Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 29 | #include <linux/fault-inject.h> |
73c10101 | 30 | #include <linux/list_sort.h> |
e3c78ca5 | 31 | #include <linux/delay.h> |
55782138 LZ |
32 | |
33 | #define CREATE_TRACE_POINTS | |
34 | #include <trace/events/block.h> | |
1da177e4 | 35 | |
8324aa91 | 36 | #include "blk.h" |
5efd6113 | 37 | #include "blk-cgroup.h" |
8324aa91 | 38 | |
d07335e5 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 41 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 42 | |
a73f730d TH |
43 | DEFINE_IDA(blk_queue_ida); |
44 | ||
1da177e4 LT |
45 | /* |
46 | * For the allocated request tables | |
47 | */ | |
5ece6c52 | 48 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
49 | |
50 | /* | |
51 | * For queue allocation | |
52 | */ | |
6728cb0e | 53 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Controlling structure to kblockd | |
57 | */ | |
ff856bad | 58 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 59 | |
26b8256e JA |
60 | static void drive_stat_acct(struct request *rq, int new_io) |
61 | { | |
28f13702 | 62 | struct hd_struct *part; |
26b8256e | 63 | int rw = rq_data_dir(rq); |
c9959059 | 64 | int cpu; |
26b8256e | 65 | |
c2553b58 | 66 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
67 | return; |
68 | ||
074a7aca | 69 | cpu = part_stat_lock(); |
c9959059 | 70 | |
09e099d4 JM |
71 | if (!new_io) { |
72 | part = rq->part; | |
074a7aca | 73 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
74 | } else { |
75 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 76 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
77 | /* |
78 | * The partition is already being removed, | |
79 | * the request will be accounted on the disk only | |
80 | * | |
81 | * We take a reference on disk->part0 although that | |
82 | * partition will never be deleted, so we can treat | |
83 | * it as any other partition. | |
84 | */ | |
85 | part = &rq->rq_disk->part0; | |
6c23a968 | 86 | hd_struct_get(part); |
09e099d4 | 87 | } |
074a7aca | 88 | part_round_stats(cpu, part); |
316d315b | 89 | part_inc_in_flight(part, rw); |
09e099d4 | 90 | rq->part = part; |
26b8256e | 91 | } |
e71bf0d0 | 92 | |
074a7aca | 93 | part_stat_unlock(); |
26b8256e JA |
94 | } |
95 | ||
8324aa91 | 96 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
97 | { |
98 | int nr; | |
99 | ||
100 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
101 | if (nr > q->nr_requests) | |
102 | nr = q->nr_requests; | |
103 | q->nr_congestion_on = nr; | |
104 | ||
105 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
106 | if (nr < 1) | |
107 | nr = 1; | |
108 | q->nr_congestion_off = nr; | |
109 | } | |
110 | ||
1da177e4 LT |
111 | /** |
112 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
113 | * @bdev: device | |
114 | * | |
115 | * Locates the passed device's request queue and returns the address of its | |
116 | * backing_dev_info | |
117 | * | |
118 | * Will return NULL if the request queue cannot be located. | |
119 | */ | |
120 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
121 | { | |
122 | struct backing_dev_info *ret = NULL; | |
165125e1 | 123 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
124 | |
125 | if (q) | |
126 | ret = &q->backing_dev_info; | |
127 | return ret; | |
128 | } | |
1da177e4 LT |
129 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
130 | ||
2a4aa30c | 131 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 132 | { |
1afb20f3 FT |
133 | memset(rq, 0, sizeof(*rq)); |
134 | ||
1da177e4 | 135 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 136 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 137 | rq->cpu = -1; |
63a71386 | 138 | rq->q = q; |
a2dec7b3 | 139 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
140 | INIT_HLIST_NODE(&rq->hash); |
141 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 142 | rq->cmd = rq->__cmd; |
e2494e1b | 143 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 144 | rq->tag = -1; |
1da177e4 | 145 | rq->ref_count = 1; |
b243ddcb | 146 | rq->start_time = jiffies; |
9195291e | 147 | set_start_time_ns(rq); |
09e099d4 | 148 | rq->part = NULL; |
1da177e4 | 149 | } |
2a4aa30c | 150 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 151 | |
