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> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 29 | #include <linux/fault-inject.h> |
73c10101 | 30 | #include <linux/list_sort.h> |
e3c78ca5 | 31 | #include <linux/delay.h> |
aaf7c680 | 32 | #include <linux/ratelimit.h> |
55782138 LZ |
33 | |
34 | #define CREATE_TRACE_POINTS | |
35 | #include <trace/events/block.h> | |
1da177e4 | 36 | |
8324aa91 | 37 | #include "blk.h" |
5efd6113 | 38 | #include "blk-cgroup.h" |
8324aa91 | 39 | |
d07335e5 | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 41 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 42 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 43 | |
a73f730d TH |
44 | DEFINE_IDA(blk_queue_ida); |
45 | ||
1da177e4 LT |
46 | /* |
47 | * For the allocated request tables | |
48 | */ | |
5ece6c52 | 49 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
50 | |
51 | /* | |
52 | * For queue allocation | |
53 | */ | |
6728cb0e | 54 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 55 | |
1da177e4 LT |
56 | /* |
57 | * Controlling structure to kblockd | |
58 | */ | |
ff856bad | 59 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 60 | |
26b8256e JA |
61 | static void drive_stat_acct(struct request *rq, int new_io) |
62 | { | |
28f13702 | 63 | struct hd_struct *part; |
26b8256e | 64 | int rw = rq_data_dir(rq); |
c9959059 | 65 | int cpu; |
26b8256e | 66 | |
c2553b58 | 67 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
68 | return; |
69 | ||
074a7aca | 70 | cpu = part_stat_lock(); |
c9959059 | 71 | |
09e099d4 JM |
72 | if (!new_io) { |
73 | part = rq->part; | |
074a7aca | 74 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
75 | } else { |
76 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 77 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
78 | /* |
79 | * The partition is already being removed, | |
80 | * the request will be accounted on the disk only | |
81 | * | |
82 | * We take a reference on disk->part0 although that | |
83 | * partition will never be deleted, so we can treat | |
84 | * it as any other partition. | |
85 | */ | |
86 | part = &rq->rq_disk->part0; | |
6c23a968 | 87 | hd_struct_get(part); |
09e099d4 | 88 | } |
074a7aca | 89 | part_round_stats(cpu, part); |
316d315b | 90 | part_inc_in_flight(part, rw); |
09e099d4 | 91 | rq->part = part; |
26b8256e | 92 | } |
e71bf0d0 | 93 | |
074a7aca | 94 | part_stat_unlock(); |
26b8256e JA |
95 | } |
96 | ||
8324aa91 | 97 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
98 | { |
99 | int nr; | |
100 | ||
101 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
102 | if (nr > q->nr_requests) | |
103 | nr = q->nr_requests; | |
104 | q->nr_congestion_on = nr; | |
105 | ||
106 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
107 | if (nr < 1) | |
108 | nr = 1; | |
109 | q->nr_congestion_off = nr; | |
110 | } | |
111 | ||
1da177e4 LT |
112 | /** |
113 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
114 | * @bdev: device | |
115 | * | |
116 | * Locates the passed device's request queue and returns the address of its | |
117 | * backing_dev_info | |
118 | * | |
119 | * Will return NULL if the request queue cannot be located. | |
120 | */ | |
121 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
122 | { | |
123 | struct backing_dev_info *ret = NULL; | |
165125e1 | 124 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
125 | |
126 | if (q) | |
127 | ret = &q->backing_dev_info; | |
128 | return ret; | |
129 | } | |
1da177e4 LT |
130 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
131 | ||
2a4aa30c | 132 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 133 | { |
1afb20f3 FT |
134 | memset(rq, 0, sizeof(*rq)); |
135 | ||
1da177e4 | 136 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 137 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 138 | rq->cpu = -1; |
63a71386 | 139 | rq->q = q; |
a2dec7b3 | 140 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
141 | INIT_HLIST_NODE(&rq->hash); |
142 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 143 | rq->cmd = rq->__cmd; |
e2494e1b | 144 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 145 | rq->tag = -1; |
1da177e4 | 146 | rq->ref_count = 1; |
b243ddcb | 147 | rq->start_time = jiffies; |
9195291e | 148 | set_start_time_ns(rq); |
09e099d4 | 149 | rq->part = NULL; |
1da177e4 | 150 | } |
2a4aa30c | 151 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 152 | |
5bb23a68 N |
153 | static void req_bio_endio(struct request *rq, struct bio *bio, |
154 | unsigned int nbytes, int error) | |
1da177e4 | 155 | { |
143a87f4 TH |
156 | if (error) |
157 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
158 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
159 | error = -EIO; | |
797e7dbb | 160 | |
143a87f4 TH |
161 | if (unlikely(nbytes > bio->bi_size)) { |
162 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
163 | __func__, nbytes, bio->bi_size); | |
164 | nbytes = bio->bi_size; | |
5bb23a68 | 165 | } |
797e7dbb | 166 | |
143a87f4 TH |
167 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
168 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 169 | |
143a87f4 TH |
170 | bio->bi_size -= nbytes; |
171 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 172 | |
143a87f4 TH |
173 | if (bio_integrity(bio)) |
174 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 175 | |
143a87f4 TH |
176 | /* don't actually finish bio if it's part of flush sequence */ |
177 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
178 | bio_endio(bio, error); | |
1da177e4 | 179 | } |
1da177e4 | 180 | |
1da177e4 LT |
181 | void blk_dump_rq_flags(struct request *rq, char *msg) |
182 | { | |
183 | int bit; | |
184 | ||
6728cb0e | 185 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
186 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
187 | rq->cmd_flags); | |
1da177e4 | 188 | |
83096ebf TH |
189 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
190 | (unsigned long long)blk_rq_pos(rq), | |
191 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 192 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 193 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 194 | |
33659ebb | 195 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 196 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 197 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
198 | printk("%02x ", rq->cmd[bit]); |
199 | printk("\n"); | |
200 | } | |
201 | } | |
1da177e4 LT |
202 | EXPORT_SYMBOL(blk_dump_rq_flags); |
203 | ||
3cca6dc1 | 204 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 205 | { |
3cca6dc1 | 206 | struct request_queue *q; |
1da177e4 | 207 | |
3cca6dc1 JA |
208 | q = container_of(work, struct request_queue, delay_work.work); |
209 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 210 | __blk_run_queue(q); |
3cca6dc1 | 211 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 212 | } |
1da177e4 LT |
213 | |
214 | /** | |
3cca6dc1 JA |
215 | * blk_delay_queue - restart queueing after defined interval |
216 | * @q: The &struct request_queue in question | |
217 | * @msecs: Delay in msecs | |
1da177e4 LT |
218 | * |
219 | * Description: | |
3cca6dc1 JA |
220 | * Sometimes queueing needs to be postponed for a little while, to allow |
221 | * resources to come back. This function will make sure that queueing is | |
222 | * restarted around the specified time. | |
223 | */ | |
224 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 225 | { |
4521cc4e JA |
226 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
227 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 228 | } |
3cca6dc1 | 229 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 230 | |
1da177e4 LT |
231 | /** |
232 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 233 | * @q: The &struct request_queue in question |
1da177e4 LT |
234 | * |
235 | * Description: | |
236 | * blk_start_queue() will clear the stop flag on the queue, and call | |
237 | * the request_fn for the queue if it was in a stopped state when | |
238 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
239 | **/ | |
165125e1 | 240 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 241 | { |
a038e253 PBG |
242 | WARN_ON(!irqs_disabled()); |
243 | ||
75ad23bc | 244 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 245 | __blk_run_queue(q); |
1da177e4 | 246 | } |
1da177e4 LT |
247 | EXPORT_SYMBOL(blk_start_queue); |
248 | ||
249 | /** | |
250 | * blk_stop_queue - stop a queue | |
165125e1 | 251 | * @q: The &struct request_queue in question |
1da177e4 LT |
252 | * |
253 | * Description: | |
254 | * The Linux block layer assumes that a block driver will consume all | |
255 | * entries on the request queue when the request_fn strategy is called. | |
256 | * Often this will not happen, because of hardware limitations (queue | |
257 | * depth settings). If a device driver gets a 'queue full' response, | |
258 | * or if it simply chooses not to queue more I/O at one point, it can | |
259 | * call this function to prevent the request_fn from being called until | |
260 | * the driver has signalled it's ready to go again. This happens by calling | |
261 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
262 | **/ | |
165125e1 | 263 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 264 | { |
ad3d9d7e | 265 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 266 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
267 | } |
268 | EXPORT_SYMBOL(blk_stop_queue); | |
269 | ||
270 | /** | |
271 | * blk_sync_queue - cancel any pending callbacks on a queue | |
272 | * @q: the queue | |
273 | * | |
274 | * Description: | |
275 | * The block layer may perform asynchronous callback activity | |
276 | * on a queue, such as calling the unplug function after a timeout. | |
277 | * A block device may call blk_sync_queue to ensure that any | |
278 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 279 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
280 | * that its ->make_request_fn will not re-add plugging prior to calling |
281 | * this function. | |
282 | * | |
da527770 VG |
283 | * This function does not cancel any asynchronous activity arising |
284 | * out of elevator or throttling code. That would require elevaotor_exit() | |
5efd6113 | 285 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 286 | * |
1da177e4 LT |
287 | */ |
288 | void blk_sync_queue(struct request_queue *q) | |
289 | { | |
70ed28b9 | 290 | del_timer_sync(&q->timeout); |
3cca6dc1 | 291 | cancel_delayed_work_sync(&q->delay_work); |
1da177e4 LT |
292 | } |
293 | EXPORT_SYMBOL(blk_sync_queue); | |
294 | ||
295 | /** | |
80a4b58e | 296 | * __blk_run_queue - run a single device queue |
1da177e4 | 297 | * @q: The queue to run |
80a4b58e JA |
298 | * |
299 | * Description: | |
300 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 301 | * held and interrupts disabled. |
1da177e4 | 302 | */ |
24ecfbe2 | 303 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 304 | { |
a538cd03 TH |
305 | if (unlikely(blk_queue_stopped(q))) |
306 | return; | |
307 | ||
c21e6beb | 308 | q->request_fn(q); |
75ad23bc NP |
309 | } |
310 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 311 | |
24ecfbe2 CH |
312 | /** |
313 | * blk_run_queue_async - run a single device queue in workqueue context | |
314 | * @q: The queue to run | |
315 | * | |
316 | * Description: | |
317 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
318 | * of us. | |
319 | */ | |
320 | void blk_run_queue_async(struct request_queue *q) | |
321 | { | |
3ec717b7 SL |
322 | if (likely(!blk_queue_stopped(q))) { |
323 | __cancel_delayed_work(&q->delay_work); | |
24ecfbe2 | 324 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
3ec717b7 | 325 | } |
24ecfbe2 | 326 | } |
c21e6beb | 327 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 328 | |
75ad23bc NP |
329 | /** |
330 | * blk_run_queue - run a single device queue | |
331 | * @q: The queue to run | |
80a4b58e JA |
332 | * |
333 | * Description: | |
334 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 335 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
336 | */ |
337 | void blk_run_queue(struct request_queue *q) | |
338 | { | |
339 | unsigned long flags; | |
340 | ||
341 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 342 | __blk_run_queue(q); |
1da177e4 LT |
343 | spin_unlock_irqrestore(q->queue_lock, flags); |
344 | } | |
345 | EXPORT_SYMBOL(blk_run_queue); | |
346 | ||
165125e1 | 347 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
348 | { |
349 | kobject_put(&q->kobj); | |
350 | } | |
d86e0e83 | 351 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 352 | |
e3c78ca5 TH |
353 | /** |
354 | * blk_drain_queue - drain requests from request_queue | |
355 | * @q: queue to drain | |
c9a929dd | 356 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 357 | * |
c9a929dd TH |
358 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
359 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
360 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 361 | */ |
c9a929dd | 362 | void blk_drain_queue(struct request_queue *q, bool drain_all) |
e3c78ca5 TH |
363 | { |
364 | while (true) { | |
481a7d64 TH |
365 | bool drain = false; |
366 | int i; | |
e3c78ca5 TH |
367 | |
368 | spin_lock_irq(q->queue_lock); | |
369 | ||
b855b04a TH |
370 | /* |
371 | * The caller might be trying to drain @q before its | |
372 | * elevator is initialized. | |
373 | */ | |
374 | if (q->elevator) | |
375 | elv_drain_elevator(q); | |
376 | ||
5efd6113 | 377 | blkcg_drain_queue(q); |
e3c78ca5 | 378 | |
4eabc941 TH |
379 | /* |
380 | * This function might be called on a queue which failed | |
b855b04a TH |
381 | * driver init after queue creation or is not yet fully |
382 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
383 | * in such cases. Kick queue iff dispatch queue has | |
384 | * something on it and @q has request_fn set. | |
4eabc941 | 385 | */ |
b855b04a | 386 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 387 | __blk_run_queue(q); |
c9a929dd | 388 | |
481a7d64 TH |
389 | drain |= q->rq.elvpriv; |
390 | ||
391 | /* | |
392 | * Unfortunately, requests are queued at and tracked from | |
