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