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> |
30abb3a6 | 37 | #include <linux/bpf.h> |
55782138 LZ |
38 | |
39 | #define CREATE_TRACE_POINTS | |
40 | #include <trace/events/block.h> | |
1da177e4 | 41 | |
8324aa91 | 42 | #include "blk.h" |
43a5e4e2 | 43 | #include "blk-mq.h" |
bd166ef1 | 44 | #include "blk-mq-sched.h" |
87760e5e | 45 | #include "blk-wbt.h" |
8324aa91 | 46 | |
18fbda91 OS |
47 | #ifdef CONFIG_DEBUG_FS |
48 | struct dentry *blk_debugfs_root; | |
49 | #endif | |
50 | ||
d07335e5 | 51 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 52 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
0a82a8d1 | 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
3291fa57 | 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
cbae8d45 | 55 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
0bfc2455 | 56 | |
a73f730d TH |
57 | DEFINE_IDA(blk_queue_ida); |
58 | ||
1da177e4 LT |
59 | /* |
60 | * For the allocated request tables | |
61 | */ | |
d674d414 | 62 | struct kmem_cache *request_cachep; |
1da177e4 LT |
63 | |
64 | /* | |
65 | * For queue allocation | |
66 | */ | |
6728cb0e | 67 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 68 | |
1da177e4 LT |
69 | /* |
70 | * Controlling structure to kblockd | |
71 | */ | |
ff856bad | 72 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 73 | |
d40f75a0 TH |
74 | static void blk_clear_congested(struct request_list *rl, int sync) |
75 | { | |
d40f75a0 TH |
76 | #ifdef CONFIG_CGROUP_WRITEBACK |
77 | clear_wb_congested(rl->blkg->wb_congested, sync); | |
78 | #else | |
482cf79c TH |
79 | /* |
80 | * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't | |
81 | * flip its congestion state for events on other blkcgs. | |
82 | */ | |
83 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 84 | clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
85 | #endif |
86 | } | |
87 | ||
88 | static void blk_set_congested(struct request_list *rl, int sync) | |
89 | { | |
d40f75a0 TH |
90 | #ifdef CONFIG_CGROUP_WRITEBACK |
91 | set_wb_congested(rl->blkg->wb_congested, sync); | |
92 | #else | |
482cf79c TH |
93 | /* see blk_clear_congested() */ |
94 | if (rl == &rl->q->root_rl) | |
dc3b17cc | 95 | set_wb_congested(rl->q->backing_dev_info->wb.congested, sync); |
d40f75a0 TH |
96 | #endif |
97 | } | |
98 | ||
8324aa91 | 99 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
100 | { |
101 | int nr; | |
102 | ||
103 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
104 | if (nr > q->nr_requests) | |
105 | nr = q->nr_requests; | |
106 | q->nr_congestion_on = nr; | |
107 | ||
108 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
109 | if (nr < 1) | |
110 | nr = 1; | |
111 | q->nr_congestion_off = nr; | |
112 | } | |
113 | ||
2a4aa30c | 114 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 115 | { |
1afb20f3 FT |
116 | memset(rq, 0, sizeof(*rq)); |
117 | ||
1da177e4 | 118 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 119 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 120 | rq->cpu = -1; |
63a71386 | 121 | rq->q = q; |
a2dec7b3 | 122 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
123 | INIT_HLIST_NODE(&rq->hash); |
124 | RB_CLEAR_NODE(&rq->rb_node); | |
63a71386 | 125 | rq->tag = -1; |
bd166ef1 | 126 | rq->internal_tag = -1; |
b243ddcb | 127 | rq->start_time = jiffies; |
9195291e | 128 | set_start_time_ns(rq); |
09e099d4 | 129 | rq->part = NULL; |
1d9bd516 TH |
130 | seqcount_init(&rq->gstate_seq); |
131 | u64_stats_init(&rq->aborted_gstate_sync); | |
1da177e4 | 132 | } |
2a4aa30c | 133 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 134 | |
2a842aca CH |
135 | static const struct { |
136 | int errno; | |
137 | const char *name; | |
138 | } blk_errors[] = { | |
139 | [BLK_STS_OK] = { 0, "" }, | |
140 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, | |
141 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, | |
142 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, | |
143 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, | |
144 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, | |
145 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, | |
146 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, | |
147 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, | |
148 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, | |
86ff7c2a | 149 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
03a07c92 | 150 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
2a842aca | 151 | |
4e4cbee9 CH |
152 | /* device mapper special case, should not leak out: */ |
153 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, | |
154 | ||
2a842aca CH |
155 | /* everything else not covered above: */ |
156 | [BLK_STS_IOERR] = { -EIO, "I/O" }, | |
157 | }; | |
158 | ||
159 | blk_status_t errno_to_blk_status(int errno) | |
160 | { | |
161 | int i; | |
162 | ||
163 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { | |
164 | if (blk_errors[i].errno == errno) | |
165 | return (__force blk_status_t)i; | |
166 | } | |
167 | ||
168 | return BLK_STS_IOERR; | |
169 | } | |
170 | EXPORT_SYMBOL_GPL(errno_to_blk_status); | |
171 | ||
172 | int blk_status_to_errno(blk_status_t status) | |
173 | { | |
174 | int idx = (__force int)status; | |
175 | ||
34bd9c1c | 176 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
177 | return -EIO; |
178 | return blk_errors[idx].errno; | |
179 | } | |
180 | EXPORT_SYMBOL_GPL(blk_status_to_errno); | |
181 | ||
182 | static void print_req_error(struct request *req, blk_status_t status) | |
183 | { | |
184 | int idx = (__force int)status; | |
185 | ||
34bd9c1c | 186 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
2a842aca CH |
187 | return; |
188 | ||
189 | printk_ratelimited(KERN_ERR "%s: %s error, dev %s, sector %llu\n", | |
190 | __func__, blk_errors[idx].name, req->rq_disk ? | |
191 | req->rq_disk->disk_name : "?", | |
192 | (unsigned long long)blk_rq_pos(req)); | |
193 | } | |
194 | ||
5bb23a68 | 195 | static void req_bio_endio(struct request *rq, struct bio *bio, |
2a842aca | 196 | unsigned int nbytes, blk_status_t error) |
1da177e4 | 197 | { |
78d8e58a | 198 | if (error) |
4e4cbee9 | 199 | bio->bi_status = error; |
797e7dbb | 200 | |
e8064021 | 201 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
b7c44ed9 | 202 | bio_set_flag(bio, BIO_QUIET); |
08bafc03 | 203 | |
f79ea416 | 204 | bio_advance(bio, nbytes); |
7ba1ba12 | 205 | |
143a87f4 | 206 | /* don't actually finish bio if it's part of flush sequence */ |
e8064021 | 207 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
4246a0b6 | 208 | bio_endio(bio); |
1da177e4 | 209 | } |
1da177e4 | 210 | |
1da177e4 LT |
211 | void blk_dump_rq_flags(struct request *rq, char *msg) |
212 | { | |
aebf526b CH |
213 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
214 | rq->rq_disk ? rq->rq_disk->disk_name : "?", | |
5953316d | 215 | (unsigned long long) rq->cmd_flags); |
1da177e4 | 216 | |
83096ebf TH |
217 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
218 | (unsigned long long)blk_rq_pos(rq), | |
219 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
b4f42e28 JA |
220 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
221 | rq->bio, rq->biotail, blk_rq_bytes(rq)); | |
1da177e4 | 222 | } |
1da177e4 LT |
223 | EXPORT_SYMBOL(blk_dump_rq_flags); |
224 | ||
3cca6dc1 | 225 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 226 | { |
3cca6dc1 | 227 | struct request_queue *q; |
1da177e4 | 228 | |
3cca6dc1 JA |
229 | q = container_of(work, struct request_queue, delay_work.work); |
230 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 231 | __blk_run_queue(q); |
3cca6dc1 | 232 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 233 | } |
1da177e4 LT |
234 | |
235 | /** | |
3cca6dc1 JA |
236 | * blk_delay_queue - restart queueing after defined interval |
237 | * @q: The &struct request_queue in question | |
238 | * @msecs: Delay in msecs | |
1da177e4 LT |
239 | * |
240 | * Description: | |
3cca6dc1 JA |
241 | * Sometimes queueing needs to be postponed for a little while, to allow |
242 | * resources to come back. This function will make sure that queueing is | |
2fff8a92 | 243 | * restarted around the specified time. |
3cca6dc1 JA |
244 | */ |
245 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 246 | { |
2fff8a92 | 247 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 248 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 249 | |
70460571 BVA |
250 | if (likely(!blk_queue_dead(q))) |
251 | queue_delayed_work(kblockd_workqueue, &q->delay_work, | |
252 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 253 | } |
3cca6dc1 | 254 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 255 | |
21491412 JA |
256 | /** |
257 | * blk_start_queue_async - asynchronously restart a previously stopped queue | |
258 | * @q: The &struct request_queue in question | |
259 | * | |
260 | * Description: | |
261 | * blk_start_queue_async() will clear the stop flag on the queue, and | |
262 | * ensure that the request_fn for the queue is run from an async | |
263 | * context. | |
264 | **/ | |
265 | void blk_start_queue_async(struct request_queue *q) | |
266 | { | |
2fff8a92 | 267 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 268 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 269 | |
21491412 JA |
270 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
271 | blk_run_queue_async(q); | |
272 | } | |
273 | EXPORT_SYMBOL(blk_start_queue_async); | |
274 | ||
1da177e4 LT |
275 | /** |
276 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 277 | * @q: The &struct request_queue in question |
1da177e4 LT |
278 | * |
279 | * Description: | |
280 | * blk_start_queue() will clear the stop flag on the queue, and call | |
281 | * the request_fn for the queue if it was in a stopped state when | |
2fff8a92 | 282 | * entered. Also see blk_stop_queue(). |
1da177e4 | 283 | **/ |
165125e1 | 284 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 285 | { |
2fff8a92 | 286 | lockdep_assert_held(q->queue_lock); |
4ddd56b0 | 287 | WARN_ON(!in_interrupt() && !irqs_disabled()); |
332ebbf7 | 288 | WARN_ON_ONCE(q->mq_ops); |
a038e253 | 289 | |
75ad23bc | 290 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 291 | __blk_run_queue(q); |
1da177e4 | 292 | } |
1da177e4 LT |
293 | EXPORT_SYMBOL(blk_start_queue); |
294 | ||
295 | /** | |
296 | * blk_stop_queue - stop a queue | |
165125e1 | 297 | * @q: The &struct request_queue in question |
1da177e4 LT |
298 | * |
299 | * Description: | |
300 | * The Linux block layer assumes that a block driver will consume all | |
301 | * entries on the request queue when the request_fn strategy is called. | |
302 | * Often this will not happen, because of hardware limitations (queue | |
303 | * depth settings). If a device driver gets a 'queue full' response, | |
304 | * or if it simply chooses not to queue more I/O at one point, it can | |
305 | * call this function to prevent the request_fn from being called until | |
306 | * the driver has signalled it's ready to go again. This happens by calling | |
2fff8a92 | 307 | * blk_start_queue() to restart queue operations. |
1da177e4 | 308 | **/ |
165125e1 | 309 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 310 | { |
2fff8a92 | 311 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 312 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 313 | |
136b5721 | 314 | cancel_delayed_work(&q->delay_work); |
75ad23bc | 315 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
316 | } |
317 | EXPORT_SYMBOL(blk_stop_queue); | |
318 | ||
319 | /** | |
320 | * blk_sync_queue - cancel any pending callbacks on a queue | |
321 | * @q: the queue | |
322 | * | |
323 | * Description: | |
324 | * The block layer may perform asynchronous callback activity | |
325 | * on a queue, such as calling the unplug function after a timeout. | |
326 | * A block device may call blk_sync_queue to ensure that any | |
327 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 328 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
329 | * that its ->make_request_fn will not re-add plugging prior to calling |
330 | * this function. | |
331 | * | |
da527770 | 332 | * This function does not cancel any asynchronous activity arising |
da3dae54 | 333 | * out of elevator or throttling code. That would require elevator_exit() |
5efd6113 | 334 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 335 | * |
1da177e4 LT |
336 | */ |
337 | void blk_sync_queue(struct request_queue *q) | |
338 | { | |
70ed28b9 | 339 | del_timer_sync(&q->timeout); |
4e9b6f20 | 340 | cancel_work_sync(&q->timeout_work); |
f04c1fe7 ML |
341 | |
342 | if (q->mq_ops) { | |
343 | struct blk_mq_hw_ctx *hctx; | |
344 | int i; | |
345 | ||
aba7afc5 | 346 | cancel_delayed_work_sync(&q->requeue_work); |
21c6e939 | 347 | queue_for_each_hw_ctx(q, hctx, i) |
9f993737 | 348 | cancel_delayed_work_sync(&hctx->run_work); |
f04c1fe7 ML |
349 | } else { |
350 | cancel_delayed_work_sync(&q->delay_work); | |
351 | } | |
1da177e4 LT |
352 | } |
353 | EXPORT_SYMBOL(blk_sync_queue); | |
354 | ||
c9254f2d BVA |
355 | /** |
356 | * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY | |
357 | * @q: request queue pointer | |
358 | * | |
359 | * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not | |
360 | * set and 1 if the flag was already set. | |
361 | */ | |
362 | int blk_set_preempt_only(struct request_queue *q) | |
363 | { | |
364 | unsigned long flags; | |
365 | int res; | |
366 | ||
367 | spin_lock_irqsave(q->queue_lock, flags); | |
368 | res = queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY, q); | |
369 | spin_unlock_irqrestore(q->queue_lock, flags); | |
370 | ||
371 | return res; | |
372 | } | |
373 | EXPORT_SYMBOL_GPL(blk_set_preempt_only); | |
374 | ||
375 | void blk_clear_preempt_only(struct request_queue *q) | |
376 | { | |
377 | unsigned long flags; | |
378 | ||
379 | spin_lock_irqsave(q->queue_lock, flags); | |
380 | queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY, q); | |
3a0a5299 | 381 | wake_up_all(&q->mq_freeze_wq); |
c9254f2d BVA |
382 | spin_unlock_irqrestore(q->queue_lock, flags); |
383 | } | |
384 | EXPORT_SYMBOL_GPL(blk_clear_preempt_only); | |
385 | ||
c246e80d BVA |
386 | /** |
387 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped | |
388 | * @q: The queue to run | |
389 | * | |
390 | * Description: | |
391 | * Invoke request handling on a queue if there are any pending requests. | |
392 | * May be used to restart request handling after a request has completed. | |
393 | * This variant runs the queue whether or not the queue has been | |
394 | * stopped. Must be called with the queue lock held and interrupts | |
395 | * disabled. See also @blk_run_queue. | |
396 | */ | |
397 | inline void __blk_run_queue_uncond(struct request_queue *q) | |
398 | { | |
2fff8a92 | 399 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 400 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 401 | |
c246e80d BVA |
402 | if (unlikely(blk_queue_dead(q))) |
403 | return; | |
404 | ||
24faf6f6 BVA |
405 | /* |
406 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock | |
407 | * the queue lock internally. As a result multiple threads may be | |
408 | * running such a request function concurrently. Keep track of the | |
409 | * number of active request_fn invocations such that blk_drain_queue() | |
410 | * can wait until all these request_fn calls have finished. | |
411 | */ | |
412 | q->request_fn_active++; | |
c246e80d | 413 | q->request_fn(q); |
24faf6f6 | 414 | q->request_fn_active--; |
c246e80d | 415 | } |
a7928c15 | 416 | EXPORT_SYMBOL_GPL(__blk_run_queue_uncond); |
c246e80d | 417 | |
1da177e4 | 418 | /** |
80a4b58e | 419 | * __blk_run_queue - run a single device queue |
1da177e4 | 420 | * @q: The queue to run |
80a4b58e JA |
421 | * |
422 | * Description: | |
2fff8a92 | 423 | * See @blk_run_queue. |
1da177e4 | 424 | */ |
24ecfbe2 | 425 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 426 | { |
2fff8a92 | 427 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 428 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 429 | |
a538cd03 TH |
430 | if (unlikely(blk_queue_stopped(q))) |
431 | return; | |
432 | ||
c246e80d | 433 | __blk_run_queue_uncond(q); |
75ad23bc NP |
434 | } |
435 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 436 | |
24ecfbe2 CH |
437 | /** |
438 | * blk_run_queue_async - run a single device queue in workqueue context | |
439 | * @q: The queue to run | |
440 | * | |
441 | * Description: | |
442 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
2fff8a92 BVA |
443 | * of us. |
444 | * | |
445 | * Note: | |
446 | * Since it is not allowed to run q->delay_work after blk_cleanup_queue() | |
447 | * has canceled q->delay_work, callers must hold the queue lock to avoid | |
448 | * race conditions between blk_cleanup_queue() and blk_run_queue_async(). | |
24ecfbe2 CH |
449 | */ |
450 | void blk_run_queue_async(struct request_queue *q) | |
451 | { | |
2fff8a92 | 452 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 453 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 454 | |
70460571 | 455 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
e7c2f967 | 456 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
24ecfbe2 | 457 | } |
c21e6beb | 458 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 459 | |
75ad23bc NP |
460 | /** |
461 | * blk_run_queue - run a single device queue | |
462 | * @q: The queue to run | |
80a4b58e JA |
463 | * |
464 | * Description: | |
465 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 466 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
467 | */ |
468 | void blk_run_queue(struct request_queue *q) | |
469 | { | |
470 | unsigned long flags; | |
471 | ||
332ebbf7 BVA |
472 | WARN_ON_ONCE(q->mq_ops); |
473 | ||
75ad23bc | 474 | spin_lock_irqsave(q->queue_lock, flags); |
24ecfbe2 | 475 | __blk_run_queue(q); |
1da177e4 LT |
476 | spin_unlock_irqrestore(q->queue_lock, flags); |
477 | } | |
478 | EXPORT_SYMBOL(blk_run_queue); | |
479 | ||
165125e1 | 480 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
481 | { |
482 | kobject_put(&q->kobj); | |
483 | } | |
d86e0e83 | 484 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 485 | |
e3c78ca5 | 486 | /** |
807592a4 | 487 | * __blk_drain_queue - drain requests from request_queue |
e3c78ca5 | 488 | * @q: queue to drain |
c9a929dd | 489 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 490 | * |
c9a929dd TH |
491 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
492 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
493 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 494 | */ |
807592a4 BVA |
495 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
496 | __releases(q->queue_lock) | |
497 | __acquires(q->queue_lock) | |
e3c78ca5 | 498 | { |
458f27a9 AH |
499 | int i; |
500 | ||
807592a4 | 501 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 502 | WARN_ON_ONCE(q->mq_ops); |
