]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
320ae51f | 19 | #include <linux/blk-mq.h> |
1da177e4 LT |
20 | #include <linux/highmem.h> |
21 | #include <linux/mm.h> | |
22 | #include <linux/kernel_stat.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/init.h> | |
1da177e4 LT |
25 | #include <linux/completion.h> |
26 | #include <linux/slab.h> | |
27 | #include <linux/swap.h> | |
28 | #include <linux/writeback.h> | |
faccbd4b | 29 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
73c10101 | 31 | #include <linux/list_sort.h> |
e3c78ca5 | 32 | #include <linux/delay.h> |
aaf7c680 | 33 | #include <linux/ratelimit.h> |
6c954667 | 34 | #include <linux/pm_runtime.h> |
eea8f41c | 35 | #include <linux/blk-cgroup.h> |
18fbda91 | 36 | #include <linux/debugfs.h> |
55782138 LZ |
37 | |
38 | #define CREATE_TRACE_POINTS | |
39 | #include <trace/events/block.h> | |
1da177e4 | 40 | |
8324aa91 | 41 | #include "blk.h" |
43a5e4e2 | 42 | #include "blk-mq.h" |
bd166ef1 | 43 | #include "blk-mq-sched.h" |
87760e5e | 44 | #include "blk-wbt.h" |
8324aa91 | 45 | |
18fbda91 OS |
46 | #ifdef CONFIG_DEBUG_FS |
47 | struct dentry *blk_debugfs_root; | |
48 | #endif | |
49 | ||
d07335e5 | 50 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 51 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 52 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 55 | |
a73f730d TH |
56 | DEFINE_IDA(blk_queue_ida); |
57 | ||
1da177e4 LT |
58 | /* |
59 | * For the allocated request tables | |
60 | */ | |
d674d414 | 61 | struct kmem_cache *request_cachep; |
1da177e4 LT |
62 | |
63 | /* | |
64 | * For queue allocation | |
65 | */ | |
6728cb0e | 66 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 67 | |
1da177e4 LT |
68 | /* |
69 | * Controlling structure to kblockd | |
70 | */ | |
ff856bad | 71 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 72 | |
d40f75a0 TH |
73 | static void blk_clear_congested(struct request_list *rl, int sync) |
74 | { | |
d40f75a0 TH |
75 | #ifdef CONFIG_CGROUP_WRITEBACK |
76 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
77 | #else | |
482cf79c TH |
78 | /* |
79 | * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't | |
80 | * flip its congestion state for events on other blkcgs. | |
81 | */ | |
82 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 83 | clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
84 | #endif |
85 | } | |
86 | ||
87 | static void blk_set_congested(struct request_list *rl, int sync) | |
88 | { | |
d40f75a0 TH |
89 | #ifdef CONFIG_CGROUP_WRITEBACK |
90 | set_wb_congested(rl->blkg->wb_congested, sync); | |
91 | #else | |
482cf79c TH |
92 | /* see blk_clear_congested() */ |
93 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 94 | set_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
95 | #endif |
96 | } | |
97 | ||
8324aa91 | 98 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
99 | { |
100 | int nr; | |
101 | ||
102 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
103 | if (nr > q->nr_requests) | |
104 | nr = q->nr_requests; | |
105 | q->nr_congestion_on = nr; | |
106 | ||
107 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
108 | if (nr < 1) | |
109 | nr = 1; | |
110 | q->nr_congestion_off = nr; | |
111 | } | |
112 | ||
2a4aa30c | 113 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 114 | { |
1afb20f3 FT |
115 | memset(rq, 0, sizeof(*rq)); |
116 | ||
1da177e4 | 117 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 118 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 119 | rq->cpu = -1; |
63a71386 | 120 | rq->q = q; |
a2dec7b3 | 121 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
122 | INIT_HLIST_NODE(&rq->hash); |
123 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 124 | rq->tag = -1; |
bd166ef1 | 125 | rq->internal_tag = -1; |
b243ddcb | 126 | rq->start_time = jiffies; |
9195291e | 127 | set_start_time_ns(rq); |
09e099d4 | 128 | rq->part = NULL; |
1da177e4 | 129 | } |
2a4aa30c | 130 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 131 | |
2a842aca CH |
132 | static const struct { |
133 | int errno; | |
134 | const char *name; | |
135 | } blk_errors[] = { | |
136 | [BLK_STS_OK] = { 0, "" }, | |
137 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
138 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
139 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
140 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
141 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
142 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
143 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
144 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
145 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
03a07c92 | 146 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 147 | |
4e4cbee9 CH |
148 | /* device mapper special case, should not leak out: */ |
149 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
150 | ||
2a842aca CH |
151 | /* everything else not covered above: */ |
152 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
153 | }; | |
154 | ||
155 | blk_status_t errno_to_blk_status(int errno) | |
156 | { | |
157 | int i; | |
158 | ||
159 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
160 | if (blk_errors[i].errno == errno) | |
161 | return (__force blk_status_t)i; | |
162 | } | |
163 | ||
164 | return BLK_STS_IOERR; | |
165 | } | |
166 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
167 | ||
168 | int blk_status_to_errno(blk_status_t status) | |
169 | { | |
170 | int idx = (__force int)status; | |
171 | ||
34bd9c1c | 172 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
173 | return -EIO; |
174 | return blk_errors[idx].errno; | |
175 | } | |
176 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
177 | ||
178 | static void print_req_error(struct request *req, blk_status_t status) | |
179 | { | |
180 | int idx = (__force int)status; | |
181 | ||
34bd9c1c | 182 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
183 | return; |
184 | ||
185 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", | |
186 | __func__, blk_errors[idx].name, req->rq_disk ? | |
187 | req->rq_disk->disk_name : "?", | |
188 | (unsigned long long)blk_rq_pos(req)); | |
189 | } | |
190 | ||
5bb23a68 | 191 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 192 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 193 | { |
78d8e58a | 194 | if (error) |
4e4cbee9 | 195 | bio->bi_status = error; |
797e7dbb | 196 | |
e8064021 | 197 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 198 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 199 | |
f79ea416 | 200 | bio_advance(bio, nbytes); |
7ba1ba12 | 201 | |
143a87f4 | 202 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 203 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 204 | bio_endio(bio); |
1da177e4 | 205 | } |
1da177e4 | 206 | |
1da177e4 LT |
207 | void blk_dump_rq_flags(struct request *rq, char *msg) |
208 | { | |
aebf526b CH |
209 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
210 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 211 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 212 | |
83096ebf TH |
213 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
214 | (unsigned long long)blk_rq_pos(rq), | |
215 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
216 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
217 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 218 | } |
1da177e4 LT |
219 | EXPORT_SYMBOL(blk_dump_rq_flags); |
220 | ||
3cca6dc1 | 221 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 222 | { |
3cca6dc1 | 223 | struct request_queue *q; |
1da177e4 | 224 | |
3cca6dc1 JA |
225 | q = container_of(work, struct request_queue, delay_work.work); |
226 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 227 | __blk_run_queue(q); |
3cca6dc1 | 228 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 229 | } |
1da177e4 LT |
230 | |
231 | /** | |
3cca6dc1 JA |
232 | * blk_delay_queue - restart queueing after defined interval |
233 | * @q: The &struct request_queue in question | |
234 | * @msecs: Delay in msecs | |
1da177e4 LT |
235 | * |
236 | * Description: | |
3cca6dc1 JA |
237 | * Sometimes queueing needs to be postponed for a little while, to allow |
238 | * resources to come back. This function will make sure that queueing is | |
2fff8a92 | 239 | * restarted around the specified time. |
3cca6dc1 JA |
240 | */ |
241 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 242 | { |
2fff8a92 | 243 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 244 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 245 | |
70460571 BVA |
246 | if (likely(!blk_queue_dead(q))) |
247 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
248 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 249 | } |
3cca6dc1 | 250 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 251 | |
21491412 JA |
252 | /** |
253 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
254 | * @q: The &struct request_queue in question | |
255 | * | |
256 | * Description: | |
257 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
258 | * ensure that the request_fn for the queue is run from an async | |
259 | * context. | |
260 | **/ | |
261 | void blk_start_queue_async(struct request_queue *q) | |
262 | { | |
2fff8a92 | 263 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 264 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 265 | |
21491412 JA |
266 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
267 | blk_run_queue_async(q); | |
268 | } | |
269 | EXPORT_SYMBOL(blk_start_queue_async); | |
270 | ||
1da177e4 LT |
271 | /** |
272 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 273 | * @q: The &struct request_queue in question |
1da177e4 LT |
274 | * |
275 | * Description: | |
276 | * blk_start_queue() will clear the stop flag on the queue, and call | |
277 | * the request_fn for the queue if it was in a stopped state when | |
2fff8a92 | 278 | * entered. Also see blk_stop_queue(). |
1da177e4 | 279 | **/ |
165125e1 | 280 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 281 | { |
2fff8a92 | 282 | lockdep_assert_held(q->queue_lock); |
4ddd56b0 | 283 | WARN_ON(!in_interrupt() && !irqs_disabled()); |
332ebbf7 | 284 | WARN_ON_ONCE(q->mq_ops); |
a038e253 | 285 | |
75ad23bc | 286 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 287 | __blk_run_queue(q); |
1da177e4 | 288 | } |
1da177e4 LT |
289 | EXPORT_SYMBOL(blk_start_queue); |
290 | ||
291 | /** | |
292 | * blk_stop_queue - stop a queue | |
165125e1 | 293 | * @q: The &struct request_queue in question |
1da177e4 LT |
294 | * |
295 | * Description: | |
296 | * The Linux block layer assumes that a block driver will consume all | |
297 | * entries on the request queue when the request_fn strategy is called. | |
298 | * Often this will not happen, because of hardware limitations (queue | |
299 | * depth settings). If a device driver gets a 'queue full' response, | |
300 | * or if it simply chooses not to queue more I/O at one point, it can | |
301 | * call this function to prevent the request_fn from being called until | |
302 | * the driver has signalled it's ready to go again. This happens by calling | |
2fff8a92 | 303 | * blk_start_queue() to restart queue operations. |
1da177e4 | 304 | **/ |
165125e1 | 305 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 306 | { |
2fff8a92 | 307 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 308 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 309 | |
136b5721 | 310 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 311 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
312 | } |
313 | EXPORT_SYMBOL(blk_stop_queue); | |
314 | ||
315 | /** | |
316 | * blk_sync_queue - cancel any pending callbacks on a queue | |
317 | * @q: the queue | |
318 | * | |
319 | * Description: | |
320 | * The block layer may perform asynchronous callback activity | |
321 | * on a queue, such as calling the unplug function after a timeout. | |
322 | * A block device may call blk_sync_queue to ensure that any | |
323 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 324 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
325 | * that its ->make_request_fn will not re-add plugging prior to calling |
326 | * this function. | |
327 | * | |
da527770 | 328 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 329 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 330 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 331 | * |
1da177e4 LT |
332 | */ |
333 | void blk_sync_queue(struct request_queue *q) | |
334 | { | |
70ed28b9 | 335 | del_timer_sync(&q->timeout); |
77a38e88 | 336 | cancel_work_sync(&q->timeout_work); |
f04c1fe7 ML |
337 | |
338 | if (q->mq_ops) { | |
339 | struct blk_mq_hw_ctx *hctx; | |
340 | int i; | |
341 | ||
21c6e939 | 342 | queue_for_each_hw_ctx(q, hctx, i) |
9f993737 | 343 | cancel_delayed_work_sync(&hctx->run_work); |
f04c1fe7 ML |
344 | } else { |
345 | cancel_delayed_work_sync(&q->delay_work); | |
346 | } | |
1da177e4 LT |
347 | } |
348 | EXPORT_SYMBOL(blk_sync_queue); | |
349 | ||
c246e80d BVA |
350 | /** |
351 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
352 | * @q: The queue to run | |
353 | * | |
354 | * Description: | |
355 | * Invoke request handling on a queue if there are any pending requests. | |
356 | * May be used to restart request handling after a request has completed. | |
357 | * This variant runs the queue whether or not the queue has been | |
358 | * stopped. Must be called with the queue lock held and interrupts | |
359 | * disabled. See also @blk_run_queue. | |
360 | */ | |
361 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
362 | { | |
2fff8a92 | 363 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 364 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 365 | |
c246e80d BVA |
366 | if (unlikely(blk_queue_dead(q))) |
367 | return; | |
368 | ||
24faf6f6 BVA |
369 | /* |
370 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
371 | * the queue lock internally. As a result multiple threads may be | |
372 | * running such a request function concurrently. Keep track of the | |
373 | * number of active request_fn invocations such that blk_drain_queue() | |
374 | * can wait until all these request_fn calls have finished. | |
375 | */ | |
376 | q->request_fn_active++; | |
c246e80d | 377 | q->request_fn(q); |
24faf6f6 | 378 | q->request_fn_active--; |
c246e80d | 379 | } |
a7928c15 | 380 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 381 | |
1da177e4 | 382 | /** |
80a4b58e | 383 | * __blk_run_queue - run a single device queue |
1da177e4 | 384 | * @q: The queue to run |
80a4b58e JA |
385 | * |
386 | * Description: | |
2fff8a92 | 387 | * See @blk_run_queue. |
1da177e4 | 388 | */ |
24ecfbe2 | 389 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 390 | { |
2fff8a92 | 391 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 392 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 393 | |
a538cd03 TH |
394 | if (unlikely(blk_queue_stopped(q))) |
395 | return; | |
396 | ||
c246e80d | 397 | __blk_run_queue_uncond(q); |
75ad23bc NP |
398 | } |
399 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 400 | |
24ecfbe2 CH |
401 | /** |
402 | * blk_run_queue_async - run a single device queue in workqueue context | |
403 | * @q: The queue to run | |
404 | * | |
405 | * Description: | |
406 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
2fff8a92 BVA |
407 | * of us. |
408 | * | |
409 | * Note: | |
410 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
411 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
412 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
24ecfbe2 CH |
413 | */ |
414 | void blk_run_queue_async(struct request_queue *q) | |
415 | { | |
2fff8a92 | 416 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 417 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 418 | |
70460571 | 419 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 420 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 421 | } |
c21e6beb | 422 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 423 | |
75ad23bc NP |
424 | /** |
425 | * blk_run_queue - run a single device queue | |
426 | * @q: The queue to run | |
80a4b58e JA |
427 | * |
428 | * Description: | |
429 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 430 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
431 | */ |
432 | void blk_run_queue(struct request_queue *q) | |
433 | { | |
434 | unsigned long flags; | |
435 | ||
332ebbf7 BVA |
436 | WARN_ON_ONCE(q->mq_ops); |
437 | ||
75ad23bc | 438 | spin_lock_irqsave(q->queue_lock, flags); |
24ecfbe2 | 439 | __blk_run_queue(q); |
1da177e4 LT |
440 | spin_unlock_irqrestore(q->queue_lock, flags); |
441 | } | |
442 | EXPORT_SYMBOL(blk_run_queue); | |
443 | ||
165125e1 | 444 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
445 | { |
446 | kobject_put(&q->kobj); | |
447 | } | |
d86e0e83 | 448 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 449 | |
e3c78ca5 | 450 | /** |
807592a4 | 451 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 452 | * @q: queue to drain |
c9a929dd | 453 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 454 | * |
c9a929dd TH |
455 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
456 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
457 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 458 | */ |
807592a4 BVA |
459 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
460 | __releases(q->queue_lock) | |
461 | __acquires(q->queue_lock) | |
