
1135 | * | |
1136 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
1137 | * it will be created after releasing queue_lock. | |
1138 | */ | |
d732580b | 1139 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
e8064021 | 1140 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1141 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1142 | if (et->icq_cache && ioc) |
1143 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1144 | } |
cb98fc8b | 1145 | |
f253b86b | 1146 | if (blk_queue_io_stat(q)) |
e8064021 | 1147 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1148 | spin_unlock_irq(q->queue_lock); |
1149 | ||
29e2b09a | 1150 | /* allocate and init request */ |
5b788ce3 | 1151 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1152 | if (!rq) |
b679281a | 1153 | goto fail_alloc; |
1da177e4 | 1154 | |
29e2b09a | 1155 | blk_rq_init(q, rq); |
a051661c | 1156 | blk_rq_set_rl(rq, rl); |
ef295ecf | 1157 | rq->cmd_flags = op; |
e8064021 | 1158 | rq->rq_flags = rq_flags; |
29e2b09a | 1159 | |
aaf7c680 | 1160 | /* init elvpriv */ |
e8064021 | 1161 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1162 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1163 | if (ioc) |
1164 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1165 | if (!icq) |
1166 | goto fail_elvpriv; | |
29e2b09a | 1167 | } |
aaf7c680 TH |
1168 | |
1169 | rq->elv.icq = icq; | |
1170 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1171 | goto fail_elvpriv; | |
1172 | ||
1173 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1174 | if (icq) |
1175 | get_io_context(icq->ioc); | |
1176 | } | |
aaf7c680 | 1177 | out: |
88ee5ef1 JA |
1178 | /* |
1179 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1180 | * OK, if the queue is under the request limit then requests need | |
1181 | * not count toward the nr_batch_requests limit. There will always | |
1182 | * be some limit enforced by BLK_BATCH_TIME. | |
1183 | */ | |
1da177e4 LT |
1184 | if (ioc_batching(q, ioc)) |
1185 | ioc->nr_batch_requests--; | |
6728cb0e | 1186 | |
e6a40b09 | 1187 | trace_block_getrq(q, bio, op); |
1da177e4 | 1188 | return rq; |
b679281a | 1189 | |
aaf7c680 TH |
1190 | fail_elvpriv: |
1191 | /* | |
1192 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1193 | * and may fail indefinitely under memory pressure and thus | |
1194 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1195 | * disturb iosched and blkcg but weird is bettern than dead. | |
1196 | */ | |
7b2b10e0 RE |
1197 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
1198 | __func__, dev_name(q->backing_dev_info.dev)); | |
aaf7c680 | 1199 | |
e8064021 | 1200 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1201 | rq->elv.icq = NULL; |
1202 | ||
1203 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1204 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1205 | spin_unlock_irq(q->queue_lock); |
1206 | goto out; | |
1207 | ||
b679281a TH |
1208 | fail_alloc: |
1209 | /* | |
1210 | * Allocation failed presumably due to memory. Undo anything we | |
1211 | * might have messed up. | |
1212 | * | |
1213 | * Allocating task should really be put onto the front of the wait | |
1214 | * queue, but this is pretty rare. | |
1215 | */ | |
1216 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1217 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1218 | |
1219 | /* | |
1220 | * in the very unlikely event that allocation failed and no | |
1221 | * requests for this direction was pending, mark us starved so that | |
1222 | * freeing of a request in the other direction will notice | |
1223 | * us. another possible fix would be to split the rq mempool into | |
1224 | * READ and WRITE | |
1225 | */ | |
1226 | rq_starved: | |
1227 | if (unlikely(rl->count[is_sync] == 0)) | |
1228 | rl->starved[is_sync] = 1; | |
a492f075 | 1229 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1230 | } |
1231 | ||
da8303c6 | 1232 | /** |
a06e05e6 | 1233 | * get_request - get a free request |
da8303c6 | 1234 | * @q: request_queue to allocate request from |
ef295ecf | 1235 | * @op: operation and flags |
da8303c6 | 1236 | * @bio: bio to allocate request for (can be %NULL) |
a06e05e6 | 1237 | * @gfp_mask: allocation mask |
da8303c6 | 1238 | * |
d0164adc MG |
1239 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, |
1240 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1241 | * |
da3dae54 | 1242 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1243 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1244 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1245 | */ |
ef295ecf CH |
1246 | static struct request *get_request(struct request_queue *q, unsigned int op, |
1247 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1248 | { |
ef295ecf | 1249 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1250 | DEFINE_WAIT(wait); |
a051661c | 1251 | struct request_list *rl; |
1da177e4 | 1252 | struct request *rq; |
a051661c TH |
1253 | |
1254 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
a06e05e6 | 1255 | retry: |
ef295ecf | 1256 | rq = __get_request(rl, op, bio, gfp_mask); |
a492f075 | 1257 | if (!IS_ERR(rq)) |
a06e05e6 | 1258 | return rq; |
1da177e4 | 1259 | |
d0164adc | 1260 | if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) { |
a051661c | 1261 | blk_put_rl(rl); |
a492f075 | 1262 | return rq; |
a051661c | 1263 | } |
1da177e4 | 1264 | |
a06e05e6 TH |
1265 | /* wait on @rl and retry */ |
1266 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1267 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1268 | |
e6a40b09 | 1269 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1270 | |
a06e05e6 TH |
1271 | spin_unlock_irq(q->queue_lock); |
1272 | io_schedule(); | |
d6344532 | 1273 | |
a06e05e6 TH |
1274 | /* |
1275 | * After sleeping, we become a "batching" process and will be able | |
1276 | * to allocate at least one request, and up to a big batch of them | |
1277 | * for a small period time. See ioc_batching, ioc_set_batching | |
1278 | */ | |
a06e05e6 | 1279 | ioc_set_batching(q, current->io_context); |
05caf8db | 1280 | |
a06e05e6 TH |
1281 | spin_lock_irq(q->queue_lock); |
1282 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1283 | |
a06e05e6 | 1284 | goto retry; |
1da177e4 LT |
1285 | } |
1286 | ||
320ae51f JA |
1287 | static struct request *blk_old_get_request(struct request_queue *q, int rw, |
1288 | gfp_t gfp_mask) | |
1da177e4 LT |
1289 | { |
1290 | struct request *rq; | |
1291 | ||
1292 | BUG_ON(rw != READ && rw != WRITE); | |
1293 | ||
7f4b35d1 TH |
1294 | /* create ioc upfront */ |
1295 | create_io_context(gfp_mask, q->node); | |
1296 | ||
d6344532 | 1297 | spin_lock_irq(q->queue_lock); |
ef295ecf | 1298 | rq = get_request(q, rw, NULL, gfp_mask); |
0c4de0f3 | 1299 | if (IS_ERR(rq)) { |
da8303c6 | 1300 | spin_unlock_irq(q->queue_lock); |
0c4de0f3 CH |
1301 | return rq; |
1302 | } | |
1da177e4 | 1303 | |
0c4de0f3 CH |
1304 | /* q->queue_lock is unlocked at this point */ |
1305 | rq->__data_len = 0; | |
1306 | rq->__sector = (sector_t) -1; | |
1307 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1308 | return rq; |
1309 | } | |
320ae51f JA |
1310 | |
1311 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
1312 | { | |
1313 | if (q->mq_ops) | |
6f3b0e8b CH |
1314 | return blk_mq_alloc_request(q, rw, |
1315 | (gfp_mask & __GFP_DIRECT_RECLAIM) ? | |
1316 | 0 : BLK_MQ_REQ_NOWAIT); | |
320ae51f JA |
1317 | else |
1318 | return blk_old_get_request(q, rw, gfp_mask); | |
1319 | } | |
1da177e4 LT |
1320 | EXPORT_SYMBOL(blk_get_request); |
1321 | ||
f27b087b | 1322 | /** |
da3dae54 | 1323 | * blk_rq_set_block_pc - initialize a request to type BLOCK_PC |
f27b087b JA |
1324 | * @rq: request to be initialized |
1325 | * | |
1326 | */ | |
1327 | void blk_rq_set_block_pc(struct request *rq) | |
1328 | { | |
1329 | rq->cmd_type = REQ_TYPE_BLOCK_PC; | |
f27b087b | 1330 | memset(rq->__cmd, 0, sizeof(rq->__cmd)); |
f27b087b JA |
1331 | } |
1332 | EXPORT_SYMBOL(blk_rq_set_block_pc); | |
1333 | ||
1da177e4 LT |
1334 | /** |
1335 | * blk_requeue_request - put a request back on queue | |
1336 | * @q: request queue where request should be inserted | |
1337 | * @rq: request to be inserted | |
1338 | * | |
1339 | * Description: | |
1340 | * Drivers often keep queueing requests until the hardware cannot accept | |
1341 | * more, when that condition happens we need to put the request back | |
1342 | * on the queue. Must be called with queue lock held. | |
1343 | */ | |
165125e1 | 1344 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1345 | { |
242f9dcb JA |
1346 | blk_delete_timer(rq); |
1347 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1348 | trace_block_rq_requeue(q, rq); |
87760e5e | 1349 | wbt_requeue(q->rq_wb, &rq->issue_stat); |
2056a782 | 1350 | |
e8064021 | 1351 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1352 | blk_queue_end_tag(q, rq); |
1353 | ||
ba396a6c JB |
1354 | BUG_ON(blk_queued_rq(rq)); |
1355 | ||
1da177e4 LT |
1356 | elv_requeue_request(q, rq); |
1357 | } | |
1da177e4 LT |
1358 | EXPORT_SYMBOL(blk_requeue_request); |
1359 | ||
73c10101 JA |
1360 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1361 | int where) | |
1362 | { | |
320ae51f | 1363 | blk_account_io_start(rq, true); |
7eaceacc | 1364 | __elv_add_request(q, rq, where); |
73c10101 JA |
1365 | } |
1366 | ||
074a7aca TH |
1367 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1368 | unsigned long now) | |
1369 | { | |
7276d02e JA |
1370 | int inflight; |
1371 | ||
074a7aca TH |
1372 | if (now == part->stamp) |
1373 | return; | |
1374 | ||
7276d02e JA |
1375 | inflight = part_in_flight(part); |
1376 | if (inflight) { | |
074a7aca | 1377 | __part_stat_add(cpu, part, time_in_queue, |
7276d02e | 1378 | inflight * (now - part->stamp)); |
074a7aca TH |
1379 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1380 | } | |
1381 | part->stamp = now; | |
1382 | } | |
1383 | ||
1384 | /** | |
496aa8a9 RD |
1385 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1386 | * @cpu: cpu number for stats access | |
1387 | * @part: target partition | |
1da177e4 LT |
1388 | * |
1389 | * The average IO queue length and utilisation statistics are maintained | |
1390 | * by observing the current state of the queue length and the amount of | |
1391 | * time it has been in this state for. | |
1392 | * | |
1393 | * Normally, that accounting is done on IO completion, but that can result | |
1394 | * in more than a second's worth of IO being accounted for within any one | |
1395 | * second, leading to >100% utilisation. To deal with that, we call this | |
1396 | * function to do a round-off before returning the results when reading | |
1397 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1398 | * the current jiffies and restarts the counters again. | |
1399 | */ | |
c9959059 | 1400 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1401 | { |
1402 | unsigned long now = jiffies; | |
1403 | ||
074a7aca TH |
1404 | if (part->partno) |
1405 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1406 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1407 | } |
074a7aca | 1408 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1409 | |
47fafbc7 | 1410 | #ifdef CONFIG_PM |
c8158819 LM |
1411 | static void blk_pm_put_request(struct request *rq) |
1412 | { | |
e8064021 | 1413 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1414 | pm_runtime_mark_last_busy(rq->q->dev); |
1415 | } | |
1416 | #else | |
1417 | static inline void blk_pm_put_request(struct request *rq) {} | |