807592a4 | 503 | |
e3c78ca5 | 504 | while (true) { |
481a7d64 | 505 | bool drain = false; |
e3c78ca5 | 506 | |
b855b04a TH |
507 | /* |
508 | * The caller might be trying to drain @q before its | |
509 | * elevator is initialized. | |
510 | */ | |
511 | if (q->elevator) | |
512 | elv_drain_elevator(q); | |
513 | ||
5efd6113 | 514 | blkcg_drain_queue(q); |
e3c78ca5 | 515 | |
4eabc941 TH |
516 | /* |
517 | * This function might be called on a queue which failed | |
b855b04a TH |
518 | * driver init after queue creation or is not yet fully |
519 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
520 | * in such cases. Kick queue iff dispatch queue has | |
521 | * something on it and @q has request_fn set. | |
4eabc941 | 522 | */ |
b855b04a | 523 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 524 | __blk_run_queue(q); |
c9a929dd | 525 | |
8a5ecdd4 | 526 | drain |= q->nr_rqs_elvpriv; |
24faf6f6 | 527 | drain |= q->request_fn_active; |
481a7d64 TH |
528 | |
529 | /* | |
530 | * Unfortunately, requests are queued at and tracked from | |
531 | * multiple places and there's no single counter which can | |
532 | * be drained. Check all the queues and counters. | |
533 | */ | |
534 | if (drain_all) { | |
e97c293c | 535 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
481a7d64 TH |
536 | drain |= !list_empty(&q->queue_head); |
537 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 538 | drain |= q->nr_rqs[i]; |
481a7d64 | 539 | drain |= q->in_flight[i]; |
7c94e1c1 ML |
540 | if (fq) |
541 | drain |= !list_empty(&fq->flush_queue[i]); | |
481a7d64 TH |
542 | } |
543 | } | |
e3c78ca5 | 544 | |
481a7d64 | 545 | if (!drain) |
e3c78ca5 | 546 | break; |
807592a4 BVA |
547 | |
548 | spin_unlock_irq(q->queue_lock); | |
549 | ||
e3c78ca5 | 550 | msleep(10); |
807592a4 BVA |
551 | |
552 | spin_lock_irq(q->queue_lock); | |
e3c78ca5 | 553 | } |
458f27a9 AH |
554 | |
555 | /* | |
556 | * With queue marked dead, any woken up waiter will fail the | |
557 | * allocation path, so the wakeup chaining is lost and we're | |
558 | * left with hung waiters. We need to wake up those waiters. | |
559 | */ | |
560 | if (q->request_fn) { | |
a051661c TH |
561 | struct request_list *rl; |
562 | ||
a051661c TH |
563 | blk_queue_for_each_rl(rl, q) |
564 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
565 | wake_up_all(&rl->wait[i]); | |
458f27a9 | 566 | } |
e3c78ca5 TH |
567 | } |
568 | ||
454be724 ML |
569 | void blk_drain_queue(struct request_queue *q) |
570 | { | |
571 | spin_lock_irq(q->queue_lock); | |
572 | __blk_drain_queue(q, true); | |
573 | spin_unlock_irq(q->queue_lock); | |
574 | } | |
575 | ||
d732580b TH |
576 | /** |
577 | * blk_queue_bypass_start - enter queue bypass mode | |
578 | * @q: queue of interest | |
579 | * | |
580 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
581 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 582 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
583 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
584 | * inside queue or RCU read lock. | |
d732580b TH |
585 | */ |
586 | void blk_queue_bypass_start(struct request_queue *q) | |
587 | { | |
332ebbf7 BVA |
588 | WARN_ON_ONCE(q->mq_ops); |
589 | ||
d732580b | 590 | spin_lock_irq(q->queue_lock); |
776687bc | 591 | q->bypass_depth++; |
d732580b TH |
592 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
593 | spin_unlock_irq(q->queue_lock); | |
594 | ||
776687bc TH |
595 | /* |
596 | * Queues start drained. Skip actual draining till init is | |
597 | * complete. This avoids lenghty delays during queue init which | |
598 | * can happen many times during boot. | |
599 | */ | |
600 | if (blk_queue_init_done(q)) { | |
807592a4 BVA |
601 | spin_lock_irq(q->queue_lock); |
602 | __blk_drain_queue(q, false); | |
603 | spin_unlock_irq(q->queue_lock); | |
604 | ||
b82d4b19 TH |
605 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
606 | synchronize_rcu(); | |
607 | } | |
d732580b TH |
608 | } |
609 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
610 | ||
611 | /** | |
612 | * blk_queue_bypass_end - leave queue bypass mode | |
613 | * @q: queue of interest | |
614 | * | |
615 | * Leave bypass mode and restore the normal queueing behavior. | |
332ebbf7 BVA |
616 | * |
617 | * Note: although blk_queue_bypass_start() is only called for blk-sq queues, | |
618 | * this function is called for both blk-sq and blk-mq queues. | |
d732580b TH |
619 | */ |
620 | void blk_queue_bypass_end(struct request_queue *q) | |
621 | { | |
622 | spin_lock_irq(q->queue_lock); | |
623 | if (!--q->bypass_depth) | |
624 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
625 | WARN_ON_ONCE(q->bypass_depth < 0); | |
626 | spin_unlock_irq(q->queue_lock); | |
627 | } | |
628 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
629 | ||
aed3ea94 JA |
630 | void blk_set_queue_dying(struct request_queue *q) |
631 | { | |
1b856086 BVA |
632 | spin_lock_irq(q->queue_lock); |
633 | queue_flag_set(QUEUE_FLAG_DYING, q); | |
634 | spin_unlock_irq(q->queue_lock); | |
aed3ea94 | 635 | |
d3cfb2a0 ML |
636 | /* |
637 | * When queue DYING flag is set, we need to block new req | |
638 | * entering queue, so we call blk_freeze_queue_start() to | |
639 | * prevent I/O from crossing blk_queue_enter(). | |
640 | */ | |
641 | blk_freeze_queue_start(q); | |
642 | ||
aed3ea94 JA |
643 | if (q->mq_ops) |
644 | blk_mq_wake_waiters(q); | |
645 | else { | |
646 | struct request_list *rl; | |
647 | ||
bbfc3c5d | 648 | spin_lock_irq(q->queue_lock); |
aed3ea94 JA |
649 | blk_queue_for_each_rl(rl, q) { |
650 | if (rl->rq_pool) { | |
34d9715a ML |
651 | wake_up_all(&rl->wait[BLK_RW_SYNC]); |
652 | wake_up_all(&rl->wait[BLK_RW_ASYNC]); | |
aed3ea94 JA |
653 | } |
654 | } | |
bbfc3c5d | 655 | spin_unlock_irq(q->queue_lock); |
aed3ea94 | 656 | } |
055f6e18 ML |
657 | |
658 | /* Make blk_queue_enter() reexamine the DYING flag. */ | |
659 | wake_up_all(&q->mq_freeze_wq); | |
aed3ea94 JA |
660 | } |
661 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); | |
662 | ||
c9a929dd TH |
663 | /** |
664 | * blk_cleanup_queue - shutdown a request queue | |
665 | * @q: request queue to shutdown | |
666 | * | |
c246e80d BVA |
667 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
668 | * put it. All future requests will be failed immediately with -ENODEV. | |
c94a96ac | 669 | */ |
6728cb0e | 670 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 671 | { |
c9a929dd | 672 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 673 | |
3f3299d5 | 674 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
483f4afc | 675 | mutex_lock(&q->sysfs_lock); |
aed3ea94 | 676 | blk_set_queue_dying(q); |
c9a929dd | 677 | spin_lock_irq(lock); |
6ecf23af | 678 | |
80fd9979 | 679 | /* |
3f3299d5 | 680 | * A dying queue is permanently in bypass mode till released. Note |
80fd9979 TH |
681 | * that, unlike blk_queue_bypass_start(), we aren't performing |
682 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
683 | * as some drivers create and destroy a lot of queues while | |
684 | * probing. This is still safe because blk_release_queue() will be | |
685 | * called only after the queue refcnt drops to zero and nothing, | |
686 | * RCU or not, would be traversing the queue by then. | |
687 | */ | |
6ecf23af TH |
688 | q->bypass_depth++; |
689 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
690 | ||
c9a929dd TH |
691 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
692 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
3f3299d5 | 693 | queue_flag_set(QUEUE_FLAG_DYING, q); |
c9a929dd TH |
694 | spin_unlock_irq(lock); |
695 | mutex_unlock(&q->sysfs_lock); | |
696 | ||
c246e80d BVA |
697 | /* |
698 | * Drain all requests queued before DYING marking. Set DEAD flag to | |
699 | * prevent that q->request_fn() gets invoked after draining finished. | |
700 | */ | |
3ef28e83 | 701 | blk_freeze_queue(q); |
9c1051aa | 702 | spin_lock_irq(lock); |
c246e80d | 703 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
807592a4 | 704 | spin_unlock_irq(lock); |
c9a929dd | 705 | |
c2856ae2 ML |
706 | /* |
707 | * make sure all in-progress dispatch are completed because | |
708 | * blk_freeze_queue() can only complete all requests, and | |
709 | * dispatch may still be in-progress since we dispatch requests | |
710 | * from more than one contexts | |
711 | */ | |
712 | if (q->mq_ops) | |
713 | blk_mq_quiesce_queue(q); | |
714 | ||
5a48fc14 DW |
715 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
716 | blk_flush_integrity(); | |
717 | ||
c9a929dd | 718 | /* @q won't process any more request, flush async actions */ |
dc3b17cc | 719 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
c9a929dd TH |
720 | blk_sync_queue(q); |
721 | ||
45a9c9d9 BVA |
722 | if (q->mq_ops) |
723 | blk_mq_free_queue(q); | |
3ef28e83 | 724 | percpu_ref_exit(&q->q_usage_counter); |
45a9c9d9 | 725 | |
5e5cfac0 AH |
726 | spin_lock_irq(lock); |
727 | if (q->queue_lock != &q->__queue_lock) | |
728 | q->queue_lock = &q->__queue_lock; | |
729 | spin_unlock_irq(lock); | |
730 | ||
c9a929dd | 731 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
732 | blk_put_queue(q); |
733 | } | |
1da177e4 LT |
734 | EXPORT_SYMBOL(blk_cleanup_queue); |
735 | ||
271508db | 736 | /* Allocate memory local to the request queue */ |
6d247d7f | 737 | static void *alloc_request_simple(gfp_t gfp_mask, void *data) |
271508db | 738 | { |
6d247d7f CH |
739 | struct request_queue *q = data; |
740 | ||
741 | return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node); | |
271508db DR |
742 | } |
743 | ||
6d247d7f | 744 | static void free_request_simple(void *element, void *data) |
271508db DR |
745 | { |
746 | kmem_cache_free(request_cachep, element); | |
747 | } | |
748 | ||
6d247d7f CH |
749 | static void *alloc_request_size(gfp_t gfp_mask, void *data) |
750 | { | |
751 | struct request_queue *q = data; | |
752 | struct request *rq; | |
753 | ||
754 | rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask, | |
755 | q->node); | |
756 | if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) { | |
757 | kfree(rq); | |
758 | rq = NULL; | |
759 | } | |
760 | return rq; | |
761 | } | |
762 | ||
763 | static void free_request_size(void *element, void *data) | |
764 | { | |
765 | struct request_queue *q = data; | |
766 | ||
767 | if (q->exit_rq_fn) | |
768 | q->exit_rq_fn(q, element); | |
769 | kfree(element); | |
770 | } | |
771 | ||
5b788ce3 TH |
772 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
773 | gfp_t gfp_mask) | |
1da177e4 | 774 | { |
85acb3ba | 775 | if (unlikely(rl->rq_pool) || q->mq_ops) |
1abec4fd MS |
776 | return 0; |
777 | ||
5b788ce3 | 778 | rl->q = q; |
1faa16d2 JA |
779 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
780 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
781 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
782 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 783 | |
6d247d7f CH |
784 | if (q->cmd_size) { |
785 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
786 | alloc_request_size, free_request_size, | |
787 | q, gfp_mask, q->node); | |
788 | } else { | |
789 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, | |
790 | alloc_request_simple, free_request_simple, | |
791 | q, gfp_mask, q->node); | |
792 | } | |
1da177e4 LT |
793 | if (!rl->rq_pool) |
794 | return -ENOMEM; | |
795 | ||
b425e504 BVA |
796 | if (rl != &q->root_rl) |
797 | WARN_ON_ONCE(!blk_get_queue(q)); | |
798 | ||
1da177e4 LT |
799 | return 0; |
800 | } | |
801 | ||
b425e504 | 802 | void blk_exit_rl(struct request_queue *q, struct request_list *rl) |
5b788ce3 | 803 | { |
b425e504 | 804 | if (rl->rq_pool) { |
5b788ce3 | 805 | mempool_destroy(rl->rq_pool); |
b425e504 BVA |
806 | if (rl != &q->root_rl) |
807 | blk_put_queue(q); | |
808 | } | |
5b788ce3 TH |
809 | } |
810 | ||
165125e1 | 811 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 812 | { |
c304a51b | 813 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
1946089a CL |
814 | } |
815 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 816 | |
3a0a5299 BVA |
817 | /** |
818 | * blk_queue_enter() - try to increase q->q_usage_counter | |
819 | * @q: request queue pointer | |
820 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT | |
821 | */ | |
9a95e4ef | 822 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
3ef28e83 | 823 | { |
3a0a5299 BVA |
824 | const bool preempt = flags & BLK_MQ_REQ_PREEMPT; |
825 | ||
3ef28e83 | 826 | while (true) { |
3a0a5299 | 827 | bool success = false; |
3ef28e83 DW |
828 | int ret; |
829 | ||
3a0a5299 BVA |
830 | rcu_read_lock_sched(); |
831 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { | |
832 | /* | |
833 | * The code that sets the PREEMPT_ONLY flag is | |
834 | * responsible for ensuring that that flag is globally | |
835 | * visible before the queue is unfrozen. | |
836 | */ | |
837 | if (preempt || !blk_queue_preempt_only(q)) { | |
838 | success = true; | |
839 | } else { | |
840 | percpu_ref_put(&q->q_usage_counter); | |
841 | } | |
842 | } | |
843 | rcu_read_unlock_sched(); | |
844 | ||
845 | if (success) | |
3ef28e83 DW |
846 | return 0; |
847 | ||
3a0a5299 | 848 | if (flags & BLK_MQ_REQ_NOWAIT) |
3ef28e83 DW |
849 | return -EBUSY; |
850 | ||
5ed61d3f | 851 | /* |
1671d522 | 852 | * read pair of barrier in blk_freeze_queue_start(), |
5ed61d3f | 853 | * we need to order reading __PERCPU_REF_DEAD flag of |
d3cfb2a0 ML |
854 | * .q_usage_counter and reading .mq_freeze_depth or |
855 | * queue dying flag, otherwise the following wait may | |
856 | * never return if the two reads are reordered. | |
5ed61d3f ML |
857 | */ |
858 | smp_rmb(); | |
859 | ||
3ef28e83 | 860 | ret = wait_event_interruptible(q->mq_freeze_wq, |
3a0a5299 BVA |
861 | (atomic_read(&q->mq_freeze_depth) == 0 && |
862 | (preempt || !blk_queue_preempt_only(q))) || | |
3ef28e83 DW |
863 | blk_queue_dying(q)); |
864 | if (blk_queue_dying(q)) | |
865 | return -ENODEV; | |
866 | if (ret) | |
867 | return ret; | |
868 | } | |
869 | } | |
870 | ||
871 | void blk_queue_exit(struct request_queue *q) | |
872 | { | |
873 | percpu_ref_put(&q->q_usage_counter); | |
874 | } | |
875 | ||
876 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) | |
877 | { | |
878 | struct request_queue *q = | |
879 | container_of(ref, struct request_queue, q_usage_counter); | |
880 | ||
881 | wake_up_all(&q->mq_freeze_wq); | |
882 | } | |
883 | ||
bca237a5 | 884 | static void blk_rq_timed_out_timer(struct timer_list *t) |
287922eb | 885 | { |
bca237a5 | 886 | struct request_queue *q = from_timer(q, t, timeout); |
287922eb CH |
887 | |
888 | kblockd_schedule_work(&q->timeout_work); | |
889 | } | |
890 | ||
165125e1 | 891 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 892 | { |
165125e1 | 893 | struct request_queue *q; |
1946089a | 894 | |
8324aa91 | 895 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 896 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
897 | if (!q) |
898 | return NULL; | |
899 | ||
00380a40 | 900 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d | 901 | if (q->id < 0) |
3d2936f4 | 902 | goto fail_q; |
a73f730d | 903 | |
93b27e72 | 904 | q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
54efd50b KO |
905 | if (!q->bio_split) |
906 | goto fail_id; | |
907 | ||
d03f6cdc JK |
908 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
909 | if (!q->backing_dev_info) | |
910 | goto fail_split; | |
911 | ||
a83b576c JA |
912 | q->stats = blk_alloc_queue_stats(); |
913 | if (!q->stats) | |
914 | goto fail_stats; | |
915 | ||
dc3b17cc | 916 | q->backing_dev_info->ra_pages = |
09cbfeaf | 917 | (VM_MAX_READAHEAD * 1024) / PAGE_SIZE; |
dc3b17cc JK |
918 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
919 | q->backing_dev_info->name = "block"; | |
5151412d | 920 | q->node = node_id; |
0989a025 | 921 | |
bca237a5 KC |
922 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
923 | laptop_mode_timer_fn, 0); | |
924 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); | |
4e9b6f20 | 925 | INIT_WORK(&q->timeout_work, NULL); |
b855b04a | 926 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 927 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 928 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 929 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 930 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 931 | #endif |
3cca6dc1 | 932 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 933 | |
8324aa91 | 934 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 935 | |
5acb3cc2 WL |
936 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
937 | mutex_init(&q->blk_trace_mutex); | |
938 | #endif | |
483f4afc | 939 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 940 | spin_lock_init(&q->__queue_lock); |
483f4afc | 941 | |
c94a96ac VG |
942 | /* |
943 | * By default initialize queue_lock to internal lock and driver can | |
944 | * override it later if need be. | |
945 | */ | |
946 | q->queue_lock = &q->__queue_lock; | |
947 | ||
b82d4b19 TH |
948 | /* |
949 | * A queue starts its life with bypass turned on to avoid | |
950 | * unnecessary bypass on/off overhead and nasty surprises during | |
749fefe6 TH |
951 | * init. The initial bypass will be finished when the queue is |
952 | * registered by blk_register_queue(). | |
b82d4b19 TH |
953 | */ |
954 | q->bypass_depth = 1; | |
955 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
956 | ||
320ae51f JA |
957 | init_waitqueue_head(&q->mq_freeze_wq); |
958 | ||
3ef28e83 DW |
959 | /* |
960 | * Init percpu_ref in atomic mode so that it's faster to shutdown. | |
961 | * See blk_register_queue() for details. | |
962 | */ | |
963 | if (percpu_ref_init(&q->q_usage_counter, | |
964 | blk_queue_usage_counter_release, | |
965 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) | |
fff4996b | 966 | goto fail_bdi; |
f51b802c | 967 | |
3ef28e83 DW |
968 | if (blkcg_init_queue(q)) |
969 | goto fail_ref; | |
970 | ||
1da177e4 | 971 | return q; |
a73f730d | 972 | |
3ef28e83 DW |
973 | fail_ref: |
974 | percpu_ref_exit(&q->q_usage_counter); | |
fff4996b | 975 | fail_bdi: |
a83b576c JA |
976 | blk_free_queue_stats(q->stats); |
977 | fail_stats: | |
d03f6cdc | 978 | bdi_put(q->backing_dev_info); |
54efd50b KO |
979 | fail_split: |
980 | bioset_free(q->bio_split); | |
a73f730d TH |
981 | fail_id: |
982 | ida_simple_remove(&blk_queue_ida, q->id); | |
983 | fail_q: | |
984 | kmem_cache_free(blk_requestq_cachep, q); | |
985 | return NULL; | |
1da177e4 | 986 | } |
1946089a | 987 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
988 | |
989 | /** | |
990 | * blk_init_queue - prepare a request queue for use with a block device | |
991 | * @rfn: The function to be called to process requests that have been | |
992 | * placed on the queue. | |
993 | * @lock: Request queue spin lock | |
994 | * | |
995 | * Description: | |
