]>
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> |
55782138 LZ |
32 | |
33 | #define CREATE_TRACE_POINTS | |
34 | #include <trace/events/block.h> | |
1da177e4 | 35 | |
8324aa91 | 36 | #include "blk.h" |
5efd6113 | 37 | #include "blk-cgroup.h" |
8324aa91 | 38 | |
d07335e5 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 41 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 42 | |
a73f730d TH |
43 | DEFINE_IDA(blk_queue_ida); |
44 | ||
1da177e4 LT |
45 | /* |
46 | * For the allocated request tables | |
47 | */ | |
5ece6c52 | 48 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
49 | |
50 | /* | |
51 | * For queue allocation | |
52 | */ | |
6728cb0e | 53 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Controlling structure to kblockd | |
57 | */ | |
ff856bad | 58 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 59 | |
26b8256e JA |
60 | static void drive_stat_acct(struct request *rq, int new_io) |
61 | { | |
28f13702 | 62 | struct hd_struct *part; |
26b8256e | 63 | int rw = rq_data_dir(rq); |
c9959059 | 64 | int cpu; |
26b8256e | 65 | |
c2553b58 | 66 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
67 | return; |
68 | ||
074a7aca | 69 | cpu = part_stat_lock(); |
c9959059 | 70 | |
09e099d4 JM |
71 | if (!new_io) { |
72 | part = rq->part; | |
074a7aca | 73 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
74 | } else { |
75 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 76 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
77 | /* |
78 | * The partition is already being removed, | |
79 | * the request will be accounted on the disk only | |
80 | * | |
81 | * We take a reference on disk->part0 although that | |
82 | * partition will never be deleted, so we can treat | |
83 | * it as any other partition. | |
84 | */ | |
85 | part = &rq->rq_disk->part0; | |
6c23a968 | 86 | hd_struct_get(part); |
09e099d4 | 87 | } |
074a7aca | 88 | part_round_stats(cpu, part); |
316d315b | 89 | part_inc_in_flight(part, rw); |
09e099d4 | 90 | rq->part = part; |
26b8256e | 91 | } |
e71bf0d0 | 92 | |
074a7aca | 93 | part_stat_unlock(); |
26b8256e JA |
94 | } |
95 | ||
8324aa91 | 96 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
97 | { |
98 | int nr; | |
99 | ||
100 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
101 | if (nr > q->nr_requests) | |
102 | nr = q->nr_requests; | |
103 | q->nr_congestion_on = nr; | |
104 | ||
105 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
106 | if (nr < 1) | |
107 | nr = 1; | |
108 | q->nr_congestion_off = nr; | |
109 | } | |
110 | ||
1da177e4 LT |
111 | /** |
112 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
113 | * @bdev: device | |
114 | * | |
115 | * Locates the passed device's request queue and returns the address of its | |
116 | * backing_dev_info | |
117 | * | |
118 | * Will return NULL if the request queue cannot be located. | |
119 | */ | |
120 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
121 | { | |
122 | struct backing_dev_info *ret = NULL; | |
165125e1 | 123 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
124 | |
125 | if (q) | |
126 | ret = &q->backing_dev_info; | |
127 | return ret; | |
128 | } | |
1da177e4 LT |
129 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
130 | ||
2a4aa30c | 131 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 132 | { |
1afb20f3 FT |
133 | memset(rq, 0, sizeof(*rq)); |
134 | ||
1da177e4 | 135 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 136 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 137 | rq->cpu = -1; |
63a71386 | 138 | rq->q = q; |
a2dec7b3 | 139 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
140 | INIT_HLIST_NODE(&rq->hash); |
141 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 142 | rq->cmd = rq->__cmd; |
e2494e1b | 143 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 144 | rq->tag = -1; |
1da177e4 | 145 | rq->ref_count = 1; |
b243ddcb | 146 | rq->start_time = jiffies; |
9195291e | 147 | set_start_time_ns(rq); |
09e099d4 | 148 | rq->part = NULL; |
1da177e4 | 149 | } |
2a4aa30c | 150 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 151 | |
5bb23a68 N |
152 | static void req_bio_endio(struct request *rq, struct bio *bio, |
153 | unsigned int nbytes, int error) | |
1da177e4 | 154 | { |
143a87f4 TH |
155 | if (error) |
156 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
157 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
158 | error = -EIO; | |
797e7dbb | 159 | |
143a87f4 TH |
160 | if (unlikely(nbytes > bio->bi_size)) { |
161 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
162 | __func__, nbytes, bio->bi_size); | |
163 | nbytes = bio->bi_size; | |
5bb23a68 | 164 | } |
797e7dbb | 165 | |
143a87f4 TH |
166 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
167 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 168 | |
143a87f4 TH |
169 | bio->bi_size -= nbytes; |
170 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 171 | |
143a87f4 TH |
172 | if (bio_integrity(bio)) |
173 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 174 | |
143a87f4 TH |
175 | /* don't actually finish bio if it's part of flush sequence */ |
176 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
177 | bio_endio(bio, error); | |
1da177e4 | 178 | } |
1da177e4 | 179 | |
1da177e4 LT |
180 | void blk_dump_rq_flags(struct request *rq, char *msg) |
181 | { | |
182 | int bit; | |
183 | ||
6728cb0e | 184 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
185 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
186 | rq->cmd_flags); | |
1da177e4 | 187 | |
83096ebf TH |
188 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
189 | (unsigned long long)blk_rq_pos(rq), | |
190 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 191 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 192 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 193 | |
33659ebb | 194 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 195 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 196 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
197 | printk("%02x ", rq->cmd[bit]); |
198 | printk("\n"); | |
199 | } | |
200 | } | |
1da177e4 LT |
201 | EXPORT_SYMBOL(blk_dump_rq_flags); |
202 | ||
3cca6dc1 | 203 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 204 | { |
3cca6dc1 | 205 | struct request_queue *q; |
1da177e4 | 206 | |
3cca6dc1 JA |
207 | q = container_of(work, struct request_queue, delay_work.work); |
208 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 209 | __blk_run_queue(q); |
3cca6dc1 | 210 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 211 | } |
1da177e4 LT |
212 | |
213 | /** | |
3cca6dc1 JA |
214 | * blk_delay_queue - restart queueing after defined interval |
215 | * @q: The &struct request_queue in question | |
216 | * @msecs: Delay in msecs | |
1da177e4 LT |
217 | * |
218 | * Description: | |
3cca6dc1 JA |
219 | * Sometimes queueing needs to be postponed for a little while, to allow |
220 | * resources to come back. This function will make sure that queueing is | |
221 | * restarted around the specified time. | |
222 | */ | |
223 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 224 | { |
4521cc4e JA |
225 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
226 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 227 | } |
3cca6dc1 | 228 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 229 | |
1da177e4 LT |
230 | /** |
231 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 232 | * @q: The &struct request_queue in question |
1da177e4 LT |
233 | * |
234 | * Description: | |
235 | * blk_start_queue() will clear the stop flag on the queue, and call | |
236 | * the request_fn for the queue if it was in a stopped state when | |
237 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
238 | **/ | |
165125e1 | 239 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 240 | { |
a038e253 PBG |
241 | WARN_ON(!irqs_disabled()); |
242 | ||
75ad23bc | 243 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 244 | __blk_run_queue(q); |
1da177e4 | 245 | } |
1da177e4 LT |
246 | EXPORT_SYMBOL(blk_start_queue); |
247 | ||
248 | /** | |
249 | * blk_stop_queue - stop a queue | |
165125e1 | 250 | * @q: The &struct request_queue in question |
1da177e4 LT |
251 | * |
252 | * Description: | |
253 | * The Linux block layer assumes that a block driver will consume all | |
254 | * entries on the request queue when the request_fn strategy is called. | |
255 | * Often this will not happen, because of hardware limitations (queue | |
256 | * depth settings). If a device driver gets a 'queue full' response, | |
257 | * or if it simply chooses not to queue more I/O at one point, it can | |
258 | * call this function to prevent the request_fn from being called until | |
259 | * the driver has signalled it's ready to go again. This happens by calling | |
260 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
261 | **/ | |
165125e1 | 262 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 263 | { |
ad3d9d7e | 264 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 265 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
266 | } |
267 | EXPORT_SYMBOL(blk_stop_queue); | |
268 | ||
269 | /** | |
270 | * blk_sync_queue - cancel any pending callbacks on a queue | |
271 | * @q: the queue | |
272 | * | |
273 | * Description: | |
274 | * The block layer may perform asynchronous callback activity | |
275 | * on a queue, such as calling the unplug function after a timeout. | |
276 | * A block device may call blk_sync_queue to ensure that any | |
277 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 278 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
279 | * that its ->make_request_fn will not re-add plugging prior to calling |
280 | * this function. | |
281 | * | |
da527770 VG |
282 | * This function does not cancel any asynchronous activity arising |
283 | * out of elevator or throttling code. That would require elevaotor_exit() | |
5efd6113 | 284 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 285 | * |
1da177e4 LT |
286 | */ |
287 | void blk_sync_queue(struct request_queue *q) | |
288 | { | |
70ed28b9 | 289 | del_timer_sync(&q->timeout); |
3cca6dc1 | 290 | cancel_delayed_work_sync(&q->delay_work); |
1da177e4 LT |
291 | } |
292 | EXPORT_SYMBOL(blk_sync_queue); | |
293 | ||
294 | /** | |
80a4b58e | 295 | * __blk_run_queue - run a single device queue |
1da177e4 | 296 | * @q: The queue to run |
80a4b58e JA |
297 | * |
298 | * Description: | |
299 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 300 | * held and interrupts disabled. |
1da177e4 | 301 | */ |
24ecfbe2 | 302 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 303 | { |
a538cd03 TH |
304 | if (unlikely(blk_queue_stopped(q))) |
305 | return; | |
306 | ||
c21e6beb | 307 | q->request_fn(q); |
75ad23bc NP |
308 | } |
309 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 310 | |
24ecfbe2 CH |
311 | /** |
312 | * blk_run_queue_async - run a single device queue in workqueue context | |
313 | * @q: The queue to run | |
314 | * | |
315 | * Description: | |
316 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
317 | * of us. | |
318 | */ | |
319 | void blk_run_queue_async(struct request_queue *q) | |
320 | { | |
3ec717b7 SL |
321 | if (likely(!blk_queue_stopped(q))) { |
322 | __cancel_delayed_work(&q->delay_work); | |
24ecfbe2 | 323 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
3ec717b7 | 324 | } |
24ecfbe2 | 325 | } |
c21e6beb | 326 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 327 | |
75ad23bc NP |
328 | /** |
329 | * blk_run_queue - run a single device queue | |
330 | * @q: The queue to run | |
80a4b58e JA |
331 | * |
332 | * Description: | |
333 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 334 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
335 | */ |
336 | void blk_run_queue(struct request_queue *q) | |
337 | { | |
338 | unsigned long flags; | |
339 | ||
340 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 341 | __blk_run_queue(q); |
1da177e4 LT |
342 | spin_unlock_irqrestore(q->queue_lock, flags); |
343 | } | |
344 | EXPORT_SYMBOL(blk_run_queue); | |
345 | ||
165125e1 | 346 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
347 | { |
348 | kobject_put(&q->kobj); | |
349 | } | |
d86e0e83 | 350 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 351 | |
e3c78ca5 TH |
352 | /** |
353 | * blk_drain_queue - drain requests from request_queue | |
354 | * @q: queue to drain | |
c9a929dd | 355 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 356 | * |
c9a929dd TH |
357 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
358 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
359 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 360 | */ |
c9a929dd | 361 | void blk_drain_queue(struct request_queue *q, bool drain_all) |
e3c78ca5 TH |
362 | { |
363 | while (true) { | |
481a7d64 TH |
364 | bool drain = false; |
365 | int i; | |
e3c78ca5 TH |
366 | |
367 | spin_lock_irq(q->queue_lock); | |
368 | ||
b855b04a TH |
369 | /* |
370 | * The caller might be trying to drain @q before its | |
371 | * elevator is initialized. | |
372 | */ | |
373 | if (q->elevator) | |
374 | elv_drain_elevator(q); | |
375 | ||
5efd6113 | 376 | blkcg_drain_queue(q); |
e3c78ca5 | 377 | |
4eabc941 TH |
378 | /* |
379 | * This function might be called on a queue which failed | |
b855b04a TH |
380 | * driver init after queue creation or is not yet fully |
381 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
382 | * in such cases. Kick queue iff dispatch queue has | |
383 | * something on it and @q has request_fn set. | |
4eabc941 | 384 | */ |
b855b04a | 385 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 386 | __blk_run_queue(q); |
c9a929dd | 387 | |
481a7d64 TH |
388 | drain |= q->rq.elvpriv; |
389 | ||
390 | /* | |
391 | * Unfortunately, requests are queued at and tracked from | |
392 | * multiple places and there's no single counter which can | |
393 | * be drained. Check all the queues and counters. | |
394 | */ | |
395 | if (drain_all) { | |
396 | drain |= !list_empty(&q->queue_head); | |