1418 | #endif | |
1419 | ||
1da177e4 LT |
1420 | /* |
1421 | * queue lock must be held | |
1422 | */ | |
165125e1 | 1423 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1424 | { |
e8064021 CH |
1425 | req_flags_t rq_flags = req->rq_flags; |
1426 | ||
1da177e4 LT |
1427 | if (unlikely(!q)) |
1428 | return; | |
1da177e4 | 1429 | |
6f5ba581 CH |
1430 | if (q->mq_ops) { |
1431 | blk_mq_free_request(req); | |
1432 | return; | |
1433 | } | |
1434 | ||
c8158819 LM |
1435 | blk_pm_put_request(req); |
1436 | ||
8922e16c TH |
1437 | elv_completed_request(q, req); |
1438 | ||
1cd96c24 BH |
1439 | /* this is a bio leak */ |
1440 | WARN_ON(req->bio != NULL); | |
1441 | ||
87760e5e JA |
1442 | wbt_done(q->rq_wb, &req->issue_stat); |
1443 | ||
1da177e4 LT |
1444 | /* |
1445 | * Request may not have originated from ll_rw_blk. if not, | |
1446 | * it didn't come out of our reserved rq pools | |
1447 | */ | |
e8064021 | 1448 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1449 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1450 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1451 | |
1da177e4 | 1452 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1453 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1454 | |
a051661c | 1455 | blk_free_request(rl, req); |
e8064021 | 1456 | freed_request(rl, sync, rq_flags); |
a051661c | 1457 | blk_put_rl(rl); |
1da177e4 LT |
1458 | } |
1459 | } | |
6e39b69e MC |
1460 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1461 | ||
1da177e4 LT |
1462 | void blk_put_request(struct request *req) |
1463 | { | |
165125e1 | 1464 | struct request_queue *q = req->q; |
8922e16c | 1465 | |
320ae51f JA |
1466 | if (q->mq_ops) |
1467 | blk_mq_free_request(req); | |
1468 | else { | |
1469 | unsigned long flags; | |
1470 | ||
1471 | spin_lock_irqsave(q->queue_lock, flags); | |
1472 | __blk_put_request(q, req); | |
1473 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1474 | } | |
1da177e4 | 1475 | } |
1da177e4 LT |
1476 | EXPORT_SYMBOL(blk_put_request); |
1477 | ||
66ac0280 CH |
1478 | /** |
1479 | * blk_add_request_payload - add a payload to a request | |
1480 | * @rq: request to update | |
1481 | * @page: page backing the payload | |
37e58237 | 1482 | * @offset: offset in page |
66ac0280 CH |
1483 | * @len: length of the payload. |
1484 | * | |
1485 | * This allows to later add a payload to an already submitted request by | |
1486 | * a block driver. The driver needs to take care of freeing the payload | |
1487 | * itself. | |
1488 | * | |
1489 | * Note that this is a quite horrible hack and nothing but handling of | |
1490 | * discard requests should ever use it. | |
1491 | */ | |
1492 | void blk_add_request_payload(struct request *rq, struct page *page, | |
37e58237 | 1493 | int offset, unsigned int len) |
66ac0280 CH |
1494 | { |
1495 | struct bio *bio = rq->bio; | |
1496 | ||
1497 | bio->bi_io_vec->bv_page = page; | |
37e58237 | 1498 | bio->bi_io_vec->bv_offset = offset; |
66ac0280 CH |
1499 | bio->bi_io_vec->bv_len = len; |
1500 | ||
4f024f37 | 1501 | bio->bi_iter.bi_size = len; |
66ac0280 CH |
1502 | bio->bi_vcnt = 1; |
1503 | bio->bi_phys_segments = 1; | |
1504 | ||
1505 | rq->__data_len = rq->resid_len = len; | |
1506 | rq->nr_phys_segments = 1; | |
66ac0280 CH |
1507 | } |
1508 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1509 | ||
320ae51f JA |
1510 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1511 | struct bio *bio) | |
73c10101 | 1512 | { |
1eff9d32 | 1513 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1514 | |
73c10101 JA |
1515 | if (!ll_back_merge_fn(q, req, bio)) |
1516 | return false; | |
1517 | ||
8c1cf6bb | 1518 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1519 | |
1520 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1521 | blk_rq_set_mixed_merge(req); | |
1522 | ||
1523 | req->biotail->bi_next = bio; | |
1524 | req->biotail = bio; | |
4f024f37 | 1525 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1526 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1527 | ||
320ae51f | 1528 | blk_account_io_start(req, false); |
73c10101 JA |
1529 | return true; |
1530 | } | |
1531 | ||
320ae51f JA |
1532 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1533 | struct bio *bio) | |
73c10101 | 1534 | { |
1eff9d32 | 1535 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1536 | |
73c10101 JA |
1537 | if (!ll_front_merge_fn(q, req, bio)) |
1538 | return false; | |
1539 | ||
8c1cf6bb | 1540 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1541 | |
1542 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1543 | blk_rq_set_mixed_merge(req); | |
1544 | ||
73c10101 JA |
1545 | bio->bi_next = req->bio; |
1546 | req->bio = bio; | |
1547 | ||
4f024f37 KO |
1548 | req->__sector = bio->bi_iter.bi_sector; |
1549 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1550 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1551 | ||
320ae51f | 1552 | blk_account_io_start(req, false); |
73c10101 JA |
1553 | return true; |
1554 | } | |
1555 | ||
bd87b589 | 1556 | /** |
320ae51f | 1557 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1558 | * @q: request_queue new bio is being queued at |
1559 | * @bio: new bio being queued | |
1560 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1561 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1562 | * another request associated with @q is found on the plug list | |
1563 | * (optional, may be %NULL) | |
bd87b589 TH |
1564 | * |
1565 | * Determine whether @bio being queued on @q can be merged with a request | |
1566 | * on %current's plugged list. Returns %true if merge was successful, | |
1567 | * otherwise %false. | |
1568 | * | |
07c2bd37 TH |
1569 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1570 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1571 | * than scheduling, and the request, while may have elvpriv data, is not | |
1572 | * added on the elevator at this point. In addition, we don't have | |
1573 | * reliable access to the elevator outside queue lock. Only check basic | |
1574 | * merging parameters without querying the elevator. | |
da41a589 RE |
1575 | * |
1576 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1577 | */ |
320ae51f | 1578 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1579 | unsigned int *request_count, |
1580 | struct request **same_queue_rq) | |
73c10101 JA |
1581 | { |
1582 | struct blk_plug *plug; | |
1583 | struct request *rq; | |
1584 | bool ret = false; | |
92f399c7 | 1585 | struct list_head *plug_list; |
73c10101 | 1586 | |
bd87b589 | 1587 | plug = current->plug; |
73c10101 JA |
1588 | if (!plug) |
1589 | goto out; | |
56ebdaf2 | 1590 | *request_count = 0; |
73c10101 | 1591 | |
92f399c7 SL |
1592 | if (q->mq_ops) |
1593 | plug_list = &plug->mq_list; | |
1594 | else | |
1595 | plug_list = &plug->list; | |
1596 | ||
1597 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
73c10101 JA |
1598 | int el_ret; |
1599 | ||
5b3f341f | 1600 | if (rq->q == q) { |
1b2e19f1 | 1601 | (*request_count)++; |
5b3f341f SL |
1602 | /* |
1603 | * Only blk-mq multiple hardware queues case checks the | |
1604 | * rq in the same queue, there should be only one such | |
1605 | * rq in a queue | |
1606 | **/ | |
1607 | if (same_queue_rq) | |
1608 | *same_queue_rq = rq; | |
1609 | } | |
56ebdaf2 | 1610 | |
07c2bd37 | 1611 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1612 | continue; |
1613 | ||
050c8ea8 | 1614 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1615 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1616 | ret = bio_attempt_back_merge(q, rq, bio); | |
1617 | if (ret) | |
1618 | break; | |
1619 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1620 | ret = bio_attempt_front_merge(q, rq, bio); | |
1621 | if (ret) | |
1622 | break; | |
1623 | } | |
1624 | } | |
1625 | out: | |
1626 | return ret; | |
1627 | } | |
1628 | ||
0809e3ac JM |
1629 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1630 | { | |
1631 | struct blk_plug *plug; | |
1632 | struct request *rq; | |
1633 | struct list_head *plug_list; | |
1634 | unsigned int ret = 0; | |
1635 | ||
1636 | plug = current->plug; | |
1637 | if (!plug) | |
1638 | goto out; | |
1639 | ||
1640 | if (q->mq_ops) | |
1641 | plug_list = &plug->mq_list; | |
1642 | else | |
1643 | plug_list = &plug->list; | |
1644 | ||
1645 | list_for_each_entry(rq, plug_list, queuelist) { | |
1646 | if (rq->q == q) | |
1647 | ret++; | |
1648 | } | |
1649 | out: | |
1650 | return ret; | |
1651 | } | |
1652 | ||
86db1e29 | 1653 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1654 | { |
4aff5e23 | 1655 | req->cmd_type = REQ_TYPE_FS; |
1eff9d32 | 1656 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1657 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1658 | |
52d9e675 | 1659 | req->errors = 0; |
4f024f37 | 1660 | req->__sector = bio->bi_iter.bi_sector; |
52d9e675 | 1661 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1662 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1663 | } |
1664 | ||
dece1635 | 1665 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1666 | { |
73c10101 | 1667 | struct blk_plug *plug; |
ef295ecf | 1668 | int el_ret, where = ELEVATOR_INSERT_SORT; |
73c10101 | 1669 | struct request *req; |
56ebdaf2 | 1670 | unsigned int request_count = 0; |
87760e5e | 1671 | unsigned int wb_acct; |
1da177e4 | 1672 | |
1da177e4 LT |
1673 | /* |
1674 | * low level driver can indicate that it wants pages above a | |
1675 | * certain limit bounced to low memory (ie for highmem, or even | |
1676 | * ISA dma in theory) | |
1677 | */ | |
1678 | blk_queue_bounce(q, &bio); | |
1679 | ||
23688bf4 JN |
1680 | blk_queue_split(q, &bio, q->bio_split); |
1681 | ||
ffecfd1a | 1682 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) { |
4246a0b6 CH |
1683 | bio->bi_error = -EIO; |
1684 | bio_endio(bio); | |
dece1635 | 1685 | return BLK_QC_T_NONE; |
ffecfd1a DW |
1686 | } |
1687 | ||
1eff9d32 | 1688 | if (bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) { |
73c10101 | 1689 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1690 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1691 | goto get_rq; |
1692 | } | |
1693 | ||
73c10101 JA |
1694 | /* |
1695 | * Check if we can merge with the plugged list before grabbing | |
1696 | * any locks. | |
1697 | */ | |
0809e3ac JM |
1698 | if (!blk_queue_nomerges(q)) { |
1699 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 1700 | return BLK_QC_T_NONE; |
0809e3ac JM |
1701 | } else |
1702 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 1703 | |
73c10101 | 1704 | spin_lock_irq(q->queue_lock); |
2056a782 | 1705 | |
73c10101 JA |
1706 | el_ret = elv_merge(q, &req, bio); |
1707 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1708 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1709 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1710 | if (!attempt_back_merge(q, req)) |
1711 | elv_merged_request(q, req, el_ret); | |
1712 | goto out_unlock; | |
1713 | } | |
1714 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1715 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1716 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1717 | if (!attempt_front_merge(q, req)) |
1718 | elv_merged_request(q, req, el_ret); | |
1719 | goto out_unlock; | |
80a761fd | 1720 | } |
1da177e4 LT |
1721 | } |
1722 | ||
450991bc | 1723 | get_rq: |
87760e5e JA |
1724 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); |
1725 | ||
1da177e4 | 1726 | /* |