5bb23a68 N |
152 | static void req_bio_endio(struct request *rq, struct bio *bio, |
153 | unsigned int nbytes, int error) | |
1da177e4 | 154 | { |
143a87f4 TH |
155 | if (error) |
156 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
157 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
158 | error = -EIO; | |
797e7dbb | 159 | |
143a87f4 TH |
160 | if (unlikely(nbytes > bio->bi_size)) { |
161 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
162 | __func__, nbytes, bio->bi_size); | |
163 | nbytes = bio->bi_size; | |
5bb23a68 | 164 | } |
797e7dbb | 165 | |
143a87f4 TH |
166 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
167 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 168 | |
143a87f4 TH |
169 | bio->bi_size -= nbytes; |
170 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 171 | |
143a87f4 TH |
172 | if (bio_integrity(bio)) |
173 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 174 | |
143a87f4 TH |
175 | /* don't actually finish bio if it's part of flush sequence */ |
176 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
177 | bio_endio(bio, error); | |
1da177e4 | 178 | } |
1da177e4 | 179 | |
1da177e4 LT |
180 | void blk_dump_rq_flags(struct request *rq, char *msg) |
181 | { | |
182 | int bit; | |
183 | ||
6728cb0e | 184 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
185 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
186 | rq->cmd_flags); | |
1da177e4 | 187 | |
83096ebf TH |
188 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
189 | (unsigned long long)blk_rq_pos(rq), | |
190 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 191 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 192 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 193 | |
33659ebb | 194 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 195 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 196 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
197 | printk("%02x ", rq->cmd[bit]); |
198 | printk("\n"); | |
199 | } | |
200 | } | |
1da177e4 LT |
201 | EXPORT_SYMBOL(blk_dump_rq_flags); |
202 | ||
3cca6dc1 | 203 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 204 | { |
3cca6dc1 | 205 | struct request_queue *q; |
1da177e4 | 206 | |
3cca6dc1 JA |
207 | q = container_of(work, struct request_queue, delay_work.work); |
208 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 209 | __blk_run_queue(q); |
3cca6dc1 | 210 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 211 | } |
1da177e4 LT |
212 | |
213 | /** | |
3cca6dc1 JA |
214 | * blk_delay_queue - restart queueing after defined interval |
215 | * @q: The &struct request_queue in question | |
216 | * @msecs: Delay in msecs | |
1da177e4 LT |
217 | * |
218 | * Description: | |
3cca6dc1 JA |
219 | * Sometimes queueing needs to be postponed for a little while, to allow |
220 | * resources to come back. This function will make sure that queueing is | |
221 | * restarted around the specified time. | |
222 | */ | |
223 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 224 | { |
4521cc4e JA |
225 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
226 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 227 | } |
3cca6dc1 | 228 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 229 | |
1da177e4 LT |
230 | /** |
231 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 232 | * @q: The &struct request_queue in question |
1da177e4 LT |
233 | * |
234 | * Description: | |
235 | * blk_start_queue() will clear the stop flag on the queue, and call | |
236 | * the request_fn for the queue if it was in a stopped state when | |
237 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
238 | **/ | |
165125e1 | 239 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 240 | { |
a038e253 PBG |
241 | WARN_ON(!irqs_disabled()); |
242 | ||
75ad23bc | 243 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 244 | __blk_run_queue(q); |
1da177e4 | 245 | } |
1da177e4 LT |
246 | EXPORT_SYMBOL(blk_start_queue); |
247 | ||
248 | /** | |
249 | * blk_stop_queue - stop a queue | |
165125e1 | 250 | * @q: The &struct request_queue in question |
1da177e4 LT |
251 | * |
252 | * Description: | |
253 | * The Linux block layer assumes that a block driver will consume all | |
