]>
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); |
4e9b6f20 | 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 | ||
aba7afc5 | 342 | cancel_delayed_work_sync(&q->requeue_work); |
21c6e939 | 343 | queue_for_each_hw_ctx(q, hctx, i) |
9f993737 | 344 | cancel_delayed_work_sync(&hctx->run_work); |
f04c1fe7 ML |
345 | } else { |
346 | cancel_delayed_work_sync(&q->delay_work); | |
347 | } | |
1da177e4 LT |
348 | } |
349 | EXPORT_SYMBOL(blk_sync_queue); | |
350 | ||
c246e80d BVA |
351 | /** |
352 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
353 | * @q: The queue to run | |
354 | * | |
355 | * Description: | |
356 | * Invoke request handling on a queue if there are any pending requests. | |
357 | * May be used to restart request handling after a request has completed. | |
358 | * This variant runs the queue whether or not the queue has been | |
359 | * stopped. Must be called with the queue lock held and interrupts | |
360 | * disabled. See also @blk_run_queue. | |
361 | */ | |
362 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
363 | { | |
2fff8a92 | 364 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 365 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 366 | |
c246e80d BVA |
367 | if (unlikely(blk_queue_dead(q))) |
368 | return; | |
369 | ||
24faf6f6 BVA |
370 | /* |
371 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
372 | * the queue lock internally. As a result multiple threads may be | |
373 | * running such a request function concurrently. Keep track of the | |
374 | * number of active request_fn invocations such that blk_drain_queue() | |
375 | * can wait until all these request_fn calls have finished. | |
376 | */ | |
377 | q->request_fn_active++; | |
c246e80d | 378 | q->request_fn(q); |
24faf6f6 | 379 | q->request_fn_active--; |
c246e80d | 380 | } |
a7928c15 | 381 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 382 | |
1da177e4 | 383 | /** |
80a4b58e | 384 | * __blk_run_queue - run a single device queue |
1da177e4 | 385 | * @q: The queue to run |
80a4b58e JA |
386 | * |
387 | * Description: | |
2fff8a92 | 388 | * See @blk_run_queue. |
1da177e4 | 389 | */ |
24ecfbe2 | 390 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 391 | { |
2fff8a92 | 392 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 393 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 394 | |
a538cd03 TH |
395 | if (unlikely(blk_queue_stopped(q))) |
396 | return; | |
397 | ||
c246e80d | 398 | __blk_run_queue_uncond(q); |
75ad23bc NP |
399 | } |
400 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 401 | |
24ecfbe2 CH |
402 | /** |
403 | * blk_run_queue_async - run a single device queue in workqueue context | |
404 | * @q: The queue to run | |
405 | * | |
406 | * Description: | |
407 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
2fff8a92 BVA |
408 | * of us. |
409 | * | |
410 | * Note: | |
411 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
412 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
413 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
24ecfbe2 CH |
414 | */ |
415 | void blk_run_queue_async(struct request_queue *q) | |
416 | { | |
2fff8a92 | 417 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 418 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 419 | |
70460571 | 420 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 421 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 422 | } |
c21e6beb | 423 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 424 | |
75ad23bc NP |
425 | /** |
426 | * blk_run_queue - run a single device queue | |
427 | * @q: The queue to run | |
80a4b58e JA |
428 | * |
429 | * Description: | |
430 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 431 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
432 | */ |
433 | void blk_run_queue(struct request_queue *q) | |
434 | { | |
435 | unsigned long flags; | |
436 | ||
332ebbf7 BVA |
437 | WARN_ON_ONCE(q->mq_ops); |
438 | ||
75ad23bc | 439 | spin_lock_irqsave(q->queue_lock, flags); |
24ecfbe2 | 440 | __blk_run_queue(q); |
1da177e4 LT |
441 | spin_unlock_irqrestore(q->queue_lock, flags); |
442 | } | |
443 | EXPORT_SYMBOL(blk_run_queue); | |
444 | ||
165125e1 | 445 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
446 | { |
447 | kobject_put(&q->kobj); | |
448 | } | |
d86e0e83 | 449 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 450 | |
e3c78ca5 | 451 | /** |
807592a4 | 452 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 453 | * @q: queue to drain |
c9a929dd | 454 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 455 | * |
c9a929dd TH |
456 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
457 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
458 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 459 | */ |
807592a4 BVA |
460 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
461 | __releases(q->queue_lock) | |
462 | __acquires(q->queue_lock) | |
e3c78ca5 | 463 | { |
458f27a9 AH |
464 | int i; |
465 | ||
807592a4 | 466 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 467 | WARN_ON_ONCE(q->mq_ops); |
807592a4 | 468 | |
e3c78ca5 | 469 | while (true) { |
481a7d64 | 470 | bool drain = false; |
e3c78ca5 | 471 | |
b855b04a TH |
472 | /* |
473 | * The caller might be trying to drain @q before its | |
474 | * elevator is initialized. | |
475 | */ | |
476 | if (q->elevator) | |
477 | elv_drain_elevator(q); | |
478 | ||
5efd6113 | 479 | blkcg_drain_queue(q); |
e3c78ca5 | 480 | |
4eabc941 TH |
481 | /* |
482 | * This function might be called on a queue which failed | |
b855b04a TH |
483 | * driver init after queue creation or is not yet fully |
484 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
485 | * in such cases. Kick queue iff dispatch queue has | |
486 | * something on it and @q has request_fn set. | |
4eabc941 | 487 | */ |
b855b04a | 488 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 489 | __blk_run_queue(q); |
c9a929dd | 490 | |
8a5ecdd4 | 491 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 492 | drain |= q->request_fn_active; |
481a7d64 TH |
493 | |
494 | /* | |
495 | * Unfortunately, requests are queued at and tracked from | |
496 | * multiple places and there's no single counter which can | |
497 | * be drained. Check all the queues and counters. | |
498 | */ | |
499 | if (drain_all) { | |
e97c293c | 500 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
501 | drain |= !list_empty(&q->queue_head); |
502 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 503 | drain |= q->nr_rqs[i]; |
481a7d64 | 504 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
505 | if (fq) |
506 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
507 | } |
508 | } | |
e3c78ca5 | 509 | |
481a7d64 | 510 | if (!drain) |
e3c78ca5 | 511 | break; |
807592a4 BVA |
512 | |
513 | spin_unlock_irq(q->queue_lock); | |
514 | ||
e3c78ca5 | 515 | msleep(10); |
807592a4 BVA |
516 | |
517 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 518 | } |
458f27a9 AH |
519 | |
520 | /* | |
521 | * With queue marked dead, any woken up waiter will fail the | |
522 | * allocation path, so the wakeup chaining is lost and we're | |
523 | * left with hung waiters. We need to wake up those waiters. | |
524 | */ | |
525 | if (q->request_fn) { | |
a051661c TH |
526 | struct request_list *rl; |
527 | ||
a051661c TH |
528 | blk_queue_for_each_rl(rl, q) |
529 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
530 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 531 | } |
e3c78ca5 TH |
532 | } |
533 | ||
d732580b TH |
534 | /** |
535 | * blk_queue_bypass_start - enter queue bypass mode | |
536 | * @q: queue of interest | |
537 | * | |
538 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
539 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 540 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
541 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
542 | * inside queue or RCU read lock. | |
d732580b TH |
543 | */ |
544 | void blk_queue_bypass_start(struct request_queue *q) | |
545 | { | |
332ebbf7 BVA |
546 | WARN_ON_ONCE(q->mq_ops); |
547 | ||
d732580b | 548 | spin_lock_irq(q->queue_lock); |
776687bc | 549 | q->bypass_depth++; |
d732580b TH |
550 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
551 | spin_unlock_irq(q->queue_lock); | |
552 | ||
776687bc TH |
553 | /* |
554 | * Queues start drained. Skip actual draining till init is | |
555 | * complete. This avoids lenghty delays during queue init which | |
556 | * can happen many times during boot. | |
557 | */ | |
558 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
559 | spin_lock_irq(q->queue_lock); |
560 | __blk_drain_queue(q, false); | |
561 | spin_unlock_irq(q->queue_lock); | |
562 | ||
b82d4b19 TH |
563 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
564 | synchronize_rcu(); | |
565 | } | |
d732580b TH |
566 | } |
567 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
568 | ||
569 | /** | |
570 | * blk_queue_bypass_end - leave queue bypass mode | |
571 | * @q: queue of interest | |
572 | * | |
573 | * Leave bypass mode and restore the normal queueing behavior. | |
332ebbf7 BVA |
574 | * |
575 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
576 | * this function is called for both blk-sq and blk-mq queues. | |
d732580b TH |
577 | */ |
578 | void blk_queue_bypass_end(struct request_queue *q) | |
579 | { | |
580 | spin_lock_irq(q->queue_lock); | |
581 | if (!--q->bypass_depth) | |
582 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
583 | WARN_ON_ONCE(q->bypass_depth < 0); | |
584 | spin_unlock_irq(q->queue_lock); | |
585 | } | |
586 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
587 | ||
aed3ea94 JA |
588 | void blk_set_queue_dying(struct request_queue *q) |
589 | { | |
1b856086 BVA |
590 | spin_lock_irq(q->queue_lock); |
591 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
592 | spin_unlock_irq(q->queue_lock); | |
aed3ea94 | 593 | |
d3cfb2a0 ML |
594 | /* |
595 | * When queue DYING flag is set, we need to block new req | |
596 | * entering queue, so we call blk_freeze_queue_start() to | |
597 | * prevent I/O from crossing blk_queue_enter(). | |
598 | */ | |
599 | blk_freeze_queue_start(q); | |
600 | ||
aed3ea94 JA |
601 | if (q->mq_ops) |
602 | blk_mq_wake_waiters(q); | |
603 | else { | |
604 | struct request_list *rl; | |
605 | ||
bbfc3c5d | 606 | spin_lock_irq(q->queue_lock); |
aed3ea94 JA |
607 | blk_queue_for_each_rl(rl, q) { |
608 | if (rl->rq_pool) { | |
609 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
610 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
611 | } | |
612 | } | |
bbfc3c5d | 613 | spin_unlock_irq(q->queue_lock); |
aed3ea94 JA |
614 | } |
615 | } | |
616 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
617 | ||
c9a929dd TH |
618 | /** |
619 | * blk_cleanup_queue - shutdown a request queue | |
620 | * @q: request queue to shutdown | |
621 | * | |
c246e80d BVA |
622 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
623 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 624 | */ |
6728cb0e | 625 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 626 | { |
c9a929dd | 627 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 628 | |
3f3299d5 | 629 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 630 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 631 | blk_set_queue_dying(q); |
c9a929dd | 632 | spin_lock_irq(lock); |
6ecf23af | 633 | |
80fd9979 | 634 | /* |
3f3299d5 | 635 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
636 | * that, unlike blk_queue_bypass_start(), we aren't performing |
637 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
638 | * as some drivers create and destroy a lot of queues while | |
639 | * probing. This is still safe because blk_release_queue() will be | |
640 | * called only after the queue refcnt drops to zero and nothing, | |
641 | * RCU or not, would be traversing the queue by then. | |
642 | */ | |
6ecf23af TH |
643 | q->bypass_depth++; |
644 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
645 | ||
c9a929dd TH |
646 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
647 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 648 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
649 | spin_unlock_irq(lock); |
650 | mutex_unlock(&q->sysfs_lock); | |
651 | ||
c246e80d BVA |
652 | /* |
653 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
654 | * prevent that q->request_fn() gets invoked after draining finished. | |
655 | */ | |
3ef28e83 | 656 | blk_freeze_queue(q); |
9c1051aa OS |
657 | spin_lock_irq(lock); |
658 | if (!q->mq_ops) | |
43a5e4e2 | 659 | __blk_drain_queue(q, true); |
c246e80d | 660 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 661 | spin_unlock_irq(lock); |
c9a929dd | 662 | |
5a48fc14 DW |
663 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
664 | blk_flush_integrity(); | |
665 | ||
c9a929dd | 666 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 667 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
668 | blk_sync_queue(q); |
669 | ||
45a9c9d9 BVA |
670 | if (q->mq_ops) |
671 | blk_mq_free_queue(q); | |
3ef28e83 | 672 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 673 | |
5e5cfac0 AH |
674 | spin_lock_irq(lock); |
675 | if (q->queue_lock != &q->__queue_lock) | |
676 | q->queue_lock = &q->__queue_lock; | |
677 | spin_unlock_irq(lock); | |
678 | ||
c9a929dd | 679 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
680 | blk_put_queue(q); |
681 | } | |
1da177e4 LT |
682 | EXPORT_SYMBOL(blk_cleanup_queue); |
683 | ||
271508db | 684 | /* Allocate memory local to the request queue */ |
6d247d7f | 685 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) |
271508db | 686 | { |
6d247d7f CH |
687 | struct request_queue *q = data; |
688 | ||
689 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
271508db DR |
690 | } |
691 | ||
6d247d7f | 692 | static void free_request_simple(void *element, void *data) |
271508db DR |
693 | { |
694 | kmem_cache_free(request_cachep, element); | |
695 | } | |
696 | ||
6d247d7f CH |
697 | static void *alloc_request_size(gfp_t gfp_mask, void *data) |
698 | { | |
699 | struct request_queue *q = data; | |
700 | struct request *rq; | |
701 | ||
702 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
703 | q->node); | |
704 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
705 | kfree(rq); | |
706 | rq = NULL; | |
707 | } | |
708 | return rq; | |
709 | } | |
710 | ||
711 | static void free_request_size(void *element, void *data) | |
712 | { | |
713 | struct request_queue *q = data; | |
714 | ||
715 | if (q->exit_rq_fn) | |
716 | q->exit_rq_fn(q, element); | |
717 | kfree(element); | |
718 | } | |
719 | ||
5b788ce3 TH |
720 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
721 | gfp_t gfp_mask) | |
1da177e4 | 722 | { |
85acb3ba | 723 | if (unlikely(rl->rq_pool) || q->mq_ops) |
1abec4fd MS |
724 | return 0; |
725 | ||
5b788ce3 | 726 | rl->q = q; |
1faa16d2 JA |
727 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
728 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
729 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
730 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 731 | |
6d247d7f CH |
732 | if (q->cmd_size) { |
733 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
734 | alloc_request_size, free_request_size, | |
735 | q, gfp_mask, q->node); | |
736 | } else { | |
737 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
738 | alloc_request_simple, free_request_simple, | |
739 | q, gfp_mask, q->node); | |
740 | } | |
1da177e4 LT |
741 | if (!rl->rq_pool) |
742 | return -ENOMEM; | |
743 | ||
b425e504 BVA |
744 | if (rl != &q->root_rl) |
745 | WARN_ON_ONCE(!blk_get_queue(q)); | |
746 | ||
1da177e4 LT |
747 | return 0; |
748 | } | |
749 | ||
b425e504 | 750 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) |
5b788ce3 | 751 | { |
b425e504 | 752 | if (rl->rq_pool) { |
5b788ce3 | 753 | mempool_destroy(rl->rq_pool); |
b425e504 BVA |
754 | if (rl != &q->root_rl) |
755 | blk_put_queue(q); | |
756 | } | |
5b788ce3 TH |
757 | } |
758 | ||
165125e1 | 759 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 760 | { |
c304a51b | 761 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
762 | } |
763 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 764 | |
6f3b0e8b | 765 | int blk_queue_enter(struct request_queue *q, bool nowait) |
3ef28e83 DW |
766 | { |
767 | while (true) { | |
768 | int ret; | |
769 | ||
770 | if (percpu_ref_tryget_live(&q->q_usage_counter)) | |