450991bc | 1727 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1728 | * Returns with the queue unlocked. |
450991bc | 1729 | */ |
ef295ecf | 1730 | req = get_request(q, bio->bi_opf, bio, GFP_NOIO); |
a492f075 | 1731 | if (IS_ERR(req)) { |
87760e5e | 1732 | __wbt_done(q->rq_wb, wb_acct); |
4246a0b6 CH |
1733 | bio->bi_error = PTR_ERR(req); |
1734 | bio_endio(bio); | |
da8303c6 TH |
1735 | goto out_unlock; |
1736 | } | |
d6344532 | 1737 | |
87760e5e JA |
1738 | wbt_track(&req->issue_stat, wb_acct); |
1739 | ||
450991bc NP |
1740 | /* |
1741 | * After dropping the lock and possibly sleeping here, our request | |
1742 | * may now be mergeable after it had proven unmergeable (above). | |
1743 | * We don't worry about that case for efficiency. It won't happen | |
1744 | * often, and the elevators are able to handle it. | |
1da177e4 | 1745 | */ |
52d9e675 | 1746 | init_request_from_bio(req, bio); |
1da177e4 | 1747 | |
9562ad9a | 1748 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1749 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1750 | |
1751 | plug = current->plug; | |
721a9602 | 1752 | if (plug) { |
dc6d36c9 JA |
1753 | /* |
1754 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1755 | * of a plug trace. |
0a6219a9 ML |
1756 | * |
1757 | * @request_count may become stale because of schedule | |
1758 | * out, so check plug list again. | |
dc6d36c9 | 1759 | */ |
0a6219a9 | 1760 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 1761 | trace_block_plug(q); |
3540d5e8 | 1762 | else { |
50d24c34 SL |
1763 | struct request *last = list_entry_rq(plug->list.prev); |
1764 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1765 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 1766 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1767 | trace_block_plug(q); |
1768 | } | |
73c10101 | 1769 | } |
73c10101 | 1770 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 1771 | blk_account_io_start(req, true); |
73c10101 JA |
1772 | } else { |
1773 | spin_lock_irq(q->queue_lock); | |
1774 | add_acct_request(q, req, where); | |
24ecfbe2 | 1775 | __blk_run_queue(q); |
73c10101 JA |
1776 | out_unlock: |
1777 | spin_unlock_irq(q->queue_lock); | |
1778 | } | |
dece1635 JA |
1779 | |
1780 | return BLK_QC_T_NONE; | |
1da177e4 LT |
1781 | } |
1782 | ||
1783 | /* | |
1784 | * If bio->bi_dev is a partition, remap the location | |
1785 | */ | |
1786 | static inline void blk_partition_remap(struct bio *bio) | |
1787 | { | |
1788 | struct block_device *bdev = bio->bi_bdev; | |
1789 | ||
778889d8 ST |
1790 | /* |
1791 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
1792 | * Include a test for the reset op code and perform the remap if needed. | |
1793 | */ | |
1794 | if (bdev != bdev->bd_contains && | |
1795 | (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) { | |
1da177e4 LT |
1796 | struct hd_struct *p = bdev->bd_part; |
1797 | ||
4f024f37 | 1798 | bio->bi_iter.bi_sector += p->start_sect; |
1da177e4 | 1799 | bio->bi_bdev = bdev->bd_contains; |
c7149d6b | 1800 | |
d07335e5 MS |
1801 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1802 | bdev->bd_dev, | |
4f024f37 | 1803 | bio->bi_iter.bi_sector - p->start_sect); |
1da177e4 LT |
1804 | } |
1805 | } | |
1806 | ||
1da177e4 LT |
1807 | static void handle_bad_sector(struct bio *bio) |
1808 | { | |
1809 | char b[BDEVNAME_SIZE]; | |
1810 | ||
1811 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 1812 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
1da177e4 | 1813 | bdevname(bio->bi_bdev, b), |
1eff9d32 | 1814 | bio->bi_opf, |
f73a1c7d | 1815 | (unsigned long long)bio_end_sector(bio), |
77304d2a | 1816 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1817 | } |
1818 | ||
c17bb495 AM |
1819 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1820 | ||
1821 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1822 | ||
1823 | static int __init setup_fail_make_request(char *str) | |
1824 | { | |
1825 | return setup_fault_attr(&fail_make_request, str); | |
1826 | } | |
1827 | __setup("fail_make_request=", setup_fail_make_request); | |
1828 | ||
b2c9cd37 | 1829 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1830 | { |
b2c9cd37 | 1831 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1832 | } |
1833 | ||
1834 | static int __init fail_make_request_debugfs(void) | |
1835 | { | |
dd48c085 AM |
1836 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1837 | NULL, &fail_make_request); | |
1838 | ||
21f9fcd8 | 1839 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
1840 | } |
1841 | ||
1842 | late_initcall(fail_make_request_debugfs); | |
1843 | ||
1844 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1845 | ||
b2c9cd37 AM |
1846 | static inline bool should_fail_request(struct hd_struct *part, |
1847 | unsigned int bytes) | |
c17bb495 | 1848 | { |
b2c9cd37 | 1849 | return false; |
c17bb495 AM |
1850 | } |
1851 | ||
1852 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1853 | ||
c07e2b41 JA |
1854 | /* |
1855 | * Check whether this bio extends beyond the end of the device. | |
1856 | */ | |
1857 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1858 | { | |
1859 | sector_t maxsector; | |
1860 | ||
1861 | if (!nr_sectors) | |
1862 | return 0; | |
1863 | ||
1864 | /* Test device or partition size, when known. */ | |
77304d2a | 1865 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 | 1866 | if (maxsector) { |
4f024f37 | 1867 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
1868 | |
1869 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1870 | /* | |
1871 | * This may well happen - the kernel calls bread() | |
1872 | * without checking the size of the device, e.g., when | |
1873 | * mounting a device. | |
1874 | */ | |
1875 | handle_bad_sector(bio); | |
1876 | return 1; | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
27a84d54 CH |
1883 | static noinline_for_stack bool |
1884 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1885 | { |
165125e1 | 1886 | struct request_queue *q; |
5a7bbad2 | 1887 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1888 | int err = -EIO; |
5a7bbad2 CH |
1889 | char b[BDEVNAME_SIZE]; |
1890 | struct hd_struct *part; | |
1da177e4 LT |
1891 | |
1892 | might_sleep(); | |
1da177e4 | 1893 | |
c07e2b41 JA |
1894 | if (bio_check_eod(bio, nr_sectors)) |
1895 | goto end_io; | |
1da177e4 | 1896 | |
5a7bbad2 CH |
1897 | q = bdev_get_queue(bio->bi_bdev); |
1898 | if (unlikely(!q)) { | |
1899 | printk(KERN_ERR | |
1900 | "generic_make_request: Trying to access " | |
1901 | "nonexistent block-device %s (%Lu)\n", | |
1902 | bdevname(bio->bi_bdev, b), | |
4f024f37 | 1903 | (long long) bio->bi_iter.bi_sector); |
5a7bbad2 CH |
1904 | goto end_io; |
1905 | } | |
c17bb495 | 1906 | |
5a7bbad2 | 1907 | part = bio->bi_bdev->bd_part; |
4f024f37 | 1908 | if (should_fail_request(part, bio->bi_iter.bi_size) || |
5a7bbad2 | 1909 | should_fail_request(&part_to_disk(part)->part0, |
4f024f37 | 1910 | bio->bi_iter.bi_size)) |
5a7bbad2 | 1911 | goto end_io; |
2056a782 | 1912 | |
5a7bbad2 CH |
1913 | /* |
1914 | * If this device has partitions, remap block n | |
1915 | * of partition p to block n+start(p) of the disk. | |
1916 | */ | |
1917 | blk_partition_remap(bio); | |
2056a782 | 1918 | |
5a7bbad2 CH |
1919 | if (bio_check_eod(bio, nr_sectors)) |
1920 | goto end_io; | |
1e87901e | 1921 | |
5a7bbad2 CH |
1922 | /* |
1923 | * Filter flush bio's early so that make_request based | |
1924 | * drivers without flush support don't have to worry | |
1925 | * about them. | |
1926 | */ | |
1eff9d32 | 1927 | if ((bio->bi_opf & (REQ_PREFLUSH | REQ_FUA)) && |
c888a8f9 | 1928 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 1929 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 CH |
1930 | if (!nr_sectors) { |
1931 | err = 0; | |
51fd77bd JA |
1932 | goto end_io; |
1933 | } | |
5a7bbad2 | 1934 | } |
5ddfe969 | 1935 | |
288dab8a CH |
1936 | switch (bio_op(bio)) { |
1937 | case REQ_OP_DISCARD: | |
1938 | if (!blk_queue_discard(q)) | |
1939 | goto not_supported; | |
1940 | break; | |
1941 | case REQ_OP_SECURE_ERASE: | |
1942 | if (!blk_queue_secure_erase(q)) | |
1943 | goto not_supported; | |
1944 | break; | |
1945 | case REQ_OP_WRITE_SAME: | |
1946 | if (!bdev_write_same(bio->bi_bdev)) | |
1947 | goto not_supported; | |
58886785 | 1948 | break; |
2d253440 ST |
1949 | case REQ_OP_ZONE_REPORT: |
1950 | case REQ_OP_ZONE_RESET: | |
1951 | if (!bdev_is_zoned(bio->bi_bdev)) | |
1952 | goto not_supported; | |
288dab8a | 1953 | break; |
a6f0788e CK |
1954 | case REQ_OP_WRITE_ZEROES: |
1955 | if (!bdev_write_zeroes_sectors(bio->bi_bdev)) | |
1956 | goto not_supported; | |
1957 | break; | |
288dab8a CH |
1958 | default: |
1959 | break; | |
5a7bbad2 | 1960 | } |
01edede4 | 1961 | |
7f4b35d1 TH |
1962 | /* |
1963 | * Various block parts want %current->io_context and lazy ioc | |
1964 | * allocation ends up trading a lot of pain for a small amount of | |
1965 | * memory. Just allocate it upfront. This may fail and block | |
1966 | * layer knows how to live with it. | |
1967 | */ | |
1968 | create_io_context(GFP_ATOMIC, q->node); | |
1969 | ||
ae118896 TH |
1970 | if (!blkcg_bio_issue_check(q, bio)) |
1971 | return false; | |
27a84d54 | 1972 | |
5a7bbad2 | 1973 | trace_block_bio_queue(q, bio); |
27a84d54 | 1974 | return true; |
a7384677 | 1975 | |
288dab8a CH |
1976 | not_supported: |
1977 | err = -EOPNOTSUPP; | |
a7384677 | 1978 | end_io: |
4246a0b6 CH |
1979 | bio->bi_error = err; |
1980 | bio_endio(bio); | |
27a84d54 | 1981 | return false; |
1da177e4 LT |
1982 | } |
1983 | ||
27a84d54 CH |
1984 | /** |
1985 | * generic_make_request - hand a buffer to its device driver for I/O | |
1986 | * @bio: The bio describing the location in memory and on the device. | |
1987 | * | |
1988 | * generic_make_request() is used to make I/O requests of block | |
1989 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1990 | * to be done. | |
1991 | * | |
1992 | * generic_make_request() does not return any status. The | |
1993 | * success/failure status of the request, along with notification of | |
1994 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1995 | * function described (one day) else where. | |
1996 | * | |
1997 | * The caller of generic_make_request must make sure that bi_io_vec | |
1998 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1999 | * set to describe the device address, and the | |
2000 | * bi_end_io and optionally bi_private are set to describe how | |
2001 | * completion notification should be signaled. | |
2002 | * | |
2003 | * generic_make_request and the drivers it calls may use bi_next if this | |
2004 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2005 | * a lower device by calling into generic_make_request recursively, which | |
2006 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 2007 | */ |
dece1635 | 2008 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 2009 | { |
bddd87c7 | 2010 | struct bio_list bio_list_on_stack; |
dece1635 | 2011 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 2012 | |
27a84d54 | 2013 | if (!generic_make_request_checks(bio)) |
dece1635 | 2014 | goto out; |
27a84d54 CH |
2015 | |
2016 | /* | |
2017 | * We only want one ->make_request_fn to be active at a time, else | |
2018 | * stack usage with stacked devices could be a problem. So use | |
2019 | * current->bio_list to keep a list of requests submited by a | |
2020 | * make_request_fn function. current->bio_list is also used as a | |
2021 | * flag to say if generic_make_request is currently active in this | |