397 | for (i = 0; i < 2; i++) { | |
398 | drain |= q->rq.count[i]; | |
399 | drain |= q->in_flight[i]; | |
400 | drain |= !list_empty(&q->flush_queue[i]); | |
401 | } | |
402 | } | |
e3c78ca5 TH |
403 | |
404 | spin_unlock_irq(q->queue_lock); | |
405 | ||
481a7d64 | 406 | if (!drain) |
e3c78ca5 TH |
407 | break; |
408 | msleep(10); | |
409 | } | |
410 | } | |
411 | ||
d732580b TH |
412 | /** |
413 | * blk_queue_bypass_start - enter queue bypass mode | |
414 | * @q: queue of interest | |
415 | * | |
416 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
417 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 418 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
419 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
420 | * inside queue or RCU read lock. | |
d732580b TH |
421 | */ |
422 | void blk_queue_bypass_start(struct request_queue *q) | |
423 | { | |
b82d4b19 TH |
424 | bool drain; |
425 | ||
d732580b | 426 | spin_lock_irq(q->queue_lock); |
b82d4b19 | 427 | drain = !q->bypass_depth++; |
d732580b TH |
428 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
429 | spin_unlock_irq(q->queue_lock); | |
430 | ||
b82d4b19 TH |
431 | if (drain) { |
432 | blk_drain_queue(q, false); | |
433 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ | |
434 | synchronize_rcu(); | |
435 | } | |
d732580b TH |
436 | } |
437 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
438 | ||
439 | /** | |
440 | * blk_queue_bypass_end - leave queue bypass mode | |
441 | * @q: queue of interest | |
442 | * | |
443 | * Leave bypass mode and restore the normal queueing behavior. | |
444 | */ | |
445 | void blk_queue_bypass_end(struct request_queue *q) | |
446 | { | |
447 | spin_lock_irq(q->queue_lock); | |
448 | if (!--q->bypass_depth) | |
449 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
450 | WARN_ON_ONCE(q->bypass_depth < 0); | |
451 | spin_unlock_irq(q->queue_lock); | |
452 | } | |
453 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
454 | ||
c9a929dd TH |
455 | /** |
456 | * blk_cleanup_queue - shutdown a request queue | |
457 | * @q: request queue to shutdown | |
458 | * | |
459 | * Mark @q DEAD, drain all pending requests, destroy and put it. All | |
460 | * future requests will be failed immediately with -ENODEV. | |
c94a96ac | 461 | */ |
6728cb0e | 462 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 463 | { |
c9a929dd | 464 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 465 | |
c9a929dd | 466 | /* mark @q DEAD, no new request or merges will be allowed afterwards */ |
483f4afc | 467 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 468 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
c9a929dd TH |
469 | |
470 | spin_lock_irq(lock); | |
6ecf23af | 471 | |
80fd9979 TH |
472 | /* |
473 | * Dead queue is permanently in bypass mode till released. Note | |
474 | * that, unlike blk_queue_bypass_start(), we aren't performing | |
475 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
476 | * as some drivers create and destroy a lot of queues while | |
477 | * probing. This is still safe because blk_release_queue() will be | |
478 | * called only after the queue refcnt drops to zero and nothing, | |
479 | * RCU or not, would be traversing the queue by then. | |
480 | */ | |
6ecf23af TH |
481 | q->bypass_depth++; |
482 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
483 | ||
c9a929dd TH |
484 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
485 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
486 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
483f4afc | 487 | |
777eb1bf HR |
488 | if (q->queue_lock != &q->__queue_lock) |
489 | q->queue_lock = &q->__queue_lock; | |
da527770 | 490 | |
c9a929dd TH |
491 | spin_unlock_irq(lock); |
492 | mutex_unlock(&q->sysfs_lock); | |
493 | ||
b855b04a TH |
494 | /* drain all requests queued before DEAD marking */ |
495 | blk_drain_queue(q, true); | |
c9a929dd TH |
496 | |
497 | /* @q won't process any more request, flush async actions */ | |
498 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
499 | blk_sync_queue(q); | |
500 | ||
501 | /* @q is and will stay empty, shutdown and put */ | |
483f4afc AV |
502 | blk_put_queue(q); |
503 | } | |
1da177e4 LT |
504 | EXPORT_SYMBOL(blk_cleanup_queue); |
505 | ||
165125e1 | 506 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
507 | { |
508 | struct request_list *rl = &q->rq; | |
509 | ||
1abec4fd MS |
510 | if (unlikely(rl->rq_pool)) |
511 | return 0; | |
512 | ||
1faa16d2 JA |
513 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
514 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 515 | rl->elvpriv = 0; |
1faa16d2 JA |
516 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
517 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 518 | |
1946089a CL |
519 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
520 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
521 | |
522 | if (!rl->rq_pool) | |
523 | return -ENOMEM; | |
524 | ||
525 | return 0; | |
526 | } | |
527 | ||
165125e1 | 528 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 529 | { |
1946089a CL |
530 | return blk_alloc_queue_node(gfp_mask, -1); |
531 | } | |
532 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 533 | |
165125e1 | 534 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 535 | { |
165125e1 | 536 | struct request_queue *q; |
e0bf68dd | 537 | int err; |
1946089a | 538 | |
8324aa91 | 539 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 540 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
541 | if (!q) |
542 | return NULL; | |
543 | ||
a73f730d TH |
544 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); |
545 | if (q->id < 0) | |
546 | goto fail_q; | |
547 | ||
0989a025 JA |
548 | q->backing_dev_info.ra_pages = |
549 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
550 | q->backing_dev_info.state = 0; | |
551 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 552 | q->backing_dev_info.name = "block"; |
5151412d | 553 | q->node = node_id; |
0989a025 | 554 | |
e0bf68dd | 555 | err = bdi_init(&q->backing_dev_info); |
a73f730d TH |
556 | if (err) |
557 | goto fail_id; | |
e0bf68dd | 558 | |
31373d09 MG |
559 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
560 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 561 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 562 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 563 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 564 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 565 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 566 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 567 | #endif |
ae1b1539 TH |
568 | INIT_LIST_HEAD(&q->flush_queue[0]); |
569 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
570 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 571 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 572 | |
8324aa91 | 573 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 574 | |
483f4afc | 575 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 576 | spin_lock_init(&q->__queue_lock); |
483f4afc | 577 | |
c94a96ac VG |
578 | /* |
579 | * By default initialize queue_lock to internal lock and driver can | |
580 | * override it later if need be. | |
581 | */ | |
582 | q->queue_lock = &q->__queue_lock; | |
583 | ||
b82d4b19 TH |
584 | /* |
585 | * A queue starts its life with bypass turned on to avoid | |
586 | * unnecessary bypass on/off overhead and nasty surprises during | |
587 | * init. The initial bypass will be finished at the end of | |
588 | * blk_init_allocated_queue(). | |
589 | */ | |
590 | q->bypass_depth = 1; | |
591 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
592 | ||
5efd6113 | 593 | if (blkcg_init_queue(q)) |
f51b802c TH |
594 | goto fail_id; |
595 | ||
1da177e4 | 596 | return q; |
a73f730d TH |
597 | |
598 | fail_id: | |
599 | ida_simple_remove(&blk_queue_ida, q->id); | |
600 | fail_q: | |
601 | kmem_cache_free(blk_requestq_cachep, q); | |
602 | return NULL; | |
1da177e4 | 603 | } |
1946089a | 604 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
605 | |
606 | /** | |
607 | * blk_init_queue - prepare a request queue for use with a block device | |
608 | * @rfn: The function to be called to process requests that have been | |
609 | * placed on the queue. | |
610 | * @lock: Request queue spin lock | |
611 | * | |
612 | * Description: | |
613 | * If a block device wishes to use the standard request handling procedures, | |
614 | * which sorts requests and coalesces adjacent requests, then it must | |
615 | * call blk_init_queue(). The function @rfn will be called when there | |
616 | * are requests on the queue that need to be processed. If the device | |
617 | * supports plugging, then @rfn may not be called immediately when requests | |
618 | * are available on the queue, but may be called at some time later instead. | |
619 | * Plugged queues are generally unplugged when a buffer belonging to one | |
620 | * of the requests on the queue is needed, or due to memory pressure. | |
621 | * | |
622 | * @rfn is not required, or even expected, to remove all requests off the | |
623 | * queue, but only as many as it can handle at a time. If it does leave | |
624 | * requests on the queue, it is responsible for arranging that the requests | |
625 | * get dealt with eventually. | |
626 | * | |
627 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
628 | * request queue; this lock will be taken also from interrupt context, so irq |
629 | * disabling is needed for it. | |
1da177e4 | 630 | * |
710027a4 | 631 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
632 | * it didn't succeed. |
633 | * | |
634 | * Note: | |
635 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
636 | * when the block device is deactivated (such as at module unload). | |
637 | **/ | |
1946089a | 638 | |
165125e1 | 639 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 640 | { |
1946089a CL |
641 | return blk_init_queue_node(rfn, lock, -1); |
642 | } | |
643 | EXPORT_SYMBOL(blk_init_queue); | |
644 | ||
165125e1 | 645 | struct request_queue * |
1946089a CL |
646 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
647 | { | |
c86d1b8a | 648 | struct request_queue *uninit_q, *q; |
1da177e4 | 649 | |
c86d1b8a MS |
650 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
651 | if (!uninit_q) | |
652 | return NULL; | |
653 | ||
5151412d | 654 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a MS |
655 | if (!q) |
656 | blk_cleanup_queue(uninit_q); | |
657 | ||
658 | return q; | |
01effb0d MS |
659 | } |
660 | EXPORT_SYMBOL(blk_init_queue_node); | |
661 | ||
662 | struct request_queue * | |
663 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
664 | spinlock_t *lock) | |
01effb0d | 665 | { |
1da177e4 LT |
666 | if (!q) |
667 | return NULL; | |
668 | ||
c86d1b8a | 669 | if (blk_init_free_list(q)) |
8669aafd | 670 | return NULL; |
1da177e4 LT |
671 | |
672 | q->request_fn = rfn; | |
1da177e4 | 673 | q->prep_rq_fn = NULL; |
28018c24 | 674 | q->unprep_rq_fn = NULL; |
bc58ba94 | 675 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
676 | |
677 | /* Override internal queue lock with supplied lock pointer */ | |
678 | if (lock) | |
679 | q->queue_lock = lock; | |
1da177e4 | 680 | |
f3b144aa JA |
681 | /* |
682 | * This also sets hw/phys segments, boundary and size | |
683 | */ | |
c20e8de2 | 684 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 685 | |
44ec9542 AS |
686 | q->sg_reserved_size = INT_MAX; |
687 | ||
b82d4b19 TH |
688 | /* init elevator */ |
689 | if (elevator_init(q, NULL)) | |
690 | return NULL; | |
1da177e4 | 691 | |
b82d4b19 TH |
692 | blk_queue_congestion_threshold(q); |
693 | ||
694 | /* all done, end the initial bypass */ | |
695 | blk_queue_bypass_end(q); | |
696 | return q; | |
1da177e4 | 697 | } |
5151412d | 698 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 699 | |
09ac46c4 | 700 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 701 | { |
34f6055c | 702 | if (likely(!blk_queue_dead(q))) { |
09ac46c4 TH |
703 | __blk_get_queue(q); |
704 | return true; | |
1da177e4 LT |
705 | } |
706 | ||
09ac46c4 | 707 | return false; |
1da177e4 | 708 | } |
d86e0e83 | 709 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 710 | |
165125e1 | 711 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 712 | { |
f1f8cc94 | 713 | if (rq->cmd_flags & REQ_ELVPRIV) { |
cb98fc8b | 714 | elv_put_request(q, rq); |
f1f8cc94 | 715 | if (rq->elv.icq) |
11a3122f | 716 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
717 | } |
718 | ||
1da177e4 LT |
719 | mempool_free(rq, q->rq.rq_pool); |
720 | } | |
721 | ||
1ea25ecb | 722 | static struct request * |
852c788f | 723 | blk_alloc_request(struct request_queue *q, struct bio *bio, struct io_cq *icq, |
f1f8cc94 | 724 | unsigned int flags, gfp_t gfp_mask) |
1da177e4 LT |
725 | { |
726 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
727 | ||
728 | if (!rq) | |
729 | return NULL; | |
730 | ||
2a4aa30c | 731 | blk_rq_init(q, rq); |
1afb20f3 | 732 | |
42dad764 | 733 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 734 | |
f1f8cc94 TH |
735 | if (flags & REQ_ELVPRIV) { |
736 | rq->elv.icq = icq; | |
852c788f | 737 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { |
f1f8cc94 TH |
738 | mempool_free(rq, q->rq.rq_pool); |
739 | return NULL; | |
740 | } | |
741 | /* @rq->elv.icq holds on to io_context until @rq is freed */ | |
742 | if (icq) | |
743 | get_io_context(icq->ioc); | |
cb98fc8b | 744 | } |
1da177e4 | 745 | |
cb98fc8b | 746 | return rq; |
1da177e4 LT |
747 | } |
748 | ||
749 | /* | |
750 | * ioc_batching returns true if the ioc is a valid batching request and | |
751 | * should be given priority access to a request. | |
752 | */ | |
165125e1 | 753 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
754 | { |
755 | if (!ioc) | |
756 | return 0; | |
757 | ||
758 | /* | |
759 | * Make sure the process is able to allocate at least 1 request | |
760 | * even if the batch times out, otherwise we could theoretically | |
761 | * lose wakeups. | |
762 | */ | |
763 | return ioc->nr_batch_requests == q->nr_batching || | |