393 | * multiple places and there's no single counter which can | |
394 | * be drained. Check all the queues and counters. | |
395 | */ | |
396 | if (drain_all) { | |
397 | drain |= !list_empty(&q->queue_head); | |
398 | for (i = 0; i < 2; i++) { | |
399 | drain |= q->rq.count[i]; | |
400 | drain |= q->in_flight[i]; | |
401 | drain |= !list_empty(&q->flush_queue[i]); | |
402 | } | |
403 | } | |
e3c78ca5 TH |
404 | |
405 | spin_unlock_irq(q->queue_lock); | |
406 | ||
481a7d64 | 407 | if (!drain) |
e3c78ca5 TH |
408 | break; |
409 | msleep(10); | |
410 | } | |
411 | } | |
412 | ||
d732580b TH |
413 | /** |
414 | * blk_queue_bypass_start - enter queue bypass mode | |
415 | * @q: queue of interest | |
416 | * | |
417 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
418 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 419 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
420 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
421 | * inside queue or RCU read lock. | |
d732580b TH |
422 | */ |
423 | void blk_queue_bypass_start(struct request_queue *q) | |
424 | { | |
b82d4b19 TH |
425 | bool drain; |
426 | ||
d732580b | 427 | spin_lock_irq(q->queue_lock); |
b82d4b19 | 428 | drain = !q->bypass_depth++; |
d732580b TH |
429 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
430 | spin_unlock_irq(q->queue_lock); | |
431 | ||
b82d4b19 TH |
432 | if (drain) { |
433 | blk_drain_queue(q, false); | |
434 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ | |
435 | synchronize_rcu(); | |
436 | } | |
d732580b TH |
437 | } |
438 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
439 | ||
440 | /** | |
441 | * blk_queue_bypass_end - leave queue bypass mode | |
442 | * @q: queue of interest | |
443 | * | |
444 | * Leave bypass mode and restore the normal queueing behavior. | |
445 | */ | |
446 | void blk_queue_bypass_end(struct request_queue *q) | |
447 | { | |
448 | spin_lock_irq(q->queue_lock); | |
449 | if (!--q->bypass_depth) | |
450 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
451 | WARN_ON_ONCE(q->bypass_depth < 0); | |
452 | spin_unlock_irq(q->queue_lock); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
455 | ||
c9a929dd TH |
456 | /** |
457 | * blk_cleanup_queue - shutdown a request queue | |
458 | * @q: request queue to shutdown | |
459 | * | |
460 | * Mark @q DEAD, drain all pending requests, destroy and put it. All | |
461 | * future requests will be failed immediately with -ENODEV. | |
c94a96ac | 462 | */ |
6728cb0e | 463 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 464 | { |
c9a929dd | 465 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 466 | |
c9a929dd | 467 | /* mark @q DEAD, no new request or merges will be allowed afterwards */ |
483f4afc | 468 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 469 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
c9a929dd TH |
470 | |
471 | spin_lock_irq(lock); | |
6ecf23af | 472 | |
80fd9979 TH |
473 | /* |
474 | * Dead queue is permanently in bypass mode till released. Note | |
475 | * that, unlike blk_queue_bypass_start(), we aren't performing | |
476 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
477 | * as some drivers create and destroy a lot of queues while | |
478 | * probing. This is still safe because blk_release_queue() will be | |
479 | * called only after the queue refcnt drops to zero and nothing, | |
480 | * RCU or not, would be traversing the queue by then. | |
481 | */ | |
6ecf23af TH |
482 | q->bypass_depth++; |
483 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
484 | ||
c9a929dd TH |
485 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
486 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
487 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
483f4afc | 488 | |
777eb1bf HR |
489 | if (q->queue_lock != &q->__queue_lock) |
490 | q->queue_lock = &q->__queue_lock; | |
da527770 | 491 | |
c9a929dd TH |
492 | spin_unlock_irq(lock); |
493 | mutex_unlock(&q->sysfs_lock); | |
494 | ||
b855b04a TH |
495 | /* drain all requests queued before DEAD marking */ |
496 | blk_drain_queue(q, true); | |
c9a929dd TH |
497 | |
498 | /* @q won't process any more request, flush async actions */ | |
499 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
500 | blk_sync_queue(q); | |
501 | ||
502 | /* @q is and will stay empty, shutdown and put */ | |
483f4afc AV |
503 | blk_put_queue(q); |
504 | } | |
1da177e4 LT |
505 | EXPORT_SYMBOL(blk_cleanup_queue); |
506 | ||
165125e1 | 507 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
508 | { |
509 | struct request_list *rl = &q->rq; | |
510 | ||
1abec4fd MS |
511 | if (unlikely(rl->rq_pool)) |
512 | return 0; | |
513 | ||
1faa16d2 JA |
514 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
515 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 516 | rl->elvpriv = 0; |
1faa16d2 JA |
517 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
518 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 519 | |
1946089a CL |
520 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
521 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
522 | |
523 | if (!rl->rq_pool) | |
524 | return -ENOMEM; | |
525 | ||
526 | return 0; | |
527 | } | |
528 | ||
165125e1 | 529 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 530 | { |
1946089a CL |
531 | return blk_alloc_queue_node(gfp_mask, -1); |
532 | } | |
533 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 534 | |
165125e1 | 535 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 536 | { |
165125e1 | 537 | struct request_queue *q; |
e0bf68dd | 538 | int err; |
1946089a | 539 | |
8324aa91 | 540 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 541 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
542 | if (!q) |
543 | return NULL; | |
544 | ||
00380a40 | 545 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d TH |
546 | if (q->id < 0) |
547 | goto fail_q; | |
548 | ||
0989a025 JA |
549 | q->backing_dev_info.ra_pages = |
550 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
551 | q->backing_dev_info.state = 0; | |
552 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 553 | q->backing_dev_info.name = "block"; |
5151412d | 554 | q->node = node_id; |
0989a025 | 555 | |
e0bf68dd | 556 | err = bdi_init(&q->backing_dev_info); |
a73f730d TH |
557 | if (err) |
558 | goto fail_id; | |
e0bf68dd | 559 | |
31373d09 MG |
560 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
561 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 562 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 563 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 564 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 565 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 566 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 567 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 568 | #endif |
ae1b1539 TH |
569 | INIT_LIST_HEAD(&q->flush_queue[0]); |
570 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
571 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 572 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 573 | |
8324aa91 | 574 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 575 | |
483f4afc | 576 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 577 | spin_lock_init(&q->__queue_lock); |
483f4afc | 578 | |
c94a96ac VG |
579 | /* |
580 | * By default initialize queue_lock to internal lock and driver can | |
581 | * override it later if need be. | |
582 | */ | |
583 | q->queue_lock = &q->__queue_lock; | |
584 | ||
b82d4b19 TH |
585 | /* |
586 | * A queue starts its life with bypass turned on to avoid | |
587 | * unnecessary bypass on/off overhead and nasty surprises during | |
588 | * init. The initial bypass will be finished at the end of | |
589 | * blk_init_allocated_queue(). | |
590 | */ | |
591 | q->bypass_depth = 1; | |
592 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
593 | ||
5efd6113 | 594 | if (blkcg_init_queue(q)) |
f51b802c TH |
595 | goto fail_id; |
596 | ||
1da177e4 | 597 | return q; |
a73f730d TH |
598 | |
599 | fail_id: | |
600 | ida_simple_remove(&blk_queue_ida, q->id); | |
601 | fail_q: | |
602 | kmem_cache_free(blk_requestq_cachep, q); | |
603 | return NULL; | |
1da177e4 | 604 | } |
1946089a | 605 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
606 | |
607 | /** | |
608 | * blk_init_queue - prepare a request queue for use with a block device | |
609 | * @rfn: The function to be called to process requests that have been | |
610 | * placed on the queue. | |
611 | * @lock: Request queue spin lock | |
612 | * | |
613 | * Description: | |
614 | * If a block device wishes to use the standard request handling procedures, | |
615 | * which sorts requests and coalesces adjacent requests, then it must | |
616 | * call blk_init_queue(). The function @rfn will be called when there | |
617 | * are requests on the queue that need to be processed. If the device | |
618 | * supports plugging, then @rfn may not be called immediately when requests | |
619 | * are available on the queue, but may be called at some time later instead. | |
620 | * Plugged queues are generally unplugged when a buffer belonging to one | |
621 | * of the requests on the queue is needed, or due to memory pressure. | |
622 | * | |
623 | * @rfn is not required, or even expected, to remove all requests off the | |
624 | * queue, but only as many as it can handle at a time. If it does leave | |
625 | * requests on the queue, it is responsible for arranging that the requests | |
626 | * get dealt with eventually. | |
627 | * | |
628 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
629 | * request queue; this lock will be taken also from interrupt context, so irq |
630 | * disabling is needed for it. | |
1da177e4 | 631 | * |
710027a4 | 632 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
633 | * it didn't succeed. |
634 | * | |
635 | * Note: | |
636 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
637 | * when the block device is deactivated (such as at module unload). | |
638 | **/ | |
1946089a | 639 | |
165125e1 | 640 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 641 | { |
1946089a CL |
642 | return blk_init_queue_node(rfn, lock, -1); |
643 | } | |
644 | EXPORT_SYMBOL(blk_init_queue); | |
645 | ||
165125e1 | 646 | struct request_queue * |
1946089a CL |
647 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
648 | { | |
c86d1b8a | 649 | struct request_queue *uninit_q, *q; |
1da177e4 | 650 | |
c86d1b8a MS |
651 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
652 | if (!uninit_q) | |
653 | return NULL; | |
654 | ||
5151412d | 655 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a MS |
656 | if (!q) |
657 | blk_cleanup_queue(uninit_q); | |
658 | ||
659 | return q; | |
01effb0d MS |
660 | } |
661 | EXPORT_SYMBOL(blk_init_queue_node); | |
662 | ||
663 | struct request_queue * | |
664 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
665 | spinlock_t *lock) | |
01effb0d | 666 | { |
1da177e4 LT |
667 | if (!q) |
668 | return NULL; | |
669 | ||
c86d1b8a | 670 | if (blk_init_free_list(q)) |
8669aafd | 671 | return NULL; |
1da177e4 LT |
672 | |
673 | q->request_fn = rfn; | |
1da177e4 | 674 | q->prep_rq_fn = NULL; |
28018c24 | 675 | q->unprep_rq_fn = NULL; |
bc58ba94 | 676 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
677 | |
678 | /* Override internal queue lock with supplied lock pointer */ | |
679 | if (lock) | |
680 | q->queue_lock = lock; | |
1da177e4 | 681 | |
f3b144aa JA |
682 | /* |
683 | * This also sets hw/phys segments, boundary and size | |
684 | */ | |
c20e8de2 | 685 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 686 | |
44ec9542 AS |
687 | q->sg_reserved_size = INT_MAX; |
688 | ||
b82d4b19 TH |
689 | /* init elevator */ |
690 | if (elevator_init(q, NULL)) | |
691 | return NULL; | |
1da177e4 | 692 | |
b82d4b19 TH |
693 | blk_queue_congestion_threshold(q); |
694 | ||
695 | /* all done, end the initial bypass */ | |
696 | blk_queue_bypass_end(q); | |
697 | return q; | |
1da177e4 | 698 | } |
5151412d | 699 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 700 | |
09ac46c4 | 701 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 702 | { |
34f6055c | 703 | if (likely(!blk_queue_dead(q))) { |
09ac46c4 TH |
704 | __blk_get_queue(q); |
705 | return true; | |
1da177e4 LT |
706 | } |
707 | ||
09ac46c4 | 708 | return false; |
1da177e4 | 709 | } |
d86e0e83 | 710 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 711 | |
165125e1 | 712 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 713 | { |
f1f8cc94 | 714 | if (rq->cmd_flags & REQ_ELVPRIV) { |
cb98fc8b | 715 | elv_put_request(q, rq); |
f1f8cc94 | 716 | if (rq->elv.icq) |
11a3122f | 717 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
718 | } |
719 | ||
1da177e4 LT |
720 | mempool_free(rq, q->rq.rq_pool); |
721 | } | |
722 | ||
1da177e4 LT |
723 | /* |
724 | * ioc_batching returns true if the ioc is a valid batching request and | |
725 | * should be given priority access to a request. | |
726 | */ | |
165125e1 | 727 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
728 | { |
729 | if (!ioc) | |
730 | return 0; | |
731 | ||
732 | /* | |
733 | * Make sure the process is able to allocate at least 1 request | |
734 | * even if the batch times out, otherwise we could theoretically | |
735 | * lose wakeups. | |
736 | */ | |
737 | return ioc->nr_batch_requests == q->nr_batching || | |
738 | (ioc->nr_batch_requests > 0 | |
739 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
740 | } | |
741 | ||
742 | /* | |
743 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
744 | * will cause the process to be a "batcher" on all queues in the system. This | |
745 | * is the behaviour we want though - once it gets a wakeup it should be given | |
746 | * a nice run. | |
747 | */ | |
165125e1 | 748 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
749 | { |
750 | if (!ioc || ioc_batching(q, ioc)) | |
751 | return; | |
752 | ||
753 | ioc->nr_batch_requests = q->nr_batching; | |
754 | ioc->last_waited = jiffies; | |
755 | } | |
756 | ||
1faa16d2 | 757 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