2022 | * task or not. If it is NULL, then no make_request is active. If | |
2023 | * it is non-NULL, then a make_request is active, and new requests | |
2024 | * should be added at the tail | |
2025 | */ | |
bddd87c7 | 2026 | if (current->bio_list) { |
bddd87c7 | 2027 | bio_list_add(current->bio_list, bio); |
dece1635 | 2028 | goto out; |
d89d8796 | 2029 | } |
27a84d54 | 2030 | |
d89d8796 NB |
2031 | /* following loop may be a bit non-obvious, and so deserves some |
2032 | * explanation. | |
2033 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2034 | * ensure that) so we have a list with a single bio. | |
2035 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2036 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2037 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2038 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2039 | * through a recursive call to generic_make_request. If it |
2040 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2041 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2042 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2043 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2044 | */ |
2045 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
2046 | bio_list_init(&bio_list_on_stack); |
2047 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 2048 | do { |
27a84d54 CH |
2049 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
2050 | ||
6f3b0e8b | 2051 | if (likely(blk_queue_enter(q, false) == 0)) { |
dece1635 | 2052 | ret = q->make_request_fn(q, bio); |
3ef28e83 DW |
2053 | |
2054 | blk_queue_exit(q); | |
27a84d54 | 2055 | |
3ef28e83 DW |
2056 | bio = bio_list_pop(current->bio_list); |
2057 | } else { | |
2058 | struct bio *bio_next = bio_list_pop(current->bio_list); | |
2059 | ||
2060 | bio_io_error(bio); | |
2061 | bio = bio_next; | |
2062 | } | |
d89d8796 | 2063 | } while (bio); |
bddd87c7 | 2064 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2065 | |
2066 | out: | |
2067 | return ret; | |
d89d8796 | 2068 | } |
1da177e4 LT |
2069 | EXPORT_SYMBOL(generic_make_request); |
2070 | ||
2071 | /** | |
710027a4 | 2072 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2073 | * @bio: The &struct bio which describes the I/O |
2074 | * | |
2075 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2076 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2077 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2078 | * |
2079 | */ | |
4e49ea4a | 2080 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2081 | { |
bf2de6f5 JA |
2082 | /* |
2083 | * If it's a regular read/write or a barrier with data attached, | |
2084 | * go through the normal accounting stuff before submission. | |
2085 | */ | |
e2a60da7 | 2086 | if (bio_has_data(bio)) { |
4363ac7c MP |
2087 | unsigned int count; |
2088 | ||
95fe6c1a | 2089 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
4363ac7c MP |
2090 | count = bdev_logical_block_size(bio->bi_bdev) >> 9; |
2091 | else | |
2092 | count = bio_sectors(bio); | |
2093 | ||
a8ebb056 | 2094 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2095 | count_vm_events(PGPGOUT, count); |
2096 | } else { | |
4f024f37 | 2097 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2098 | count_vm_events(PGPGIN, count); |
2099 | } | |
2100 | ||
2101 | if (unlikely(block_dump)) { | |
2102 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2103 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2104 | current->comm, task_pid_nr(current), |
a8ebb056 | 2105 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2106 | (unsigned long long)bio->bi_iter.bi_sector, |
8dcbdc74 SM |
2107 | bdevname(bio->bi_bdev, b), |
2108 | count); | |
bf2de6f5 | 2109 | } |
1da177e4 LT |
2110 | } |
2111 | ||
dece1635 | 2112 | return generic_make_request(bio); |
1da177e4 | 2113 | } |
1da177e4 LT |
2114 | EXPORT_SYMBOL(submit_bio); |
2115 | ||
82124d60 | 2116 | /** |
bf4e6b4e HR |
2117 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2118 | * for new the queue limits | |
82124d60 KU |
2119 | * @q: the queue |
2120 | * @rq: the request being checked | |
2121 | * | |
2122 | * Description: | |
2123 | * @rq may have been made based on weaker limitations of upper-level queues | |
2124 | * in request stacking drivers, and it may violate the limitation of @q. | |
2125 | * Since the block layer and the underlying device driver trust @rq | |
2126 | * after it is inserted to @q, it should be checked against @q before | |
2127 | * the insertion using this generic function. | |
2128 | * | |
82124d60 | 2129 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2130 | * limits when retrying requests on other queues. Those requests need |
2131 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2132 | */ |
bf4e6b4e HR |
2133 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2134 | struct request *rq) | |
82124d60 | 2135 | { |
8fe0d473 | 2136 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2137 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2138 | return -EIO; | |
2139 | } | |
2140 | ||
2141 | /* | |
2142 | * queue's settings related to segment counting like q->bounce_pfn | |
2143 | * may differ from that of other stacking queues. | |
2144 | * Recalculate it to check the request correctly on this queue's | |
2145 | * limitation. | |
2146 | */ | |
2147 | blk_recalc_rq_segments(rq); | |
8a78362c | 2148 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2149 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2150 | return -EIO; | |
2151 | } | |
2152 | ||
2153 | return 0; | |
2154 | } | |
82124d60 KU |
2155 | |
2156 | /** | |
2157 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2158 | * @q: the queue to submit the request | |
2159 | * @rq: the request being queued | |
2160 | */ | |
2161 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
2162 | { | |
2163 | unsigned long flags; | |
4853abaa | 2164 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2165 | |
bf4e6b4e | 2166 | if (blk_cloned_rq_check_limits(q, rq)) |
82124d60 KU |
2167 | return -EIO; |
2168 | ||
b2c9cd37 AM |
2169 | if (rq->rq_disk && |
2170 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 2171 | return -EIO; |
82124d60 | 2172 | |
7fb4898e KB |
2173 | if (q->mq_ops) { |
2174 | if (blk_queue_io_stat(q)) | |
2175 | blk_account_io_start(rq, true); | |
6acfe68b | 2176 | blk_mq_insert_request(rq, false, true, false); |
7fb4898e KB |
2177 | return 0; |
2178 | } | |
2179 | ||
82124d60 | 2180 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2181 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
2182 | spin_unlock_irqrestore(q->queue_lock, flags); |
2183 | return -ENODEV; | |
2184 | } | |
82124d60 KU |
2185 | |
2186 | /* | |
2187 | * Submitting request must be dequeued before calling this function | |
2188 | * because it will be linked to another request_queue | |
2189 | */ | |
2190 | BUG_ON(blk_queued_rq(rq)); | |
2191 | ||
28a8f0d3 | 2192 | if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA)) |
4853abaa JM |
2193 | where = ELEVATOR_INSERT_FLUSH; |
2194 | ||
2195 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2196 | if (where == ELEVATOR_INSERT_FLUSH) |
2197 | __blk_run_queue(q); | |
82124d60 KU |
2198 | spin_unlock_irqrestore(q->queue_lock, flags); |
2199 | ||
2200 | return 0; | |
2201 | } | |
2202 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2203 | ||
80a761fd TH |
2204 | /** |
2205 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2206 | * @rq: request to examine | |
2207 | * | |
2208 | * Description: | |
2209 | * A request could be merge of IOs which require different failure | |
2210 | * handling. This function determines the number of bytes which | |
2211 | * can be failed from the beginning of the request without | |
2212 | * crossing into area which need to be retried further. | |
2213 | * | |
2214 | * Return: | |
2215 | * The number of bytes to fail. | |
2216 | * | |
2217 | * Context: | |
2218 | * queue_lock must be held. | |
2219 | */ | |
2220 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2221 | { | |
2222 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2223 | unsigned int bytes = 0; | |
2224 | struct bio *bio; | |
2225 | ||
e8064021 | 2226 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2227 | return blk_rq_bytes(rq); |
2228 | ||
2229 | /* | |
2230 | * Currently the only 'mixing' which can happen is between | |
2231 | * different fastfail types. We can safely fail portions | |
2232 | * which have all the failfast bits that the first one has - | |
2233 | * the ones which are at least as eager to fail as the first | |
2234 | * one. | |
2235 | */ | |
2236 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2237 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2238 | break; |
4f024f37 | 2239 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2240 | } |
2241 | ||
2242 | /* this could lead to infinite loop */ | |
2243 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2244 | return bytes; | |
2245 | } | |
2246 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2247 | ||
320ae51f | 2248 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2249 | { |
c2553b58 | 2250 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2251 | const int rw = rq_data_dir(req); |
2252 | struct hd_struct *part; | |
2253 | int cpu; | |
2254 | ||
2255 | cpu = part_stat_lock(); | |
09e099d4 | 2256 | part = req->part; |
bc58ba94 JA |
2257 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2258 | part_stat_unlock(); | |
2259 | } | |
2260 | } | |
2261 | ||
320ae51f | 2262 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2263 | { |
bc58ba94 | 2264 | /* |
dd4c133f TH |
2265 | * Account IO completion. flush_rq isn't accounted as a |
2266 | * normal IO on queueing nor completion. Accounting the | |
2267 | * containing request is enough. | |
bc58ba94 | 2268 | */ |
e8064021 | 2269 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
bc58ba94 JA |
2270 | unsigned long duration = jiffies - req->start_time; |
2271 | const int rw = rq_data_dir(req); | |
2272 | struct hd_struct *part; | |
2273 | int cpu; | |
2274 | ||
2275 | cpu = part_stat_lock(); | |
09e099d4 | 2276 | part = req->part; |
bc58ba94 JA |
2277 | |
2278 | part_stat_inc(cpu, part, ios[rw]); | |
2279 | part_stat_add(cpu, part, ticks[rw], duration); | |
2280 | part_round_stats(cpu, part); | |
316d315b | 2281 | part_dec_in_flight(part, rw); |
bc58ba94 | 2282 | |
6c23a968 | 2283 | hd_struct_put(part); |
bc58ba94 JA |
2284 | part_stat_unlock(); |
2285 | } | |
2286 | } | |
2287 | ||
47fafbc7 | 2288 | #ifdef CONFIG_PM |
c8158819 LM |
2289 | /* |
2290 | * Don't process normal requests when queue is suspended | |
2291 | * or in the process of suspending/resuming | |
2292 | */ | |
2293 | static struct request *blk_pm_peek_request(struct request_queue *q, | |
2294 | struct request *rq) | |
2295 | { | |
2296 | if (q->dev && (q->rpm_status == RPM_SUSPENDED || | |
e8064021 | 2297 | (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM)))) |
c8158819 LM |
2298 | return NULL; |
2299 | else | |
2300 | return rq; | |
2301 | } | |
2302 | #else | |
2303 | static inline struct request *blk_pm_peek_request(struct request_queue *q, | |
2304 | struct request *rq) | |
2305 | { | |
2306 | return rq; | |
2307 | } | |
2308 | #endif | |
2309 | ||
320ae51f JA |
2310 | void blk_account_io_start(struct request *rq, bool new_io) |
2311 | { | |
2312 | struct hd_struct *part; | |
2313 | int rw = rq_data_dir(rq); | |
2314 | int cpu; | |
2315 | ||
2316 | if (!blk_do_io_stat(rq)) | |