254 | * entries on the request queue when the request_fn strategy is called. | |
255 | * Often this will not happen, because of hardware limitations (queue | |
256 | * depth settings). If a device driver gets a 'queue full' response, | |
257 | * or if it simply chooses not to queue more I/O at one point, it can | |
258 | * call this function to prevent the request_fn from being called until | |
259 | * the driver has signalled it's ready to go again. This happens by calling | |
260 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
261 | **/ | |
165125e1 | 262 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 263 | { |
ad3d9d7e | 264 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 265 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
266 | } |
267 | EXPORT_SYMBOL(blk_stop_queue); | |
268 | ||
269 | /** | |
270 | * blk_sync_queue - cancel any pending callbacks on a queue | |
271 | * @q: the queue | |
272 | * | |
273 | * Description: | |
274 | * The block layer may perform asynchronous callback activity | |
275 | * on a queue, such as calling the unplug function after a timeout. | |
276 | * A block device may call blk_sync_queue to ensure that any | |
277 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 278 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
279 | * that its ->make_request_fn will not re-add plugging prior to calling |
280 | * this function. | |
281 | * | |
da527770 VG |
282 | * This function does not cancel any asynchronous activity arising |
283 | * out of elevator or throttling code. That would require elevaotor_exit() | |
5efd6113 | 284 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 285 | * |
1da177e4 LT |
286 | */ |
287 | void blk_sync_queue(struct request_queue *q) | |
288 | { | |
70ed28b9 | 289 | del_timer_sync(&q->timeout); |
3cca6dc1 | 290 | cancel_delayed_work_sync(&q->delay_work); |
1da177e4 LT |
291 | } |
292 | EXPORT_SYMBOL(blk_sync_queue); | |
293 | ||
294 | /** | |
80a4b58e | 295 | * __blk_run_queue - run a single device queue |
1da177e4 | 296 | * @q: The queue to run |
80a4b58e JA |
297 | * |
298 | * Description: | |
299 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 300 | * held and interrupts disabled. |
1da177e4 | 301 | */ |
24ecfbe2 | 302 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 303 | { |
a538cd03 TH |
304 | if (unlikely(blk_queue_stopped(q))) |
305 | return; | |
306 | ||
c21e6beb | 307 | q->request_fn(q); |
75ad23bc NP |
308 | } |
309 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 310 | |
24ecfbe2 CH |
311 | /** |
312 | * blk_run_queue_async - run a single device queue in workqueue context | |
313 | * @q: The queue to run | |
314 | * | |
315 | * Description: | |
316 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
317 | * of us. | |
318 | */ | |
319 | void blk_run_queue_async(struct request_queue *q) | |
320 | { | |
3ec717b7 SL |
321 | if (likely(!blk_queue_stopped(q))) { |
322 | __cancel_delayed_work(&q->delay_work); | |
24ecfbe2 | 323 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
3ec717b7 | 324 | } |
24ecfbe2 | 325 | } |
c21e6beb | 326 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 327 | |
75ad23bc NP |
328 | /** |
329 | * blk_run_queue - run a single device queue | |
330 | * @q: The queue to run | |
80a4b58e JA |
331 | * |
332 | * Description: | |
333 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 334 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
335 | */ |
336 | void blk_run_queue(struct request_queue *q) | |
337 | { | |
338 | unsigned long flags; | |
339 | ||
340 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 341 | __blk_run_queue(q); |
1da177e4 LT |
342 | spin_unlock_irqrestore(q->queue_lock, flags); |
343 | } | |
344 | EXPORT_SYMBOL(blk_run_queue); | |
345 | ||
165125e1 | 346 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
347 | { |
348 | kobject_put(&q->kobj); | |
349 | } | |
d86e0e83 | 350 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 351 | |
e3c78ca5 TH |
352 | /** |
353 | * blk_drain_queue - drain requests from request_queue | |
354 | * @q: queue to drain | |
c9a929dd | 355 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 356 | * |
c9a929dd TH |
357 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