996 | * If a block device wishes to use the standard request handling procedures, | |
997 | * which sorts requests and coalesces adjacent requests, then it must | |
998 | * call blk_init_queue(). The function @rfn will be called when there | |
999 | * are requests on the queue that need to be processed. If the device | |
1000 | * supports plugging, then @rfn may not be called immediately when requests | |
1001 | * are available on the queue, but may be called at some time later instead. | |
1002 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1003 | * of the requests on the queue is needed, or due to memory pressure. | |
1004 | * | |
1005 | * @rfn is not required, or even expected, to remove all requests off the | |
1006 | * queue, but only as many as it can handle at a time. If it does leave | |
1007 | * requests on the queue, it is responsible for arranging that the requests | |
1008 | * get dealt with eventually. | |
1009 | * | |
1010 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1011 | * request queue; this lock will be taken also from interrupt context, so irq |
1012 | * disabling is needed for it. | |
1da177e4 | 1013 | * |
710027a4 | 1014 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
1015 | * it didn't succeed. |
1016 | * | |
1017 | * Note: | |
1018 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1019 | * when the block device is deactivated (such as at module unload). | |
1020 | **/ | |
1946089a | 1021 | |
165125e1 | 1022 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 1023 | { |
c304a51b | 1024 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
1946089a CL |
1025 | } |
1026 | EXPORT_SYMBOL(blk_init_queue); | |
1027 | ||
165125e1 | 1028 | struct request_queue * |
1946089a CL |
1029 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
1030 | { | |
5ea708d1 | 1031 | struct request_queue *q; |
1da177e4 | 1032 | |
5ea708d1 CH |
1033 | q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1034 | if (!q) | |
c86d1b8a MS |
1035 | return NULL; |
1036 | ||
5ea708d1 CH |
1037 | q->request_fn = rfn; |
1038 | if (lock) | |
1039 | q->queue_lock = lock; | |
1040 | if (blk_init_allocated_queue(q) < 0) { | |
1041 | blk_cleanup_queue(q); | |
1042 | return NULL; | |
1043 | } | |
18741986 | 1044 | |
7982e90c | 1045 | return q; |
01effb0d MS |
1046 | } |
1047 | EXPORT_SYMBOL(blk_init_queue_node); | |
1048 | ||
dece1635 | 1049 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio); |
336b7e1f | 1050 | |
1da177e4 | 1051 | |
5ea708d1 CH |
1052 | int blk_init_allocated_queue(struct request_queue *q) |
1053 | { | |
332ebbf7 BVA |
1054 | WARN_ON_ONCE(q->mq_ops); |
1055 | ||
6d247d7f | 1056 | q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size); |
ba483388 | 1057 | if (!q->fq) |
5ea708d1 | 1058 | return -ENOMEM; |
7982e90c | 1059 | |
6d247d7f CH |
1060 | if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL)) |
1061 | goto out_free_flush_queue; | |
7982e90c | 1062 | |
a051661c | 1063 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
6d247d7f | 1064 | goto out_exit_flush_rq; |
1da177e4 | 1065 | |
287922eb | 1066 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
60ea8226 | 1067 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
c94a96ac | 1068 | |
f3b144aa JA |
1069 | /* |
1070 | * This also sets hw/phys segments, boundary and size | |
1071 | */ | |
c20e8de2 | 1072 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 1073 | |
44ec9542 AS |
1074 | q->sg_reserved_size = INT_MAX; |
1075 | ||
eb1c160b TS |
1076 | /* Protect q->elevator from elevator_change */ |
1077 | mutex_lock(&q->sysfs_lock); | |
1078 | ||
b82d4b19 | 1079 | /* init elevator */ |
eb1c160b TS |
1080 | if (elevator_init(q, NULL)) { |
1081 | mutex_unlock(&q->sysfs_lock); | |
6d247d7f | 1082 | goto out_exit_flush_rq; |
eb1c160b TS |
1083 | } |
1084 | ||
1085 | mutex_unlock(&q->sysfs_lock); | |
5ea708d1 | 1086 | return 0; |
eb1c160b | 1087 | |
6d247d7f CH |
1088 | out_exit_flush_rq: |
1089 | if (q->exit_rq_fn) | |
1090 | q->exit_rq_fn(q, q->fq->flush_rq); | |
1091 | out_free_flush_queue: | |
ba483388 | 1092 | blk_free_flush_queue(q->fq); |
5ea708d1 | 1093 | return -ENOMEM; |
1da177e4 | 1094 | } |
5151412d | 1095 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 1096 | |
09ac46c4 | 1097 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 1098 | { |
3f3299d5 | 1099 | if (likely(!blk_queue_dying(q))) { |
09ac46c4 TH |
1100 | __blk_get_queue(q); |
1101 | return true; | |
1da177e4 LT |
1102 | } |
1103 | ||
09ac46c4 | 1104 | return false; |
1da177e4 | 1105 | } |
d86e0e83 | 1106 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 1107 | |
5b788ce3 | 1108 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 1109 | { |
e8064021 | 1110 | if (rq->rq_flags & RQF_ELVPRIV) { |
5b788ce3 | 1111 | elv_put_request(rl->q, rq); |
f1f8cc94 | 1112 | if (rq->elv.icq) |
11a3122f | 1113 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
1114 | } |
1115 | ||
5b788ce3 | 1116 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
1117 | } |
1118 | ||
1da177e4 LT |
1119 | /* |
1120 | * ioc_batching returns true if the ioc is a valid batching request and | |
1121 | * should be given priority access to a request. | |
1122 | */ | |
165125e1 | 1123 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1124 | { |
1125 | if (!ioc) | |
1126 | return 0; | |
1127 | ||
1128 | /* | |
1129 | * Make sure the process is able to allocate at least 1 request | |
1130 | * even if the batch times out, otherwise we could theoretically | |
1131 | * lose wakeups. | |
1132 | */ | |
1133 | return ioc->nr_batch_requests == q->nr_batching || | |
1134 | (ioc->nr_batch_requests > 0 | |
1135 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
1140 | * will cause the process to be a "batcher" on all queues in the system. This | |
1141 | * is the behaviour we want though - once it gets a wakeup it should be given | |
1142 | * a nice run. | |
1143 | */ | |
165125e1 | 1144 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
1145 | { |
1146 | if (!ioc || ioc_batching(q, ioc)) | |
1147 | return; | |
1148 | ||
1149 | ioc->nr_batch_requests = q->nr_batching; | |
1150 | ioc->last_waited = jiffies; | |
1151 | } | |
1152 | ||
5b788ce3 | 1153 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 1154 | { |
5b788ce3 | 1155 | struct request_queue *q = rl->q; |
1da177e4 | 1156 | |
d40f75a0 TH |
1157 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
1158 | blk_clear_congested(rl, sync); | |
1da177e4 | 1159 | |
1faa16d2 JA |
1160 | if (rl->count[sync] + 1 <= q->nr_requests) { |
1161 | if (waitqueue_active(&rl->wait[sync])) | |
1162 | wake_up(&rl->wait[sync]); | |
1da177e4 | 1163 | |
5b788ce3 | 1164 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
1165 | } |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * A request has just been released. Account for it, update the full and | |
1170 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
1171 | */ | |
e8064021 CH |
1172 | static void freed_request(struct request_list *rl, bool sync, |
1173 | req_flags_t rq_flags) | |
1da177e4 | 1174 | { |
5b788ce3 | 1175 | struct request_queue *q = rl->q; |
1da177e4 | 1176 | |
8a5ecdd4 | 1177 | q->nr_rqs[sync]--; |
1faa16d2 | 1178 | rl->count[sync]--; |
e8064021 | 1179 | if (rq_flags & RQF_ELVPRIV) |
8a5ecdd4 | 1180 | q->nr_rqs_elvpriv--; |
1da177e4 | 1181 | |
5b788ce3 | 1182 | __freed_request(rl, sync); |
1da177e4 | 1183 | |
1faa16d2 | 1184 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 1185 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
1186 | } |
1187 | ||
e3a2b3f9 JA |
1188 | int blk_update_nr_requests(struct request_queue *q, unsigned int nr) |
1189 | { | |
1190 | struct request_list *rl; | |
d40f75a0 | 1191 | int on_thresh, off_thresh; |
e3a2b3f9 | 1192 | |
332ebbf7 BVA |
1193 | WARN_ON_ONCE(q->mq_ops); |
1194 | ||
e3a2b3f9 JA |
1195 | spin_lock_irq(q->queue_lock); |
1196 | q->nr_requests = nr; | |
1197 | blk_queue_congestion_threshold(q); | |
d40f75a0 TH |
1198 | on_thresh = queue_congestion_on_threshold(q); |
1199 | off_thresh = queue_congestion_off_threshold(q); | |
e3a2b3f9 | 1200 | |
d40f75a0 TH |
1201 | blk_queue_for_each_rl(rl, q) { |
1202 | if (rl->count[BLK_RW_SYNC] >= on_thresh) | |
1203 | blk_set_congested(rl, BLK_RW_SYNC); | |
1204 | else if (rl->count[BLK_RW_SYNC] < off_thresh) | |
1205 | blk_clear_congested(rl, BLK_RW_SYNC); | |
e3a2b3f9 | 1206 | |
d40f75a0 TH |
1207 | if (rl->count[BLK_RW_ASYNC] >= on_thresh) |
1208 | blk_set_congested(rl, BLK_RW_ASYNC); | |
1209 | else if (rl->count[BLK_RW_ASYNC] < off_thresh) | |
1210 | blk_clear_congested(rl, BLK_RW_ASYNC); | |
e3a2b3f9 | 1211 | |
e3a2b3f9 JA |
1212 | if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { |
1213 | blk_set_rl_full(rl, BLK_RW_SYNC); | |
1214 | } else { | |
1215 | blk_clear_rl_full(rl, BLK_RW_SYNC); | |
1216 | wake_up(&rl->wait[BLK_RW_SYNC]); | |
1217 | } | |
1218 | ||
1219 | if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { | |
1220 | blk_set_rl_full(rl, BLK_RW_ASYNC); | |
1221 | } else { | |
1222 | blk_clear_rl_full(rl, BLK_RW_ASYNC); | |
1223 | wake_up(&rl->wait[BLK_RW_ASYNC]); | |
1224 | } | |
1225 | } | |
1226 | ||
1227 | spin_unlock_irq(q->queue_lock); | |
1228 | return 0; | |
1229 | } | |
1230 | ||
da8303c6 | 1231 | /** |
a06e05e6 | 1232 | * __get_request - get a free request |
5b788ce3 | 1233 | * @rl: request list to allocate from |
ef295ecf | 1234 | * @op: operation and flags |
da8303c6 | 1235 | * @bio: bio to allocate request for (can be %NULL) |
6a15674d | 1236 | * @flags: BLQ_MQ_REQ_* flags |
da8303c6 TH |
1237 | * |
1238 | * Get a free request from @q. This function may fail under memory | |
1239 | * pressure or if @q is dead. | |
1240 | * | |
da3dae54 | 1241 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1242 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1243 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1244 | */ |
ef295ecf | 1245 | static struct request *__get_request(struct request_list *rl, unsigned int op, |
9a95e4ef | 1246 | struct bio *bio, blk_mq_req_flags_t flags) |
1da177e4 | 1247 | { |
5b788ce3 | 1248 | struct request_queue *q = rl->q; |
b679281a | 1249 | struct request *rq; |
7f4b35d1 TH |
1250 | struct elevator_type *et = q->elevator->type; |
1251 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 1252 | struct io_cq *icq = NULL; |
ef295ecf | 1253 | const bool is_sync = op_is_sync(op); |
75eb6c37 | 1254 | int may_queue; |
6a15674d BVA |
1255 | gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : |
1256 | __GFP_DIRECT_RECLAIM; | |
e8064021 | 1257 | req_flags_t rq_flags = RQF_ALLOCED; |
88ee5ef1 | 1258 | |
2fff8a92 BVA |
1259 | lockdep_assert_held(q->queue_lock); |
1260 | ||
3f3299d5 | 1261 | if (unlikely(blk_queue_dying(q))) |
a492f075 | 1262 | return ERR_PTR(-ENODEV); |
da8303c6 | 1263 | |
ef295ecf | 1264 | may_queue = elv_may_queue(q, op); |
88ee5ef1 JA |
1265 | if (may_queue == ELV_MQUEUE_NO) |
1266 | goto rq_starved; | |
1267 | ||
1faa16d2 JA |
1268 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
1269 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
1270 | /* |
1271 | * The queue will fill after this allocation, so set | |
1272 | * it as full, and mark this process as "batching". | |
1273 | * This process will be allowed to complete a batch of | |
1274 | * requests, others will be blocked. | |
1275 | */ | |
5b788ce3 | 1276 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 1277 | ioc_set_batching(q, ioc); |
5b788ce3 | 1278 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
1279 | } else { |
1280 | if (may_queue != ELV_MQUEUE_MUST | |
1281 | && !ioc_batching(q, ioc)) { | |
1282 | /* | |
1283 | * The queue is full and the allocating | |
1284 | * process is not a "batcher", and not | |
1285 | * exempted by the IO scheduler | |
1286 | */ | |
a492f075 | 1287 | return ERR_PTR(-ENOMEM); |
88ee5ef1 JA |
1288 | } |
1289 | } | |
1da177e4 | 1290 | } |
d40f75a0 | 1291 | blk_set_congested(rl, is_sync); |
1da177e4 LT |
1292 | } |
1293 | ||
082cf69e JA |
1294 | /* |
1295 | * Only allow batching queuers to allocate up to 50% over the defined | |
1296 | * limit of requests, otherwise we could have thousands of requests | |
1297 | * allocated with any setting of ->nr_requests | |
1298 | */ | |
1faa16d2 | 1299 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
a492f075 | 1300 | return ERR_PTR(-ENOMEM); |
fd782a4a | 1301 | |
8a5ecdd4 | 1302 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
1303 | rl->count[is_sync]++; |
1304 | rl->starved[is_sync] = 0; | |
cb98fc8b | 1305 | |
f1f8cc94 TH |
1306 | /* |
1307 | * Decide whether the new request will be managed by elevator. If | |
e8064021 | 1308 | * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will |
f1f8cc94 TH |
1309 | * prevent the current elevator from being destroyed until the new |
1310 | * request is freed. This guarantees icq's won't be destroyed and | |
1311 | * makes creating new ones safe. | |
1312 | * | |
e6f7f93d CH |
1313 | * Flush requests do not use the elevator so skip initialization. |
1314 | * This allows a request to share the flush and elevator data. | |
1315 | * | |
f1f8cc94 TH |
1316 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
1317 | * it will be created after releasing queue_lock. | |
1318 | */ | |
e6f7f93d | 1319 | if (!op_is_flush(op) && !blk_queue_bypass(q)) { |
e8064021 | 1320 | rq_flags |= RQF_ELVPRIV; |
8a5ecdd4 | 1321 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
1322 | if (et->icq_cache && ioc) |
1323 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 1324 | } |
cb98fc8b | 1325 | |
f253b86b | 1326 | if (blk_queue_io_stat(q)) |
e8064021 | 1327 | rq_flags |= RQF_IO_STAT; |
1da177e4 LT |
1328 | spin_unlock_irq(q->queue_lock); |
1329 | ||
29e2b09a | 1330 | /* allocate and init request */ |
5b788ce3 | 1331 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 1332 | if (!rq) |
b679281a | 1333 | goto fail_alloc; |
1da177e4 | 1334 | |
29e2b09a | 1335 | blk_rq_init(q, rq); |
a051661c | 1336 | blk_rq_set_rl(rq, rl); |
ef295ecf | 1337 | rq->cmd_flags = op; |
e8064021 | 1338 | rq->rq_flags = rq_flags; |
1b6d65a0 BVA |
1339 | if (flags & BLK_MQ_REQ_PREEMPT) |
1340 | rq->rq_flags |= RQF_PREEMPT; | |
29e2b09a | 1341 | |
aaf7c680 | 1342 | /* init elvpriv */ |
e8064021 | 1343 | if (rq_flags & RQF_ELVPRIV) { |
aaf7c680 | 1344 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
1345 | if (ioc) |
1346 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
1347 | if (!icq) |
1348 | goto fail_elvpriv; | |
29e2b09a | 1349 | } |
aaf7c680 TH |
1350 | |
1351 | rq->elv.icq = icq; | |
1352 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
1353 | goto fail_elvpriv; | |
1354 | ||
1355 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
1356 | if (icq) |
1357 | get_io_context(icq->ioc); | |
1358 | } | |
aaf7c680 | 1359 | out: |
88ee5ef1 JA |
1360 | /* |
1361 | * ioc may be NULL here, and ioc_batching will be false. That's | |
1362 | * OK, if the queue is under the request limit then requests need | |
1363 | * not count toward the nr_batch_requests limit. There will always | |
1364 | * be some limit enforced by BLK_BATCH_TIME. | |
1365 | */ | |
1da177e4 LT |
1366 | if (ioc_batching(q, ioc)) |
1367 | ioc->nr_batch_requests--; | |
6728cb0e | 1368 | |
e6a40b09 | 1369 | trace_block_getrq(q, bio, op); |
1da177e4 | 1370 | return rq; |
b679281a | 1371 | |
aaf7c680 TH |
1372 | fail_elvpriv: |
1373 | /* | |
1374 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
1375 | * and may fail indefinitely under memory pressure and thus | |
1376 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
1377 | * disturb iosched and blkcg but weird is bettern than dead. | |
1378 | */ | |
7b2b10e0 | 1379 | printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n", |
dc3b17cc | 1380 | __func__, dev_name(q->backing_dev_info->dev)); |
aaf7c680 | 1381 | |
e8064021 | 1382 | rq->rq_flags &= ~RQF_ELVPRIV; |
aaf7c680 TH |
1383 | rq->elv.icq = NULL; |
1384 | ||
1385 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1386 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1387 | spin_unlock_irq(q->queue_lock); |
1388 | goto out; | |
1389 | ||
b679281a TH |
1390 | fail_alloc: |
1391 | /* | |
1392 | * Allocation failed presumably due to memory. Undo anything we | |
1393 | * might have messed up. | |
1394 | * | |
1395 | * Allocating task should really be put onto the front of the wait | |
1396 | * queue, but this is pretty rare. | |
1397 | */ | |
1398 | spin_lock_irq(q->queue_lock); | |
e8064021 | 1399 | freed_request(rl, is_sync, rq_flags); |
b679281a TH |
1400 | |
1401 | /* | |
1402 | * in the very unlikely event that allocation failed and no | |
1403 | * requests for this direction was pending, mark us starved so that | |
1404 | * freeing of a request in the other direction will notice | |
1405 | * us. another possible fix would be to split the rq mempool into | |
1406 | * READ and WRITE | |
1407 | */ | |
1408 | rq_starved: | |
1409 | if (unlikely(rl->count[is_sync] == 0)) | |
1410 | rl->starved[is_sync] = 1; | |
a492f075 | 1411 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1412 | } |
1413 | ||
da8303c6 | 1414 | /** |
a06e05e6 | 1415 | * get_request - get a free request |
da8303c6 | 1416 | * @q: request_queue to allocate request from |
ef295ecf | 1417 | * @op: operation and flags |
da8303c6 | 1418 | * @bio: bio to allocate request for (can be %NULL) |
6a15674d | 1419 | * @flags: BLK_MQ_REQ_* flags. |
da8303c6 | 1420 | * |
d0164adc MG |
1421 | * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask, |
1422 | * this function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1423 | * |
da3dae54 | 1424 | * Must be called with @q->queue_lock held and, |
a492f075 JL |
1425 | * Returns ERR_PTR on failure, with @q->queue_lock held. |
1426 | * Returns request pointer on success, with @q->queue_lock *not held*. | |
1da177e4 | 1427 | */ |
ef295ecf | 1428 | static struct request *get_request(struct request_queue *q, unsigned int op, |
9a95e4ef | 1429 | struct bio *bio, blk_mq_req_flags_t flags) |
1da177e4 | 1430 | { |
ef295ecf | 1431 | const bool is_sync = op_is_sync(op); |
a06e05e6 | 1432 | DEFINE_WAIT(wait); |
a051661c | 1433 | struct request_list *rl; |
1da177e4 | 1434 | struct request *rq; |
a051661c | 1435 | |
2fff8a92 | 1436 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1437 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1438 | |
a051661c | 1439 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ |
a06e05e6 | 1440 | retry: |
6a15674d | 1441 | rq = __get_request(rl, op, bio, flags); |
a492f075 | 1442 | if (!IS_ERR(rq)) |
a06e05e6 | 1443 | return rq; |
1da177e4 | 1444 | |
03a07c92 GR |
1445 | if (op & REQ_NOWAIT) { |
1446 | blk_put_rl(rl); | |
1447 | return ERR_PTR(-EAGAIN); | |
1448 | } | |
1449 | ||
6a15674d | 1450 | if ((flags & BLK_MQ_REQ_NOWAIT) || unlikely(blk_queue_dying(q))) { |
a051661c | 1451 | blk_put_rl(rl); |