764 | (ioc->nr_batch_requests > 0 | |
765 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
766 | } | |
767 | ||
768 | /* | |
769 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
770 | * will cause the process to be a "batcher" on all queues in the system. This | |
771 | * is the behaviour we want though - once it gets a wakeup it should be given | |
772 | * a nice run. | |
773 | */ | |
165125e1 | 774 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
775 | { |
776 | if (!ioc || ioc_batching(q, ioc)) | |
777 | return; | |
778 | ||
779 | ioc->nr_batch_requests = q->nr_batching; | |
780 | ioc->last_waited = jiffies; | |
781 | } | |
782 | ||
1faa16d2 | 783 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
784 | { |
785 | struct request_list *rl = &q->rq; | |
786 | ||
1faa16d2 JA |
787 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
788 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 789 | |
1faa16d2 JA |
790 | if (rl->count[sync] + 1 <= q->nr_requests) { |
791 | if (waitqueue_active(&rl->wait[sync])) | |
792 | wake_up(&rl->wait[sync]); | |
1da177e4 | 793 | |
1faa16d2 | 794 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
795 | } |
796 | } | |
797 | ||
798 | /* | |
799 | * A request has just been released. Account for it, update the full and | |
800 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
801 | */ | |
75eb6c37 | 802 | static void freed_request(struct request_queue *q, unsigned int flags) |
1da177e4 LT |
803 | { |
804 | struct request_list *rl = &q->rq; | |
75eb6c37 | 805 | int sync = rw_is_sync(flags); |
1da177e4 | 806 | |
1faa16d2 | 807 | rl->count[sync]--; |
75eb6c37 | 808 | if (flags & REQ_ELVPRIV) |
cb98fc8b | 809 | rl->elvpriv--; |
1da177e4 | 810 | |
1faa16d2 | 811 | __freed_request(q, sync); |
1da177e4 | 812 | |
1faa16d2 JA |
813 | if (unlikely(rl->starved[sync ^ 1])) |
814 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
815 | } |
816 | ||
9d5a4e94 MS |
817 | /* |
818 | * Determine if elevator data should be initialized when allocating the | |
819 | * request associated with @bio. | |
820 | */ | |
821 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
822 | { | |
823 | if (!bio) | |
824 | return true; | |
825 | ||
826 | /* | |
827 | * Flush requests do not use the elevator so skip initialization. | |
828 | * This allows a request to share the flush and elevator data. | |
829 | */ | |
830 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
831 | return false; | |
832 | ||
833 | return true; | |
834 | } | |
835 | ||
852c788f TH |
836 | /** |
837 | * rq_ioc - determine io_context for request allocation | |
838 | * @bio: request being allocated is for this bio (can be %NULL) | |
839 | * | |
840 | * Determine io_context to use for request allocation for @bio. May return | |
841 | * %NULL if %current->io_context doesn't exist. | |
842 | */ | |
843 | static struct io_context *rq_ioc(struct bio *bio) | |
844 | { | |
845 | #ifdef CONFIG_BLK_CGROUP | |
846 | if (bio && bio->bi_ioc) | |
847 | return bio->bi_ioc; | |
848 | #endif | |
849 | return current->io_context; | |
850 | } | |
851 | ||
da8303c6 TH |
852 | /** |
853 | * get_request - get a free request | |
854 | * @q: request_queue to allocate request from | |
855 | * @rw_flags: RW and SYNC flags | |
856 | * @bio: bio to allocate request for (can be %NULL) | |
857 | * @gfp_mask: allocation mask | |
858 | * | |
859 | * Get a free request from @q. This function may fail under memory | |
860 | * pressure or if @q is dead. | |
861 | * | |
862 | * Must be callled with @q->queue_lock held and, | |
863 | * Returns %NULL on failure, with @q->queue_lock held. | |
864 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 865 | */ |
165125e1 | 866 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 867 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 868 | { |
b679281a | 869 | struct request *rq; |
1da177e4 | 870 | struct request_list *rl = &q->rq; |
f1f8cc94 | 871 | struct elevator_type *et; |
f2dbd76a | 872 | struct io_context *ioc; |
f1f8cc94 | 873 | struct io_cq *icq = NULL; |
1faa16d2 | 874 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
f2dbd76a | 875 | bool retried = false; |
75eb6c37 | 876 | int may_queue; |
f2dbd76a | 877 | retry: |
f1f8cc94 | 878 | et = q->elevator->type; |
852c788f | 879 | ioc = rq_ioc(bio); |
88ee5ef1 | 880 | |
34f6055c | 881 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
882 | return NULL; |
883 | ||
7749a8d4 | 884 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
885 | if (may_queue == ELV_MQUEUE_NO) |
886 | goto rq_starved; | |
887 | ||
1faa16d2 JA |
888 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
889 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
f2dbd76a TH |
890 | /* |
891 | * We want ioc to record batching state. If it's | |
892 | * not already there, creating a new one requires | |
893 | * dropping queue_lock, which in turn requires | |
894 | * retesting conditions to avoid queue hang. | |
895 | */ | |
896 | if (!ioc && !retried) { | |
897 | spin_unlock_irq(q->queue_lock); | |
24acfc34 | 898 | create_io_context(gfp_mask, q->node); |
f2dbd76a TH |
899 | spin_lock_irq(q->queue_lock); |
900 | retried = true; | |
901 | goto retry; | |
902 | } | |
903 | ||
88ee5ef1 JA |
904 | /* |
905 | * The queue will fill after this allocation, so set | |
906 | * it as full, and mark this process as "batching". | |
907 | * This process will be allowed to complete a batch of | |
908 | * requests, others will be blocked. | |
909 | */ | |
1faa16d2 | 910 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 911 | ioc_set_batching(q, ioc); |
1faa16d2 | 912 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
913 | } else { |
914 | if (may_queue != ELV_MQUEUE_MUST | |
915 | && !ioc_batching(q, ioc)) { | |
916 | /* | |
917 | * The queue is full and the allocating | |
918 | * process is not a "batcher", and not | |
919 | * exempted by the IO scheduler | |
920 | */ | |
b679281a | 921 | return NULL; |
88ee5ef1 JA |
922 | } |
923 | } | |
1da177e4 | 924 | } |
1faa16d2 | 925 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
926 | } |
927 | ||
082cf69e JA |
928 | /* |
929 | * Only allow batching queuers to allocate up to 50% over the defined | |
930 | * limit of requests, otherwise we could have thousands of requests | |
931 | * allocated with any setting of ->nr_requests | |
932 | */ | |
1faa16d2 | 933 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
b679281a | 934 | return NULL; |
fd782a4a | 935 | |
1faa16d2 JA |
936 | rl->count[is_sync]++; |
937 | rl->starved[is_sync] = 0; | |
cb98fc8b | 938 | |
f1f8cc94 TH |
939 | /* |
940 | * Decide whether the new request will be managed by elevator. If | |
941 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will | |
942 | * prevent the current elevator from being destroyed until the new | |
943 | * request is freed. This guarantees icq's won't be destroyed and | |
944 | * makes creating new ones safe. | |
945 | * | |
946 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
947 | * it will be created after releasing queue_lock. | |
948 | */ | |
d732580b | 949 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
75eb6c37 TH |
950 | rw_flags |= REQ_ELVPRIV; |
951 | rl->elvpriv++; | |
f1f8cc94 TH |
952 | if (et->icq_cache && ioc) |
953 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 954 | } |
cb98fc8b | 955 | |
f253b86b JA |
956 | if (blk_queue_io_stat(q)) |
957 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
958 | spin_unlock_irq(q->queue_lock); |
959 | ||
f1f8cc94 | 960 | /* create icq if missing */ |
05c30b95 | 961 | if ((rw_flags & REQ_ELVPRIV) && unlikely(et->icq_cache && !icq)) { |
852c788f TH |
962 | create_io_context(gfp_mask, q->node); |
963 | ioc = rq_ioc(bio); | |
964 | if (!ioc) | |
965 | goto fail_alloc; | |
966 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
05c30b95 | 967 | if (!icq) |
b679281a | 968 | goto fail_alloc; |
05c30b95 | 969 | } |
f1f8cc94 | 970 | |
852c788f | 971 | rq = blk_alloc_request(q, bio, icq, rw_flags, gfp_mask); |
b679281a TH |
972 | if (unlikely(!rq)) |
973 | goto fail_alloc; | |
1da177e4 | 974 | |
88ee5ef1 JA |
975 | /* |
976 | * ioc may be NULL here, and ioc_batching will be false. That's | |
977 | * OK, if the queue is under the request limit then requests need | |
978 | * not count toward the nr_batch_requests limit. There will always | |
979 | * be some limit enforced by BLK_BATCH_TIME. | |
980 | */ | |
1da177e4 LT |
981 | if (ioc_batching(q, ioc)) |
982 | ioc->nr_batch_requests--; | |
6728cb0e | 983 | |
1faa16d2 | 984 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 985 | return rq; |
b679281a TH |
986 | |
987 | fail_alloc: | |
988 | /* | |
989 | * Allocation failed presumably due to memory. Undo anything we | |
990 | * might have messed up. | |
991 | * | |
992 | * Allocating task should really be put onto the front of the wait | |
993 | * queue, but this is pretty rare. | |
994 | */ | |
995 | spin_lock_irq(q->queue_lock); | |
996 | freed_request(q, rw_flags); | |
997 | ||
998 | /* | |
999 | * in the very unlikely event that allocation failed and no | |
1000 | * requests for this direction was pending, mark us starved so that | |
1001 | * freeing of a request in the other direction will notice | |
1002 | * us. another possible fix would be to split the rq mempool into | |
1003 | * READ and WRITE | |
1004 | */ | |
1005 | rq_starved: | |
1006 | if (unlikely(rl->count[is_sync] == 0)) | |
1007 | rl->starved[is_sync] = 1; | |
1008 | return NULL; | |
1da177e4 LT |
1009 | } |
1010 | ||
da8303c6 TH |
1011 | /** |
1012 | * get_request_wait - get a free request with retry | |
1013 | * @q: request_queue to allocate request from | |
1014 | * @rw_flags: RW and SYNC flags | |
1015 | * @bio: bio to allocate request for (can be %NULL) | |
1016 | * | |
1017 | * Get a free request from @q. This function keeps retrying under memory | |
1018 | * pressure and fails iff @q is dead. | |
d6344532 | 1019 | * |
da8303c6 TH |
1020 | * Must be callled with @q->queue_lock held and, |
1021 | * Returns %NULL on failure, with @q->queue_lock held. | |
1022 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 1023 | */ |
165125e1 | 1024 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 1025 | struct bio *bio) |
1da177e4 | 1026 | { |
1faa16d2 | 1027 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
1028 | struct request *rq; |
1029 | ||
7749a8d4 | 1030 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
1031 | while (!rq) { |
1032 | DEFINE_WAIT(wait); | |
1da177e4 LT |
1033 | struct request_list *rl = &q->rq; |
1034 | ||
34f6055c | 1035 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
1036 | return NULL; |
1037 | ||
1faa16d2 | 1038 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
1039 | TASK_UNINTERRUPTIBLE); |
1040 | ||
1faa16d2 | 1041 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 1042 | |
05caf8db ZY |
1043 | spin_unlock_irq(q->queue_lock); |
1044 | io_schedule(); | |
1da177e4 | 1045 | |
05caf8db ZY |
1046 | /* |
1047 | * After sleeping, we become a "batching" process and | |
1048 | * will be able to allocate at least one request, and | |
1049 | * up to a big batch of them for a small period time. | |
1050 | * See ioc_batching, ioc_set_batching | |
1051 | */ | |
24acfc34 | 1052 | create_io_context(GFP_NOIO, q->node); |
f2dbd76a | 1053 | ioc_set_batching(q, current->io_context); |
d6344532 | 1054 | |
05caf8db | 1055 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 1056 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
1057 | |
1058 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
1059 | }; | |
1da177e4 LT |
1060 | |
1061 | return rq; | |
1062 | } | |
1063 | ||
165125e1 | 1064 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
1065 | { |
1066 | struct request *rq; | |
1067 | ||
1068 | BUG_ON(rw != READ && rw != WRITE); | |
1069 | ||
d6344532 | 1070 | spin_lock_irq(q->queue_lock); |
da8303c6 | 1071 | if (gfp_mask & __GFP_WAIT) |
22e2c507 | 1072 | rq = get_request_wait(q, rw, NULL); |
da8303c6 | 1073 | else |
22e2c507 | 1074 | rq = get_request(q, rw, NULL, gfp_mask); |
da8303c6 TH |
1075 | if (!rq) |
1076 | spin_unlock_irq(q->queue_lock); | |
d6344532 | 1077 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
1078 | |
1079 | return rq; | |
1080 | } | |
1da177e4 LT |
1081 | EXPORT_SYMBOL(blk_get_request); |
1082 | ||
dc72ef4a | 1083 | /** |
79eb63e9 | 1084 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 1085 | * @q: target request queue |
79eb63e9 BH |
1086 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
1087 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 1088 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 1089 | * |
79eb63e9 BH |
1090 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
1091 | * type commands. Where the struct request needs to be farther initialized by | |
1092 | * the caller. It is passed a &struct bio, which describes the memory info of | |
1093 | * the I/O transfer. | |
dc72ef4a | 1094 | * |
79eb63e9 BH |
1095 | * The caller of blk_make_request must make sure that bi_io_vec |
1096 | * are set to describe the memory buffers. That bio_data_dir() will return | |
1097 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
1098 | * are properly set accordingly) | |
1099 | * | |
1100 | * If called under none-sleepable conditions, mapped bio buffers must not | |
1101 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
1102 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
1103 | * BUG. | |
53674ac5 JA |
1104 | * |
1105 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
1106 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
1107 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
1108 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
1109 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
1110 | * of bio_alloc(), as that avoids the mempool deadlock. | |
1111 | * If possible a big IO should be split into smaller parts when allocation | |
1112 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 1113 | */ |
79eb63e9 BH |
1114 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