771 | return 0; | |
772 | ||
6f3b0e8b | 773 | if (nowait) |
3ef28e83 DW |
774 | return -EBUSY; |
775 | ||
5ed61d3f | 776 | /* |
1671d522 | 777 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 778 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
779 | * .q_usage_counter and reading .mq_freeze_depth or |
780 | * queue dying flag, otherwise the following wait may | |
781 | * never return if the two reads are reordered. | |
5ed61d3f ML |
782 | */ |
783 | smp_rmb(); | |
784 | ||
3ef28e83 DW |
785 | ret = wait_event_interruptible(q->mq_freeze_wq, |
786 | !atomic_read(&q->mq_freeze_depth) || | |
787 | blk_queue_dying(q)); | |
788 | if (blk_queue_dying(q)) | |
789 | return -ENODEV; | |
790 | if (ret) | |
791 | return ret; | |
792 | } | |
793 | } | |
794 | ||
795 | void blk_queue_exit(struct request_queue *q) | |
796 | { | |
797 | percpu_ref_put(&q->q_usage_counter); | |
798 | } | |
799 | ||
800 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
801 | { | |
802 | struct request_queue *q = | |
803 | container_of(ref, struct request_queue, q_usage_counter); | |
804 | ||
805 | wake_up_all(&q->mq_freeze_wq); | |
806 | } | |
807 | ||
287922eb CH |
808 | static void blk_rq_timed_out_timer(unsigned long data) |
809 | { | |
810 | struct request_queue *q = (struct request_queue *)data; | |
811 | ||
812 | kblockd_schedule_work(&q->timeout_work); | |
813 | } | |
814 | ||
165125e1 | 815 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 816 | { |
165125e1 | 817 | struct request_queue *q; |
1946089a | 818 | |
8324aa91 | 819 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 820 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
821 | if (!q) |
822 | return NULL; | |
823 | ||
00380a40 | 824 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 825 | if (q->id < 0) |
3d2936f4 | 826 | goto fail_q; |
a73f730d | 827 | |
93b27e72 | 828 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
54efd50b KO |
829 | if (!q->bio_split) |
830 | goto fail_id; | |
831 | ||
d03f6cdc JK |
832 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
833 | if (!q->backing_dev_info) | |
834 | goto fail_split; | |
835 | ||
a83b576c JA |
836 | q->stats = blk_alloc_queue_stats(); |
837 | if (!q->stats) | |
838 | goto fail_stats; | |
839 | ||
dc3b17cc | 840 | q->backing_dev_info->ra_pages = |
09cbfeaf | 841 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
dc3b17cc JK |
842 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
843 | q->backing_dev_info->name = "block"; | |
5151412d | 844 | q->node = node_id; |
0989a025 | 845 | |
dc3b17cc | 846 | setup_timer(&q->backing_dev_info->laptop_mode_wb_timer, |
31373d09 | 847 | laptop_mode_timer_fn, (unsigned long) q); |
242f9dcb | 848 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
4e9b6f20 | 849 | INIT_WORK(&q->timeout_work, NULL); |
b855b04a | 850 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 851 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 852 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 853 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 854 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 855 | #endif |
3cca6dc1 | 856 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 857 | |
8324aa91 | 858 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 859 | |
5acb3cc2 WL |
860 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
861 | mutex_init(&q->blk_trace_mutex); | |
862 | #endif | |
483f4afc | 863 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 864 | spin_lock_init(&q->__queue_lock); |
483f4afc | 865 | |
c94a96ac VG |
866 | /* |
867 | * By default initialize queue_lock to internal lock and driver can | |
868 | * override it later if need be. | |
869 | */ | |
870 | q->queue_lock = &q->__queue_lock; | |
871 | ||
b82d4b19 TH |
872 | /* |
873 | * A queue starts its life with bypass turned on to avoid | |
874 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
875 | * init. The initial bypass will be finished when the queue is |
876 | * registered by blk_register_queue(). | |
b82d4b19 TH |
877 | */ |
878 | q->bypass_depth = 1; | |
879 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
880 | ||
320ae51f JA |
881 | init_waitqueue_head(&q->mq_freeze_wq); |
882 | ||
3ef28e83 DW |
883 | /* |
884 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
885 | * See blk_register_queue() for details. | |
886 | */ | |
887 | if (percpu_ref_init(&q->q_usage_counter, | |
888 | blk_queue_usage_counter_release, | |
889 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 890 | goto fail_bdi; |
f51b802c | 891 | |
3ef28e83 DW |
892 | if (blkcg_init_queue(q)) |
893 | goto fail_ref; | |
894 | ||
1da177e4 | 895 | return q; |
a73f730d | 896 | |
3ef28e83 DW |
897 | fail_ref: |
898 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 899 | fail_bdi: |
a83b576c JA |
900 | blk_free_queue_stats(q->stats); |
901 | fail_stats: | |
d03f6cdc | 902 | bdi_put(q->backing_dev_info); |
54efd50b KO |
903 | fail_split: |
904 | bioset_free(q->bio_split); | |
a73f730d TH |
905 | fail_id: |
906 | ida_simple_remove(&blk_queue_ida, q->id); | |
907 | fail_q: | |
908 | kmem_cache_free(blk_requestq_cachep, q); | |
909 | return NULL; | |
1da177e4 | 910 | } |
1946089a | 911 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
912 | |
913 | /** | |
914 | * blk_init_queue - prepare a request queue for use with a block device | |
915 | * @rfn: The function to be called to process requests that have been | |
916 | * placed on the queue. | |
917 | * @lock: Request queue spin lock | |
918 | * | |
919 | * Description: | |
920 | * If a block device wishes to use the standard request handling procedures, | |
921 | * which sorts requests and coalesces adjacent requests, then it must | |
922 | * call blk_init_queue(). The function @rfn will be called when there | |
923 | * are requests on the queue that need to be processed. If the device | |
924 | * supports plugging, then @rfn may not be called immediately when requests | |
925 | * are available on the queue, but may be called at some time later instead. | |
926 | * Plugged queues are generally unplugged when a buffer belonging to one | |
927 | * of the requests on the queue is needed, or due to memory pressure. | |
928 | * | |
929 | * @rfn is not required, or even expected, to remove all requests off the | |
930 | * queue, but only as many as it can handle at a time. If it does leave | |
931 | * requests on the queue, it is responsible for arranging that the requests | |
932 | * get dealt with eventually. | |
933 | * | |
934 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
935 | * request queue; this lock will be taken also from interrupt context, so irq |
936 | * disabling is needed for it. | |
1da177e4 | 937 | * |
710027a4 | 938 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
939 | * it didn't succeed. |
940 | * | |
941 | * Note: | |
942 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
943 | * when the block device is deactivated (such as at module unload). | |
944 | **/ | |
1946089a | 945 | |
165125e1 | 946 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 947 | { |
c304a51b | 948 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
949 | } |
950 | EXPORT_SYMBOL(blk_init_queue); | |
951 | ||
165125e1 | 952 | struct request_queue * |
1946089a CL |
953 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
954 | { | |
5ea708d1 | 955 | struct request_queue *q; |
1da177e4 | 956 | |
5ea708d1 CH |
957 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
958 | if (!q) | |
c86d1b8a MS |
959 | return NULL; |
960 | ||
5ea708d1 CH |
961 | q->request_fn = rfn; |
962 | if (lock) | |
963 | q->queue_lock = lock; | |
964 | if (blk_init_allocated_queue(q) < 0) { | |
965 | blk_cleanup_queue(q); | |
966 | return NULL; | |
967 | } | |
18741986 | 968 | |
7982e90c | 969 | return q; |
01effb0d MS |
970 | } |
971 | EXPORT_SYMBOL(blk_init_queue_node); | |
972 | ||
dece1635 | 973 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 974 | |
1da177e4 | 975 | |
5ea708d1 CH |
976 | int blk_init_allocated_queue(struct request_queue *q) |
977 | { | |
332ebbf7 BVA |
978 | WARN_ON_ONCE(q->mq_ops); |
979 | ||
6d247d7f | 980 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); |
ba483388 | 981 | if (!q->fq) |
5ea708d1 | 982 | return -ENOMEM; |
7982e90c | 983 | |
6d247d7f CH |
984 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) |
985 | goto out_free_flush_queue; | |
7982e90c | 986 | |
a051661c | 987 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
6d247d7f | 988 | goto out_exit_flush_rq; |
1da177e4 | 989 | |
287922eb | 990 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
60ea8226 | 991 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac | 992 | |
f3b144aa JA |
993 | /* |
994 | * This also sets hw/phys segments, boundary and size | |
995 | */ | |
c20e8de2 | 996 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 997 | |
44ec9542 AS |
998 | q->sg_reserved_size = INT_MAX; |
999 | ||
eb1c160b TS |
1000 | /* Protect q->elevator from elevator_change */ |
1001 | mutex_lock(&q->sysfs_lock); | |
1002 | ||
b82d4b19 | 1003 | /* init elevator */ |
eb1c160b TS |
1004 | if (elevator_init(q, NULL)) { |
1005 | mutex_unlock(&q->sysfs_lock); | |
6d247d7f | 1006 | goto out_exit_flush_rq; |
eb1c160b TS |
1007 | } |
1008 | ||
1009 | mutex_unlock(&q->sysfs_lock); | |
5ea708d1 | 1010 | return 0; |
eb1c160b | 1011 | |
6d247d7f CH |
1012 | out_exit_flush_rq: |
1013 | if (q->exit_rq_fn) | |
1014 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1015 | out_free_flush_queue: | |
ba483388 | 1016 | blk_free_flush_queue(q->fq); |
5ea708d1 | 1017 | return -ENOMEM; |
1da177e4 | 1018 | } |
5151412d | 1019 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 1020 | |
09ac46c4 | 1021 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 1022 | { |
3f3299d5 | 1023 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
1024 | __blk_get_queue(q); |
1025 | return true; | |
1da177e4 LT |
1026 | } |
1027 | ||
09ac46c4 | 1028 | return false; |
1da177e4 | 1029 | } |
d86e0e83 | 1030 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 1031 | |
5b788ce3 | 1032 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 1033 | { |
e8064021 | 1034 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 1035 | elv_put_request(rl->q, rq); |
f1f8cc94 | 1036 | if (rq->elv.icq) |
11a3122f | 1037 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
1038 | } |
1039 | ||
5b788ce3 | 1040 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
1041 | } |
1042 | ||
1da177e4 LT |
1043 | /* |
1044 | * ioc_batching returns true if the ioc is a valid batching request and | |
1045 | * should be given priority access to a request. | |
1046 | */ | |
165125e1 | 1047 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1048 | { |
1049 | if (!ioc) | |
1050 | return 0; | |
1051 | ||
1052 | /* | |
1053 | * Make sure the process is able to allocate at least 1 request | |
1054 | * even if the batch times out, otherwise we could theoretically | |
1055 | * lose wakeups. | |
1056 | */ | |
1057 | return ioc->nr_batch_requests == q->nr_batching || | |
1058 | (ioc->nr_batch_requests > 0 | |
1059 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1064 | * will cause the process to be a "batcher" on all queues in the system. This | |
1065 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1066 | * a nice run. | |
1067 | */ | |
165125e1 | 1068 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1069 | { |
1070 | if (!ioc || ioc_batching(q, ioc)) | |
1071 | return; | |
1072 | ||
1073 | ioc->nr_batch_requests = q->nr_batching; | |
1074 | ioc->last_waited = jiffies; | |
1075 | } | |
1076 | ||
5b788ce3 | 1077 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 1078 | { |
5b788ce3 | 1079 | struct request_queue *q = rl->q; |
1da177e4 | 1080 | |
d40f75a0 TH |
1081 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
1082 | blk_clear_congested(rl, sync); | |
1da177e4 | 1083 | |
1faa16d2 JA |
1084 | if (rl->count[sync] + 1 <= q->nr_requests) { |
1085 | if (waitqueue_active(&rl->wait[sync])) | |
1086 | wake_up(&rl->wait[sync]); | |
1da177e4 | 1087 | |
5b788ce3 | 1088 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
1089 | } |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * A request has just been released. Account for it, update the full and | |
1094 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1095 | */ | |
e8064021 CH |
1096 | static void freed_request(struct request_list *rl, bool sync, |
1097 | req_flags_t rq_flags) | |
1da177e4 | 1098 | { |
5b788ce3 | 1099 | struct request_queue *q = rl->q; |
1da177e4 | 1100 | |
8a5ecdd4 | 1101 | q->nr_rqs[sync]--; |
1faa16d2 | 1102 | rl->count[sync]--; |
e8064021 | 1103 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 1104 | q->nr_rqs_elvpriv--; |
1da177e4 | 1105 | |
5b788ce3 | 1106 | __freed_request(rl, sync); |
1da177e4 | 1107 | |
1faa16d2 | 1108 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 1109 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
1110 | } |
1111 | ||
e3a2b3f9 JA |
1112 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
1113 | { | |
1114 | struct request_list *rl; | |
d40f75a0 | 1115 | int on_thresh, off_thresh; |
e3a2b3f9 | 1116 | |
332ebbf7 BVA |
1117 | WARN_ON_ONCE(q->mq_ops); |
1118 | ||
e3a2b3f9 JA |
1119 | spin_lock_irq(q->queue_lock); |
1120 | q->nr_requests = nr; | |
1121 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
1122 | on_thresh = queue_congestion_on_threshold(q); |
1123 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 1124 | |
d40f75a0 TH |
1125 | blk_queue_for_each_rl(rl, q) { |
1126 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1127 | blk_set_congested(rl, BLK_RW_SYNC); | |
1128 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1129 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 1130 | |
d40f75a0 TH |
1131 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1132 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1133 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1134 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1135 | |
e3a2b3f9 JA |
1136 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1137 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1138 | } else { | |
1139 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1140 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1141 | } | |
1142 | ||
1143 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1144 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1145 | } else { | |
1146 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1147 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1148 | } | |
1149 | } | |
1150 | ||
1151 | spin_unlock_irq(q->queue_lock); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
da8303c6 | 1155 | /** |
a06e05e6 | 1156 | * __get_request - get a free request |
5b788ce3 | 1157 | * @rl: request list to allocate from |
ef295ecf | 1158 | * @op: operation and flags |
da8303c6 TH |
1159 | * @bio: bio to allocate request for (can be %NULL) |
1160 | * @gfp_mask: allocation mask | |
1161 | * | |
1162 | * Get a free request from @q. This function may fail under memory | |
1163 | * pressure or if @q is dead. | |
1164 | * | |
da3dae54 | 1165 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1166 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1167 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1168 | */ |
ef295ecf CH |
1169 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
1170 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1171 | { |
5b788ce3 | 1172 | struct request_queue *q = rl->q; |
b679281a | 1173 | struct request *rq; |
7f4b35d1 TH |
1174 | struct elevator_type *et = q->elevator->type; |
1175 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1176 | struct io_cq *icq = NULL; |
ef295ecf | 1177 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1178 | int may_queue; |
e8064021 | 1179 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1180 | |
2fff8a92 BVA |
1181 | lockdep_assert_held(q->queue_lock); |
1182 | ||
3f3299d5 | 1183 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1184 | return ERR_PTR(-ENODEV); |
da8303c6 | 1185 | |
ef295ecf | 1186 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1187 | if (may_queue == ELV_MQUEUE_NO) |
1188 | goto rq_starved; | |
1189 | ||
1faa16d2 JA |
1190 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1191 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1192 | /* |