e3c78ca5 | 462 | { |
458f27a9 AH |
463 | int i; |
464 | ||
807592a4 | 465 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 466 | WARN_ON_ONCE(q->mq_ops); |
807592a4 | 467 | |
e3c78ca5 | 468 | while (true) { |
481a7d64 | 469 | bool drain = false; |
e3c78ca5 | 470 | |
b855b04a TH |
471 | /* |
472 | * The caller might be trying to drain @q before its | |
473 | * elevator is initialized. | |
474 | */ | |
475 | if (q->elevator) | |
476 | elv_drain_elevator(q); | |
477 | ||
5efd6113 | 478 | blkcg_drain_queue(q); |
e3c78ca5 | 479 | |
4eabc941 TH |
480 | /* |
481 | * This function might be called on a queue which failed | |
b855b04a TH |
482 | * driver init after queue creation or is not yet fully |
483 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
484 | * in such cases. Kick queue iff dispatch queue has | |
485 | * something on it and @q has request_fn set. | |
4eabc941 | 486 | */ |
b855b04a | 487 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 488 | __blk_run_queue(q); |
c9a929dd | 489 | |
8a5ecdd4 | 490 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 491 | drain |= q->request_fn_active; |
481a7d64 TH |
492 | |
493 | /* | |
494 | * Unfortunately, requests are queued at and tracked from | |
495 | * multiple places and there's no single counter which can | |
496 | * be drained. Check all the queues and counters. | |
497 | */ | |
498 | if (drain_all) { | |
e97c293c | 499 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
500 | drain |= !list_empty(&q->queue_head); |
501 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 502 | drain |= q->nr_rqs[i]; |
481a7d64 | 503 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
504 | if (fq) |
505 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
506 | } |
507 | } | |
e3c78ca5 | 508 | |
481a7d64 | 509 | if (!drain) |
e3c78ca5 | 510 | break; |
807592a4 BVA |
511 | |
512 | spin_unlock_irq(q->queue_lock); | |
513 | ||
e3c78ca5 | 514 | msleep(10); |
807592a4 BVA |
515 | |
516 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 517 | } |
458f27a9 AH |
518 | |
519 | /* | |
520 | * With queue marked dead, any woken up waiter will fail the | |
521 | * allocation path, so the wakeup chaining is lost and we're | |
522 | * left with hung waiters. We need to wake up those waiters. | |
523 | */ | |
524 | if (q->request_fn) { | |
a051661c TH |
525 | struct request_list *rl; |
526 | ||
a051661c TH |
527 | blk_queue_for_each_rl(rl, q) |
528 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
529 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 530 | } |
e3c78ca5 TH |
531 | } |
532 | ||
d732580b TH |
533 | /** |
534 | * blk_queue_bypass_start - enter queue bypass mode | |
535 | * @q: queue of interest | |
536 | * | |
537 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
538 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 539 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
540 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
541 | * inside queue or RCU read lock. | |
d732580b TH |
542 | */ |
543 | void blk_queue_bypass_start(struct request_queue *q) | |
544 | { | |
332ebbf7 BVA |
545 | WARN_ON_ONCE(q->mq_ops); |
546 | ||
d732580b | 547 | spin_lock_irq(q->queue_lock); |
776687bc | 548 | q->bypass_depth++; |
d732580b TH |
549 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
550 | spin_unlock_irq(q->queue_lock); | |
551 | ||
776687bc TH |
552 | /* |
553 | * Queues start drained. Skip actual draining till init is | |
554 | * complete. This avoids lenghty delays during queue init which | |
555 | * can happen many times during boot. | |
556 | */ | |
557 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
558 | spin_lock_irq(q->queue_lock); |
559 | __blk_drain_queue(q, false); | |
560 | spin_unlock_irq(q->queue_lock); | |
561 | ||
b82d4b19 TH |
562 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
563 | synchronize_rcu(); | |
564 | } | |
d732580b TH |
565 | } |
566 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
567 | ||
568 | /** | |
569 | * blk_queue_bypass_end - leave queue bypass mode | |
570 | * @q: queue of interest | |
571 | * | |
572 | * Leave bypass mode and restore the normal queueing behavior. | |
332ebbf7 BVA |
573 | * |
574 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
575 | * this function is called for both blk-sq and blk-mq queues. | |
d732580b TH |
576 | */ |
577 | void blk_queue_bypass_end(struct request_queue *q) | |
578 | { | |
579 | spin_lock_irq(q->queue_lock); | |
580 | if (!--q->bypass_depth) | |
581 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
582 | WARN_ON_ONCE(q->bypass_depth < 0); | |
583 | spin_unlock_irq(q->queue_lock); | |
584 | } | |
585 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
586 | ||
aed3ea94 JA |
587 | void blk_set_queue_dying(struct request_queue *q) |
588 | { | |
1b856086 BVA |
589 | spin_lock_irq(q->queue_lock); |
590 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
591 | spin_unlock_irq(q->queue_lock); | |
aed3ea94 | 592 | |
d3cfb2a0 ML |
593 | /* |
594 | * When queue DYING flag is set, we need to block new req | |
595 | * entering queue, so we call blk_freeze_queue_start() to | |
596 | * prevent I/O from crossing blk_queue_enter(). | |
597 | */ | |
598 | blk_freeze_queue_start(q); | |
599 | ||
aed3ea94 JA |
600 | if (q->mq_ops) |
601 | blk_mq_wake_waiters(q); | |
602 | else { | |
603 | struct request_list *rl; | |
604 | ||
bbfc3c5d | 605 | spin_lock_irq(q->queue_lock); |
aed3ea94 JA |
606 | blk_queue_for_each_rl(rl, q) { |
607 | if (rl->rq_pool) { | |
60bed713 ML |
608 | wake_up_all(&rl->wait[BLK_RW_SYNC]); |
609 | wake_up_all(&rl->wait[BLK_RW_ASYNC]); | |
aed3ea94 JA |
610 | } |
611 | } | |
bbfc3c5d | 612 | spin_unlock_irq(q->queue_lock); |
aed3ea94 JA |
613 | } |
614 | } | |
615 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
616 | ||
c9a929dd TH |
617 | /** |
618 | * blk_cleanup_queue - shutdown a request queue | |
619 | * @q: request queue to shutdown | |
620 | * | |
c246e80d BVA |
621 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
622 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 623 | */ |
6728cb0e | 624 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 625 | { |
c9a929dd | 626 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 627 | |
3f3299d5 | 628 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 629 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 630 | blk_set_queue_dying(q); |
c9a929dd | 631 | spin_lock_irq(lock); |
6ecf23af | 632 | |
80fd9979 | 633 | /* |
3f3299d5 | 634 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
635 | * that, unlike blk_queue_bypass_start(), we aren't performing |
636 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
637 | * as some drivers create and destroy a lot of queues while | |
638 | * probing. This is still safe because blk_release_queue() will be | |
639 | * called only after the queue refcnt drops to zero and nothing, | |
640 | * RCU or not, would be traversing the queue by then. | |
641 | */ | |
6ecf23af TH |
642 | q->bypass_depth++; |
643 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
644 | ||
c9a929dd TH |
645 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
646 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 647 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
648 | spin_unlock_irq(lock); |
649 | mutex_unlock(&q->sysfs_lock); | |
650 | ||
c246e80d BVA |
651 | /* |
652 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
653 | * prevent that q->request_fn() gets invoked after draining finished. | |
654 | */ | |
3ef28e83 | 655 | blk_freeze_queue(q); |
9c1051aa OS |
656 | spin_lock_irq(lock); |
657 | if (!q->mq_ops) | |
43a5e4e2 | 658 | __blk_drain_queue(q, true); |
c246e80d | 659 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 660 | spin_unlock_irq(lock); |
c9a929dd | 661 | |
5a48fc14 DW |
662 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
663 | blk_flush_integrity(); | |
664 | ||
c9a929dd | 665 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 666 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
667 | blk_sync_queue(q); |
668 | ||
45a9c9d9 BVA |
669 | if (q->mq_ops) |
670 | blk_mq_free_queue(q); | |
3ef28e83 | 671 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 672 | |
5e5cfac0 AH |
673 | spin_lock_irq(lock); |
674 | if (q->queue_lock != &q->__queue_lock) | |
675 | q->queue_lock = &q->__queue_lock; | |
676 | spin_unlock_irq(lock); | |
677 | ||
c9a929dd | 678 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
679 | blk_put_queue(q); |
680 | } | |
1da177e4 LT |
681 | EXPORT_SYMBOL(blk_cleanup_queue); |
682 | ||
271508db | 683 | /* Allocate memory local to the request queue */ |
6d247d7f | 684 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) |
271508db | 685 | { |
6d247d7f CH |
686 | struct request_queue *q = data; |
687 | ||
688 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
271508db DR |
689 | } |
690 | ||
6d247d7f | 691 | static void free_request_simple(void *element, void *data) |
271508db DR |
692 | { |
693 | kmem_cache_free(request_cachep, element); | |
694 | } | |
695 | ||
6d247d7f CH |
696 | static void *alloc_request_size(gfp_t gfp_mask, void *data) |
697 | { | |
698 | struct request_queue *q = data; | |
699 | struct request *rq; | |
700 | ||
701 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
702 | q->node); | |
703 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
704 | kfree(rq); | |
705 | rq = NULL; | |
706 | } | |
707 | return rq; | |
708 | } | |
709 | ||
710 | static void free_request_size(void *element, void *data) | |
711 | { | |
712 | struct request_queue *q = data; | |
713 | ||
714 | if (q->exit_rq_fn) | |
715 | q->exit_rq_fn(q, element); | |
716 | kfree(element); | |
717 | } | |
718 | ||
5b788ce3 TH |
719 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
720 | gfp_t gfp_mask) | |
1da177e4 | 721 | { |
1abec4fd MS |
722 | if (unlikely(rl->rq_pool)) |
723 | return 0; | |
724 | ||
5b788ce3 | 725 | rl->q = q; |
1faa16d2 JA |
726 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
727 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
728 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
729 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 730 | |
6d247d7f CH |
731 | if (q->cmd_size) { |
732 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
733 | alloc_request_size, free_request_size, | |
734 | q, gfp_mask, q->node); | |
735 | } else { | |
736 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
737 | alloc_request_simple, free_request_simple, | |
738 | q, gfp_mask, q->node); | |
739 | } | |
1da177e4 LT |
740 | if (!rl->rq_pool) |
741 | return -ENOMEM; | |
742 | ||
b425e504 BVA |
743 | if (rl != &q->root_rl) |
744 | WARN_ON_ONCE(!blk_get_queue(q)); | |
745 | ||
1da177e4 LT |
746 | return 0; |
747 | } | |
748 | ||
b425e504 | 749 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) |
5b788ce3 | 750 | { |
b425e504 | 751 | if (rl->rq_pool) { |
5b788ce3 | 752 | mempool_destroy(rl->rq_pool); |
b425e504 BVA |
753 | if (rl != &q->root_rl) |
754 | blk_put_queue(q); | |
755 | } | |
5b788ce3 TH |
756 | } |
757 | ||
165125e1 | 758 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 759 | { |
c304a51b | 760 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
761 | } |
762 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 763 | |
6f3b0e8b | 764 | int blk_queue_enter(struct request_queue *q, bool nowait) |
3ef28e83 DW |
765 | { |
766 | while (true) { | |
767 | int ret; | |
768 | ||
769 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
770 | return 0; | |
771 | ||
6f3b0e8b | 772 | if (nowait) |
3ef28e83 DW |
773 | return -EBUSY; |
774 | ||
5ed61d3f | 775 | /* |
1671d522 | 776 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 777 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
778 | * .q_usage_counter and reading .mq_freeze_depth or |
779 | * queue dying flag, otherwise the following wait may | |
780 | * never return if the two reads are reordered. | |
5ed61d3f ML |
781 | */ |
782 | smp_rmb(); | |
783 | ||
3ef28e83 DW |
784 | ret = wait_event_interruptible(q->mq_freeze_wq, |
785 | !atomic_read(&q->mq_freeze_depth) || | |
786 | blk_queue_dying(q)); | |
787 | if (blk_queue_dying(q)) | |
788 | return -ENODEV; | |
789 | if (ret) | |
790 | return ret; | |
791 | } | |
792 | } | |
793 | ||
794 | void blk_queue_exit(struct request_queue *q) | |
795 | { | |
796 | percpu_ref_put(&q->q_usage_counter); | |
797 | } | |
798 | ||
799 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
800 | { | |
801 | struct request_queue *q = | |
802 | container_of(ref, struct request_queue, q_usage_counter); | |
803 | ||
804 | wake_up_all(&q->mq_freeze_wq); | |
805 | } | |
806 | ||
287922eb CH |
807 | static void blk_rq_timed_out_timer(unsigned long data) |
808 | { | |
809 | struct request_queue *q = (struct request_queue *)data; | |
810 | ||
811 | kblockd_schedule_work(&q->timeout_work); | |
812 | } | |
813 | ||
165125e1 | 814 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 815 | { |
165125e1 | 816 | struct request_queue *q; |
1946089a | 817 | |
8324aa91 | 818 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 819 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
820 | if (!q) |
821 | return NULL; | |
822 | ||
00380a40 | 823 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 824 | if (q->id < 0) |
3d2936f4 | 825 | goto fail_q; |
a73f730d | 826 | |
93b27e72 | 827 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
54efd50b KO |
828 | if (!q->bio_split) |
829 | goto fail_id; | |
830 | ||
d03f6cdc JK |
831 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
832 | if (!q->backing_dev_info) | |
833 | goto fail_split; | |
834 | ||
a83b576c JA |
835 | q->stats = blk_alloc_queue_stats(); |
836 | if (!q->stats) | |
837 | goto fail_stats; | |
838 | ||
dc3b17cc | 839 | q->backing_dev_info->ra_pages = |
09cbfeaf | 840 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
dc3b17cc JK |
841 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
842 | q->backing_dev_info->name = "block"; | |
5151412d | 843 | q->node = node_id; |
0989a025 | 844 | |
dc3b17cc | 845 | setup_timer(&q->backing_dev_info->laptop_mode_wb_timer, |
31373d09 | 846 | laptop_mode_timer_fn, (unsigned long) q); |
242f9dcb | 847 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
77a38e88 | 848 | INIT_WORK(&q->timeout_work, NULL); |
b855b04a | 849 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 850 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 851 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 852 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 853 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 854 | #endif |
3cca6dc1 | 855 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 856 | |
8324aa91 | 857 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 858 | |
5acb3cc2 WL |
859 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
860 | mutex_init(&q->blk_trace_mutex); | |
861 | #endif | |
483f4afc | 862 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 863 | spin_lock_init(&q->__queue_lock); |
483f4afc | 864 | |
c94a96ac VG |
865 | /* |
866 | * By default initialize queue_lock to internal lock and driver can | |
867 | * override it later if need be. | |
868 | */ | |
869 | q->queue_lock = &q->__queue_lock; | |
870 | ||
b82d4b19 TH |
871 | /* |
872 | * A queue starts its life with bypass turned on to avoid | |
873 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
874 | * init. The initial bypass will be finished when the queue is |
875 | * registered by blk_register_queue(). | |
b82d4b19 TH |
876 | */ |
877 | q->bypass_depth = 1; | |
878 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
879 | ||
320ae51f JA |
880 | init_waitqueue_head(&q->mq_freeze_wq); |
881 | ||
3ef28e83 DW |
882 | /* |
883 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
884 | * See blk_register_queue() for details. | |
885 | */ | |
886 | if (percpu_ref_init(&q->q_usage_counter, | |
887 | blk_queue_usage_counter_release, | |
888 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 889 | goto fail_bdi; |
f51b802c | 890 | |
3ef28e83 DW |
891 | if (blkcg_init_queue(q)) |
892 | goto fail_ref; | |
893 | ||
1da177e4 | 894 | return q; |
a73f730d | 895 | |
3ef28e83 DW |
896 | fail_ref: |
897 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 898 | fail_bdi: |
a83b576c JA |
899 | blk_free_queue_stats(q->stats); |
900 | fail_stats: | |
d03f6cdc | 901 | bdi_put(q->backing_dev_info); |
54efd50b KO |
902 | fail_split: |
903 | bioset_free(q->bio_split); | |
a73f730d TH |
904 | fail_id: |
905 | ida_simple_remove(&blk_queue_ida, q->id); | |
906 | fail_q: | |
907 | kmem_cache_free(blk_requestq_cachep, q); | |
908 | return NULL; | |
1da177e4 | 909 | } |
1946089a | 910 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
911 | |
912 | /** | |
913 | * blk_init_queue - prepare a request queue for use with a block device | |
914 | * @rfn: The function to be called to process requests that have been | |
915 | * placed on the queue. | |
916 | * @lock: Request queue spin lock | |
917 | * | |
918 | * Description: | |
919 | * If a block device wishes to use the standard request handling procedures, | |
920 | * which sorts requests and coalesces adjacent requests, then it must | |
921 | * call blk_init_queue(). The function @rfn will be called when there | |
922 | * are requests on the queue that need to be processed. If the device | |
923 | * supports plugging, then @rfn may not be called immediately when requests | |