2317 | return; | |
2318 | ||
2319 | cpu = part_stat_lock(); | |
2320 | ||
2321 | if (!new_io) { | |
2322 | part = rq->part; | |
2323 | part_stat_inc(cpu, part, merges[rw]); | |
2324 | } else { | |
2325 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2326 | if (!hd_struct_try_get(part)) { | |
2327 | /* | |
2328 | * The partition is already being removed, | |
2329 | * the request will be accounted on the disk only | |
2330 | * | |
2331 | * We take a reference on disk->part0 although that | |
2332 | * partition will never be deleted, so we can treat | |
2333 | * it as any other partition. | |
2334 | */ | |
2335 | part = &rq->rq_disk->part0; | |
2336 | hd_struct_get(part); | |
2337 | } | |
2338 | part_round_stats(cpu, part); | |
2339 | part_inc_in_flight(part, rw); | |
2340 | rq->part = part; | |
2341 | } | |
2342 | ||
2343 | part_stat_unlock(); | |
2344 | } | |
2345 | ||
3bcddeac | 2346 | /** |
9934c8c0 TH |
2347 | * blk_peek_request - peek at the top of a request queue |
2348 | * @q: request queue to peek at | |
2349 | * | |
2350 | * Description: | |
2351 | * Return the request at the top of @q. The returned request | |
2352 | * should be started using blk_start_request() before LLD starts | |
2353 | * processing it. | |
2354 | * | |
2355 | * Return: | |
2356 | * Pointer to the request at the top of @q if available. Null | |
2357 | * otherwise. | |
2358 | * | |
2359 | * Context: | |
2360 | * queue_lock must be held. | |
2361 | */ | |
2362 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2363 | { |
2364 | struct request *rq; | |
2365 | int ret; | |
2366 | ||
2367 | while ((rq = __elv_next_request(q)) != NULL) { | |
c8158819 LM |
2368 | |
2369 | rq = blk_pm_peek_request(q, rq); | |
2370 | if (!rq) | |
2371 | break; | |
2372 | ||
e8064021 | 2373 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2374 | /* |
2375 | * This is the first time the device driver | |
2376 | * sees this request (possibly after | |
2377 | * requeueing). Notify IO scheduler. | |
2378 | */ | |
e8064021 | 2379 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2380 | elv_activate_rq(q, rq); |
2381 | ||
2382 | /* | |
2383 | * just mark as started even if we don't start | |
2384 | * it, a request that has been delayed should | |
2385 | * not be passed by new incoming requests | |
2386 | */ | |
e8064021 | 2387 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2388 | trace_block_rq_issue(q, rq); |
2389 | } | |
2390 | ||
2391 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2392 | q->end_sector = rq_end_sector(rq); | |
2393 | q->boundary_rq = NULL; | |
2394 | } | |
2395 | ||
e8064021 | 2396 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2397 | break; |
2398 | ||
2e46e8b2 | 2399 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2400 | /* |
2401 | * make sure space for the drain appears we | |
2402 | * know we can do this because max_hw_segments | |
2403 | * has been adjusted to be one fewer than the | |
2404 | * device can handle | |
2405 | */ | |
2406 | rq->nr_phys_segments++; | |
2407 | } | |
2408 | ||
2409 | if (!q->prep_rq_fn) | |
2410 | break; | |
2411 | ||
2412 | ret = q->prep_rq_fn(q, rq); | |
2413 | if (ret == BLKPREP_OK) { | |
2414 | break; | |
2415 | } else if (ret == BLKPREP_DEFER) { | |
2416 | /* | |
2417 | * the request may have been (partially) prepped. | |
2418 | * we need to keep this request in the front to | |
e8064021 | 2419 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2420 | * prevent other fs requests from passing this one. |
2421 | */ | |
2e46e8b2 | 2422 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2423 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2424 | /* |
2425 | * remove the space for the drain we added | |
2426 | * so that we don't add it again | |
2427 | */ | |
2428 | --rq->nr_phys_segments; | |
2429 | } | |
2430 | ||
2431 | rq = NULL; | |
2432 | break; | |
0fb5b1fb MP |
2433 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
2434 | int err = (ret == BLKPREP_INVALID) ? -EREMOTEIO : -EIO; | |
2435 | ||
e8064021 | 2436 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2437 | /* |
2438 | * Mark this request as started so we don't trigger | |
2439 | * any debug logic in the end I/O path. | |
2440 | */ | |
2441 | blk_start_request(rq); | |
0fb5b1fb | 2442 | __blk_end_request_all(rq, err); |
158dbda0 TH |
2443 | } else { |
2444 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2445 | break; | |
2446 | } | |
2447 | } | |
2448 | ||
2449 | return rq; | |
2450 | } | |
9934c8c0 | 2451 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2452 | |
9934c8c0 | 2453 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2454 | { |
9934c8c0 TH |
2455 | struct request_queue *q = rq->q; |
2456 | ||
158dbda0 TH |
2457 | BUG_ON(list_empty(&rq->queuelist)); |
2458 | BUG_ON(ELV_ON_HASH(rq)); | |
2459 | ||
2460 | list_del_init(&rq->queuelist); | |
2461 | ||
2462 | /* | |
2463 | * the time frame between a request being removed from the lists | |
2464 | * and to it is freed is accounted as io that is in progress at | |
2465 | * the driver side. | |
2466 | */ | |
9195291e | 2467 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2468 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2469 | set_io_start_time_ns(rq); |
2470 | } | |
158dbda0 TH |
2471 | } |
2472 | ||
9934c8c0 TH |
2473 | /** |
2474 | * blk_start_request - start request processing on the driver | |
2475 | * @req: request to dequeue | |
2476 | * | |
2477 | * Description: | |
2478 | * Dequeue @req and start timeout timer on it. This hands off the | |
2479 | * request to the driver. | |
2480 | * | |
2481 | * Block internal functions which don't want to start timer should | |
2482 | * call blk_dequeue_request(). | |
2483 | * | |
2484 | * Context: | |
2485 | * queue_lock must be held. | |
2486 | */ | |
2487 | void blk_start_request(struct request *req) | |
2488 | { | |
2489 | blk_dequeue_request(req); | |
2490 | ||
cf43e6be JA |
2491 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
2492 | blk_stat_set_issue_time(&req->issue_stat); | |
2493 | req->rq_flags |= RQF_STATS; | |
87760e5e | 2494 | wbt_issue(req->q->rq_wb, &req->issue_stat); |
cf43e6be JA |
2495 | } |
2496 | ||
9934c8c0 | 2497 | /* |
5f49f631 TH |
2498 | * We are now handing the request to the hardware, initialize |
2499 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2500 | */ |
5f49f631 | 2501 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2502 | if (unlikely(blk_bidi_rq(req))) |
2503 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2504 | ||
4912aa6c | 2505 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); |
9934c8c0 TH |
2506 | blk_add_timer(req); |
2507 | } | |
2508 | EXPORT_SYMBOL(blk_start_request); | |
2509 | ||
2510 | /** | |
2511 | * blk_fetch_request - fetch a request from a request queue | |
2512 | * @q: request queue to fetch a request from | |
2513 | * | |
2514 | * Description: | |
2515 | * Return the request at the top of @q. The request is started on | |
2516 | * return and LLD can start processing it immediately. | |
2517 | * | |
2518 | * Return: | |
2519 | * Pointer to the request at the top of @q if available. Null | |
2520 | * otherwise. | |
2521 | * | |
2522 | * Context: | |
2523 | * queue_lock must be held. | |
2524 | */ | |
2525 | struct request *blk_fetch_request(struct request_queue *q) | |
2526 | { | |
2527 | struct request *rq; | |
2528 | ||
2529 | rq = blk_peek_request(q); | |
2530 | if (rq) | |
2531 | blk_start_request(rq); | |
2532 | return rq; | |
2533 | } | |
2534 | EXPORT_SYMBOL(blk_fetch_request); | |
2535 | ||
3bcddeac | 2536 | /** |
2e60e022 | 2537 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2538 | * @req: the request being processed |
710027a4 | 2539 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2540 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2541 | * |
2542 | * Description: | |
8ebf9756 RD |
2543 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2544 | * the request structure even if @req doesn't have leftover. | |
2545 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2546 | * |
2547 | * This special helper function is only for request stacking drivers | |
2548 | * (e.g. request-based dm) so that they can handle partial completion. | |
2549 | * Actual device drivers should use blk_end_request instead. | |
2550 | * | |
2551 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2552 | * %false return from this function. | |
3bcddeac KU |
2553 | * |
2554 | * Return: | |
2e60e022 TH |
2555 | * %false - this request doesn't have any more data |
2556 | * %true - this request has more data | |
3bcddeac | 2557 | **/ |
2e60e022 | 2558 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2559 | { |
f79ea416 | 2560 | int total_bytes; |
1da177e4 | 2561 | |
4a0efdc9 HR |
2562 | trace_block_rq_complete(req->q, req, nr_bytes); |
2563 | ||
2e60e022 TH |
2564 | if (!req->bio) |
2565 | return false; | |
2566 | ||
1da177e4 | 2567 | /* |
6f41469c TH |
2568 | * For fs requests, rq is just carrier of independent bio's |
2569 | * and each partial completion should be handled separately. | |
2570 | * Reset per-request error on each partial completion. | |
2571 | * | |
2572 | * TODO: tj: This is too subtle. It would be better to let | |
2573 | * low level drivers do what they see fit. | |
1da177e4 | 2574 | */ |
33659ebb | 2575 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2576 | req->errors = 0; |
2577 | ||
33659ebb | 2578 | if (error && req->cmd_type == REQ_TYPE_FS && |
e8064021 | 2579 | !(req->rq_flags & RQF_QUIET)) { |
79775567 HR |
2580 | char *error_type; |
2581 | ||
2582 | switch (error) { | |
2583 | case -ENOLINK: | |
2584 | error_type = "recoverable transport"; | |
2585 | break; | |
2586 | case -EREMOTEIO: | |
2587 | error_type = "critical target"; | |
2588 | break; | |
2589 | case -EBADE: | |
2590 | error_type = "critical nexus"; | |
2591 | break; | |
d1ffc1f8 HR |
2592 | case -ETIMEDOUT: |
2593 | error_type = "timeout"; | |
2594 | break; | |
a9d6ceb8 HR |
2595 | case -ENOSPC: |
2596 | error_type = "critical space allocation"; | |
2597 | break; | |
7e782af5 HR |
2598 | case -ENODATA: |
2599 | error_type = "critical medium"; | |
2600 | break; | |
79775567 HR |
2601 | case -EIO: |
2602 | default: | |
2603 | error_type = "I/O"; | |
2604 | break; | |
2605 | } | |
ef3ecb66 RE |
2606 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", |
2607 | __func__, error_type, req->rq_disk ? | |
37d7b34f YZ |
2608 | req->rq_disk->disk_name : "?", |
2609 | (unsigned long long)blk_rq_pos(req)); | |
2610 | ||
1da177e4 LT |
2611 | } |
2612 | ||
bc58ba94 | 2613 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2614 | |
f79ea416 KO |
2615 | total_bytes = 0; |
2616 | while (req->bio) { | |
2617 | struct bio *bio = req->bio; | |
4f024f37 | 2618 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2619 | |
4f024f37 | 2620 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2621 | req->bio = bio->bi_next; |
1da177e4 | 2622 | |
f79ea416 | 2623 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2624 | |
f79ea416 KO |
2625 | total_bytes += bio_bytes; |
2626 | nr_bytes -= bio_bytes; | |
1da177e4 | 2627 | |
f79ea416 KO |
2628 | if (!nr_bytes) |
2629 | break; | |
1da177e4 LT |
2630 | } |
2631 | ||
2632 | /* | |
2633 | * completely done | |
2634 | */ | |
2e60e022 TH |
2635 | if (!req->bio) { |
2636 | /* | |
2637 | * Reset counters so that the request stacking driver | |
2638 | * can find how many bytes remain in the request | |
2639 | * later. | |
2640 | */ | |