a492f075 | 1452 | return rq; |
a051661c | 1453 | } |
1da177e4 | 1454 | |
a06e05e6 TH |
1455 | /* wait on @rl and retry */ |
1456 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1457 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1458 | |
e6a40b09 | 1459 | trace_block_sleeprq(q, bio, op); |
1da177e4 | 1460 | |
a06e05e6 TH |
1461 | spin_unlock_irq(q->queue_lock); |
1462 | io_schedule(); | |
d6344532 | 1463 | |
a06e05e6 TH |
1464 | /* |
1465 | * After sleeping, we become a "batching" process and will be able | |
1466 | * to allocate at least one request, and up to a big batch of them | |
1467 | * for a small period time. See ioc_batching, ioc_set_batching | |
1468 | */ | |
a06e05e6 | 1469 | ioc_set_batching(q, current->io_context); |
05caf8db | 1470 | |
a06e05e6 TH |
1471 | spin_lock_irq(q->queue_lock); |
1472 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1473 | |
a06e05e6 | 1474 | goto retry; |
1da177e4 LT |
1475 | } |
1476 | ||
6a15674d | 1477 | /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */ |
cd6ce148 | 1478 | static struct request *blk_old_get_request(struct request_queue *q, |
9a95e4ef | 1479 | unsigned int op, blk_mq_req_flags_t flags) |
1da177e4 LT |
1480 | { |
1481 | struct request *rq; | |
6a15674d BVA |
1482 | gfp_t gfp_mask = flags & BLK_MQ_REQ_NOWAIT ? GFP_ATOMIC : |
1483 | __GFP_DIRECT_RECLAIM; | |
055f6e18 | 1484 | int ret = 0; |
1da177e4 | 1485 | |
332ebbf7 BVA |
1486 | WARN_ON_ONCE(q->mq_ops); |
1487 | ||
7f4b35d1 TH |
1488 | /* create ioc upfront */ |
1489 | create_io_context(gfp_mask, q->node); | |
1490 | ||
3a0a5299 | 1491 | ret = blk_queue_enter(q, flags); |
055f6e18 ML |
1492 | if (ret) |
1493 | return ERR_PTR(ret); | |
d6344532 | 1494 | spin_lock_irq(q->queue_lock); |
6a15674d | 1495 | rq = get_request(q, op, NULL, flags); |
0c4de0f3 | 1496 | if (IS_ERR(rq)) { |
da8303c6 | 1497 | spin_unlock_irq(q->queue_lock); |
055f6e18 | 1498 | blk_queue_exit(q); |
0c4de0f3 CH |
1499 | return rq; |
1500 | } | |
1da177e4 | 1501 | |
0c4de0f3 CH |
1502 | /* q->queue_lock is unlocked at this point */ |
1503 | rq->__data_len = 0; | |
1504 | rq->__sector = (sector_t) -1; | |
1505 | rq->bio = rq->biotail = NULL; | |
1da177e4 LT |
1506 | return rq; |
1507 | } | |
320ae51f | 1508 | |
6a15674d BVA |
1509 | /** |
1510 | * blk_get_request_flags - allocate a request | |
1511 | * @q: request queue to allocate a request for | |
1512 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. | |
1513 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. | |
1514 | */ | |
1515 | struct request *blk_get_request_flags(struct request_queue *q, unsigned int op, | |
9a95e4ef | 1516 | blk_mq_req_flags_t flags) |
320ae51f | 1517 | { |
d280bab3 BVA |
1518 | struct request *req; |
1519 | ||
6a15674d | 1520 | WARN_ON_ONCE(op & REQ_NOWAIT); |
1b6d65a0 | 1521 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); |
6a15674d | 1522 | |
d280bab3 | 1523 | if (q->mq_ops) { |
6a15674d | 1524 | req = blk_mq_alloc_request(q, op, flags); |
d280bab3 BVA |
1525 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) |
1526 | q->mq_ops->initialize_rq_fn(req); | |
1527 | } else { | |
6a15674d | 1528 | req = blk_old_get_request(q, op, flags); |
d280bab3 BVA |
1529 | if (!IS_ERR(req) && q->initialize_rq_fn) |
1530 | q->initialize_rq_fn(req); | |
1531 | } | |
1532 | ||
1533 | return req; | |
320ae51f | 1534 | } |
6a15674d BVA |
1535 | EXPORT_SYMBOL(blk_get_request_flags); |
1536 | ||
1537 | struct request *blk_get_request(struct request_queue *q, unsigned int op, | |
1538 | gfp_t gfp_mask) | |
1539 | { | |
1540 | return blk_get_request_flags(q, op, gfp_mask & __GFP_DIRECT_RECLAIM ? | |
1541 | 0 : BLK_MQ_REQ_NOWAIT); | |
1542 | } | |
1da177e4 LT |
1543 | EXPORT_SYMBOL(blk_get_request); |
1544 | ||
1545 | /** | |
1546 | * blk_requeue_request - put a request back on queue | |
1547 | * @q: request queue where request should be inserted | |
1548 | * @rq: request to be inserted | |
1549 | * | |
1550 | * Description: | |
1551 | * Drivers often keep queueing requests until the hardware cannot accept | |
1552 | * more, when that condition happens we need to put the request back | |
1553 | * on the queue. Must be called with queue lock held. | |
1554 | */ | |
165125e1 | 1555 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1556 | { |
2fff8a92 | 1557 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 1558 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 1559 | |
242f9dcb JA |
1560 | blk_delete_timer(rq); |
1561 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1562 | trace_block_rq_requeue(q, rq); |
87760e5e | 1563 | wbt_requeue(q->rq_wb, &rq->issue_stat); |
2056a782 | 1564 | |
e8064021 | 1565 | if (rq->rq_flags & RQF_QUEUED) |
1da177e4 LT |
1566 | blk_queue_end_tag(q, rq); |
1567 | ||
ba396a6c JB |
1568 | BUG_ON(blk_queued_rq(rq)); |
1569 | ||
1da177e4 LT |
1570 | elv_requeue_request(q, rq); |
1571 | } | |
1da177e4 LT |
1572 | EXPORT_SYMBOL(blk_requeue_request); |
1573 | ||
73c10101 JA |
1574 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1575 | int where) | |
1576 | { | |
320ae51f | 1577 | blk_account_io_start(rq, true); |
7eaceacc | 1578 | __elv_add_request(q, rq, where); |
73c10101 JA |
1579 | } |
1580 | ||
d62e26b3 | 1581 | static void part_round_stats_single(struct request_queue *q, int cpu, |
b8d62b3a JA |
1582 | struct hd_struct *part, unsigned long now, |
1583 | unsigned int inflight) | |
074a7aca | 1584 | { |
b8d62b3a | 1585 | if (inflight) { |
074a7aca | 1586 | __part_stat_add(cpu, part, time_in_queue, |
b8d62b3a | 1587 | inflight * (now - part->stamp)); |
074a7aca TH |
1588 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1589 | } | |
1590 | part->stamp = now; | |
1591 | } | |
1592 | ||
1593 | /** | |
496aa8a9 | 1594 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
d62e26b3 | 1595 | * @q: target block queue |
496aa8a9 RD |
1596 | * @cpu: cpu number for stats access |
1597 | * @part: target partition | |
1da177e4 LT |
1598 | * |
1599 | * The average IO queue length and utilisation statistics are maintained | |
1600 | * by observing the current state of the queue length and the amount of | |
1601 | * time it has been in this state for. | |
1602 | * | |
1603 | * Normally, that accounting is done on IO completion, but that can result | |
1604 | * in more than a second's worth of IO being accounted for within any one | |
1605 | * second, leading to >100% utilisation. To deal with that, we call this | |
1606 | * function to do a round-off before returning the results when reading | |
1607 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1608 | * the current jiffies and restarts the counters again. | |
1609 | */ | |
d62e26b3 | 1610 | void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part) |
6f2576af | 1611 | { |
b8d62b3a | 1612 | struct hd_struct *part2 = NULL; |
6f2576af | 1613 | unsigned long now = jiffies; |
b8d62b3a JA |
1614 | unsigned int inflight[2]; |
1615 | int stats = 0; | |
1616 | ||
1617 | if (part->stamp != now) | |
1618 | stats |= 1; | |
1619 | ||
1620 | if (part->partno) { | |
1621 | part2 = &part_to_disk(part)->part0; | |
1622 | if (part2->stamp != now) | |
1623 | stats |= 2; | |
1624 | } | |
1625 | ||
1626 | if (!stats) | |
1627 | return; | |
1628 | ||
1629 | part_in_flight(q, part, inflight); | |
6f2576af | 1630 | |
b8d62b3a JA |
1631 | if (stats & 2) |
1632 | part_round_stats_single(q, cpu, part2, now, inflight[1]); | |
1633 | if (stats & 1) | |
1634 | part_round_stats_single(q, cpu, part, now, inflight[0]); | |
6f2576af | 1635 | } |
074a7aca | 1636 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1637 | |
47fafbc7 | 1638 | #ifdef CONFIG_PM |
c8158819 LM |
1639 | static void blk_pm_put_request(struct request *rq) |
1640 | { | |
e8064021 | 1641 | if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending) |
c8158819 LM |
1642 | pm_runtime_mark_last_busy(rq->q->dev); |
1643 | } | |
1644 | #else | |
1645 | static inline void blk_pm_put_request(struct request *rq) {} | |
1646 | #endif | |
1647 | ||
165125e1 | 1648 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1649 | { |
e8064021 CH |
1650 | req_flags_t rq_flags = req->rq_flags; |
1651 | ||
1da177e4 LT |
1652 | if (unlikely(!q)) |
1653 | return; | |
1da177e4 | 1654 | |
6f5ba581 CH |
1655 | if (q->mq_ops) { |
1656 | blk_mq_free_request(req); | |
1657 | return; | |
1658 | } | |
1659 | ||
2fff8a92 BVA |
1660 | lockdep_assert_held(q->queue_lock); |
1661 | ||
6cc77e9c | 1662 | blk_req_zone_write_unlock(req); |
c8158819 LM |
1663 | blk_pm_put_request(req); |
1664 | ||
8922e16c TH |
1665 | elv_completed_request(q, req); |
1666 | ||
1cd96c24 BH |
1667 | /* this is a bio leak */ |
1668 | WARN_ON(req->bio != NULL); | |
1669 | ||
87760e5e JA |
1670 | wbt_done(q->rq_wb, &req->issue_stat); |
1671 | ||
1da177e4 LT |
1672 | /* |
1673 | * Request may not have originated from ll_rw_blk. if not, | |
1674 | * it didn't come out of our reserved rq pools | |
1675 | */ | |
e8064021 | 1676 | if (rq_flags & RQF_ALLOCED) { |
a051661c | 1677 | struct request_list *rl = blk_rq_rl(req); |
ef295ecf | 1678 | bool sync = op_is_sync(req->cmd_flags); |
1da177e4 | 1679 | |
1da177e4 | 1680 | BUG_ON(!list_empty(&req->queuelist)); |
360f92c2 | 1681 | BUG_ON(ELV_ON_HASH(req)); |
1da177e4 | 1682 | |
a051661c | 1683 | blk_free_request(rl, req); |
e8064021 | 1684 | freed_request(rl, sync, rq_flags); |
a051661c | 1685 | blk_put_rl(rl); |
055f6e18 | 1686 | blk_queue_exit(q); |
1da177e4 LT |
1687 | } |
1688 | } | |
6e39b69e MC |
1689 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1690 | ||
1da177e4 LT |
1691 | void blk_put_request(struct request *req) |
1692 | { | |
165125e1 | 1693 | struct request_queue *q = req->q; |
8922e16c | 1694 | |
320ae51f JA |
1695 | if (q->mq_ops) |
1696 | blk_mq_free_request(req); | |
1697 | else { | |
1698 | unsigned long flags; | |
1699 | ||
1700 | spin_lock_irqsave(q->queue_lock, flags); | |
1701 | __blk_put_request(q, req); | |
1702 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1703 | } | |
1da177e4 | 1704 | } |
1da177e4 LT |
1705 | EXPORT_SYMBOL(blk_put_request); |
1706 | ||
320ae51f JA |
1707 | bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1708 | struct bio *bio) | |
73c10101 | 1709 | { |
1eff9d32 | 1710 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1711 | |
73c10101 JA |
1712 | if (!ll_back_merge_fn(q, req, bio)) |
1713 | return false; | |
1714 | ||
8c1cf6bb | 1715 | trace_block_bio_backmerge(q, req, bio); |
73c10101 JA |
1716 | |
1717 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1718 | blk_rq_set_mixed_merge(req); | |
1719 | ||
1720 | req->biotail->bi_next = bio; | |
1721 | req->biotail = bio; | |
4f024f37 | 1722 | req->__data_len += bio->bi_iter.bi_size; |
73c10101 JA |
1723 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1724 | ||
320ae51f | 1725 | blk_account_io_start(req, false); |
73c10101 JA |
1726 | return true; |
1727 | } | |
1728 | ||
320ae51f JA |
1729 | bool bio_attempt_front_merge(struct request_queue *q, struct request *req, |
1730 | struct bio *bio) | |
73c10101 | 1731 | { |
1eff9d32 | 1732 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
73c10101 | 1733 | |
73c10101 JA |
1734 | if (!ll_front_merge_fn(q, req, bio)) |
1735 | return false; | |
1736 | ||
8c1cf6bb | 1737 | trace_block_bio_frontmerge(q, req, bio); |
73c10101 JA |
1738 | |
1739 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1740 | blk_rq_set_mixed_merge(req); | |
1741 | ||
73c10101 JA |
1742 | bio->bi_next = req->bio; |
1743 | req->bio = bio; | |
1744 | ||
4f024f37 KO |
1745 | req->__sector = bio->bi_iter.bi_sector; |
1746 | req->__data_len += bio->bi_iter.bi_size; | |
73c10101 JA |
1747 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
1748 | ||
320ae51f | 1749 | blk_account_io_start(req, false); |
73c10101 JA |
1750 | return true; |
1751 | } | |
1752 | ||
1e739730 CH |
1753 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
1754 | struct bio *bio) | |
1755 | { | |
1756 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1757 | ||
1758 | if (segments >= queue_max_discard_segments(q)) | |
1759 | goto no_merge; | |
1760 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1761 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1762 | goto no_merge; | |
1763 | ||
1764 | req->biotail->bi_next = bio; | |
1765 | req->biotail = bio; | |
1766 | req->__data_len += bio->bi_iter.bi_size; | |
1767 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1768 | req->nr_phys_segments = segments + 1; | |
1769 | ||
1770 | blk_account_io_start(req, false); | |
1771 | return true; | |
1772 | no_merge: | |
1773 | req_set_nomerge(q, req); | |
1774 | return false; | |
1775 | } | |
1776 | ||
bd87b589 | 1777 | /** |
320ae51f | 1778 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
bd87b589 TH |
1779 | * @q: request_queue new bio is being queued at |
1780 | * @bio: new bio being queued | |
1781 | * @request_count: out parameter for number of traversed plugged requests | |
ccc2600b RD |
1782 | * @same_queue_rq: pointer to &struct request that gets filled in when |
1783 | * another request associated with @q is found on the plug list | |
1784 | * (optional, may be %NULL) | |
bd87b589 TH |
1785 | * |
1786 | * Determine whether @bio being queued on @q can be merged with a request | |
1787 | * on %current's plugged list. Returns %true if merge was successful, | |
1788 | * otherwise %false. | |
1789 | * | |
07c2bd37 TH |
1790 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1791 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1792 | * than scheduling, and the request, while may have elvpriv data, is not | |
1793 | * added on the elevator at this point. In addition, we don't have | |
1794 | * reliable access to the elevator outside queue lock. Only check basic | |
1795 | * merging parameters without querying the elevator. | |
da41a589 RE |
1796 | * |
1797 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
73c10101 | 1798 | */ |
320ae51f | 1799 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
5b3f341f SL |
1800 | unsigned int *request_count, |
1801 | struct request **same_queue_rq) | |
73c10101 JA |
1802 | { |
1803 | struct blk_plug *plug; | |
1804 | struct request *rq; | |
92f399c7 | 1805 | struct list_head *plug_list; |
73c10101 | 1806 | |
bd87b589 | 1807 | plug = current->plug; |
73c10101 | 1808 | if (!plug) |
34fe7c05 | 1809 | return false; |
56ebdaf2 | 1810 | *request_count = 0; |
73c10101 | 1811 | |
92f399c7 SL |
1812 | if (q->mq_ops) |
1813 | plug_list = &plug->mq_list; | |
1814 | else | |
1815 | plug_list = &plug->list; | |
1816 | ||
1817 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
34fe7c05 | 1818 | bool merged = false; |
73c10101 | 1819 | |
5b3f341f | 1820 | if (rq->q == q) { |
1b2e19f1 | 1821 | (*request_count)++; |
5b3f341f SL |
1822 | /* |
1823 | * Only blk-mq multiple hardware queues case checks the | |
1824 | * rq in the same queue, there should be only one such | |
1825 | * rq in a queue | |
1826 | **/ | |
1827 | if (same_queue_rq) | |
1828 | *same_queue_rq = rq; | |
1829 | } | |
56ebdaf2 | 1830 | |
07c2bd37 | 1831 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1832 | continue; |
1833 | ||
34fe7c05 CH |
1834 | switch (blk_try_merge(rq, bio)) { |
1835 | case ELEVATOR_BACK_MERGE: | |
1836 | merged = bio_attempt_back_merge(q, rq, bio); | |
1837 | break; | |
1838 | case ELEVATOR_FRONT_MERGE: | |
1839 | merged = bio_attempt_front_merge(q, rq, bio); | |
1840 | break; | |
1e739730 CH |
1841 | case ELEVATOR_DISCARD_MERGE: |
1842 | merged = bio_attempt_discard_merge(q, rq, bio); | |
1843 | break; | |
34fe7c05 CH |
1844 | default: |
1845 | break; | |
73c10101 | 1846 | } |
34fe7c05 CH |
1847 | |
1848 | if (merged) | |
1849 | return true; | |
73c10101 | 1850 | } |
34fe7c05 CH |
1851 | |
1852 | return false; | |
73c10101 JA |
1853 | } |
1854 | ||
0809e3ac JM |
1855 | unsigned int blk_plug_queued_count(struct request_queue *q) |
1856 | { | |
1857 | struct blk_plug *plug; | |
1858 | struct request *rq; | |
1859 | struct list_head *plug_list; | |
1860 | unsigned int ret = 0; | |
1861 | ||
1862 | plug = current->plug; | |
1863 | if (!plug) | |
1864 | goto out; | |
1865 | ||
1866 | if (q->mq_ops) | |
1867 | plug_list = &plug->mq_list; | |
1868 | else | |
1869 | plug_list = &plug->list; | |
1870 | ||
1871 | list_for_each_entry(rq, plug_list, queuelist) { | |
1872 | if (rq->q == q) | |
1873 | ret++; | |
1874 | } | |
1875 | out: | |
1876 | return ret; | |
1877 | } | |
1878 | ||
da8d7f07 | 1879 | void blk_init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1880 | { |
0be0dee6 BVA |
1881 | struct io_context *ioc = rq_ioc(bio); |
1882 | ||
1eff9d32 | 1883 | if (bio->bi_opf & REQ_RAHEAD) |
a82afdfc | 1884 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1885 | |
4f024f37 | 1886 | req->__sector = bio->bi_iter.bi_sector; |
5dc8b362 AM |
1887 | if (ioprio_valid(bio_prio(bio))) |
1888 | req->ioprio = bio_prio(bio); | |
0be0dee6 BVA |
1889 | else if (ioc) |
1890 | req->ioprio = ioc->ioprio; | |
1891 | else | |
1892 | req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); | |
cb6934f8 | 1893 | req->write_hint = bio->bi_write_hint; |
bc1c56fd | 1894 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 | 1895 | } |
da8d7f07 | 1896 | EXPORT_SYMBOL_GPL(blk_init_request_from_bio); |
52d9e675 | 1897 | |
dece1635 | 1898 | static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1899 | { |
73c10101 | 1900 | struct blk_plug *plug; |
34fe7c05 | 1901 | int where = ELEVATOR_INSERT_SORT; |
e4d750c9 | 1902 | struct request *req, *free; |
56ebdaf2 | 1903 | unsigned int request_count = 0; |
87760e5e | 1904 | unsigned int wb_acct; |
1da177e4 | 1905 | |
1da177e4 LT |
1906 | /* |
1907 | * low level driver can indicate that it wants pages above a | |
1908 | * certain limit bounced to low memory (ie for highmem, or even | |
1909 | * ISA dma in theory) | |
1910 | */ | |
1911 | blk_queue_bounce(q, &bio); | |
1912 | ||
af67c31f | 1913 | blk_queue_split(q, &bio); |
23688bf4 | 1914 | |
e23947bd | 1915 | if (!bio_integrity_prep(bio)) |
dece1635 | 1916 | return BLK_QC_T_NONE; |
ffecfd1a | 1917 | |
f73f44eb | 1918 | if (op_is_flush(bio->bi_opf)) { |
73c10101 | 1919 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1920 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1921 | goto get_rq; |
1922 | } | |
1923 | ||
73c10101 JA |
1924 | /* |
1925 | * Check if we can merge with the plugged list before grabbing | |
1926 | * any locks. | |
1927 | */ | |
0809e3ac JM |
1928 | if (!blk_queue_nomerges(q)) { |
1929 | if (blk_attempt_plug_merge(q, bio, &request_count, NULL)) | |
dece1635 | 1930 | return BLK_QC_T_NONE; |
0809e3ac JM |
1931 | } else |
1932 | request_count = blk_plug_queued_count(q); | |
1da177e4 | 1933 | |
73c10101 | 1934 | spin_lock_irq(q->queue_lock); |
2056a782 | 1935 | |
34fe7c05 CH |
1936 | switch (elv_merge(q, &req, bio)) { |
1937 | case ELEVATOR_BACK_MERGE: | |
1938 | if (!bio_attempt_back_merge(q, req, bio)) | |