358 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
359 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 360 | */ |
c9a929dd | 361 | void blk_drain_queue(struct request_queue *q, bool drain_all) |
e3c78ca5 TH |
362 | { |
363 | while (true) { | |
481a7d64 TH |
364 | bool drain = false; |
365 | int i; | |
e3c78ca5 TH |
366 | |
367 | spin_lock_irq(q->queue_lock); | |
368 | ||
b855b04a TH |
369 | /* |
370 | * The caller might be trying to drain @q before its | |
371 | * elevator is initialized. | |
372 | */ | |
373 | if (q->elevator) | |
374 | elv_drain_elevator(q); | |
375 | ||
5efd6113 | 376 | blkcg_drain_queue(q); |
e3c78ca5 | 377 | |
4eabc941 TH |
378 | /* |
379 | * This function might be called on a queue which failed | |
b855b04a TH |
380 | * driver init after queue creation or is not yet fully |
381 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
382 | * in such cases. Kick queue iff dispatch queue has | |
383 | * something on it and @q has request_fn set. | |
4eabc941 | 384 | */ |
b855b04a | 385 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 386 | __blk_run_queue(q); |
c9a929dd | 387 | |
481a7d64 TH |
388 | drain |= q->rq.elvpriv; |
389 | ||
390 | /* | |
391 | * Unfortunately, requests are queued at and tracked from | |
392 | * multiple places and there's no single counter which can | |
393 | * be drained. Check all the queues and counters. | |
394 | */ | |
395 | if (drain_all) { | |
396 | drain |= !list_empty(&q->queue_head); | |
397 | for (i = 0; i < 2; i++) { | |
398 | drain |= q->rq.count[i]; | |
399 | drain |= q->in_flight[i]; | |
400 | drain |= !list_empty(&q->flush_queue[i]); | |
401 | } | |
402 | } | |
e3c78ca5 TH |
403 | |
404 | spin_unlock_irq(q->queue_lock); | |
405 | ||
481a7d64 | 406 | if (!drain) |
e3c78ca5 TH |
407 | break; |
408 | msleep(10); | |
409 | } | |
410 | } | |
411 | ||
d732580b TH |
412 | /** |
413 | * blk_queue_bypass_start - enter queue bypass mode | |
414 | * @q: queue of interest | |
415 | * | |
416 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
417 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af TH |
418 | * throttled or issued before. On return, it's guaranteed that no request |
419 | * is being throttled or has ELVPRIV set. | |
d732580b TH |
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); | |
6ecf23af TH |
464 | |
465 | /* dead queue is permanently in bypass mode till released */ | |
466 | q->bypass_depth++; | |
467 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
468 | ||
c9a929dd TH |
469 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
470 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
471 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
483f4afc | 472 | |
777eb1bf HR |
473 | if (q->queue_lock != &q->__queue_lock) |
474 | q->queue_lock = &q->__queue_lock; | |
da527770 | 475 | |
c9a929dd TH |
476 | spin_unlock_irq(lock); |
477 | mutex_unlock(&q->sysfs_lock); | |
478 | ||
b855b04a TH |
479 | /* drain all requests queued before DEAD marking */ |
480 | blk_drain_queue(q, true); | |
c9a929dd TH |
481 | |
482 | /* @q won't process any more request, flush async actions */ | |
483 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
484 | blk_sync_queue(q); | |
485 | ||
486 | /* @q is and will stay empty, shutdown and put */ | |
483f4afc AV |
487 | blk_put_queue(q); |
488 | } | |
1da177e4 LT |
489 | EXPORT_SYMBOL(blk_cleanup_queue); |
490 | ||
165125e1 | 491 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
492 | { |
493 | struct request_list *rl = &q->rq; | |
494 | ||
1abec4fd MS |
495 | if (unlikely(rl->rq_pool)) |
496 | return 0; | |
497 | ||
1faa16d2 JA |
498 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
499 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 500 | rl->elvpriv = 0; |
1faa16d2 JA |
501 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
502 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 503 | |
1946089a CL |
504 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
505 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
506 | |
507 | if (!rl->rq_pool) | |
508 | return -ENOMEM; | |
509 | ||
510 | return 0; | |
511 | } | |
512 | ||
165125e1 | 513 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 514 | { |