a2dec7b3 | 2641 | req->__data_len = 0; |
2e60e022 TH |
2642 | return false; |
2643 | } | |
1da177e4 | 2644 | |
a2dec7b3 | 2645 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2646 | |
2647 | /* update sector only for requests with clear definition of sector */ | |
e2a60da7 | 2648 | if (req->cmd_type == REQ_TYPE_FS) |
a2dec7b3 | 2649 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2650 | |
80a761fd | 2651 | /* mixed attributes always follow the first bio */ |
e8064021 | 2652 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 2653 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 2654 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
2655 | } |
2656 | ||
2e46e8b2 TH |
2657 | /* |
2658 | * If total number of sectors is less than the first segment | |
2659 | * size, something has gone terribly wrong. | |
2660 | */ | |
2661 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2662 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2663 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2664 | } |
2665 | ||
2666 | /* recalculate the number of segments */ | |
1da177e4 | 2667 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2668 | |
2e60e022 | 2669 | return true; |
1da177e4 | 2670 | } |
2e60e022 | 2671 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2672 | |
2e60e022 TH |
2673 | static bool blk_update_bidi_request(struct request *rq, int error, |
2674 | unsigned int nr_bytes, | |
2675 | unsigned int bidi_bytes) | |
5efccd17 | 2676 | { |
2e60e022 TH |
2677 | if (blk_update_request(rq, error, nr_bytes)) |
2678 | return true; | |
5efccd17 | 2679 | |
2e60e022 TH |
2680 | /* Bidi request must be completed as a whole */ |
2681 | if (unlikely(blk_bidi_rq(rq)) && | |
2682 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2683 | return true; | |
5efccd17 | 2684 | |
e2e1a148 JA |
2685 | if (blk_queue_add_random(rq->q)) |
2686 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2687 | |
2688 | return false; | |
1da177e4 LT |
2689 | } |
2690 | ||
28018c24 JB |
2691 | /** |
2692 | * blk_unprep_request - unprepare a request | |
2693 | * @req: the request | |
2694 | * | |
2695 | * This function makes a request ready for complete resubmission (or | |
2696 | * completion). It happens only after all error handling is complete, | |
2697 | * so represents the appropriate moment to deallocate any resources | |
2698 | * that were allocated to the request in the prep_rq_fn. The queue | |
2699 | * lock is held when calling this. | |
2700 | */ | |
2701 | void blk_unprep_request(struct request *req) | |
2702 | { | |
2703 | struct request_queue *q = req->q; | |
2704 | ||
e8064021 | 2705 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
2706 | if (q->unprep_rq_fn) |
2707 | q->unprep_rq_fn(q, req); | |
2708 | } | |
2709 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2710 | ||
1da177e4 LT |
2711 | /* |
2712 | * queue lock must be held | |
2713 | */ | |
12120077 | 2714 | void blk_finish_request(struct request *req, int error) |
1da177e4 | 2715 | { |
cf43e6be JA |
2716 | struct request_queue *q = req->q; |
2717 | ||
2718 | if (req->rq_flags & RQF_STATS) | |
2719 | blk_stat_add(&q->rq_stats[rq_data_dir(req)], req); | |
2720 | ||
e8064021 | 2721 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 2722 | blk_queue_end_tag(q, req); |
b8286239 | 2723 | |
ba396a6c | 2724 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2725 | |
33659ebb | 2726 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2727 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2728 | |
e78042e5 MA |
2729 | blk_delete_timer(req); |
2730 | ||
e8064021 | 2731 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
2732 | blk_unprep_request(req); |
2733 | ||
bc58ba94 | 2734 | blk_account_io_done(req); |
b8286239 | 2735 | |
87760e5e JA |
2736 | if (req->end_io) { |
2737 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
8ffdc655 | 2738 | req->end_io(req, error); |
87760e5e | 2739 | } else { |
b8286239 KU |
2740 | if (blk_bidi_rq(req)) |
2741 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2742 | ||
cf43e6be | 2743 | __blk_put_request(q, req); |
b8286239 | 2744 | } |
1da177e4 | 2745 | } |
12120077 | 2746 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 2747 | |
3b11313a | 2748 | /** |
2e60e022 TH |
2749 | * blk_end_bidi_request - Complete a bidi request |
2750 | * @rq: the request to complete | |
2751 | * @error: %0 for success, < %0 for error | |
2752 | * @nr_bytes: number of bytes to complete @rq | |
2753 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2754 | * |
2755 | * Description: | |
e3a04fe3 | 2756 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2757 | * Drivers that supports bidi can safely call this member for any |
2758 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2759 | * just ignored. | |
336cdb40 KU |
2760 | * |
2761 | * Return: | |
2e60e022 TH |
2762 | * %false - we are done with this request |
2763 | * %true - still buffers pending for this request | |
a0cd1285 | 2764 | **/ |
b1f74493 | 2765 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2766 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2767 | { | |
336cdb40 | 2768 | struct request_queue *q = rq->q; |
2e60e022 | 2769 | unsigned long flags; |
32fab448 | 2770 | |
2e60e022 TH |
2771 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2772 | return true; | |
32fab448 | 2773 | |
336cdb40 | 2774 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2775 | blk_finish_request(rq, error); |
336cdb40 KU |
2776 | spin_unlock_irqrestore(q->queue_lock, flags); |
2777 | ||
2e60e022 | 2778 | return false; |
32fab448 KU |
2779 | } |
2780 | ||
336cdb40 | 2781 | /** |
2e60e022 TH |
2782 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2783 | * @rq: the request to complete | |
710027a4 | 2784 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2785 | * @nr_bytes: number of bytes to complete @rq |
2786 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2787 | * |
2788 | * Description: | |
2e60e022 TH |
2789 | * Identical to blk_end_bidi_request() except that queue lock is |
2790 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2791 | * |
2792 | * Return: | |
2e60e022 TH |
2793 | * %false - we are done with this request |
2794 | * %true - still buffers pending for this request | |
336cdb40 | 2795 | **/ |
4853abaa | 2796 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2797 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2798 | { |
2e60e022 TH |
2799 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2800 | return true; | |
336cdb40 | 2801 | |
2e60e022 | 2802 | blk_finish_request(rq, error); |
336cdb40 | 2803 | |
2e60e022 | 2804 | return false; |
336cdb40 | 2805 | } |
e19a3ab0 KU |
2806 | |
2807 | /** | |
2808 | * blk_end_request - Helper function for drivers to complete the request. | |
2809 | * @rq: the request being processed | |
710027a4 | 2810 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2811 | * @nr_bytes: number of bytes to complete |
2812 | * | |
2813 | * Description: | |
2814 | * Ends I/O on a number of bytes attached to @rq. | |
2815 | * If @rq has leftover, sets it up for the next range of segments. | |
2816 | * | |
2817 | * Return: | |
b1f74493 FT |
2818 | * %false - we are done with this request |
2819 | * %true - still buffers pending for this request | |
e19a3ab0 | 2820 | **/ |
b1f74493 | 2821 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2822 | { |
b1f74493 | 2823 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2824 | } |
56ad1740 | 2825 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2826 | |
2827 | /** | |
b1f74493 FT |
2828 | * blk_end_request_all - Helper function for drives to finish the request. |
2829 | * @rq: the request to finish | |
8ebf9756 | 2830 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2831 | * |
2832 | * Description: | |
b1f74493 FT |
2833 | * Completely finish @rq. |
2834 | */ | |
2835 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2836 | { |
b1f74493 FT |
2837 | bool pending; |
2838 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2839 | |
b1f74493 FT |
2840 | if (unlikely(blk_bidi_rq(rq))) |
2841 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2842 | |
b1f74493 FT |
2843 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2844 | BUG_ON(pending); | |
2845 | } | |
56ad1740 | 2846 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2847 | |
b1f74493 FT |
2848 | /** |
2849 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2850 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2851 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2852 | * |
2853 | * Description: | |
2854 | * Complete the current consecutively mapped chunk from @rq. | |
2855 | * | |
2856 | * Return: | |
2857 | * %false - we are done with this request | |
2858 | * %true - still buffers pending for this request | |
2859 | */ | |
2860 | bool blk_end_request_cur(struct request *rq, int error) | |
2861 | { | |
2862 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2863 | } |
56ad1740 | 2864 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2865 | |
80a761fd TH |
2866 | /** |
2867 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2868 | * @rq: the request to finish till the next failure boundary for | |
2869 | * @error: must be negative errno | |
2870 | * | |
2871 | * Description: | |
2872 | * Complete @rq till the next failure boundary. | |
2873 | * | |
2874 | * Return: | |
2875 | * %false - we are done with this request | |
2876 | * %true - still buffers pending for this request | |
2877 | */ | |
2878 | bool blk_end_request_err(struct request *rq, int error) | |
2879 | { | |
2880 | WARN_ON(error >= 0); | |
2881 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2882 | } | |
2883 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2884 | ||
e3a04fe3 | 2885 | /** |
b1f74493 FT |
2886 | * __blk_end_request - Helper function for drivers to complete the request. |
2887 | * @rq: the request being processed | |
2888 | * @error: %0 for success, < %0 for error | |
2889 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2890 | * |
2891 | * Description: | |
b1f74493 | 2892 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2893 | * |
2894 | * Return: | |
b1f74493 FT |
2895 | * %false - we are done with this request |
2896 | * %true - still buffers pending for this request | |
e3a04fe3 | 2897 | **/ |
b1f74493 | 2898 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2899 | { |
b1f74493 | 2900 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2901 | } |
56ad1740 | 2902 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2903 | |
32fab448 | 2904 | /** |
b1f74493 FT |
2905 | * __blk_end_request_all - Helper function for drives to finish the request. |
2906 | * @rq: the request to finish | |
8ebf9756 | 2907 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2908 | * |
2909 | * Description: | |
b1f74493 | 2910 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2911 | */ |
b1f74493 | 2912 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2913 | { |
b1f74493 FT |
2914 | bool pending; |
2915 | unsigned int bidi_bytes = 0; | |
2916 | ||
2917 | if (unlikely(blk_bidi_rq(rq))) | |
2918 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2919 | ||
2920 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2921 | BUG_ON(pending); | |
32fab448 | 2922 | } |
56ad1740 | 2923 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2924 | |
e19a3ab0 | 2925 | /** |
b1f74493 FT |