1115 | gfp_t gfp_mask) | |
dc72ef4a | 1116 | { |
79eb63e9 BH |
1117 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1118 | ||
1119 | if (unlikely(!rq)) | |
1120 | return ERR_PTR(-ENOMEM); | |
1121 | ||
1122 | for_each_bio(bio) { | |
1123 | struct bio *bounce_bio = bio; | |
1124 | int ret; | |
1125 | ||
1126 | blk_queue_bounce(q, &bounce_bio); | |
1127 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1128 | if (unlikely(ret)) { | |
1129 | blk_put_request(rq); | |
1130 | return ERR_PTR(ret); | |
1131 | } | |
1132 | } | |
1133 | ||
1134 | return rq; | |
dc72ef4a | 1135 | } |
79eb63e9 | 1136 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1137 | |
1da177e4 LT |
1138 | /** |
1139 | * blk_requeue_request - put a request back on queue | |
1140 | * @q: request queue where request should be inserted | |
1141 | * @rq: request to be inserted | |
1142 | * | |
1143 | * Description: | |
1144 | * Drivers often keep queueing requests until the hardware cannot accept | |
1145 | * more, when that condition happens we need to put the request back | |
1146 | * on the queue. Must be called with queue lock held. | |
1147 | */ | |
165125e1 | 1148 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1149 | { |
242f9dcb JA |
1150 | blk_delete_timer(rq); |
1151 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1152 | trace_block_rq_requeue(q, rq); |
2056a782 | 1153 | |
1da177e4 LT |
1154 | if (blk_rq_tagged(rq)) |
1155 | blk_queue_end_tag(q, rq); | |
1156 | ||
ba396a6c JB |
1157 | BUG_ON(blk_queued_rq(rq)); |
1158 | ||
1da177e4 LT |
1159 | elv_requeue_request(q, rq); |
1160 | } | |
1da177e4 LT |
1161 | EXPORT_SYMBOL(blk_requeue_request); |
1162 | ||
73c10101 JA |
1163 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1164 | int where) | |
1165 | { | |
1166 | drive_stat_acct(rq, 1); | |
7eaceacc | 1167 | __elv_add_request(q, rq, where); |
73c10101 JA |
1168 | } |
1169 | ||
074a7aca TH |
1170 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1171 | unsigned long now) | |
1172 | { | |
1173 | if (now == part->stamp) | |
1174 | return; | |
1175 | ||
316d315b | 1176 | if (part_in_flight(part)) { |
074a7aca | 1177 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1178 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1179 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1180 | } | |
1181 | part->stamp = now; | |
1182 | } | |
1183 | ||
1184 | /** | |
496aa8a9 RD |
1185 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1186 | * @cpu: cpu number for stats access | |
1187 | * @part: target partition | |
1da177e4 LT |
1188 | * |
1189 | * The average IO queue length and utilisation statistics are maintained | |
1190 | * by observing the current state of the queue length and the amount of | |
1191 | * time it has been in this state for. | |
1192 | * | |
1193 | * Normally, that accounting is done on IO completion, but that can result | |
1194 | * in more than a second's worth of IO being accounted for within any one | |
1195 | * second, leading to >100% utilisation. To deal with that, we call this | |
1196 | * function to do a round-off before returning the results when reading | |
1197 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1198 | * the current jiffies and restarts the counters again. | |
1199 | */ | |
c9959059 | 1200 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1201 | { |
1202 | unsigned long now = jiffies; | |
1203 | ||
074a7aca TH |
1204 | if (part->partno) |
1205 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1206 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1207 | } |
074a7aca | 1208 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1209 | |
1da177e4 LT |
1210 | /* |
1211 | * queue lock must be held | |
1212 | */ | |
165125e1 | 1213 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1214 | { |
1da177e4 LT |
1215 | if (unlikely(!q)) |
1216 | return; | |
1217 | if (unlikely(--req->ref_count)) | |
1218 | return; | |
1219 | ||
8922e16c TH |
1220 | elv_completed_request(q, req); |
1221 | ||
1cd96c24 BH |
1222 | /* this is a bio leak */ |
1223 | WARN_ON(req->bio != NULL); | |
1224 | ||
1da177e4 LT |
1225 | /* |
1226 | * Request may not have originated from ll_rw_blk. if not, | |
1227 | * it didn't come out of our reserved rq pools | |
1228 | */ | |
49171e5c | 1229 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1230 | unsigned int flags = req->cmd_flags; |
1da177e4 | 1231 | |
1da177e4 | 1232 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1233 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1234 | |
1235 | blk_free_request(q, req); | |
75eb6c37 | 1236 | freed_request(q, flags); |
1da177e4 LT |
1237 | } |
1238 | } | |
6e39b69e MC |
1239 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1240 | ||
1da177e4 LT |
1241 | void blk_put_request(struct request *req) |
1242 | { | |
8922e16c | 1243 | unsigned long flags; |
165125e1 | 1244 | struct request_queue *q = req->q; |
8922e16c | 1245 | |
52a93ba8 FT |
1246 | spin_lock_irqsave(q->queue_lock, flags); |
1247 | __blk_put_request(q, req); | |
1248 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1249 | } |
1da177e4 LT |
1250 | EXPORT_SYMBOL(blk_put_request); |
1251 | ||
66ac0280 CH |
1252 | /** |
1253 | * blk_add_request_payload - add a payload to a request | |
1254 | * @rq: request to update | |
1255 | * @page: page backing the payload | |
1256 | * @len: length of the payload. | |
1257 | * | |
1258 | * This allows to later add a payload to an already submitted request by | |
1259 | * a block driver. The driver needs to take care of freeing the payload | |
1260 | * itself. | |
1261 | * | |
1262 | * Note that this is a quite horrible hack and nothing but handling of | |
1263 | * discard requests should ever use it. | |
1264 | */ | |
1265 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1266 | unsigned int len) | |
1267 | { | |
1268 | struct bio *bio = rq->bio; | |
1269 | ||
1270 | bio->bi_io_vec->bv_page = page; | |
1271 | bio->bi_io_vec->bv_offset = 0; | |
1272 | bio->bi_io_vec->bv_len = len; | |
1273 | ||
1274 | bio->bi_size = len; | |
1275 | bio->bi_vcnt = 1; | |
1276 | bio->bi_phys_segments = 1; | |
1277 | ||
1278 | rq->__data_len = rq->resid_len = len; | |
1279 | rq->nr_phys_segments = 1; | |
1280 | rq->buffer = bio_data(bio); | |
1281 | } | |
1282 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1283 | ||
73c10101 JA |
1284 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1285 | struct bio *bio) | |
1286 | { | |
1287 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1288 | ||
73c10101 JA |
1289 | if (!ll_back_merge_fn(q, req, bio)) |
1290 | return false; | |
1291 | ||
1292 | trace_block_bio_backmerge(q, bio); | |
1293 | ||
1294 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1295 | blk_rq_set_mixed_merge(req); | |
1296 | ||
1297 | req->biotail->bi_next = bio; | |
1298 | req->biotail = bio; | |
1299 | req->__data_len += bio->bi_size; | |
1300 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1301 | ||
1302 | drive_stat_acct(req, 0); | |
1303 | return true; | |
1304 | } | |
1305 | ||
1306 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1307 | struct request *req, struct bio *bio) | |
1308 | { | |
1309 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1310 | |
73c10101 JA |
1311 | if (!ll_front_merge_fn(q, req, bio)) |
1312 | return false; | |
1313 | ||
1314 | trace_block_bio_frontmerge(q, bio); | |
1315 | ||
1316 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1317 | blk_rq_set_mixed_merge(req); | |
1318 | ||
73c10101 JA |
1319 | bio->bi_next = req->bio; |
1320 | req->bio = bio; | |
1321 | ||
1322 | /* | |
1323 | * may not be valid. if the low level driver said | |
1324 | * it didn't need a bounce buffer then it better | |
1325 | * not touch req->buffer either... | |
1326 | */ | |
1327 | req->buffer = bio_data(bio); | |
1328 | req->__sector = bio->bi_sector; | |
1329 | req->__data_len += bio->bi_size; | |
1330 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1331 | ||
1332 | drive_stat_acct(req, 0); | |
1333 | return true; | |
1334 | } | |
1335 | ||
bd87b589 TH |
1336 | /** |
1337 | * attempt_plug_merge - try to merge with %current's plugged list | |
1338 | * @q: request_queue new bio is being queued at | |
1339 | * @bio: new bio being queued | |
1340 | * @request_count: out parameter for number of traversed plugged requests | |
1341 | * | |
1342 | * Determine whether @bio being queued on @q can be merged with a request | |
1343 | * on %current's plugged list. Returns %true if merge was successful, | |
1344 | * otherwise %false. | |
1345 | * | |
07c2bd37 TH |
1346 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1347 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1348 | * than scheduling, and the request, while may have elvpriv data, is not | |
1349 | * added on the elevator at this point. In addition, we don't have | |
1350 | * reliable access to the elevator outside queue lock. Only check basic | |
1351 | * merging parameters without querying the elevator. | |
73c10101 | 1352 | */ |
bd87b589 TH |
1353 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1354 | unsigned int *request_count) | |
73c10101 JA |
1355 | { |
1356 | struct blk_plug *plug; | |
1357 | struct request *rq; | |
1358 | bool ret = false; | |
1359 | ||
bd87b589 | 1360 | plug = current->plug; |
73c10101 JA |
1361 | if (!plug) |
1362 | goto out; | |
56ebdaf2 | 1363 | *request_count = 0; |
73c10101 JA |
1364 | |
1365 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1366 | int el_ret; | |
1367 | ||
56ebdaf2 SL |
1368 | (*request_count)++; |
1369 | ||
07c2bd37 | 1370 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1371 | continue; |
1372 | ||
050c8ea8 | 1373 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1374 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1375 | ret = bio_attempt_back_merge(q, rq, bio); | |
1376 | if (ret) | |
1377 | break; | |
1378 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1379 | ret = bio_attempt_front_merge(q, rq, bio); | |
1380 | if (ret) | |
1381 | break; | |
1382 | } | |
1383 | } | |
1384 | out: | |
1385 | return ret; | |
1386 | } | |
1387 | ||
86db1e29 | 1388 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1389 | { |
4aff5e23 | 1390 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1391 | |
7b6d91da CH |
1392 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1393 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1394 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1395 | |
52d9e675 | 1396 | req->errors = 0; |
a2dec7b3 | 1397 | req->__sector = bio->bi_sector; |
52d9e675 | 1398 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1399 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1400 | } |
1401 | ||
5a7bbad2 | 1402 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1403 | { |
5e00d1b5 | 1404 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1405 | struct blk_plug *plug; |
1406 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1407 | struct request *req; | |
56ebdaf2 | 1408 | unsigned int request_count = 0; |
1da177e4 | 1409 | |
1da177e4 LT |
1410 | /* |
1411 | * low level driver can indicate that it wants pages above a | |
1412 | * certain limit bounced to low memory (ie for highmem, or even | |
1413 | * ISA dma in theory) | |
1414 | */ | |
1415 | blk_queue_bounce(q, &bio); | |
1416 | ||
4fed947c | 1417 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1418 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1419 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1420 | goto get_rq; |
1421 | } | |
1422 | ||
73c10101 JA |
1423 | /* |
1424 | * Check if we can merge with the plugged list before grabbing | |
1425 | * any locks. | |
1426 | */ | |
bd87b589 | 1427 | if (attempt_plug_merge(q, bio, &request_count)) |
5a7bbad2 | 1428 | return; |
1da177e4 | 1429 | |
73c10101 | 1430 | spin_lock_irq(q->queue_lock); |
2056a782 | 1431 | |
73c10101 JA |
1432 | el_ret = elv_merge(q, &req, bio); |
1433 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1434 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1435 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1436 | if (!attempt_back_merge(q, req)) |
1437 | elv_merged_request(q, req, el_ret); | |
1438 | goto out_unlock; | |
1439 | } | |
1440 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1441 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1442 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1443 | if (!attempt_front_merge(q, req)) |
1444 | elv_merged_request(q, req, el_ret); | |
1445 | goto out_unlock; | |
80a761fd | 1446 | } |
1da177e4 LT |
1447 | } |
1448 | ||
450991bc | 1449 | get_rq: |
7749a8d4 JA |
1450 | /* |
1451 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1452 | * but we need to set it earlier to expose the sync flag to the | |
1453 | * rq allocator and io schedulers. | |
1454 | */ | |
1455 | rw_flags = bio_data_dir(bio); | |
1456 | if (sync) | |
7b6d91da | 1457 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1458 | |
1da177e4 | 1459 | /* |
450991bc | 1460 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1461 | * Returns with the queue unlocked. |
450991bc | 1462 | */ |
7749a8d4 | 1463 | req = get_request_wait(q, rw_flags, bio); |
da8303c6 TH |
1464 | if (unlikely(!req)) { |
1465 | bio_endio(bio, -ENODEV); /* @q is dead */ | |
1466 | goto out_unlock; | |
1467 | } | |
d6344532 | 1468 | |
450991bc NP |
1469 | /* |
1470 | * After dropping the lock and possibly sleeping here, our request | |
1471 | * may now be mergeable after it had proven unmergeable (above). | |
1472 | * We don't worry about that case for efficiency. It won't happen | |
1473 | * often, and the elevators are able to handle it. | |
1da177e4 | 1474 | */ |
52d9e675 | 1475 | init_request_from_bio(req, bio); |
1da177e4 | 1476 | |
9562ad9a | 1477 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1478 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1479 | |
1480 | plug = current->plug; | |
721a9602 | 1481 | if (plug) { |
dc6d36c9 JA |
1482 | /* |
1483 | * If this is the first request added after a plug, fire | |