1193 | * The queue will fill after this allocation, so set | |
1194 | * it as full, and mark this process as "batching". | |
1195 | * This process will be allowed to complete a batch of | |
1196 | * requests, others will be blocked. | |
1197 | */ | |
5b788ce3 | 1198 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1199 | ioc_set_batching(q, ioc); |
5b788ce3 | 1200 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1201 | } else { |
1202 | if (may_queue != ELV_MQUEUE_MUST | |
1203 | && !ioc_batching(q, ioc)) { | |
1204 | /* | |
1205 | * The queue is full and the allocating | |
1206 | * process is not a "batcher", and not | |
1207 | * exempted by the IO scheduler | |
1208 | */ | |
a492f075 | 1209 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1210 | } |
1211 | } | |
1da177e4 | 1212 | } |
d40f75a0 | 1213 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1214 | } |
1215 | ||
082cf69e JA |
1216 | /* |
1217 | * Only allow batching queuers to allocate up to 50% over the defined | |
1218 | * limit of requests, otherwise we could have thousands of requests | |
1219 | * allocated with any setting of ->nr_requests | |
1220 | */ | |
1faa16d2 | 1221 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1222 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1223 | |
8a5ecdd4 | 1224 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1225 | rl->count[is_sync]++; |
1226 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1227 | |
f1f8cc94 TH |
1228 | /* |
1229 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1230 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1231 | * prevent the current elevator from being destroyed until the new |
1232 | * request is freed. This guarantees icq's won't be destroyed and | |
1233 | * makes creating new ones safe. | |
1234 | * | |
e6f7f93d CH |
1235 | * Flush requests do not use the elevator so skip initialization. |
1236 | * This allows a request to share the flush and elevator data. | |
1237 | * | |
f1f8cc94 TH |
1238 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
1239 | * it will be created after releasing queue_lock. | |
1240 | */ | |
e6f7f93d | 1241 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { |
e8064021 | 1242 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1243 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1244 | if (et->icq_cache && ioc) |
1245 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1246 | } |
cb98fc8b | 1247 | |
f253b86b | 1248 | if (blk_queue_io_stat(q)) |
e8064021 | 1249 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1250 | spin_unlock_irq(q->queue_lock); |
1251 | ||
29e2b09a | 1252 | /* allocate and init request */ |
5b788ce3 | 1253 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1254 | if (!rq) |
b679281a | 1255 | goto fail_alloc; |
1da177e4 | 1256 | |
29e2b09a | 1257 | blk_rq_init(q, rq); |
a051661c | 1258 | blk_rq_set_rl(rq, rl); |
ef295ecf | 1259 | rq->cmd_flags = op; |
e8064021 | 1260 | rq->rq_flags = rq_flags; |
29e2b09a | 1261 | |
aaf7c680 | 1262 | /* init elvpriv */ |
e8064021 | 1263 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1264 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1265 | if (ioc) |
1266 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1267 | if (!icq) |
1268 | goto fail_elvpriv; | |
29e2b09a | 1269 | } |
aaf7c680 TH |
1270 | |
1271 | rq->elv.icq = icq; | |
1272 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1273 | goto fail_elvpriv; | |
1274 | ||
1275 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1276 | if (icq) |
1277 | get_io_context(icq->ioc); | |
1278 | } | |
aaf7c680 | 1279 | out: |
88ee5ef1 JA |
1280 | /* |
1281 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1282 | * OK, if the queue is under the request limit then requests need | |
1283 | * not count toward the nr_batch_requests limit. There will always | |
1284 | * be some limit enforced by BLK_BATCH_TIME. | |
1285 | */ | |
1da177e4 LT |
1286 | if (ioc_batching(q, ioc)) |
1287 | ioc->nr_batch_requests--; | |
6728cb0e | 1288 | |
e6a40b09 | 1289 | trace_block_getrq(q, bio, op); |
1da177e4 | 1290 | return rq; |
b679281a | 1291 | |
aaf7c680 TH |
1292 | fail_elvpriv: |
1293 | /* | |
1294 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1295 | * and may fail indefinitely under memory pressure and thus | |
1296 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1297 | * disturb iosched and blkcg but weird is bettern than dead. | |
1298 | */ | |
7b2b10e0 | 1299 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
dc3b17cc | 1300 | __func__, dev_name(q->backing_dev_info->dev)); |
aaf7c680 | 1301 | |
e8064021 | 1302 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1303 | rq->elv.icq = NULL; |
1304 | ||
1305 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1306 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1307 | spin_unlock_irq(q->queue_lock); |
1308 | goto out; | |
1309 | ||
b679281a TH |
1310 | fail_alloc: |
1311 | /* | |
1312 | * Allocation failed presumably due to memory. Undo anything we | |
1313 | * might have messed up. | |
1314 | * | |
1315 | * Allocating task should really be put onto the front of the wait | |
1316 | * queue, but this is pretty rare. | |
1317 | */ | |
1318 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1319 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1320 | |
1321 | /* | |
1322 | * in the very unlikely event that allocation failed and no | |
1323 | * requests for this direction was pending, mark us starved so that | |
1324 | * freeing of a request in the other direction will notice | |
1325 | * us. another possible fix would be to split the rq mempool into | |
1326 | * READ and WRITE | |
1327 | */ | |
1328 | rq_starved: | |
1329 | if (unlikely(rl->count[is_sync] == 0)) | |
1330 | rl->starved[is_sync] = 1; | |
a492f075 | 1331 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1332 | } |
1333 | ||
da8303c6 | 1334 | /** |
a06e05e6 | 1335 | * get_request - get a free request |
da8303c6 | 1336 | * @q: request_queue to allocate request from |
ef295ecf | 1337 | * @op: operation and flags |
da8303c6 | 1338 | * @bio: bio to allocate request for (can be %NULL) |
a06e05e6 | 1339 | * @gfp_mask: allocation mask |
da8303c6 | 1340 | * |
d0164adc MG |
1341 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, |
1342 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1343 | * |
da3dae54 | 1344 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1345 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1346 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1347 | */ |
ef295ecf CH |
1348 | static struct request *get_request(struct request_queue *q, unsigned int op, |
1349 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1350 | { |
ef295ecf | 1351 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1352 | DEFINE_WAIT(wait); |
a051661c | 1353 | struct request_list *rl; |
1da177e4 | 1354 | struct request *rq; |
a051661c | 1355 | |
2fff8a92 | 1356 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1357 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1358 | |
a051661c | 1359 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ |
a06e05e6 | 1360 | retry: |
ef295ecf | 1361 | rq = __get_request(rl, op, bio, gfp_mask); |
a492f075 | 1362 | if (!IS_ERR(rq)) |
a06e05e6 | 1363 | return rq; |
1da177e4 | 1364 | |
03a07c92 GR |
1365 | if (op & REQ_NOWAIT) { |
1366 | blk_put_rl(rl); | |
1367 | return ERR_PTR(-EAGAIN); | |
1368 | } | |
1369 | ||
d0164adc | 1370 | if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) { |
a051661c | 1371 | blk_put_rl(rl); |
a492f075 | 1372 | return rq; |
a051661c | 1373 | } |
1da177e4 | 1374 | |
a06e05e6 TH |
1375 | /* wait on @rl and retry */ |
1376 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1377 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1378 | |
e6a40b09 | 1379 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1380 | |
a06e05e6 TH |
1381 | spin_unlock_irq(q->queue_lock); |
1382 | io_schedule(); | |
d6344532 | 1383 | |
a06e05e6 TH |
1384 | /* |
1385 | * After sleeping, we become a "batching" process and will be able | |
1386 | * to allocate at least one request, and up to a big batch of them | |
1387 | * for a small period time. See ioc_batching, ioc_set_batching | |
1388 | */ | |
a06e05e6 | 1389 | ioc_set_batching(q, current->io_context); |
05caf8db | 1390 | |
a06e05e6 TH |
1391 | spin_lock_irq(q->queue_lock); |
1392 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1393 | |
a06e05e6 | 1394 | goto retry; |
1da177e4 LT |
1395 | } |
1396 | ||
cd6ce148 BVA |
1397 | static struct request *blk_old_get_request(struct request_queue *q, |
1398 | unsigned int op, gfp_t gfp_mask) | |
1da177e4 LT |
1399 | { |
1400 | struct request *rq; | |
1401 | ||
332ebbf7 BVA |
1402 | WARN_ON_ONCE(q->mq_ops); |
1403 | ||
7f4b35d1 TH |
1404 | /* create ioc upfront */ |
1405 | create_io_context(gfp_mask, q->node); | |
1406 | ||
d6344532 | 1407 | spin_lock_irq(q->queue_lock); |
cd6ce148 | 1408 | rq = get_request(q, op, NULL, gfp_mask); |
0c4de0f3 | 1409 | if (IS_ERR(rq)) { |
da8303c6 | 1410 | spin_unlock_irq(q->queue_lock); |
0c4de0f3 CH |
1411 | return rq; |
1412 | } | |
1da177e4 | 1413 | |
0c4de0f3 CH |
1414 | /* q->queue_lock is unlocked at this point */ |
1415 | rq->__data_len = 0; | |
1416 | rq->__sector = (sector_t) -1; | |
1417 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1418 | return rq; |
1419 | } | |
320ae51f | 1420 | |
cd6ce148 BVA |
1421 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
1422 | gfp_t gfp_mask) | |
320ae51f | 1423 | { |
d280bab3 BVA |
1424 | struct request *req; |
1425 | ||
1426 | if (q->mq_ops) { | |
1427 | req = blk_mq_alloc_request(q, op, | |
6f3b0e8b CH |
1428 | (gfp_mask & __GFP_DIRECT_RECLAIM) ? |
1429 | 0 : BLK_MQ_REQ_NOWAIT); | |
d280bab3 BVA |
1430 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) |
1431 | q->mq_ops->initialize_rq_fn(req); | |
1432 | } else { | |
1433 | req = blk_old_get_request(q, op, gfp_mask); | |
1434 | if (!IS_ERR(req) && q->initialize_rq_fn) | |
1435 | q->initialize_rq_fn(req); | |
1436 | } | |
1437 | ||
1438 | return req; | |
320ae51f | 1439 | } |
1da177e4 LT |
1440 | EXPORT_SYMBOL(blk_get_request); |
1441 | ||
1442 | /** | |
1443 | * blk_requeue_request - put a request back on queue | |
1444 | * @q: request queue where request should be inserted | |
1445 | * @rq: request to be inserted | |
1446 | * | |
1447 | * Description: | |
1448 | * Drivers often keep queueing requests until the hardware cannot accept | |
1449 | * more, when that condition happens we need to put the request back | |
1450 | * on the queue. Must be called with queue lock held. | |
1451 | */ | |
165125e1 | 1452 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1453 | { |
2fff8a92 | 1454 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1455 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1456 | |
242f9dcb JA |
1457 | blk_delete_timer(rq); |
1458 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1459 | trace_block_rq_requeue(q, rq); |
87760e5e | 1460 | wbt_requeue(q->rq_wb, &rq->issue_stat); |
2056a782 | 1461 | |
e8064021 | 1462 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1463 | blk_queue_end_tag(q, rq); |
1464 | ||
ba396a6c JB |
1465 | BUG_ON(blk_queued_rq(rq)); |
1466 | ||
1da177e4 LT |
1467 | elv_requeue_request(q, rq); |
1468 | } | |
1da177e4 LT |
1469 | EXPORT_SYMBOL(blk_requeue_request); |
1470 | ||
73c10101 JA |
1471 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1472 | int where) | |
1473 | { | |
320ae51f | 1474 | blk_account_io_start(rq, true); |
7eaceacc | 1475 | __elv_add_request(q, rq, where); |
73c10101 JA |
1476 | } |
1477 | ||
d62e26b3 | 1478 | static void part_round_stats_single(struct request_queue *q, int cpu, |
b8d62b3a JA |
1479 | struct hd_struct *part, unsigned long now, |
1480 | unsigned int inflight) | |
074a7aca | 1481 | { |
b8d62b3a | 1482 | if (inflight) { |
074a7aca | 1483 | __part_stat_add(cpu, part, time_in_queue, |
b8d62b3a | 1484 | inflight * (now - part->stamp)); |
074a7aca TH |
1485 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1486 | } | |
1487 | part->stamp = now; | |
1488 | } | |
1489 | ||
1490 | /** | |
496aa8a9 | 1491 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
d62e26b3 | 1492 | * @q: target block queue |
496aa8a9 RD |
1493 | * @cpu: cpu number for stats access |
1494 | * @part: target partition | |
1da177e4 LT |
1495 | * |
1496 | * The average IO queue length and utilisation statistics are maintained | |
1497 | * by observing the current state of the queue length and the amount of | |
1498 | * time it has been in this state for. | |
1499 | * | |
1500 | * Normally, that accounting is done on IO completion, but that can result | |
1501 | * in more than a second's worth of IO being accounted for within any one | |
1502 | * second, leading to >100% utilisation. To deal with that, we call this | |
1503 | * function to do a round-off before returning the results when reading | |
1504 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1505 | * the current jiffies and restarts the counters again. | |
1506 | */ | |
d62e26b3 | 1507 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) |
6f2576af | 1508 | { |
b8d62b3a | 1509 | struct hd_struct *part2 = NULL; |
6f2576af | 1510 | unsigned long now = jiffies; |
b8d62b3a JA |
1511 | unsigned int inflight[2]; |
1512 | int stats = 0; | |
1513 | ||
1514 | if (part->stamp != now) | |
1515 | stats |= 1; | |
1516 | ||
1517 | if (part->partno) { | |
1518 | part2 = &part_to_disk(part)->part0; | |
1519 | if (part2->stamp != now) | |
1520 | stats |= 2; | |
1521 | } | |
1522 | ||
1523 | if (!stats) | |
1524 | return; | |
1525 | ||
1526 | part_in_flight(q, part, inflight); | |
6f2576af | 1527 | |
b8d62b3a JA |
1528 | if (stats & 2) |
1529 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1530 | if (stats & 1) | |
1531 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
6f2576af | 1532 | } |
074a7aca | 1533 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1534 | |
47fafbc7 | 1535 | #ifdef CONFIG_PM |
c8158819 LM |
1536 | static void blk_pm_put_request(struct request *rq) |
1537 | { | |
e8064021 | 1538 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1539 | pm_runtime_mark_last_busy(rq->q->dev); |
1540 | } | |
1541 | #else | |
1542 | static inline void blk_pm_put_request(struct request *rq) {} | |
1543 | #endif | |
1544 | ||
165125e1 | 1545 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1546 | { |
e8064021 CH |
1547 | req_flags_t rq_flags = req->rq_flags; |
1548 | ||
1da177e4 LT |
1549 | if (unlikely(!q)) |
1550 | return; | |
1da177e4 | 1551 | |
6f5ba581 CH |
1552 | if (q->mq_ops) { |
1553 | blk_mq_free_request(req); | |
1554 | return; | |
1555 | } | |
1556 | ||
2fff8a92 BVA |
1557 | lockdep_assert_held(q->queue_lock); |
1558 | ||
c8158819 LM |
1559 | blk_pm_put_request(req); |
1560 | ||
8922e16c TH |
1561 | elv_completed_request(q, req); |
1562 | ||
1cd96c24 BH |
1563 | /* this is a bio leak */ |
1564 | WARN_ON(req->bio != NULL); | |
1565 | ||
87760e5e JA |
1566 | wbt_done(q->rq_wb, &req->issue_stat); |
1567 | ||
1da177e4 LT |
1568 | /* |
1569 | * Request may not have originated from ll_rw_blk. if not, | |
1570 | * it didn't come out of our reserved rq pools | |
1571 | */ | |
e8064021 | 1572 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1573 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1574 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1575 | |
1da177e4 | 1576 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1577 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1578 | |
a051661c | 1579 | blk_free_request(rl, req); |
e8064021 | 1580 | freed_request(rl, sync, rq_flags); |
a051661c | 1581 | blk_put_rl(rl); |
1da177e4 LT |
1582 | } |
1583 | } | |
6e39b69e MC |
1584 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1585 | ||
1da177e4 LT |
1586 | void blk_put_request(struct request *req) |
1587 | { | |
165125e1 | 1588 | struct request_queue *q = req->q; |
8922e16c | 1589 | |
320ae51f JA |
1590 | if (q->mq_ops) |
1591 | blk_mq_free_request(req); | |
1592 | else { | |
1593 | unsigned long flags; | |
1594 | ||
1595 | spin_lock_irqsave(q->queue_lock, flags); | |
1596 | __blk_put_request(q, req); | |
1597 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1598 | } | |
1da177e4 | 1599 | } |
1da177e4 LT |
1600 | EXPORT_SYMBOL(blk_put_request); |
1601 | ||
320ae51f JA |
1602 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1603 | struct bio *bio) | |