1946089a CL |
515 | return blk_alloc_queue_node(gfp_mask, -1); |
516 | } | |
517 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 518 | |
165125e1 | 519 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 520 | { |
165125e1 | 521 | struct request_queue *q; |
e0bf68dd | 522 | int err; |
1946089a | 523 | |
8324aa91 | 524 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 525 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
526 | if (!q) |
527 | return NULL; | |
528 | ||
a73f730d TH |
529 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL); |
530 | if (q->id < 0) | |
531 | goto fail_q; | |
532 | ||
0989a025 JA |
533 | q->backing_dev_info.ra_pages = |
534 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
535 | q->backing_dev_info.state = 0; | |
536 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 537 | q->backing_dev_info.name = "block"; |
5151412d | 538 | q->node = node_id; |
0989a025 | 539 | |
e0bf68dd | 540 | err = bdi_init(&q->backing_dev_info); |
a73f730d TH |
541 | if (err) |
542 | goto fail_id; | |
e0bf68dd | 543 | |
31373d09 MG |
544 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
545 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 546 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 547 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 548 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 549 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 550 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 551 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 552 | #endif |
ae1b1539 TH |
553 | INIT_LIST_HEAD(&q->flush_queue[0]); |
554 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
555 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 556 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 557 | |
8324aa91 | 558 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 559 | |
483f4afc | 560 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 561 | spin_lock_init(&q->__queue_lock); |
483f4afc | 562 | |
c94a96ac VG |
563 | /* |
564 | * By default initialize queue_lock to internal lock and driver can | |
565 | * override it later if need be. | |
566 | */ | |
567 | q->queue_lock = &q->__queue_lock; | |
568 | ||
5efd6113 | 569 | if (blkcg_init_queue(q)) |
f51b802c TH |
570 | goto fail_id; |
571 | ||
1da177e4 | 572 | return q; |
a73f730d TH |
573 | |
574 | fail_id: | |
575 | ida_simple_remove(&blk_queue_ida, q->id); | |
576 | fail_q: | |
577 | kmem_cache_free(blk_requestq_cachep, q); | |
578 | return NULL; | |
1da177e4 | 579 | } |
1946089a | 580 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
581 | |
582 | /** | |
583 | * blk_init_queue - prepare a request queue for use with a block device | |
584 | * @rfn: The function to be called to process requests that have been | |
585 | * placed on the queue. | |
586 | * @lock: Request queue spin lock | |
587 | * | |
588 | * Description: | |
589 | * If a block device wishes to use the standard request handling procedures, | |
590 | * which sorts requests and coalesces adjacent requests, then it must | |
591 | * call blk_init_queue(). The function @rfn will be called when there | |
592 | * are requests on the queue that need to be processed. If the device | |
593 | * supports plugging, then @rfn may not be called immediately when requests | |
594 | * are available on the queue, but may be called at some time later instead. | |
595 | * Plugged queues are generally unplugged when a buffer belonging to one | |
596 | * of the requests on the queue is needed, or due to memory pressure. | |
597 | * | |
598 | * @rfn is not required, or even expected, to remove all requests off the | |
599 | * queue, but only as many as it can handle at a time. If it does leave | |
600 | * requests on the queue, it is responsible for arranging that the requests | |
601 | * get dealt with eventually. | |
602 | * | |
603 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
604 | * request queue; this lock will be taken also from interrupt context, so irq |
605 | * disabling is needed for it. | |
1da177e4 | 606 | * |
710027a4 | 607 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
608 | * it didn't succeed. |
609 | * | |
610 | * Note: | |