924 | * are available on the queue, but may be called at some time later instead. | |
925 | * Plugged queues are generally unplugged when a buffer belonging to one | |
926 | * of the requests on the queue is needed, or due to memory pressure. | |
927 | * | |
928 | * @rfn is not required, or even expected, to remove all requests off the | |
929 | * queue, but only as many as it can handle at a time. If it does leave | |
930 | * requests on the queue, it is responsible for arranging that the requests | |
931 | * get dealt with eventually. | |
932 | * | |
933 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
934 | * request queue; this lock will be taken also from interrupt context, so irq |
935 | * disabling is needed for it. | |
1da177e4 | 936 | * |
710027a4 | 937 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
938 | * it didn't succeed. |
939 | * | |
940 | * Note: | |
941 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
942 | * when the block device is deactivated (such as at module unload). | |
943 | **/ | |
1946089a | 944 | |
165125e1 | 945 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 946 | { |
c304a51b | 947 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
948 | } |
949 | EXPORT_SYMBOL(blk_init_queue); | |
950 | ||
165125e1 | 951 | struct request_queue * |
1946089a CL |
952 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
953 | { | |
5ea708d1 | 954 | struct request_queue *q; |
1da177e4 | 955 | |
5ea708d1 CH |
956 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
957 | if (!q) | |
c86d1b8a MS |
958 | return NULL; |
959 | ||
5ea708d1 CH |
960 | q->request_fn = rfn; |
961 | if (lock) | |
962 | q->queue_lock = lock; | |
963 | if (blk_init_allocated_queue(q) < 0) { | |
964 | blk_cleanup_queue(q); | |
965 | return NULL; | |
966 | } | |
18741986 | 967 | |
7982e90c | 968 | return q; |
01effb0d MS |
969 | } |
970 | EXPORT_SYMBOL(blk_init_queue_node); | |
971 | ||
dece1635 | 972 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 973 | |
1da177e4 | 974 | |
5ea708d1 CH |
975 | int blk_init_allocated_queue(struct request_queue *q) |
976 | { | |
332ebbf7 BVA |
977 | WARN_ON_ONCE(q->mq_ops); |
978 | ||
6d247d7f | 979 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); |
ba483388 | 980 | if (!q->fq) |
5ea708d1 | 981 | return -ENOMEM; |
7982e90c | 982 | |
6d247d7f CH |
983 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) |
984 | goto out_free_flush_queue; | |
7982e90c | 985 | |
a051661c | 986 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
6d247d7f | 987 | goto out_exit_flush_rq; |
1da177e4 | 988 | |
287922eb | 989 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
60ea8226 | 990 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac | 991 | |
f3b144aa JA |
992 | /* |
993 | * This also sets hw/phys segments, boundary and size | |
994 | */ | |
c20e8de2 | 995 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 996 | |
44ec9542 AS |
997 | q->sg_reserved_size = INT_MAX; |
998 | ||
eb1c160b TS |
999 | /* Protect q->elevator from elevator_change */ |
1000 | mutex_lock(&q->sysfs_lock); | |
1001 | ||
b82d4b19 | 1002 | /* init elevator */ |
eb1c160b TS |
1003 | if (elevator_init(q, NULL)) { |
1004 | mutex_unlock(&q->sysfs_lock); | |
6d247d7f | 1005 | goto out_exit_flush_rq; |
eb1c160b TS |
1006 | } |
1007 | ||
1008 | mutex_unlock(&q->sysfs_lock); | |
5ea708d1 | 1009 | return 0; |
eb1c160b | 1010 | |
6d247d7f CH |
1011 | out_exit_flush_rq: |
1012 | if (q->exit_rq_fn) | |
1013 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1014 | out_free_flush_queue: | |
ba483388 | 1015 | blk_free_flush_queue(q->fq); |
5ea708d1 | 1016 | return -ENOMEM; |
1da177e4 | 1017 | } |
5151412d | 1018 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 1019 | |
09ac46c4 | 1020 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 1021 | { |
3f3299d5 | 1022 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
1023 | __blk_get_queue(q); |
1024 | return true; | |
1da177e4 LT |
1025 | } |
1026 | ||
09ac46c4 | 1027 | return false; |
1da177e4 | 1028 | } |
d86e0e83 | 1029 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 1030 | |
5b788ce3 | 1031 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 1032 | { |
e8064021 | 1033 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 1034 | elv_put_request(rl->q, rq); |
f1f8cc94 | 1035 | if (rq->elv.icq) |
11a3122f | 1036 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
1037 | } |
1038 | ||
5b788ce3 | 1039 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
1040 | } |
1041 | ||
1da177e4 LT |
1042 | /* |
1043 | * ioc_batching returns true if the ioc is a valid batching request and | |
1044 | * should be given priority access to a request. | |
1045 | */ | |
165125e1 | 1046 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1047 | { |
1048 | if (!ioc) | |
1049 | return 0; | |
1050 | ||
1051 | /* | |
1052 | * Make sure the process is able to allocate at least 1 request | |
1053 | * even if the batch times out, otherwise we could theoretically | |
1054 | * lose wakeups. | |
1055 | */ | |
1056 | return ioc->nr_batch_requests == q->nr_batching || | |
1057 | (ioc->nr_batch_requests > 0 | |
1058 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1063 | * will cause the process to be a "batcher" on all queues in the system. This | |
1064 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1065 | * a nice run. | |
1066 | */ | |
165125e1 | 1067 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1068 | { |
1069 | if (!ioc || ioc_batching(q, ioc)) | |
1070 | return; | |
1071 | ||
1072 | ioc->nr_batch_requests = q->nr_batching; | |
1073 | ioc->last_waited = jiffies; | |
1074 | } | |
1075 | ||
5b788ce3 | 1076 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 1077 | { |
5b788ce3 | 1078 | struct request_queue *q = rl->q; |
1da177e4 | 1079 | |
d40f75a0 TH |
1080 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
1081 | blk_clear_congested(rl, sync); | |
1da177e4 | 1082 | |
1faa16d2 JA |
1083 | if (rl->count[sync] + 1 <= q->nr_requests) { |
1084 | if (waitqueue_active(&rl->wait[sync])) | |
1085 | wake_up(&rl->wait[sync]); | |
1da177e4 | 1086 | |
5b788ce3 | 1087 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
1088 | } |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * A request has just been released. Account for it, update the full and | |
1093 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1094 | */ | |
e8064021 CH |
1095 | static void freed_request(struct request_list *rl, bool sync, |
1096 | req_flags_t rq_flags) | |
1da177e4 | 1097 | { |
5b788ce3 | 1098 | struct request_queue *q = rl->q; |
1da177e4 | 1099 | |
8a5ecdd4 | 1100 | q->nr_rqs[sync]--; |
1faa16d2 | 1101 | rl->count[sync]--; |
e8064021 | 1102 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 1103 | q->nr_rqs_elvpriv--; |
1da177e4 | 1104 | |
5b788ce3 | 1105 | __freed_request(rl, sync); |
1da177e4 | 1106 | |
1faa16d2 | 1107 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 1108 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
1109 | } |
1110 | ||
e3a2b3f9 JA |
1111 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
1112 | { | |
1113 | struct request_list *rl; | |
d40f75a0 | 1114 | int on_thresh, off_thresh; |
e3a2b3f9 | 1115 | |
332ebbf7 BVA |
1116 | WARN_ON_ONCE(q->mq_ops); |
1117 | ||
e3a2b3f9 JA |
1118 | spin_lock_irq(q->queue_lock); |
1119 | q->nr_requests = nr; | |
1120 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
1121 | on_thresh = queue_congestion_on_threshold(q); |
1122 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 1123 | |
d40f75a0 TH |
1124 | blk_queue_for_each_rl(rl, q) { |
1125 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1126 | blk_set_congested(rl, BLK_RW_SYNC); | |
1127 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1128 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 1129 | |
d40f75a0 TH |
1130 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1131 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1132 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1133 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1134 | |
e3a2b3f9 JA |
1135 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1136 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1137 | } else { | |
1138 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1139 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1140 | } | |
1141 | ||
1142 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1143 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1144 | } else { | |
1145 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1146 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | spin_unlock_irq(q->queue_lock); | |
1151 | return 0; | |
1152 | } | |
1153 | ||
da8303c6 | 1154 | /** |
a06e05e6 | 1155 | * __get_request - get a free request |
5b788ce3 | 1156 | * @rl: request list to allocate from |
ef295ecf | 1157 | * @op: operation and flags |
da8303c6 TH |
1158 | * @bio: bio to allocate request for (can be %NULL) |
1159 | * @gfp_mask: allocation mask | |
1160 | * | |
1161 | * Get a free request from @q. This function may fail under memory | |
1162 | * pressure or if @q is dead. | |
1163 | * | |
da3dae54 | 1164 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1165 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1166 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1167 | */ |
ef295ecf CH |
1168 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
1169 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1170 | { |
5b788ce3 | 1171 | struct request_queue *q = rl->q; |
b679281a | 1172 | struct request *rq; |
7f4b35d1 TH |
1173 | struct elevator_type *et = q->elevator->type; |
1174 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1175 | struct io_cq *icq = NULL; |
ef295ecf | 1176 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1177 | int may_queue; |
e8064021 | 1178 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1179 | |
2fff8a92 BVA |
1180 | lockdep_assert_held(q->queue_lock); |
1181 | ||
3f3299d5 | 1182 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1183 | return ERR_PTR(-ENODEV); |
da8303c6 | 1184 | |
ef295ecf | 1185 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1186 | if (may_queue == ELV_MQUEUE_NO) |
1187 | goto rq_starved; | |
1188 | ||
1faa16d2 JA |
1189 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1190 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1191 | /* |
1192 | * The queue will fill after this allocation, so set | |
1193 | * it as full, and mark this process as "batching". | |
1194 | * This process will be allowed to complete a batch of | |
1195 | * requests, others will be blocked. | |
1196 | */ | |
5b788ce3 | 1197 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1198 | ioc_set_batching(q, ioc); |
5b788ce3 | 1199 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1200 | } else { |
1201 | if (may_queue != ELV_MQUEUE_MUST | |
1202 | && !ioc_batching(q, ioc)) { | |
1203 | /* | |
1204 | * The queue is full and the allocating | |
1205 | * process is not a "batcher", and not | |
1206 | * exempted by the IO scheduler | |
1207 | */ | |
a492f075 | 1208 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1209 | } |
1210 | } | |
1da177e4 | 1211 | } |
d40f75a0 | 1212 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1213 | } |
1214 | ||
082cf69e JA |
1215 | /* |
1216 | * Only allow batching queuers to allocate up to 50% over the defined | |
1217 | * limit of requests, otherwise we could have thousands of requests | |
1218 | * allocated with any setting of ->nr_requests | |
1219 | */ | |
1faa16d2 | 1220 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1221 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1222 | |
8a5ecdd4 | 1223 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1224 | rl->count[is_sync]++; |
1225 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1226 | |
f1f8cc94 TH |
1227 | /* |
1228 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1229 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1230 | * prevent the current elevator from being destroyed until the new |
1231 | * request is freed. This guarantees icq's won't be destroyed and | |
1232 | * makes creating new ones safe. | |
1233 | * | |
e6f7f93d CH |
1234 | * Flush requests do not use the elevator so skip initialization. |
1235 | * This allows a request to share the flush and elevator data. | |
1236 | * | |
f1f8cc94 TH |
1237 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
1238 | * it will be created after releasing queue_lock. | |
1239 | */ | |
e6f7f93d | 1240 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { |
e8064021 | 1241 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1242 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1243 | if (et->icq_cache && ioc) |
1244 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1245 | } |
cb98fc8b | 1246 | |
f253b86b | 1247 | if (blk_queue_io_stat(q)) |
e8064021 | 1248 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1249 | spin_unlock_irq(q->queue_lock); |
1250 | ||
29e2b09a | 1251 | /* allocate and init request */ |
5b788ce3 | 1252 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1253 | if (!rq) |
b679281a | 1254 | goto fail_alloc; |
1da177e4 | 1255 | |
29e2b09a | 1256 | blk_rq_init(q, rq); |
a051661c | 1257 | blk_rq_set_rl(rq, rl); |
ef295ecf | 1258 | rq->cmd_flags = op; |
e8064021 | 1259 | rq->rq_flags = rq_flags; |
29e2b09a | 1260 | |
aaf7c680 | 1261 | /* init elvpriv */ |
e8064021 | 1262 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1263 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1264 | if (ioc) |
1265 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1266 | if (!icq) |
1267 | goto fail_elvpriv; | |
29e2b09a | 1268 | } |
aaf7c680 TH |
1269 | |
1270 | rq->elv.icq = icq; | |
1271 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1272 | goto fail_elvpriv; | |
1273 | ||
1274 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1275 | if (icq) |
1276 | get_io_context(icq->ioc); | |
1277 | } | |
aaf7c680 | 1278 | out: |
88ee5ef1 JA |
1279 | /* |
1280 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1281 | * OK, if the queue is under the request limit then requests need | |
1282 | * not count toward the nr_batch_requests limit. There will always | |
1283 | * be some limit enforced by BLK_BATCH_TIME. | |
1284 | */ | |
1da177e4 LT |
1285 | if (ioc_batching(q, ioc)) |
1286 | ioc->nr_batch_requests--; | |
6728cb0e | 1287 | |
e6a40b09 | 1288 | trace_block_getrq(q, bio, op); |
1da177e4 | 1289 | return rq; |
b679281a | 1290 | |
aaf7c680 TH |
1291 | fail_elvpriv: |
1292 | /* | |
1293 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1294 | * and may fail indefinitely under memory pressure and thus | |
1295 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1296 | * disturb iosched and blkcg but weird is bettern than dead. | |
1297 | */ | |
7b2b10e0 | 1298 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
dc3b17cc | 1299 | __func__, dev_name(q->backing_dev_info->dev)); |
aaf7c680 | 1300 | |
e8064021 | 1301 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1302 | rq->elv.icq = NULL; |
1303 | ||
1304 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1305 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1306 | spin_unlock_irq(q->queue_lock); |
1307 | goto out; | |
1308 | ||
b679281a TH |
1309 | fail_alloc: |
1310 | /* | |
1311 | * Allocation failed presumably due to memory. Undo anything we | |
1312 | * might have messed up. | |
1313 | * | |
1314 | * Allocating task should really be put onto the front of the wait | |
1315 | * queue, but this is pretty rare. | |
1316 | */ | |
1317 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1318 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1319 | |
1320 | /* | |
1321 | * in the very unlikely event that allocation failed and no | |
1322 | * requests for this direction was pending, mark us starved so that | |
1323 | * freeing of a request in the other direction will notice | |
1324 | * us. another possible fix would be to split the rq mempool into | |
1325 | * READ and WRITE | |
1326 | */ | |
1327 | rq_starved: | |
1328 | if (unlikely(rl->count[is_sync] == 0)) | |
1329 | rl->starved[is_sync] = 1; | |
a492f075 | 1330 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1331 | } |
1332 | ||
da8303c6 | 1333 | /** |
a06e05e6 | 1334 | * get_request - get a free request |
da8303c6 | 1335 | * @q: request_queue to allocate request from |
ef295ecf | 1336 | * @op: operation and flags |
da8303c6 | 1337 | * @bio: bio to allocate request for (can be %NULL) |
a06e05e6 | 1338 | * @gfp_mask: allocation mask |
da8303c6 | 1339 | * |
d0164adc MG |
1340 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, |
1341 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1342 | * |
da3dae54 | 1343 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1344 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1345 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1346 | */ |
ef295ecf CH |
1347 | static struct request *get_request(struct request_queue *q, unsigned int op, |
1348 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1349 | { |
ef295ecf | 1350 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1351 | DEFINE_WAIT(wait); |
a051661c | 1352 | struct request_list *rl; |
1da177e4 | 1353 | struct request *rq; |
a051661c | 1354 | |
2fff8a92 | 1355 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1356 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1357 | |
a051661c | 1358 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ |
a06e05e6 | 1359 | retry: |
ef295ecf | 1360 | rq = __get_request(rl, op, bio, gfp_mask); |
a492f075 | 1361 | if (!IS_ERR(rq)) |
a06e05e6 | 1362 | return rq; |
1da177e4 | 1363 | |
03a07c92 GR |
1364 | if (op & REQ_NOWAIT) { |
1365 | blk_put_rl(rl); | |
1366 | return ERR_PTR(-EAGAIN); | |
1367 | } | |
1368 | ||
d0164adc | 1369 | if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) { |
a051661c | 1370 | blk_put_rl(rl); |
a492f075 | 1371 | return rq; |
a051661c | 1372 | } |
1da177e4 | 1373 | |
a06e05e6 TH |
1374 | /* wait on @rl and retry */ |
1375 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1376 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1377 | |
e6a40b09 | 1378 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1379 | |
a06e05e6 TH |
1380 | spin_unlock_irq(q->queue_lock); |
1381 | io_schedule(); | |
d6344532 | 1382 | |
a06e05e6 TH |
1383 | /* |
1384 | * After sleeping, we become a "batching" process and will be able | |
1385 | * to allocate at least one request, and up to a big batch of them | |
1386 | * for a small period time. See ioc_batching, ioc_set_batching | |
1387 | */ | |
a06e05e6 | 1388 | ioc_set_batching(q, current->io_context); |
05caf8db | 1389 | |
a06e05e6 TH |
1390 | spin_lock_irq(q->queue_lock); |
1391 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1392 | |
a06e05e6 | 1393 | goto retry; |
1da177e4 LT |
1394 | } |
1395 | ||
cd6ce148 BVA |
1396 | static struct request *blk_old_get_request(struct request_queue *q, |
1397 | unsigned int op, gfp_t gfp_mask) | |
1da177e4 LT |
1398 | { |
1399 | struct request *rq; | |
1400 | ||
332ebbf7 BVA |
1401 | WARN_ON_ONCE(q->mq_ops); |
1402 | ||
7f4b35d1 TH |
1403 | /* create ioc upfront */ |
1404 | create_io_context(gfp_mask, q->node); | |
1405 | ||
d6344532 | 1406 | spin_lock_irq(q->queue_lock); |
cd6ce148 | 1407 | rq = get_request(q, op, NULL, gfp_mask); |
0c4de0f3 | 1408 | if (IS_ERR(rq)) { |
da8303c6 | 1409 | spin_unlock_irq(q->queue_lock); |
0c4de0f3 CH |
1410 | return rq; |
1411 | } | |
1da177e4 | 1412 | |
0c4de0f3 CH |
1413 | /* q->queue_lock is unlocked at this point */ |
1414 | rq->__data_len = 0; | |
1415 | rq->__sector = (sector_t) -1; | |
1416 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1417 | return rq; |
1418 | } | |
320ae51f | 1419 | |
cd6ce148 BVA |
1420 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
1421 | gfp_t gfp_mask) | |
320ae51f | 1422 | { |
d280bab3 BVA |
1423 | struct request *req; |
1424 | ||
1425 | if (q->mq_ops) { | |
1426 | req = blk_mq_alloc_request(q, op, | |
6f3b0e8b CH |
1427 | (gfp_mask & __GFP_DIRECT_RECLAIM) ? |
1428 | 0 : BLK_MQ_REQ_NOWAIT); | |
d280bab3 BVA |
1429 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) |
1430 | q->mq_ops->initialize_rq_fn(req); | |
1431 | } else { | |
1432 | req = blk_old_get_request(q, op, gfp_mask); | |
1433 | if (!IS_ERR(req) && q->initialize_rq_fn) | |
1434 | q->initialize_rq_fn(req); | |
1435 | } | |
1436 | ||
1437 | return req; | |
320ae51f | 1438 | } |
1da177e4 LT |
1439 | EXPORT_SYMBOL(blk_get_request); |
1440 | ||
1441 | /** | |
1442 | * blk_requeue_request - put a request back on queue | |
1443 | * @q: request queue where request should be inserted | |
1444 | * @rq: request to be inserted | |
1445 | * | |
1446 | * Description: | |
1447 | * Drivers often keep queueing requests until the hardware cannot accept | |
1448 | * more, when that condition happens we need to put the request back | |
1449 | * on the queue. Must be called with queue lock held. | |
1450 | */ | |
165125e1 | 1451 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1452 | { |
2fff8a92 | 1453 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1454 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1455 | |
242f9dcb JA |
1456 | blk_delete_timer(rq); |
1457 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1458 | trace_block_rq_requeue(q, rq); |
87760e5e | 1459 | wbt_requeue(q->rq_wb, &rq->issue_stat); |
2056a782 | 1460 | |
e8064021 | 1461 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1462 | blk_queue_end_tag(q, rq); |
1463 | ||
ba396a6c JB |
1464 | BUG_ON(blk_queued_rq(rq)); |
1465 | ||
1da177e4 LT |
1466 | elv_requeue_request(q, rq); |
1467 | } | |
1da177e4 LT |
1468 | EXPORT_SYMBOL(blk_requeue_request); |
1469 | ||
73c10101 JA |
1470 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1471 | int where) | |
1472 | { | |
320ae51f | 1473 | blk_account_io_start(rq, true); |
7eaceacc | 1474 | __elv_add_request(q, rq, where); |
73c10101 JA |
1475 | } |
1476 | ||
d62e26b3 | 1477 | static void part_round_stats_single(struct request_queue *q, int cpu, |
b8d62b3a JA |
1478 | struct hd_struct *part, unsigned long now, |
1479 | unsigned int inflight) | |
074a7aca | 1480 | { |
b8d62b3a | 1481 | if (inflight) { |
074a7aca | 1482 | __part_stat_add(cpu, part, time_in_queue, |
b8d62b3a | 1483 | inflight * (now - part->stamp)); |
074a7aca TH |
1484 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1485 | } | |
1486 | part->stamp = now; | |
1487 | } | |
1488 | ||
1489 | /** | |
496aa8a9 | 1490 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
d62e26b3 | 1491 | * @q: target block queue |
496aa8a9 RD |
1492 | * @cpu: cpu number for stats access |
1493 | * @part: target partition | |
1da177e4 LT |
1494 | * |
1495 | * The average IO queue length and utilisation statistics are maintained | |
1496 | * by observing the current state of the queue length and the amount of | |
1497 | * time it has been in this state for. | |
1498 | * | |
1499 | * Normally, that accounting is done on IO completion, but that can result | |
1500 | * in more than a second's worth of IO being accounted for within any one | |
1501 | * second, leading to >100% utilisation. To deal with that, we call this | |
1502 | * function to do a round-off before returning the results when reading | |
1503 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1504 | * the current jiffies and restarts the counters again. | |
1505 | */ | |
d62e26b3 | 1506 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) |
6f2576af | 1507 | { |
b8d62b3a | 1508 | struct hd_struct *part2 = NULL; |
6f2576af | 1509 | unsigned long now = jiffies; |
b8d62b3a JA |
1510 | unsigned int inflight[2]; |
1511 | int stats = 0; | |
1512 | ||
1513 | if (part->stamp != now) | |
1514 | stats |= 1; | |
1515 | ||
1516 | if (part->partno) { | |
1517 | part2 = &part_to_disk(part)->part0; | |
1518 | if (part2->stamp != now) | |
1519 | stats |= 2; | |
1520 | } | |
1521 | ||
1522 | if (!stats) | |
1523 | return; | |
1524 | ||
1525 | part_in_flight(q, part, inflight); | |
6f2576af | 1526 | |
b8d62b3a JA |
1527 | if (stats & 2) |
1528 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1529 | if (stats & 1) | |
1530 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
6f2576af | 1531 | } |
074a7aca | 1532 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1533 | |
47fafbc7 | 1534 | #ifdef CONFIG_PM |
c8158819 LM |
1535 | static void blk_pm_put_request(struct request *rq) |
1536 | { | |
e8064021 | 1537 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1538 | pm_runtime_mark_last_busy(rq->q->dev); |
1539 | } | |
1540 | #else | |
1541 | static inline void blk_pm_put_request(struct request *rq) {} | |
1542 | #endif | |
1543 | ||
165125e1 | 1544 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1545 | { |
e8064021 CH |
1546 | req_flags_t rq_flags = req->rq_flags; |
1547 | ||
1da177e4 LT |
1548 | if (unlikely(!q)) |
1549 | return; | |
1da177e4 | 1550 | |
6f5ba581 CH |
1551 | if (q->mq_ops) { |
1552 | blk_mq_free_request(req); | |
1553 | return; | |
1554 | } | |
1555 | ||
2fff8a92 BVA |
1556 | lockdep_assert_held(q->queue_lock); |
1557 | ||
c8158819 LM |
1558 | blk_pm_put_request(req); |
1559 | ||
8922e16c TH |
1560 | elv_completed_request(q, req); |
1561 | ||
1cd96c24 BH |
1562 | /* this is a bio leak */ |
1563 | WARN_ON(req->bio != NULL); | |
1564 | ||
87760e5e JA |
1565 | wbt_done(q->rq_wb, &req->issue_stat); |
1566 | ||
1da177e4 LT |
1567 | /* |
1568 | * Request may not have originated from ll_rw_blk. if not, | |
1569 | * it didn't come out of our reserved rq pools | |
1570 | */ | |
e8064021 | 1571 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1572 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1573 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1574 | |
1da177e4 | 1575 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1576 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1577 | |
a051661c | 1578 | blk_free_request(rl, req); |
e8064021 | 1579 | freed_request(rl, sync, rq_flags); |
a051661c | 1580 | blk_put_rl(rl); |
1da177e4 LT |
1581 | } |
1582 | } | |
6e39b69e MC |
1583 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1584 | ||
1da177e4 LT |
1585 | void blk_put_request(struct request *req) |
1586 | { | |
165125e1 | 1587 | struct request_queue *q = req->q; |
8922e16c | 1588 | |
320ae51f JA |
1589 | if (q->mq_ops) |
1590 | blk_mq_free_request(req); | |
1591 | else { | |
1592 | unsigned long flags; | |
1593 | ||
1594 | spin_lock_irqsave(q->queue_lock, flags); | |
1595 | __blk_put_request(q, req); | |
1596 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1597 | } | |
1da177e4 | 1598 | } |
1da177e4 LT |
1599 | EXPORT_SYMBOL(blk_put_request); |
1600 | ||
320ae51f JA |
1601 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1602 | struct bio *bio) | |
73c10101 | 1603 | { |
1eff9d32 | 1604 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1605 | |
73c10101 JA |
1606 | if (!ll_back_merge_fn(q, req, bio)) |
1607 | return false; | |
1608 | ||
8c1cf6bb | 1609 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1610 | |
1611 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1612 | blk_rq_set_mixed_merge(req); | |
1613 | ||
1614 | req->biotail->bi_next = bio; | |
1615 | req->biotail = bio; | |
4f024f37 | 1616 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1617 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1618 | ||
320ae51f | 1619 | blk_account_io_start(req, false); |
73c10101 JA |
1620 | return true; |
1621 | } | |
1622 | ||
320ae51f JA |
1623 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1624 | struct bio *bio) | |
73c10101 | 1625 | { |
1eff9d32 | 1626 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1627 | |
73c10101 JA |
1628 | if (!ll_front_merge_fn(q, req, bio)) |
1629 | return false; | |
1630 | ||
8c1cf6bb | 1631 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1632 | |
1633 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1634 | blk_rq_set_mixed_merge(req); | |
1635 | ||
73c10101 JA |
1636 | bio->bi_next = req->bio; |
1637 | req->bio = bio; | |
1638 | ||
4f024f37 KO |
1639 | req->__sector = bio->bi_iter.bi_sector; |
1640 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1641 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1642 | ||
320ae51f | 1643 | blk_account_io_start(req, false); |
73c10101 JA |
1644 | return true; |
1645 | } | |
1646 | ||
1e739730 CH |
1647 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
1648 | struct bio *bio) | |
1649 | { | |
1650 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1651 | ||
1652 | if (segments >= queue_max_discard_segments(q)) | |
1653 | goto no_merge; | |
1654 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1655 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1656 | goto no_merge; | |
1657 | ||
1658 | req->biotail->bi_next = bio; | |
1659 | req->biotail = bio; | |
1660 | req->__data_len += bio->bi_iter.bi_size; | |
1661 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1662 | req->nr_phys_segments = segments + 1; | |
1663 | ||
1664 | blk_account_io_start(req, false); | |
1665 | return true; | |
1666 | no_merge: | |
1667 | req_set_nomerge(q, req); | |
1668 | return false; | |
1669 | } | |
1670 | ||
bd87b589 | 1671 | /** |
320ae51f | 1672 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1673 | * @q: request_queue new bio is being queued at |
1674 | * @bio: new bio being queued | |
1675 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1676 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1677 | * another request associated with @q is found on the plug list | |
1678 | * (optional, may be %NULL) | |
bd87b589 TH |
1679 | * |
1680 | * Determine whether @bio being queued on @q can be merged with a request | |
1681 | * on %current's plugged list. Returns %true if merge was successful, | |
1682 | * otherwise %false. | |
1683 | * | |
07c2bd37 TH |
1684 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1685 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1686 | * than scheduling, and the request, while may have elvpriv data, is not | |
1687 | * added on the elevator at this point. In addition, we don't have | |
1688 | * reliable access to the elevator outside queue lock. Only check basic | |
1689 | * merging parameters without querying the elevator. | |
da41a589 RE |
1690 | * |
1691 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1692 | */ |
320ae51f | 1693 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1694 | unsigned int *request_count, |
1695 | struct request **same_queue_rq) | |
73c10101 JA |
1696 | { |
1697 | struct blk_plug *plug; | |
1698 | struct request *rq; | |
92f399c7 | 1699 | struct list_head *plug_list; |
73c10101 | 1700 | |
bd87b589 | 1701 | plug = current->plug; |
73c10101 | 1702 | if (!plug) |
34fe7c05 | 1703 | return false; |
56ebdaf2 | 1704 | *request_count = 0; |
73c10101 | 1705 | |
92f399c7 SL |
1706 | if (q->mq_ops) |
1707 | plug_list = &plug->mq_list; | |
1708 | else | |
1709 | plug_list = &plug->list; | |
1710 | ||
1711 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 1712 | bool merged = false; |
73c10101 | 1713 | |
5b3f341f | 1714 | if (rq->q == q) { |
1b2e19f1 | 1715 | (*request_count)++; |
5b3f341f SL |
1716 | /* |
1717 | * Only blk-mq multiple hardware queues case checks the | |
1718 | * rq in the same queue, there should be only one such | |
1719 | * rq in a queue | |
1720 | **/ | |
1721 | if (same_queue_rq) | |
1722 | *same_queue_rq = rq; | |
1723 | } | |
56ebdaf2 | 1724 | |
07c2bd37 | 1725 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1726 | continue; |
1727 | ||
34fe7c05 CH |
1728 | switch (blk_try_merge(rq, bio)) { |
1729 | case ELEVATOR_BACK_MERGE: | |
1730 | merged = bio_attempt_back_merge(q, rq, bio); | |
1731 | break; | |
1732 | case ELEVATOR_FRONT_MERGE: | |
1733 | merged = bio_attempt_front_merge(q, rq, bio); | |
1734 | break; | |
1e739730 CH |
1735 | case ELEVATOR_DISCARD_MERGE: |
1736 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1737 | break; | |
34fe7c05 CH |
1738 | default: |
1739 | break; | |
73c10101 | 1740 | } |
34fe7c05 CH |
1741 | |
1742 | if (merged) | |
1743 | return true; | |
73c10101 | 1744 | } |
34fe7c05 CH |
1745 | |
1746 | return false; | |
73c10101 JA |
1747 | } |
1748 | ||
0809e3ac JM |
1749 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1750 | { | |
1751 | struct blk_plug *plug; | |
1752 | struct request *rq; | |
1753 | struct list_head *plug_list; | |
1754 | unsigned int ret = 0; | |
1755 | ||
1756 | plug = current->plug; | |
1757 | if (!plug) | |
1758 | goto out; | |
1759 | ||
1760 | if (q->mq_ops) | |
1761 | plug_list = &plug->mq_list; | |
1762 | else | |
1763 | plug_list = &plug->list; | |
1764 | ||
1765 | list_for_each_entry(rq, plug_list, queuelist) { | |
1766 | if (rq->q == q) | |
1767 | ret++; | |
1768 | } | |
1769 | out: | |
1770 | return ret; | |
1771 | } | |
1772 | ||
da8d7f07 | 1773 | void blk_init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1774 | { |
0be0dee6 BVA |
1775 | struct io_context *ioc = rq_ioc(bio); |
1776 | ||
1eff9d32 | 1777 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1778 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1779 | |
4f024f37 | 1780 | req->__sector = bio->bi_iter.bi_sector; |
5dc8b362 AM |
1781 | if (ioprio_valid(bio_prio(bio))) |
1782 | req->ioprio = bio_prio(bio); | |
0be0dee6 BVA |
1783 | else if (ioc) |
1784 | req->ioprio = ioc->ioprio; | |
1785 | else | |
1786 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
cb6934f8 | 1787 | req->write_hint = bio->bi_write_hint; |
bc1c56fd | 1788 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 | 1789 | } |
da8d7f07 | 1790 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); |
52d9e675 | 1791 | |
dece1635 | 1792 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1793 | { |
73c10101 | 1794 | struct blk_plug *plug; |
34fe7c05 | 1795 | int where = ELEVATOR_INSERT_SORT; |
e4d750c9 | 1796 | struct request *req, *free; |
56ebdaf2 | 1797 | unsigned int request_count = 0; |
87760e5e | 1798 | unsigned int wb_acct; |
1da177e4 | 1799 | |
1da177e4 LT |
1800 | /* |
1801 | * low level driver can indicate that it wants pages above a | |
1802 | * certain limit bounced to low memory (ie for highmem, or even | |
1803 | * ISA dma in theory) | |
1804 | */ | |
1805 | blk_queue_bounce(q, &bio); | |
1806 | ||
af67c31f | 1807 | blk_queue_split(q, &bio); |
23688bf4 | 1808 | |
e23947bd | 1809 | if (!bio_integrity_prep(bio)) |
dece1635 | 1810 | return BLK_QC_T_NONE; |
ffecfd1a | 1811 | |
f73f44eb | 1812 | if (op_is_flush(bio->bi_opf)) { |
73c10101 | 1813 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1814 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1815 | goto get_rq; |
1816 | } | |
1817 | ||
73c10101 JA |
1818 | /* |
1819 | * Check if we can merge with the plugged list before grabbing | |
1820 | * any locks. | |
1821 | */ | |
0809e3ac JM |
1822 | if (!blk_queue_nomerges(q)) { |
1823 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 1824 | return BLK_QC_T_NONE; |