1484 | * of a plug trace. If others have been added before, check | |
1485 | * if we have multiple devices in this plug. If so, make a | |
1486 | * note to sort the list before dispatch. | |
1487 | */ | |
1488 | if (list_empty(&plug->list)) | |
1489 | trace_block_plug(q); | |
3540d5e8 SL |
1490 | else { |
1491 | if (!plug->should_sort) { | |
1492 | struct request *__rq; | |
73c10101 | 1493 | |
3540d5e8 SL |
1494 | __rq = list_entry_rq(plug->list.prev); |
1495 | if (__rq->q != q) | |
1496 | plug->should_sort = 1; | |
1497 | } | |
019ceb7d | 1498 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
3540d5e8 | 1499 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1500 | trace_block_plug(q); |
1501 | } | |
73c10101 | 1502 | } |
73c10101 JA |
1503 | list_add_tail(&req->queuelist, &plug->list); |
1504 | drive_stat_acct(req, 1); | |
1505 | } else { | |
1506 | spin_lock_irq(q->queue_lock); | |
1507 | add_acct_request(q, req, where); | |
24ecfbe2 | 1508 | __blk_run_queue(q); |
73c10101 JA |
1509 | out_unlock: |
1510 | spin_unlock_irq(q->queue_lock); | |
1511 | } | |
1da177e4 | 1512 | } |
c20e8de2 | 1513 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1514 | |
1515 | /* | |
1516 | * If bio->bi_dev is a partition, remap the location | |
1517 | */ | |
1518 | static inline void blk_partition_remap(struct bio *bio) | |
1519 | { | |
1520 | struct block_device *bdev = bio->bi_bdev; | |
1521 | ||
bf2de6f5 | 1522 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1523 | struct hd_struct *p = bdev->bd_part; |
1524 | ||
1da177e4 LT |
1525 | bio->bi_sector += p->start_sect; |
1526 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1527 | |
d07335e5 MS |
1528 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1529 | bdev->bd_dev, | |
1530 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1531 | } |
1532 | } | |
1533 | ||
1da177e4 LT |
1534 | static void handle_bad_sector(struct bio *bio) |
1535 | { | |
1536 | char b[BDEVNAME_SIZE]; | |
1537 | ||
1538 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1539 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1540 | bdevname(bio->bi_bdev, b), | |
1541 | bio->bi_rw, | |
1542 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1543 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1544 | |
1545 | set_bit(BIO_EOF, &bio->bi_flags); | |
1546 | } | |
1547 | ||
c17bb495 AM |
1548 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1549 | ||
1550 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1551 | ||
1552 | static int __init setup_fail_make_request(char *str) | |
1553 | { | |
1554 | return setup_fault_attr(&fail_make_request, str); | |
1555 | } | |
1556 | __setup("fail_make_request=", setup_fail_make_request); | |
1557 | ||
b2c9cd37 | 1558 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1559 | { |
b2c9cd37 | 1560 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1561 | } |
1562 | ||
1563 | static int __init fail_make_request_debugfs(void) | |
1564 | { | |
dd48c085 AM |
1565 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1566 | NULL, &fail_make_request); | |
1567 | ||
1568 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; | |
c17bb495 AM |
1569 | } |
1570 | ||
1571 | late_initcall(fail_make_request_debugfs); | |
1572 | ||
1573 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1574 | ||
b2c9cd37 AM |
1575 | static inline bool should_fail_request(struct hd_struct *part, |
1576 | unsigned int bytes) | |
c17bb495 | 1577 | { |
b2c9cd37 | 1578 | return false; |
c17bb495 AM |
1579 | } |
1580 | ||
1581 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1582 | ||
c07e2b41 JA |
1583 | /* |
1584 | * Check whether this bio extends beyond the end of the device. | |
1585 | */ | |
1586 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1587 | { | |
1588 | sector_t maxsector; | |
1589 | ||
1590 | if (!nr_sectors) | |
1591 | return 0; | |
1592 | ||
1593 | /* Test device or partition size, when known. */ | |
77304d2a | 1594 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1595 | if (maxsector) { |
1596 | sector_t sector = bio->bi_sector; | |
1597 | ||
1598 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1599 | /* | |
1600 | * This may well happen - the kernel calls bread() | |
1601 | * without checking the size of the device, e.g., when | |
1602 | * mounting a device. | |
1603 | */ | |
1604 | handle_bad_sector(bio); | |
1605 | return 1; | |
1606 | } | |
1607 | } | |
1608 | ||
1609 | return 0; | |
1610 | } | |
1611 | ||
27a84d54 CH |
1612 | static noinline_for_stack bool |
1613 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1614 | { |
165125e1 | 1615 | struct request_queue *q; |
5a7bbad2 | 1616 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1617 | int err = -EIO; |
5a7bbad2 CH |
1618 | char b[BDEVNAME_SIZE]; |
1619 | struct hd_struct *part; | |
1da177e4 LT |
1620 | |
1621 | might_sleep(); | |
1da177e4 | 1622 | |
c07e2b41 JA |
1623 | if (bio_check_eod(bio, nr_sectors)) |
1624 | goto end_io; | |
1da177e4 | 1625 | |
5a7bbad2 CH |
1626 | q = bdev_get_queue(bio->bi_bdev); |
1627 | if (unlikely(!q)) { | |
1628 | printk(KERN_ERR | |
1629 | "generic_make_request: Trying to access " | |
1630 | "nonexistent block-device %s (%Lu)\n", | |
1631 | bdevname(bio->bi_bdev, b), | |
1632 | (long long) bio->bi_sector); | |
1633 | goto end_io; | |
1634 | } | |
c17bb495 | 1635 | |
5a7bbad2 CH |
1636 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
1637 | nr_sectors > queue_max_hw_sectors(q))) { | |
1638 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1639 | bdevname(bio->bi_bdev, b), | |
1640 | bio_sectors(bio), | |
1641 | queue_max_hw_sectors(q)); | |
1642 | goto end_io; | |
1643 | } | |
1da177e4 | 1644 | |
5a7bbad2 CH |
1645 | part = bio->bi_bdev->bd_part; |
1646 | if (should_fail_request(part, bio->bi_size) || | |
1647 | should_fail_request(&part_to_disk(part)->part0, | |
1648 | bio->bi_size)) | |
1649 | goto end_io; | |
2056a782 | 1650 | |
5a7bbad2 CH |
1651 | /* |
1652 | * If this device has partitions, remap block n | |
1653 | * of partition p to block n+start(p) of the disk. | |
1654 | */ | |
1655 | blk_partition_remap(bio); | |
2056a782 | 1656 | |
5a7bbad2 CH |
1657 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1658 | goto end_io; | |
a7384677 | 1659 | |
5a7bbad2 CH |
1660 | if (bio_check_eod(bio, nr_sectors)) |
1661 | goto end_io; | |
1e87901e | 1662 | |
5a7bbad2 CH |
1663 | /* |
1664 | * Filter flush bio's early so that make_request based | |
1665 | * drivers without flush support don't have to worry | |
1666 | * about them. | |
1667 | */ | |
1668 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1669 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1670 | if (!nr_sectors) { | |
1671 | err = 0; | |
51fd77bd JA |
1672 | goto end_io; |
1673 | } | |
5a7bbad2 | 1674 | } |
5ddfe969 | 1675 | |
5a7bbad2 CH |
1676 | if ((bio->bi_rw & REQ_DISCARD) && |
1677 | (!blk_queue_discard(q) || | |
1678 | ((bio->bi_rw & REQ_SECURE) && | |
1679 | !blk_queue_secdiscard(q)))) { | |
1680 | err = -EOPNOTSUPP; | |
1681 | goto end_io; | |
1682 | } | |
01edede4 | 1683 | |
bc16a4f9 TH |
1684 | if (blk_throtl_bio(q, bio)) |
1685 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1686 | |
5a7bbad2 | 1687 | trace_block_bio_queue(q, bio); |
27a84d54 | 1688 | return true; |
a7384677 TH |
1689 | |
1690 | end_io: | |
1691 | bio_endio(bio, err); | |
27a84d54 | 1692 | return false; |
1da177e4 LT |
1693 | } |
1694 | ||
27a84d54 CH |
1695 | /** |
1696 | * generic_make_request - hand a buffer to its device driver for I/O | |
1697 | * @bio: The bio describing the location in memory and on the device. | |
1698 | * | |
1699 | * generic_make_request() is used to make I/O requests of block | |
1700 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1701 | * to be done. | |
1702 | * | |
1703 | * generic_make_request() does not return any status. The | |
1704 | * success/failure status of the request, along with notification of | |
1705 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1706 | * function described (one day) else where. | |
1707 | * | |
1708 | * The caller of generic_make_request must make sure that bi_io_vec | |
1709 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1710 | * set to describe the device address, and the | |
1711 | * bi_end_io and optionally bi_private are set to describe how | |
1712 | * completion notification should be signaled. | |
1713 | * | |
1714 | * generic_make_request and the drivers it calls may use bi_next if this | |
1715 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1716 | * a lower device by calling into generic_make_request recursively, which | |
1717 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1718 | */ |
1719 | void generic_make_request(struct bio *bio) | |
1720 | { | |
bddd87c7 AM |
1721 | struct bio_list bio_list_on_stack; |
1722 | ||
27a84d54 CH |
1723 | if (!generic_make_request_checks(bio)) |
1724 | return; | |
1725 | ||
1726 | /* | |
1727 | * We only want one ->make_request_fn to be active at a time, else | |
1728 | * stack usage with stacked devices could be a problem. So use | |
1729 | * current->bio_list to keep a list of requests submited by a | |
1730 | * make_request_fn function. current->bio_list is also used as a | |
1731 | * flag to say if generic_make_request is currently active in this | |
1732 | * task or not. If it is NULL, then no make_request is active. If | |
1733 | * it is non-NULL, then a make_request is active, and new requests | |
1734 | * should be added at the tail | |
1735 | */ | |
bddd87c7 | 1736 | if (current->bio_list) { |
bddd87c7 | 1737 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1738 | return; |
1739 | } | |
27a84d54 | 1740 | |
d89d8796 NB |
1741 | /* following loop may be a bit non-obvious, and so deserves some |
1742 | * explanation. | |
1743 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1744 | * ensure that) so we have a list with a single bio. | |
1745 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1746 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1747 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1748 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1749 | * through a recursive call to generic_make_request. If it |
1750 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1751 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1752 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1753 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1754 | */ |
1755 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1756 | bio_list_init(&bio_list_on_stack); |
1757 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1758 | do { |
27a84d54 CH |
1759 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1760 | ||
1761 | q->make_request_fn(q, bio); | |
1762 | ||
bddd87c7 | 1763 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1764 | } while (bio); |
bddd87c7 | 1765 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1766 | } |
1da177e4 LT |
1767 | EXPORT_SYMBOL(generic_make_request); |
1768 | ||
1769 | /** | |
710027a4 | 1770 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1771 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1772 | * @bio: The &struct bio which describes the I/O | |
1773 | * | |
1774 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1775 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1776 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1777 | * |
1778 | */ | |
1779 | void submit_bio(int rw, struct bio *bio) | |
1780 | { | |
1781 | int count = bio_sectors(bio); | |
1782 | ||
22e2c507 | 1783 | bio->bi_rw |= rw; |
1da177e4 | 1784 | |
bf2de6f5 JA |
1785 | /* |
1786 | * If it's a regular read/write or a barrier with data attached, | |
1787 | * go through the normal accounting stuff before submission. | |
1788 | */ | |
3ffb52e7 | 1789 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1790 | if (rw & WRITE) { |
1791 | count_vm_events(PGPGOUT, count); | |
1792 | } else { | |
1793 | task_io_account_read(bio->bi_size); | |
1794 | count_vm_events(PGPGIN, count); | |
1795 | } | |
1796 | ||
1797 | if (unlikely(block_dump)) { | |
1798 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1799 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1800 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1801 | (rw & WRITE) ? "WRITE" : "READ", |
1802 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1803 | bdevname(bio->bi_bdev, b), |
1804 | count); | |
bf2de6f5 | 1805 | } |
1da177e4 LT |
1806 | } |
1807 | ||
1808 | generic_make_request(bio); | |
1809 | } | |
1da177e4 LT |
1810 | EXPORT_SYMBOL(submit_bio); |
1811 | ||
82124d60 KU |
1812 | /** |
1813 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1814 | * @q: the queue | |
1815 | * @rq: the request being checked | |
1816 | * | |
1817 | * Description: | |
1818 | * @rq may have been made based on weaker limitations of upper-level queues | |
1819 | * in request stacking drivers, and it may violate the limitation of @q. | |
1820 | * Since the block layer and the underlying device driver trust @rq | |
1821 | * after it is inserted to @q, it should be checked against @q before | |
1822 | * the insertion using this generic function. | |
1823 | * | |
1824 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1825 | * in some cases below, so export this function. |
82124d60 KU |
1826 | * Request stacking drivers like request-based dm may change the queue |
1827 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1828 | * Such request stacking drivers should check those requests agaist | |
1829 | * the new queue limits again when they dispatch those requests, | |
1830 | * although such checkings are also done against the old queue limits | |
1831 | * when submitting requests. | |
1832 | */ | |
1833 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1834 | { | |
3383977f S |
1835 | if (rq->cmd_flags & REQ_DISCARD) |
1836 | return 0; | |
1837 | ||
ae03bf63 MP |
1838 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1839 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1840 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1841 | return -EIO; | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * queue's settings related to segment counting like q->bounce_pfn | |