758 | { |
759 | struct request_list *rl = &q->rq; | |
760 | ||
1faa16d2 JA |
761 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
762 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 763 | |
1faa16d2 JA |
764 | if (rl->count[sync] + 1 <= q->nr_requests) { |
765 | if (waitqueue_active(&rl->wait[sync])) | |
766 | wake_up(&rl->wait[sync]); | |
1da177e4 | 767 | |
1faa16d2 | 768 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
769 | } |
770 | } | |
771 | ||
772 | /* | |
773 | * A request has just been released. Account for it, update the full and | |
774 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
775 | */ | |
75eb6c37 | 776 | static void freed_request(struct request_queue *q, unsigned int flags) |
1da177e4 LT |
777 | { |
778 | struct request_list *rl = &q->rq; | |
75eb6c37 | 779 | int sync = rw_is_sync(flags); |
1da177e4 | 780 | |
1faa16d2 | 781 | rl->count[sync]--; |
75eb6c37 | 782 | if (flags & REQ_ELVPRIV) |
cb98fc8b | 783 | rl->elvpriv--; |
1da177e4 | 784 | |
1faa16d2 | 785 | __freed_request(q, sync); |
1da177e4 | 786 | |
1faa16d2 JA |
787 | if (unlikely(rl->starved[sync ^ 1])) |
788 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
789 | } |
790 | ||
9d5a4e94 MS |
791 | /* |
792 | * Determine if elevator data should be initialized when allocating the | |
793 | * request associated with @bio. | |
794 | */ | |
795 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
796 | { | |
797 | if (!bio) | |
798 | return true; | |
799 | ||
800 | /* | |
801 | * Flush requests do not use the elevator so skip initialization. | |
802 | * This allows a request to share the flush and elevator data. | |
803 | */ | |
804 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
805 | return false; | |
806 | ||
807 | return true; | |
808 | } | |
809 | ||
852c788f TH |
810 | /** |
811 | * rq_ioc - determine io_context for request allocation | |
812 | * @bio: request being allocated is for this bio (can be %NULL) | |
813 | * | |
814 | * Determine io_context to use for request allocation for @bio. May return | |
815 | * %NULL if %current->io_context doesn't exist. | |
816 | */ | |
817 | static struct io_context *rq_ioc(struct bio *bio) | |
818 | { | |
819 | #ifdef CONFIG_BLK_CGROUP | |
820 | if (bio && bio->bi_ioc) | |
821 | return bio->bi_ioc; | |
822 | #endif | |
823 | return current->io_context; | |
824 | } | |
825 | ||
da8303c6 TH |
826 | /** |
827 | * get_request - get a free request | |
828 | * @q: request_queue to allocate request from | |
829 | * @rw_flags: RW and SYNC flags | |
830 | * @bio: bio to allocate request for (can be %NULL) | |
831 | * @gfp_mask: allocation mask | |
832 | * | |
833 | * Get a free request from @q. This function may fail under memory | |
834 | * pressure or if @q is dead. | |
835 | * | |
836 | * Must be callled with @q->queue_lock held and, | |
837 | * Returns %NULL on failure, with @q->queue_lock held. | |
838 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 839 | */ |
165125e1 | 840 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 841 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 842 | { |
b679281a | 843 | struct request *rq; |
1da177e4 | 844 | struct request_list *rl = &q->rq; |
f1f8cc94 | 845 | struct elevator_type *et; |
f2dbd76a | 846 | struct io_context *ioc; |
f1f8cc94 | 847 | struct io_cq *icq = NULL; |
1faa16d2 | 848 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
f2dbd76a | 849 | bool retried = false; |
75eb6c37 | 850 | int may_queue; |
f2dbd76a | 851 | retry: |
f1f8cc94 | 852 | et = q->elevator->type; |
852c788f | 853 | ioc = rq_ioc(bio); |
88ee5ef1 | 854 | |
34f6055c | 855 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
856 | return NULL; |
857 | ||
7749a8d4 | 858 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
859 | if (may_queue == ELV_MQUEUE_NO) |
860 | goto rq_starved; | |
861 | ||
1faa16d2 JA |
862 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
863 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
f2dbd76a TH |
864 | /* |
865 | * We want ioc to record batching state. If it's | |
866 | * not already there, creating a new one requires | |
867 | * dropping queue_lock, which in turn requires | |
868 | * retesting conditions to avoid queue hang. | |
869 | */ | |
870 | if (!ioc && !retried) { | |
871 | spin_unlock_irq(q->queue_lock); | |
24acfc34 | 872 | create_io_context(gfp_mask, q->node); |
f2dbd76a TH |
873 | spin_lock_irq(q->queue_lock); |
874 | retried = true; | |
875 | goto retry; | |
876 | } | |
877 | ||
88ee5ef1 JA |
878 | /* |
879 | * The queue will fill after this allocation, so set | |
880 | * it as full, and mark this process as "batching". | |
881 | * This process will be allowed to complete a batch of | |
882 | * requests, others will be blocked. | |
883 | */ | |
1faa16d2 | 884 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 885 | ioc_set_batching(q, ioc); |
1faa16d2 | 886 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
887 | } else { |
888 | if (may_queue != ELV_MQUEUE_MUST | |
889 | && !ioc_batching(q, ioc)) { | |
890 | /* | |
891 | * The queue is full and the allocating | |
892 | * process is not a "batcher", and not | |
893 | * exempted by the IO scheduler | |
894 | */ | |
b679281a | 895 | return NULL; |
88ee5ef1 JA |
896 | } |
897 | } | |
1da177e4 | 898 | } |
1faa16d2 | 899 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
900 | } |
901 | ||
082cf69e JA |
902 | /* |
903 | * Only allow batching queuers to allocate up to 50% over the defined | |
904 | * limit of requests, otherwise we could have thousands of requests | |
905 | * allocated with any setting of ->nr_requests | |
906 | */ | |
1faa16d2 | 907 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
b679281a | 908 | return NULL; |
fd782a4a | 909 | |
1faa16d2 JA |
910 | rl->count[is_sync]++; |
911 | rl->starved[is_sync] = 0; | |
cb98fc8b | 912 | |
f1f8cc94 TH |
913 | /* |
914 | * Decide whether the new request will be managed by elevator. If | |
915 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will | |
916 | * prevent the current elevator from being destroyed until the new | |
917 | * request is freed. This guarantees icq's won't be destroyed and | |
918 | * makes creating new ones safe. | |
919 | * | |
920 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
921 | * it will be created after releasing queue_lock. | |
922 | */ | |
d732580b | 923 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
75eb6c37 TH |
924 | rw_flags |= REQ_ELVPRIV; |
925 | rl->elvpriv++; | |
f1f8cc94 TH |
926 | if (et->icq_cache && ioc) |
927 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 928 | } |
cb98fc8b | 929 | |
f253b86b JA |
930 | if (blk_queue_io_stat(q)) |
931 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
932 | spin_unlock_irq(q->queue_lock); |
933 | ||
29e2b09a TH |
934 | /* allocate and init request */ |
935 | rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
936 | if (!rq) | |
b679281a | 937 | goto fail_alloc; |
1da177e4 | 938 | |
29e2b09a TH |
939 | blk_rq_init(q, rq); |
940 | rq->cmd_flags = rw_flags | REQ_ALLOCED; | |
941 | ||
aaf7c680 | 942 | /* init elvpriv */ |
29e2b09a | 943 | if (rw_flags & REQ_ELVPRIV) { |
aaf7c680 TH |
944 | if (unlikely(et->icq_cache && !icq)) { |
945 | create_io_context(gfp_mask, q->node); | |
946 | ioc = rq_ioc(bio); | |
947 | if (!ioc) | |
948 | goto fail_elvpriv; | |
949 | ||
950 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
951 | if (!icq) | |
952 | goto fail_elvpriv; | |
29e2b09a | 953 | } |
aaf7c680 TH |
954 | |
955 | rq->elv.icq = icq; | |
956 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
957 | goto fail_elvpriv; | |
958 | ||
959 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
960 | if (icq) |
961 | get_io_context(icq->ioc); | |
962 | } | |
aaf7c680 | 963 | out: |
88ee5ef1 JA |
964 | /* |
965 | * ioc may be NULL here, and ioc_batching will be false. That's | |
966 | * OK, if the queue is under the request limit then requests need | |
967 | * not count toward the nr_batch_requests limit. There will always | |
968 | * be some limit enforced by BLK_BATCH_TIME. | |
969 | */ | |
1da177e4 LT |
970 | if (ioc_batching(q, ioc)) |
971 | ioc->nr_batch_requests--; | |
6728cb0e | 972 | |
1faa16d2 | 973 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 974 | return rq; |
b679281a | 975 | |
aaf7c680 TH |
976 | fail_elvpriv: |
977 | /* | |
978 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
979 | * and may fail indefinitely under memory pressure and thus | |
980 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
981 | * disturb iosched and blkcg but weird is bettern than dead. | |
982 | */ | |
983 | printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n", | |
984 | dev_name(q->backing_dev_info.dev)); | |
985 | ||
986 | rq->cmd_flags &= ~REQ_ELVPRIV; | |
987 | rq->elv.icq = NULL; | |
988 | ||
989 | spin_lock_irq(q->queue_lock); | |
990 | rl->elvpriv--; | |
991 | spin_unlock_irq(q->queue_lock); | |
992 | goto out; | |
993 | ||
b679281a TH |
994 | fail_alloc: |
995 | /* | |
996 | * Allocation failed presumably due to memory. Undo anything we | |
997 | * might have messed up. | |
998 | * | |
999 | * Allocating task should really be put onto the front of the wait | |
1000 | * queue, but this is pretty rare. | |
1001 | */ | |
1002 | spin_lock_irq(q->queue_lock); | |
1003 | freed_request(q, rw_flags); | |
1004 | ||
1005 | /* | |
1006 | * in the very unlikely event that allocation failed and no | |
1007 | * requests for this direction was pending, mark us starved so that | |
1008 | * freeing of a request in the other direction will notice | |
1009 | * us. another possible fix would be to split the rq mempool into | |
1010 | * READ and WRITE | |
1011 | */ | |
1012 | rq_starved: | |
1013 | if (unlikely(rl->count[is_sync] == 0)) | |
1014 | rl->starved[is_sync] = 1; | |
1015 | return NULL; | |
1da177e4 LT |
1016 | } |
1017 | ||
da8303c6 TH |
1018 | /** |
1019 | * get_request_wait - get a free request with retry | |
1020 | * @q: request_queue to allocate request from | |
1021 | * @rw_flags: RW and SYNC flags | |
1022 | * @bio: bio to allocate request for (can be %NULL) | |
1023 | * | |
1024 | * Get a free request from @q. This function keeps retrying under memory | |
1025 | * pressure and fails iff @q is dead. | |
d6344532 | 1026 | * |
da8303c6 TH |
1027 | * Must be callled with @q->queue_lock held and, |
1028 | * Returns %NULL on failure, with @q->queue_lock held. | |
1029 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 1030 | */ |
165125e1 | 1031 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 1032 | struct bio *bio) |
1da177e4 | 1033 | { |
1faa16d2 | 1034 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
1035 | struct request *rq; |
1036 | ||
7749a8d4 | 1037 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
1038 | while (!rq) { |
1039 | DEFINE_WAIT(wait); | |
1da177e4 LT |
1040 | struct request_list *rl = &q->rq; |
1041 | ||
34f6055c | 1042 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
1043 | return NULL; |
1044 | ||
1faa16d2 | 1045 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
1046 | TASK_UNINTERRUPTIBLE); |
1047 | ||
1faa16d2 | 1048 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 1049 | |
05caf8db ZY |
1050 | spin_unlock_irq(q->queue_lock); |
1051 | io_schedule(); | |
1da177e4 | 1052 | |
05caf8db ZY |
1053 | /* |
1054 | * After sleeping, we become a "batching" process and | |
1055 | * will be able to allocate at least one request, and | |
1056 | * up to a big batch of them for a small period time. | |
1057 | * See ioc_batching, ioc_set_batching | |
1058 | */ | |
24acfc34 | 1059 | create_io_context(GFP_NOIO, q->node); |
f2dbd76a | 1060 | ioc_set_batching(q, current->io_context); |
d6344532 | 1061 | |
05caf8db | 1062 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 1063 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
1064 | |
1065 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
1066 | }; | |
1da177e4 LT |
1067 | |
1068 | return rq; | |
1069 | } | |
1070 | ||
165125e1 | 1071 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
1072 | { |
1073 | struct request *rq; | |
1074 | ||
1075 | BUG_ON(rw != READ && rw != WRITE); | |
1076 | ||
d6344532 | 1077 | spin_lock_irq(q->queue_lock); |
da8303c6 | 1078 | if (gfp_mask & __GFP_WAIT) |
22e2c507 | 1079 | rq = get_request_wait(q, rw, NULL); |
da8303c6 | 1080 | else |
22e2c507 | 1081 | rq = get_request(q, rw, NULL, gfp_mask); |
da8303c6 TH |
1082 | if (!rq) |
1083 | spin_unlock_irq(q->queue_lock); | |
d6344532 | 1084 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
1085 | |
1086 | return rq; | |
1087 | } | |
1da177e4 LT |
1088 | EXPORT_SYMBOL(blk_get_request); |
1089 | ||
dc72ef4a | 1090 | /** |
79eb63e9 | 1091 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 1092 | * @q: target request queue |
79eb63e9 BH |
1093 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
1094 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 1095 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 1096 | * |
79eb63e9 BH |
1097 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
1098 | * type commands. Where the struct request needs to be farther initialized by | |
1099 | * the caller. It is passed a &struct bio, which describes the memory info of | |
1100 | * the I/O transfer. | |
dc72ef4a | 1101 | * |
79eb63e9 BH |
1102 | * The caller of blk_make_request must make sure that bi_io_vec |
1103 | * are set to describe the memory buffers. That bio_data_dir() will return | |
1104 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
1105 | * are properly set accordingly) | |
1106 | * | |
1107 | * If called under none-sleepable conditions, mapped bio buffers must not | |
1108 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
1109 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
1110 | * BUG. | |
53674ac5 JA |
1111 | * |
1112 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
1113 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
1114 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
1115 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
1116 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