73c10101 | 1604 | { |
1eff9d32 | 1605 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1606 | |
73c10101 JA |
1607 | if (!ll_back_merge_fn(q, req, bio)) |
1608 | return false; | |
1609 | ||
8c1cf6bb | 1610 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1611 | |
1612 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1613 | blk_rq_set_mixed_merge(req); | |
1614 | ||
1615 | req->biotail->bi_next = bio; | |
1616 | req->biotail = bio; | |
4f024f37 | 1617 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1618 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1619 | ||
320ae51f | 1620 | blk_account_io_start(req, false); |
73c10101 JA |
1621 | return true; |
1622 | } | |
1623 | ||
320ae51f JA |
1624 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1625 | struct bio *bio) | |
73c10101 | 1626 | { |
1eff9d32 | 1627 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1628 | |
73c10101 JA |
1629 | if (!ll_front_merge_fn(q, req, bio)) |
1630 | return false; | |
1631 | ||
8c1cf6bb | 1632 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1633 | |
1634 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1635 | blk_rq_set_mixed_merge(req); | |
1636 | ||
73c10101 JA |
1637 | bio->bi_next = req->bio; |
1638 | req->bio = bio; | |
1639 | ||
4f024f37 KO |
1640 | req->__sector = bio->bi_iter.bi_sector; |
1641 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1642 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1643 | ||
320ae51f | 1644 | blk_account_io_start(req, false); |
73c10101 JA |
1645 | return true; |
1646 | } | |
1647 | ||
1e739730 CH |
1648 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
1649 | struct bio *bio) | |
1650 | { | |
1651 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1652 | ||
1653 | if (segments >= queue_max_discard_segments(q)) | |
1654 | goto no_merge; | |
1655 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1656 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1657 | goto no_merge; | |
1658 | ||
1659 | req->biotail->bi_next = bio; | |
1660 | req->biotail = bio; | |
1661 | req->__data_len += bio->bi_iter.bi_size; | |
1662 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1663 | req->nr_phys_segments = segments + 1; | |
1664 | ||
1665 | blk_account_io_start(req, false); | |
1666 | return true; | |
1667 | no_merge: | |
1668 | req_set_nomerge(q, req); | |
1669 | return false; | |
1670 | } | |
1671 | ||
bd87b589 | 1672 | /** |
320ae51f | 1673 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1674 | * @q: request_queue new bio is being queued at |
1675 | * @bio: new bio being queued | |
1676 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1677 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1678 | * another request associated with @q is found on the plug list | |
1679 | * (optional, may be %NULL) | |
bd87b589 TH |
1680 | * |
1681 | * Determine whether @bio being queued on @q can be merged with a request | |
1682 | * on %current's plugged list. Returns %true if merge was successful, | |
1683 | * otherwise %false. | |
1684 | * | |
07c2bd37 TH |
1685 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1686 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1687 | * than scheduling, and the request, while may have elvpriv data, is not | |
1688 | * added on the elevator at this point. In addition, we don't have | |
1689 | * reliable access to the elevator outside queue lock. Only check basic | |
1690 | * merging parameters without querying the elevator. | |
da41a589 RE |
1691 | * |
1692 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1693 | */ |
320ae51f | 1694 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1695 | unsigned int *request_count, |
1696 | struct request **same_queue_rq) | |
73c10101 JA |
1697 | { |
1698 | struct blk_plug *plug; | |
1699 | struct request *rq; | |
92f399c7 | 1700 | struct list_head *plug_list; |
73c10101 | 1701 | |
bd87b589 | 1702 | plug = current->plug; |
73c10101 | 1703 | if (!plug) |
34fe7c05 | 1704 | return false; |
56ebdaf2 | 1705 | *request_count = 0; |
73c10101 | 1706 | |
92f399c7 SL |
1707 | if (q->mq_ops) |
1708 | plug_list = &plug->mq_list; | |
1709 | else | |
1710 | plug_list = &plug->list; | |
1711 | ||
1712 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 1713 | bool merged = false; |
73c10101 | 1714 | |
5b3f341f | 1715 | if (rq->q == q) { |
1b2e19f1 | 1716 | (*request_count)++; |
5b3f341f SL |
1717 | /* |
1718 | * Only blk-mq multiple hardware queues case checks the | |
1719 | * rq in the same queue, there should be only one such | |
1720 | * rq in a queue | |
1721 | **/ | |
1722 | if (same_queue_rq) | |
1723 | *same_queue_rq = rq; | |
1724 | } | |
56ebdaf2 | 1725 | |
07c2bd37 | 1726 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1727 | continue; |
1728 | ||
34fe7c05 CH |
1729 | switch (blk_try_merge(rq, bio)) { |
1730 | case ELEVATOR_BACK_MERGE: | |
1731 | merged = bio_attempt_back_merge(q, rq, bio); | |
1732 | break; | |
1733 | case ELEVATOR_FRONT_MERGE: | |
1734 | merged = bio_attempt_front_merge(q, rq, bio); | |
1735 | break; | |
1e739730 CH |
1736 | case ELEVATOR_DISCARD_MERGE: |
1737 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1738 | break; | |
34fe7c05 CH |
1739 | default: |
1740 | break; | |
73c10101 | 1741 | } |
34fe7c05 CH |
1742 | |
1743 | if (merged) | |
1744 | return true; | |
73c10101 | 1745 | } |
34fe7c05 CH |
1746 | |
1747 | return false; | |
73c10101 JA |
1748 | } |
1749 | ||
0809e3ac JM |
1750 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1751 | { | |
1752 | struct blk_plug *plug; | |
1753 | struct request *rq; | |
1754 | struct list_head *plug_list; | |
1755 | unsigned int ret = 0; | |
1756 | ||
1757 | plug = current->plug; | |
1758 | if (!plug) | |
1759 | goto out; | |
1760 | ||
1761 | if (q->mq_ops) | |
1762 | plug_list = &plug->mq_list; | |
1763 | else | |
1764 | plug_list = &plug->list; | |
1765 | ||
1766 | list_for_each_entry(rq, plug_list, queuelist) { | |
1767 | if (rq->q == q) | |
1768 | ret++; | |
1769 | } | |
1770 | out: | |
1771 | return ret; | |
1772 | } | |
1773 | ||
da8d7f07 | 1774 | void blk_init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1775 | { |
0be0dee6 BVA |
1776 | struct io_context *ioc = rq_ioc(bio); |
1777 | ||
1eff9d32 | 1778 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1779 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1780 | |
4f024f37 | 1781 | req->__sector = bio->bi_iter.bi_sector; |
5dc8b362 AM |
1782 | if (ioprio_valid(bio_prio(bio))) |
1783 | req->ioprio = bio_prio(bio); | |
0be0dee6 BVA |
1784 | else if (ioc) |
1785 | req->ioprio = ioc->ioprio; | |
1786 | else | |
1787 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
cb6934f8 | 1788 | req->write_hint = bio->bi_write_hint; |
bc1c56fd | 1789 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 | 1790 | } |
da8d7f07 | 1791 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); |
52d9e675 | 1792 | |
dece1635 | 1793 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1794 | { |
73c10101 | 1795 | struct blk_plug *plug; |
34fe7c05 | 1796 | int where = ELEVATOR_INSERT_SORT; |
e4d750c9 | 1797 | struct request *req, *free; |
56ebdaf2 | 1798 | unsigned int request_count = 0; |
87760e5e | 1799 | unsigned int wb_acct; |
1da177e4 | 1800 | |
1da177e4 LT |
1801 | /* |
1802 | * low level driver can indicate that it wants pages above a | |
1803 | * certain limit bounced to low memory (ie for highmem, or even | |
1804 | * ISA dma in theory) | |
1805 | */ | |
1806 | blk_queue_bounce(q, &bio); | |
1807 | ||
af67c31f | 1808 | blk_queue_split(q, &bio); |
23688bf4 | 1809 | |
e23947bd | 1810 | if (!bio_integrity_prep(bio)) |
dece1635 | 1811 | return BLK_QC_T_NONE; |
ffecfd1a | 1812 | |
f73f44eb | 1813 | if (op_is_flush(bio->bi_opf)) { |
73c10101 | 1814 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1815 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1816 | goto get_rq; |
1817 | } | |
1818 | ||
73c10101 JA |
1819 | /* |
1820 | * Check if we can merge with the plugged list before grabbing | |
1821 | * any locks. | |
1822 | */ | |
0809e3ac JM |
1823 | if (!blk_queue_nomerges(q)) { |
1824 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 1825 | return BLK_QC_T_NONE; |
0809e3ac JM |
1826 | } else |
1827 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 1828 | |
73c10101 | 1829 | spin_lock_irq(q->queue_lock); |
2056a782 | 1830 | |
34fe7c05 CH |
1831 | switch (elv_merge(q, &req, bio)) { |
1832 | case ELEVATOR_BACK_MERGE: | |
1833 | if (!bio_attempt_back_merge(q, req, bio)) | |
1834 | break; | |
1835 | elv_bio_merged(q, req, bio); | |
1836 | free = attempt_back_merge(q, req); | |
1837 | if (free) | |
1838 | __blk_put_request(q, free); | |
1839 | else | |
1840 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1841 | goto out_unlock; | |
1842 | case ELEVATOR_FRONT_MERGE: | |
1843 | if (!bio_attempt_front_merge(q, req, bio)) | |
1844 | break; | |
1845 | elv_bio_merged(q, req, bio); | |
1846 | free = attempt_front_merge(q, req); | |
1847 | if (free) | |
1848 | __blk_put_request(q, free); | |
1849 | else | |
1850 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1851 | goto out_unlock; | |
1852 | default: | |
1853 | break; | |
1da177e4 LT |
1854 | } |
1855 | ||
450991bc | 1856 | get_rq: |
87760e5e JA |
1857 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); |
1858 | ||
1da177e4 | 1859 | /* |
450991bc | 1860 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1861 | * Returns with the queue unlocked. |
450991bc | 1862 | */ |
ef295ecf | 1863 | req = get_request(q, bio->bi_opf, bio, GFP_NOIO); |
a492f075 | 1864 | if (IS_ERR(req)) { |
87760e5e | 1865 | __wbt_done(q->rq_wb, wb_acct); |
4e4cbee9 CH |
1866 | if (PTR_ERR(req) == -ENOMEM) |
1867 | bio->bi_status = BLK_STS_RESOURCE; | |
1868 | else | |
1869 | bio->bi_status = BLK_STS_IOERR; | |
4246a0b6 | 1870 | bio_endio(bio); |
da8303c6 TH |
1871 | goto out_unlock; |
1872 | } | |
d6344532 | 1873 | |
87760e5e JA |
1874 | wbt_track(&req->issue_stat, wb_acct); |
1875 | ||
450991bc NP |
1876 | /* |
1877 | * After dropping the lock and possibly sleeping here, our request | |
1878 | * may now be mergeable after it had proven unmergeable (above). | |
1879 | * We don't worry about that case for efficiency. It won't happen | |
1880 | * often, and the elevators are able to handle it. | |
1da177e4 | 1881 | */ |
da8d7f07 | 1882 | blk_init_request_from_bio(req, bio); |
1da177e4 | 1883 | |
9562ad9a | 1884 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1885 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1886 | |
1887 | plug = current->plug; | |
721a9602 | 1888 | if (plug) { |
dc6d36c9 JA |
1889 | /* |
1890 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1891 | * of a plug trace. |
0a6219a9 ML |
1892 | * |
1893 | * @request_count may become stale because of schedule | |
1894 | * out, so check plug list again. | |
dc6d36c9 | 1895 | */ |
0a6219a9 | 1896 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 1897 | trace_block_plug(q); |
3540d5e8 | 1898 | else { |
50d24c34 SL |
1899 | struct request *last = list_entry_rq(plug->list.prev); |
1900 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
1901 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 1902 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1903 | trace_block_plug(q); |
1904 | } | |
73c10101 | 1905 | } |
73c10101 | 1906 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 1907 | blk_account_io_start(req, true); |
73c10101 JA |
1908 | } else { |
1909 | spin_lock_irq(q->queue_lock); | |
1910 | add_acct_request(q, req, where); | |
24ecfbe2 | 1911 | __blk_run_queue(q); |
73c10101 JA |
1912 | out_unlock: |
1913 | spin_unlock_irq(q->queue_lock); | |
1914 | } | |
dece1635 JA |
1915 | |
1916 | return BLK_QC_T_NONE; | |
1da177e4 LT |
1917 | } |
1918 | ||
1da177e4 LT |
1919 | static void handle_bad_sector(struct bio *bio) |
1920 | { | |
1921 | char b[BDEVNAME_SIZE]; | |
1922 | ||
1923 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 1924 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 1925 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 1926 | (unsigned long long)bio_end_sector(bio), |
74d46992 | 1927 | (long long)get_capacity(bio->bi_disk)); |
1da177e4 LT |
1928 | } |
1929 | ||
c17bb495 AM |
1930 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1931 | ||
1932 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1933 | ||
1934 | static int __init setup_fail_make_request(char *str) | |
1935 | { | |
1936 | return setup_fault_attr(&fail_make_request, str); | |
1937 | } | |
1938 | __setup("fail_make_request=", setup_fail_make_request); | |
1939 | ||
b2c9cd37 | 1940 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1941 | { |
b2c9cd37 | 1942 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1943 | } |
1944 | ||
1945 | static int __init fail_make_request_debugfs(void) | |
1946 | { | |
dd48c085 AM |
1947 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1948 | NULL, &fail_make_request); | |
1949 | ||
21f9fcd8 | 1950 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
1951 | } |
1952 | ||
1953 | late_initcall(fail_make_request_debugfs); | |
1954 | ||
1955 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1956 | ||
b2c9cd37 AM |
1957 | static inline bool should_fail_request(struct hd_struct *part, |
1958 | unsigned int bytes) | |
c17bb495 | 1959 | { |
b2c9cd37 | 1960 | return false; |
c17bb495 AM |
1961 | } |
1962 | ||
1963 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1964 | ||
74d46992 CH |
1965 | /* |
1966 | * Remap block n of partition p to block n+start(p) of the disk. | |
1967 | */ | |
1968 | static inline int blk_partition_remap(struct bio *bio) | |
1969 | { | |
1970 | struct hd_struct *p; | |
1971 | int ret = 0; | |
1972 | ||
1973 | /* | |
1974 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
1975 | * Include a test for the reset op code and perform the remap if needed. | |
1976 | */ | |
1977 | if (!bio->bi_partno || | |
1978 | (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET)) | |
1979 | return 0; | |
1980 | ||
1981 | rcu_read_lock(); | |
1982 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
1983 | if (likely(p && !should_fail_request(p, bio->bi_iter.bi_size))) { | |
1984 | bio->bi_iter.bi_sector += p->start_sect; | |
1985 | bio->bi_partno = 0; | |
1986 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), | |
1987 | bio->bi_iter.bi_sector - p->start_sect); | |
1988 | } else { | |
1989 | printk("%s: fail for partition %d\n", __func__, bio->bi_partno); | |
1990 | ret = -EIO; | |
1991 | } | |
1992 | rcu_read_unlock(); | |
1993 | ||
1994 | return ret; | |
1995 | } | |
1996 | ||
c07e2b41 JA |
1997 | /* |
1998 | * Check whether this bio extends beyond the end of the device. | |
1999 | */ | |
2000 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
2001 | { | |
2002 | sector_t maxsector; | |
2003 | ||
2004 | if (!nr_sectors) | |
2005 | return 0; | |
2006 | ||
2007 | /* Test device or partition size, when known. */ | |
74d46992 | 2008 | maxsector = get_capacity(bio->bi_disk); |
c07e2b41 | 2009 | if (maxsector) { |
4f024f37 | 2010 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
2011 | |
2012 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2013 | /* | |
2014 | * This may well happen - the kernel calls bread() | |
2015 | * without checking the size of the device, e.g., when | |
2016 | * mounting a device. | |
2017 | */ | |
2018 | handle_bad_sector(bio); | |
2019 | return 1; | |
2020 | } | |
2021 | } | |
2022 | ||
2023 | return 0; | |
2024 | } | |
2025 | ||
27a84d54 CH |
2026 | static noinline_for_stack bool |
2027 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 2028 | { |
165125e1 | 2029 | struct request_queue *q; |
5a7bbad2 | 2030 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 2031 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 2032 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
2033 | |
2034 | might_sleep(); | |
1da177e4 | 2035 | |
c07e2b41 JA |
2036 | if (bio_check_eod(bio, nr_sectors)) |
2037 | goto end_io; | |
1da177e4 | 2038 | |
74d46992 | 2039 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
2040 | if (unlikely(!q)) { |
2041 | printk(KERN_ERR | |
2042 | "generic_make_request: Trying to access " | |
2043 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 2044 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
2045 | goto end_io; |
2046 | } | |
c17bb495 | 2047 | |
03a07c92 GR |
2048 | /* |