1846 | * may differ from that of other stacking queues. | |
1847 | * Recalculate it to check the request correctly on this queue's | |
1848 | * limitation. | |
1849 | */ | |
1850 | blk_recalc_rq_segments(rq); | |
8a78362c | 1851 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1852 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1853 | return -EIO; | |
1854 | } | |
1855 | ||
1856 | return 0; | |
1857 | } | |
1858 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1859 | ||
1860 | /** | |
1861 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1862 | * @q: the queue to submit the request | |
1863 | * @rq: the request being queued | |
1864 | */ | |
1865 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1866 | { | |
1867 | unsigned long flags; | |
4853abaa | 1868 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
1869 | |
1870 | if (blk_rq_check_limits(q, rq)) | |
1871 | return -EIO; | |
1872 | ||
b2c9cd37 AM |
1873 | if (rq->rq_disk && |
1874 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 1875 | return -EIO; |
82124d60 KU |
1876 | |
1877 | spin_lock_irqsave(q->queue_lock, flags); | |
8ba61435 TH |
1878 | if (unlikely(blk_queue_dead(q))) { |
1879 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1880 | return -ENODEV; | |
1881 | } | |
82124d60 KU |
1882 | |
1883 | /* | |
1884 | * Submitting request must be dequeued before calling this function | |
1885 | * because it will be linked to another request_queue | |
1886 | */ | |
1887 | BUG_ON(blk_queued_rq(rq)); | |
1888 | ||
4853abaa JM |
1889 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
1890 | where = ELEVATOR_INSERT_FLUSH; | |
1891 | ||
1892 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
1893 | if (where == ELEVATOR_INSERT_FLUSH) |
1894 | __blk_run_queue(q); | |
82124d60 KU |
1895 | spin_unlock_irqrestore(q->queue_lock, flags); |
1896 | ||
1897 | return 0; | |
1898 | } | |
1899 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1900 | ||
80a761fd TH |
1901 | /** |
1902 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1903 | * @rq: request to examine | |
1904 | * | |
1905 | * Description: | |
1906 | * A request could be merge of IOs which require different failure | |
1907 | * handling. This function determines the number of bytes which | |
1908 | * can be failed from the beginning of the request without | |
1909 | * crossing into area which need to be retried further. | |
1910 | * | |
1911 | * Return: | |
1912 | * The number of bytes to fail. | |
1913 | * | |
1914 | * Context: | |
1915 | * queue_lock must be held. | |
1916 | */ | |
1917 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1918 | { | |
1919 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1920 | unsigned int bytes = 0; | |
1921 | struct bio *bio; | |
1922 | ||
1923 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1924 | return blk_rq_bytes(rq); | |
1925 | ||
1926 | /* | |
1927 | * Currently the only 'mixing' which can happen is between | |
1928 | * different fastfail types. We can safely fail portions | |
1929 | * which have all the failfast bits that the first one has - | |
1930 | * the ones which are at least as eager to fail as the first | |
1931 | * one. | |
1932 | */ | |
1933 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1934 | if ((bio->bi_rw & ff) != ff) | |
1935 | break; | |
1936 | bytes += bio->bi_size; | |
1937 | } | |
1938 | ||
1939 | /* this could lead to infinite loop */ | |
1940 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1941 | return bytes; | |
1942 | } | |
1943 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1944 | ||
bc58ba94 JA |
1945 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1946 | { | |
c2553b58 | 1947 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1948 | const int rw = rq_data_dir(req); |
1949 | struct hd_struct *part; | |
1950 | int cpu; | |
1951 | ||
1952 | cpu = part_stat_lock(); | |
09e099d4 | 1953 | part = req->part; |
bc58ba94 JA |
1954 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1955 | part_stat_unlock(); | |
1956 | } | |
1957 | } | |
1958 | ||
1959 | static void blk_account_io_done(struct request *req) | |
1960 | { | |
bc58ba94 | 1961 | /* |
dd4c133f TH |
1962 | * Account IO completion. flush_rq isn't accounted as a |
1963 | * normal IO on queueing nor completion. Accounting the | |
1964 | * containing request is enough. | |
bc58ba94 | 1965 | */ |
414b4ff5 | 1966 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1967 | unsigned long duration = jiffies - req->start_time; |
1968 | const int rw = rq_data_dir(req); | |
1969 | struct hd_struct *part; | |
1970 | int cpu; | |
1971 | ||
1972 | cpu = part_stat_lock(); | |
09e099d4 | 1973 | part = req->part; |
bc58ba94 JA |
1974 | |
1975 | part_stat_inc(cpu, part, ios[rw]); | |
1976 | part_stat_add(cpu, part, ticks[rw], duration); | |
1977 | part_round_stats(cpu, part); | |
316d315b | 1978 | part_dec_in_flight(part, rw); |
bc58ba94 | 1979 | |
6c23a968 | 1980 | hd_struct_put(part); |
bc58ba94 JA |
1981 | part_stat_unlock(); |
1982 | } | |
1983 | } | |
1984 | ||
3bcddeac | 1985 | /** |
9934c8c0 TH |
1986 | * blk_peek_request - peek at the top of a request queue |
1987 | * @q: request queue to peek at | |
1988 | * | |
1989 | * Description: | |
1990 | * Return the request at the top of @q. The returned request | |
1991 | * should be started using blk_start_request() before LLD starts | |
1992 | * processing it. | |
1993 | * | |
1994 | * Return: | |
1995 | * Pointer to the request at the top of @q if available. Null | |
1996 | * otherwise. | |
1997 | * | |
1998 | * Context: | |
1999 | * queue_lock must be held. | |
2000 | */ | |
2001 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2002 | { |
2003 | struct request *rq; | |
2004 | int ret; | |
2005 | ||
2006 | while ((rq = __elv_next_request(q)) != NULL) { | |
2007 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
2008 | /* | |
2009 | * This is the first time the device driver | |
2010 | * sees this request (possibly after | |
2011 | * requeueing). Notify IO scheduler. | |
2012 | */ | |
33659ebb | 2013 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
2014 | elv_activate_rq(q, rq); |
2015 | ||
2016 | /* | |
2017 | * just mark as started even if we don't start | |
2018 | * it, a request that has been delayed should | |
2019 | * not be passed by new incoming requests | |
2020 | */ | |
2021 | rq->cmd_flags |= REQ_STARTED; | |
2022 | trace_block_rq_issue(q, rq); | |
2023 | } | |
2024 | ||
2025 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2026 | q->end_sector = rq_end_sector(rq); | |
2027 | q->boundary_rq = NULL; | |
2028 | } | |
2029 | ||
2030 | if (rq->cmd_flags & REQ_DONTPREP) | |
2031 | break; | |
2032 | ||
2e46e8b2 | 2033 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2034 | /* |
2035 | * make sure space for the drain appears we | |
2036 | * know we can do this because max_hw_segments | |
2037 | * has been adjusted to be one fewer than the | |
2038 | * device can handle | |
2039 | */ | |
2040 | rq->nr_phys_segments++; | |
2041 | } | |
2042 | ||
2043 | if (!q->prep_rq_fn) | |
2044 | break; | |
2045 | ||
2046 | ret = q->prep_rq_fn(q, rq); | |
2047 | if (ret == BLKPREP_OK) { | |
2048 | break; | |
2049 | } else if (ret == BLKPREP_DEFER) { | |
2050 | /* | |
2051 | * the request may have been (partially) prepped. | |
2052 | * we need to keep this request in the front to | |
2053 | * avoid resource deadlock. REQ_STARTED will | |
2054 | * prevent other fs requests from passing this one. | |
2055 | */ | |
2e46e8b2 | 2056 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
2057 | !(rq->cmd_flags & REQ_DONTPREP)) { |
2058 | /* | |
2059 | * remove the space for the drain we added | |
2060 | * so that we don't add it again | |
2061 | */ | |
2062 | --rq->nr_phys_segments; | |
2063 | } | |
2064 | ||
2065 | rq = NULL; | |
2066 | break; | |
2067 | } else if (ret == BLKPREP_KILL) { | |
2068 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
2069 | /* |
2070 | * Mark this request as started so we don't trigger | |
2071 | * any debug logic in the end I/O path. | |
2072 | */ | |
2073 | blk_start_request(rq); | |
40cbbb78 | 2074 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
2075 | } else { |
2076 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2077 | break; | |
2078 | } | |
2079 | } | |
2080 | ||
2081 | return rq; | |
2082 | } | |
9934c8c0 | 2083 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2084 | |
9934c8c0 | 2085 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2086 | { |
9934c8c0 TH |
2087 | struct request_queue *q = rq->q; |
2088 | ||
158dbda0 TH |
2089 | BUG_ON(list_empty(&rq->queuelist)); |
2090 | BUG_ON(ELV_ON_HASH(rq)); | |
2091 | ||
2092 | list_del_init(&rq->queuelist); | |
2093 | ||
2094 | /* | |
2095 | * the time frame between a request being removed from the lists | |
2096 | * and to it is freed is accounted as io that is in progress at | |
2097 | * the driver side. | |
2098 | */ | |
9195291e | 2099 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2100 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2101 | set_io_start_time_ns(rq); |
2102 | } | |
158dbda0 TH |
2103 | } |
2104 | ||
9934c8c0 TH |
2105 | /** |
2106 | * blk_start_request - start request processing on the driver | |
2107 | * @req: request to dequeue | |
2108 | * | |
2109 | * Description: | |
2110 | * Dequeue @req and start timeout timer on it. This hands off the | |
2111 | * request to the driver. | |
2112 | * | |
2113 | * Block internal functions which don't want to start timer should | |
2114 | * call blk_dequeue_request(). | |
2115 | * | |
2116 | * Context: | |
2117 | * queue_lock must be held. | |
2118 | */ | |
2119 | void blk_start_request(struct request *req) | |
2120 | { | |
2121 | blk_dequeue_request(req); | |
2122 | ||
2123 | /* | |
5f49f631 TH |
2124 | * We are now handing the request to the hardware, initialize |
2125 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2126 | */ |
5f49f631 | 2127 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2128 | if (unlikely(blk_bidi_rq(req))) |
2129 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2130 | ||
9934c8c0 TH |
2131 | blk_add_timer(req); |
2132 | } | |
2133 | EXPORT_SYMBOL(blk_start_request); | |
2134 | ||
2135 | /** | |
2136 | * blk_fetch_request - fetch a request from a request queue | |
2137 | * @q: request queue to fetch a request from | |
2138 | * | |
2139 | * Description: | |
2140 | * Return the request at the top of @q. The request is started on | |
2141 | * return and LLD can start processing it immediately. | |
2142 | * | |
2143 | * Return: | |
2144 | * Pointer to the request at the top of @q if available. Null | |
2145 | * otherwise. | |
2146 | * | |
2147 | * Context: | |
2148 | * queue_lock must be held. | |
2149 | */ | |
2150 | struct request *blk_fetch_request(struct request_queue *q) | |
2151 | { | |
2152 | struct request *rq; | |
2153 | ||
2154 | rq = blk_peek_request(q); | |
2155 | if (rq) | |
2156 | blk_start_request(rq); | |
2157 | return rq; | |
2158 | } | |
2159 | EXPORT_SYMBOL(blk_fetch_request); | |
2160 | ||
3bcddeac | 2161 | /** |
2e60e022 | 2162 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2163 | * @req: the request being processed |
710027a4 | 2164 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2165 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2166 | * |
2167 | * Description: | |
8ebf9756 RD |
2168 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2169 | * the request structure even if @req doesn't have leftover. | |
2170 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2171 | * |
2172 | * This special helper function is only for request stacking drivers | |
2173 | * (e.g. request-based dm) so that they can handle partial completion. | |
2174 | * Actual device drivers should use blk_end_request instead. | |
2175 | * | |
2176 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2177 | * %false return from this function. | |
3bcddeac KU |
2178 | * |
2179 | * Return: | |
2e60e022 TH |
2180 | * %false - this request doesn't have any more data |
2181 | * %true - this request has more data | |
3bcddeac | 2182 | **/ |
2e60e022 | 2183 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2184 | { |
5450d3e1 | 2185 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2186 | struct bio *bio; |
2187 | ||
2e60e022 TH |
2188 | if (!req->bio) |
2189 | return false; | |
2190 | ||
5f3ea37c | 2191 | trace_block_rq_complete(req->q, req); |
2056a782 | 2192 | |
1da177e4 | 2193 | /* |
6f41469c TH |
2194 | * For fs requests, rq is just carrier of independent bio's |
2195 | * and each partial completion should be handled separately. | |
2196 | * Reset per-request error on each partial completion. | |
2197 | * | |
2198 | * TODO: tj: This is too subtle. It would be better to let | |
2199 | * low level drivers do what they see fit. | |
1da177e4 | 2200 | */ |
33659ebb | 2201 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2202 | req->errors = 0; |
2203 | ||
33659ebb CH |
2204 | if (error && req->cmd_type == REQ_TYPE_FS && |
2205 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2206 | char *error_type; |
2207 | ||
2208 | switch (error) { | |
2209 | case -ENOLINK: | |
2210 | error_type = "recoverable transport"; | |
2211 | break; | |
2212 | case -EREMOTEIO: | |
2213 | error_type = "critical target"; | |
2214 | break; | |
2215 | case -EBADE: | |
2216 | error_type = "critical nexus"; | |
2217 | break; | |
2218 | case -EIO: | |
2219 | default: | |
2220 | error_type = "I/O"; | |
2221 | break; | |
2222 | } | |
2223 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2224 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2225 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2226 | } |
2227 | ||
bc58ba94 | 2228 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2229 | |
1da177e4 LT |
2230 | total_bytes = bio_nbytes = 0; |
2231 | while ((bio = req->bio) != NULL) { | |
2232 | int nbytes; | |
2233 | ||