1939 | break; | |
1940 | elv_bio_merged(q, req, bio); | |
1941 | free = attempt_back_merge(q, req); | |
1942 | if (free) | |
1943 | __blk_put_request(q, free); | |
1944 | else | |
1945 | elv_merged_request(q, req, ELEVATOR_BACK_MERGE); | |
1946 | goto out_unlock; | |
1947 | case ELEVATOR_FRONT_MERGE: | |
1948 | if (!bio_attempt_front_merge(q, req, bio)) | |
1949 | break; | |
1950 | elv_bio_merged(q, req, bio); | |
1951 | free = attempt_front_merge(q, req); | |
1952 | if (free) | |
1953 | __blk_put_request(q, free); | |
1954 | else | |
1955 | elv_merged_request(q, req, ELEVATOR_FRONT_MERGE); | |
1956 | goto out_unlock; | |
1957 | default: | |
1958 | break; | |
1da177e4 LT |
1959 | } |
1960 | ||
450991bc | 1961 | get_rq: |
87760e5e JA |
1962 | wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock); |
1963 | ||
1da177e4 | 1964 | /* |
450991bc | 1965 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1966 | * Returns with the queue unlocked. |
450991bc | 1967 | */ |
055f6e18 | 1968 | blk_queue_enter_live(q); |
6a15674d | 1969 | req = get_request(q, bio->bi_opf, bio, 0); |
a492f075 | 1970 | if (IS_ERR(req)) { |
055f6e18 | 1971 | blk_queue_exit(q); |
87760e5e | 1972 | __wbt_done(q->rq_wb, wb_acct); |
4e4cbee9 CH |
1973 | if (PTR_ERR(req) == -ENOMEM) |
1974 | bio->bi_status = BLK_STS_RESOURCE; | |
1975 | else | |
1976 | bio->bi_status = BLK_STS_IOERR; | |
4246a0b6 | 1977 | bio_endio(bio); |
da8303c6 TH |
1978 | goto out_unlock; |
1979 | } | |
d6344532 | 1980 | |
87760e5e JA |
1981 | wbt_track(&req->issue_stat, wb_acct); |
1982 | ||
450991bc NP |
1983 | /* |
1984 | * After dropping the lock and possibly sleeping here, our request | |
1985 | * may now be mergeable after it had proven unmergeable (above). | |
1986 | * We don't worry about that case for efficiency. It won't happen | |
1987 | * often, and the elevators are able to handle it. | |
1da177e4 | 1988 | */ |
da8d7f07 | 1989 | blk_init_request_from_bio(req, bio); |
1da177e4 | 1990 | |
9562ad9a | 1991 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1992 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1993 | |
1994 | plug = current->plug; | |
721a9602 | 1995 | if (plug) { |
dc6d36c9 JA |
1996 | /* |
1997 | * If this is the first request added after a plug, fire | |
7aef2e78 | 1998 | * of a plug trace. |
0a6219a9 ML |
1999 | * |
2000 | * @request_count may become stale because of schedule | |
2001 | * out, so check plug list again. | |
dc6d36c9 | 2002 | */ |
0a6219a9 | 2003 | if (!request_count || list_empty(&plug->list)) |
dc6d36c9 | 2004 | trace_block_plug(q); |
3540d5e8 | 2005 | else { |
50d24c34 SL |
2006 | struct request *last = list_entry_rq(plug->list.prev); |
2007 | if (request_count >= BLK_MAX_REQUEST_COUNT || | |
2008 | blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) { | |
3540d5e8 | 2009 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
2010 | trace_block_plug(q); |
2011 | } | |
73c10101 | 2012 | } |
73c10101 | 2013 | list_add_tail(&req->queuelist, &plug->list); |
320ae51f | 2014 | blk_account_io_start(req, true); |
73c10101 JA |
2015 | } else { |
2016 | spin_lock_irq(q->queue_lock); | |
2017 | add_acct_request(q, req, where); | |
24ecfbe2 | 2018 | __blk_run_queue(q); |
73c10101 JA |
2019 | out_unlock: |
2020 | spin_unlock_irq(q->queue_lock); | |
2021 | } | |
dece1635 JA |
2022 | |
2023 | return BLK_QC_T_NONE; | |
1da177e4 LT |
2024 | } |
2025 | ||
1da177e4 LT |
2026 | static void handle_bad_sector(struct bio *bio) |
2027 | { | |
2028 | char b[BDEVNAME_SIZE]; | |
2029 | ||
2030 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
6296b960 | 2031 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
74d46992 | 2032 | bio_devname(bio, b), bio->bi_opf, |
f73a1c7d | 2033 | (unsigned long long)bio_end_sector(bio), |
74d46992 | 2034 | (long long)get_capacity(bio->bi_disk)); |
1da177e4 LT |
2035 | } |
2036 | ||
c17bb495 AM |
2037 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
2038 | ||
2039 | static DECLARE_FAULT_ATTR(fail_make_request); | |
2040 | ||
2041 | static int __init setup_fail_make_request(char *str) | |
2042 | { | |
2043 | return setup_fault_attr(&fail_make_request, str); | |
2044 | } | |
2045 | __setup("fail_make_request=", setup_fail_make_request); | |
2046 | ||
b2c9cd37 | 2047 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 2048 | { |
b2c9cd37 | 2049 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
2050 | } |
2051 | ||
2052 | static int __init fail_make_request_debugfs(void) | |
2053 | { | |
dd48c085 AM |
2054 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
2055 | NULL, &fail_make_request); | |
2056 | ||
21f9fcd8 | 2057 | return PTR_ERR_OR_ZERO(dir); |
c17bb495 AM |
2058 | } |
2059 | ||
2060 | late_initcall(fail_make_request_debugfs); | |
2061 | ||
2062 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
2063 | ||
b2c9cd37 AM |
2064 | static inline bool should_fail_request(struct hd_struct *part, |
2065 | unsigned int bytes) | |
c17bb495 | 2066 | { |
b2c9cd37 | 2067 | return false; |
c17bb495 AM |
2068 | } |
2069 | ||
2070 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
2071 | ||
721c7fc7 ID |
2072 | static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) |
2073 | { | |
2074 | if (part->policy && op_is_write(bio_op(bio))) { | |
2075 | char b[BDEVNAME_SIZE]; | |
2076 | ||
2077 | printk(KERN_ERR | |
2078 | "generic_make_request: Trying to write " | |
2079 | "to read-only block-device %s (partno %d)\n", | |
2080 | bio_devname(bio, b), part->partno); | |
2081 | return true; | |
2082 | } | |
2083 | ||
2084 | return false; | |
2085 | } | |
2086 | ||
30abb3a6 HM |
2087 | static noinline int should_fail_bio(struct bio *bio) |
2088 | { | |
2089 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) | |
2090 | return -EIO; | |
2091 | return 0; | |
2092 | } | |
2093 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); | |
2094 | ||
74d46992 CH |
2095 | /* |
2096 | * Remap block n of partition p to block n+start(p) of the disk. | |
2097 | */ | |
2098 | static inline int blk_partition_remap(struct bio *bio) | |
2099 | { | |
2100 | struct hd_struct *p; | |
2101 | int ret = 0; | |
2102 | ||
721c7fc7 ID |
2103 | rcu_read_lock(); |
2104 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
2105 | if (unlikely(!p || should_fail_request(p, bio->bi_iter.bi_size) || | |
2106 | bio_check_ro(bio, p))) { | |
2107 | ret = -EIO; | |
2108 | goto out; | |
2109 | } | |
2110 | ||
74d46992 CH |
2111 | /* |
2112 | * Zone reset does not include bi_size so bio_sectors() is always 0. | |
2113 | * Include a test for the reset op code and perform the remap if needed. | |
2114 | */ | |
721c7fc7 ID |
2115 | if (!bio_sectors(bio) && bio_op(bio) != REQ_OP_ZONE_RESET) |
2116 | goto out; | |
74d46992 | 2117 | |
721c7fc7 ID |
2118 | bio->bi_iter.bi_sector += p->start_sect; |
2119 | bio->bi_partno = 0; | |
2120 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), | |
2121 | bio->bi_iter.bi_sector - p->start_sect); | |
74d46992 | 2122 | |
721c7fc7 ID |
2123 | out: |
2124 | rcu_read_unlock(); | |
74d46992 CH |
2125 | return ret; |
2126 | } | |
2127 | ||
c07e2b41 JA |
2128 | /* |
2129 | * Check whether this bio extends beyond the end of the device. | |
2130 | */ | |
2131 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
2132 | { | |
2133 | sector_t maxsector; | |
2134 | ||
2135 | if (!nr_sectors) | |
2136 | return 0; | |
2137 | ||
2138 | /* Test device or partition size, when known. */ | |
74d46992 | 2139 | maxsector = get_capacity(bio->bi_disk); |
c07e2b41 | 2140 | if (maxsector) { |
4f024f37 | 2141 | sector_t sector = bio->bi_iter.bi_sector; |
c07e2b41 JA |
2142 | |
2143 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
2144 | /* | |
2145 | * This may well happen - the kernel calls bread() | |
2146 | * without checking the size of the device, e.g., when | |
2147 | * mounting a device. | |
2148 | */ | |
2149 | handle_bad_sector(bio); | |
2150 | return 1; | |
2151 | } | |
2152 | } | |
2153 | ||
2154 | return 0; | |
2155 | } | |
2156 | ||
27a84d54 CH |
2157 | static noinline_for_stack bool |
2158 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 2159 | { |
165125e1 | 2160 | struct request_queue *q; |
5a7bbad2 | 2161 | int nr_sectors = bio_sectors(bio); |
4e4cbee9 | 2162 | blk_status_t status = BLK_STS_IOERR; |
5a7bbad2 | 2163 | char b[BDEVNAME_SIZE]; |
1da177e4 LT |
2164 | |
2165 | might_sleep(); | |
1da177e4 | 2166 | |
c07e2b41 JA |
2167 | if (bio_check_eod(bio, nr_sectors)) |
2168 | goto end_io; | |
1da177e4 | 2169 | |
74d46992 | 2170 | q = bio->bi_disk->queue; |
5a7bbad2 CH |
2171 | if (unlikely(!q)) { |
2172 | printk(KERN_ERR | |
2173 | "generic_make_request: Trying to access " | |
2174 | "nonexistent block-device %s (%Lu)\n", | |
74d46992 | 2175 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
5a7bbad2 CH |
2176 | goto end_io; |
2177 | } | |
c17bb495 | 2178 | |
03a07c92 GR |
2179 | /* |
2180 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP | |
2181 | * if queue is not a request based queue. | |
2182 | */ | |
03a07c92 GR |
2183 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q)) |
2184 | goto not_supported; | |
2185 | ||
30abb3a6 | 2186 | if (should_fail_bio(bio)) |
5a7bbad2 | 2187 | goto end_io; |
2056a782 | 2188 | |
721c7fc7 ID |
2189 | if (!bio->bi_partno) { |
2190 | if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) | |
2191 | goto end_io; | |
2192 | } else { | |
2193 | if (blk_partition_remap(bio)) | |
2194 | goto end_io; | |
2195 | } | |
2056a782 | 2196 | |
5a7bbad2 CH |
2197 | if (bio_check_eod(bio, nr_sectors)) |
2198 | goto end_io; | |
1e87901e | 2199 | |
5a7bbad2 CH |
2200 | /* |
2201 | * Filter flush bio's early so that make_request based | |
2202 | * drivers without flush support don't have to worry | |
2203 | * about them. | |
2204 | */ | |
f3a8ab7d | 2205 | if (op_is_flush(bio->bi_opf) && |
c888a8f9 | 2206 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
1eff9d32 | 2207 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
5a7bbad2 | 2208 | if (!nr_sectors) { |
4e4cbee9 | 2209 | status = BLK_STS_OK; |
51fd77bd JA |
2210 | goto end_io; |
2211 | } | |
5a7bbad2 | 2212 | } |
5ddfe969 | 2213 | |
288dab8a CH |
2214 | switch (bio_op(bio)) { |
2215 | case REQ_OP_DISCARD: | |
2216 | if (!blk_queue_discard(q)) | |
2217 | goto not_supported; | |
2218 | break; | |
2219 | case REQ_OP_SECURE_ERASE: | |
2220 | if (!blk_queue_secure_erase(q)) | |
2221 | goto not_supported; | |
2222 | break; | |
2223 | case REQ_OP_WRITE_SAME: | |
74d46992 | 2224 | if (!q->limits.max_write_same_sectors) |
288dab8a | 2225 | goto not_supported; |
58886785 | 2226 | break; |
2d253440 ST |
2227 | case REQ_OP_ZONE_REPORT: |
2228 | case REQ_OP_ZONE_RESET: | |
74d46992 | 2229 | if (!blk_queue_is_zoned(q)) |
2d253440 | 2230 | goto not_supported; |
288dab8a | 2231 | break; |
a6f0788e | 2232 | case REQ_OP_WRITE_ZEROES: |
74d46992 | 2233 | if (!q->limits.max_write_zeroes_sectors) |
a6f0788e CK |
2234 | goto not_supported; |
2235 | break; | |
288dab8a CH |
2236 | default: |
2237 | break; | |
5a7bbad2 | 2238 | } |
01edede4 | 2239 | |
7f4b35d1 TH |
2240 | /* |
2241 | * Various block parts want %current->io_context and lazy ioc | |
2242 | * allocation ends up trading a lot of pain for a small amount of | |
2243 | * memory. Just allocate it upfront. This may fail and block | |
2244 | * layer knows how to live with it. | |
2245 | */ | |
2246 | create_io_context(GFP_ATOMIC, q->node); | |
2247 | ||
ae118896 TH |
2248 | if (!blkcg_bio_issue_check(q, bio)) |
2249 | return false; | |
27a84d54 | 2250 | |
fbbaf700 N |
2251 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
2252 | trace_block_bio_queue(q, bio); | |
2253 | /* Now that enqueuing has been traced, we need to trace | |
2254 | * completion as well. | |
2255 | */ | |
2256 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
2257 | } | |
27a84d54 | 2258 | return true; |
a7384677 | 2259 | |
288dab8a | 2260 | not_supported: |
4e4cbee9 | 2261 | status = BLK_STS_NOTSUPP; |
a7384677 | 2262 | end_io: |
4e4cbee9 | 2263 | bio->bi_status = status; |
4246a0b6 | 2264 | bio_endio(bio); |
27a84d54 | 2265 | return false; |
1da177e4 LT |
2266 | } |
2267 | ||
27a84d54 CH |
2268 | /** |
2269 | * generic_make_request - hand a buffer to its device driver for I/O | |
2270 | * @bio: The bio describing the location in memory and on the device. | |
2271 | * | |
2272 | * generic_make_request() is used to make I/O requests of block | |
2273 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
2274 | * to be done. | |
2275 | * | |
2276 | * generic_make_request() does not return any status. The | |
2277 | * success/failure status of the request, along with notification of | |
2278 | * completion, is delivered asynchronously through the bio->bi_end_io | |
2279 | * function described (one day) else where. | |
2280 | * | |
2281 | * The caller of generic_make_request must make sure that bi_io_vec | |
2282 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
2283 | * set to describe the device address, and the | |
2284 | * bi_end_io and optionally bi_private are set to describe how | |
2285 | * completion notification should be signaled. | |
2286 | * | |
2287 | * generic_make_request and the drivers it calls may use bi_next if this | |
2288 | * bio happens to be merged with someone else, and may resubmit the bio to | |
2289 | * a lower device by calling into generic_make_request recursively, which | |
2290 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 | 2291 | */ |
dece1635 | 2292 | blk_qc_t generic_make_request(struct bio *bio) |
d89d8796 | 2293 | { |
f5fe1b51 N |
2294 | /* |
2295 | * bio_list_on_stack[0] contains bios submitted by the current | |
2296 | * make_request_fn. | |
2297 | * bio_list_on_stack[1] contains bios that were submitted before | |
2298 | * the current make_request_fn, but that haven't been processed | |
2299 | * yet. | |
2300 | */ | |
2301 | struct bio_list bio_list_on_stack[2]; | |
dece1635 | 2302 | blk_qc_t ret = BLK_QC_T_NONE; |
bddd87c7 | 2303 | |
27a84d54 | 2304 | if (!generic_make_request_checks(bio)) |
dece1635 | 2305 | goto out; |
27a84d54 CH |
2306 | |
2307 | /* | |
2308 | * We only want one ->make_request_fn to be active at a time, else | |
2309 | * stack usage with stacked devices could be a problem. So use | |
2310 | * current->bio_list to keep a list of requests submited by a | |
2311 | * make_request_fn function. current->bio_list is also used as a | |
2312 | * flag to say if generic_make_request is currently active in this | |
2313 | * task or not. If it is NULL, then no make_request is active. If | |
2314 | * it is non-NULL, then a make_request is active, and new requests | |
2315 | * should be added at the tail | |
2316 | */ | |
bddd87c7 | 2317 | if (current->bio_list) { |
f5fe1b51 | 2318 | bio_list_add(¤t->bio_list[0], bio); |
dece1635 | 2319 | goto out; |
d89d8796 | 2320 | } |
27a84d54 | 2321 | |
d89d8796 NB |
2322 | /* following loop may be a bit non-obvious, and so deserves some |
2323 | * explanation. | |
2324 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
2325 | * ensure that) so we have a list with a single bio. | |
2326 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
2327 | * we assign bio_list to a pointer to the bio_list_on_stack, |
2328 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 2329 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
2330 | * through a recursive call to generic_make_request. If it |
2331 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
2332 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 2333 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 2334 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
2335 | */ |
2336 | BUG_ON(bio->bi_next); | |
f5fe1b51 N |
2337 | bio_list_init(&bio_list_on_stack[0]); |
2338 | current->bio_list = bio_list_on_stack; | |
d89d8796 | 2339 | do { |
74d46992 | 2340 | struct request_queue *q = bio->bi_disk->queue; |
9a95e4ef | 2341 | blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ? |
3a0a5299 | 2342 | BLK_MQ_REQ_NOWAIT : 0; |
27a84d54 | 2343 | |
3a0a5299 | 2344 | if (likely(blk_queue_enter(q, flags) == 0)) { |
79bd9959 N |
2345 | struct bio_list lower, same; |
2346 | ||
2347 | /* Create a fresh bio_list for all subordinate requests */ | |
f5fe1b51 N |
2348 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
2349 | bio_list_init(&bio_list_on_stack[0]); | |
dece1635 | 2350 | ret = q->make_request_fn(q, bio); |
3ef28e83 DW |
2351 | |
2352 | blk_queue_exit(q); | |
27a84d54 | 2353 | |
79bd9959 N |
2354 | /* sort new bios into those for a lower level |
2355 | * and those for the same level | |
2356 | */ | |
2357 | bio_list_init(&lower); | |
2358 | bio_list_init(&same); | |
f5fe1b51 | 2359 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
74d46992 | 2360 | if (q == bio->bi_disk->queue) |
79bd9959 N |
2361 | bio_list_add(&same, bio); |
2362 | else | |
2363 | bio_list_add(&lower, bio); | |
2364 | /* now assemble so we handle the lowest level first */ | |
f5fe1b51 N |
2365 | bio_list_merge(&bio_list_on_stack[0], &lower); |
2366 | bio_list_merge(&bio_list_on_stack[0], &same); | |
2367 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); | |
3ef28e83 | 2368 | } else { |
03a07c92 GR |
2369 | if (unlikely(!blk_queue_dying(q) && |
2370 | (bio->bi_opf & REQ_NOWAIT))) | |
2371 | bio_wouldblock_error(bio); | |
2372 | else | |
2373 | bio_io_error(bio); | |
3ef28e83 | 2374 | } |
f5fe1b51 | 2375 | bio = bio_list_pop(&bio_list_on_stack[0]); |
d89d8796 | 2376 | } while (bio); |
bddd87c7 | 2377 | current->bio_list = NULL; /* deactivate */ |
dece1635 JA |
2378 | |
2379 | out: | |
2380 | return ret; | |
d89d8796 | 2381 | } |
1da177e4 LT |
2382 | EXPORT_SYMBOL(generic_make_request); |
2383 | ||
f421e1d9 CH |
2384 | /** |
2385 | * direct_make_request - hand a buffer directly to its device driver for I/O | |
2386 | * @bio: The bio describing the location in memory and on the device. | |
2387 | * | |
2388 | * This function behaves like generic_make_request(), but does not protect | |
2389 | * against recursion. Must only be used if the called driver is known | |
2390 | * to not call generic_make_request (or direct_make_request) again from | |
2391 | * its make_request function. (Calling direct_make_request again from | |
2392 | * a workqueue is perfectly fine as that doesn't recurse). | |
2393 | */ | |
2394 | blk_qc_t direct_make_request(struct bio *bio) | |
2395 | { | |
2396 | struct request_queue *q = bio->bi_disk->queue; | |
2397 | bool nowait = bio->bi_opf & REQ_NOWAIT; | |
2398 | blk_qc_t ret; | |
2399 | ||
2400 | if (!generic_make_request_checks(bio)) | |
2401 | return BLK_QC_T_NONE; | |
2402 | ||
3a0a5299 | 2403 | if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) { |
f421e1d9 CH |
2404 | if (nowait && !blk_queue_dying(q)) |
2405 | bio->bi_status = BLK_STS_AGAIN; | |
2406 | else | |
2407 | bio->bi_status = BLK_STS_IOERR; | |