1117 | * of bio_alloc(), as that avoids the mempool deadlock. | |
1118 | * If possible a big IO should be split into smaller parts when allocation | |
1119 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 1120 | */ |
79eb63e9 BH |
1121 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
1122 | gfp_t gfp_mask) | |
dc72ef4a | 1123 | { |
79eb63e9 BH |
1124 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1125 | ||
1126 | if (unlikely(!rq)) | |
1127 | return ERR_PTR(-ENOMEM); | |
1128 | ||
1129 | for_each_bio(bio) { | |
1130 | struct bio *bounce_bio = bio; | |
1131 | int ret; | |
1132 | ||
1133 | blk_queue_bounce(q, &bounce_bio); | |
1134 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1135 | if (unlikely(ret)) { | |
1136 | blk_put_request(rq); | |
1137 | return ERR_PTR(ret); | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | return rq; | |
dc72ef4a | 1142 | } |
79eb63e9 | 1143 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1144 | |
1da177e4 LT |
1145 | /** |
1146 | * blk_requeue_request - put a request back on queue | |
1147 | * @q: request queue where request should be inserted | |
1148 | * @rq: request to be inserted | |
1149 | * | |
1150 | * Description: | |
1151 | * Drivers often keep queueing requests until the hardware cannot accept | |
1152 | * more, when that condition happens we need to put the request back | |
1153 | * on the queue. Must be called with queue lock held. | |
1154 | */ | |
165125e1 | 1155 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1156 | { |
242f9dcb JA |
1157 | blk_delete_timer(rq); |
1158 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1159 | trace_block_rq_requeue(q, rq); |
2056a782 | 1160 | |
1da177e4 LT |
1161 | if (blk_rq_tagged(rq)) |
1162 | blk_queue_end_tag(q, rq); | |
1163 | ||
ba396a6c JB |
1164 | BUG_ON(blk_queued_rq(rq)); |
1165 | ||
1da177e4 LT |
1166 | elv_requeue_request(q, rq); |
1167 | } | |
1da177e4 LT |
1168 | EXPORT_SYMBOL(blk_requeue_request); |
1169 | ||
73c10101 JA |
1170 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1171 | int where) | |
1172 | { | |
1173 | drive_stat_acct(rq, 1); | |
7eaceacc | 1174 | __elv_add_request(q, rq, where); |
73c10101 JA |
1175 | } |
1176 | ||
074a7aca TH |
1177 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1178 | unsigned long now) | |
1179 | { | |
1180 | if (now == part->stamp) | |
1181 | return; | |
1182 | ||
316d315b | 1183 | if (part_in_flight(part)) { |
074a7aca | 1184 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1185 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1186 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1187 | } | |
1188 | part->stamp = now; | |
1189 | } | |
1190 | ||
1191 | /** | |
496aa8a9 RD |
1192 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1193 | * @cpu: cpu number for stats access | |
1194 | * @part: target partition | |
1da177e4 LT |
1195 | * |
1196 | * The average IO queue length and utilisation statistics are maintained | |
1197 | * by observing the current state of the queue length and the amount of | |
1198 | * time it has been in this state for. | |
1199 | * | |
1200 | * Normally, that accounting is done on IO completion, but that can result | |
1201 | * in more than a second's worth of IO being accounted for within any one | |
1202 | * second, leading to >100% utilisation. To deal with that, we call this | |
1203 | * function to do a round-off before returning the results when reading | |
1204 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1205 | * the current jiffies and restarts the counters again. | |
1206 | */ | |
c9959059 | 1207 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1208 | { |
1209 | unsigned long now = jiffies; | |
1210 | ||
074a7aca TH |
1211 | if (part->partno) |
1212 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1213 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1214 | } |
074a7aca | 1215 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1216 | |
1da177e4 LT |
1217 | /* |
1218 | * queue lock must be held | |
1219 | */ | |
165125e1 | 1220 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1221 | { |
1da177e4 LT |
1222 | if (unlikely(!q)) |
1223 | return; | |
1224 | if (unlikely(--req->ref_count)) | |
1225 | return; | |
1226 | ||
8922e16c TH |
1227 | elv_completed_request(q, req); |
1228 | ||
1cd96c24 BH |
1229 | /* this is a bio leak */ |
1230 | WARN_ON(req->bio != NULL); | |
1231 | ||
1da177e4 LT |
1232 | /* |
1233 | * Request may not have originated from ll_rw_blk. if not, | |
1234 | * it didn't come out of our reserved rq pools | |
1235 | */ | |
49171e5c | 1236 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1237 | unsigned int flags = req->cmd_flags; |
1da177e4 | 1238 | |
1da177e4 | 1239 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1240 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1241 | |
1242 | blk_free_request(q, req); | |
75eb6c37 | 1243 | freed_request(q, flags); |
1da177e4 LT |
1244 | } |
1245 | } | |
6e39b69e MC |
1246 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1247 | ||
1da177e4 LT |
1248 | void blk_put_request(struct request *req) |
1249 | { | |
8922e16c | 1250 | unsigned long flags; |
165125e1 | 1251 | struct request_queue *q = req->q; |
8922e16c | 1252 | |
52a93ba8 FT |
1253 | spin_lock_irqsave(q->queue_lock, flags); |
1254 | __blk_put_request(q, req); | |
1255 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1256 | } |
1da177e4 LT |
1257 | EXPORT_SYMBOL(blk_put_request); |
1258 | ||
66ac0280 CH |
1259 | /** |
1260 | * blk_add_request_payload - add a payload to a request | |
1261 | * @rq: request to update | |
1262 | * @page: page backing the payload | |
1263 | * @len: length of the payload. | |
1264 | * | |
1265 | * This allows to later add a payload to an already submitted request by | |
1266 | * a block driver. The driver needs to take care of freeing the payload | |
1267 | * itself. | |
1268 | * | |
1269 | * Note that this is a quite horrible hack and nothing but handling of | |
1270 | * discard requests should ever use it. | |
1271 | */ | |
1272 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1273 | unsigned int len) | |
1274 | { | |
1275 | struct bio *bio = rq->bio; | |
1276 | ||
1277 | bio->bi_io_vec->bv_page = page; | |
1278 | bio->bi_io_vec->bv_offset = 0; | |
1279 | bio->bi_io_vec->bv_len = len; | |
1280 | ||
1281 | bio->bi_size = len; | |
1282 | bio->bi_vcnt = 1; | |
1283 | bio->bi_phys_segments = 1; | |
1284 | ||
1285 | rq->__data_len = rq->resid_len = len; | |
1286 | rq->nr_phys_segments = 1; | |
1287 | rq->buffer = bio_data(bio); | |
1288 | } | |
1289 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1290 | ||
73c10101 JA |
1291 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1292 | struct bio *bio) | |
1293 | { | |
1294 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1295 | ||
73c10101 JA |
1296 | if (!ll_back_merge_fn(q, req, bio)) |
1297 | return false; | |
1298 | ||
1299 | trace_block_bio_backmerge(q, bio); | |
1300 | ||
1301 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1302 | blk_rq_set_mixed_merge(req); | |
1303 | ||
1304 | req->biotail->bi_next = bio; | |
1305 | req->biotail = bio; | |
1306 | req->__data_len += bio->bi_size; | |
1307 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1308 | ||
1309 | drive_stat_acct(req, 0); | |
1310 | return true; | |
1311 | } | |
1312 | ||
1313 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1314 | struct request *req, struct bio *bio) | |
1315 | { | |
1316 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1317 | |
73c10101 JA |
1318 | if (!ll_front_merge_fn(q, req, bio)) |
1319 | return false; | |
1320 | ||
1321 | trace_block_bio_frontmerge(q, bio); | |
1322 | ||
1323 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1324 | blk_rq_set_mixed_merge(req); | |
1325 | ||
73c10101 JA |
1326 | bio->bi_next = req->bio; |
1327 | req->bio = bio; | |
1328 | ||
1329 | /* | |
1330 | * may not be valid. if the low level driver said | |
1331 | * it didn't need a bounce buffer then it better | |
1332 | * not touch req->buffer either... | |
1333 | */ | |
1334 | req->buffer = bio_data(bio); | |
1335 | req->__sector = bio->bi_sector; | |
1336 | req->__data_len += bio->bi_size; | |
1337 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1338 | ||
1339 | drive_stat_acct(req, 0); | |
1340 | return true; | |
1341 | } | |
1342 | ||
bd87b589 TH |
1343 | /** |
1344 | * attempt_plug_merge - try to merge with %current's plugged list | |
1345 | * @q: request_queue new bio is being queued at | |
1346 | * @bio: new bio being queued | |
1347 | * @request_count: out parameter for number of traversed plugged requests | |
1348 | * | |
1349 | * Determine whether @bio being queued on @q can be merged with a request | |
1350 | * on %current's plugged list. Returns %true if merge was successful, | |
1351 | * otherwise %false. | |
1352 | * | |
07c2bd37 TH |
1353 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1354 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1355 | * than scheduling, and the request, while may have elvpriv data, is not | |
1356 | * added on the elevator at this point. In addition, we don't have | |
1357 | * reliable access to the elevator outside queue lock. Only check basic | |
1358 | * merging parameters without querying the elevator. | |
73c10101 | 1359 | */ |
bd87b589 TH |
1360 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1361 | unsigned int *request_count) | |
73c10101 JA |
1362 | { |
1363 | struct blk_plug *plug; | |
1364 | struct request *rq; | |
1365 | bool ret = false; | |
1366 | ||
bd87b589 | 1367 | plug = current->plug; |
73c10101 JA |
1368 | if (!plug) |
1369 | goto out; | |
56ebdaf2 | 1370 | *request_count = 0; |
73c10101 JA |
1371 | |
1372 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1373 | int el_ret; | |
1374 | ||
1b2e19f1 SL |
1375 | if (rq->q == q) |
1376 | (*request_count)++; | |
56ebdaf2 | 1377 | |
07c2bd37 | 1378 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1379 | continue; |
1380 | ||
050c8ea8 | 1381 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1382 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1383 | ret = bio_attempt_back_merge(q, rq, bio); | |
1384 | if (ret) | |
1385 | break; | |
1386 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1387 | ret = bio_attempt_front_merge(q, rq, bio); | |
1388 | if (ret) | |
1389 | break; | |
1390 | } | |
1391 | } | |
1392 | out: | |
1393 | return ret; | |
1394 | } | |
1395 | ||
86db1e29 | 1396 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1397 | { |
4aff5e23 | 1398 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1399 | |
7b6d91da CH |
1400 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1401 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1402 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1403 | |
52d9e675 | 1404 | req->errors = 0; |
a2dec7b3 | 1405 | req->__sector = bio->bi_sector; |
52d9e675 | 1406 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1407 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1408 | } |
1409 | ||
5a7bbad2 | 1410 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1411 | { |
5e00d1b5 | 1412 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1413 | struct blk_plug *plug; |
1414 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1415 | struct request *req; | |
56ebdaf2 | 1416 | unsigned int request_count = 0; |
1da177e4 | 1417 | |
1da177e4 LT |
1418 | /* |
1419 | * low level driver can indicate that it wants pages above a | |
1420 | * certain limit bounced to low memory (ie for highmem, or even | |
1421 | * ISA dma in theory) | |
1422 | */ | |
1423 | blk_queue_bounce(q, &bio); | |
1424 | ||
4fed947c | 1425 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1426 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1427 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1428 | goto get_rq; |
1429 | } | |
1430 | ||
73c10101 JA |
1431 | /* |
1432 | * Check if we can merge with the plugged list before grabbing | |
1433 | * any locks. | |
1434 | */ | |
bd87b589 | 1435 | if (attempt_plug_merge(q, bio, &request_count)) |
5a7bbad2 | 1436 | return; |
1da177e4 | 1437 | |
73c10101 | 1438 | spin_lock_irq(q->queue_lock); |
2056a782 | 1439 | |
73c10101 JA |
1440 | el_ret = elv_merge(q, &req, bio); |
1441 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1442 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1443 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1444 | if (!attempt_back_merge(q, req)) |
1445 | elv_merged_request(q, req, el_ret); | |
1446 | goto out_unlock; | |
1447 | } | |
1448 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1449 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1450 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1451 | if (!attempt_front_merge(q, req)) |
1452 | elv_merged_request(q, req, el_ret); | |
1453 | goto out_unlock; | |
80a761fd | 1454 | } |
1da177e4 LT |
1455 | } |
1456 | ||
450991bc | 1457 | get_rq: |
7749a8d4 JA |
1458 | /* |
1459 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1460 | * but we need to set it earlier to expose the sync flag to the | |
1461 | * rq allocator and io schedulers. | |
1462 | */ | |
1463 | rw_flags = bio_data_dir(bio); | |
1464 | if (sync) | |
7b6d91da | 1465 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1466 | |
1da177e4 | 1467 | /* |
450991bc | 1468 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1469 | * Returns with the queue unlocked. |
450991bc | 1470 | */ |
7749a8d4 | 1471 | req = get_request_wait(q, rw_flags, bio); |
da8303c6 TH |
1472 | if (unlikely(!req)) { |
1473 | bio_endio(bio, -ENODEV); /* @q is dead */ | |
1474 | goto out_unlock; | |
1475 | } | |
d6344532 | 1476 | |
450991bc NP |
1477 | /* |
1478 | * After dropping the lock and possibly sleeping here, our request | |
1479 | * may now be mergeable after it had proven unmergeable (above). | |
1480 | * We don't worry about that case for efficiency. It won't happen | |
1481 | * often, and the elevators are able to handle it. | |
1da177e4 | 1482 | */ |
52d9e675 | 1483 | init_request_from_bio(req, bio); |
1da177e4 | 1484 | |