2049 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2050 | * if queue is not a request based queue. | |
2051 | */ | |
2052 | ||
2053 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) | |
2054 | goto not_supported; | |
2055 | ||
74d46992 | 2056 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) |
5a7bbad2 | 2057 | goto end_io; |
2056a782 | 2058 | |
74d46992 CH |
2059 | if (blk_partition_remap(bio)) |
2060 | goto end_io; | |
2056a782 | 2061 | |
5a7bbad2 CH |
2062 | if (bio_check_eod(bio, nr_sectors)) |
2063 | goto end_io; | |
1e87901e | 2064 | |
5a7bbad2 CH |
2065 | /* |
2066 | * Filter flush bio's early so that make_request based | |
2067 | * drivers without flush support don't have to worry | |
2068 | * about them. | |
2069 | */ | |
f3a8ab7d | 2070 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 2071 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 2072 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 2073 | if (!nr_sectors) { |
4e4cbee9 | 2074 | status = BLK_STS_OK; |
51fd77bd JA |
2075 | goto end_io; |
2076 | } | |
5a7bbad2 | 2077 | } |
5ddfe969 | 2078 | |
288dab8a CH |
2079 | switch (bio_op(bio)) { |
2080 | case REQ_OP_DISCARD: | |
2081 | if (!blk_queue_discard(q)) | |
2082 | goto not_supported; | |
2083 | break; | |
2084 | case REQ_OP_SECURE_ERASE: | |
2085 | if (!blk_queue_secure_erase(q)) | |
2086 | goto not_supported; | |
2087 | break; | |
2088 | case REQ_OP_WRITE_SAME: | |
74d46992 | 2089 | if (!q->limits.max_write_same_sectors) |
288dab8a | 2090 | goto not_supported; |
58886785 | 2091 | break; |
2d253440 ST |
2092 | case REQ_OP_ZONE_REPORT: |
2093 | case REQ_OP_ZONE_RESET: | |
74d46992 | 2094 | if (!blk_queue_is_zoned(q)) |
2d253440 | 2095 | goto not_supported; |
288dab8a | 2096 | break; |
a6f0788e | 2097 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 2098 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
2099 | goto not_supported; |
2100 | break; | |
288dab8a CH |
2101 | default: |
2102 | break; | |
5a7bbad2 | 2103 | } |
01edede4 | 2104 | |
7f4b35d1 TH |
2105 | /* |
2106 | * Various block parts want %current->io_context and lazy ioc | |
2107 | * allocation ends up trading a lot of pain for a small amount of | |
2108 | * memory. Just allocate it upfront. This may fail and block | |
2109 | * layer knows how to live with it. | |
2110 | */ | |
2111 | create_io_context(GFP_ATOMIC, q->node); | |
2112 | ||
ae118896 TH |
2113 | if (!blkcg_bio_issue_check(q, bio)) |
2114 | return false; | |
27a84d54 | 2115 | |
fbbaf700 N |
2116 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
2117 | trace_block_bio_queue(q, bio); | |
2118 | /* Now that enqueuing has been traced, we need to trace | |
2119 | * completion as well. | |
2120 | */ | |
2121 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2122 | } | |
27a84d54 | 2123 | return true; |
a7384677 | 2124 | |
288dab8a | 2125 | not_supported: |
4e4cbee9 | 2126 | status = BLK_STS_NOTSUPP; |
a7384677 | 2127 | end_io: |
4e4cbee9 | 2128 | bio->bi_status = status; |
4246a0b6 | 2129 | bio_endio(bio); |
27a84d54 | 2130 | return false; |
1da177e4 LT |
2131 | } |
2132 | ||
27a84d54 CH |
2133 | /** |
2134 | * generic_make_request - hand a buffer to its device driver for I/O | |
2135 | * @bio: The bio describing the location in memory and on the device. | |
2136 | * | |
2137 | * generic_make_request() is used to make I/O requests of block | |
2138 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2139 | * to be done. | |
2140 | * | |
2141 | * generic_make_request() does not return any status. The | |
2142 | * success/failure status of the request, along with notification of | |
2143 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2144 | * function described (one day) else where. | |
2145 | * | |
2146 | * The caller of generic_make_request must make sure that bi_io_vec | |
2147 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2148 | * set to describe the device address, and the | |
2149 | * bi_end_io and optionally bi_private are set to describe how | |
2150 | * completion notification should be signaled. | |
2151 | * | |
2152 | * generic_make_request and the drivers it calls may use bi_next if this | |
2153 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2154 | * a lower device by calling into generic_make_request recursively, which | |
2155 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 2156 | */ |
dece1635 | 2157 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 2158 | { |
f5fe1b51 N |
2159 | /* |
2160 | * bio_list_on_stack[0] contains bios submitted by the current | |
2161 | * make_request_fn. | |
2162 | * bio_list_on_stack[1] contains bios that were submitted before | |
2163 | * the current make_request_fn, but that haven't been processed | |
2164 | * yet. | |
2165 | */ | |
2166 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 2167 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 2168 | |
27a84d54 | 2169 | if (!generic_make_request_checks(bio)) |
dece1635 | 2170 | goto out; |
27a84d54 CH |
2171 | |
2172 | /* | |
2173 | * We only want one ->make_request_fn to be active at a time, else | |
2174 | * stack usage with stacked devices could be a problem. So use | |
2175 | * current->bio_list to keep a list of requests submited by a | |
2176 | * make_request_fn function. current->bio_list is also used as a | |
2177 | * flag to say if generic_make_request is currently active in this | |
2178 | * task or not. If it is NULL, then no make_request is active. If | |
2179 | * it is non-NULL, then a make_request is active, and new requests | |
2180 | * should be added at the tail | |
2181 | */ | |
bddd87c7 | 2182 | if (current->bio_list) { |
f5fe1b51 | 2183 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 2184 | goto out; |
d89d8796 | 2185 | } |
27a84d54 | 2186 | |
d89d8796 NB |
2187 | /* following loop may be a bit non-obvious, and so deserves some |
2188 | * explanation. | |
2189 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2190 | * ensure that) so we have a list with a single bio. | |
2191 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2192 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2193 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2194 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2195 | * through a recursive call to generic_make_request. If it |
2196 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2197 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2198 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2199 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2200 | */ |
2201 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
2202 | bio_list_init(&bio_list_on_stack[0]); |
2203 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 2204 | do { |
74d46992 | 2205 | struct request_queue *q = bio->bi_disk->queue; |
27a84d54 | 2206 | |
03a07c92 | 2207 | if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) { |
79bd9959 N |
2208 | struct bio_list lower, same; |
2209 | ||
2210 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
2211 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
2212 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 2213 | ret = q->make_request_fn(q, bio); |
3ef28e83 DW |
2214 | |
2215 | blk_queue_exit(q); | |
27a84d54 | 2216 | |
79bd9959 N |
2217 | /* sort new bios into those for a lower level |
2218 | * and those for the same level | |
2219 | */ | |
2220 | bio_list_init(&lower); | |
2221 | bio_list_init(&same); | |
f5fe1b51 | 2222 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 2223 | if (q == bio->bi_disk->queue) |
79bd9959 N |
2224 | bio_list_add(&same, bio); |
2225 | else | |
2226 | bio_list_add(&lower, bio); | |
2227 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
2228 | bio_list_merge(&bio_list_on_stack[0], &lower); |
2229 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2230 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 2231 | } else { |
03a07c92 GR |
2232 | if (unlikely(!blk_queue_dying(q) && |
2233 | (bio->bi_opf & REQ_NOWAIT))) | |
2234 | bio_wouldblock_error(bio); | |
2235 | else | |
2236 | bio_io_error(bio); | |
3ef28e83 | 2237 | } |
f5fe1b51 | 2238 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 2239 | } while (bio); |
bddd87c7 | 2240 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2241 | |
2242 | out: | |
2243 | return ret; | |
d89d8796 | 2244 | } |
1da177e4 LT |
2245 | EXPORT_SYMBOL(generic_make_request); |
2246 | ||
f421e1d9 CH |
2247 | /** |
2248 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
2249 | * @bio: The bio describing the location in memory and on the device. | |
2250 | * | |
2251 | * This function behaves like generic_make_request(), but does not protect | |
2252 | * against recursion. Must only be used if the called driver is known | |
2253 | * to not call generic_make_request (or direct_make_request) again from | |
2254 | * its make_request function. (Calling direct_make_request again from | |
2255 | * a workqueue is perfectly fine as that doesn't recurse). | |
2256 | */ | |
2257 | blk_qc_t direct_make_request(struct bio *bio) | |
2258 | { | |
2259 | struct request_queue *q = bio->bi_disk->queue; | |
2260 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
2261 | blk_qc_t ret; | |
2262 | ||
2263 | if (!generic_make_request_checks(bio)) | |
2264 | return BLK_QC_T_NONE; | |
2265 | ||
2266 | if (unlikely(blk_queue_enter(q, nowait))) { | |
2267 | if (nowait && !blk_queue_dying(q)) | |
2268 | bio->bi_status = BLK_STS_AGAIN; | |
2269 | else | |
2270 | bio->bi_status = BLK_STS_IOERR; | |
2271 | bio_endio(bio); | |
2272 | return BLK_QC_T_NONE; | |
2273 | } | |
2274 | ||
2275 | ret = q->make_request_fn(q, bio); | |
2276 | blk_queue_exit(q); | |
2277 | return ret; | |
2278 | } | |
2279 | EXPORT_SYMBOL_GPL(direct_make_request); | |
2280 | ||
1da177e4 | 2281 | /** |
710027a4 | 2282 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2283 | * @bio: The &struct bio which describes the I/O |
2284 | * | |
2285 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2286 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2287 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2288 | * |
2289 | */ | |
4e49ea4a | 2290 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2291 | { |
bf2de6f5 JA |
2292 | /* |
2293 | * If it's a regular read/write or a barrier with data attached, | |
2294 | * go through the normal accounting stuff before submission. | |
2295 | */ | |
e2a60da7 | 2296 | if (bio_has_data(bio)) { |
4363ac7c MP |
2297 | unsigned int count; |
2298 | ||
95fe6c1a | 2299 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
74d46992 | 2300 | count = queue_logical_block_size(bio->bi_disk->queue); |
4363ac7c MP |
2301 | else |
2302 | count = bio_sectors(bio); | |
2303 | ||
a8ebb056 | 2304 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2305 | count_vm_events(PGPGOUT, count); |
2306 | } else { | |
4f024f37 | 2307 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2308 | count_vm_events(PGPGIN, count); |
2309 | } | |
2310 | ||
2311 | if (unlikely(block_dump)) { | |
2312 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2313 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2314 | current->comm, task_pid_nr(current), |
a8ebb056 | 2315 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2316 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 2317 | bio_devname(bio, b), count); |
bf2de6f5 | 2318 | } |
1da177e4 LT |
2319 | } |
2320 | ||
dece1635 | 2321 | return generic_make_request(bio); |
1da177e4 | 2322 | } |
1da177e4 LT |
2323 | EXPORT_SYMBOL(submit_bio); |
2324 | ||
ea435e1b CH |
2325 | bool blk_poll(struct request_queue *q, blk_qc_t cookie) |
2326 | { | |
2327 | if (!q->poll_fn || !blk_qc_t_valid(cookie)) | |
2328 | return false; | |
2329 | ||
2330 | if (current->plug) | |
2331 | blk_flush_plug_list(current->plug, false); | |
2332 | return q->poll_fn(q, cookie); | |
2333 | } | |
2334 | EXPORT_SYMBOL_GPL(blk_poll); | |
2335 | ||
82124d60 | 2336 | /** |
bf4e6b4e HR |
2337 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2338 | * for new the queue limits | |
82124d60 KU |
2339 | * @q: the queue |
2340 | * @rq: the request being checked | |
2341 | * | |
2342 | * Description: | |
2343 | * @rq may have been made based on weaker limitations of upper-level queues | |
2344 | * in request stacking drivers, and it may violate the limitation of @q. | |
2345 | * Since the block layer and the underlying device driver trust @rq | |
2346 | * after it is inserted to @q, it should be checked against @q before | |
2347 | * the insertion using this generic function. | |
2348 | * | |
82124d60 | 2349 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2350 | * limits when retrying requests on other queues. Those requests need |
2351 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2352 | */ |
bf4e6b4e HR |
2353 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2354 | struct request *rq) | |
82124d60 | 2355 | { |
8fe0d473 | 2356 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2357 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2358 | return -EIO; | |
2359 | } | |
2360 | ||
2361 | /* | |
2362 | * queue's settings related to segment counting like q->bounce_pfn | |
2363 | * may differ from that of other stacking queues. | |
2364 | * Recalculate it to check the request correctly on this queue's | |
2365 | * limitation. | |
2366 | */ | |
2367 | blk_recalc_rq_segments(rq); | |
8a78362c | 2368 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2369 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2370 | return -EIO; | |
2371 | } | |
2372 | ||
2373 | return 0; | |
2374 | } | |
82124d60 KU |
2375 | |
2376 | /** | |
2377 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2378 | * @q: the queue to submit the request | |
2379 | * @rq: the request being queued | |
2380 | */ | |
2a842aca | 2381 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 KU |
2382 | { |
2383 | unsigned long flags; | |
4853abaa | 2384 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2385 | |
bf4e6b4e | 2386 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 2387 | return BLK_STS_IOERR; |
82124d60 | 2388 | |
b2c9cd37 AM |
2389 | if (rq->rq_disk && |
2390 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 2391 | return BLK_STS_IOERR; |
82124d60 | 2392 | |
7fb4898e KB |
2393 | if (q->mq_ops) { |
2394 | if (blk_queue_io_stat(q)) | |
2395 | blk_account_io_start(rq, true); | |
157f377b JA |
2396 | /* |
2397 | * Since we have a scheduler attached on the top device, | |
2398 | * bypass a potential scheduler on the bottom device for | |
2399 | * insert. | |
2400 | */ | |
b0850297 | 2401 | blk_mq_request_bypass_insert(rq, true); |
2a842aca | 2402 | return BLK_STS_OK; |
7fb4898e KB |
2403 | } |
2404 | ||
82124d60 | 2405 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2406 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 | 2407 | spin_unlock_irqrestore(q->queue_lock, flags); |
2a842aca | 2408 | return BLK_STS_IOERR; |
8ba61435 | 2409 | } |
82124d60 KU |
2410 | |
2411 | /* | |
2412 | * Submitting request must be dequeued before calling this function | |
2413 | * because it will be linked to another request_queue | |
2414 | */ | |
2415 | BUG_ON(blk_queued_rq(rq)); | |
2416 | ||
f73f44eb | 2417 | if (op_is_flush(rq->cmd_flags)) |
4853abaa JM |
2418 | where = ELEVATOR_INSERT_FLUSH; |
2419 | ||
2420 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2421 | if (where == ELEVATOR_INSERT_FLUSH) |
2422 | __blk_run_queue(q); | |
82124d60 KU |
2423 | spin_unlock_irqrestore(q->queue_lock, flags); |
2424 | ||
2a842aca | 2425 | return BLK_STS_OK; |
82124d60 KU |
2426 | } |
2427 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2428 | ||
80a761fd TH |
2429 | /** |
2430 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2431 | * @rq: request to examine | |
2432 | * | |
2433 | * Description: | |
2434 | * A request could be merge of IOs which require different failure | |
2435 | * handling. This function determines the number of bytes which | |
2436 | * can be failed from the beginning of the request without | |
2437 | * crossing into area which need to be retried further. | |
2438 | * | |
2439 | * Return: | |
2440 | * The number of bytes to fail. | |
80a761fd TH |
2441 | */ |
2442 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2443 | { | |
2444 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2445 | unsigned int bytes = 0; | |
2446 | struct bio *bio; | |
2447 | ||