2408 | bio_endio(bio); | |
2409 | return BLK_QC_T_NONE; | |
2410 | } | |
2411 | ||
2412 | ret = q->make_request_fn(q, bio); | |
2413 | blk_queue_exit(q); | |
2414 | return ret; | |
2415 | } | |
2416 | EXPORT_SYMBOL_GPL(direct_make_request); | |
2417 | ||
1da177e4 | 2418 | /** |
710027a4 | 2419 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
2420 | * @bio: The &struct bio which describes the I/O |
2421 | * | |
2422 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
2423 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 2424 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
2425 | * |
2426 | */ | |
4e49ea4a | 2427 | blk_qc_t submit_bio(struct bio *bio) |
1da177e4 | 2428 | { |
bf2de6f5 JA |
2429 | /* |
2430 | * If it's a regular read/write or a barrier with data attached, | |
2431 | * go through the normal accounting stuff before submission. | |
2432 | */ | |
e2a60da7 | 2433 | if (bio_has_data(bio)) { |
4363ac7c MP |
2434 | unsigned int count; |
2435 | ||
95fe6c1a | 2436 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
74d46992 | 2437 | count = queue_logical_block_size(bio->bi_disk->queue); |
4363ac7c MP |
2438 | else |
2439 | count = bio_sectors(bio); | |
2440 | ||
a8ebb056 | 2441 | if (op_is_write(bio_op(bio))) { |
bf2de6f5 JA |
2442 | count_vm_events(PGPGOUT, count); |
2443 | } else { | |
4f024f37 | 2444 | task_io_account_read(bio->bi_iter.bi_size); |
bf2de6f5 JA |
2445 | count_vm_events(PGPGIN, count); |
2446 | } | |
2447 | ||
2448 | if (unlikely(block_dump)) { | |
2449 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 2450 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 2451 | current->comm, task_pid_nr(current), |
a8ebb056 | 2452 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
4f024f37 | 2453 | (unsigned long long)bio->bi_iter.bi_sector, |
74d46992 | 2454 | bio_devname(bio, b), count); |
bf2de6f5 | 2455 | } |
1da177e4 LT |
2456 | } |
2457 | ||
dece1635 | 2458 | return generic_make_request(bio); |
1da177e4 | 2459 | } |
1da177e4 LT |
2460 | EXPORT_SYMBOL(submit_bio); |
2461 | ||
ea435e1b CH |
2462 | bool blk_poll(struct request_queue *q, blk_qc_t cookie) |
2463 | { | |
2464 | if (!q->poll_fn || !blk_qc_t_valid(cookie)) | |
2465 | return false; | |
2466 | ||
2467 | if (current->plug) | |
2468 | blk_flush_plug_list(current->plug, false); | |
2469 | return q->poll_fn(q, cookie); | |
2470 | } | |
2471 | EXPORT_SYMBOL_GPL(blk_poll); | |
2472 | ||
82124d60 | 2473 | /** |
bf4e6b4e HR |
2474 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
2475 | * for new the queue limits | |
82124d60 KU |
2476 | * @q: the queue |
2477 | * @rq: the request being checked | |
2478 | * | |
2479 | * Description: | |
2480 | * @rq may have been made based on weaker limitations of upper-level queues | |
2481 | * in request stacking drivers, and it may violate the limitation of @q. | |
2482 | * Since the block layer and the underlying device driver trust @rq | |
2483 | * after it is inserted to @q, it should be checked against @q before | |
2484 | * the insertion using this generic function. | |
2485 | * | |
82124d60 | 2486 | * Request stacking drivers like request-based dm may change the queue |
bf4e6b4e HR |
2487 | * limits when retrying requests on other queues. Those requests need |
2488 | * to be checked against the new queue limits again during dispatch. | |
82124d60 | 2489 | */ |
bf4e6b4e HR |
2490 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
2491 | struct request *rq) | |
82124d60 | 2492 | { |
8fe0d473 | 2493 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
82124d60 KU |
2494 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
2495 | return -EIO; | |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * queue's settings related to segment counting like q->bounce_pfn | |
2500 | * may differ from that of other stacking queues. | |
2501 | * Recalculate it to check the request correctly on this queue's | |
2502 | * limitation. | |
2503 | */ | |
2504 | blk_recalc_rq_segments(rq); | |
8a78362c | 2505 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
2506 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
2507 | return -EIO; | |
2508 | } | |
2509 | ||
2510 | return 0; | |
2511 | } | |
82124d60 KU |
2512 | |
2513 | /** | |
2514 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
2515 | * @q: the queue to submit the request | |
2516 | * @rq: the request being queued | |
2517 | */ | |
2a842aca | 2518 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
82124d60 KU |
2519 | { |
2520 | unsigned long flags; | |
4853abaa | 2521 | int where = ELEVATOR_INSERT_BACK; |
82124d60 | 2522 | |
bf4e6b4e | 2523 | if (blk_cloned_rq_check_limits(q, rq)) |
2a842aca | 2524 | return BLK_STS_IOERR; |
82124d60 | 2525 | |
b2c9cd37 AM |
2526 | if (rq->rq_disk && |
2527 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
2a842aca | 2528 | return BLK_STS_IOERR; |
82124d60 | 2529 | |
7fb4898e KB |
2530 | if (q->mq_ops) { |
2531 | if (blk_queue_io_stat(q)) | |
2532 | blk_account_io_start(rq, true); | |
157f377b JA |
2533 | /* |
2534 | * Since we have a scheduler attached on the top device, | |
2535 | * bypass a potential scheduler on the bottom device for | |
2536 | * insert. | |
2537 | */ | |
c77ff7fd | 2538 | return blk_mq_request_issue_directly(rq); |
7fb4898e KB |
2539 | } |
2540 | ||
82124d60 | 2541 | spin_lock_irqsave(q->queue_lock, flags); |
3f3299d5 | 2542 | if (unlikely(blk_queue_dying(q))) { |
8ba61435 | 2543 | spin_unlock_irqrestore(q->queue_lock, flags); |
2a842aca | 2544 | return BLK_STS_IOERR; |
8ba61435 | 2545 | } |
82124d60 KU |
2546 | |
2547 | /* | |
2548 | * Submitting request must be dequeued before calling this function | |
2549 | * because it will be linked to another request_queue | |
2550 | */ | |
2551 | BUG_ON(blk_queued_rq(rq)); | |
2552 | ||
f73f44eb | 2553 | if (op_is_flush(rq->cmd_flags)) |
4853abaa JM |
2554 | where = ELEVATOR_INSERT_FLUSH; |
2555 | ||
2556 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
2557 | if (where == ELEVATOR_INSERT_FLUSH) |
2558 | __blk_run_queue(q); | |
82124d60 KU |
2559 | spin_unlock_irqrestore(q->queue_lock, flags); |
2560 | ||
2a842aca | 2561 | return BLK_STS_OK; |
82124d60 KU |
2562 | } |
2563 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
2564 | ||
80a761fd TH |
2565 | /** |
2566 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
2567 | * @rq: request to examine | |
2568 | * | |
2569 | * Description: | |
2570 | * A request could be merge of IOs which require different failure | |
2571 | * handling. This function determines the number of bytes which | |
2572 | * can be failed from the beginning of the request without | |
2573 | * crossing into area which need to be retried further. | |
2574 | * | |
2575 | * Return: | |
2576 | * The number of bytes to fail. | |
80a761fd TH |
2577 | */ |
2578 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
2579 | { | |
2580 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
2581 | unsigned int bytes = 0; | |
2582 | struct bio *bio; | |
2583 | ||
e8064021 | 2584 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
80a761fd TH |
2585 | return blk_rq_bytes(rq); |
2586 | ||
2587 | /* | |
2588 | * Currently the only 'mixing' which can happen is between | |
2589 | * different fastfail types. We can safely fail portions | |
2590 | * which have all the failfast bits that the first one has - | |
2591 | * the ones which are at least as eager to fail as the first | |
2592 | * one. | |
2593 | */ | |
2594 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 | 2595 | if ((bio->bi_opf & ff) != ff) |
80a761fd | 2596 | break; |
4f024f37 | 2597 | bytes += bio->bi_iter.bi_size; |
80a761fd TH |
2598 | } |
2599 | ||
2600 | /* this could lead to infinite loop */ | |
2601 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
2602 | return bytes; | |
2603 | } | |
2604 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
2605 | ||
320ae51f | 2606 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
bc58ba94 | 2607 | { |
c2553b58 | 2608 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
2609 | const int rw = rq_data_dir(req); |
2610 | struct hd_struct *part; | |
2611 | int cpu; | |
2612 | ||
2613 | cpu = part_stat_lock(); | |
09e099d4 | 2614 | part = req->part; |
bc58ba94 JA |
2615 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
2616 | part_stat_unlock(); | |
2617 | } | |
2618 | } | |
2619 | ||
320ae51f | 2620 | void blk_account_io_done(struct request *req) |
bc58ba94 | 2621 | { |
bc58ba94 | 2622 | /* |
dd4c133f TH |
2623 | * Account IO completion. flush_rq isn't accounted as a |
2624 | * normal IO on queueing nor completion. Accounting the | |
2625 | * containing request is enough. | |
bc58ba94 | 2626 | */ |
e8064021 | 2627 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
bc58ba94 JA |
2628 | unsigned long duration = jiffies - req->start_time; |
2629 | const int rw = rq_data_dir(req); | |
2630 | struct hd_struct *part; | |
2631 | int cpu; | |
2632 | ||
2633 | cpu = part_stat_lock(); | |
09e099d4 | 2634 | part = req->part; |
bc58ba94 JA |
2635 | |
2636 | part_stat_inc(cpu, part, ios[rw]); | |
2637 | part_stat_add(cpu, part, ticks[rw], duration); | |
d62e26b3 JA |
2638 | part_round_stats(req->q, cpu, part); |
2639 | part_dec_in_flight(req->q, part, rw); | |
bc58ba94 | 2640 | |
6c23a968 | 2641 | hd_struct_put(part); |
bc58ba94 JA |
2642 | part_stat_unlock(); |
2643 | } | |
2644 | } | |
2645 | ||
47fafbc7 | 2646 | #ifdef CONFIG_PM |
c8158819 LM |
2647 | /* |
2648 | * Don't process normal requests when queue is suspended | |
2649 | * or in the process of suspending/resuming | |
2650 | */ | |
e4f36b24 | 2651 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2652 | { |
e4f36b24 CH |
2653 | switch (rq->q->rpm_status) { |
2654 | case RPM_RESUMING: | |
2655 | case RPM_SUSPENDING: | |
2656 | return rq->rq_flags & RQF_PM; | |
2657 | case RPM_SUSPENDED: | |
2658 | return false; | |
2659 | } | |
2660 | ||
2661 | return true; | |
c8158819 LM |
2662 | } |
2663 | #else | |
e4f36b24 | 2664 | static bool blk_pm_allow_request(struct request *rq) |
c8158819 | 2665 | { |
e4f36b24 | 2666 | return true; |
c8158819 LM |
2667 | } |
2668 | #endif | |
2669 | ||
320ae51f JA |
2670 | void blk_account_io_start(struct request *rq, bool new_io) |
2671 | { | |
2672 | struct hd_struct *part; | |
2673 | int rw = rq_data_dir(rq); | |
2674 | int cpu; | |
2675 | ||
2676 | if (!blk_do_io_stat(rq)) | |
2677 | return; | |
2678 | ||
2679 | cpu = part_stat_lock(); | |
2680 | ||
2681 | if (!new_io) { | |
2682 | part = rq->part; | |
2683 | part_stat_inc(cpu, part, merges[rw]); | |
2684 | } else { | |
2685 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
2686 | if (!hd_struct_try_get(part)) { | |
2687 | /* | |
2688 | * The partition is already being removed, | |
2689 | * the request will be accounted on the disk only | |
2690 | * | |
2691 | * We take a reference on disk->part0 although that | |
2692 | * partition will never be deleted, so we can treat | |
2693 | * it as any other partition. | |
2694 | */ | |
2695 | part = &rq->rq_disk->part0; | |
2696 | hd_struct_get(part); | |
2697 | } | |
d62e26b3 JA |
2698 | part_round_stats(rq->q, cpu, part); |
2699 | part_inc_in_flight(rq->q, part, rw); | |
320ae51f JA |
2700 | rq->part = part; |
2701 | } | |
2702 | ||
2703 | part_stat_unlock(); | |
2704 | } | |
2705 | ||
9c988374 CH |
2706 | static struct request *elv_next_request(struct request_queue *q) |
2707 | { | |
2708 | struct request *rq; | |
2709 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); | |
2710 | ||
2711 | WARN_ON_ONCE(q->mq_ops); | |
2712 | ||
2713 | while (1) { | |
e4f36b24 CH |
2714 | list_for_each_entry(rq, &q->queue_head, queuelist) { |
2715 | if (blk_pm_allow_request(rq)) | |
2716 | return rq; | |
2717 | ||
2718 | if (rq->rq_flags & RQF_SOFTBARRIER) | |
2719 | break; | |
9c988374 CH |
2720 | } |
2721 | ||
2722 | /* | |
2723 | * Flush request is running and flush request isn't queueable | |
2724 | * in the drive, we can hold the queue till flush request is | |
2725 | * finished. Even we don't do this, driver can't dispatch next | |
2726 | * requests and will requeue them. And this can improve | |
2727 | * throughput too. For example, we have request flush1, write1, | |
2728 | * flush 2. flush1 is dispatched, then queue is hold, write1 | |
2729 | * isn't inserted to queue. After flush1 is finished, flush2 | |
2730 | * will be dispatched. Since disk cache is already clean, | |
2731 | * flush2 will be finished very soon, so looks like flush2 is | |
2732 | * folded to flush1. | |
2733 | * Since the queue is hold, a flag is set to indicate the queue | |
2734 | * should be restarted later. Please see flush_end_io() for | |
2735 | * details. | |
2736 | */ | |
2737 | if (fq->flush_pending_idx != fq->flush_running_idx && | |
2738 | !queue_flush_queueable(q)) { | |
2739 | fq->flush_queue_delayed = 1; | |
2740 | return NULL; | |
2741 | } | |
2742 | if (unlikely(blk_queue_bypass(q)) || | |
2743 | !q->elevator->type->ops.sq.elevator_dispatch_fn(q, 0)) | |
2744 | return NULL; | |
2745 | } | |
2746 | } | |
2747 | ||
3bcddeac | 2748 | /** |
9934c8c0 TH |
2749 | * blk_peek_request - peek at the top of a request queue |
2750 | * @q: request queue to peek at | |
2751 | * | |
2752 | * Description: | |
2753 | * Return the request at the top of @q. The returned request | |
2754 | * should be started using blk_start_request() before LLD starts | |
2755 | * processing it. | |
2756 | * | |
2757 | * Return: | |
2758 | * Pointer to the request at the top of @q if available. Null | |
2759 | * otherwise. | |
9934c8c0 TH |
2760 | */ |
2761 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2762 | { |
2763 | struct request *rq; | |
2764 | int ret; | |
2765 | ||
2fff8a92 | 2766 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2767 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2768 | |
9c988374 | 2769 | while ((rq = elv_next_request(q)) != NULL) { |
e8064021 | 2770 | if (!(rq->rq_flags & RQF_STARTED)) { |
158dbda0 TH |
2771 | /* |
2772 | * This is the first time the device driver | |
2773 | * sees this request (possibly after | |
2774 | * requeueing). Notify IO scheduler. | |
2775 | */ | |
e8064021 | 2776 | if (rq->rq_flags & RQF_SORTED) |
158dbda0 TH |
2777 | elv_activate_rq(q, rq); |
2778 | ||
2779 | /* | |
2780 | * just mark as started even if we don't start | |
2781 | * it, a request that has been delayed should | |
2782 | * not be passed by new incoming requests | |
2783 | */ | |
e8064021 | 2784 | rq->rq_flags |= RQF_STARTED; |
158dbda0 TH |
2785 | trace_block_rq_issue(q, rq); |
2786 | } | |
2787 | ||
2788 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2789 | q->end_sector = rq_end_sector(rq); | |
2790 | q->boundary_rq = NULL; | |
2791 | } | |
2792 | ||
e8064021 | 2793 | if (rq->rq_flags & RQF_DONTPREP) |
158dbda0 TH |
2794 | break; |
2795 | ||
2e46e8b2 | 2796 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2797 | /* |
2798 | * make sure space for the drain appears we | |
2799 | * know we can do this because max_hw_segments | |
2800 | * has been adjusted to be one fewer than the | |
2801 | * device can handle | |
2802 | */ | |
2803 | rq->nr_phys_segments++; | |
2804 | } | |
2805 | ||
2806 | if (!q->prep_rq_fn) | |
2807 | break; | |
2808 | ||
2809 | ret = q->prep_rq_fn(q, rq); | |
2810 | if (ret == BLKPREP_OK) { | |
2811 | break; | |
2812 | } else if (ret == BLKPREP_DEFER) { | |
2813 | /* | |
2814 | * the request may have been (partially) prepped. | |
2815 | * we need to keep this request in the front to | |
e8064021 | 2816 | * avoid resource deadlock. RQF_STARTED will |
158dbda0 TH |
2817 | * prevent other fs requests from passing this one. |
2818 | */ | |
2e46e8b2 | 2819 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
e8064021 | 2820 | !(rq->rq_flags & RQF_DONTPREP)) { |
158dbda0 TH |
2821 | /* |
2822 | * remove the space for the drain we added | |
2823 | * so that we don't add it again | |
2824 | */ | |
2825 | --rq->nr_phys_segments; | |
2826 | } | |
2827 | ||
2828 | rq = NULL; | |
2829 | break; | |
0fb5b1fb | 2830 | } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) { |
e8064021 | 2831 | rq->rq_flags |= RQF_QUIET; |
c143dc90 JB |
2832 | /* |
2833 | * Mark this request as started so we don't trigger | |
2834 | * any debug logic in the end I/O path. | |
2835 | */ | |
2836 | blk_start_request(rq); | |
2a842aca CH |
2837 | __blk_end_request_all(rq, ret == BLKPREP_INVALID ? |
2838 | BLK_STS_TARGET : BLK_STS_IOERR); | |
158dbda0 TH |
2839 | } else { |
2840 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2841 | break; | |
2842 | } | |
2843 | } | |
2844 | ||
2845 | return rq; | |
2846 | } | |
9934c8c0 | 2847 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2848 | |
5034435c | 2849 | static void blk_dequeue_request(struct request *rq) |
158dbda0 | 2850 | { |
9934c8c0 TH |
2851 | struct request_queue *q = rq->q; |
2852 | ||
158dbda0 TH |
2853 | BUG_ON(list_empty(&rq->queuelist)); |
2854 | BUG_ON(ELV_ON_HASH(rq)); | |
2855 | ||
2856 | list_del_init(&rq->queuelist); | |
2857 | ||
2858 | /* | |
2859 | * the time frame between a request being removed from the lists | |
2860 | * and to it is freed is accounted as io that is in progress at | |
2861 | * the driver side. | |
2862 | */ | |
9195291e | 2863 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2864 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2865 | set_io_start_time_ns(rq); |
2866 | } | |
158dbda0 TH |
2867 | } |
2868 | ||
9934c8c0 TH |
2869 | /** |
2870 | * blk_start_request - start request processing on the driver | |
2871 | * @req: request to dequeue | |
2872 | * | |
2873 | * Description: | |
2874 | * Dequeue @req and start timeout timer on it. This hands off the | |
2875 | * request to the driver. | |
9934c8c0 TH |
2876 | */ |
2877 | void blk_start_request(struct request *req) | |
2878 | { | |
2fff8a92 | 2879 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 2880 | WARN_ON_ONCE(req->q->mq_ops); |
2fff8a92 | 2881 | |
9934c8c0 TH |
2882 | blk_dequeue_request(req); |
2883 | ||
cf43e6be | 2884 | if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) { |
88eeca49 | 2885 | blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req)); |
cf43e6be | 2886 | req->rq_flags |= RQF_STATS; |
87760e5e | 2887 | wbt_issue(req->q->rq_wb, &req->issue_stat); |
cf43e6be JA |
2888 | } |
2889 | ||
e14575b3 | 2890 | BUG_ON(blk_rq_is_complete(req)); |
9934c8c0 TH |
2891 | blk_add_timer(req); |
2892 | } | |
2893 | EXPORT_SYMBOL(blk_start_request); | |
2894 | ||
2895 | /** | |
2896 | * blk_fetch_request - fetch a request from a request queue | |
2897 | * @q: request queue to fetch a request from | |
2898 | * | |
2899 | * Description: | |
2900 | * Return the request at the top of @q. The request is started on | |
2901 | * return and LLD can start processing it immediately. | |
2902 | * | |
2903 | * Return: | |