9562ad9a | 1485 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1486 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1487 | |
1488 | plug = current->plug; | |
721a9602 | 1489 | if (plug) { |
dc6d36c9 JA |
1490 | /* |
1491 | * If this is the first request added after a plug, fire | |
1492 | * of a plug trace. If others have been added before, check | |
1493 | * if we have multiple devices in this plug. If so, make a | |
1494 | * note to sort the list before dispatch. | |
1495 | */ | |
1496 | if (list_empty(&plug->list)) | |
1497 | trace_block_plug(q); | |
3540d5e8 SL |
1498 | else { |
1499 | if (!plug->should_sort) { | |
1500 | struct request *__rq; | |
73c10101 | 1501 | |
3540d5e8 SL |
1502 | __rq = list_entry_rq(plug->list.prev); |
1503 | if (__rq->q != q) | |
1504 | plug->should_sort = 1; | |
1505 | } | |
019ceb7d | 1506 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
3540d5e8 | 1507 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1508 | trace_block_plug(q); |
1509 | } | |
73c10101 | 1510 | } |
73c10101 JA |
1511 | list_add_tail(&req->queuelist, &plug->list); |
1512 | drive_stat_acct(req, 1); | |
1513 | } else { | |
1514 | spin_lock_irq(q->queue_lock); | |
1515 | add_acct_request(q, req, where); | |
24ecfbe2 | 1516 | __blk_run_queue(q); |
73c10101 JA |
1517 | out_unlock: |
1518 | spin_unlock_irq(q->queue_lock); | |
1519 | } | |
1da177e4 | 1520 | } |
c20e8de2 | 1521 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1522 | |
1523 | /* | |
1524 | * If bio->bi_dev is a partition, remap the location | |
1525 | */ | |
1526 | static inline void blk_partition_remap(struct bio *bio) | |
1527 | { | |
1528 | struct block_device *bdev = bio->bi_bdev; | |
1529 | ||
bf2de6f5 | 1530 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1531 | struct hd_struct *p = bdev->bd_part; |
1532 | ||
1da177e4 LT |
1533 | bio->bi_sector += p->start_sect; |
1534 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1535 | |
d07335e5 MS |
1536 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1537 | bdev->bd_dev, | |
1538 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1539 | } |
1540 | } | |
1541 | ||
1da177e4 LT |
1542 | static void handle_bad_sector(struct bio *bio) |
1543 | { | |
1544 | char b[BDEVNAME_SIZE]; | |
1545 | ||
1546 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1547 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1548 | bdevname(bio->bi_bdev, b), | |
1549 | bio->bi_rw, | |
1550 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1551 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1552 | |
1553 | set_bit(BIO_EOF, &bio->bi_flags); | |
1554 | } | |
1555 | ||
c17bb495 AM |
1556 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1557 | ||
1558 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1559 | ||
1560 | static int __init setup_fail_make_request(char *str) | |
1561 | { | |
1562 | return setup_fault_attr(&fail_make_request, str); | |
1563 | } | |
1564 | __setup("fail_make_request=", setup_fail_make_request); | |
1565 | ||
b2c9cd37 | 1566 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1567 | { |
b2c9cd37 | 1568 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1569 | } |
1570 | ||
1571 | static int __init fail_make_request_debugfs(void) | |
1572 | { | |
dd48c085 AM |
1573 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1574 | NULL, &fail_make_request); | |
1575 | ||
1576 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; | |
c17bb495 AM |
1577 | } |
1578 | ||
1579 | late_initcall(fail_make_request_debugfs); | |
1580 | ||
1581 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1582 | ||
b2c9cd37 AM |
1583 | static inline bool should_fail_request(struct hd_struct *part, |
1584 | unsigned int bytes) | |
c17bb495 | 1585 | { |
b2c9cd37 | 1586 | return false; |
c17bb495 AM |
1587 | } |
1588 | ||
1589 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1590 | ||
c07e2b41 JA |
1591 | /* |
1592 | * Check whether this bio extends beyond the end of the device. | |
1593 | */ | |
1594 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1595 | { | |
1596 | sector_t maxsector; | |
1597 | ||
1598 | if (!nr_sectors) | |
1599 | return 0; | |
1600 | ||
1601 | /* Test device or partition size, when known. */ | |
77304d2a | 1602 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1603 | if (maxsector) { |
1604 | sector_t sector = bio->bi_sector; | |
1605 | ||
1606 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1607 | /* | |
1608 | * This may well happen - the kernel calls bread() | |
1609 | * without checking the size of the device, e.g., when | |
1610 | * mounting a device. | |
1611 | */ | |
1612 | handle_bad_sector(bio); | |
1613 | return 1; | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | return 0; | |
1618 | } | |
1619 | ||
27a84d54 CH |
1620 | static noinline_for_stack bool |
1621 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1622 | { |
165125e1 | 1623 | struct request_queue *q; |
5a7bbad2 | 1624 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1625 | int err = -EIO; |
5a7bbad2 CH |
1626 | char b[BDEVNAME_SIZE]; |
1627 | struct hd_struct *part; | |
1da177e4 LT |
1628 | |
1629 | might_sleep(); | |
1da177e4 | 1630 | |
c07e2b41 JA |
1631 | if (bio_check_eod(bio, nr_sectors)) |
1632 | goto end_io; | |
1da177e4 | 1633 | |
5a7bbad2 CH |
1634 | q = bdev_get_queue(bio->bi_bdev); |
1635 | if (unlikely(!q)) { | |
1636 | printk(KERN_ERR | |
1637 | "generic_make_request: Trying to access " | |
1638 | "nonexistent block-device %s (%Lu)\n", | |
1639 | bdevname(bio->bi_bdev, b), | |
1640 | (long long) bio->bi_sector); | |
1641 | goto end_io; | |
1642 | } | |
c17bb495 | 1643 | |
5a7bbad2 CH |
1644 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
1645 | nr_sectors > queue_max_hw_sectors(q))) { | |
1646 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1647 | bdevname(bio->bi_bdev, b), | |
1648 | bio_sectors(bio), | |
1649 | queue_max_hw_sectors(q)); | |
1650 | goto end_io; | |
1651 | } | |
1da177e4 | 1652 | |
5a7bbad2 CH |
1653 | part = bio->bi_bdev->bd_part; |
1654 | if (should_fail_request(part, bio->bi_size) || | |
1655 | should_fail_request(&part_to_disk(part)->part0, | |
1656 | bio->bi_size)) | |
1657 | goto end_io; | |
2056a782 | 1658 | |
5a7bbad2 CH |
1659 | /* |
1660 | * If this device has partitions, remap block n | |
1661 | * of partition p to block n+start(p) of the disk. | |
1662 | */ | |
1663 | blk_partition_remap(bio); | |
2056a782 | 1664 | |
5a7bbad2 CH |
1665 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1666 | goto end_io; | |
a7384677 | 1667 | |
5a7bbad2 CH |
1668 | if (bio_check_eod(bio, nr_sectors)) |
1669 | goto end_io; | |
1e87901e | 1670 | |
5a7bbad2 CH |
1671 | /* |
1672 | * Filter flush bio's early so that make_request based | |
1673 | * drivers without flush support don't have to worry | |
1674 | * about them. | |
1675 | */ | |
1676 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1677 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1678 | if (!nr_sectors) { | |
1679 | err = 0; | |
51fd77bd JA |
1680 | goto end_io; |
1681 | } | |
5a7bbad2 | 1682 | } |
5ddfe969 | 1683 | |
5a7bbad2 CH |
1684 | if ((bio->bi_rw & REQ_DISCARD) && |
1685 | (!blk_queue_discard(q) || | |
1686 | ((bio->bi_rw & REQ_SECURE) && | |
1687 | !blk_queue_secdiscard(q)))) { | |
1688 | err = -EOPNOTSUPP; | |
1689 | goto end_io; | |
1690 | } | |
01edede4 | 1691 | |
bc16a4f9 TH |
1692 | if (blk_throtl_bio(q, bio)) |
1693 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1694 | |
5a7bbad2 | 1695 | trace_block_bio_queue(q, bio); |
27a84d54 | 1696 | return true; |
a7384677 TH |
1697 | |
1698 | end_io: | |
1699 | bio_endio(bio, err); | |
27a84d54 | 1700 | return false; |
1da177e4 LT |
1701 | } |
1702 | ||
27a84d54 CH |
1703 | /** |
1704 | * generic_make_request - hand a buffer to its device driver for I/O | |
1705 | * @bio: The bio describing the location in memory and on the device. | |
1706 | * | |
1707 | * generic_make_request() is used to make I/O requests of block | |
1708 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1709 | * to be done. | |
1710 | * | |
1711 | * generic_make_request() does not return any status. The | |
1712 | * success/failure status of the request, along with notification of | |
1713 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1714 | * function described (one day) else where. | |
1715 | * | |
1716 | * The caller of generic_make_request must make sure that bi_io_vec | |
1717 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1718 | * set to describe the device address, and the | |
1719 | * bi_end_io and optionally bi_private are set to describe how | |
1720 | * completion notification should be signaled. | |
1721 | * | |
1722 | * generic_make_request and the drivers it calls may use bi_next if this | |
1723 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1724 | * a lower device by calling into generic_make_request recursively, which | |
1725 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1726 | */ |
1727 | void generic_make_request(struct bio *bio) | |
1728 | { | |
bddd87c7 AM |
1729 | struct bio_list bio_list_on_stack; |
1730 | ||
27a84d54 CH |
1731 | if (!generic_make_request_checks(bio)) |
1732 | return; | |
1733 | ||
1734 | /* | |
1735 | * We only want one ->make_request_fn to be active at a time, else | |
1736 | * stack usage with stacked devices could be a problem. So use | |
1737 | * current->bio_list to keep a list of requests submited by a | |
1738 | * make_request_fn function. current->bio_list is also used as a | |
1739 | * flag to say if generic_make_request is currently active in this | |
1740 | * task or not. If it is NULL, then no make_request is active. If | |
1741 | * it is non-NULL, then a make_request is active, and new requests | |
1742 | * should be added at the tail | |
1743 | */ | |
bddd87c7 | 1744 | if (current->bio_list) { |
bddd87c7 | 1745 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1746 | return; |
1747 | } | |
27a84d54 | 1748 | |
d89d8796 NB |
1749 | /* following loop may be a bit non-obvious, and so deserves some |
1750 | * explanation. | |
1751 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1752 | * ensure that) so we have a list with a single bio. | |
1753 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1754 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1755 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1756 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1757 | * through a recursive call to generic_make_request. If it |
1758 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1759 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1760 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1761 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1762 | */ |
1763 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1764 | bio_list_init(&bio_list_on_stack); |
1765 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1766 | do { |
27a84d54 CH |
1767 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1768 | ||
1769 | q->make_request_fn(q, bio); | |
1770 | ||
bddd87c7 | 1771 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1772 | } while (bio); |
bddd87c7 | 1773 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1774 | } |
1da177e4 LT |
1775 | EXPORT_SYMBOL(generic_make_request); |
1776 | ||
1777 | /** | |
710027a4 | 1778 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1779 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1780 | * @bio: The &struct bio which describes the I/O | |
1781 | * | |
1782 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1783 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1784 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1785 | * |
1786 | */ | |
1787 | void submit_bio(int rw, struct bio *bio) | |
1788 | { | |
1789 | int count = bio_sectors(bio); | |
1790 | ||
22e2c507 | 1791 | bio->bi_rw |= rw; |
1da177e4 | 1792 | |
bf2de6f5 JA |
1793 | /* |
1794 | * If it's a regular read/write or a barrier with data attached, | |
1795 | * go through the normal accounting stuff before submission. | |
1796 | */ | |
3ffb52e7 | 1797 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1798 | if (rw & WRITE) { |
1799 | count_vm_events(PGPGOUT, count); | |
1800 | } else { | |
1801 | task_io_account_read(bio->bi_size); | |
1802 | count_vm_events(PGPGIN, count); | |
1803 | } | |
1804 | ||
1805 | if (unlikely(block_dump)) { | |
1806 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1807 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1808 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1809 | (rw & WRITE) ? "WRITE" : "READ", |
1810 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1811 | bdevname(bio->bi_bdev, b), |
1812 | count); | |
bf2de6f5 | 1813 | } |
1da177e4 LT |
1814 | } |
1815 | ||
1816 | generic_make_request(bio); | |
1817 | } | |
1da177e4 LT |
1818 | EXPORT_SYMBOL(submit_bio); |
1819 | ||
82124d60 KU |
1820 | /** |
1821 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1822 | * @q: the queue | |
1823 | * @rq: the request being checked | |
1824 | * | |
1825 | * Description: | |
1826 | * @rq may have been made based on weaker limitations of upper-level queues | |
1827 | * in request stacking drivers, and it may violate the limitation of @q. | |
1828 | * Since the block layer and the underlying device driver trust @rq | |
1829 | * after it is inserted to @q, it should be checked against @q before | |
1830 | * the insertion using this generic function. | |
1831 | * | |
1832 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1833 | * in some cases below, so export this function. |
82124d60 KU |
1834 | * Request stacking drivers like request-based dm may change the queue |
1835 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1836 | * Such request stacking drivers should check those requests agaist | |
1837 | * the new queue limits again when they dispatch those requests, | |
1838 | * although such checkings are also done against the old queue limits | |
1839 | * when submitting requests. | |
1840 | */ | |
1841 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1842 | { | |
3383977f S |
1843 | if (rq->cmd_flags & REQ_DISCARD) |
1844 | return 0; | |
1845 | ||