2234 | if (nr_bytes >= bio->bi_size) { | |
2235 | req->bio = bio->bi_next; | |
2236 | nbytes = bio->bi_size; | |
5bb23a68 | 2237 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2238 | next_idx = 0; |
2239 | bio_nbytes = 0; | |
2240 | } else { | |
2241 | int idx = bio->bi_idx + next_idx; | |
2242 | ||
af498d7f | 2243 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2244 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2245 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2246 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2247 | break; |
2248 | } | |
2249 | ||
2250 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2251 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2252 | ||
2253 | /* | |
2254 | * not a complete bvec done | |
2255 | */ | |
2256 | if (unlikely(nbytes > nr_bytes)) { | |
2257 | bio_nbytes += nr_bytes; | |
2258 | total_bytes += nr_bytes; | |
2259 | break; | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * advance to the next vector | |
2264 | */ | |
2265 | next_idx++; | |
2266 | bio_nbytes += nbytes; | |
2267 | } | |
2268 | ||
2269 | total_bytes += nbytes; | |
2270 | nr_bytes -= nbytes; | |
2271 | ||
6728cb0e JA |
2272 | bio = req->bio; |
2273 | if (bio) { | |
1da177e4 LT |
2274 | /* |
2275 | * end more in this run, or just return 'not-done' | |
2276 | */ | |
2277 | if (unlikely(nr_bytes <= 0)) | |
2278 | break; | |
2279 | } | |
2280 | } | |
2281 | ||
2282 | /* | |
2283 | * completely done | |
2284 | */ | |
2e60e022 TH |
2285 | if (!req->bio) { |
2286 | /* | |
2287 | * Reset counters so that the request stacking driver | |
2288 | * can find how many bytes remain in the request | |
2289 | * later. | |
2290 | */ | |
a2dec7b3 | 2291 | req->__data_len = 0; |
2e60e022 TH |
2292 | return false; |
2293 | } | |
1da177e4 LT |
2294 | |
2295 | /* | |
2296 | * if the request wasn't completed, update state | |
2297 | */ | |
2298 | if (bio_nbytes) { | |
5bb23a68 | 2299 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2300 | bio->bi_idx += next_idx; |
2301 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2302 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2303 | } | |
2304 | ||
a2dec7b3 | 2305 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2306 | req->buffer = bio_data(req->bio); |
2307 | ||
2308 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2309 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2310 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2311 | |
80a761fd TH |
2312 | /* mixed attributes always follow the first bio */ |
2313 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2314 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2315 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2316 | } | |
2317 | ||
2e46e8b2 TH |
2318 | /* |
2319 | * If total number of sectors is less than the first segment | |
2320 | * size, something has gone terribly wrong. | |
2321 | */ | |
2322 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2323 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2324 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2325 | } |
2326 | ||
2327 | /* recalculate the number of segments */ | |
1da177e4 | 2328 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2329 | |
2e60e022 | 2330 | return true; |
1da177e4 | 2331 | } |
2e60e022 | 2332 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2333 | |
2e60e022 TH |
2334 | static bool blk_update_bidi_request(struct request *rq, int error, |
2335 | unsigned int nr_bytes, | |
2336 | unsigned int bidi_bytes) | |
5efccd17 | 2337 | { |
2e60e022 TH |
2338 | if (blk_update_request(rq, error, nr_bytes)) |
2339 | return true; | |
5efccd17 | 2340 | |
2e60e022 TH |
2341 | /* Bidi request must be completed as a whole */ |
2342 | if (unlikely(blk_bidi_rq(rq)) && | |
2343 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2344 | return true; | |
5efccd17 | 2345 | |
e2e1a148 JA |
2346 | if (blk_queue_add_random(rq->q)) |
2347 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2348 | |
2349 | return false; | |
1da177e4 LT |
2350 | } |
2351 | ||
28018c24 JB |
2352 | /** |
2353 | * blk_unprep_request - unprepare a request | |
2354 | * @req: the request | |
2355 | * | |
2356 | * This function makes a request ready for complete resubmission (or | |
2357 | * completion). It happens only after all error handling is complete, | |
2358 | * so represents the appropriate moment to deallocate any resources | |
2359 | * that were allocated to the request in the prep_rq_fn. The queue | |
2360 | * lock is held when calling this. | |
2361 | */ | |
2362 | void blk_unprep_request(struct request *req) | |
2363 | { | |
2364 | struct request_queue *q = req->q; | |
2365 | ||
2366 | req->cmd_flags &= ~REQ_DONTPREP; | |
2367 | if (q->unprep_rq_fn) | |
2368 | q->unprep_rq_fn(q, req); | |
2369 | } | |
2370 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2371 | ||
1da177e4 LT |
2372 | /* |
2373 | * queue lock must be held | |
2374 | */ | |
2e60e022 | 2375 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2376 | { |
b8286239 KU |
2377 | if (blk_rq_tagged(req)) |
2378 | blk_queue_end_tag(req->q, req); | |
2379 | ||
ba396a6c | 2380 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2381 | |
33659ebb | 2382 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2383 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2384 | |
e78042e5 MA |
2385 | blk_delete_timer(req); |
2386 | ||
28018c24 JB |
2387 | if (req->cmd_flags & REQ_DONTPREP) |
2388 | blk_unprep_request(req); | |
2389 | ||
2390 | ||
bc58ba94 | 2391 | blk_account_io_done(req); |
b8286239 | 2392 | |
1da177e4 | 2393 | if (req->end_io) |
8ffdc655 | 2394 | req->end_io(req, error); |
b8286239 KU |
2395 | else { |
2396 | if (blk_bidi_rq(req)) | |
2397 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2398 | ||
1da177e4 | 2399 | __blk_put_request(req->q, req); |
b8286239 | 2400 | } |
1da177e4 LT |
2401 | } |
2402 | ||
3b11313a | 2403 | /** |
2e60e022 TH |
2404 | * blk_end_bidi_request - Complete a bidi request |
2405 | * @rq: the request to complete | |
2406 | * @error: %0 for success, < %0 for error | |
2407 | * @nr_bytes: number of bytes to complete @rq | |
2408 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2409 | * |
2410 | * Description: | |
e3a04fe3 | 2411 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2412 | * Drivers that supports bidi can safely call this member for any |
2413 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2414 | * just ignored. | |
336cdb40 KU |
2415 | * |
2416 | * Return: | |
2e60e022 TH |
2417 | * %false - we are done with this request |
2418 | * %true - still buffers pending for this request | |
a0cd1285 | 2419 | **/ |
b1f74493 | 2420 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2421 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2422 | { | |
336cdb40 | 2423 | struct request_queue *q = rq->q; |
2e60e022 | 2424 | unsigned long flags; |
32fab448 | 2425 | |
2e60e022 TH |
2426 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2427 | return true; | |
32fab448 | 2428 | |
336cdb40 | 2429 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2430 | blk_finish_request(rq, error); |
336cdb40 KU |
2431 | spin_unlock_irqrestore(q->queue_lock, flags); |
2432 | ||
2e60e022 | 2433 | return false; |
32fab448 KU |
2434 | } |
2435 | ||
336cdb40 | 2436 | /** |
2e60e022 TH |
2437 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2438 | * @rq: the request to complete | |
710027a4 | 2439 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2440 | * @nr_bytes: number of bytes to complete @rq |
2441 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2442 | * |
2443 | * Description: | |
2e60e022 TH |
2444 | * Identical to blk_end_bidi_request() except that queue lock is |
2445 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2446 | * |
2447 | * Return: | |
2e60e022 TH |
2448 | * %false - we are done with this request |
2449 | * %true - still buffers pending for this request | |
336cdb40 | 2450 | **/ |
4853abaa | 2451 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2452 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2453 | { |
2e60e022 TH |
2454 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2455 | return true; | |
336cdb40 | 2456 | |
2e60e022 | 2457 | blk_finish_request(rq, error); |
336cdb40 | 2458 | |
2e60e022 | 2459 | return false; |
336cdb40 | 2460 | } |
e19a3ab0 KU |
2461 | |
2462 | /** | |
2463 | * blk_end_request - Helper function for drivers to complete the request. | |
2464 | * @rq: the request being processed | |
710027a4 | 2465 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2466 | * @nr_bytes: number of bytes to complete |
2467 | * | |
2468 | * Description: | |
2469 | * Ends I/O on a number of bytes attached to @rq. | |
2470 | * If @rq has leftover, sets it up for the next range of segments. | |
2471 | * | |
2472 | * Return: | |
b1f74493 FT |
2473 | * %false - we are done with this request |
2474 | * %true - still buffers pending for this request | |
e19a3ab0 | 2475 | **/ |
b1f74493 | 2476 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2477 | { |
b1f74493 | 2478 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2479 | } |
56ad1740 | 2480 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2481 | |
2482 | /** | |
b1f74493 FT |
2483 | * blk_end_request_all - Helper function for drives to finish the request. |
2484 | * @rq: the request to finish | |
8ebf9756 | 2485 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2486 | * |
2487 | * Description: | |
b1f74493 FT |
2488 | * Completely finish @rq. |
2489 | */ | |
2490 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2491 | { |
b1f74493 FT |
2492 | bool pending; |
2493 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2494 | |
b1f74493 FT |
2495 | if (unlikely(blk_bidi_rq(rq))) |
2496 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2497 | |
b1f74493 FT |
2498 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2499 | BUG_ON(pending); | |
2500 | } | |
56ad1740 | 2501 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2502 | |
b1f74493 FT |
2503 | /** |
2504 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2505 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2506 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2507 | * |
2508 | * Description: | |
2509 | * Complete the current consecutively mapped chunk from @rq. | |
2510 | * | |
2511 | * Return: | |
2512 | * %false - we are done with this request | |
2513 | * %true - still buffers pending for this request | |
2514 | */ | |
2515 | bool blk_end_request_cur(struct request *rq, int error) | |
2516 | { | |
2517 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2518 | } |
56ad1740 | 2519 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2520 | |
80a761fd TH |
2521 | /** |
2522 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2523 | * @rq: the request to finish till the next failure boundary for | |
2524 | * @error: must be negative errno | |
2525 | * | |
2526 | * Description: | |
2527 | * Complete @rq till the next failure boundary. | |
2528 | * | |
2529 | * Return: | |
2530 | * %false - we are done with this request | |
2531 | * %true - still buffers pending for this request | |
2532 | */ | |
2533 | bool blk_end_request_err(struct request *rq, int error) | |
2534 | { | |
2535 | WARN_ON(error >= 0); | |
2536 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2537 | } | |
2538 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2539 | ||
e3a04fe3 | 2540 | /** |
b1f74493 FT |
2541 | * __blk_end_request - Helper function for drivers to complete the request. |
2542 | * @rq: the request being processed | |
2543 | * @error: %0 for success, < %0 for error | |
2544 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2545 | * |
2546 | * Description: | |
b1f74493 | 2547 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2548 | * |
2549 | * Return: | |
b1f74493 FT |
2550 | * %false - we are done with this request |
2551 | * %true - still buffers pending for this request | |
e3a04fe3 | 2552 | **/ |
b1f74493 | 2553 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2554 | { |
b1f74493 | 2555 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2556 | } |
56ad1740 | 2557 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2558 | |
32fab448 | 2559 | /** |
b1f74493 FT |
2560 | * __blk_end_request_all - Helper function for drives to finish the request. |
2561 | * @rq: the request to finish | |
8ebf9756 | 2562 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2563 | * |
2564 | * Description: | |
b1f74493 | 2565 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2566 | */ |
b1f74493 | 2567 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2568 | { |
b1f74493 FT |
2569 | bool pending; |
2570 | unsigned int bidi_bytes = 0; | |
2571 | ||
2572 | if (unlikely(blk_bidi_rq(rq))) | |
2573 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2574 | ||
2575 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2576 | BUG_ON(pending); | |
32fab448 | 2577 | } |
56ad1740 | 2578 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2579 | |
e19a3ab0 | 2580 | /** |
b1f74493 FT |
2581 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2582 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2583 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2584 | * |
2585 | * Description: | |
b1f74493 FT |
2586 | * Complete the current consecutively mapped chunk from @rq. Must |
2587 | * be called with queue lock held. | |
e19a3ab0 KU |
2588 | * |
2589 | * Return: | |
b1f74493 FT |
2590 | * %false - we are done with this request |
2591 | * %true - still buffers pending for this request | |
2592 | */ | |
2593 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2594 | { |
b1f74493 | 2595 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2596 | } |
56ad1740 | 2597 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2598 | |
80a761fd TH |
2599 | /** |
2600 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2601 | * @rq: the request to finish till the next failure boundary for | |
2602 | * @error: must be negative errno | |
2603 | * | |