0809e3ac JM |
1825 | } else |
1826 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 1827 | |
73c10101 | 1828 | spin_lock_irq(q->queue_lock); |
2056a782 | 1829 | |
34fe7c05 CH |
1830 | switch (elv_merge(q, &req, bio)) { |
1831 | case ELEVATOR_BACK_MERGE: | |
1832 | if (!bio_attempt_back_merge(q, req, bio)) | |
1833 | break; | |
1834 | elv_bio_merged(q, req, bio); | |
1835 | free = attempt_back_merge(q, req); | |
1836 | if (free) | |
1837 | __blk_put_request(q, free); | |
1838 | else | |
1839 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1840 | goto out_unlock; | |
1841 | case ELEVATOR_FRONT_MERGE: | |
1842 | if (!bio_attempt_front_merge(q, req, bio)) | |
1843 | break; | |
1844 | elv_bio_merged(q, req, bio); | |
1845 | free = attempt_front_merge(q, req); | |
1846 | if (free) | |
1847 | __blk_put_request(q, free); | |
1848 | else | |
1849 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1850 | goto out_unlock; | |
1851 | default: | |
1852 | break; | |
1da177e4 LT |
1853 | } |
1854 | ||
450991bc | 1855 | get_rq: |
87760e5e JA |
1856 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); |
1857 | ||
1da177e4 | 1858 | /* |
450991bc | 1859 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1860 | * Returns with the queue unlocked. |
450991bc | 1861 | */ |
ef295ecf | 1862 | req = get_request(q, bio->bi_opf, bio, GFP_NOIO); |
a492f075 | 1863 | if (IS_ERR(req)) { |
87760e5e | 1864 | __wbt_done(q->rq_wb, wb_acct); |
4e4cbee9 CH |
1865 | if (PTR_ERR(req) == -ENOMEM) |
1866 | bio->bi_status = BLK_STS_RESOURCE; | |
1867 | else | |
1868 | bio->bi_status = BLK_STS_IOERR; | |
4246a0b6 | 1869 | bio_endio(bio); |
da8303c6 TH |
1870 | goto out_unlock; |
1871 | } | |
d6344532 | 1872 | |
87760e5e JA |
1873 | wbt_track(&req->issue_stat, wb_acct); |
1874 | ||
450991bc NP |
1875 | /* |
1876 | * After dropping the lock and possibly sleeping here, our request | |
1877 | * may now be mergeable after it had proven unmergeable (above). | |
1878 | * We don't worry about that case for efficiency. It won't happen | |
1879 | * often, and the elevators are able to handle it. | |
1da177e4 | 1880 | */ |
da8d7f07 | 1881 | blk_init_request_from_bio(req, bio); |
1da177e4 | 1882 | |
9562ad9a | 1883 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1884 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1885 | |
1886 | plug = current->plug; | |
721a9602 | 1887 | if (plug) { |
dc6d36c9 JA |
1888 | /* |
1889 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1890 | * of a plug trace. |
0a6219a9 ML |
1891 | * |
1892 | * @request_count may become stale because of schedule | |
1893 | * out, so check plug list again. | |
dc6d36c9 | 1894 | */ |
0a6219a9 | 1895 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 1896 | trace_block_plug(q); |
3540d5e8 | 1897 | else { |
50d24c34 SL |
1898 | struct request *last = list_entry_rq(plug->list.prev); |
1899 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1900 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 1901 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1902 | trace_block_plug(q); |
1903 | } | |
73c10101 | 1904 | } |
73c10101 | 1905 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 1906 | blk_account_io_start(req, true); |
73c10101 JA |
1907 | } else { |
1908 | spin_lock_irq(q->queue_lock); | |
1909 | add_acct_request(q, req, where); | |
24ecfbe2 | 1910 | __blk_run_queue(q); |
73c10101 JA |
1911 | out_unlock: |
1912 | spin_unlock_irq(q->queue_lock); | |
1913 | } | |
dece1635 JA |
1914 | |
1915 | return BLK_QC_T_NONE; | |
1da177e4 LT |
1916 | } |
1917 | ||
1da177e4 LT |
1918 | static void handle_bad_sector(struct bio *bio) |
1919 | { | |
1920 | char b[BDEVNAME_SIZE]; | |
1921 | ||
1922 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 1923 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 1924 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 1925 | (unsigned long long)bio_end_sector(bio), |
74d46992 | 1926 | (long long)get_capacity(bio->bi_disk)); |
1da177e4 LT |
1927 | } |
1928 | ||
c17bb495 AM |
1929 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1930 | ||
1931 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1932 | ||
1933 | static int __init setup_fail_make_request(char *str) | |
1934 | { | |
1935 | return setup_fault_attr(&fail_make_request, str); | |
1936 | } | |
1937 | __setup("fail_make_request=", setup_fail_make_request); | |
1938 | ||
b2c9cd37 | 1939 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1940 | { |
b2c9cd37 | 1941 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1942 | } |
1943 | ||
1944 | static int __init fail_make_request_debugfs(void) | |
1945 | { | |
dd48c085 AM |
1946 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1947 | NULL, &fail_make_request); | |
1948 | ||
21f9fcd8 | 1949 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
1950 | } |
1951 | ||
1952 | late_initcall(fail_make_request_debugfs); | |
1953 | ||
1954 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1955 | ||
b2c9cd37 AM |
1956 | static inline bool should_fail_request(struct hd_struct *part, |
1957 | unsigned int bytes) | |
c17bb495 | 1958 | { |
b2c9cd37 | 1959 | return false; |
c17bb495 AM |
1960 | } |
1961 | ||
1962 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1963 | ||
74d46992 CH |
1964 | /* |
1965 | * Remap block n of partition p to block n+start(p) of the disk. | |
1966 | */ | |
1967 | static inline int blk_partition_remap(struct bio *bio) | |
1968 | { | |
1969 | struct hd_struct *p; | |
1970 | int ret = 0; | |
1971 | ||
1972 | /* | |
1973 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
1974 | * Include a test for the reset op code and perform the remap if needed. | |
1975 | */ | |
1976 | if (!bio->bi_partno || | |
1977 | (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET)) | |
1978 | return 0; | |
1979 | ||
1980 | rcu_read_lock(); | |
1981 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
1982 | if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) { | |
1983 | bio->bi_iter.bi_sector += p->start_sect; | |
1984 | bio->bi_partno = 0; | |
1985 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), | |
1986 | bio->bi_iter.bi_sector - p->start_sect); | |
1987 | } else { | |
1988 | printk("%s: fail for partition %d\n", __func__, bio->bi_partno); | |
1989 | ret = -EIO; | |
1990 | } | |
1991 | rcu_read_unlock(); | |
1992 | ||
1993 | return ret; | |
1994 | } | |
1995 | ||
c07e2b41 JA |
1996 | /* |
1997 | * Check whether this bio extends beyond the end of the device. | |
1998 | */ | |
1999 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
2000 | { | |
2001 | sector_t maxsector; | |
2002 | ||
2003 | if (!nr_sectors) | |
2004 | return 0; | |
2005 | ||
2006 | /* Test device or partition size, when known. */ | |
74d46992 | 2007 | maxsector = get_capacity(bio->bi_disk); |
c07e2b41 | 2008 | if (maxsector) { |
4f024f37 | 2009 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
2010 | |
2011 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2012 | /* | |
2013 | * This may well happen - the kernel calls bread() | |
2014 | * without checking the size of the device, e.g., when | |
2015 | * mounting a device. | |
2016 | */ | |
2017 | handle_bad_sector(bio); | |
2018 | return 1; | |
2019 | } | |
2020 | } | |
2021 | ||
2022 | return 0; | |
2023 | } | |
2024 | ||
27a84d54 CH |
2025 | static noinline_for_stack bool |
2026 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 2027 | { |
165125e1 | 2028 | struct request_queue *q; |
5a7bbad2 | 2029 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 2030 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 2031 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
2032 | |
2033 | might_sleep(); | |
1da177e4 | 2034 | |
c07e2b41 JA |
2035 | if (bio_check_eod(bio, nr_sectors)) |
2036 | goto end_io; | |
1da177e4 | 2037 | |
74d46992 | 2038 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
2039 | if (unlikely(!q)) { |
2040 | printk(KERN_ERR | |
2041 | "generic_make_request: Trying to access " | |
2042 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 2043 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
2044 | goto end_io; |
2045 | } | |
c17bb495 | 2046 | |
03a07c92 GR |
2047 | /* |
2048 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2049 | * if queue is not a request based queue. | |
2050 | */ | |
2051 | ||
2052 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) | |
2053 | goto not_supported; | |
2054 | ||
74d46992 | 2055 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) |
5a7bbad2 | 2056 | goto end_io; |
2056a782 | 2057 | |
74d46992 CH |
2058 | if (blk_partition_remap(bio)) |
2059 | goto end_io; | |
2056a782 | 2060 | |
5a7bbad2 CH |
2061 | if (bio_check_eod(bio, nr_sectors)) |
2062 | goto end_io; | |
1e87901e | 2063 | |
5a7bbad2 CH |
2064 | /* |
2065 | * Filter flush bio's early so that make_request based | |
2066 | * drivers without flush support don't have to worry | |
2067 | * about them. | |
2068 | */ | |
f3a8ab7d | 2069 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 2070 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 2071 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 2072 | if (!nr_sectors) { |
4e4cbee9 | 2073 | status = BLK_STS_OK; |
51fd77bd JA |
2074 | goto end_io; |
2075 | } | |
5a7bbad2 | 2076 | } |
5ddfe969 | 2077 | |
288dab8a CH |
2078 | switch (bio_op(bio)) { |
2079 | case REQ_OP_DISCARD: | |
2080 | if (!blk_queue_discard(q)) | |
2081 | goto not_supported; | |
2082 | break; | |
2083 | case REQ_OP_SECURE_ERASE: | |
2084 | if (!blk_queue_secure_erase(q)) | |
2085 | goto not_supported; | |
2086 | break; | |
2087 | case REQ_OP_WRITE_SAME: | |
74d46992 | 2088 | if (!q->limits.max_write_same_sectors) |
288dab8a | 2089 | goto not_supported; |
58886785 | 2090 | break; |
2d253440 ST |
2091 | case REQ_OP_ZONE_REPORT: |
2092 | case REQ_OP_ZONE_RESET: | |
74d46992 | 2093 | if (!blk_queue_is_zoned(q)) |
2d253440 | 2094 | goto not_supported; |
288dab8a | 2095 | break; |
a6f0788e | 2096 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 2097 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
2098 | goto not_supported; |
2099 | break; | |
288dab8a CH |
2100 | default: |
2101 | break; | |
5a7bbad2 | 2102 | } |
01edede4 | 2103 | |
7f4b35d1 TH |
2104 | /* |
2105 | * Various block parts want %current->io_context and lazy ioc | |
2106 | * allocation ends up trading a lot of pain for a small amount of | |
2107 | * memory. Just allocate it upfront. This may fail and block | |
2108 | * layer knows how to live with it. | |
2109 | */ | |
2110 | create_io_context(GFP_ATOMIC, q->node); | |
2111 | ||
ae118896 TH |
2112 | if (!blkcg_bio_issue_check(q, bio)) |
2113 | return false; | |
27a84d54 | 2114 | |
fbbaf700 N |
2115 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
2116 | trace_block_bio_queue(q, bio); | |
2117 | /* Now that enqueuing has been traced, we need to trace | |
2118 | * completion as well. | |
2119 | */ | |
2120 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2121 | } | |
27a84d54 | 2122 | return true; |
a7384677 | 2123 | |
288dab8a | 2124 | not_supported: |
4e4cbee9 | 2125 | status = BLK_STS_NOTSUPP; |
a7384677 | 2126 | end_io: |
4e4cbee9 | 2127 | bio->bi_status = status; |
4246a0b6 | 2128 | bio_endio(bio); |
27a84d54 | 2129 | return false; |
1da177e4 LT |
2130 | } |
2131 | ||
27a84d54 CH |
2132 | /** |
2133 | * generic_make_request - hand a buffer to its device driver for I/O | |
2134 | * @bio: The bio describing the location in memory and on the device. | |
2135 | * | |
2136 | * generic_make_request() is used to make I/O requests of block | |
2137 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2138 | * to be done. | |
2139 | * | |
2140 | * generic_make_request() does not return any status. The | |
2141 | * success/failure status of the request, along with notification of | |
2142 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2143 | * function described (one day) else where. | |
2144 | * | |
2145 | * The caller of generic_make_request must make sure that bi_io_vec | |
2146 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2147 | * set to describe the device address, and the | |
2148 | * bi_end_io and optionally bi_private are set to describe how | |
2149 | * completion notification should be signaled. | |
2150 | * | |
2151 | * generic_make_request and the drivers it calls may use bi_next if this | |
2152 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2153 | * a lower device by calling into generic_make_request recursively, which | |
2154 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 2155 | */ |
dece1635 | 2156 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 2157 | { |
f5fe1b51 N |
2158 | /* |
2159 | * bio_list_on_stack[0] contains bios submitted by the current | |
2160 | * make_request_fn. | |
2161 | * bio_list_on_stack[1] contains bios that were submitted before | |
2162 | * the current make_request_fn, but that haven't been processed | |
2163 | * yet. | |
2164 | */ | |
2165 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 2166 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 2167 | |
27a84d54 | 2168 | if (!generic_make_request_checks(bio)) |
dece1635 | 2169 | goto out; |
27a84d54 CH |
2170 | |
2171 | /* | |
2172 | * We only want one ->make_request_fn to be active at a time, else | |
2173 | * stack usage with stacked devices could be a problem. So use | |
2174 | * current->bio_list to keep a list of requests submited by a | |
2175 | * make_request_fn function. current->bio_list is also used as a | |
2176 | * flag to say if generic_make_request is currently active in this | |
2177 | * task or not. If it is NULL, then no make_request is active. If | |
2178 | * it is non-NULL, then a make_request is active, and new requests | |
2179 | * should be added at the tail | |
2180 | */ | |
bddd87c7 | 2181 | if (current->bio_list) { |
f5fe1b51 | 2182 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 2183 | goto out; |
d89d8796 | 2184 | } |
27a84d54 | 2185 | |
d89d8796 NB |
2186 | /* following loop may be a bit non-obvious, and so deserves some |
2187 | * explanation. | |
2188 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2189 | * ensure that) so we have a list with a single bio. | |
2190 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2191 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2192 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2193 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2194 | * through a recursive call to generic_make_request. If it |
2195 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2196 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2197 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2198 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2199 | */ |
2200 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
2201 | bio_list_init(&bio_list_on_stack[0]); |
2202 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 2203 | do { |
74d46992 | 2204 | struct request_queue *q = bio->bi_disk->queue; |
27a84d54 | 2205 | |
03a07c92 | 2206 | if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) { |
79bd9959 N |
2207 | struct bio_list lower, same; |
2208 | ||
2209 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
2210 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
2211 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 2212 | ret = q->make_request_fn(q, bio); |
3ef28e83 DW |
2213 | |
2214 | blk_queue_exit(q); | |
27a84d54 | 2215 | |
79bd9959 N |
2216 | /* sort new bios into those for a lower level |
2217 | * and those for the same level | |
2218 | */ | |
2219 | bio_list_init(&lower); | |
2220 | bio_list_init(&same); | |
f5fe1b51 | 2221 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 2222 | if (q == bio->bi_disk->queue) |
79bd9959 N |
2223 | bio_list_add(&same, bio); |
2224 | else | |
2225 | bio_list_add(&lower, bio); | |
2226 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
2227 | bio_list_merge(&bio_list_on_stack[0], &lower); |
2228 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2229 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 2230 | } else { |
03a07c92 GR |
2231 | if (unlikely(!blk_queue_dying(q) && |
2232 | (bio->bi_opf & REQ_NOWAIT))) | |
2233 | bio_wouldblock_error(bio); | |
2234 | else | |
2235 | bio_io_error(bio); | |
3ef28e83 | 2236 | } |
f5fe1b51 | 2237 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 2238 | } while (bio); |
bddd87c7 | 2239 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2240 | |
2241 | out: | |
2242 | return ret; | |
d89d8796 | 2243 | } |
1da177e4 LT |
2244 | EXPORT_SYMBOL(generic_make_request); |
2245 | ||
2246 | /** | |
710027a4 | 2247 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2248 | * @bio: The &struct bio which describes the I/O |
2249 | * | |
2250 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2251 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2252 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2253 | * |
2254 | */ | |
4e49ea4a | 2255 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2256 | { |
bf2de6f5 JA |
2257 | /* |
2258 | * If it's a regular read/write or a barrier with data attached, | |
2259 | * go through the normal accounting stuff before submission. | |
2260 | */ | |
e2a60da7 | 2261 | if (bio_has_data(bio)) { |
4363ac7c MP |
2262 | unsigned int count; |
2263 | ||
95fe6c1a | 2264 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
74d46992 | 2265 | count = queue_logical_block_size(bio->bi_disk->queue); |