2926 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2927 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2928 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2929 | * |
2930 | * Description: | |
b1f74493 FT |
2931 | * Complete the current consecutively mapped chunk from @rq. Must |
2932 | * be called with queue lock held. | |
e19a3ab0 KU |
2933 | * |
2934 | * Return: | |
b1f74493 FT |
2935 | * %false - we are done with this request |
2936 | * %true - still buffers pending for this request | |
2937 | */ | |
2938 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2939 | { |
b1f74493 | 2940 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2941 | } |
56ad1740 | 2942 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2943 | |
80a761fd TH |
2944 | /** |
2945 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2946 | * @rq: the request to finish till the next failure boundary for | |
2947 | * @error: must be negative errno | |
2948 | * | |
2949 | * Description: | |
2950 | * Complete @rq till the next failure boundary. Must be called | |
2951 | * with queue lock held. | |
2952 | * | |
2953 | * Return: | |
2954 | * %false - we are done with this request | |
2955 | * %true - still buffers pending for this request | |
2956 | */ | |
2957 | bool __blk_end_request_err(struct request *rq, int error) | |
2958 | { | |
2959 | WARN_ON(error >= 0); | |
2960 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2961 | } | |
2962 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2963 | ||
86db1e29 JA |
2964 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2965 | struct bio *bio) | |
1da177e4 | 2966 | { |
b4f42e28 | 2967 | if (bio_has_data(bio)) |
fb2dce86 | 2968 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
b4f42e28 | 2969 | |
4f024f37 | 2970 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 2971 | rq->bio = rq->biotail = bio; |
1da177e4 | 2972 | |
66846572 N |
2973 | if (bio->bi_bdev) |
2974 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2975 | } | |
1da177e4 | 2976 | |
2d4dc890 IL |
2977 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2978 | /** | |
2979 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2980 | * @rq: the request to be flushed | |
2981 | * | |
2982 | * Description: | |
2983 | * Flush all pages in @rq. | |
2984 | */ | |
2985 | void rq_flush_dcache_pages(struct request *rq) | |
2986 | { | |
2987 | struct req_iterator iter; | |
7988613b | 2988 | struct bio_vec bvec; |
2d4dc890 IL |
2989 | |
2990 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 2991 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
2992 | } |
2993 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2994 | #endif | |
2995 | ||
ef9e3fac KU |
2996 | /** |
2997 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2998 | * @q : the queue of the device being checked | |
2999 | * | |
3000 | * Description: | |
3001 | * Check if underlying low-level drivers of a device are busy. | |
3002 | * If the drivers want to export their busy state, they must set own | |
3003 | * exporting function using blk_queue_lld_busy() first. | |
3004 | * | |
3005 | * Basically, this function is used only by request stacking drivers | |
3006 | * to stop dispatching requests to underlying devices when underlying | |
3007 | * devices are busy. This behavior helps more I/O merging on the queue | |
3008 | * of the request stacking driver and prevents I/O throughput regression | |
3009 | * on burst I/O load. | |
3010 | * | |
3011 | * Return: | |
3012 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3013 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3014 | */ | |
3015 | int blk_lld_busy(struct request_queue *q) | |
3016 | { | |
3017 | if (q->lld_busy_fn) | |
3018 | return q->lld_busy_fn(q); | |
3019 | ||
3020 | return 0; | |
3021 | } | |
3022 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3023 | ||
78d8e58a MS |
3024 | /** |
3025 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3026 | * @rq: the clone request to be cleaned up | |
3027 | * | |
3028 | * Description: | |
3029 | * Free all bios in @rq for a cloned request. | |
3030 | */ | |
3031 | void blk_rq_unprep_clone(struct request *rq) | |
3032 | { | |
3033 | struct bio *bio; | |
3034 | ||
3035 | while ((bio = rq->bio) != NULL) { | |
3036 | rq->bio = bio->bi_next; | |
3037 | ||
3038 | bio_put(bio); | |
3039 | } | |
3040 | } | |
3041 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3042 | ||
3043 | /* | |
3044 | * Copy attributes of the original request to the clone request. | |
3045 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3046 | */ | |
3047 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3048 | { |
3049 | dst->cpu = src->cpu; | |
ef295ecf | 3050 | dst->cmd_flags = src->cmd_flags | REQ_NOMERGE; |
b0fd271d KU |
3051 | dst->cmd_type = src->cmd_type; |
3052 | dst->__sector = blk_rq_pos(src); | |
3053 | dst->__data_len = blk_rq_bytes(src); | |
3054 | dst->nr_phys_segments = src->nr_phys_segments; | |
3055 | dst->ioprio = src->ioprio; | |
3056 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3057 | } |
3058 | ||
3059 | /** | |
3060 | * blk_rq_prep_clone - Helper function to setup clone request | |
3061 | * @rq: the request to be setup | |
3062 | * @rq_src: original request to be cloned | |
3063 | * @bs: bio_set that bios for clone are allocated from | |
3064 | * @gfp_mask: memory allocation mask for bio | |
3065 | * @bio_ctr: setup function to be called for each clone bio. | |
3066 | * Returns %0 for success, non %0 for failure. | |
3067 | * @data: private data to be passed to @bio_ctr | |
3068 | * | |
3069 | * Description: | |
3070 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3071 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3072 | * are not copied, and copying such parts is the caller's responsibility. | |
3073 | * Also, pages which the original bios are pointing to are not copied | |
3074 | * and the cloned bios just point same pages. | |
3075 | * So cloned bios must be completed before original bios, which means | |
3076 | * the caller must complete @rq before @rq_src. | |
3077 | */ | |
3078 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3079 | struct bio_set *bs, gfp_t gfp_mask, | |
3080 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3081 | void *data) | |
3082 | { | |
3083 | struct bio *bio, *bio_src; | |
3084 | ||
3085 | if (!bs) | |
3086 | bs = fs_bio_set; | |
3087 | ||
3088 | __rq_for_each_bio(bio_src, rq_src) { | |
3089 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3090 | if (!bio) | |
3091 | goto free_and_out; | |
3092 | ||
3093 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3094 | goto free_and_out; | |
3095 | ||
3096 | if (rq->bio) { | |
3097 | rq->biotail->bi_next = bio; | |
3098 | rq->biotail = bio; | |
3099 | } else | |
3100 | rq->bio = rq->biotail = bio; | |
3101 | } | |
3102 | ||
3103 | __blk_rq_prep_clone(rq, rq_src); | |
3104 | ||
3105 | return 0; | |
3106 | ||
3107 | free_and_out: | |
3108 | if (bio) | |
3109 | bio_put(bio); | |
3110 | blk_rq_unprep_clone(rq); | |
3111 | ||
3112 | return -ENOMEM; | |
b0fd271d KU |
3113 | } |
3114 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3115 | ||
59c3d45e | 3116 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3117 | { |
3118 | return queue_work(kblockd_workqueue, work); | |
3119 | } | |
1da177e4 LT |
3120 | EXPORT_SYMBOL(kblockd_schedule_work); |
3121 | ||
ee63cfa7 JA |
3122 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3123 | { | |
3124 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3125 | } | |
3126 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3127 | ||
59c3d45e JA |
3128 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, |
3129 | unsigned long delay) | |
e43473b7 VG |
3130 | { |
3131 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3132 | } | |
3133 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3134 | ||
8ab14595 JA |
3135 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3136 | unsigned long delay) | |
3137 | { | |
3138 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3139 | } | |
3140 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3141 | ||
75df7136 SJ |
3142 | /** |
3143 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3144 | * @plug: The &struct blk_plug that needs to be initialized | |
3145 | * | |
3146 | * Description: | |
3147 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3148 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3149 | * blk_finish_plug(). This is important from a performance perspective, but | |
3150 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3151 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3152 | * page belonging to that request that is currently residing in our private | |
3153 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3154 | * this kind of deadlock. | |
3155 | */ | |
73c10101 JA |
3156 | void blk_start_plug(struct blk_plug *plug) |
3157 | { | |
3158 | struct task_struct *tsk = current; | |
3159 | ||
dd6cf3e1 SL |
3160 | /* |
3161 | * If this is a nested plug, don't actually assign it. | |
3162 | */ | |
3163 | if (tsk->plug) | |
3164 | return; | |
3165 | ||
73c10101 | 3166 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3167 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3168 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3169 | /* |
dd6cf3e1 SL |
3170 | * Store ordering should not be needed here, since a potential |
3171 | * preempt will imply a full memory barrier | |
73c10101 | 3172 | */ |
dd6cf3e1 | 3173 | tsk->plug = plug; |
73c10101 JA |
3174 | } |
3175 | EXPORT_SYMBOL(blk_start_plug); | |
3176 | ||
3177 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3178 | { | |
3179 | struct request *rqa = container_of(a, struct request, queuelist); | |
3180 | struct request *rqb = container_of(b, struct request, queuelist); | |
3181 | ||
975927b9 JM |
3182 | return !(rqa->q < rqb->q || |
3183 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3184 | } |
3185 | ||
49cac01e JA |
3186 | /* |
3187 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3188 | * until a safe kblockd context. We due this to avoid accidental big | |
3189 | * additional stack usage in driver dispatch, in places where the originally | |
3190 | * plugger did not intend it. | |
3191 | */ | |
f6603783 | 3192 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3193 | bool from_schedule) |
99e22598 | 3194 | __releases(q->queue_lock) |
94b5eb28 | 3195 | { |
49cac01e | 3196 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3197 | |
70460571 | 3198 | if (from_schedule) |
24ecfbe2 | 3199 | blk_run_queue_async(q); |
70460571 | 3200 | else |
24ecfbe2 | 3201 | __blk_run_queue(q); |
70460571 | 3202 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3203 | } |
3204 | ||
74018dc3 | 3205 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3206 | { |
3207 | LIST_HEAD(callbacks); | |
3208 | ||
2a7d5559 SL |
3209 | while (!list_empty(&plug->cb_list)) { |
3210 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3211 | |
2a7d5559 SL |
3212 | while (!list_empty(&callbacks)) { |
3213 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3214 | struct blk_plug_cb, |
3215 | list); | |
2a7d5559 | 3216 | list_del(&cb->list); |
74018dc3 | 3217 | cb->callback(cb, from_schedule); |
2a7d5559 | 3218 | } |
048c9374 N |
3219 | } |
3220 | } | |
3221 | ||
9cbb1750 N |
3222 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3223 | int size) | |
3224 | { | |
3225 | struct blk_plug *plug = current->plug; | |
3226 | struct blk_plug_cb *cb; | |
3227 | ||
3228 | if (!plug) | |