611 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
612 | * when the block device is deactivated (such as at module unload). | |
613 | **/ | |
1946089a | 614 | |
165125e1 | 615 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 616 | { |
1946089a CL |
617 | return blk_init_queue_node(rfn, lock, -1); |
618 | } | |
619 | EXPORT_SYMBOL(blk_init_queue); | |
620 | ||
165125e1 | 621 | struct request_queue * |
1946089a CL |
622 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
623 | { | |
c86d1b8a | 624 | struct request_queue *uninit_q, *q; |
1da177e4 | 625 | |
c86d1b8a MS |
626 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
627 | if (!uninit_q) | |
628 | return NULL; | |
629 | ||
5151412d | 630 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a MS |
631 | if (!q) |
632 | blk_cleanup_queue(uninit_q); | |
633 | ||
634 | return q; | |
01effb0d MS |
635 | } |
636 | EXPORT_SYMBOL(blk_init_queue_node); | |
637 | ||
638 | struct request_queue * | |
639 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
640 | spinlock_t *lock) | |
01effb0d | 641 | { |
1da177e4 LT |
642 | if (!q) |
643 | return NULL; | |
644 | ||
c86d1b8a | 645 | if (blk_init_free_list(q)) |
8669aafd | 646 | return NULL; |
1da177e4 LT |
647 | |
648 | q->request_fn = rfn; | |
1da177e4 | 649 | q->prep_rq_fn = NULL; |
28018c24 | 650 | q->unprep_rq_fn = NULL; |
bc58ba94 | 651 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
652 | |
653 | /* Override internal queue lock with supplied lock pointer */ | |
654 | if (lock) | |
655 | q->queue_lock = lock; | |
1da177e4 | 656 | |
f3b144aa JA |
657 | /* |
658 | * This also sets hw/phys segments, boundary and size | |
659 | */ | |
c20e8de2 | 660 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 661 | |
44ec9542 AS |
662 | q->sg_reserved_size = INT_MAX; |
663 | ||
1da177e4 LT |
664 | /* |
665 | * all done | |
666 | */ | |
667 | if (!elevator_init(q, NULL)) { | |
668 | blk_queue_congestion_threshold(q); | |
669 | return q; | |
670 | } | |
671 | ||
1da177e4 LT |
672 | return NULL; |
673 | } | |
5151412d | 674 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 675 | |
09ac46c4 | 676 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 677 | { |
34f6055c | 678 | if (likely(!blk_queue_dead(q))) { |
09ac46c4 TH |
679 | __blk_get_queue(q); |
680 | return true; | |
1da177e4 LT |
681 | } |
682 | ||
09ac46c4 | 683 | return false; |
1da177e4 | 684 | } |
d86e0e83 | 685 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 686 | |
165125e1 | 687 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 688 | { |
f1f8cc94 | 689 | if (rq->cmd_flags & REQ_ELVPRIV) { |
cb98fc8b | 690 | elv_put_request(q, rq); |
f1f8cc94 | 691 | if (rq->elv.icq) |
11a3122f | 692 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
693 | } |
694 | ||
1da177e4 LT |
695 | mempool_free(rq, q->rq.rq_pool); |
696 | } | |
697 | ||
1ea25ecb | 698 | static struct request * |
852c788f | 699 | blk_alloc_request(struct request_queue *q, struct bio *bio, struct io_cq *icq, |
f1f8cc94 | 700 | unsigned int flags, gfp_t gfp_mask) |
1da177e4 LT |
701 | { |
702 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
703 | ||
704 | if (!rq) | |
705 | return NULL; | |
706 | ||
2a4aa30c | 707 | blk_rq_init(q, rq); |
1afb20f3 | 708 | |
42dad764 | 709 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 710 | |
f1f8cc94 TH |
711 | if (flags & REQ_ELVPRIV) { |
712 | rq->elv.icq = icq; | |
852c788f | 713 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { |
f1f8cc94 TH |
714 | mempool_free(rq, q->rq.rq_pool); |
715 | return NULL; | |
716 | } | |
717 | /* @rq->elv.icq holds on to io_context until @rq is freed */ | |
718 | if (icq) | |
719 | get_io_context(icq->ioc); | |
cb98fc8b | 720 | } |
1da177e4 | 721 | |
cb98fc8b | 722 | return rq; |
1da177e4 LT |
723 | } |
724 | ||
725 | /* | |
726 | * ioc_batching returns true if the ioc is a valid batching request and | |
727 | * should be given priority access to a request. | |
728 | */ | |
165125e1 | 729 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
730 | { |
731 | if (!ioc) | |
732 | return 0; | |
733 | ||
734 | /* | |
735 | * Make sure the process is able to allocate at least 1 request | |
736 | * even if the batch times out, otherwise we could theoretically | |