4363ac7c MP |
2266 | else |
2267 | count = bio_sectors(bio); | |
2268 | ||
a8ebb056 | 2269 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2270 | count_vm_events(PGPGOUT, count); |
2271 | } else { | |
4f024f37 | 2272 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2273 | count_vm_events(PGPGIN, count); |
2274 | } | |
2275 | ||
2276 | if (unlikely(block_dump)) { | |
2277 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2278 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2279 | current->comm, task_pid_nr(current), |
a8ebb056 | 2280 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2281 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 2282 | bio_devname(bio, b), count); |
bf2de6f5 | 2283 | } |
1da177e4 LT |
2284 | } |
2285 | ||
dece1635 | 2286 | return generic_make_request(bio); |
1da177e4 | 2287 | } |
1da177e4 LT |
2288 | EXPORT_SYMBOL(submit_bio); |
2289 | ||
82124d60 | 2290 | /** |
bf4e6b4e HR |
2291 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2292 | * for new the queue limits | |
82124d60 KU |
2293 | * @q: the queue |
2294 | * @rq: the request being checked | |
2295 | * | |
2296 | * Description: | |
2297 | * @rq may have been made based on weaker limitations of upper-level queues | |
2298 | * in request stacking drivers, and it may violate the limitation of @q. | |
2299 | * Since the block layer and the underlying device driver trust @rq | |
2300 | * after it is inserted to @q, it should be checked against @q before | |
2301 | * the insertion using this generic function. | |
2302 | * | |
82124d60 | 2303 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2304 | * limits when retrying requests on other queues. Those requests need |
2305 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2306 | */ |
bf4e6b4e HR |
2307 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2308 | struct request *rq) | |
82124d60 | 2309 | { |
8fe0d473 | 2310 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2311 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2312 | return -EIO; | |
2313 | } | |
2314 | ||
2315 | /* | |
2316 | * queue's settings related to segment counting like q->bounce_pfn | |
2317 | * may differ from that of other stacking queues. | |
2318 | * Recalculate it to check the request correctly on this queue's | |
2319 | * limitation. | |
2320 | */ | |
2321 | blk_recalc_rq_segments(rq); | |
8a78362c | 2322 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2323 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2324 | return -EIO; | |
2325 | } | |
2326 | ||
2327 | return 0; | |
2328 | } | |
82124d60 KU |
2329 | |
2330 | /** | |
2331 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2332 | * @q: the queue to submit the request | |
2333 | * @rq: the request being queued | |
2334 | */ | |
2a842aca | 2335 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 KU |
2336 | { |
2337 | unsigned long flags; | |
4853abaa | 2338 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2339 | |
bf4e6b4e | 2340 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 2341 | return BLK_STS_IOERR; |
82124d60 | 2342 | |
b2c9cd37 AM |
2343 | if (rq->rq_disk && |
2344 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 2345 | return BLK_STS_IOERR; |
82124d60 | 2346 | |
7fb4898e KB |
2347 | if (q->mq_ops) { |
2348 | if (blk_queue_io_stat(q)) | |
2349 | blk_account_io_start(rq, true); | |
157f377b JA |
2350 | /* |
2351 | * Since we have a scheduler attached on the top device, | |
2352 | * bypass a potential scheduler on the bottom device for | |
2353 | * insert. | |
2354 | */ | |
2355 | blk_mq_request_bypass_insert(rq); | |
2a842aca | 2356 | return BLK_STS_OK; |
7fb4898e KB |
2357 | } |
2358 | ||
82124d60 | 2359 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2360 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 | 2361 | spin_unlock_irqrestore(q->queue_lock, flags); |
2a842aca | 2362 | return BLK_STS_IOERR; |
8ba61435 | 2363 | } |
82124d60 KU |
2364 | |
2365 | /* | |
2366 | * Submitting request must be dequeued before calling this function | |
2367 | * because it will be linked to another request_queue | |
2368 | */ | |
2369 | BUG_ON(blk_queued_rq(rq)); | |
2370 | ||
f73f44eb | 2371 | if (op_is_flush(rq->cmd_flags)) |
4853abaa JM |
2372 | where = ELEVATOR_INSERT_FLUSH; |
2373 | ||
2374 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2375 | if (where == ELEVATOR_INSERT_FLUSH) |
2376 | __blk_run_queue(q); | |
82124d60 KU |
2377 | spin_unlock_irqrestore(q->queue_lock, flags); |
2378 | ||
2a842aca | 2379 | return BLK_STS_OK; |
82124d60 KU |
2380 | } |
2381 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2382 | ||
80a761fd TH |
2383 | /** |
2384 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2385 | * @rq: request to examine | |
2386 | * | |
2387 | * Description: | |
2388 | * A request could be merge of IOs which require different failure | |
2389 | * handling. This function determines the number of bytes which | |
2390 | * can be failed from the beginning of the request without | |
2391 | * crossing into area which need to be retried further. | |
2392 | * | |
2393 | * Return: | |
2394 | * The number of bytes to fail. | |
80a761fd TH |
2395 | */ |
2396 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2397 | { | |
2398 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2399 | unsigned int bytes = 0; | |
2400 | struct bio *bio; | |
2401 | ||
e8064021 | 2402 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2403 | return blk_rq_bytes(rq); |
2404 | ||
2405 | /* | |
2406 | * Currently the only 'mixing' which can happen is between | |
2407 | * different fastfail types. We can safely fail portions | |
2408 | * which have all the failfast bits that the first one has - | |
2409 | * the ones which are at least as eager to fail as the first | |
2410 | * one. | |
2411 | */ | |
2412 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2413 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2414 | break; |
4f024f37 | 2415 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2416 | } |
2417 | ||
2418 | /* this could lead to infinite loop */ | |
2419 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2420 | return bytes; | |
2421 | } | |
2422 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2423 | ||
320ae51f | 2424 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2425 | { |
c2553b58 | 2426 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2427 | const int rw = rq_data_dir(req); |
2428 | struct hd_struct *part; | |
2429 | int cpu; | |
2430 | ||
2431 | cpu = part_stat_lock(); | |
09e099d4 | 2432 | part = req->part; |
bc58ba94 JA |
2433 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2434 | part_stat_unlock(); | |
2435 | } | |
2436 | } | |
2437 | ||
320ae51f | 2438 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2439 | { |
bc58ba94 | 2440 | /* |
dd4c133f TH |
2441 | * Account IO completion. flush_rq isn't accounted as a |
2442 | * normal IO on queueing nor completion. Accounting the | |
2443 | * containing request is enough. | |
bc58ba94 | 2444 | */ |
e8064021 | 2445 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
bc58ba94 JA |
2446 | unsigned long duration = jiffies - req->start_time; |
2447 | const int rw = rq_data_dir(req); | |
2448 | struct hd_struct *part; | |
2449 | int cpu; | |
2450 | ||
2451 | cpu = part_stat_lock(); | |
09e099d4 | 2452 | part = req->part; |
bc58ba94 JA |
2453 | |
2454 | part_stat_inc(cpu, part, ios[rw]); | |
2455 | part_stat_add(cpu, part, ticks[rw], duration); | |
d62e26b3 JA |
2456 | part_round_stats(req->q, cpu, part); |
2457 | part_dec_in_flight(req->q, part, rw); | |
bc58ba94 | 2458 | |
6c23a968 | 2459 | hd_struct_put(part); |
bc58ba94 JA |
2460 | part_stat_unlock(); |
2461 | } | |
2462 | } | |
2463 | ||
47fafbc7 | 2464 | #ifdef CONFIG_PM |
c8158819 LM |
2465 | /* |
2466 | * Don't process normal requests when queue is suspended | |
2467 | * or in the process of suspending/resuming | |
2468 | */ | |
2469 | static struct request *blk_pm_peek_request(struct request_queue *q, | |
2470 | struct request *rq) | |
2471 | { | |
2472 | if (q->dev && (q->rpm_status == RPM_SUSPENDED || | |
e8064021 | 2473 | (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM)))) |
c8158819 LM |
2474 | return NULL; |
2475 | else | |
2476 | return rq; | |
2477 | } | |
2478 | #else | |
2479 | static inline struct request *blk_pm_peek_request(struct request_queue *q, | |
2480 | struct request *rq) | |
2481 | { | |
2482 | return rq; | |
2483 | } | |
2484 | #endif | |
2485 | ||
320ae51f JA |
2486 | void blk_account_io_start(struct request *rq, bool new_io) |
2487 | { | |
2488 | struct hd_struct *part; | |
2489 | int rw = rq_data_dir(rq); | |
2490 | int cpu; | |
2491 | ||
2492 | if (!blk_do_io_stat(rq)) | |
2493 | return; | |
2494 | ||
2495 | cpu = part_stat_lock(); | |
2496 | ||
2497 | if (!new_io) { | |
2498 | part = rq->part; | |
2499 | part_stat_inc(cpu, part, merges[rw]); | |
2500 | } else { | |
2501 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2502 | if (!hd_struct_try_get(part)) { | |
2503 | /* | |
2504 | * The partition is already being removed, | |
2505 | * the request will be accounted on the disk only | |
2506 | * | |
2507 | * We take a reference on disk->part0 although that | |
2508 | * partition will never be deleted, so we can treat | |
2509 | * it as any other partition. | |
2510 | */ | |
2511 | part = &rq->rq_disk->part0; | |
2512 | hd_struct_get(part); | |
2513 | } | |
d62e26b3 JA |
2514 | part_round_stats(rq->q, cpu, part); |
2515 | part_inc_in_flight(rq->q, part, rw); | |
320ae51f JA |
2516 | rq->part = part; |
2517 | } | |
2518 | ||
2519 | part_stat_unlock(); | |
2520 | } | |
2521 | ||
3bcddeac | 2522 | /** |
9934c8c0 TH |
2523 | * blk_peek_request - peek at the top of a request queue |
2524 | * @q: request queue to peek at | |
2525 | * | |
2526 | * Description: | |
2527 | * Return the request at the top of @q. The returned request | |
2528 | * should be started using blk_start_request() before LLD starts | |
2529 | * processing it. | |
2530 | * | |
2531 | * Return: | |
2532 | * Pointer to the request at the top of @q if available. Null | |
2533 | * otherwise. | |
9934c8c0 TH |
2534 | */ |
2535 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2536 | { |
2537 | struct request *rq; | |
2538 | int ret; | |
2539 | ||
2fff8a92 | 2540 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2541 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2542 | |
158dbda0 | 2543 | while ((rq = __elv_next_request(q)) != NULL) { |
c8158819 LM |
2544 | |
2545 | rq = blk_pm_peek_request(q, rq); | |
2546 | if (!rq) | |
2547 | break; | |
2548 | ||
e8064021 | 2549 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2550 | /* |
2551 | * This is the first time the device driver | |
2552 | * sees this request (possibly after | |
2553 | * requeueing). Notify IO scheduler. | |
2554 | */ | |
e8064021 | 2555 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2556 | elv_activate_rq(q, rq); |
2557 | ||
2558 | /* | |
2559 | * just mark as started even if we don't start | |
2560 | * it, a request that has been delayed should | |
2561 | * not be passed by new incoming requests | |
2562 | */ | |
e8064021 | 2563 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2564 | trace_block_rq_issue(q, rq); |
2565 | } | |
2566 | ||
2567 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2568 | q->end_sector = rq_end_sector(rq); | |
2569 | q->boundary_rq = NULL; | |
2570 | } | |
2571 | ||
e8064021 | 2572 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2573 | break; |
2574 | ||
2e46e8b2 | 2575 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2576 | /* |
2577 | * make sure space for the drain appears we | |
2578 | * know we can do this because max_hw_segments | |
2579 | * has been adjusted to be one fewer than the | |
2580 | * device can handle | |
2581 | */ | |
2582 | rq->nr_phys_segments++; | |
2583 | } | |
2584 | ||
2585 | if (!q->prep_rq_fn) | |
2586 | break; | |
2587 | ||
2588 | ret = q->prep_rq_fn(q, rq); | |
2589 | if (ret == BLKPREP_OK) { | |
2590 | break; | |
2591 | } else if (ret == BLKPREP_DEFER) { | |
2592 | /* | |
2593 | * the request may have been (partially) prepped. | |
2594 | * we need to keep this request in the front to | |
e8064021 | 2595 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2596 | * prevent other fs requests from passing this one. |
2597 | */ | |
2e46e8b2 | 2598 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2599 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2600 | /* |
2601 | * remove the space for the drain we added | |
2602 | * so that we don't add it again | |
2603 | */ | |
2604 | --rq->nr_phys_segments; | |
2605 | } | |
2606 | ||
2607 | rq = NULL; | |
2608 | break; | |
0fb5b1fb | 2609 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
e8064021 | 2610 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2611 | /* |
2612 | * Mark this request as started so we don't trigger | |
2613 | * any debug logic in the end I/O path. | |
2614 | */ | |
2615 | blk_start_request(rq); | |
2a842aca CH |
2616 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? |
2617 | BLK_STS_TARGET : BLK_STS_IOERR); | |
158dbda0 TH |
2618 | } else { |
2619 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2620 | break; | |
2621 | } | |
2622 | } | |
2623 | ||
2624 | return rq; | |
2625 | } | |
9934c8c0 | 2626 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2627 | |
5034435c | 2628 | static void blk_dequeue_request(struct request *rq) |
158dbda0 | 2629 | { |
9934c8c0 TH |
2630 | struct request_queue *q = rq->q; |
2631 | ||
158dbda0 TH |
2632 | BUG_ON(list_empty(&rq->queuelist)); |
2633 | BUG_ON(ELV_ON_HASH(rq)); | |
2634 | ||
2635 | list_del_init(&rq->queuelist); | |
2636 | ||
2637 | /* | |
2638 | * the time frame between a request being removed from the lists | |
2639 | * and to it is freed is accounted as io that is in progress at | |
2640 | * the driver side. | |
2641 | */ | |
9195291e | 2642 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2643 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2644 | set_io_start_time_ns(rq); |
2645 | } | |
158dbda0 TH |
2646 | } |
2647 | ||
9934c8c0 TH |
2648 | /** |
2649 | * blk_start_request - start request processing on the driver | |
2650 | * @req: request to dequeue | |
2651 | * | |
2652 | * Description: | |
2653 | * Dequeue @req and start timeout timer on it. This hands off the | |
2654 | * request to the driver. | |
9934c8c0 TH |
2655 | */ |
2656 | void blk_start_request(struct request *req) | |
2657 | { | |
2fff8a92 | 2658 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2659 | WARN_ON_ONCE(req->q->mq_ops); |
2fff8a92 | 2660 | |
9934c8c0 TH |
2661 | blk_dequeue_request(req); |
2662 | ||
cf43e6be | 2663 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
88eeca49 | 2664 | blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req)); |
cf43e6be | 2665 | req->rq_flags |= RQF_STATS; |
87760e5e | 2666 | wbt_issue(req->q->rq_wb, &req->issue_stat); |
cf43e6be JA |
2667 | } |
2668 | ||
4912aa6c | 2669 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); |
9934c8c0 TH |
2670 | blk_add_timer(req); |
2671 | } | |
2672 | EXPORT_SYMBOL(blk_start_request); | |
2673 | ||
2674 | /** | |
2675 | * blk_fetch_request - fetch a request from a request queue | |
2676 | * @q: request queue to fetch a request from | |
2677 | * | |
2678 | * Description: | |
2679 | * Return the request at the top of @q. The request is started on | |
2680 | * return and LLD can start processing it immediately. | |
2681 | * | |
2682 | * Return: | |
2683 | * Pointer to the request at the top of @q if available. Null | |
2684 | * otherwise. | |
9934c8c0 TH |
2685 | */ |
2686 | struct request *blk_fetch_request(struct request_queue *q) | |
2687 | { | |
2688 | struct request *rq; | |
2689 | ||
2fff8a92 | 2690 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2691 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2692 | |
9934c8c0 TH |
2693 | rq = blk_peek_request(q); |
2694 | if (rq) | |
2695 | blk_start_request(rq); | |
2696 | return rq; | |
2697 | } | |
2698 | EXPORT_SYMBOL(blk_fetch_request); | |
2699 | ||
3bcddeac | 2700 | /** |
2e60e022 | 2701 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2702 | * @req: the request being processed |
2a842aca | 2703 | * @error: block status code |
8ebf9756 | 2704 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2705 | * |
2706 | * Description: | |
8ebf9756 RD |
2707 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2708 | * the request structure even if @req doesn't have leftover. | |
2709 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2710 | * |
2711 | * This special helper function is only for request stacking drivers | |
2712 | * (e.g. request-based dm) so that they can handle partial completion. | |
2713 | * Actual device drivers should use blk_end_request instead. | |
2714 | * | |
2715 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2716 | * %false return from this function. | |
3bcddeac KU |
2717 | * |
2718 | * Return: | |
2e60e022 TH |
2719 | * %false - this request doesn't have any more data |
2720 | * %true - this request has more data | |
3bcddeac | 2721 | **/ |
2a842aca CH |
2722 | bool blk_update_request(struct request *req, blk_status_t error, |
2723 | unsigned int nr_bytes) | |
1da177e4 | 2724 | { |
f79ea416 | 2725 | int total_bytes; |
1da177e4 | 2726 | |
2a842aca | 2727 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 2728 | |
2e60e022 TH |