3229 | return NULL; | |
3230 | ||
3231 | list_for_each_entry(cb, &plug->cb_list, list) | |
3232 | if (cb->callback == unplug && cb->data == data) | |
3233 | return cb; | |
3234 | ||
3235 | /* Not currently on the callback list */ | |
3236 | BUG_ON(size < sizeof(*cb)); | |
3237 | cb = kzalloc(size, GFP_ATOMIC); | |
3238 | if (cb) { | |
3239 | cb->data = data; | |
3240 | cb->callback = unplug; | |
3241 | list_add(&cb->list, &plug->cb_list); | |
3242 | } | |
3243 | return cb; | |
3244 | } | |
3245 | EXPORT_SYMBOL(blk_check_plugged); | |
3246 | ||
49cac01e | 3247 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3248 | { |
3249 | struct request_queue *q; | |
3250 | unsigned long flags; | |
3251 | struct request *rq; | |
109b8129 | 3252 | LIST_HEAD(list); |
94b5eb28 | 3253 | unsigned int depth; |
73c10101 | 3254 | |
74018dc3 | 3255 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3256 | |
3257 | if (!list_empty(&plug->mq_list)) | |
3258 | blk_mq_flush_plug_list(plug, from_schedule); | |
3259 | ||
73c10101 JA |
3260 | if (list_empty(&plug->list)) |
3261 | return; | |
3262 | ||
109b8129 N |
3263 | list_splice_init(&plug->list, &list); |
3264 | ||
422765c2 | 3265 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3266 | |
3267 | q = NULL; | |
94b5eb28 | 3268 | depth = 0; |
18811272 JA |
3269 | |
3270 | /* | |
3271 | * Save and disable interrupts here, to avoid doing it for every | |
3272 | * queue lock we have to take. | |
3273 | */ | |
73c10101 | 3274 | local_irq_save(flags); |
109b8129 N |
3275 | while (!list_empty(&list)) { |
3276 | rq = list_entry_rq(list.next); | |
73c10101 | 3277 | list_del_init(&rq->queuelist); |
73c10101 JA |
3278 | BUG_ON(!rq->q); |
3279 | if (rq->q != q) { | |
99e22598 JA |
3280 | /* |
3281 | * This drops the queue lock | |
3282 | */ | |
3283 | if (q) | |
49cac01e | 3284 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3285 | q = rq->q; |
94b5eb28 | 3286 | depth = 0; |
73c10101 JA |
3287 | spin_lock(q->queue_lock); |
3288 | } | |
8ba61435 TH |
3289 | |
3290 | /* | |
3291 | * Short-circuit if @q is dead | |
3292 | */ | |
3f3299d5 | 3293 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 TH |
3294 | __blk_end_request_all(rq, -ENODEV); |
3295 | continue; | |
3296 | } | |
3297 | ||
73c10101 JA |
3298 | /* |
3299 | * rq is already accounted, so use raw insert | |
3300 | */ | |
28a8f0d3 | 3301 | if (rq->cmd_flags & (REQ_PREFLUSH | REQ_FUA)) |
401a18e9 JA |
3302 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3303 | else | |
3304 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3305 | |
3306 | depth++; | |
73c10101 JA |
3307 | } |
3308 | ||
99e22598 JA |
3309 | /* |
3310 | * This drops the queue lock | |
3311 | */ | |
3312 | if (q) | |
49cac01e | 3313 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3314 | |
73c10101 JA |
3315 | local_irq_restore(flags); |
3316 | } | |
73c10101 JA |
3317 | |
3318 | void blk_finish_plug(struct blk_plug *plug) | |
3319 | { | |
dd6cf3e1 SL |
3320 | if (plug != current->plug) |
3321 | return; | |
f6603783 | 3322 | blk_flush_plug_list(plug, false); |
73c10101 | 3323 | |
dd6cf3e1 | 3324 | current->plug = NULL; |
73c10101 | 3325 | } |
88b996cd | 3326 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3327 | |
47fafbc7 | 3328 | #ifdef CONFIG_PM |
6c954667 LM |
3329 | /** |
3330 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3331 | * @q: the queue of the device | |
3332 | * @dev: the device the queue belongs to | |
3333 | * | |
3334 | * Description: | |
3335 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3336 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3337 | * should call this function after @dev has been initialized, and its | |
3338 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3339 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3340 | * cases, driver should call this function before any I/O has taken place. | |
3341 | * | |
3342 | * This function takes care of setting up using auto suspend for the device, | |
3343 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3344 | * until an updated value is either set by user or by driver. Drivers do | |
3345 | * not need to touch other autosuspend settings. | |
3346 | * | |
3347 | * The block layer runtime PM is request based, so only works for drivers | |
3348 | * that use request as their IO unit instead of those directly use bio's. | |
3349 | */ | |
3350 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3351 | { | |
3352 | q->dev = dev; | |
3353 | q->rpm_status = RPM_ACTIVE; | |
3354 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3355 | pm_runtime_use_autosuspend(q->dev); | |
3356 | } | |
3357 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3358 | ||
3359 | /** | |
3360 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3361 | * @q: the queue of the device | |
3362 | * | |
3363 | * Description: | |
3364 | * This function will check if runtime suspend is allowed for the device | |
3365 | * by examining if there are any requests pending in the queue. If there | |
3366 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3367 | * the queue's status will be updated to SUSPENDING and the driver can | |
3368 | * proceed to suspend the device. | |
3369 | * | |
3370 | * For the not allowed case, we mark last busy for the device so that | |
3371 | * runtime PM core will try to autosuspend it some time later. | |
3372 | * | |
3373 | * This function should be called near the start of the device's | |
3374 | * runtime_suspend callback. | |
3375 | * | |
3376 | * Return: | |
3377 | * 0 - OK to runtime suspend the device | |
3378 | * -EBUSY - Device should not be runtime suspended | |
3379 | */ | |
3380 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3381 | { | |
3382 | int ret = 0; | |
3383 | ||
4fd41a85 KX |
3384 | if (!q->dev) |
3385 | return ret; | |
3386 | ||
6c954667 LM |
3387 | spin_lock_irq(q->queue_lock); |
3388 | if (q->nr_pending) { | |
3389 | ret = -EBUSY; | |
3390 | pm_runtime_mark_last_busy(q->dev); | |
3391 | } else { | |
3392 | q->rpm_status = RPM_SUSPENDING; | |
3393 | } | |
3394 | spin_unlock_irq(q->queue_lock); | |
3395 | return ret; | |
3396 | } | |
3397 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3398 | ||
3399 | /** | |
3400 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3401 | * @q: the queue of the device | |
3402 | * @err: return value of the device's runtime_suspend function | |
3403 | * | |
3404 | * Description: | |
3405 | * Update the queue's runtime status according to the return value of the | |
3406 | * device's runtime suspend function and mark last busy for the device so | |
3407 | * that PM core will try to auto suspend the device at a later time. | |
3408 | * | |
3409 | * This function should be called near the end of the device's | |
3410 | * runtime_suspend callback. | |
3411 | */ | |
3412 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3413 | { | |
4fd41a85 KX |
3414 | if (!q->dev) |
3415 | return; | |
3416 | ||
6c954667 LM |
3417 | spin_lock_irq(q->queue_lock); |
3418 | if (!err) { | |
3419 | q->rpm_status = RPM_SUSPENDED; | |
3420 | } else { | |
3421 | q->rpm_status = RPM_ACTIVE; | |
3422 | pm_runtime_mark_last_busy(q->dev); | |
3423 | } | |
3424 | spin_unlock_irq(q->queue_lock); | |
3425 | } | |
3426 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3427 | ||
3428 | /** | |
3429 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3430 | * @q: the queue of the device | |
3431 | * | |
3432 | * Description: | |
3433 | * Update the queue's runtime status to RESUMING in preparation for the | |
3434 | * runtime resume of the device. | |
3435 | * | |
3436 | * This function should be called near the start of the device's | |
3437 | * runtime_resume callback. | |
3438 | */ | |
3439 | void blk_pre_runtime_resume(struct request_queue *q) | |
3440 | { | |
4fd41a85 KX |
3441 | if (!q->dev) |
3442 | return; | |
3443 | ||
6c954667 LM |
3444 | spin_lock_irq(q->queue_lock); |
3445 | q->rpm_status = RPM_RESUMING; | |
3446 | spin_unlock_irq(q->queue_lock); | |
3447 | } | |
3448 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3449 | ||
3450 | /** | |
3451 | * blk_post_runtime_resume - Post runtime resume processing | |
3452 | * @q: the queue of the device | |
3453 | * @err: return value of the device's runtime_resume function | |
3454 | * | |
3455 | * Description: | |
3456 | * Update the queue's runtime status according to the return value of the | |
3457 | * device's runtime_resume function. If it is successfully resumed, process | |
3458 | * the requests that are queued into the device's queue when it is resuming | |
3459 | * and then mark last busy and initiate autosuspend for it. | |
3460 | * | |
3461 | * This function should be called near the end of the device's | |
3462 | * runtime_resume callback. | |
3463 | */ | |
3464 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3465 | { | |
4fd41a85 KX |
3466 | if (!q->dev) |
3467 | return; | |
3468 | ||
6c954667 LM |
3469 | spin_lock_irq(q->queue_lock); |
3470 | if (!err) { | |
3471 | q->rpm_status = RPM_ACTIVE; | |
3472 | __blk_run_queue(q); | |
3473 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3474 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3475 | } else { |
3476 | q->rpm_status = RPM_SUSPENDED; | |
3477 | } | |
3478 | spin_unlock_irq(q->queue_lock); | |
3479 | } | |
3480 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3481 | |
3482 | /** | |
3483 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3484 | * @q: the queue of the device | |
3485 | * | |
3486 | * If the device is left runtime suspended during system suspend the resume | |
3487 | * hook typically resumes the device and corrects runtime status | |
3488 | * accordingly. However, that does not affect the queue runtime PM status | |
3489 | * which is still "suspended". This prevents processing requests from the | |
3490 | * queue. | |
3491 | * | |
3492 | * This function can be used in driver's resume hook to correct queue | |
3493 | * runtime PM status and re-enable peeking requests from the queue. It | |
3494 | * should be called before first request is added to the queue. | |
3495 | */ | |
3496 | void blk_set_runtime_active(struct request_queue *q) | |
3497 | { | |
3498 | spin_lock_irq(q->queue_lock); | |
3499 | q->rpm_status = RPM_ACTIVE; | |
3500 | pm_runtime_mark_last_busy(q->dev); | |
3501 | pm_request_autosuspend(q->dev); | |
3502 | spin_unlock_irq(q->queue_lock); | |
3503 | } | |
3504 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3505 | #endif |
3506 | ||
1da177e4 LT |
3507 | int __init blk_dev_init(void) |
3508 | { | |
ef295ecf CH |
3509 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3510 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3511 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3512 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3513 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3514 | |
89b90be2 TH |
3515 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3516 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3517 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3518 | if (!kblockd_workqueue) |
3519 | panic("Failed to create kblockd\n"); | |
3520 | ||
3521 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3522 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3523 | |
c2789bd4 | 3524 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3525 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3526 | |
d38ecf93 | 3527 | return 0; |
1da177e4 | 3528 | } |