2904 | * Pointer to the request at the top of @q if available. Null | |
2905 | * otherwise. | |
9934c8c0 TH |
2906 | */ |
2907 | struct request *blk_fetch_request(struct request_queue *q) | |
2908 | { | |
2909 | struct request *rq; | |
2910 | ||
2fff8a92 | 2911 | lockdep_assert_held(q->queue_lock); |
332ebbf7 | 2912 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 2913 | |
9934c8c0 TH |
2914 | rq = blk_peek_request(q); |
2915 | if (rq) | |
2916 | blk_start_request(rq); | |
2917 | return rq; | |
2918 | } | |
2919 | EXPORT_SYMBOL(blk_fetch_request); | |
2920 | ||
ef71de8b CH |
2921 | /* |
2922 | * Steal bios from a request and add them to a bio list. | |
2923 | * The request must not have been partially completed before. | |
2924 | */ | |
2925 | void blk_steal_bios(struct bio_list *list, struct request *rq) | |
2926 | { | |
2927 | if (rq->bio) { | |
2928 | if (list->tail) | |
2929 | list->tail->bi_next = rq->bio; | |
2930 | else | |
2931 | list->head = rq->bio; | |
2932 | list->tail = rq->biotail; | |
2933 | ||
2934 | rq->bio = NULL; | |
2935 | rq->biotail = NULL; | |
2936 | } | |
2937 | ||
2938 | rq->__data_len = 0; | |
2939 | } | |
2940 | EXPORT_SYMBOL_GPL(blk_steal_bios); | |
2941 | ||
3bcddeac | 2942 | /** |
2e60e022 | 2943 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2944 | * @req: the request being processed |
2a842aca | 2945 | * @error: block status code |
8ebf9756 | 2946 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2947 | * |
2948 | * Description: | |
8ebf9756 RD |
2949 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2950 | * the request structure even if @req doesn't have leftover. | |
2951 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2952 | * |
2953 | * This special helper function is only for request stacking drivers | |
2954 | * (e.g. request-based dm) so that they can handle partial completion. | |
2955 | * Actual device drivers should use blk_end_request instead. | |
2956 | * | |
2957 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2958 | * %false return from this function. | |
3bcddeac KU |
2959 | * |
2960 | * Return: | |
2e60e022 TH |
2961 | * %false - this request doesn't have any more data |
2962 | * %true - this request has more data | |
3bcddeac | 2963 | **/ |
2a842aca CH |
2964 | bool blk_update_request(struct request *req, blk_status_t error, |
2965 | unsigned int nr_bytes) | |
1da177e4 | 2966 | { |
f79ea416 | 2967 | int total_bytes; |
1da177e4 | 2968 | |
2a842aca | 2969 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
4a0efdc9 | 2970 | |
2e60e022 TH |
2971 | if (!req->bio) |
2972 | return false; | |
2973 | ||
2a842aca CH |
2974 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
2975 | !(req->rq_flags & RQF_QUIET))) | |
2976 | print_req_error(req, error); | |
1da177e4 | 2977 | |
bc58ba94 | 2978 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2979 | |
f79ea416 KO |
2980 | total_bytes = 0; |
2981 | while (req->bio) { | |
2982 | struct bio *bio = req->bio; | |
4f024f37 | 2983 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
1da177e4 | 2984 | |
4f024f37 | 2985 | if (bio_bytes == bio->bi_iter.bi_size) |
1da177e4 | 2986 | req->bio = bio->bi_next; |
1da177e4 | 2987 | |
fbbaf700 N |
2988 | /* Completion has already been traced */ |
2989 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
f79ea416 | 2990 | req_bio_endio(req, bio, bio_bytes, error); |
1da177e4 | 2991 | |
f79ea416 KO |
2992 | total_bytes += bio_bytes; |
2993 | nr_bytes -= bio_bytes; | |
1da177e4 | 2994 | |
f79ea416 KO |
2995 | if (!nr_bytes) |
2996 | break; | |
1da177e4 LT |
2997 | } |
2998 | ||
2999 | /* | |
3000 | * completely done | |
3001 | */ | |
2e60e022 TH |
3002 | if (!req->bio) { |
3003 | /* | |
3004 | * Reset counters so that the request stacking driver | |
3005 | * can find how many bytes remain in the request | |
3006 | * later. | |
3007 | */ | |
a2dec7b3 | 3008 | req->__data_len = 0; |
2e60e022 TH |
3009 | return false; |
3010 | } | |
1da177e4 | 3011 | |
a2dec7b3 | 3012 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
3013 | |
3014 | /* update sector only for requests with clear definition of sector */ | |
57292b58 | 3015 | if (!blk_rq_is_passthrough(req)) |
a2dec7b3 | 3016 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 3017 | |
80a761fd | 3018 | /* mixed attributes always follow the first bio */ |
e8064021 | 3019 | if (req->rq_flags & RQF_MIXED_MERGE) { |
80a761fd | 3020 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
1eff9d32 | 3021 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
80a761fd TH |
3022 | } |
3023 | ||
ed6565e7 CH |
3024 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
3025 | /* | |
3026 | * If total number of sectors is less than the first segment | |
3027 | * size, something has gone terribly wrong. | |
3028 | */ | |
3029 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
3030 | blk_dump_rq_flags(req, "request botched"); | |
3031 | req->__data_len = blk_rq_cur_bytes(req); | |
3032 | } | |
2e46e8b2 | 3033 | |
ed6565e7 CH |
3034 | /* recalculate the number of segments */ |
3035 | blk_recalc_rq_segments(req); | |
3036 | } | |
2e46e8b2 | 3037 | |
2e60e022 | 3038 | return true; |
1da177e4 | 3039 | } |
2e60e022 | 3040 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 3041 | |
2a842aca | 3042 | static bool blk_update_bidi_request(struct request *rq, blk_status_t error, |
2e60e022 TH |
3043 | unsigned int nr_bytes, |
3044 | unsigned int bidi_bytes) | |
5efccd17 | 3045 | { |
2e60e022 TH |
3046 | if (blk_update_request(rq, error, nr_bytes)) |
3047 | return true; | |
5efccd17 | 3048 | |
2e60e022 TH |
3049 | /* Bidi request must be completed as a whole */ |
3050 | if (unlikely(blk_bidi_rq(rq)) && | |
3051 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
3052 | return true; | |
5efccd17 | 3053 | |
e2e1a148 JA |
3054 | if (blk_queue_add_random(rq->q)) |
3055 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
3056 | |
3057 | return false; | |
1da177e4 LT |
3058 | } |
3059 | ||
28018c24 JB |
3060 | /** |
3061 | * blk_unprep_request - unprepare a request | |
3062 | * @req: the request | |
3063 | * | |
3064 | * This function makes a request ready for complete resubmission (or | |
3065 | * completion). It happens only after all error handling is complete, | |
3066 | * so represents the appropriate moment to deallocate any resources | |
3067 | * that were allocated to the request in the prep_rq_fn. The queue | |
3068 | * lock is held when calling this. | |
3069 | */ | |
3070 | void blk_unprep_request(struct request *req) | |
3071 | { | |
3072 | struct request_queue *q = req->q; | |
3073 | ||
e8064021 | 3074 | req->rq_flags &= ~RQF_DONTPREP; |
28018c24 JB |
3075 | if (q->unprep_rq_fn) |
3076 | q->unprep_rq_fn(q, req); | |
3077 | } | |
3078 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
3079 | ||
2a842aca | 3080 | void blk_finish_request(struct request *req, blk_status_t error) |
1da177e4 | 3081 | { |
cf43e6be JA |
3082 | struct request_queue *q = req->q; |
3083 | ||
2fff8a92 | 3084 | lockdep_assert_held(req->q->queue_lock); |
332ebbf7 | 3085 | WARN_ON_ONCE(q->mq_ops); |
2fff8a92 | 3086 | |
cf43e6be | 3087 | if (req->rq_flags & RQF_STATS) |
34dbad5d | 3088 | blk_stat_add(req); |
cf43e6be | 3089 | |
e8064021 | 3090 | if (req->rq_flags & RQF_QUEUED) |
cf43e6be | 3091 | blk_queue_end_tag(q, req); |
b8286239 | 3092 | |
ba396a6c | 3093 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 3094 | |
57292b58 | 3095 | if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req)) |
dc3b17cc | 3096 | laptop_io_completion(req->q->backing_dev_info); |
1da177e4 | 3097 | |
e78042e5 MA |
3098 | blk_delete_timer(req); |
3099 | ||
e8064021 | 3100 | if (req->rq_flags & RQF_DONTPREP) |
28018c24 JB |
3101 | blk_unprep_request(req); |
3102 | ||
bc58ba94 | 3103 | blk_account_io_done(req); |
b8286239 | 3104 | |
87760e5e JA |
3105 | if (req->end_io) { |
3106 | wbt_done(req->q->rq_wb, &req->issue_stat); | |
8ffdc655 | 3107 | req->end_io(req, error); |
87760e5e | 3108 | } else { |
b8286239 KU |
3109 | if (blk_bidi_rq(req)) |
3110 | __blk_put_request(req->next_rq->q, req->next_rq); | |
3111 | ||
cf43e6be | 3112 | __blk_put_request(q, req); |
b8286239 | 3113 | } |
1da177e4 | 3114 | } |
12120077 | 3115 | EXPORT_SYMBOL(blk_finish_request); |
1da177e4 | 3116 | |
3b11313a | 3117 | /** |
2e60e022 TH |
3118 | * blk_end_bidi_request - Complete a bidi request |
3119 | * @rq: the request to complete | |
2a842aca | 3120 | * @error: block status code |
2e60e022 TH |
3121 | * @nr_bytes: number of bytes to complete @rq |
3122 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
3123 | * |
3124 | * Description: | |
e3a04fe3 | 3125 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
3126 | * Drivers that supports bidi can safely call this member for any |
3127 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
3128 | * just ignored. | |
336cdb40 KU |
3129 | * |
3130 | * Return: | |
2e60e022 TH |
3131 | * %false - we are done with this request |
3132 | * %true - still buffers pending for this request | |
a0cd1285 | 3133 | **/ |
2a842aca | 3134 | static bool blk_end_bidi_request(struct request *rq, blk_status_t error, |
32fab448 KU |
3135 | unsigned int nr_bytes, unsigned int bidi_bytes) |
3136 | { | |
336cdb40 | 3137 | struct request_queue *q = rq->q; |
2e60e022 | 3138 | unsigned long flags; |
32fab448 | 3139 | |
332ebbf7 BVA |
3140 | WARN_ON_ONCE(q->mq_ops); |
3141 | ||
2e60e022 TH |
3142 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3143 | return true; | |
32fab448 | 3144 | |
336cdb40 | 3145 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 3146 | blk_finish_request(rq, error); |
336cdb40 KU |
3147 | spin_unlock_irqrestore(q->queue_lock, flags); |
3148 | ||
2e60e022 | 3149 | return false; |
32fab448 KU |
3150 | } |
3151 | ||
336cdb40 | 3152 | /** |
2e60e022 TH |
3153 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
3154 | * @rq: the request to complete | |
2a842aca | 3155 | * @error: block status code |
e3a04fe3 KU |
3156 | * @nr_bytes: number of bytes to complete @rq |
3157 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
3158 | * |
3159 | * Description: | |
2e60e022 TH |
3160 | * Identical to blk_end_bidi_request() except that queue lock is |
3161 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
3162 | * |
3163 | * Return: | |
2e60e022 TH |
3164 | * %false - we are done with this request |
3165 | * %true - still buffers pending for this request | |
336cdb40 | 3166 | **/ |
2a842aca | 3167 | static bool __blk_end_bidi_request(struct request *rq, blk_status_t error, |
b1f74493 | 3168 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 3169 | { |
2fff8a92 | 3170 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3171 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3172 | |
2e60e022 TH |
3173 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
3174 | return true; | |
336cdb40 | 3175 | |
2e60e022 | 3176 | blk_finish_request(rq, error); |
336cdb40 | 3177 | |
2e60e022 | 3178 | return false; |
336cdb40 | 3179 | } |
e19a3ab0 KU |
3180 | |
3181 | /** | |
3182 | * blk_end_request - Helper function for drivers to complete the request. | |
3183 | * @rq: the request being processed | |
2a842aca | 3184 | * @error: block status code |
e19a3ab0 KU |
3185 | * @nr_bytes: number of bytes to complete |
3186 | * | |
3187 | * Description: | |
3188 | * Ends I/O on a number of bytes attached to @rq. | |
3189 | * If @rq has leftover, sets it up for the next range of segments. | |
3190 | * | |
3191 | * Return: | |
b1f74493 FT |
3192 | * %false - we are done with this request |
3193 | * %true - still buffers pending for this request | |
e19a3ab0 | 3194 | **/ |
2a842aca CH |
3195 | bool blk_end_request(struct request *rq, blk_status_t error, |
3196 | unsigned int nr_bytes) | |
e19a3ab0 | 3197 | { |
332ebbf7 | 3198 | WARN_ON_ONCE(rq->q->mq_ops); |
b1f74493 | 3199 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 3200 | } |
56ad1740 | 3201 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
3202 | |
3203 | /** | |
b1f74493 FT |
3204 | * blk_end_request_all - Helper function for drives to finish the request. |
3205 | * @rq: the request to finish | |
2a842aca | 3206 | * @error: block status code |
336cdb40 KU |
3207 | * |
3208 | * Description: | |
b1f74493 FT |
3209 | * Completely finish @rq. |
3210 | */ | |
2a842aca | 3211 | void blk_end_request_all(struct request *rq, blk_status_t error) |
336cdb40 | 3212 | { |
b1f74493 FT |
3213 | bool pending; |
3214 | unsigned int bidi_bytes = 0; | |
336cdb40 | 3215 | |
b1f74493 FT |
3216 | if (unlikely(blk_bidi_rq(rq))) |
3217 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 3218 | |
b1f74493 FT |
3219 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
3220 | BUG_ON(pending); | |
3221 | } | |
56ad1740 | 3222 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 3223 | |
e3a04fe3 | 3224 | /** |
b1f74493 FT |
3225 | * __blk_end_request - Helper function for drivers to complete the request. |
3226 | * @rq: the request being processed | |
2a842aca | 3227 | * @error: block status code |
b1f74493 | 3228 | * @nr_bytes: number of bytes to complete |
e3a04fe3 KU |
3229 | * |
3230 | * Description: | |
b1f74493 | 3231 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
3232 | * |
3233 | * Return: | |
b1f74493 FT |
3234 | * %false - we are done with this request |
3235 | * %true - still buffers pending for this request | |
e3a04fe3 | 3236 | **/ |
2a842aca CH |
3237 | bool __blk_end_request(struct request *rq, blk_status_t error, |
3238 | unsigned int nr_bytes) | |
e3a04fe3 | 3239 | { |
2fff8a92 | 3240 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3241 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3242 | |
b1f74493 | 3243 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 3244 | } |
56ad1740 | 3245 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 3246 | |
32fab448 | 3247 | /** |
b1f74493 FT |
3248 | * __blk_end_request_all - Helper function for drives to finish the request. |
3249 | * @rq: the request to finish | |
2a842aca | 3250 | * @error: block status code |
32fab448 KU |
3251 | * |
3252 | * Description: | |
b1f74493 | 3253 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 3254 | */ |
2a842aca | 3255 | void __blk_end_request_all(struct request *rq, blk_status_t error) |
32fab448 | 3256 | { |
b1f74493 FT |
3257 | bool pending; |
3258 | unsigned int bidi_bytes = 0; | |
3259 | ||
2fff8a92 | 3260 | lockdep_assert_held(rq->q->queue_lock); |
332ebbf7 | 3261 | WARN_ON_ONCE(rq->q->mq_ops); |
2fff8a92 | 3262 | |
b1f74493 FT |
3263 | if (unlikely(blk_bidi_rq(rq))) |
3264 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
3265 | ||
3266 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
3267 | BUG_ON(pending); | |
32fab448 | 3268 | } |
56ad1740 | 3269 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 3270 | |
e19a3ab0 | 3271 | /** |
b1f74493 FT |
3272 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
3273 | * @rq: the request to finish the current chunk for | |
2a842aca | 3274 | * @error: block status code |
e19a3ab0 KU |
3275 | * |
3276 | * Description: | |
b1f74493 FT |
3277 | * Complete the current consecutively mapped chunk from @rq. Must |
3278 | * be called with queue lock held. | |
e19a3ab0 KU |
3279 | * |
3280 | * Return: | |
b1f74493 FT |
3281 | * %false - we are done with this request |
3282 | * %true - still buffers pending for this request | |
3283 | */ | |
2a842aca | 3284 | bool __blk_end_request_cur(struct request *rq, blk_status_t error) |
e19a3ab0 | 3285 | { |
b1f74493 | 3286 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 3287 | } |
56ad1740 | 3288 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 3289 | |
86db1e29 JA |
3290 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3291 | struct bio *bio) | |
1da177e4 | 3292 | { |
b4f42e28 | 3293 | if (bio_has_data(bio)) |
fb2dce86 | 3294 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
445251d0 JA |
3295 | else if (bio_op(bio) == REQ_OP_DISCARD) |
3296 | rq->nr_phys_segments = 1; | |
b4f42e28 | 3297 | |
4f024f37 | 3298 | rq->__data_len = bio->bi_iter.bi_size; |
1da177e4 | 3299 | rq->bio = rq->biotail = bio; |
1da177e4 | 3300 | |
74d46992 CH |
3301 | if (bio->bi_disk) |
3302 | rq->rq_disk = bio->bi_disk; | |
66846572 | 3303 | } |
1da177e4 | 3304 | |
2d4dc890 IL |
3305 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
3306 | /** | |
3307 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
3308 | * @rq: the request to be flushed | |
3309 | * | |
3310 | * Description: | |
3311 | * Flush all pages in @rq. | |
3312 | */ | |
3313 | void rq_flush_dcache_pages(struct request *rq) | |
3314 | { | |
3315 | struct req_iterator iter; | |
7988613b | 3316 | struct bio_vec bvec; |
2d4dc890 IL |
3317 | |
3318 | rq_for_each_segment(bvec, rq, iter) | |
7988613b | 3319 | flush_dcache_page(bvec.bv_page); |
2d4dc890 IL |
3320 | } |
3321 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
3322 | #endif | |
3323 | ||
ef9e3fac KU |
3324 | /** |
3325 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
3326 | * @q : the queue of the device being checked | |
3327 | * | |
3328 | * Description: | |
3329 | * Check if underlying low-level drivers of a device are busy. | |
3330 | * If the drivers want to export their busy state, they must set own | |
3331 | * exporting function using blk_queue_lld_busy() first. | |
3332 | * | |
3333 | * Basically, this function is used only by request stacking drivers | |
3334 | * to stop dispatching requests to underlying devices when underlying | |
3335 | * devices are busy. This behavior helps more I/O merging on the queue | |
3336 | * of the request stacking driver and prevents I/O throughput regression | |
3337 | * on burst I/O load. | |
3338 | * | |
3339 | * Return: | |
3340 | * 0 - Not busy (The request stacking driver should dispatch request) | |
3341 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
3342 | */ | |
3343 | int blk_lld_busy(struct request_queue *q) | |
3344 | { | |
3345 | if (q->lld_busy_fn) | |
3346 | return q->lld_busy_fn(q); | |
3347 | ||
3348 | return 0; | |
3349 | } | |
3350 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
3351 | ||
78d8e58a MS |
3352 | /** |
3353 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
3354 | * @rq: the clone request to be cleaned up | |
3355 | * | |
3356 | * Description: | |
3357 | * Free all bios in @rq for a cloned request. | |
3358 | */ | |
3359 | void blk_rq_unprep_clone(struct request *rq) | |
3360 | { | |
3361 | struct bio *bio; | |
3362 | ||
3363 | while ((bio = rq->bio) != NULL) { | |
3364 | rq->bio = bio->bi_next; | |
3365 | ||
3366 | bio_put(bio); | |
3367 | } | |
3368 | } | |
3369 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
3370 | ||
3371 | /* | |
3372 | * Copy attributes of the original request to the clone request. | |
3373 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. | |
3374 | */ | |