2729 | if (!req->bio) |
2730 | return false; | |
2731 | ||
2a842aca CH |
2732 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
2733 | !(req->rq_flags & RQF_QUIET))) | |
2734 | print_req_error(req, error); | |
1da177e4 | 2735 | |
bc58ba94 | 2736 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2737 | |
f79ea416 KO |
2738 | total_bytes = 0; |
2739 | while (req->bio) { | |
2740 | struct bio *bio = req->bio; | |
4f024f37 | 2741 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2742 | |
4f024f37 | 2743 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2744 | req->bio = bio->bi_next; |
1da177e4 | 2745 | |
fbbaf700 N |
2746 | /* Completion has already been traced */ |
2747 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 2748 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2749 | |
f79ea416 KO |
2750 | total_bytes += bio_bytes; |
2751 | nr_bytes -= bio_bytes; | |
1da177e4 | 2752 | |
f79ea416 KO |
2753 | if (!nr_bytes) |
2754 | break; | |
1da177e4 LT |
2755 | } |
2756 | ||
2757 | /* | |
2758 | * completely done | |
2759 | */ | |
2e60e022 TH |
2760 | if (!req->bio) { |
2761 | /* | |
2762 | * Reset counters so that the request stacking driver | |
2763 | * can find how many bytes remain in the request | |
2764 | * later. | |
2765 | */ | |
a2dec7b3 | 2766 | req->__data_len = 0; |
2e60e022 TH |
2767 | return false; |
2768 | } | |
1da177e4 | 2769 | |
a2dec7b3 | 2770 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2771 | |
2772 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 2773 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 2774 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2775 | |
80a761fd | 2776 | /* mixed attributes always follow the first bio */ |
e8064021 | 2777 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 2778 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 2779 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
2780 | } |
2781 | ||
ed6565e7 CH |
2782 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
2783 | /* | |
2784 | * If total number of sectors is less than the first segment | |
2785 | * size, something has gone terribly wrong. | |
2786 | */ | |
2787 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2788 | blk_dump_rq_flags(req, "request botched"); | |
2789 | req->__data_len = blk_rq_cur_bytes(req); | |
2790 | } | |
2e46e8b2 | 2791 | |
ed6565e7 CH |
2792 | /* recalculate the number of segments */ |
2793 | blk_recalc_rq_segments(req); | |
2794 | } | |
2e46e8b2 | 2795 | |
2e60e022 | 2796 | return true; |
1da177e4 | 2797 | } |
2e60e022 | 2798 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2799 | |
2a842aca | 2800 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, |
2e60e022 TH |
2801 | unsigned int nr_bytes, |
2802 | unsigned int bidi_bytes) | |
5efccd17 | 2803 | { |
2e60e022 TH |
2804 | if (blk_update_request(rq, error, nr_bytes)) |
2805 | return true; | |
5efccd17 | 2806 | |
2e60e022 TH |
2807 | /* Bidi request must be completed as a whole */ |
2808 | if (unlikely(blk_bidi_rq(rq)) && | |
2809 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2810 | return true; | |
5efccd17 | 2811 | |
e2e1a148 JA |
2812 | if (blk_queue_add_random(rq->q)) |
2813 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2814 | |
2815 | return false; | |
1da177e4 LT |
2816 | } |
2817 | ||
28018c24 JB |
2818 | /** |
2819 | * blk_unprep_request - unprepare a request | |
2820 | * @req: the request | |
2821 | * | |
2822 | * This function makes a request ready for complete resubmission (or | |
2823 | * completion). It happens only after all error handling is complete, | |
2824 | * so represents the appropriate moment to deallocate any resources | |
2825 | * that were allocated to the request in the prep_rq_fn. The queue | |
2826 | * lock is held when calling this. | |
2827 | */ | |
2828 | void blk_unprep_request(struct request *req) | |
2829 | { | |
2830 | struct request_queue *q = req->q; | |
2831 | ||
e8064021 | 2832 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
2833 | if (q->unprep_rq_fn) |
2834 | q->unprep_rq_fn(q, req); | |
2835 | } | |
2836 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2837 | ||
2a842aca | 2838 | void blk_finish_request(struct request *req, blk_status_t error) |
1da177e4 | 2839 | { |
cf43e6be JA |
2840 | struct request_queue *q = req->q; |
2841 | ||
2fff8a92 | 2842 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2843 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2844 | |
cf43e6be | 2845 | if (req->rq_flags & RQF_STATS) |
34dbad5d | 2846 | blk_stat_add(req); |
cf43e6be | 2847 | |
e8064021 | 2848 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 2849 | blk_queue_end_tag(q, req); |
b8286239 | 2850 | |
ba396a6c | 2851 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2852 | |
57292b58 | 2853 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) |
dc3b17cc | 2854 | laptop_io_completion(req->q->backing_dev_info); |
1da177e4 | 2855 | |
e78042e5 MA |
2856 | blk_delete_timer(req); |
2857 | ||
e8064021 | 2858 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
2859 | blk_unprep_request(req); |
2860 | ||
bc58ba94 | 2861 | blk_account_io_done(req); |
b8286239 | 2862 | |
87760e5e JA |
2863 | if (req->end_io) { |
2864 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
8ffdc655 | 2865 | req->end_io(req, error); |
87760e5e | 2866 | } else { |
b8286239 KU |
2867 | if (blk_bidi_rq(req)) |
2868 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2869 | ||
cf43e6be | 2870 | __blk_put_request(q, req); |
b8286239 | 2871 | } |
1da177e4 | 2872 | } |
12120077 | 2873 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 2874 | |
3b11313a | 2875 | /** |
2e60e022 TH |
2876 | * blk_end_bidi_request - Complete a bidi request |
2877 | * @rq: the request to complete | |
2a842aca | 2878 | * @error: block status code |
2e60e022 TH |
2879 | * @nr_bytes: number of bytes to complete @rq |
2880 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2881 | * |
2882 | * Description: | |
e3a04fe3 | 2883 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2884 | * Drivers that supports bidi can safely call this member for any |
2885 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2886 | * just ignored. | |
336cdb40 KU |
2887 | * |
2888 | * Return: | |
2e60e022 TH |
2889 | * %false - we are done with this request |
2890 | * %true - still buffers pending for this request | |
a0cd1285 | 2891 | **/ |
2a842aca | 2892 | static bool blk_end_bidi_request(struct request *rq, blk_status_t error, |
32fab448 KU |
2893 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2894 | { | |
336cdb40 | 2895 | struct request_queue *q = rq->q; |
2e60e022 | 2896 | unsigned long flags; |
32fab448 | 2897 | |
332ebbf7 BVA |
2898 | WARN_ON_ONCE(q->mq_ops); |
2899 | ||
2e60e022 TH |
2900 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2901 | return true; | |
32fab448 | 2902 | |
336cdb40 | 2903 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2904 | blk_finish_request(rq, error); |
336cdb40 KU |
2905 | spin_unlock_irqrestore(q->queue_lock, flags); |
2906 | ||
2e60e022 | 2907 | return false; |
32fab448 KU |
2908 | } |
2909 | ||
336cdb40 | 2910 | /** |
2e60e022 TH |
2911 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2912 | * @rq: the request to complete | |
2a842aca | 2913 | * @error: block status code |
e3a04fe3 KU |
2914 | * @nr_bytes: number of bytes to complete @rq |
2915 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2916 | * |
2917 | * Description: | |
2e60e022 TH |
2918 | * Identical to blk_end_bidi_request() except that queue lock is |
2919 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2920 | * |
2921 | * Return: | |
2e60e022 TH |
2922 | * %false - we are done with this request |
2923 | * %true - still buffers pending for this request | |
336cdb40 | 2924 | **/ |
2a842aca | 2925 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, |
b1f74493 | 2926 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2927 | { |
2fff8a92 | 2928 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 2929 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 2930 | |
2e60e022 TH |
2931 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2932 | return true; | |
336cdb40 | 2933 | |
2e60e022 | 2934 | blk_finish_request(rq, error); |
336cdb40 | 2935 | |
2e60e022 | 2936 | return false; |
336cdb40 | 2937 | } |
e19a3ab0 KU |
2938 | |
2939 | /** | |
2940 | * blk_end_request - Helper function for drivers to complete the request. | |
2941 | * @rq: the request being processed | |
2a842aca | 2942 | * @error: block status code |
e19a3ab0 KU |
2943 | * @nr_bytes: number of bytes to complete |
2944 | * | |
2945 | * Description: | |
2946 | * Ends I/O on a number of bytes attached to @rq. | |
2947 | * If @rq has leftover, sets it up for the next range of segments. | |
2948 | * | |
2949 | * Return: | |
b1f74493 FT |
2950 | * %false - we are done with this request |
2951 | * %true - still buffers pending for this request | |
e19a3ab0 | 2952 | **/ |
2a842aca CH |
2953 | bool blk_end_request(struct request *rq, blk_status_t error, |
2954 | unsigned int nr_bytes) | |
e19a3ab0 | 2955 | { |
332ebbf7 | 2956 | WARN_ON_ONCE(rq->q->mq_ops); |
b1f74493 | 2957 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2958 | } |
56ad1740 | 2959 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2960 | |
2961 | /** | |
b1f74493 FT |
2962 | * blk_end_request_all - Helper function for drives to finish the request. |
2963 | * @rq: the request to finish | |
2a842aca | 2964 | * @error: block status code |
336cdb40 KU |
2965 | * |
2966 | * Description: | |
b1f74493 FT |
2967 | * Completely finish @rq. |
2968 | */ | |
2a842aca | 2969 | void blk_end_request_all(struct request *rq, blk_status_t error) |
336cdb40 | 2970 | { |
b1f74493 FT |
2971 | bool pending; |
2972 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2973 | |
b1f74493 FT |
2974 | if (unlikely(blk_bidi_rq(rq))) |
2975 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2976 | |
b1f74493 FT |
2977 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2978 | BUG_ON(pending); | |
2979 | } | |
56ad1740 | 2980 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2981 | |
e3a04fe3 | 2982 | /** |
b1f74493 FT |
2983 | * __blk_end_request - Helper function for drivers to complete the request. |
2984 | * @rq: the request being processed | |
2a842aca | 2985 | * @error: block status code |
b1f74493 | 2986 | * @nr_bytes: number of bytes to complete |
e3a04fe3 KU |
2987 | * |
2988 | * Description: | |
b1f74493 | 2989 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2990 | * |
2991 | * Return: | |
b1f74493 FT |
2992 | * %false - we are done with this request |
2993 | * %true - still buffers pending for this request | |
e3a04fe3 | 2994 | **/ |
2a842aca CH |
2995 | bool __blk_end_request(struct request *rq, blk_status_t error, |
2996 | unsigned int nr_bytes) | |
e3a04fe3 | 2997 | { |
2fff8a92 | 2998 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 2999 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3000 | |
b1f74493 | 3001 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 3002 | } |
56ad1740 | 3003 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 3004 | |
32fab448 | 3005 | /** |
b1f74493 FT |
3006 | * __blk_end_request_all - Helper function for drives to finish the request. |
3007 | * @rq: the request to finish | |
2a842aca | 3008 | * @error: block status code |
32fab448 KU |
3009 | * |
3010 | * Description: | |
b1f74493 | 3011 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 3012 | */ |
2a842aca | 3013 | void __blk_end_request_all(struct request *rq, blk_status_t error) |
32fab448 | 3014 | { |
b1f74493 FT |
3015 | bool pending; |
3016 | unsigned int bidi_bytes = 0; | |
3017 | ||
2fff8a92 | 3018 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3019 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3020 | |
b1f74493 FT |
3021 | if (unlikely(blk_bidi_rq(rq))) |
3022 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3023 | ||
3024 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3025 | BUG_ON(pending); | |
32fab448 | 3026 | } |
56ad1740 | 3027 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 3028 | |
e19a3ab0 | 3029 | /** |
b1f74493 FT |
3030 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
3031 | * @rq: the request to finish the current chunk for | |
2a842aca | 3032 | * @error: block status code |
e19a3ab0 KU |
3033 | * |
3034 | * Description: | |
b1f74493 FT |
3035 | * Complete the current consecutively mapped chunk from @rq. Must |
3036 | * be called with queue lock held. | |
e19a3ab0 KU |
3037 | * |
3038 | * Return: | |
b1f74493 FT |
3039 | * %false - we are done with this request |
3040 | * %true - still buffers pending for this request | |
3041 | */ | |
2a842aca | 3042 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) |
e19a3ab0 | 3043 | { |
b1f74493 | 3044 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 3045 | } |
56ad1740 | 3046 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 3047 | |
86db1e29 JA |
3048 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3049 | struct bio *bio) | |
1da177e4 | 3050 | { |
b4f42e28 | 3051 | if (bio_has_data(bio)) |
fb2dce86 | 3052 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
b4f42e28 | 3053 | |
4f024f37 | 3054 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 3055 | rq->bio = rq->biotail = bio; |
1da177e4 | 3056 | |
74d46992 CH |
3057 | if (bio->bi_disk) |
3058 | rq->rq_disk = bio->bi_disk; | |
66846572 | 3059 | } |
1da177e4 | 3060 | |
2d4dc890 IL |
3061 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
3062 | /** | |
3063 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3064 | * @rq: the request to be flushed | |
3065 | * | |
3066 | * Description: | |
3067 | * Flush all pages in @rq. | |
3068 | */ | |
3069 | void rq_flush_dcache_pages(struct request *rq) | |
3070 | { | |
3071 | struct req_iterator iter; | |
7988613b | 3072 | struct bio_vec bvec; |
2d4dc890 IL |
3073 | |
3074 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 3075 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
3076 | } |
3077 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3078 | #endif | |
3079 | ||
ef9e3fac KU |
3080 | /** |
3081 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3082 | * @q : the queue of the device being checked | |
3083 | * | |
3084 | * Description: | |
3085 | * Check if underlying low-level drivers of a device are busy. | |
3086 | * If the drivers want to export their busy state, they must set own | |
3087 | * exporting function using blk_queue_lld_busy() first. | |
3088 | * | |
3089 | * Basically, this function is used only by request stacking drivers | |
3090 | * to stop dispatching requests to underlying devices when underlying | |
3091 | * devices are busy. This behavior helps more I/O merging on the queue | |
3092 | * of the request stacking driver and prevents I/O throughput regression | |
3093 | * on burst I/O load. | |
3094 | * | |
3095 | * Return: | |
3096 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3097 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3098 | */ | |
3099 | int blk_lld_busy(struct request_queue *q) | |
3100 | { | |
3101 | if (q->lld_busy_fn) | |
3102 | return q->lld_busy_fn(q); | |
3103 | ||
3104 | return 0; | |
3105 | } | |
3106 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3107 | ||
78d8e58a MS |
3108 | /** |
3109 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3110 | * @rq: the clone request to be cleaned up | |
3111 | * | |
3112 | * Description: | |
3113 | * Free all bios in @rq for a cloned request. | |
3114 | */ | |
3115 | void blk_rq_unprep_clone(struct request *rq) | |
3116 | { | |
3117 | struct bio *bio; | |
3118 | ||
3119 | while ((bio = rq->bio) != NULL) { | |
3120 | rq->bio = bio->bi_next; | |
3121 | ||
3122 | bio_put(bio); | |
3123 | } | |
3124 | } | |
3125 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3126 | ||
3127 | /* | |
3128 | * Copy attributes of the original request to the clone request. | |
3129 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3130 | */ | |
3131 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3132 | { |
3133 | dst->cpu = src->cpu; | |
b0fd271d KU |
3134 | dst->__sector = blk_rq_pos(src); |
3135 | dst->__data_len = blk_rq_bytes(src); | |
3136 | dst->nr_phys_segments = src->nr_phys_segments; | |
3137 | dst->ioprio = src->ioprio; | |
3138 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3139 | } |
3140 | ||
3141 | /** | |
3142 | * blk_rq_prep_clone - Helper function to setup clone request | |
3143 | * @rq: the request to be setup | |
3144 | * @rq_src: original request to be cloned | |
3145 | * @bs: bio_set that bios for clone are allocated from | |
3146 | * @gfp_mask: memory allocation mask for bio | |
3147 | * @bio_ctr: setup function to be called for each clone bio. | |
3148 | * Returns %0 for success, non %0 for failure. | |
3149 | * @data: private data to be passed to @bio_ctr | |
3150 | * | |
3151 | * Description: | |
3152 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3153 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3154 | * are not copied, and copying such parts is the caller's responsibility. | |
3155 | * Also, pages which the original bios are pointing to are not copied | |
3156 | * and the cloned bios just point same pages. | |
3157 | * So cloned bios must be completed before original bios, which means | |
3158 | * the caller must complete @rq before @rq_src. | |
3159 | */ | |
3160 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3161 | struct bio_set *bs, gfp_t gfp_mask, | |