ae03bf63 MP |
1846 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1847 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1848 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1849 | return -EIO; | |
1850 | } | |
1851 | ||
1852 | /* | |
1853 | * queue's settings related to segment counting like q->bounce_pfn | |
1854 | * may differ from that of other stacking queues. | |
1855 | * Recalculate it to check the request correctly on this queue's | |
1856 | * limitation. | |
1857 | */ | |
1858 | blk_recalc_rq_segments(rq); | |
8a78362c | 1859 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1860 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1861 | return -EIO; | |
1862 | } | |
1863 | ||
1864 | return 0; | |
1865 | } | |
1866 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1867 | ||
1868 | /** | |
1869 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1870 | * @q: the queue to submit the request | |
1871 | * @rq: the request being queued | |
1872 | */ | |
1873 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1874 | { | |
1875 | unsigned long flags; | |
4853abaa | 1876 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
1877 | |
1878 | if (blk_rq_check_limits(q, rq)) | |
1879 | return -EIO; | |
1880 | ||
b2c9cd37 AM |
1881 | if (rq->rq_disk && |
1882 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 1883 | return -EIO; |
82124d60 KU |
1884 | |
1885 | spin_lock_irqsave(q->queue_lock, flags); | |
8ba61435 TH |
1886 | if (unlikely(blk_queue_dead(q))) { |
1887 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1888 | return -ENODEV; | |
1889 | } | |
82124d60 KU |
1890 | |
1891 | /* | |
1892 | * Submitting request must be dequeued before calling this function | |
1893 | * because it will be linked to another request_queue | |
1894 | */ | |
1895 | BUG_ON(blk_queued_rq(rq)); | |
1896 | ||
4853abaa JM |
1897 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
1898 | where = ELEVATOR_INSERT_FLUSH; | |
1899 | ||
1900 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
1901 | if (where == ELEVATOR_INSERT_FLUSH) |
1902 | __blk_run_queue(q); | |
82124d60 KU |
1903 | spin_unlock_irqrestore(q->queue_lock, flags); |
1904 | ||
1905 | return 0; | |
1906 | } | |
1907 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1908 | ||
80a761fd TH |
1909 | /** |
1910 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1911 | * @rq: request to examine | |
1912 | * | |
1913 | * Description: | |
1914 | * A request could be merge of IOs which require different failure | |
1915 | * handling. This function determines the number of bytes which | |
1916 | * can be failed from the beginning of the request without | |
1917 | * crossing into area which need to be retried further. | |
1918 | * | |
1919 | * Return: | |
1920 | * The number of bytes to fail. | |
1921 | * | |
1922 | * Context: | |
1923 | * queue_lock must be held. | |
1924 | */ | |
1925 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1926 | { | |
1927 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1928 | unsigned int bytes = 0; | |
1929 | struct bio *bio; | |
1930 | ||
1931 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1932 | return blk_rq_bytes(rq); | |
1933 | ||
1934 | /* | |
1935 | * Currently the only 'mixing' which can happen is between | |
1936 | * different fastfail types. We can safely fail portions | |
1937 | * which have all the failfast bits that the first one has - | |
1938 | * the ones which are at least as eager to fail as the first | |
1939 | * one. | |
1940 | */ | |
1941 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1942 | if ((bio->bi_rw & ff) != ff) | |
1943 | break; | |
1944 | bytes += bio->bi_size; | |
1945 | } | |
1946 | ||
1947 | /* this could lead to infinite loop */ | |
1948 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1949 | return bytes; | |
1950 | } | |
1951 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1952 | ||
bc58ba94 JA |
1953 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1954 | { | |
c2553b58 | 1955 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1956 | const int rw = rq_data_dir(req); |
1957 | struct hd_struct *part; | |
1958 | int cpu; | |
1959 | ||
1960 | cpu = part_stat_lock(); | |
09e099d4 | 1961 | part = req->part; |
bc58ba94 JA |
1962 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1963 | part_stat_unlock(); | |
1964 | } | |
1965 | } | |
1966 | ||
1967 | static void blk_account_io_done(struct request *req) | |
1968 | { | |
bc58ba94 | 1969 | /* |
dd4c133f TH |
1970 | * Account IO completion. flush_rq isn't accounted as a |
1971 | * normal IO on queueing nor completion. Accounting the | |
1972 | * containing request is enough. | |
bc58ba94 | 1973 | */ |
414b4ff5 | 1974 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1975 | unsigned long duration = jiffies - req->start_time; |
1976 | const int rw = rq_data_dir(req); | |
1977 | struct hd_struct *part; | |
1978 | int cpu; | |
1979 | ||
1980 | cpu = part_stat_lock(); | |
09e099d4 | 1981 | part = req->part; |
bc58ba94 JA |
1982 | |
1983 | part_stat_inc(cpu, part, ios[rw]); | |
1984 | part_stat_add(cpu, part, ticks[rw], duration); | |
1985 | part_round_stats(cpu, part); | |
316d315b | 1986 | part_dec_in_flight(part, rw); |
bc58ba94 | 1987 | |
6c23a968 | 1988 | hd_struct_put(part); |
bc58ba94 JA |
1989 | part_stat_unlock(); |
1990 | } | |
1991 | } | |
1992 | ||
3bcddeac | 1993 | /** |
9934c8c0 TH |
1994 | * blk_peek_request - peek at the top of a request queue |
1995 | * @q: request queue to peek at | |
1996 | * | |
1997 | * Description: | |
1998 | * Return the request at the top of @q. The returned request | |
1999 | * should be started using blk_start_request() before LLD starts | |
2000 | * processing it. | |
2001 | * | |
2002 | * Return: | |
2003 | * Pointer to the request at the top of @q if available. Null | |
2004 | * otherwise. | |
2005 | * | |
2006 | * Context: | |
2007 | * queue_lock must be held. | |
2008 | */ | |
2009 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2010 | { |
2011 | struct request *rq; | |
2012 | int ret; | |
2013 | ||
2014 | while ((rq = __elv_next_request(q)) != NULL) { | |
2015 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
2016 | /* | |
2017 | * This is the first time the device driver | |
2018 | * sees this request (possibly after | |
2019 | * requeueing). Notify IO scheduler. | |
2020 | */ | |
33659ebb | 2021 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
2022 | elv_activate_rq(q, rq); |
2023 | ||
2024 | /* | |
2025 | * just mark as started even if we don't start | |
2026 | * it, a request that has been delayed should | |
2027 | * not be passed by new incoming requests | |
2028 | */ | |
2029 | rq->cmd_flags |= REQ_STARTED; | |
2030 | trace_block_rq_issue(q, rq); | |
2031 | } | |
2032 | ||
2033 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2034 | q->end_sector = rq_end_sector(rq); | |
2035 | q->boundary_rq = NULL; | |
2036 | } | |
2037 | ||
2038 | if (rq->cmd_flags & REQ_DONTPREP) | |
2039 | break; | |
2040 | ||
2e46e8b2 | 2041 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2042 | /* |
2043 | * make sure space for the drain appears we | |
2044 | * know we can do this because max_hw_segments | |
2045 | * has been adjusted to be one fewer than the | |
2046 | * device can handle | |
2047 | */ | |
2048 | rq->nr_phys_segments++; | |
2049 | } | |
2050 | ||
2051 | if (!q->prep_rq_fn) | |
2052 | break; | |
2053 | ||
2054 | ret = q->prep_rq_fn(q, rq); | |
2055 | if (ret == BLKPREP_OK) { | |
2056 | break; | |
2057 | } else if (ret == BLKPREP_DEFER) { | |
2058 | /* | |
2059 | * the request may have been (partially) prepped. | |
2060 | * we need to keep this request in the front to | |
2061 | * avoid resource deadlock. REQ_STARTED will | |
2062 | * prevent other fs requests from passing this one. | |
2063 | */ | |
2e46e8b2 | 2064 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
2065 | !(rq->cmd_flags & REQ_DONTPREP)) { |
2066 | /* | |
2067 | * remove the space for the drain we added | |
2068 | * so that we don't add it again | |
2069 | */ | |
2070 | --rq->nr_phys_segments; | |
2071 | } | |
2072 | ||
2073 | rq = NULL; | |
2074 | break; | |
2075 | } else if (ret == BLKPREP_KILL) { | |
2076 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
2077 | /* |
2078 | * Mark this request as started so we don't trigger | |
2079 | * any debug logic in the end I/O path. | |
2080 | */ | |
2081 | blk_start_request(rq); | |
40cbbb78 | 2082 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
2083 | } else { |
2084 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2085 | break; | |
2086 | } | |
2087 | } | |
2088 | ||
2089 | return rq; | |
2090 | } | |
9934c8c0 | 2091 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2092 | |
9934c8c0 | 2093 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2094 | { |
9934c8c0 TH |
2095 | struct request_queue *q = rq->q; |
2096 | ||
158dbda0 TH |
2097 | BUG_ON(list_empty(&rq->queuelist)); |
2098 | BUG_ON(ELV_ON_HASH(rq)); | |
2099 | ||
2100 | list_del_init(&rq->queuelist); | |
2101 | ||
2102 | /* | |
2103 | * the time frame between a request being removed from the lists | |
2104 | * and to it is freed is accounted as io that is in progress at | |
2105 | * the driver side. | |
2106 | */ | |
9195291e | 2107 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2108 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2109 | set_io_start_time_ns(rq); |
2110 | } | |
158dbda0 TH |
2111 | } |
2112 | ||
9934c8c0 TH |
2113 | /** |
2114 | * blk_start_request - start request processing on the driver | |
2115 | * @req: request to dequeue | |
2116 | * | |
2117 | * Description: | |
2118 | * Dequeue @req and start timeout timer on it. This hands off the | |
2119 | * request to the driver. | |
2120 | * | |
2121 | * Block internal functions which don't want to start timer should | |
2122 | * call blk_dequeue_request(). | |
2123 | * | |
2124 | * Context: | |
2125 | * queue_lock must be held. | |
2126 | */ | |
2127 | void blk_start_request(struct request *req) | |
2128 | { | |
2129 | blk_dequeue_request(req); | |
2130 | ||
2131 | /* | |
5f49f631 TH |
2132 | * We are now handing the request to the hardware, initialize |
2133 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2134 | */ |
5f49f631 | 2135 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2136 | if (unlikely(blk_bidi_rq(req))) |
2137 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2138 | ||
9934c8c0 TH |
2139 | blk_add_timer(req); |
2140 | } | |
2141 | EXPORT_SYMBOL(blk_start_request); | |
2142 | ||
2143 | /** | |
2144 | * blk_fetch_request - fetch a request from a request queue | |
2145 | * @q: request queue to fetch a request from | |
2146 | * | |
2147 | * Description: | |
2148 | * Return the request at the top of @q. The request is started on | |
2149 | * return and LLD can start processing it immediately. | |
2150 | * | |
2151 | * Return: | |
2152 | * Pointer to the request at the top of @q if available. Null | |
2153 | * otherwise. | |
2154 | * | |
2155 | * Context: | |
2156 | * queue_lock must be held. | |
2157 | */ | |
2158 | struct request *blk_fetch_request(struct request_queue *q) | |
2159 | { | |
2160 | struct request *rq; | |
2161 | ||
2162 | rq = blk_peek_request(q); | |
2163 | if (rq) | |
2164 | blk_start_request(rq); | |
2165 | return rq; | |
2166 | } | |
2167 | EXPORT_SYMBOL(blk_fetch_request); | |
2168 | ||
3bcddeac | 2169 | /** |
2e60e022 | 2170 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2171 | * @req: the request being processed |
710027a4 | 2172 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2173 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2174 | * |
2175 | * Description: | |
8ebf9756 RD |
2176 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2177 | * the request structure even if @req doesn't have leftover. | |
2178 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2179 | * |
2180 | * This special helper function is only for request stacking drivers | |
2181 | * (e.g. request-based dm) so that they can handle partial completion. | |
2182 | * Actual device drivers should use blk_end_request instead. | |
2183 | * | |
2184 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2185 | * %false return from this function. | |
3bcddeac KU |
2186 | * |
2187 | * Return: | |
2e60e022 TH |
2188 | * %false - this request doesn't have any more data |
2189 | * %true - this request has more data | |
3bcddeac | 2190 | **/ |
2e60e022 | 2191 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2192 | { |
5450d3e1 | 2193 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2194 | struct bio *bio; |
2195 | ||
2e60e022 TH |
2196 | if (!req->bio) |
2197 | return false; | |
2198 | ||
5f3ea37c | 2199 | trace_block_rq_complete(req->q, req); |
2056a782 | 2200 | |
1da177e4 | 2201 | /* |
6f41469c TH |
2202 | * For fs requests, rq is just carrier of independent bio's |
2203 | * and each partial completion should be handled separately. | |
2204 | * Reset per-request error on each partial completion. | |
2205 | * | |
2206 | * TODO: tj: This is too subtle. It would be better to let | |
2207 | * low level drivers do what they see fit. | |
1da177e4 | 2208 | */ |
33659ebb | 2209 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2210 | req->errors = 0; |
2211 | ||
33659ebb CH |
2212 | if (error && req->cmd_type == REQ_TYPE_FS && |
2213 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2214 | char *error_type; |
2215 | ||
2216 | switch (error) { | |
2217 | case -ENOLINK: | |
2218 | error_type = "recoverable transport"; | |
2219 | break; | |
2220 | case -EREMOTEIO: | |
2221 | error_type = "critical target"; | |
2222 | break; | |
2223 | case -EBADE: | |
2224 | error_type = "critical nexus"; | |
2225 | break; | |
2226 | case -EIO: | |
2227 | default: | |
2228 | error_type = "I/O"; | |
2229 | break; | |
2230 | } | |
2231 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2232 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2233 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2234 | } |
2235 | ||
bc58ba94 | 2236 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2237 | |
1da177e4 LT |