3375 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
b0fd271d KU |
3376 | { |
3377 | dst->cpu = src->cpu; | |
b0fd271d KU |
3378 | dst->__sector = blk_rq_pos(src); |
3379 | dst->__data_len = blk_rq_bytes(src); | |
3380 | dst->nr_phys_segments = src->nr_phys_segments; | |
3381 | dst->ioprio = src->ioprio; | |
3382 | dst->extra_len = src->extra_len; | |
78d8e58a MS |
3383 | } |
3384 | ||
3385 | /** | |
3386 | * blk_rq_prep_clone - Helper function to setup clone request | |
3387 | * @rq: the request to be setup | |
3388 | * @rq_src: original request to be cloned | |
3389 | * @bs: bio_set that bios for clone are allocated from | |
3390 | * @gfp_mask: memory allocation mask for bio | |
3391 | * @bio_ctr: setup function to be called for each clone bio. | |
3392 | * Returns %0 for success, non %0 for failure. | |
3393 | * @data: private data to be passed to @bio_ctr | |
3394 | * | |
3395 | * Description: | |
3396 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
3397 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) | |
3398 | * are not copied, and copying such parts is the caller's responsibility. | |
3399 | * Also, pages which the original bios are pointing to are not copied | |
3400 | * and the cloned bios just point same pages. | |
3401 | * So cloned bios must be completed before original bios, which means | |
3402 | * the caller must complete @rq before @rq_src. | |
3403 | */ | |
3404 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
3405 | struct bio_set *bs, gfp_t gfp_mask, | |
3406 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
3407 | void *data) | |
3408 | { | |
3409 | struct bio *bio, *bio_src; | |
3410 | ||
3411 | if (!bs) | |
3412 | bs = fs_bio_set; | |
3413 | ||
3414 | __rq_for_each_bio(bio_src, rq_src) { | |
3415 | bio = bio_clone_fast(bio_src, gfp_mask, bs); | |
3416 | if (!bio) | |
3417 | goto free_and_out; | |
3418 | ||
3419 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
3420 | goto free_and_out; | |
3421 | ||
3422 | if (rq->bio) { | |
3423 | rq->biotail->bi_next = bio; | |
3424 | rq->biotail = bio; | |
3425 | } else | |
3426 | rq->bio = rq->biotail = bio; | |
3427 | } | |
3428 | ||
3429 | __blk_rq_prep_clone(rq, rq_src); | |
3430 | ||
3431 | return 0; | |
3432 | ||
3433 | free_and_out: | |
3434 | if (bio) | |
3435 | bio_put(bio); | |
3436 | blk_rq_unprep_clone(rq); | |
3437 | ||
3438 | return -ENOMEM; | |
b0fd271d KU |
3439 | } |
3440 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
3441 | ||
59c3d45e | 3442 | int kblockd_schedule_work(struct work_struct *work) |
1da177e4 LT |
3443 | { |
3444 | return queue_work(kblockd_workqueue, work); | |
3445 | } | |
1da177e4 LT |
3446 | EXPORT_SYMBOL(kblockd_schedule_work); |
3447 | ||
ee63cfa7 JA |
3448 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
3449 | { | |
3450 | return queue_work_on(cpu, kblockd_workqueue, work); | |
3451 | } | |
3452 | EXPORT_SYMBOL(kblockd_schedule_work_on); | |
3453 | ||
818cd1cb JA |
3454 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
3455 | unsigned long delay) | |
3456 | { | |
3457 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); | |
3458 | } | |
3459 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); | |
3460 | ||
75df7136 SJ |
3461 | /** |
3462 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
3463 | * @plug: The &struct blk_plug that needs to be initialized | |
3464 | * | |
3465 | * Description: | |
3466 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
3467 | * pending I/O should the task end up blocking between blk_start_plug() and | |
3468 | * blk_finish_plug(). This is important from a performance perspective, but | |
3469 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
3470 | * for a memory allocation, memory reclaim could end up wanting to free a | |
3471 | * page belonging to that request that is currently residing in our private | |
3472 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
3473 | * this kind of deadlock. | |
3474 | */ | |
73c10101 JA |
3475 | void blk_start_plug(struct blk_plug *plug) |
3476 | { | |
3477 | struct task_struct *tsk = current; | |
3478 | ||
dd6cf3e1 SL |
3479 | /* |
3480 | * If this is a nested plug, don't actually assign it. | |
3481 | */ | |
3482 | if (tsk->plug) | |
3483 | return; | |
3484 | ||
73c10101 | 3485 | INIT_LIST_HEAD(&plug->list); |
320ae51f | 3486 | INIT_LIST_HEAD(&plug->mq_list); |
048c9374 | 3487 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 | 3488 | /* |
dd6cf3e1 SL |
3489 | * Store ordering should not be needed here, since a potential |
3490 | * preempt will imply a full memory barrier | |
73c10101 | 3491 | */ |
dd6cf3e1 | 3492 | tsk->plug = plug; |
73c10101 JA |
3493 | } |
3494 | EXPORT_SYMBOL(blk_start_plug); | |
3495 | ||
3496 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
3497 | { | |
3498 | struct request *rqa = container_of(a, struct request, queuelist); | |
3499 | struct request *rqb = container_of(b, struct request, queuelist); | |
3500 | ||
975927b9 JM |
3501 | return !(rqa->q < rqb->q || |
3502 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); | |
73c10101 JA |
3503 | } |
3504 | ||
49cac01e JA |
3505 | /* |
3506 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
3507 | * until a safe kblockd context. We due this to avoid accidental big | |
3508 | * additional stack usage in driver dispatch, in places where the originally | |
3509 | * plugger did not intend it. | |
3510 | */ | |
f6603783 | 3511 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 3512 | bool from_schedule) |
99e22598 | 3513 | __releases(q->queue_lock) |
94b5eb28 | 3514 | { |
2fff8a92 BVA |
3515 | lockdep_assert_held(q->queue_lock); |
3516 | ||
49cac01e | 3517 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 3518 | |
70460571 | 3519 | if (from_schedule) |
24ecfbe2 | 3520 | blk_run_queue_async(q); |
70460571 | 3521 | else |
24ecfbe2 | 3522 | __blk_run_queue(q); |
70460571 | 3523 | spin_unlock(q->queue_lock); |
94b5eb28 JA |
3524 | } |
3525 | ||
74018dc3 | 3526 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
3527 | { |
3528 | LIST_HEAD(callbacks); | |
3529 | ||
2a7d5559 SL |
3530 | while (!list_empty(&plug->cb_list)) { |
3531 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 3532 | |
2a7d5559 SL |
3533 | while (!list_empty(&callbacks)) { |
3534 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
3535 | struct blk_plug_cb, |
3536 | list); | |
2a7d5559 | 3537 | list_del(&cb->list); |
74018dc3 | 3538 | cb->callback(cb, from_schedule); |
2a7d5559 | 3539 | } |
048c9374 N |
3540 | } |
3541 | } | |
3542 | ||
9cbb1750 N |
3543 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
3544 | int size) | |
3545 | { | |
3546 | struct blk_plug *plug = current->plug; | |
3547 | struct blk_plug_cb *cb; | |
3548 | ||
3549 | if (!plug) | |
3550 | return NULL; | |
3551 | ||
3552 | list_for_each_entry(cb, &plug->cb_list, list) | |
3553 | if (cb->callback == unplug && cb->data == data) | |
3554 | return cb; | |
3555 | ||
3556 | /* Not currently on the callback list */ | |
3557 | BUG_ON(size < sizeof(*cb)); | |
3558 | cb = kzalloc(size, GFP_ATOMIC); | |
3559 | if (cb) { | |
3560 | cb->data = data; | |
3561 | cb->callback = unplug; | |
3562 | list_add(&cb->list, &plug->cb_list); | |
3563 | } | |
3564 | return cb; | |
3565 | } | |
3566 | EXPORT_SYMBOL(blk_check_plugged); | |
3567 | ||
49cac01e | 3568 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
3569 | { |
3570 | struct request_queue *q; | |
3571 | unsigned long flags; | |
3572 | struct request *rq; | |
109b8129 | 3573 | LIST_HEAD(list); |
94b5eb28 | 3574 | unsigned int depth; |
73c10101 | 3575 | |
74018dc3 | 3576 | flush_plug_callbacks(plug, from_schedule); |
320ae51f JA |
3577 | |
3578 | if (!list_empty(&plug->mq_list)) | |
3579 | blk_mq_flush_plug_list(plug, from_schedule); | |
3580 | ||
73c10101 JA |
3581 | if (list_empty(&plug->list)) |
3582 | return; | |
3583 | ||
109b8129 N |
3584 | list_splice_init(&plug->list, &list); |
3585 | ||
422765c2 | 3586 | list_sort(NULL, &list, plug_rq_cmp); |
73c10101 JA |
3587 | |
3588 | q = NULL; | |
94b5eb28 | 3589 | depth = 0; |
18811272 JA |
3590 | |
3591 | /* | |
3592 | * Save and disable interrupts here, to avoid doing it for every | |
3593 | * queue lock we have to take. | |
3594 | */ | |
73c10101 | 3595 | local_irq_save(flags); |
109b8129 N |
3596 | while (!list_empty(&list)) { |
3597 | rq = list_entry_rq(list.next); | |
73c10101 | 3598 | list_del_init(&rq->queuelist); |
73c10101 JA |
3599 | BUG_ON(!rq->q); |
3600 | if (rq->q != q) { | |
99e22598 JA |
3601 | /* |
3602 | * This drops the queue lock | |
3603 | */ | |
3604 | if (q) | |
49cac01e | 3605 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3606 | q = rq->q; |
94b5eb28 | 3607 | depth = 0; |
73c10101 JA |
3608 | spin_lock(q->queue_lock); |
3609 | } | |
8ba61435 TH |
3610 | |
3611 | /* | |
3612 | * Short-circuit if @q is dead | |
3613 | */ | |
3f3299d5 | 3614 | if (unlikely(blk_queue_dying(q))) { |
2a842aca | 3615 | __blk_end_request_all(rq, BLK_STS_IOERR); |
8ba61435 TH |
3616 | continue; |
3617 | } | |
3618 | ||
73c10101 JA |
3619 | /* |
3620 | * rq is already accounted, so use raw insert | |
3621 | */ | |
f73f44eb | 3622 | if (op_is_flush(rq->cmd_flags)) |
401a18e9 JA |
3623 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
3624 | else | |
3625 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3626 | |
3627 | depth++; | |
73c10101 JA |
3628 | } |
3629 | ||
99e22598 JA |
3630 | /* |
3631 | * This drops the queue lock | |
3632 | */ | |
3633 | if (q) | |
49cac01e | 3634 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3635 | |
73c10101 JA |
3636 | local_irq_restore(flags); |
3637 | } | |
73c10101 JA |
3638 | |
3639 | void blk_finish_plug(struct blk_plug *plug) | |
3640 | { | |
dd6cf3e1 SL |
3641 | if (plug != current->plug) |
3642 | return; | |
f6603783 | 3643 | blk_flush_plug_list(plug, false); |
73c10101 | 3644 | |
dd6cf3e1 | 3645 | current->plug = NULL; |
73c10101 | 3646 | } |
88b996cd | 3647 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3648 | |
47fafbc7 | 3649 | #ifdef CONFIG_PM |
6c954667 LM |
3650 | /** |
3651 | * blk_pm_runtime_init - Block layer runtime PM initialization routine | |
3652 | * @q: the queue of the device | |
3653 | * @dev: the device the queue belongs to | |
3654 | * | |
3655 | * Description: | |
3656 | * Initialize runtime-PM-related fields for @q and start auto suspend for | |
3657 | * @dev. Drivers that want to take advantage of request-based runtime PM | |
3658 | * should call this function after @dev has been initialized, and its | |
3659 | * request queue @q has been allocated, and runtime PM for it can not happen | |
3660 | * yet(either due to disabled/forbidden or its usage_count > 0). In most | |
3661 | * cases, driver should call this function before any I/O has taken place. | |
3662 | * | |
3663 | * This function takes care of setting up using auto suspend for the device, | |
3664 | * the autosuspend delay is set to -1 to make runtime suspend impossible | |
3665 | * until an updated value is either set by user or by driver. Drivers do | |
3666 | * not need to touch other autosuspend settings. | |
3667 | * | |
3668 | * The block layer runtime PM is request based, so only works for drivers | |
3669 | * that use request as their IO unit instead of those directly use bio's. | |
3670 | */ | |
3671 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) | |
3672 | { | |
765e40b6 CH |
3673 | /* not support for RQF_PM and ->rpm_status in blk-mq yet */ |
3674 | if (q->mq_ops) | |
3675 | return; | |
3676 | ||
6c954667 LM |
3677 | q->dev = dev; |
3678 | q->rpm_status = RPM_ACTIVE; | |
3679 | pm_runtime_set_autosuspend_delay(q->dev, -1); | |
3680 | pm_runtime_use_autosuspend(q->dev); | |
3681 | } | |
3682 | EXPORT_SYMBOL(blk_pm_runtime_init); | |
3683 | ||
3684 | /** | |
3685 | * blk_pre_runtime_suspend - Pre runtime suspend check | |
3686 | * @q: the queue of the device | |
3687 | * | |
3688 | * Description: | |
3689 | * This function will check if runtime suspend is allowed for the device | |
3690 | * by examining if there are any requests pending in the queue. If there | |
3691 | * are requests pending, the device can not be runtime suspended; otherwise, | |
3692 | * the queue's status will be updated to SUSPENDING and the driver can | |
3693 | * proceed to suspend the device. | |
3694 | * | |
3695 | * For the not allowed case, we mark last busy for the device so that | |
3696 | * runtime PM core will try to autosuspend it some time later. | |
3697 | * | |
3698 | * This function should be called near the start of the device's | |
3699 | * runtime_suspend callback. | |
3700 | * | |
3701 | * Return: | |
3702 | * 0 - OK to runtime suspend the device | |
3703 | * -EBUSY - Device should not be runtime suspended | |
3704 | */ | |
3705 | int blk_pre_runtime_suspend(struct request_queue *q) | |
3706 | { | |
3707 | int ret = 0; | |
3708 | ||
4fd41a85 KX |
3709 | if (!q->dev) |
3710 | return ret; | |
3711 | ||
6c954667 LM |
3712 | spin_lock_irq(q->queue_lock); |
3713 | if (q->nr_pending) { | |
3714 | ret = -EBUSY; | |
3715 | pm_runtime_mark_last_busy(q->dev); | |
3716 | } else { | |
3717 | q->rpm_status = RPM_SUSPENDING; | |
3718 | } | |
3719 | spin_unlock_irq(q->queue_lock); | |
3720 | return ret; | |
3721 | } | |
3722 | EXPORT_SYMBOL(blk_pre_runtime_suspend); | |
3723 | ||
3724 | /** | |
3725 | * blk_post_runtime_suspend - Post runtime suspend processing | |
3726 | * @q: the queue of the device | |
3727 | * @err: return value of the device's runtime_suspend function | |
3728 | * | |
3729 | * Description: | |
3730 | * Update the queue's runtime status according to the return value of the | |
3731 | * device's runtime suspend function and mark last busy for the device so | |
3732 | * that PM core will try to auto suspend the device at a later time. | |
3733 | * | |
3734 | * This function should be called near the end of the device's | |
3735 | * runtime_suspend callback. | |
3736 | */ | |
3737 | void blk_post_runtime_suspend(struct request_queue *q, int err) | |
3738 | { | |
4fd41a85 KX |
3739 | if (!q->dev) |
3740 | return; | |
3741 | ||
6c954667 LM |
3742 | spin_lock_irq(q->queue_lock); |
3743 | if (!err) { | |
3744 | q->rpm_status = RPM_SUSPENDED; | |
3745 | } else { | |
3746 | q->rpm_status = RPM_ACTIVE; | |
3747 | pm_runtime_mark_last_busy(q->dev); | |
3748 | } | |
3749 | spin_unlock_irq(q->queue_lock); | |
3750 | } | |
3751 | EXPORT_SYMBOL(blk_post_runtime_suspend); | |
3752 | ||
3753 | /** | |
3754 | * blk_pre_runtime_resume - Pre runtime resume processing | |
3755 | * @q: the queue of the device | |
3756 | * | |
3757 | * Description: | |
3758 | * Update the queue's runtime status to RESUMING in preparation for the | |
3759 | * runtime resume of the device. | |
3760 | * | |
3761 | * This function should be called near the start of the device's | |
3762 | * runtime_resume callback. | |
3763 | */ | |
3764 | void blk_pre_runtime_resume(struct request_queue *q) | |
3765 | { | |
4fd41a85 KX |
3766 | if (!q->dev) |
3767 | return; | |
3768 | ||
6c954667 LM |
3769 | spin_lock_irq(q->queue_lock); |
3770 | q->rpm_status = RPM_RESUMING; | |
3771 | spin_unlock_irq(q->queue_lock); | |
3772 | } | |
3773 | EXPORT_SYMBOL(blk_pre_runtime_resume); | |
3774 | ||
3775 | /** | |
3776 | * blk_post_runtime_resume - Post runtime resume processing | |
3777 | * @q: the queue of the device | |
3778 | * @err: return value of the device's runtime_resume function | |
3779 | * | |
3780 | * Description: | |
3781 | * Update the queue's runtime status according to the return value of the | |
3782 | * device's runtime_resume function. If it is successfully resumed, process | |
3783 | * the requests that are queued into the device's queue when it is resuming | |
3784 | * and then mark last busy and initiate autosuspend for it. | |
3785 | * | |
3786 | * This function should be called near the end of the device's | |
3787 | * runtime_resume callback. | |
3788 | */ | |
3789 | void blk_post_runtime_resume(struct request_queue *q, int err) | |
3790 | { | |
4fd41a85 KX |
3791 | if (!q->dev) |
3792 | return; | |
3793 | ||
6c954667 LM |
3794 | spin_lock_irq(q->queue_lock); |
3795 | if (!err) { | |
3796 | q->rpm_status = RPM_ACTIVE; | |
3797 | __blk_run_queue(q); | |
3798 | pm_runtime_mark_last_busy(q->dev); | |
c60855cd | 3799 | pm_request_autosuspend(q->dev); |
6c954667 LM |
3800 | } else { |
3801 | q->rpm_status = RPM_SUSPENDED; | |
3802 | } | |
3803 | spin_unlock_irq(q->queue_lock); | |
3804 | } | |
3805 | EXPORT_SYMBOL(blk_post_runtime_resume); | |
d07ab6d1 MW |
3806 | |
3807 | /** | |
3808 | * blk_set_runtime_active - Force runtime status of the queue to be active | |
3809 | * @q: the queue of the device | |
3810 | * | |
3811 | * If the device is left runtime suspended during system suspend the resume | |
3812 | * hook typically resumes the device and corrects runtime status | |
3813 | * accordingly. However, that does not affect the queue runtime PM status | |
3814 | * which is still "suspended". This prevents processing requests from the | |
3815 | * queue. | |
3816 | * | |
3817 | * This function can be used in driver's resume hook to correct queue | |
3818 | * runtime PM status and re-enable peeking requests from the queue. It | |
3819 | * should be called before first request is added to the queue. | |
3820 | */ | |
3821 | void blk_set_runtime_active(struct request_queue *q) | |
3822 | { | |
3823 | spin_lock_irq(q->queue_lock); | |
3824 | q->rpm_status = RPM_ACTIVE; | |
3825 | pm_runtime_mark_last_busy(q->dev); | |
3826 | pm_request_autosuspend(q->dev); | |
3827 | spin_unlock_irq(q->queue_lock); | |
3828 | } | |
3829 | EXPORT_SYMBOL(blk_set_runtime_active); | |
6c954667 LM |
3830 | #endif |
3831 | ||
1da177e4 LT |
3832 | int __init blk_dev_init(void) |
3833 | { | |
ef295ecf CH |
3834 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
3835 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * | |
0762b23d | 3836 | FIELD_SIZEOF(struct request, cmd_flags)); |
ef295ecf CH |
3837 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
3838 | FIELD_SIZEOF(struct bio, bi_opf)); | |
9eb55b03 | 3839 | |
89b90be2 TH |
3840 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3841 | kblockd_workqueue = alloc_workqueue("kblockd", | |
28747fcd | 3842 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
1da177e4 LT |
3843 | if (!kblockd_workqueue) |
3844 | panic("Failed to create kblockd\n"); | |
3845 | ||
3846 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3847 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3848 | |
c2789bd4 | 3849 | blk_requestq_cachep = kmem_cache_create("request_queue", |
165125e1 | 3850 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3851 | |
18fbda91 OS |
3852 | #ifdef CONFIG_DEBUG_FS |
3853 | blk_debugfs_root = debugfs_create_dir("block", NULL); | |
3854 | #endif | |
3855 | ||
d38ecf93 | 3856 | return 0; |
1da177e4 | 3857 | } |