737 | * lose wakeups. | |
738 | */ | |
739 | return ioc->nr_batch_requests == q->nr_batching || | |
740 | (ioc->nr_batch_requests > 0 | |
741 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
742 | } | |
743 | ||
744 | /* | |
745 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
746 | * will cause the process to be a "batcher" on all queues in the system. This | |
747 | * is the behaviour we want though - once it gets a wakeup it should be given | |
748 | * a nice run. | |
749 | */ | |
165125e1 | 750 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
751 | { |
752 | if (!ioc || ioc_batching(q, ioc)) | |
753 | return; | |
754 | ||
755 | ioc->nr_batch_requests = q->nr_batching; | |
756 | ioc->last_waited = jiffies; | |
757 | } | |
758 | ||
1faa16d2 | 759 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
760 | { |
761 | struct request_list *rl = &q->rq; | |
762 | ||
1faa16d2 JA |
763 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
764 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 765 | |
1faa16d2 JA |
766 | if (rl->count[sync] + 1 <= q->nr_requests) { |
767 | if (waitqueue_active(&rl->wait[sync])) | |
768 | wake_up(&rl->wait[sync]); | |
1da177e4 | 769 | |
1faa16d2 | 770 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
771 | } |
772 | } | |
773 | ||
774 | /* | |
775 | * A request has just been released. Account for it, update the full and | |
776 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
777 | */ | |
75eb6c37 | 778 | static void freed_request(struct request_queue *q, unsigned int flags) |
1da177e4 LT |
779 | { |
780 | struct request_list *rl = &q->rq; | |
75eb6c37 | 781 | int sync = rw_is_sync(flags); |
1da177e4 | 782 | |
1faa16d2 | 783 | rl->count[sync]--; |
75eb6c37 | 784 | if (flags & REQ_ELVPRIV) |
cb98fc8b | 785 | rl->elvpriv--; |
1da177e4 | 786 | |
1faa16d2 | 787 | __freed_request(q, sync); |
1da177e4 | 788 | |
1faa16d2 JA |
789 | if (unlikely(rl->starved[sync ^ 1])) |
790 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
791 | } |
792 | ||
9d5a4e94 MS |
793 | /* |
794 | * Determine if elevator data should be initialized when allocating the | |
795 | * request associated with @bio. | |
796 | */ | |
797 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
798 | { | |
799 | if (!bio) | |
800 | return true; | |
801 | ||
802 | /* | |
803 | * Flush requests do not use the elevator so skip initialization. | |
804 | * This allows a request to share the flush and elevator data. | |
805 | */ | |
806 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
807 | return false; | |
808 | ||
809 | return true; | |
810 | } | |
811 | ||
852c788f TH |
812 | /** |
813 | * rq_ioc - determine io_context for request allocation | |
814 | * @bio: request being allocated is for this bio (can be %NULL) | |
815 | * | |
816 | * Determine io_context to use for request allocation for @bio. May return | |
817 | * %NULL if %current->io_context doesn't exist. | |
818 | */ | |
819 | static struct io_context *rq_ioc(struct bio *bio) | |
820 | { | |
821 | #ifdef CONFIG_BLK_CGROUP | |
822 | if (bio && bio->bi_ioc) | |
823 | return bio->bi_ioc; | |
824 | #endif | |
825 | return current->io_context; | |
826 | } | |
827 | ||
da8303c6 TH |
828 | /** |
829 | * get_request - get a free request | |
830 | * @q: request_queue to allocate request from | |
831 | * @rw_flags: RW and SYNC flags | |
832 | * @bio: bio to allocate request for (can be %NULL) | |
833 | * @gfp_mask: allocation mask | |
834 | * | |
835 | * Get a free request from @q. This function may fail under memory | |
836 | * pressure or if @q is dead. | |
837 | * | |
838 | * Must be callled with @q->queue_lock held and, | |
839 | * Returns %NULL on failure, with @q->queue_lock held. | |
840 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 841 | */ |
165125e1 | 842 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 843 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 844 | { |
b679281a | 845 | struct request *rq; |
1da177e4 | 846 | struct request_list *rl = &q->rq; |
f1f8cc94 | 847 | struct elevator_type *et; |
f2dbd76a | 848 | struct io_context *ioc; |
f1f8cc94 | 849 | struct io_cq *icq = NULL; |
1faa16d2 | 850 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