2238 | total_bytes = bio_nbytes = 0; |
2239 | while ((bio = req->bio) != NULL) { | |
2240 | int nbytes; | |
2241 | ||
2242 | if (nr_bytes >= bio->bi_size) { | |
2243 | req->bio = bio->bi_next; | |
2244 | nbytes = bio->bi_size; | |
5bb23a68 | 2245 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2246 | next_idx = 0; |
2247 | bio_nbytes = 0; | |
2248 | } else { | |
2249 | int idx = bio->bi_idx + next_idx; | |
2250 | ||
af498d7f | 2251 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2252 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2253 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2254 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2255 | break; |
2256 | } | |
2257 | ||
2258 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2259 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2260 | ||
2261 | /* | |
2262 | * not a complete bvec done | |
2263 | */ | |
2264 | if (unlikely(nbytes > nr_bytes)) { | |
2265 | bio_nbytes += nr_bytes; | |
2266 | total_bytes += nr_bytes; | |
2267 | break; | |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * advance to the next vector | |
2272 | */ | |
2273 | next_idx++; | |
2274 | bio_nbytes += nbytes; | |
2275 | } | |
2276 | ||
2277 | total_bytes += nbytes; | |
2278 | nr_bytes -= nbytes; | |
2279 | ||
6728cb0e JA |
2280 | bio = req->bio; |
2281 | if (bio) { | |
1da177e4 LT |
2282 | /* |
2283 | * end more in this run, or just return 'not-done' | |
2284 | */ | |
2285 | if (unlikely(nr_bytes <= 0)) | |
2286 | break; | |
2287 | } | |
2288 | } | |
2289 | ||
2290 | /* | |
2291 | * completely done | |
2292 | */ | |
2e60e022 TH |
2293 | if (!req->bio) { |
2294 | /* | |
2295 | * Reset counters so that the request stacking driver | |
2296 | * can find how many bytes remain in the request | |
2297 | * later. | |
2298 | */ | |
a2dec7b3 | 2299 | req->__data_len = 0; |
2e60e022 TH |
2300 | return false; |
2301 | } | |
1da177e4 LT |
2302 | |
2303 | /* | |
2304 | * if the request wasn't completed, update state | |
2305 | */ | |
2306 | if (bio_nbytes) { | |
5bb23a68 | 2307 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2308 | bio->bi_idx += next_idx; |
2309 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2310 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2311 | } | |
2312 | ||
a2dec7b3 | 2313 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2314 | req->buffer = bio_data(req->bio); |
2315 | ||
2316 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2317 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2318 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2319 | |
80a761fd TH |
2320 | /* mixed attributes always follow the first bio */ |
2321 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2322 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2323 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2324 | } | |
2325 | ||
2e46e8b2 TH |
2326 | /* |
2327 | * If total number of sectors is less than the first segment | |
2328 | * size, something has gone terribly wrong. | |
2329 | */ | |
2330 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2331 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2332 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2333 | } |
2334 | ||
2335 | /* recalculate the number of segments */ | |
1da177e4 | 2336 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2337 | |
2e60e022 | 2338 | return true; |
1da177e4 | 2339 | } |
2e60e022 | 2340 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2341 | |
2e60e022 TH |
2342 | static bool blk_update_bidi_request(struct request *rq, int error, |
2343 | unsigned int nr_bytes, | |
2344 | unsigned int bidi_bytes) | |
5efccd17 | 2345 | { |
2e60e022 TH |
2346 | if (blk_update_request(rq, error, nr_bytes)) |
2347 | return true; | |
5efccd17 | 2348 | |
2e60e022 TH |
2349 | /* Bidi request must be completed as a whole */ |
2350 | if (unlikely(blk_bidi_rq(rq)) && | |
2351 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2352 | return true; | |
5efccd17 | 2353 | |
e2e1a148 JA |
2354 | if (blk_queue_add_random(rq->q)) |
2355 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2356 | |
2357 | return false; | |
1da177e4 LT |
2358 | } |
2359 | ||
28018c24 JB |
2360 | /** |
2361 | * blk_unprep_request - unprepare a request | |
2362 | * @req: the request | |
2363 | * | |
2364 | * This function makes a request ready for complete resubmission (or | |
2365 | * completion). It happens only after all error handling is complete, | |
2366 | * so represents the appropriate moment to deallocate any resources | |
2367 | * that were allocated to the request in the prep_rq_fn. The queue | |
2368 | * lock is held when calling this. | |
2369 | */ | |
2370 | void blk_unprep_request(struct request *req) | |
2371 | { | |
2372 | struct request_queue *q = req->q; | |
2373 | ||
2374 | req->cmd_flags &= ~REQ_DONTPREP; | |
2375 | if (q->unprep_rq_fn) | |
2376 | q->unprep_rq_fn(q, req); | |
2377 | } | |
2378 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2379 | ||
1da177e4 LT |
2380 | /* |
2381 | * queue lock must be held | |
2382 | */ | |
2e60e022 | 2383 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2384 | { |
b8286239 KU |
2385 | if (blk_rq_tagged(req)) |
2386 | blk_queue_end_tag(req->q, req); | |
2387 | ||
ba396a6c | 2388 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2389 | |
33659ebb | 2390 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2391 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2392 | |
e78042e5 MA |
2393 | blk_delete_timer(req); |
2394 | ||
28018c24 JB |
2395 | if (req->cmd_flags & REQ_DONTPREP) |
2396 | blk_unprep_request(req); | |
2397 | ||
2398 | ||
bc58ba94 | 2399 | blk_account_io_done(req); |
b8286239 | 2400 | |
1da177e4 | 2401 | if (req->end_io) |
8ffdc655 | 2402 | req->end_io(req, error); |
b8286239 KU |
2403 | else { |
2404 | if (blk_bidi_rq(req)) | |
2405 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2406 | ||
1da177e4 | 2407 | __blk_put_request(req->q, req); |
b8286239 | 2408 | } |
1da177e4 LT |
2409 | } |
2410 | ||
3b11313a | 2411 | /** |
2e60e022 TH |
2412 | * blk_end_bidi_request - Complete a bidi request |
2413 | * @rq: the request to complete | |
2414 | * @error: %0 for success, < %0 for error | |
2415 | * @nr_bytes: number of bytes to complete @rq | |
2416 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2417 | * |
2418 | * Description: | |
e3a04fe3 | 2419 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2420 | * Drivers that supports bidi can safely call this member for any |
2421 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2422 | * just ignored. | |
336cdb40 KU |
2423 | * |
2424 | * Return: | |
2e60e022 TH |
2425 | * %false - we are done with this request |
2426 | * %true - still buffers pending for this request | |
a0cd1285 | 2427 | **/ |
b1f74493 | 2428 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2429 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2430 | { | |
336cdb40 | 2431 | struct request_queue *q = rq->q; |
2e60e022 | 2432 | unsigned long flags; |
32fab448 | 2433 | |
2e60e022 TH |
2434 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2435 | return true; | |
32fab448 | 2436 | |
336cdb40 | 2437 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2438 | blk_finish_request(rq, error); |
336cdb40 KU |
2439 | spin_unlock_irqrestore(q->queue_lock, flags); |
2440 | ||
2e60e022 | 2441 | return false; |
32fab448 KU |
2442 | } |
2443 | ||
336cdb40 | 2444 | /** |
2e60e022 TH |
2445 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2446 | * @rq: the request to complete | |
710027a4 | 2447 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2448 | * @nr_bytes: number of bytes to complete @rq |
2449 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2450 | * |
2451 | * Description: | |
2e60e022 TH |
2452 | * Identical to blk_end_bidi_request() except that queue lock is |
2453 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2454 | * |
2455 | * Return: | |
2e60e022 TH |
2456 | * %false - we are done with this request |
2457 | * %true - still buffers pending for this request | |
336cdb40 | 2458 | **/ |
4853abaa | 2459 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2460 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2461 | { |
2e60e022 TH |
2462 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2463 | return true; | |
336cdb40 | 2464 | |
2e60e022 | 2465 | blk_finish_request(rq, error); |
336cdb40 | 2466 | |
2e60e022 | 2467 | return false; |
336cdb40 | 2468 | } |
e19a3ab0 KU |
2469 | |
2470 | /** | |
2471 | * blk_end_request - Helper function for drivers to complete the request. | |
2472 | * @rq: the request being processed | |
710027a4 | 2473 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2474 | * @nr_bytes: number of bytes to complete |
2475 | * | |
2476 | * Description: | |
2477 | * Ends I/O on a number of bytes attached to @rq. | |
2478 | * If @rq has leftover, sets it up for the next range of segments. | |
2479 | * | |
2480 | * Return: | |
b1f74493 FT |
2481 | * %false - we are done with this request |
2482 | * %true - still buffers pending for this request | |
e19a3ab0 | 2483 | **/ |
b1f74493 | 2484 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2485 | { |
b1f74493 | 2486 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2487 | } |
56ad1740 | 2488 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2489 | |
2490 | /** | |
b1f74493 FT |
2491 | * blk_end_request_all - Helper function for drives to finish the request. |
2492 | * @rq: the request to finish | |
8ebf9756 | 2493 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2494 | * |
2495 | * Description: | |
b1f74493 FT |
2496 | * Completely finish @rq. |
2497 | */ | |
2498 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2499 | { |
b1f74493 FT |
2500 | bool pending; |
2501 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2502 | |
b1f74493 FT |
2503 | if (unlikely(blk_bidi_rq(rq))) |
2504 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2505 | |
b1f74493 FT |
2506 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2507 | BUG_ON(pending); | |
2508 | } | |
56ad1740 | 2509 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2510 | |
b1f74493 FT |
2511 | /** |
2512 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2513 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2514 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2515 | * |
2516 | * Description: | |
2517 | * Complete the current consecutively mapped chunk from @rq. | |
2518 | * | |
2519 | * Return: | |
2520 | * %false - we are done with this request | |
2521 | * %true - still buffers pending for this request | |
2522 | */ | |
2523 | bool blk_end_request_cur(struct request *rq, int error) | |
2524 | { | |
2525 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2526 | } |
56ad1740 | 2527 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2528 | |
80a761fd TH |
2529 | /** |
2530 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2531 | * @rq: the request to finish till the next failure boundary for | |
2532 | * @error: must be negative errno | |
2533 | * | |
2534 | * Description: | |
2535 | * Complete @rq till the next failure boundary. | |
2536 | * | |
2537 | * Return: | |
2538 | * %false - we are done with this request | |
2539 | * %true - still buffers pending for this request | |
2540 | */ | |
2541 | bool blk_end_request_err(struct request *rq, int error) | |
2542 | { | |
2543 | WARN_ON(error >= 0); | |
2544 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2545 | } | |
2546 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2547 | ||
e3a04fe3 | 2548 | /** |
b1f74493 FT |
2549 | * __blk_end_request - Helper function for drivers to complete the request. |
2550 | * @rq: the request being processed | |
2551 | * @error: %0 for success, < %0 for error | |
2552 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2553 | * |
2554 | * Description: | |
b1f74493 | 2555 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2556 | * |
2557 | * Return: | |
b1f74493 FT |
2558 | * %false - we are done with this request |
2559 | * %true - still buffers pending for this request | |
e3a04fe3 | 2560 | **/ |
b1f74493 | 2561 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2562 | { |
b1f74493 | 2563 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2564 | } |
56ad1740 | 2565 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2566 | |
32fab448 | 2567 | /** |
b1f74493 FT |
2568 | * __blk_end_request_all - Helper function for drives to finish the request. |
2569 | * @rq: the request to finish | |
8ebf9756 | 2570 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2571 | * |
2572 | * Description: | |
b1f74493 | 2573 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2574 | */ |
b1f74493 | 2575 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2576 | { |
b1f74493 FT |
2577 | bool pending; |
2578 | unsigned int bidi_bytes = 0; | |
2579 | ||
2580 | if (unlikely(blk_bidi_rq(rq))) | |
2581 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2582 | ||
2583 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2584 | BUG_ON(pending); | |
32fab448 | 2585 | } |
56ad1740 | 2586 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2587 | |
e19a3ab0 | 2588 | /** |
b1f74493 FT |
2589 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2590 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2591 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2592 | * |
2593 | * Description: | |
b1f74493 FT |
2594 | * Complete the current consecutively mapped chunk from @rq. Must |
2595 | * be called with queue lock held. | |
e19a3ab0 KU |
2596 | * |
2597 | * Return: | |
b1f74493 FT |
2598 | * %false - we are done with this request |
2599 | * %true - still buffers pending for this request | |
2600 | */ | |
2601 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2602 | { |
b1f74493 | 2603 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2604 | } |
56ad1740 | 2605 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2606 | |
80a761fd TH |
2607 | /** |
2608 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2609 | * @rq: the request to finish till the next failure boundary for | |