3162 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3163 | void *data) | |
3164 | { | |
3165 | struct bio *bio, *bio_src; | |
3166 | ||
3167 | if (!bs) | |
3168 | bs = fs_bio_set; | |
3169 | ||
3170 | __rq_for_each_bio(bio_src, rq_src) { | |
3171 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3172 | if (!bio) | |
3173 | goto free_and_out; | |
3174 | ||
3175 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3176 | goto free_and_out; | |
3177 | ||
3178 | if (rq->bio) { | |
3179 | rq->biotail->bi_next = bio; | |
3180 | rq->biotail = bio; | |
3181 | } else | |
3182 | rq->bio = rq->biotail = bio; | |
3183 | } | |
3184 | ||
3185 | __blk_rq_prep_clone(rq, rq_src); | |
3186 | ||
3187 | return 0; | |
3188 | ||
3189 | free_and_out: | |
3190 | if (bio) | |
3191 | bio_put(bio); | |
3192 | blk_rq_unprep_clone(rq); | |
3193 | ||
3194 | return -ENOMEM; | |
b0fd271d KU |
3195 | } |
3196 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3197 | ||
59c3d45e | 3198 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3199 | { |
3200 | return queue_work(kblockd_workqueue, work); | |
3201 | } | |
1da177e4 LT |
3202 | EXPORT_SYMBOL(kblockd_schedule_work); |
3203 | ||
ee63cfa7 JA |
3204 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3205 | { | |
3206 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3207 | } | |
3208 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3209 | ||
818cd1cb JA |
3210 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
3211 | unsigned long delay) | |
3212 | { | |
3213 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3214 | } | |
3215 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3216 | ||
59c3d45e JA |
3217 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, |
3218 | unsigned long delay) | |
e43473b7 VG |
3219 | { |
3220 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3221 | } | |
3222 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3223 | ||
8ab14595 JA |
3224 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3225 | unsigned long delay) | |
3226 | { | |
3227 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3228 | } | |
3229 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3230 | ||
75df7136 SJ |
3231 | /** |
3232 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3233 | * @plug: The &struct blk_plug that needs to be initialized | |
3234 | * | |
3235 | * Description: | |
3236 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3237 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3238 | * blk_finish_plug(). This is important from a performance perspective, but | |
3239 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3240 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3241 | * page belonging to that request that is currently residing in our private | |
3242 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3243 | * this kind of deadlock. | |
3244 | */ | |
73c10101 JA |
3245 | void blk_start_plug(struct blk_plug *plug) |
3246 | { | |
3247 | struct task_struct *tsk = current; | |
3248 | ||
dd6cf3e1 SL |
3249 | /* |
3250 | * If this is a nested plug, don't actually assign it. | |
3251 | */ | |
3252 | if (tsk->plug) | |
3253 | return; | |
3254 | ||
73c10101 | 3255 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3256 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3257 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3258 | /* |
dd6cf3e1 SL |
3259 | * Store ordering should not be needed here, since a potential |
3260 | * preempt will imply a full memory barrier | |
73c10101 | 3261 | */ |
dd6cf3e1 | 3262 | tsk->plug = plug; |
73c10101 JA |
3263 | } |
3264 | EXPORT_SYMBOL(blk_start_plug); | |
3265 | ||
3266 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3267 | { | |
3268 | struct request *rqa = container_of(a, struct request, queuelist); | |
3269 | struct request *rqb = container_of(b, struct request, queuelist); | |
3270 | ||
975927b9 JM |
3271 | return !(rqa->q < rqb->q || |
3272 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3273 | } |
3274 | ||
49cac01e JA |
3275 | /* |
3276 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3277 | * until a safe kblockd context. We due this to avoid accidental big | |
3278 | * additional stack usage in driver dispatch, in places where the originally | |
3279 | * plugger did not intend it. | |
3280 | */ | |
f6603783 | 3281 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3282 | bool from_schedule) |
99e22598 | 3283 | __releases(q->queue_lock) |
94b5eb28 | 3284 | { |
2fff8a92 BVA |
3285 | lockdep_assert_held(q->queue_lock); |
3286 | ||
49cac01e | 3287 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3288 | |
70460571 | 3289 | if (from_schedule) |
24ecfbe2 | 3290 | blk_run_queue_async(q); |
70460571 | 3291 | else |
24ecfbe2 | 3292 | __blk_run_queue(q); |
70460571 | 3293 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3294 | } |
3295 | ||
74018dc3 | 3296 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3297 | { |
3298 | LIST_HEAD(callbacks); | |
3299 | ||
2a7d5559 SL |
3300 | while (!list_empty(&plug->cb_list)) { |
3301 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3302 | |
2a7d5559 SL |
3303 | while (!list_empty(&callbacks)) { |
3304 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3305 | struct blk_plug_cb, |
3306 | list); | |
2a7d5559 | 3307 | list_del(&cb->list); |
74018dc3 | 3308 | cb->callback(cb, from_schedule); |
2a7d5559 | 3309 | } |
048c9374 N |
3310 | } |
3311 | } | |
3312 | ||
9cbb1750 N |
3313 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3314 | int size) | |
3315 | { | |
3316 | struct blk_plug *plug = current->plug; | |
3317 | struct blk_plug_cb *cb; | |
3318 | ||
3319 | if (!plug) | |
3320 | return NULL; | |
3321 | ||
3322 | list_for_each_entry(cb, &plug->cb_list, list) | |
3323 | if (cb->callback == unplug && cb->data == data) | |
3324 | return cb; | |
3325 | ||
3326 | /* Not currently on the callback list */ | |
3327 | BUG_ON(size < sizeof(*cb)); | |
3328 | cb = kzalloc(size, GFP_ATOMIC); | |
3329 | if (cb) { | |
3330 | cb->data = data; | |
3331 | cb->callback = unplug; | |
3332 | list_add(&cb->list, &plug->cb_list); | |
3333 | } | |
3334 | return cb; | |
3335 | } | |
3336 | EXPORT_SYMBOL(blk_check_plugged); | |
3337 | ||
49cac01e | 3338 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3339 | { |
3340 | struct request_queue *q; | |
3341 | unsigned long flags; | |
3342 | struct request *rq; | |
109b8129 | 3343 | LIST_HEAD(list); |
94b5eb28 | 3344 | unsigned int depth; |
73c10101 | 3345 | |
74018dc3 | 3346 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3347 | |
3348 | if (!list_empty(&plug->mq_list)) | |
3349 | blk_mq_flush_plug_list(plug, from_schedule); | |
3350 | ||
73c10101 JA |
3351 | if (list_empty(&plug->list)) |
3352 | return; | |
3353 | ||
109b8129 N |
3354 | list_splice_init(&plug->list, &list); |
3355 | ||
422765c2 | 3356 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3357 | |
3358 | q = NULL; | |
94b5eb28 | 3359 | depth = 0; |
18811272 JA |
3360 | |
3361 | /* | |
3362 | * Save and disable interrupts here, to avoid doing it for every | |
3363 | * queue lock we have to take. | |
3364 | */ | |
73c10101 | 3365 | local_irq_save(flags); |
109b8129 N |
3366 | while (!list_empty(&list)) { |
3367 | rq = list_entry_rq(list.next); | |
73c10101 | 3368 | list_del_init(&rq->queuelist); |
73c10101 JA |
3369 | BUG_ON(!rq->q); |
3370 | if (rq->q != q) { | |
99e22598 JA |
3371 | /* |
3372 | * This drops the queue lock | |
3373 | */ | |
3374 | if (q) | |
49cac01e | 3375 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3376 | q = rq->q; |
94b5eb28 | 3377 | depth = 0; |
73c10101 JA |
3378 | spin_lock(q->queue_lock); |
3379 | } | |
8ba61435 TH |
3380 | |
3381 | /* | |
3382 | * Short-circuit if @q is dead | |
3383 | */ | |
3f3299d5 | 3384 | if (unlikely(blk_queue_dying(q))) { |
2a842aca | 3385 | __blk_end_request_all(rq, BLK_STS_IOERR); |
8ba61435 TH |
3386 | continue; |
3387 | } | |
3388 | ||
73c10101 JA |
3389 | /* |
3390 | * rq is already accounted, so use raw insert | |
3391 | */ | |
f73f44eb | 3392 | if (op_is_flush(rq->cmd_flags)) |
401a18e9 JA |
3393 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3394 | else | |
3395 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3396 | |
3397 | depth++; | |
73c10101 JA |
3398 | } |
3399 | ||
99e22598 JA |
3400 | /* |
3401 | * This drops the queue lock | |
3402 | */ | |
3403 | if (q) | |
49cac01e | 3404 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3405 | |
73c10101 JA |
3406 | local_irq_restore(flags); |
3407 | } | |
73c10101 JA |
3408 | |
3409 | void blk_finish_plug(struct blk_plug *plug) | |
3410 | { | |
dd6cf3e1 SL |
3411 | if (plug != current->plug) |
3412 | return; | |
f6603783 | 3413 | blk_flush_plug_list(plug, false); |
73c10101 | 3414 | |
dd6cf3e1 | 3415 | current->plug = NULL; |
73c10101 | 3416 | } |
88b996cd | 3417 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3418 | |
47fafbc7 | 3419 | #ifdef CONFIG_PM |
6c954667 LM |
3420 | /** |
3421 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3422 | * @q: the queue of the device | |
3423 | * @dev: the device the queue belongs to | |
3424 | * | |
3425 | * Description: | |
3426 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3427 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3428 | * should call this function after @dev has been initialized, and its | |
3429 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3430 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3431 | * cases, driver should call this function before any I/O has taken place. | |
3432 | * | |
3433 | * This function takes care of setting up using auto suspend for the device, | |
3434 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3435 | * until an updated value is either set by user or by driver. Drivers do | |
3436 | * not need to touch other autosuspend settings. | |
3437 | * | |
3438 | * The block layer runtime PM is request based, so only works for drivers | |
3439 | * that use request as their IO unit instead of those directly use bio's. | |
3440 | */ | |
3441 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3442 | { | |
765e40b6 CH |
3443 | /* not support for RQF_PM and ->rpm_status in blk-mq yet */ |
3444 | if (q->mq_ops) | |
3445 | return; | |
3446 | ||
6c954667 LM |
3447 | q->dev = dev; |
3448 | q->rpm_status = RPM_ACTIVE; | |
3449 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3450 | pm_runtime_use_autosuspend(q->dev); | |
3451 | } | |
3452 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3453 | ||
3454 | /** | |
3455 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3456 | * @q: the queue of the device | |
3457 | * | |
3458 | * Description: | |
3459 | * This function will check if runtime suspend is allowed for the device | |
3460 | * by examining if there are any requests pending in the queue. If there | |
3461 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3462 | * the queue's status will be updated to SUSPENDING and the driver can | |
3463 | * proceed to suspend the device. | |
3464 | * | |
3465 | * For the not allowed case, we mark last busy for the device so that | |
3466 | * runtime PM core will try to autosuspend it some time later. | |
3467 | * | |
3468 | * This function should be called near the start of the device's | |
3469 | * runtime_suspend callback. | |
3470 | * | |
3471 | * Return: | |
3472 | * 0 - OK to runtime suspend the device | |
3473 | * -EBUSY - Device should not be runtime suspended | |
3474 | */ | |
3475 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3476 | { | |
3477 | int ret = 0; | |
3478 | ||
4fd41a85 KX |
3479 | if (!q->dev) |
3480 | return ret; | |
3481 | ||
6c954667 LM |
3482 | spin_lock_irq(q->queue_lock); |
3483 | if (q->nr_pending) { | |
3484 | ret = -EBUSY; | |
3485 | pm_runtime_mark_last_busy(q->dev); | |
3486 | } else { | |
3487 | q->rpm_status = RPM_SUSPENDING; | |
3488 | } | |
3489 | spin_unlock_irq(q->queue_lock); | |
3490 | return ret; | |
3491 | } | |
3492 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3493 | ||
3494 | /** | |
3495 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3496 | * @q: the queue of the device | |
3497 | * @err: return value of the device's runtime_suspend function | |
3498 | * | |
3499 | * Description: | |
3500 | * Update the queue's runtime status according to the return value of the | |
3501 | * device's runtime suspend function and mark last busy for the device so | |
3502 | * that PM core will try to auto suspend the device at a later time. | |
3503 | * | |
3504 | * This function should be called near the end of the device's | |
3505 | * runtime_suspend callback. | |
3506 | */ | |
3507 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3508 | { | |
4fd41a85 KX |
3509 | if (!q->dev) |
3510 | return; | |
3511 | ||
6c954667 LM |
3512 | spin_lock_irq(q->queue_lock); |
3513 | if (!err) { | |
3514 | q->rpm_status = RPM_SUSPENDED; | |
3515 | } else { | |
3516 | q->rpm_status = RPM_ACTIVE; | |
3517 | pm_runtime_mark_last_busy(q->dev); | |
3518 | } | |
3519 | spin_unlock_irq(q->queue_lock); | |
3520 | } | |
3521 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3522 | ||
3523 | /** | |
3524 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3525 | * @q: the queue of the device | |
3526 | * | |
3527 | * Description: | |
3528 | * Update the queue's runtime status to RESUMING in preparation for the | |
3529 | * runtime resume of the device. | |
3530 | * | |
3531 | * This function should be called near the start of the device's | |
3532 | * runtime_resume callback. | |
3533 | */ | |
3534 | void blk_pre_runtime_resume(struct request_queue *q) | |
3535 | { | |
4fd41a85 KX |
3536 | if (!q->dev) |
3537 | return; | |
3538 | ||
6c954667 LM |
3539 | spin_lock_irq(q->queue_lock); |
3540 | q->rpm_status = RPM_RESUMING; | |
3541 | spin_unlock_irq(q->queue_lock); | |
3542 | } | |
3543 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3544 | ||
3545 | /** | |
3546 | * blk_post_runtime_resume - Post runtime resume processing | |
3547 | * @q: the queue of the device | |
3548 | * @err: return value of the device's runtime_resume function | |
3549 | * | |
3550 | * Description: | |
3551 | * Update the queue's runtime status according to the return value of the | |
3552 | * device's runtime_resume function. If it is successfully resumed, process | |
3553 | * the requests that are queued into the device's queue when it is resuming | |
3554 | * and then mark last busy and initiate autosuspend for it. | |
3555 | * | |
3556 | * This function should be called near the end of the device's | |
3557 | * runtime_resume callback. | |
3558 | */ | |
3559 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3560 | { | |
4fd41a85 KX |
3561 | if (!q->dev) |
3562 | return; | |
3563 | ||
6c954667 LM |
3564 | spin_lock_irq(q->queue_lock); |
3565 | if (!err) { | |
3566 | q->rpm_status = RPM_ACTIVE; | |
3567 | __blk_run_queue(q); | |
3568 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3569 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3570 | } else { |
3571 | q->rpm_status = RPM_SUSPENDED; | |
3572 | } | |
3573 | spin_unlock_irq(q->queue_lock); | |
3574 | } | |
3575 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3576 | |
3577 | /** | |
3578 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3579 | * @q: the queue of the device | |
3580 | * | |
3581 | * If the device is left runtime suspended during system suspend the resume | |
3582 | * hook typically resumes the device and corrects runtime status | |
3583 | * accordingly. However, that does not affect the queue runtime PM status | |
3584 | * which is still "suspended". This prevents processing requests from the | |
3585 | * queue. | |
3586 | * | |
3587 | * This function can be used in driver's resume hook to correct queue | |
3588 | * runtime PM status and re-enable peeking requests from the queue. It | |
3589 | * should be called before first request is added to the queue. | |
3590 | */ | |
3591 | void blk_set_runtime_active(struct request_queue *q) | |
3592 | { | |
3593 | spin_lock_irq(q->queue_lock); | |
3594 | q->rpm_status = RPM_ACTIVE; | |
3595 | pm_runtime_mark_last_busy(q->dev); | |
3596 | pm_request_autosuspend(q->dev); | |
3597 | spin_unlock_irq(q->queue_lock); | |
3598 | } | |
3599 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3600 | #endif |
3601 | ||
1da177e4 LT |
3602 | int __init blk_dev_init(void) |
3603 | { | |
ef295ecf CH |
3604 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3605 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3606 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3607 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3608 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3609 | |
89b90be2 TH |
3610 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3611 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3612 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3613 | if (!kblockd_workqueue) |
3614 | panic("Failed to create kblockd\n"); | |
3615 | ||
3616 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3617 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3618 | |
c2789bd4 | 3619 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3620 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3621 | |
18fbda91 OS |
3622 | #ifdef CONFIG_DEBUG_FS |
3623 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3624 | #endif | |
3625 | ||
d38ecf93 | 3626 | return 0; |
1da177e4 | 3627 | } |