e8064021 | 2448 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2449 | return blk_rq_bytes(rq); |
2450 | ||
2451 | /* | |
2452 | * Currently the only 'mixing' which can happen is between | |
2453 | * different fastfail types. We can safely fail portions | |
2454 | * which have all the failfast bits that the first one has - | |
2455 | * the ones which are at least as eager to fail as the first | |
2456 | * one. | |
2457 | */ | |
2458 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2459 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2460 | break; |
4f024f37 | 2461 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2462 | } |
2463 | ||
2464 | /* this could lead to infinite loop */ | |
2465 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2466 | return bytes; | |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2469 | ||
320ae51f | 2470 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2471 | { |
c2553b58 | 2472 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2473 | const int rw = rq_data_dir(req); |
2474 | struct hd_struct *part; | |
2475 | int cpu; | |
2476 | ||
2477 | cpu = part_stat_lock(); | |
09e099d4 | 2478 | part = req->part; |
bc58ba94 JA |
2479 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2480 | part_stat_unlock(); | |
2481 | } | |
2482 | } | |
2483 | ||
320ae51f | 2484 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2485 | { |
bc58ba94 | 2486 | /* |
dd4c133f TH |
2487 | * Account IO completion. flush_rq isn't accounted as a |
2488 | * normal IO on queueing nor completion. Accounting the | |
2489 | * containing request is enough. | |
bc58ba94 | 2490 | */ |
e8064021 | 2491 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
bc58ba94 JA |
2492 | unsigned long duration = jiffies - req->start_time; |
2493 | const int rw = rq_data_dir(req); | |
2494 | struct hd_struct *part; | |
2495 | int cpu; | |
2496 | ||
2497 | cpu = part_stat_lock(); | |
09e099d4 | 2498 | part = req->part; |
bc58ba94 JA |
2499 | |
2500 | part_stat_inc(cpu, part, ios[rw]); | |
2501 | part_stat_add(cpu, part, ticks[rw], duration); | |
d62e26b3 JA |
2502 | part_round_stats(req->q, cpu, part); |
2503 | part_dec_in_flight(req->q, part, rw); | |
bc58ba94 | 2504 | |
6c23a968 | 2505 | hd_struct_put(part); |
bc58ba94 JA |
2506 | part_stat_unlock(); |
2507 | } | |
2508 | } | |
2509 | ||
47fafbc7 | 2510 | #ifdef CONFIG_PM |
c8158819 LM |
2511 | /* |
2512 | * Don't process normal requests when queue is suspended | |
2513 | * or in the process of suspending/resuming | |
2514 | */ | |
e4f36b24 | 2515 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2516 | { |
e4f36b24 CH |
2517 | switch (rq->q->rpm_status) { |
2518 | case RPM_RESUMING: | |
2519 | case RPM_SUSPENDING: | |
2520 | return rq->rq_flags & RQF_PM; | |
2521 | case RPM_SUSPENDED: | |
2522 | return false; | |
2523 | } | |
2524 | ||
2525 | return true; | |
c8158819 LM |
2526 | } |
2527 | #else | |
e4f36b24 | 2528 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2529 | { |
e4f36b24 | 2530 | return true; |
c8158819 LM |
2531 | } |
2532 | #endif | |
2533 | ||
320ae51f JA |
2534 | void blk_account_io_start(struct request *rq, bool new_io) |
2535 | { | |
2536 | struct hd_struct *part; | |
2537 | int rw = rq_data_dir(rq); | |
2538 | int cpu; | |
2539 | ||
2540 | if (!blk_do_io_stat(rq)) | |
2541 | return; | |
2542 | ||
2543 | cpu = part_stat_lock(); | |
2544 | ||
2545 | if (!new_io) { | |
2546 | part = rq->part; | |
2547 | part_stat_inc(cpu, part, merges[rw]); | |
2548 | } else { | |
2549 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2550 | if (!hd_struct_try_get(part)) { | |
2551 | /* | |
2552 | * The partition is already being removed, | |
2553 | * the request will be accounted on the disk only | |
2554 | * | |
2555 | * We take a reference on disk->part0 although that | |
2556 | * partition will never be deleted, so we can treat | |
2557 | * it as any other partition. | |
2558 | */ | |
2559 | part = &rq->rq_disk->part0; | |
2560 | hd_struct_get(part); | |
2561 | } | |
d62e26b3 JA |
2562 | part_round_stats(rq->q, cpu, part); |
2563 | part_inc_in_flight(rq->q, part, rw); | |
320ae51f JA |
2564 | rq->part = part; |
2565 | } | |
2566 | ||
2567 | part_stat_unlock(); | |
2568 | } | |
2569 | ||
9c988374 CH |
2570 | static struct request *elv_next_request(struct request_queue *q) |
2571 | { | |
2572 | struct request *rq; | |
2573 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
2574 | ||
2575 | WARN_ON_ONCE(q->mq_ops); | |
2576 | ||
2577 | while (1) { | |
e4f36b24 CH |
2578 | list_for_each_entry(rq, &q->queue_head, queuelist) { |
2579 | if (blk_pm_allow_request(rq)) | |
2580 | return rq; | |
2581 | ||
2582 | if (rq->rq_flags & RQF_SOFTBARRIER) | |
2583 | break; | |
9c988374 CH |
2584 | } |
2585 | ||
2586 | /* | |
2587 | * Flush request is running and flush request isn't queueable | |
2588 | * in the drive, we can hold the queue till flush request is | |
2589 | * finished. Even we don't do this, driver can't dispatch next | |
2590 | * requests and will requeue them. And this can improve | |
2591 | * throughput too. For example, we have request flush1, write1, | |
2592 | * flush 2. flush1 is dispatched, then queue is hold, write1 | |
2593 | * isn't inserted to queue. After flush1 is finished, flush2 | |
2594 | * will be dispatched. Since disk cache is already clean, | |
2595 | * flush2 will be finished very soon, so looks like flush2 is | |
2596 | * folded to flush1. | |
2597 | * Since the queue is hold, a flag is set to indicate the queue | |
2598 | * should be restarted later. Please see flush_end_io() for | |
2599 | * details. | |
2600 | */ | |
2601 | if (fq->flush_pending_idx != fq->flush_running_idx && | |
2602 | !queue_flush_queueable(q)) { | |
2603 | fq->flush_queue_delayed = 1; | |
2604 | return NULL; | |
2605 | } | |
2606 | if (unlikely(blk_queue_bypass(q)) || | |
2607 | !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0)) | |
2608 | return NULL; | |
2609 | } | |
2610 | } | |
2611 | ||
3bcddeac | 2612 | /** |
9934c8c0 TH |
2613 | * blk_peek_request - peek at the top of a request queue |
2614 | * @q: request queue to peek at | |
2615 | * | |
2616 | * Description: | |
2617 | * Return the request at the top of @q. The returned request | |
2618 | * should be started using blk_start_request() before LLD starts | |
2619 | * processing it. | |
2620 | * | |
2621 | * Return: | |
2622 | * Pointer to the request at the top of @q if available. Null | |
2623 | * otherwise. | |
9934c8c0 TH |
2624 | */ |
2625 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2626 | { |
2627 | struct request *rq; | |
2628 | int ret; | |
2629 | ||
2fff8a92 | 2630 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2631 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2632 | |
9c988374 | 2633 | while ((rq = elv_next_request(q)) != NULL) { |
e8064021 | 2634 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2635 | /* |
2636 | * This is the first time the device driver | |
2637 | * sees this request (possibly after | |
2638 | * requeueing). Notify IO scheduler. | |
2639 | */ | |
e8064021 | 2640 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2641 | elv_activate_rq(q, rq); |
2642 | ||
2643 | /* | |
2644 | * just mark as started even if we don't start | |
2645 | * it, a request that has been delayed should | |
2646 | * not be passed by new incoming requests | |
2647 | */ | |
e8064021 | 2648 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2649 | trace_block_rq_issue(q, rq); |
2650 | } | |
2651 | ||
2652 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2653 | q->end_sector = rq_end_sector(rq); | |
2654 | q->boundary_rq = NULL; | |
2655 | } | |
2656 | ||
e8064021 | 2657 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2658 | break; |
2659 | ||
2e46e8b2 | 2660 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2661 | /* |
2662 | * make sure space for the drain appears we | |
2663 | * know we can do this because max_hw_segments | |
2664 | * has been adjusted to be one fewer than the | |
2665 | * device can handle | |
2666 | */ | |
2667 | rq->nr_phys_segments++; | |
2668 | } | |
2669 | ||
2670 | if (!q->prep_rq_fn) | |
2671 | break; | |
2672 | ||
2673 | ret = q->prep_rq_fn(q, rq); | |
2674 | if (ret == BLKPREP_OK) { | |
2675 | break; | |
2676 | } else if (ret == BLKPREP_DEFER) { | |
2677 | /* | |
2678 | * the request may have been (partially) prepped. | |
2679 | * we need to keep this request in the front to | |
e8064021 | 2680 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2681 | * prevent other fs requests from passing this one. |
2682 | */ | |
2e46e8b2 | 2683 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2684 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2685 | /* |
2686 | * remove the space for the drain we added | |
2687 | * so that we don't add it again | |
2688 | */ | |
2689 | --rq->nr_phys_segments; | |
2690 | } | |
2691 | ||
2692 | rq = NULL; | |
2693 | break; | |
0fb5b1fb | 2694 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
e8064021 | 2695 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2696 | /* |
2697 | * Mark this request as started so we don't trigger | |
2698 | * any debug logic in the end I/O path. | |
2699 | */ | |
2700 | blk_start_request(rq); | |
2a842aca CH |
2701 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? |
2702 | BLK_STS_TARGET : BLK_STS_IOERR); | |
158dbda0 TH |
2703 | } else { |
2704 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2705 | break; | |
2706 | } | |
2707 | } | |
2708 | ||
2709 | return rq; | |
2710 | } | |
9934c8c0 | 2711 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2712 | |
5034435c | 2713 | static void blk_dequeue_request(struct request *rq) |
158dbda0 | 2714 | { |
9934c8c0 TH |
2715 | struct request_queue *q = rq->q; |
2716 | ||
158dbda0 TH |
2717 | BUG_ON(list_empty(&rq->queuelist)); |
2718 | BUG_ON(ELV_ON_HASH(rq)); | |
2719 | ||
2720 | list_del_init(&rq->queuelist); | |
2721 | ||
2722 | /* | |
2723 | * the time frame between a request being removed from the lists | |
2724 | * and to it is freed is accounted as io that is in progress at | |
2725 | * the driver side. | |
2726 | */ | |
9195291e | 2727 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2728 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2729 | set_io_start_time_ns(rq); |
2730 | } | |
158dbda0 TH |
2731 | } |
2732 | ||
9934c8c0 TH |
2733 | /** |
2734 | * blk_start_request - start request processing on the driver | |
2735 | * @req: request to dequeue | |
2736 | * | |
2737 | * Description: | |
2738 | * Dequeue @req and start timeout timer on it. This hands off the | |
2739 | * request to the driver. | |
9934c8c0 TH |
2740 | */ |
2741 | void blk_start_request(struct request *req) | |
2742 | { | |
2fff8a92 | 2743 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2744 | WARN_ON_ONCE(req->q->mq_ops); |
2fff8a92 | 2745 | |
9934c8c0 TH |
2746 | blk_dequeue_request(req); |
2747 | ||
cf43e6be | 2748 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
88eeca49 | 2749 | blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req)); |
cf43e6be | 2750 | req->rq_flags |= RQF_STATS; |
87760e5e | 2751 | wbt_issue(req->q->rq_wb, &req->issue_stat); |
cf43e6be JA |
2752 | } |
2753 | ||
4912aa6c | 2754 | BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags)); |
9934c8c0 TH |
2755 | blk_add_timer(req); |
2756 | } | |
2757 | EXPORT_SYMBOL(blk_start_request); | |
2758 | ||
2759 | /** | |
2760 | * blk_fetch_request - fetch a request from a request queue | |
2761 | * @q: request queue to fetch a request from | |
2762 | * | |
2763 | * Description: | |
2764 | * Return the request at the top of @q. The request is started on | |
2765 | * return and LLD can start processing it immediately. | |
2766 | * | |
2767 | * Return: | |
2768 | * Pointer to the request at the top of @q if available. Null | |
2769 | * otherwise. | |
9934c8c0 TH |
2770 | */ |
2771 | struct request *blk_fetch_request(struct request_queue *q) | |
2772 | { | |
2773 | struct request *rq; | |
2774 | ||
2fff8a92 | 2775 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2776 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2777 | |
9934c8c0 TH |
2778 | rq = blk_peek_request(q); |
2779 | if (rq) | |
2780 | blk_start_request(rq); | |
2781 | return rq; | |
2782 | } | |
2783 | EXPORT_SYMBOL(blk_fetch_request); | |
2784 | ||
ef71de8b CH |
2785 | /* |
2786 | * Steal bios from a request and add them to a bio list. | |
2787 | * The request must not have been partially completed before. | |
2788 | */ | |
2789 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
2790 | { | |
2791 | if (rq->bio) { | |
2792 | if (list->tail) | |
2793 | list->tail->bi_next = rq->bio; | |
2794 | else | |
2795 | list->head = rq->bio; | |
2796 | list->tail = rq->biotail; | |
2797 | ||
2798 | rq->bio = NULL; | |
2799 | rq->biotail = NULL; | |
2800 | } | |
2801 | ||
2802 | rq->__data_len = 0; | |
2803 | } | |
2804 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
2805 | ||
3bcddeac | 2806 | /** |
2e60e022 | 2807 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2808 | * @req: the request being processed |
2a842aca | 2809 | * @error: block status code |
8ebf9756 | 2810 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2811 | * |
2812 | * Description: | |
8ebf9756 RD |
2813 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2814 | * the request structure even if @req doesn't have leftover. | |
2815 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2816 | * |
2817 | * This special helper function is only for request stacking drivers | |
2818 | * (e.g. request-based dm) so that they can handle partial completion. | |
2819 | * Actual device drivers should use blk_end_request instead. | |
2820 | * | |
2821 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2822 | * %false return from this function. | |
3bcddeac KU |
2823 | * |
2824 | * Return: | |
2e60e022 TH |
2825 | * %false - this request doesn't have any more data |
2826 | * %true - this request has more data | |
3bcddeac | 2827 | **/ |
2a842aca CH |
2828 | bool blk_update_request(struct request *req, blk_status_t error, |
2829 | unsigned int nr_bytes) | |
1da177e4 | 2830 | { |
f79ea416 | 2831 | int total_bytes; |
1da177e4 | 2832 | |
2a842aca | 2833 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 2834 | |
2e60e022 TH |
2835 | if (!req->bio) |
2836 | return false; | |
2837 | ||
2a842aca CH |
2838 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
2839 | !(req->rq_flags & RQF_QUIET))) | |
2840 | print_req_error(req, error); | |
1da177e4 | 2841 | |
bc58ba94 | 2842 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2843 | |
f79ea416 KO |
2844 | total_bytes = 0; |
2845 | while (req->bio) { | |
2846 | struct bio *bio = req->bio; | |
4f024f37 | 2847 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2848 | |
4f024f37 | 2849 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2850 | req->bio = bio->bi_next; |
1da177e4 | 2851 | |
fbbaf700 N |
2852 | /* Completion has already been traced */ |
2853 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 2854 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2855 | |
f79ea416 KO |
2856 | total_bytes += bio_bytes; |
2857 | nr_bytes -= bio_bytes; | |
1da177e4 | 2858 | |
f79ea416 KO |
2859 | if (!nr_bytes) |
2860 | break; | |
1da177e4 LT |
2861 | } |
2862 | ||
2863 | /* | |
2864 | * completely done | |
2865 | */ | |
2e60e022 TH |
2866 | if (!req->bio) { |
2867 | /* | |
2868 | * Reset counters so that the request stacking driver | |
2869 | * can find how many bytes remain in the request | |
2870 | * later. | |
2871 | */ | |
a2dec7b3 | 2872 | req->__data_len = 0; |
2e60e022 TH |
2873 | return false; |
2874 | } | |
1da177e4 | 2875 | |
a2dec7b3 | 2876 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2877 | |
2878 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 2879 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 2880 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2881 | |
80a761fd | 2882 | /* mixed attributes always follow the first bio */ |
e8064021 | 2883 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 2884 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 2885 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
2886 | } |
2887 | ||
ed6565e7 CH |
2888 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
2889 | /* | |
2890 | * If total number of sectors is less than the first segment | |
2891 | * size, something has gone terribly wrong. | |
2892 | */ | |