2610 | * @error: must be negative errno | |
2611 | * | |
2612 | * Description: | |
2613 | * Complete @rq till the next failure boundary. Must be called | |
2614 | * with queue lock held. | |
2615 | * | |
2616 | * Return: | |
2617 | * %false - we are done with this request | |
2618 | * %true - still buffers pending for this request | |
2619 | */ | |
2620 | bool __blk_end_request_err(struct request *rq, int error) | |
2621 | { | |
2622 | WARN_ON(error >= 0); | |
2623 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2624 | } | |
2625 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2626 | ||
86db1e29 JA |
2627 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2628 | struct bio *bio) | |
1da177e4 | 2629 | { |
a82afdfc | 2630 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2631 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2632 | |
fb2dce86 DW |
2633 | if (bio_has_data(bio)) { |
2634 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2635 | rq->buffer = bio_data(bio); |
2636 | } | |
a2dec7b3 | 2637 | rq->__data_len = bio->bi_size; |
1da177e4 | 2638 | rq->bio = rq->biotail = bio; |
1da177e4 | 2639 | |
66846572 N |
2640 | if (bio->bi_bdev) |
2641 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2642 | } | |
1da177e4 | 2643 | |
2d4dc890 IL |
2644 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2645 | /** | |
2646 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2647 | * @rq: the request to be flushed | |
2648 | * | |
2649 | * Description: | |
2650 | * Flush all pages in @rq. | |
2651 | */ | |
2652 | void rq_flush_dcache_pages(struct request *rq) | |
2653 | { | |
2654 | struct req_iterator iter; | |
2655 | struct bio_vec *bvec; | |
2656 | ||
2657 | rq_for_each_segment(bvec, rq, iter) | |
2658 | flush_dcache_page(bvec->bv_page); | |
2659 | } | |
2660 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2661 | #endif | |
2662 | ||
ef9e3fac KU |
2663 | /** |
2664 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2665 | * @q : the queue of the device being checked | |
2666 | * | |
2667 | * Description: | |
2668 | * Check if underlying low-level drivers of a device are busy. | |
2669 | * If the drivers want to export their busy state, they must set own | |
2670 | * exporting function using blk_queue_lld_busy() first. | |
2671 | * | |
2672 | * Basically, this function is used only by request stacking drivers | |
2673 | * to stop dispatching requests to underlying devices when underlying | |
2674 | * devices are busy. This behavior helps more I/O merging on the queue | |
2675 | * of the request stacking driver and prevents I/O throughput regression | |
2676 | * on burst I/O load. | |
2677 | * | |
2678 | * Return: | |
2679 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2680 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2681 | */ | |
2682 | int blk_lld_busy(struct request_queue *q) | |
2683 | { | |
2684 | if (q->lld_busy_fn) | |
2685 | return q->lld_busy_fn(q); | |
2686 | ||
2687 | return 0; | |
2688 | } | |
2689 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2690 | ||
b0fd271d KU |
2691 | /** |
2692 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2693 | * @rq: the clone request to be cleaned up | |
2694 | * | |
2695 | * Description: | |
2696 | * Free all bios in @rq for a cloned request. | |
2697 | */ | |
2698 | void blk_rq_unprep_clone(struct request *rq) | |
2699 | { | |
2700 | struct bio *bio; | |
2701 | ||
2702 | while ((bio = rq->bio) != NULL) { | |
2703 | rq->bio = bio->bi_next; | |
2704 | ||
2705 | bio_put(bio); | |
2706 | } | |
2707 | } | |
2708 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2709 | ||
2710 | /* | |
2711 | * Copy attributes of the original request to the clone request. | |
2712 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2713 | */ | |
2714 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2715 | { | |
2716 | dst->cpu = src->cpu; | |
3a2edd0d | 2717 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2718 | dst->cmd_type = src->cmd_type; |
2719 | dst->__sector = blk_rq_pos(src); | |
2720 | dst->__data_len = blk_rq_bytes(src); | |
2721 | dst->nr_phys_segments = src->nr_phys_segments; | |
2722 | dst->ioprio = src->ioprio; | |
2723 | dst->extra_len = src->extra_len; | |
2724 | } | |
2725 | ||
2726 | /** | |
2727 | * blk_rq_prep_clone - Helper function to setup clone request | |
2728 | * @rq: the request to be setup | |
2729 | * @rq_src: original request to be cloned | |
2730 | * @bs: bio_set that bios for clone are allocated from | |
2731 | * @gfp_mask: memory allocation mask for bio | |
2732 | * @bio_ctr: setup function to be called for each clone bio. | |
2733 | * Returns %0 for success, non %0 for failure. | |
2734 | * @data: private data to be passed to @bio_ctr | |
2735 | * | |
2736 | * Description: | |
2737 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2738 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2739 | * are not copied, and copying such parts is the caller's responsibility. | |
2740 | * Also, pages which the original bios are pointing to are not copied | |
2741 | * and the cloned bios just point same pages. | |
2742 | * So cloned bios must be completed before original bios, which means | |
2743 | * the caller must complete @rq before @rq_src. | |
2744 | */ | |
2745 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2746 | struct bio_set *bs, gfp_t gfp_mask, | |
2747 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2748 | void *data) | |
2749 | { | |
2750 | struct bio *bio, *bio_src; | |
2751 | ||
2752 | if (!bs) | |
2753 | bs = fs_bio_set; | |
2754 | ||
2755 | blk_rq_init(NULL, rq); | |
2756 | ||
2757 | __rq_for_each_bio(bio_src, rq_src) { | |
2758 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2759 | if (!bio) | |
2760 | goto free_and_out; | |
2761 | ||
2762 | __bio_clone(bio, bio_src); | |
2763 | ||
2764 | if (bio_integrity(bio_src) && | |
7878cba9 | 2765 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2766 | goto free_and_out; |
2767 | ||
2768 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2769 | goto free_and_out; | |
2770 | ||
2771 | if (rq->bio) { | |
2772 | rq->biotail->bi_next = bio; | |
2773 | rq->biotail = bio; | |
2774 | } else | |
2775 | rq->bio = rq->biotail = bio; | |
2776 | } | |
2777 | ||
2778 | __blk_rq_prep_clone(rq, rq_src); | |
2779 | ||
2780 | return 0; | |
2781 | ||
2782 | free_and_out: | |
2783 | if (bio) | |
2784 | bio_free(bio, bs); | |
2785 | blk_rq_unprep_clone(rq); | |
2786 | ||
2787 | return -ENOMEM; | |
2788 | } | |
2789 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2790 | ||
18887ad9 | 2791 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2792 | { |
2793 | return queue_work(kblockd_workqueue, work); | |
2794 | } | |
1da177e4 LT |
2795 | EXPORT_SYMBOL(kblockd_schedule_work); |
2796 | ||
e43473b7 VG |
2797 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2798 | struct delayed_work *dwork, unsigned long delay) | |
2799 | { | |
2800 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2801 | } | |
2802 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2803 | ||
73c10101 JA |
2804 | #define PLUG_MAGIC 0x91827364 |
2805 | ||
75df7136 SJ |
2806 | /** |
2807 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2808 | * @plug: The &struct blk_plug that needs to be initialized | |
2809 | * | |
2810 | * Description: | |
2811 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2812 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2813 | * blk_finish_plug(). This is important from a performance perspective, but | |
2814 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2815 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2816 | * page belonging to that request that is currently residing in our private | |
2817 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2818 | * this kind of deadlock. | |
2819 | */ | |
73c10101 JA |
2820 | void blk_start_plug(struct blk_plug *plug) |
2821 | { | |
2822 | struct task_struct *tsk = current; | |
2823 | ||
2824 | plug->magic = PLUG_MAGIC; | |
2825 | INIT_LIST_HEAD(&plug->list); | |
048c9374 | 2826 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 JA |
2827 | plug->should_sort = 0; |
2828 | ||
2829 | /* | |
2830 | * If this is a nested plug, don't actually assign it. It will be | |
2831 | * flushed on its own. | |
2832 | */ | |
2833 | if (!tsk->plug) { | |
2834 | /* | |
2835 | * Store ordering should not be needed here, since a potential | |
2836 | * preempt will imply a full memory barrier | |
2837 | */ | |
2838 | tsk->plug = plug; | |
2839 | } | |
2840 | } | |
2841 | EXPORT_SYMBOL(blk_start_plug); | |
2842 | ||
2843 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2844 | { | |
2845 | struct request *rqa = container_of(a, struct request, queuelist); | |
2846 | struct request *rqb = container_of(b, struct request, queuelist); | |
2847 | ||
f83e8261 | 2848 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2849 | } |
2850 | ||
49cac01e JA |
2851 | /* |
2852 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2853 | * until a safe kblockd context. We due this to avoid accidental big | |
2854 | * additional stack usage in driver dispatch, in places where the originally | |
2855 | * plugger did not intend it. | |
2856 | */ | |
f6603783 | 2857 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 2858 | bool from_schedule) |
99e22598 | 2859 | __releases(q->queue_lock) |
94b5eb28 | 2860 | { |
49cac01e | 2861 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 2862 | |
8ba61435 TH |
2863 | /* |
2864 | * Don't mess with dead queue. | |
2865 | */ | |
2866 | if (unlikely(blk_queue_dead(q))) { | |
2867 | spin_unlock(q->queue_lock); | |
2868 | return; | |
2869 | } | |
2870 | ||
99e22598 JA |
2871 | /* |
2872 | * If we are punting this to kblockd, then we can safely drop | |
2873 | * the queue_lock before waking kblockd (which needs to take | |
2874 | * this lock). | |
2875 | */ | |
2876 | if (from_schedule) { | |
2877 | spin_unlock(q->queue_lock); | |
24ecfbe2 | 2878 | blk_run_queue_async(q); |
99e22598 | 2879 | } else { |
24ecfbe2 | 2880 | __blk_run_queue(q); |
99e22598 JA |
2881 | spin_unlock(q->queue_lock); |
2882 | } | |
2883 | ||
94b5eb28 JA |
2884 | } |
2885 | ||
048c9374 N |
2886 | static void flush_plug_callbacks(struct blk_plug *plug) |
2887 | { | |
2888 | LIST_HEAD(callbacks); | |
2889 | ||
2890 | if (list_empty(&plug->cb_list)) | |
2891 | return; | |
2892 | ||
2893 | list_splice_init(&plug->cb_list, &callbacks); | |
2894 | ||
2895 | while (!list_empty(&callbacks)) { | |
2896 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
2897 | struct blk_plug_cb, | |
2898 | list); | |
2899 | list_del(&cb->list); | |
2900 | cb->callback(cb); | |
2901 | } | |
2902 | } | |
2903 | ||
49cac01e | 2904 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
2905 | { |
2906 | struct request_queue *q; | |
2907 | unsigned long flags; | |
2908 | struct request *rq; | |
109b8129 | 2909 | LIST_HEAD(list); |
94b5eb28 | 2910 | unsigned int depth; |
73c10101 JA |
2911 | |
2912 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2913 | ||
048c9374 | 2914 | flush_plug_callbacks(plug); |
73c10101 JA |
2915 | if (list_empty(&plug->list)) |
2916 | return; | |
2917 | ||
109b8129 N |
2918 | list_splice_init(&plug->list, &list); |
2919 | ||
2920 | if (plug->should_sort) { | |
2921 | list_sort(NULL, &list, plug_rq_cmp); | |
2922 | plug->should_sort = 0; | |
2923 | } | |
73c10101 JA |
2924 | |
2925 | q = NULL; | |
94b5eb28 | 2926 | depth = 0; |
18811272 JA |
2927 | |
2928 | /* | |
2929 | * Save and disable interrupts here, to avoid doing it for every | |
2930 | * queue lock we have to take. | |
2931 | */ | |
73c10101 | 2932 | local_irq_save(flags); |
109b8129 N |
2933 | while (!list_empty(&list)) { |
2934 | rq = list_entry_rq(list.next); | |
73c10101 | 2935 | list_del_init(&rq->queuelist); |
73c10101 JA |
2936 | BUG_ON(!rq->q); |
2937 | if (rq->q != q) { | |
99e22598 JA |
2938 | /* |
2939 | * This drops the queue lock | |
2940 | */ | |
2941 | if (q) | |
49cac01e | 2942 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2943 | q = rq->q; |
94b5eb28 | 2944 | depth = 0; |
73c10101 JA |
2945 | spin_lock(q->queue_lock); |
2946 | } | |
8ba61435 TH |
2947 | |
2948 | /* | |
2949 | * Short-circuit if @q is dead | |
2950 | */ | |
2951 | if (unlikely(blk_queue_dead(q))) { | |
2952 | __blk_end_request_all(rq, -ENODEV); | |
2953 | continue; | |
2954 | } | |
2955 | ||
73c10101 JA |
2956 | /* |
2957 | * rq is already accounted, so use raw insert | |
2958 | */ | |
401a18e9 JA |
2959 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
2960 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
2961 | else | |
2962 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
2963 | |
2964 | depth++; | |
73c10101 JA |
2965 | } |
2966 | ||
99e22598 JA |
2967 | /* |
2968 | * This drops the queue lock | |
2969 | */ | |
2970 | if (q) | |
49cac01e | 2971 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2972 | |
73c10101 JA |
2973 | local_irq_restore(flags); |
2974 | } | |
73c10101 JA |
2975 | |
2976 | void blk_finish_plug(struct blk_plug *plug) | |
2977 | { | |
f6603783 | 2978 | blk_flush_plug_list(plug, false); |
73c10101 | 2979 | |
88b996cd CH |
2980 | if (plug == current->plug) |
2981 | current->plug = NULL; | |
73c10101 | 2982 | } |
88b996cd | 2983 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 2984 | |
1da177e4 LT |
2985 | int __init blk_dev_init(void) |
2986 | { | |
9eb55b03 NK |
2987 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2988 | sizeof(((struct request *)0)->cmd_flags)); | |
2989 | ||
89b90be2 TH |
2990 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2991 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2992 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2993 | if (!kblockd_workqueue) |
2994 | panic("Failed to create kblockd\n"); | |
2995 | ||
2996 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2997 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2998 | |
8324aa91 | 2999 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 3000 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3001 | |
d38ecf93 | 3002 | return 0; |
1da177e4 | 3003 | } |