2604 | * Description: | |
2605 | * Complete @rq till the next failure boundary. Must be called | |
2606 | * with queue lock held. | |
2607 | * | |
2608 | * Return: | |
2609 | * %false - we are done with this request | |
2610 | * %true - still buffers pending for this request | |
2611 | */ | |
2612 | bool __blk_end_request_err(struct request *rq, int error) | |
2613 | { | |
2614 | WARN_ON(error >= 0); | |
2615 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2616 | } | |
2617 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2618 | ||
86db1e29 JA |
2619 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2620 | struct bio *bio) | |
1da177e4 | 2621 | { |
a82afdfc | 2622 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2623 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2624 | |
fb2dce86 DW |
2625 | if (bio_has_data(bio)) { |
2626 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2627 | rq->buffer = bio_data(bio); |
2628 | } | |
a2dec7b3 | 2629 | rq->__data_len = bio->bi_size; |
1da177e4 | 2630 | rq->bio = rq->biotail = bio; |
1da177e4 | 2631 | |
66846572 N |
2632 | if (bio->bi_bdev) |
2633 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2634 | } | |
1da177e4 | 2635 | |
2d4dc890 IL |
2636 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2637 | /** | |
2638 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2639 | * @rq: the request to be flushed | |
2640 | * | |
2641 | * Description: | |
2642 | * Flush all pages in @rq. | |
2643 | */ | |
2644 | void rq_flush_dcache_pages(struct request *rq) | |
2645 | { | |
2646 | struct req_iterator iter; | |
2647 | struct bio_vec *bvec; | |
2648 | ||
2649 | rq_for_each_segment(bvec, rq, iter) | |
2650 | flush_dcache_page(bvec->bv_page); | |
2651 | } | |
2652 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2653 | #endif | |
2654 | ||
ef9e3fac KU |
2655 | /** |
2656 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2657 | * @q : the queue of the device being checked | |
2658 | * | |
2659 | * Description: | |
2660 | * Check if underlying low-level drivers of a device are busy. | |
2661 | * If the drivers want to export their busy state, they must set own | |
2662 | * exporting function using blk_queue_lld_busy() first. | |
2663 | * | |
2664 | * Basically, this function is used only by request stacking drivers | |
2665 | * to stop dispatching requests to underlying devices when underlying | |
2666 | * devices are busy. This behavior helps more I/O merging on the queue | |
2667 | * of the request stacking driver and prevents I/O throughput regression | |
2668 | * on burst I/O load. | |
2669 | * | |
2670 | * Return: | |
2671 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2672 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2673 | */ | |
2674 | int blk_lld_busy(struct request_queue *q) | |
2675 | { | |
2676 | if (q->lld_busy_fn) | |
2677 | return q->lld_busy_fn(q); | |
2678 | ||
2679 | return 0; | |
2680 | } | |
2681 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2682 | ||
b0fd271d KU |
2683 | /** |
2684 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2685 | * @rq: the clone request to be cleaned up | |
2686 | * | |
2687 | * Description: | |
2688 | * Free all bios in @rq for a cloned request. | |
2689 | */ | |
2690 | void blk_rq_unprep_clone(struct request *rq) | |
2691 | { | |
2692 | struct bio *bio; | |
2693 | ||
2694 | while ((bio = rq->bio) != NULL) { | |
2695 | rq->bio = bio->bi_next; | |
2696 | ||
2697 | bio_put(bio); | |
2698 | } | |
2699 | } | |
2700 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2701 | ||
2702 | /* | |
2703 | * Copy attributes of the original request to the clone request. | |
2704 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2705 | */ | |
2706 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2707 | { | |
2708 | dst->cpu = src->cpu; | |
3a2edd0d | 2709 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2710 | dst->cmd_type = src->cmd_type; |
2711 | dst->__sector = blk_rq_pos(src); | |
2712 | dst->__data_len = blk_rq_bytes(src); | |
2713 | dst->nr_phys_segments = src->nr_phys_segments; | |
2714 | dst->ioprio = src->ioprio; | |
2715 | dst->extra_len = src->extra_len; | |
2716 | } | |
2717 | ||
2718 | /** | |
2719 | * blk_rq_prep_clone - Helper function to setup clone request | |
2720 | * @rq: the request to be setup | |
2721 | * @rq_src: original request to be cloned | |
2722 | * @bs: bio_set that bios for clone are allocated from | |
2723 | * @gfp_mask: memory allocation mask for bio | |
2724 | * @bio_ctr: setup function to be called for each clone bio. | |
2725 | * Returns %0 for success, non %0 for failure. | |
2726 | * @data: private data to be passed to @bio_ctr | |
2727 | * | |
2728 | * Description: | |
2729 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2730 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2731 | * are not copied, and copying such parts is the caller's responsibility. | |
2732 | * Also, pages which the original bios are pointing to are not copied | |
2733 | * and the cloned bios just point same pages. | |
2734 | * So cloned bios must be completed before original bios, which means | |
2735 | * the caller must complete @rq before @rq_src. | |
2736 | */ | |
2737 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2738 | struct bio_set *bs, gfp_t gfp_mask, | |
2739 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2740 | void *data) | |
2741 | { | |
2742 | struct bio *bio, *bio_src; | |
2743 | ||
2744 | if (!bs) | |
2745 | bs = fs_bio_set; | |
2746 | ||
2747 | blk_rq_init(NULL, rq); | |
2748 | ||
2749 | __rq_for_each_bio(bio_src, rq_src) { | |
2750 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2751 | if (!bio) | |
2752 | goto free_and_out; | |
2753 | ||
2754 | __bio_clone(bio, bio_src); | |
2755 | ||
2756 | if (bio_integrity(bio_src) && | |
7878cba9 | 2757 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2758 | goto free_and_out; |
2759 | ||
2760 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2761 | goto free_and_out; | |
2762 | ||
2763 | if (rq->bio) { | |
2764 | rq->biotail->bi_next = bio; | |
2765 | rq->biotail = bio; | |
2766 | } else | |
2767 | rq->bio = rq->biotail = bio; | |
2768 | } | |
2769 | ||
2770 | __blk_rq_prep_clone(rq, rq_src); | |
2771 | ||
2772 | return 0; | |
2773 | ||
2774 | free_and_out: | |
2775 | if (bio) | |
2776 | bio_free(bio, bs); | |
2777 | blk_rq_unprep_clone(rq); | |
2778 | ||
2779 | return -ENOMEM; | |
2780 | } | |
2781 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2782 | ||
18887ad9 | 2783 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2784 | { |
2785 | return queue_work(kblockd_workqueue, work); | |
2786 | } | |
1da177e4 LT |
2787 | EXPORT_SYMBOL(kblockd_schedule_work); |
2788 | ||
e43473b7 VG |
2789 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2790 | struct delayed_work *dwork, unsigned long delay) | |
2791 | { | |
2792 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2793 | } | |
2794 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2795 | ||
73c10101 JA |
2796 | #define PLUG_MAGIC 0x91827364 |
2797 | ||
75df7136 SJ |
2798 | /** |
2799 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2800 | * @plug: The &struct blk_plug that needs to be initialized | |
2801 | * | |
2802 | * Description: | |
2803 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2804 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2805 | * blk_finish_plug(). This is important from a performance perspective, but | |
2806 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2807 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2808 | * page belonging to that request that is currently residing in our private | |
2809 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2810 | * this kind of deadlock. | |
2811 | */ | |
73c10101 JA |
2812 | void blk_start_plug(struct blk_plug *plug) |
2813 | { | |
2814 | struct task_struct *tsk = current; | |
2815 | ||
2816 | plug->magic = PLUG_MAGIC; | |
2817 | INIT_LIST_HEAD(&plug->list); | |
048c9374 | 2818 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 JA |
2819 | plug->should_sort = 0; |
2820 | ||
2821 | /* | |
2822 | * If this is a nested plug, don't actually assign it. It will be | |
2823 | * flushed on its own. | |
2824 | */ | |
2825 | if (!tsk->plug) { | |
2826 | /* | |
2827 | * Store ordering should not be needed here, since a potential | |
2828 | * preempt will imply a full memory barrier | |
2829 | */ | |
2830 | tsk->plug = plug; | |
2831 | } | |
2832 | } | |
2833 | EXPORT_SYMBOL(blk_start_plug); | |
2834 | ||
2835 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2836 | { | |
2837 | struct request *rqa = container_of(a, struct request, queuelist); | |
2838 | struct request *rqb = container_of(b, struct request, queuelist); | |
2839 | ||
f83e8261 | 2840 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2841 | } |
2842 | ||
49cac01e JA |
2843 | /* |
2844 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2845 | * until a safe kblockd context. We due this to avoid accidental big | |
2846 | * additional stack usage in driver dispatch, in places where the originally | |
2847 | * plugger did not intend it. | |
2848 | */ | |
f6603783 | 2849 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 2850 | bool from_schedule) |
99e22598 | 2851 | __releases(q->queue_lock) |
94b5eb28 | 2852 | { |
49cac01e | 2853 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 2854 | |
8ba61435 TH |
2855 | /* |
2856 | * Don't mess with dead queue. | |
2857 | */ | |
2858 | if (unlikely(blk_queue_dead(q))) { | |
2859 | spin_unlock(q->queue_lock); | |
2860 | return; | |
2861 | } | |
2862 | ||
99e22598 JA |
2863 | /* |
2864 | * If we are punting this to kblockd, then we can safely drop | |
2865 | * the queue_lock before waking kblockd (which needs to take | |
2866 | * this lock). | |
2867 | */ | |
2868 | if (from_schedule) { | |
2869 | spin_unlock(q->queue_lock); | |
24ecfbe2 | 2870 | blk_run_queue_async(q); |
99e22598 | 2871 | } else { |
24ecfbe2 | 2872 | __blk_run_queue(q); |
99e22598 JA |
2873 | spin_unlock(q->queue_lock); |
2874 | } | |
2875 | ||
94b5eb28 JA |
2876 | } |
2877 | ||
048c9374 N |
2878 | static void flush_plug_callbacks(struct blk_plug *plug) |
2879 | { | |
2880 | LIST_HEAD(callbacks); | |
2881 | ||
2882 | if (list_empty(&plug->cb_list)) | |
2883 | return; | |
2884 | ||
2885 | list_splice_init(&plug->cb_list, &callbacks); | |
2886 | ||
2887 | while (!list_empty(&callbacks)) { | |
2888 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
2889 | struct blk_plug_cb, | |
2890 | list); | |
2891 | list_del(&cb->list); | |
2892 | cb->callback(cb); | |
2893 | } | |
2894 | } | |
2895 | ||
49cac01e | 2896 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
2897 | { |
2898 | struct request_queue *q; | |
2899 | unsigned long flags; | |
2900 | struct request *rq; | |
109b8129 | 2901 | LIST_HEAD(list); |
94b5eb28 | 2902 | unsigned int depth; |
73c10101 JA |
2903 | |
2904 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2905 | ||
048c9374 | 2906 | flush_plug_callbacks(plug); |
73c10101 JA |
2907 | if (list_empty(&plug->list)) |
2908 | return; | |
2909 | ||
109b8129 N |
2910 | list_splice_init(&plug->list, &list); |
2911 | ||
2912 | if (plug->should_sort) { | |
2913 | list_sort(NULL, &list, plug_rq_cmp); | |
2914 | plug->should_sort = 0; | |
2915 | } | |
73c10101 JA |
2916 | |
2917 | q = NULL; | |
94b5eb28 | 2918 | depth = 0; |
18811272 JA |
2919 | |
2920 | /* | |
2921 | * Save and disable interrupts here, to avoid doing it for every | |
2922 | * queue lock we have to take. | |
2923 | */ | |
73c10101 | 2924 | local_irq_save(flags); |
109b8129 N |
2925 | while (!list_empty(&list)) { |
2926 | rq = list_entry_rq(list.next); | |
73c10101 | 2927 | list_del_init(&rq->queuelist); |
73c10101 JA |
2928 | BUG_ON(!rq->q); |
2929 | if (rq->q != q) { | |
99e22598 JA |
2930 | /* |
2931 | * This drops the queue lock | |
2932 | */ | |
2933 | if (q) | |
49cac01e | 2934 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2935 | q = rq->q; |
94b5eb28 | 2936 | depth = 0; |
73c10101 JA |
2937 | spin_lock(q->queue_lock); |
2938 | } | |
8ba61435 TH |
2939 | |
2940 | /* | |
2941 | * Short-circuit if @q is dead | |
2942 | */ | |
2943 | if (unlikely(blk_queue_dead(q))) { | |
2944 | __blk_end_request_all(rq, -ENODEV); | |
2945 | continue; | |
2946 | } | |
2947 | ||
73c10101 JA |
2948 | /* |
2949 | * rq is already accounted, so use raw insert | |
2950 | */ | |
401a18e9 JA |
2951 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
2952 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
2953 | else | |
2954 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
2955 | |
2956 | depth++; | |
73c10101 JA |
2957 | } |
2958 | ||
99e22598 JA |
2959 | /* |
2960 | * This drops the queue lock | |
2961 | */ | |
2962 | if (q) | |
49cac01e | 2963 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2964 | |
73c10101 JA |
2965 | local_irq_restore(flags); |
2966 | } | |
73c10101 JA |
2967 | |
2968 | void blk_finish_plug(struct blk_plug *plug) | |
2969 | { | |
f6603783 | 2970 | blk_flush_plug_list(plug, false); |
73c10101 | 2971 | |
88b996cd CH |
2972 | if (plug == current->plug) |
2973 | current->plug = NULL; | |
73c10101 | 2974 | } |
88b996cd | 2975 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 2976 | |
1da177e4 LT |
2977 | int __init blk_dev_init(void) |
2978 | { | |
9eb55b03 NK |
2979 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2980 | sizeof(((struct request *)0)->cmd_flags)); | |
2981 | ||
89b90be2 TH |
2982 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2983 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2984 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2985 | if (!kblockd_workqueue) |
2986 | panic("Failed to create kblockd\n"); | |
2987 | ||
2988 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2989 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2990 | |
8324aa91 | 2991 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2992 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2993 | |
d38ecf93 | 2994 | return 0; |
1da177e4 | 2995 | } |