2893 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2894 | blk_dump_rq_flags(req, "request botched"); | |
2895 | req->__data_len = blk_rq_cur_bytes(req); | |
2896 | } | |
2e46e8b2 | 2897 | |
ed6565e7 CH |
2898 | /* recalculate the number of segments */ |
2899 | blk_recalc_rq_segments(req); | |
2900 | } | |
2e46e8b2 | 2901 | |
2e60e022 | 2902 | return true; |
1da177e4 | 2903 | } |
2e60e022 | 2904 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2905 | |
2a842aca | 2906 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, |
2e60e022 TH |
2907 | unsigned int nr_bytes, |
2908 | unsigned int bidi_bytes) | |
5efccd17 | 2909 | { |
2e60e022 TH |
2910 | if (blk_update_request(rq, error, nr_bytes)) |
2911 | return true; | |
5efccd17 | 2912 | |
2e60e022 TH |
2913 | /* Bidi request must be completed as a whole */ |
2914 | if (unlikely(blk_bidi_rq(rq)) && | |
2915 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2916 | return true; | |
5efccd17 | 2917 | |
e2e1a148 JA |
2918 | if (blk_queue_add_random(rq->q)) |
2919 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2920 | |
2921 | return false; | |
1da177e4 LT |
2922 | } |
2923 | ||
28018c24 JB |
2924 | /** |
2925 | * blk_unprep_request - unprepare a request | |
2926 | * @req: the request | |
2927 | * | |
2928 | * This function makes a request ready for complete resubmission (or | |
2929 | * completion). It happens only after all error handling is complete, | |
2930 | * so represents the appropriate moment to deallocate any resources | |
2931 | * that were allocated to the request in the prep_rq_fn. The queue | |
2932 | * lock is held when calling this. | |
2933 | */ | |
2934 | void blk_unprep_request(struct request *req) | |
2935 | { | |
2936 | struct request_queue *q = req->q; | |
2937 | ||
e8064021 | 2938 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
2939 | if (q->unprep_rq_fn) |
2940 | q->unprep_rq_fn(q, req); | |
2941 | } | |
2942 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2943 | ||
2a842aca | 2944 | void blk_finish_request(struct request *req, blk_status_t error) |
1da177e4 | 2945 | { |
cf43e6be JA |
2946 | struct request_queue *q = req->q; |
2947 | ||
2fff8a92 | 2948 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2949 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2950 | |
cf43e6be | 2951 | if (req->rq_flags & RQF_STATS) |
34dbad5d | 2952 | blk_stat_add(req); |
cf43e6be | 2953 | |
e8064021 | 2954 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 2955 | blk_queue_end_tag(q, req); |
b8286239 | 2956 | |
ba396a6c | 2957 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2958 | |
57292b58 | 2959 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) |
dc3b17cc | 2960 | laptop_io_completion(req->q->backing_dev_info); |
1da177e4 | 2961 | |
e78042e5 MA |
2962 | blk_delete_timer(req); |
2963 | ||
e8064021 | 2964 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
2965 | blk_unprep_request(req); |
2966 | ||
bc58ba94 | 2967 | blk_account_io_done(req); |
b8286239 | 2968 | |
87760e5e JA |
2969 | if (req->end_io) { |
2970 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
8ffdc655 | 2971 | req->end_io(req, error); |
87760e5e | 2972 | } else { |
b8286239 KU |
2973 | if (blk_bidi_rq(req)) |
2974 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2975 | ||
cf43e6be | 2976 | __blk_put_request(q, req); |
b8286239 | 2977 | } |
1da177e4 | 2978 | } |
12120077 | 2979 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 2980 | |
3b11313a | 2981 | /** |
2e60e022 TH |
2982 | * blk_end_bidi_request - Complete a bidi request |
2983 | * @rq: the request to complete | |
2a842aca | 2984 | * @error: block status code |
2e60e022 TH |
2985 | * @nr_bytes: number of bytes to complete @rq |
2986 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2987 | * |
2988 | * Description: | |
e3a04fe3 | 2989 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2990 | * Drivers that supports bidi can safely call this member for any |
2991 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2992 | * just ignored. | |
336cdb40 KU |
2993 | * |
2994 | * Return: | |
2e60e022 TH |
2995 | * %false - we are done with this request |
2996 | * %true - still buffers pending for this request | |
a0cd1285 | 2997 | **/ |
2a842aca | 2998 | static bool blk_end_bidi_request(struct request *rq, blk_status_t error, |
32fab448 KU |
2999 | unsigned int nr_bytes, unsigned int bidi_bytes) |
3000 | { | |
336cdb40 | 3001 | struct request_queue *q = rq->q; |
2e60e022 | 3002 | unsigned long flags; |
32fab448 | 3003 | |
332ebbf7 BVA |
3004 | WARN_ON_ONCE(q->mq_ops); |
3005 | ||
2e60e022 TH |
3006 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3007 | return true; | |
32fab448 | 3008 | |
336cdb40 | 3009 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 3010 | blk_finish_request(rq, error); |
336cdb40 KU |
3011 | spin_unlock_irqrestore(q->queue_lock, flags); |
3012 | ||
2e60e022 | 3013 | return false; |
32fab448 KU |
3014 | } |
3015 | ||
336cdb40 | 3016 | /** |
2e60e022 TH |
3017 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
3018 | * @rq: the request to complete | |
2a842aca | 3019 | * @error: block status code |
e3a04fe3 KU |
3020 | * @nr_bytes: number of bytes to complete @rq |
3021 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
3022 | * |
3023 | * Description: | |
2e60e022 TH |
3024 | * Identical to blk_end_bidi_request() except that queue lock is |
3025 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
3026 | * |
3027 | * Return: | |
2e60e022 TH |
3028 | * %false - we are done with this request |
3029 | * %true - still buffers pending for this request | |
336cdb40 | 3030 | **/ |
2a842aca | 3031 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, |
b1f74493 | 3032 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 3033 | { |
2fff8a92 | 3034 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3035 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3036 | |
2e60e022 TH |
3037 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3038 | return true; | |
336cdb40 | 3039 | |
2e60e022 | 3040 | blk_finish_request(rq, error); |
336cdb40 | 3041 | |
2e60e022 | 3042 | return false; |
336cdb40 | 3043 | } |
e19a3ab0 KU |
3044 | |
3045 | /** | |
3046 | * blk_end_request - Helper function for drivers to complete the request. | |
3047 | * @rq: the request being processed | |
2a842aca | 3048 | * @error: block status code |
e19a3ab0 KU |
3049 | * @nr_bytes: number of bytes to complete |
3050 | * | |
3051 | * Description: | |
3052 | * Ends I/O on a number of bytes attached to @rq. | |
3053 | * If @rq has leftover, sets it up for the next range of segments. | |
3054 | * | |
3055 | * Return: | |
b1f74493 FT |
3056 | * %false - we are done with this request |
3057 | * %true - still buffers pending for this request | |
e19a3ab0 | 3058 | **/ |
2a842aca CH |
3059 | bool blk_end_request(struct request *rq, blk_status_t error, |
3060 | unsigned int nr_bytes) | |
e19a3ab0 | 3061 | { |
332ebbf7 | 3062 | WARN_ON_ONCE(rq->q->mq_ops); |
b1f74493 | 3063 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 3064 | } |
56ad1740 | 3065 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
3066 | |
3067 | /** | |
b1f74493 FT |
3068 | * blk_end_request_all - Helper function for drives to finish the request. |
3069 | * @rq: the request to finish | |
2a842aca | 3070 | * @error: block status code |
336cdb40 KU |
3071 | * |
3072 | * Description: | |
b1f74493 FT |
3073 | * Completely finish @rq. |
3074 | */ | |
2a842aca | 3075 | void blk_end_request_all(struct request *rq, blk_status_t error) |
336cdb40 | 3076 | { |
b1f74493 FT |
3077 | bool pending; |
3078 | unsigned int bidi_bytes = 0; | |
336cdb40 | 3079 | |
b1f74493 FT |
3080 | if (unlikely(blk_bidi_rq(rq))) |
3081 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 3082 | |
b1f74493 FT |
3083 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
3084 | BUG_ON(pending); | |
3085 | } | |
56ad1740 | 3086 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 3087 | |
e3a04fe3 | 3088 | /** |
b1f74493 FT |
3089 | * __blk_end_request - Helper function for drivers to complete the request. |
3090 | * @rq: the request being processed | |
2a842aca | 3091 | * @error: block status code |
b1f74493 | 3092 | * @nr_bytes: number of bytes to complete |
e3a04fe3 KU |
3093 | * |
3094 | * Description: | |
b1f74493 | 3095 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
3096 | * |
3097 | * Return: | |
b1f74493 FT |
3098 | * %false - we are done with this request |
3099 | * %true - still buffers pending for this request | |
e3a04fe3 | 3100 | **/ |
2a842aca CH |
3101 | bool __blk_end_request(struct request *rq, blk_status_t error, |
3102 | unsigned int nr_bytes) | |
e3a04fe3 | 3103 | { |
2fff8a92 | 3104 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3105 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3106 | |
b1f74493 | 3107 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 3108 | } |
56ad1740 | 3109 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 3110 | |
32fab448 | 3111 | /** |
b1f74493 FT |
3112 | * __blk_end_request_all - Helper function for drives to finish the request. |
3113 | * @rq: the request to finish | |
2a842aca | 3114 | * @error: block status code |
32fab448 KU |
3115 | * |
3116 | * Description: | |
b1f74493 | 3117 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 3118 | */ |
2a842aca | 3119 | void __blk_end_request_all(struct request *rq, blk_status_t error) |
32fab448 | 3120 | { |
b1f74493 FT |
3121 | bool pending; |
3122 | unsigned int bidi_bytes = 0; | |
3123 | ||
2fff8a92 | 3124 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3125 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3126 | |
b1f74493 FT |
3127 | if (unlikely(blk_bidi_rq(rq))) |
3128 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3129 | ||
3130 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3131 | BUG_ON(pending); | |
32fab448 | 3132 | } |
56ad1740 | 3133 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 3134 | |
e19a3ab0 | 3135 | /** |
b1f74493 FT |
3136 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
3137 | * @rq: the request to finish the current chunk for | |
2a842aca | 3138 | * @error: block status code |
e19a3ab0 KU |
3139 | * |
3140 | * Description: | |
b1f74493 FT |
3141 | * Complete the current consecutively mapped chunk from @rq. Must |
3142 | * be called with queue lock held. | |
e19a3ab0 KU |
3143 | * |
3144 | * Return: | |
b1f74493 FT |
3145 | * %false - we are done with this request |
3146 | * %true - still buffers pending for this request | |
3147 | */ | |
2a842aca | 3148 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) |
e19a3ab0 | 3149 | { |
b1f74493 | 3150 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 3151 | } |
56ad1740 | 3152 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 3153 | |
86db1e29 JA |
3154 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3155 | struct bio *bio) | |
1da177e4 | 3156 | { |
b4f42e28 | 3157 | if (bio_has_data(bio)) |
fb2dce86 | 3158 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
b4f42e28 | 3159 | |
4f024f37 | 3160 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 3161 | rq->bio = rq->biotail = bio; |
1da177e4 | 3162 | |
74d46992 CH |
3163 | if (bio->bi_disk) |
3164 | rq->rq_disk = bio->bi_disk; | |
66846572 | 3165 | } |
1da177e4 | 3166 | |
2d4dc890 IL |
3167 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
3168 | /** | |
3169 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3170 | * @rq: the request to be flushed | |
3171 | * | |
3172 | * Description: | |
3173 | * Flush all pages in @rq. | |
3174 | */ | |
3175 | void rq_flush_dcache_pages(struct request *rq) | |
3176 | { | |
3177 | struct req_iterator iter; | |
7988613b | 3178 | struct bio_vec bvec; |
2d4dc890 IL |
3179 | |
3180 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 3181 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
3182 | } |
3183 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3184 | #endif | |
3185 | ||
ef9e3fac KU |
3186 | /** |
3187 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3188 | * @q : the queue of the device being checked | |
3189 | * | |
3190 | * Description: | |
3191 | * Check if underlying low-level drivers of a device are busy. | |
3192 | * If the drivers want to export their busy state, they must set own | |
3193 | * exporting function using blk_queue_lld_busy() first. | |
3194 | * | |
3195 | * Basically, this function is used only by request stacking drivers | |
3196 | * to stop dispatching requests to underlying devices when underlying | |
3197 | * devices are busy. This behavior helps more I/O merging on the queue | |
3198 | * of the request stacking driver and prevents I/O throughput regression | |
3199 | * on burst I/O load. | |
3200 | * | |
3201 | * Return: | |
3202 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3203 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3204 | */ | |
3205 | int blk_lld_busy(struct request_queue *q) | |
3206 | { | |
3207 | if (q->lld_busy_fn) | |
3208 | return q->lld_busy_fn(q); | |
3209 | ||
3210 | return 0; | |
3211 | } | |
3212 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3213 | ||
78d8e58a MS |
3214 | /** |
3215 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3216 | * @rq: the clone request to be cleaned up | |
3217 | * | |
3218 | * Description: | |
3219 | * Free all bios in @rq for a cloned request. | |
3220 | */ | |
3221 | void blk_rq_unprep_clone(struct request *rq) | |
3222 | { | |
3223 | struct bio *bio; | |
3224 | ||
3225 | while ((bio = rq->bio) != NULL) { | |
3226 | rq->bio = bio->bi_next; | |
3227 | ||
3228 | bio_put(bio); | |
3229 | } | |
3230 | } | |
3231 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3232 | ||
3233 | /* | |
3234 | * Copy attributes of the original request to the clone request. | |
3235 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3236 | */ | |
3237 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3238 | { |
3239 | dst->cpu = src->cpu; | |
b0fd271d KU |
3240 | dst->__sector = blk_rq_pos(src); |
3241 | dst->__data_len = blk_rq_bytes(src); | |
3242 | dst->nr_phys_segments = src->nr_phys_segments; | |
3243 | dst->ioprio = src->ioprio; | |
3244 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3245 | } |
3246 | ||
3247 | /** | |
3248 | * blk_rq_prep_clone - Helper function to setup clone request | |
3249 | * @rq: the request to be setup | |
3250 | * @rq_src: original request to be cloned | |
3251 | * @bs: bio_set that bios for clone are allocated from | |
3252 | * @gfp_mask: memory allocation mask for bio | |
3253 | * @bio_ctr: setup function to be called for each clone bio. | |
3254 | * Returns %0 for success, non %0 for failure. | |
3255 | * @data: private data to be passed to @bio_ctr | |
3256 | * | |
3257 | * Description: | |
3258 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3259 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3260 | * are not copied, and copying such parts is the caller's responsibility. | |
3261 | * Also, pages which the original bios are pointing to are not copied | |
3262 | * and the cloned bios just point same pages. | |
3263 | * So cloned bios must be completed before original bios, which means | |
3264 | * the caller must complete @rq before @rq_src. | |
3265 | */ | |
3266 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3267 | struct bio_set *bs, gfp_t gfp_mask, | |
3268 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3269 | void *data) | |
3270 | { | |
3271 | struct bio *bio, *bio_src; | |
3272 | ||
3273 | if (!bs) | |
3274 | bs = fs_bio_set; | |
3275 | ||
3276 | __rq_for_each_bio(bio_src, rq_src) { | |
3277 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3278 | if (!bio) | |
3279 | goto free_and_out; | |
3280 | ||
3281 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3282 | goto free_and_out; | |
3283 | ||
3284 | if (rq->bio) { | |
3285 | rq->biotail->bi_next = bio; | |
3286 | rq->biotail = bio; | |
3287 | } else | |
3288 | rq->bio = rq->biotail = bio; | |
3289 | } | |
3290 | ||
3291 | __blk_rq_prep_clone(rq, rq_src); | |
3292 | ||
3293 | return 0; | |
3294 | ||
3295 | free_and_out: | |
3296 | if (bio) | |
3297 | bio_put(bio); | |
3298 | blk_rq_unprep_clone(rq); | |
3299 | ||
3300 | return -ENOMEM; | |
b0fd271d KU |
3301 | } |
3302 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3303 | ||
59c3d45e | 3304 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3305 | { |
3306 | return queue_work(kblockd_workqueue, work); | |
3307 | } | |
1da177e4 LT |
3308 | EXPORT_SYMBOL(kblockd_schedule_work); |
3309 | ||
ee63cfa7 JA |
3310 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3311 | { | |
3312 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3313 | } | |
3314 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3315 | ||
818cd1cb JA |
3316 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
3317 | unsigned long delay) | |
3318 | { | |
3319 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3320 | } | |
3321 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3322 | ||
59c3d45e JA |
3323 | int kblockd_schedule_delayed_work(struct delayed_work *dwork, |
3324 | unsigned long delay) | |
e43473b7 VG |
3325 | { |
3326 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
3327 | } | |
3328 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
3329 | ||
8ab14595 JA |
3330 | int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3331 | unsigned long delay) | |
3332 | { | |
3333 | return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3334 | } | |
3335 | EXPORT_SYMBOL(kblockd_schedule_delayed_work_on); | |
3336 | ||
75df7136 SJ |
3337 | /** |
3338 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3339 | * @plug: The &struct blk_plug that needs to be initialized | |
3340 | * | |
3341 | * Description: | |
3342 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3343 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3344 | * blk_finish_plug(). This is important from a performance perspective, but | |
3345 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3346 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3347 | * page belonging to that request that is currently residing in our private | |
3348 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3349 | * this kind of deadlock. | |
3350 | */ | |
73c10101 JA |
3351 | void blk_start_plug(struct blk_plug *plug) |
3352 | { | |
3353 | struct task_struct *tsk = current; | |
3354 | ||
dd6cf3e1 SL |
3355 | /* |
3356 | * If this is a nested plug, don't actually assign it. | |
3357 | */ | |
3358 | if (tsk->plug) | |
3359 | return; | |
3360 | ||
73c10101 | 3361 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3362 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3363 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3364 | /* |
dd6cf3e1 SL |
3365 | * Store ordering should not be needed here, since a potential |
3366 | * preempt will imply a full memory barrier | |
73c10101 | 3367 | */ |
dd6cf3e1 | 3368 | tsk->plug = plug; |
73c10101 JA |
3369 | } |
3370 | EXPORT_SYMBOL(blk_start_plug); | |
3371 | ||
3372 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3373 | { | |
3374 | struct request *rqa = container_of(a, struct request, queuelist); | |
3375 | struct request *rqb = container_of(b, struct request, queuelist); | |
3376 | ||
975927b9 JM |
3377 | return !(rqa->q < rqb->q || |
3378 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3379 | } |
3380 | ||
49cac01e JA |
3381 | /* |
3382 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3383 | * until a safe kblockd context. We due this to avoid accidental big | |
3384 | * additional stack usage in driver dispatch, in places where the originally | |
3385 | * plugger did not intend it. | |
3386 | */ | |
f6603783 | 3387 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3388 | bool from_schedule) |
99e22598 | 3389 | __releases(q->queue_lock) |
94b5eb28 | 3390 | { |
2fff8a92 BVA |
3391 | lockdep_assert_held(q->queue_lock); |
3392 | ||
49cac01e | 3393 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3394 | |
70460571 | 3395 | if (from_schedule) |
24ecfbe2 | 3396 | blk_run_queue_async(q); |
70460571 | 3397 | else |
24ecfbe2 | 3398 | __blk_run_queue(q); |
70460571 | 3399 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3400 | } |
3401 | ||
74018dc3 | 3402 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3403 | { |
3404 | LIST_HEAD(callbacks); | |
3405 | ||
2a7d5559 SL |
3406 | while (!list_empty(&plug->cb_list)) { |
3407 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3408 | |
2a7d5559 SL |
3409 | while (!list_empty(&callbacks)) { |
3410 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3411 | struct blk_plug_cb, |
3412 | list); | |
2a7d5559 | 3413 | list_del(&cb->list); |
74018dc3 | 3414 | cb->callback(cb, from_schedule); |
2a7d5559 | 3415 | } |
048c9374 N |
3416 | } |
3417 | } | |
3418 | ||
9cbb1750 N |
3419 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3420 | int size) | |
3421 | { | |
3422 | struct blk_plug *plug = current->plug; | |
3423 | struct blk_plug_cb *cb; | |
3424 | ||
3425 | if (!plug) | |
3426 | return NULL; | |
3427 | ||
3428 | list_for_each_entry(cb, &plug->cb_list, list) | |
3429 | if (cb->callback == unplug && cb->data == data) | |
3430 | return cb; | |
3431 | ||
3432 | /* Not currently on the callback list */ | |
3433 | BUG_ON(size < sizeof(*cb)); | |
3434 | cb = kzalloc(size, GFP_ATOMIC); | |
3435 | if (cb) { | |
3436 | cb->data = data; | |
3437 | cb->callback = unplug; | |
3438 | list_add(&cb->list, &plug->cb_list); | |
3439 | } | |
3440 | return cb; | |
3441 | } | |
3442 | EXPORT_SYMBOL(blk_check_plugged); | |
3443 | ||
49cac01e | 3444 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3445 | { |
3446 | struct request_queue *q; | |
3447 | unsigned long flags; | |
3448 | struct request *rq; | |
109b8129 | 3449 | LIST_HEAD(list); |
94b5eb28 | 3450 | unsigned int depth; |
73c10101 | 3451 | |
74018dc3 | 3452 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3453 | |
3454 | if (!list_empty(&plug->mq_list)) | |
3455 | blk_mq_flush_plug_list(plug, from_schedule); | |
3456 | ||
73c10101 JA |
3457 | if (list_empty(&plug->list)) |
3458 | return; | |
3459 | ||
109b8129 N |
3460 | list_splice_init(&plug->list, &list); |
3461 | ||
422765c2 | 3462 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3463 | |
3464 | q = NULL; | |
94b5eb28 | 3465 | depth = 0; |
18811272 JA |
3466 | |
3467 | /* | |
3468 | * Save and disable interrupts here, to avoid doing it for every | |
3469 | * queue lock we have to take. | |
3470 | */ | |
73c10101 | 3471 | local_irq_save(flags); |
109b8129 N |
3472 | while (!list_empty(&list)) { |
3473 | rq = list_entry_rq(list.next); | |
73c10101 | 3474 | list_del_init(&rq->queuelist); |
73c10101 JA |
3475 | BUG_ON(!rq->q); |
3476 | if (rq->q != q) { | |
99e22598 JA |
3477 | /* |
3478 | * This drops the queue lock | |
3479 | */ | |
3480 | if (q) | |
49cac01e | 3481 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3482 | q = rq->q; |
94b5eb28 | 3483 | depth = 0; |
73c10101 JA |
3484 | spin_lock(q->queue_lock); |
3485 | } | |
8ba61435 TH |
3486 | |
3487 | /* | |
3488 | * Short-circuit if @q is dead | |
3489 | */ | |
3f3299d5 | 3490 | if (unlikely(blk_queue_dying(q))) { |
2a842aca | 3491 | __blk_end_request_all(rq, BLK_STS_IOERR); |
8ba61435 TH |
3492 | continue; |
3493 | } | |
3494 | ||
73c10101 JA |
3495 | /* |
3496 | * rq is already accounted, so use raw insert | |
3497 | */ | |
f73f44eb | 3498 | if (op_is_flush(rq->cmd_flags)) |
401a18e9 JA |
3499 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3500 | else | |
3501 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3502 | |
3503 | depth++; | |
73c10101 JA |
3504 | } |
3505 | ||
99e22598 JA |
3506 | /* |
3507 | * This drops the queue lock | |
3508 | */ | |
3509 | if (q) | |
49cac01e | 3510 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3511 | |
73c10101 JA |
3512 | local_irq_restore(flags); |
3513 | } | |
73c10101 JA |
3514 | |
3515 | void blk_finish_plug(struct blk_plug *plug) | |
3516 | { | |
dd6cf3e1 SL |
3517 | if (plug != current->plug) |
3518 | return; | |
f6603783 | 3519 | blk_flush_plug_list(plug, false); |
73c10101 | 3520 | |
dd6cf3e1 | 3521 | current->plug = NULL; |
73c10101 | 3522 | } |
88b996cd | 3523 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3524 | |
47fafbc7 | 3525 | #ifdef CONFIG_PM |
6c954667 LM |
3526 | /** |
3527 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3528 | * @q: the queue of the device | |
3529 | * @dev: the device the queue belongs to | |
3530 | * | |
3531 | * Description: | |
3532 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3533 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3534 | * should call this function after @dev has been initialized, and its | |
3535 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3536 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3537 | * cases, driver should call this function before any I/O has taken place. | |
3538 | * | |
3539 | * This function takes care of setting up using auto suspend for the device, | |
3540 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3541 | * until an updated value is either set by user or by driver. Drivers do | |
3542 | * not need to touch other autosuspend settings. | |
3543 | * | |
3544 | * The block layer runtime PM is request based, so only works for drivers | |
3545 | * that use request as their IO unit instead of those directly use bio's. | |
3546 | */ | |
3547 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3548 | { | |
765e40b6 CH |
3549 | /* not support for RQF_PM and ->rpm_status in blk-mq yet */ |
3550 | if (q->mq_ops) | |
3551 | return; | |
3552 | ||
6c954667 LM |
3553 | q->dev = dev; |
3554 | q->rpm_status = RPM_ACTIVE; | |
3555 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3556 | pm_runtime_use_autosuspend(q->dev); | |
3557 | } | |
3558 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3559 | ||
3560 | /** | |
3561 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3562 | * @q: the queue of the device | |
3563 | * | |
3564 | * Description: | |
3565 | * This function will check if runtime suspend is allowed for the device | |
3566 | * by examining if there are any requests pending in the queue. If there | |
3567 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3568 | * the queue's status will be updated to SUSPENDING and the driver can | |
3569 | * proceed to suspend the device. | |
3570 | * | |
3571 | * For the not allowed case, we mark last busy for the device so that | |
3572 | * runtime PM core will try to autosuspend it some time later. | |
3573 | * | |
3574 | * This function should be called near the start of the device's | |
3575 | * runtime_suspend callback. | |
3576 | * | |
3577 | * Return: | |
3578 | * 0 - OK to runtime suspend the device | |
3579 | * -EBUSY - Device should not be runtime suspended | |
3580 | */ | |
3581 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3582 | { | |
3583 | int ret = 0; | |
3584 | ||
4fd41a85 KX |
3585 | if (!q->dev) |
3586 | return ret; | |
3587 | ||
6c954667 LM |
3588 | spin_lock_irq(q->queue_lock); |
3589 | if (q->nr_pending) { | |
3590 | ret = -EBUSY; | |
3591 | pm_runtime_mark_last_busy(q->dev); | |
3592 | } else { | |
3593 | q->rpm_status = RPM_SUSPENDING; | |
3594 | } | |
3595 | spin_unlock_irq(q->queue_lock); | |
3596 | return ret; | |
3597 | } | |
3598 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3599 | ||
3600 | /** | |
3601 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3602 | * @q: the queue of the device | |
3603 | * @err: return value of the device's runtime_suspend function | |
3604 | * | |
3605 | * Description: | |
3606 | * Update the queue's runtime status according to the return value of the | |
3607 | * device's runtime suspend function and mark last busy for the device so | |
3608 | * that PM core will try to auto suspend the device at a later time. | |
3609 | * | |
3610 | * This function should be called near the end of the device's | |
3611 | * runtime_suspend callback. | |
3612 | */ | |
3613 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3614 | { | |
4fd41a85 KX |
3615 | if (!q->dev) |
3616 | return; | |
3617 | ||
6c954667 LM |
3618 | spin_lock_irq(q->queue_lock); |
3619 | if (!err) { | |
3620 | q->rpm_status = RPM_SUSPENDED; | |
3621 | } else { | |
3622 | q->rpm_status = RPM_ACTIVE; | |
3623 | pm_runtime_mark_last_busy(q->dev); | |
3624 | } | |
3625 | spin_unlock_irq(q->queue_lock); | |
3626 | } | |
3627 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3628 | ||
3629 | /** | |
3630 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3631 | * @q: the queue of the device | |
3632 | * | |
3633 | * Description: | |
3634 | * Update the queue's runtime status to RESUMING in preparation for the | |
3635 | * runtime resume of the device. | |
3636 | * | |
3637 | * This function should be called near the start of the device's | |
3638 | * runtime_resume callback. | |
3639 | */ | |
3640 | void blk_pre_runtime_resume(struct request_queue *q) | |
3641 | { | |
4fd41a85 KX |
3642 | if (!q->dev) |
3643 | return; | |
3644 | ||
6c954667 LM |
3645 | spin_lock_irq(q->queue_lock); |
3646 | q->rpm_status = RPM_RESUMING; | |
3647 | spin_unlock_irq(q->queue_lock); | |
3648 | } | |
3649 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3650 | ||
3651 | /** | |
3652 | * blk_post_runtime_resume - Post runtime resume processing | |
3653 | * @q: the queue of the device | |
3654 | * @err: return value of the device's runtime_resume function | |
3655 | * | |
3656 | * Description: | |
3657 | * Update the queue's runtime status according to the return value of the | |
3658 | * device's runtime_resume function. If it is successfully resumed, process | |
3659 | * the requests that are queued into the device's queue when it is resuming | |
3660 | * and then mark last busy and initiate autosuspend for it. | |
3661 | * | |
3662 | * This function should be called near the end of the device's | |
3663 | * runtime_resume callback. | |
3664 | */ | |
3665 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3666 | { | |
4fd41a85 KX |
3667 | if (!q->dev) |
3668 | return; | |
3669 | ||
6c954667 LM |
3670 | spin_lock_irq(q->queue_lock); |
3671 | if (!err) { | |
3672 | q->rpm_status = RPM_ACTIVE; | |
3673 | __blk_run_queue(q); | |
3674 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3675 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3676 | } else { |
3677 | q->rpm_status = RPM_SUSPENDED; | |
3678 | } | |
3679 | spin_unlock_irq(q->queue_lock); | |
3680 | } | |
3681 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3682 | |
3683 | /** | |
3684 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3685 | * @q: the queue of the device | |
3686 | * | |
3687 | * If the device is left runtime suspended during system suspend the resume | |
3688 | * hook typically resumes the device and corrects runtime status | |
3689 | * accordingly. However, that does not affect the queue runtime PM status | |
3690 | * which is still "suspended". This prevents processing requests from the | |
3691 | * queue. | |
3692 | * | |
3693 | * This function can be used in driver's resume hook to correct queue | |
3694 | * runtime PM status and re-enable peeking requests from the queue. It | |
3695 | * should be called before first request is added to the queue. | |
3696 | */ | |
3697 | void blk_set_runtime_active(struct request_queue *q) | |
3698 | { | |
3699 | spin_lock_irq(q->queue_lock); | |
3700 | q->rpm_status = RPM_ACTIVE; | |
3701 | pm_runtime_mark_last_busy(q->dev); | |
3702 | pm_request_autosuspend(q->dev); | |
3703 | spin_unlock_irq(q->queue_lock); | |
3704 | } | |
3705 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3706 | #endif |
3707 | ||
1da177e4 LT |
3708 | int __init blk_dev_init(void) |
3709 | { | |
ef295ecf CH |
3710 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3711 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3712 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3713 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3714 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3715 | |
89b90be2 TH |
3716 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3717 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3718 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3719 | if (!kblockd_workqueue) |
3720 | panic("Failed to create kblockd\n"); | |
3721 | ||
3722 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3723 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3724 | |
c2789bd4 | 3725 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3726 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3727 | |
18fbda91 OS |
3728 | #ifdef CONFIG_DEBUG_FS |
3729 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3730 | #endif | |
3731 | ||
d38ecf93 | 3732 | return 0; |
1da177e4 | 3733 | } |