f2dbd76a | 851 | bool retried = false; |
75eb6c37 | 852 | int may_queue; |
f2dbd76a | 853 | retry: |
f1f8cc94 | 854 | et = q->elevator->type; |
852c788f | 855 | ioc = rq_ioc(bio); |
88ee5ef1 | 856 | |
34f6055c | 857 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
858 | return NULL; |
859 | ||
7749a8d4 | 860 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
861 | if (may_queue == ELV_MQUEUE_NO) |
862 | goto rq_starved; | |
863 | ||
1faa16d2 JA |
864 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
865 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
f2dbd76a TH |
866 | /* |
867 | * We want ioc to record batching state. If it's | |
868 | * not already there, creating a new one requires | |
869 | * dropping queue_lock, which in turn requires | |
870 | * retesting conditions to avoid queue hang. | |
871 | */ | |
872 | if (!ioc && !retried) { | |
873 | spin_unlock_irq(q->queue_lock); | |
24acfc34 | 874 | create_io_context(gfp_mask, q->node); |
f2dbd76a TH |
875 | spin_lock_irq(q->queue_lock); |
876 | retried = true; | |
877 | goto retry; | |
878 | } | |
879 | ||
88ee5ef1 JA |
880 | /* |
881 | * The queue will fill after this allocation, so set | |
882 | * it as full, and mark this process as "batching". | |
883 | * This process will be allowed to complete a batch of | |
884 | * requests, others will be blocked. | |
885 | */ | |
1faa16d2 | 886 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 887 | ioc_set_batching(q, ioc); |
1faa16d2 | 888 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
889 | } else { |
890 | if (may_queue != ELV_MQUEUE_MUST | |
891 | && !ioc_batching(q, ioc)) { | |
892 | /* | |
893 | * The queue is full and the allocating | |
894 | * process is not a "batcher", and not | |
895 | * exempted by the IO scheduler | |
896 | */ | |
b679281a | 897 | return NULL; |
88ee5ef1 JA |
898 | } |
899 | } | |
1da177e4 | 900 | } |
1faa16d2 | 901 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
902 | } |
903 | ||
082cf69e JA |
904 | /* |
905 | * Only allow batching queuers to allocate up to 50% over the defined | |
906 | * limit of requests, otherwise we could have thousands of requests | |
907 | * allocated with any setting of ->nr_requests | |
908 | */ | |
1faa16d2 | 909 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
b679281a | 910 | return NULL; |
fd782a4a | 911 | |
1faa16d2 JA |
912 | rl->count[is_sync]++; |
913 | rl->starved[is_sync] = 0; | |
cb98fc8b | 914 | |
f1f8cc94 TH |
915 | /* |
916 | * Decide whether the new request will be managed by elevator. If | |
917 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will | |
918 | * prevent the current elevator from being destroyed until the new | |
919 | * request is freed. This guarantees icq's won't be destroyed and | |
920 | * makes creating new ones safe. | |
921 | * | |
922 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
923 | * it will be created after releasing queue_lock. | |
924 | */ | |
d732580b | 925 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
75eb6c37 TH |
926 | rw_flags |= REQ_ELVPRIV; |
927 | rl->elvpriv++; | |
f1f8cc94 TH |
928 | if (et->icq_cache && ioc) |
929 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 930 | } |
cb98fc8b | 931 | |
f253b86b JA |
932 | if (blk_queue_io_stat(q)) |
933 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
934 | spin_unlock_irq(q->queue_lock); |
935 | ||
f1f8cc94 | 936 | /* create icq if missing */ |
05c30b95 | 937 | if ((rw_flags & REQ_ELVPRIV) && unlikely(et->icq_cache && !icq)) { |
852c788f TH |
938 | create_io_context(gfp_mask, q->node); |
939 | ioc = rq_ioc(bio); | |
940 | if (!ioc) | |
941 | goto fail_alloc; | |
942 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
05c30b95 | 943 | if (!icq) |
b679281a | 944 | goto fail_alloc; |
05c30b95 | 945 | } |
f1f8cc94 | 946 | |
852c788f | 947 | rq = blk_alloc_request(q, bio, icq, rw_flags, gfp_mask); |
b679281a TH |
948 | if (unlikely(!rq)) |
949 | goto fail_alloc; | |
1da177e4 | 950 | |
88ee5ef1 JA |
951 | /* |
952 | * ioc may be NULL here, and ioc_batching will be false. That's | |
953 | * OK, if the queue is under the request limit then requests need | |
954 | * not count toward the nr_batch_requests limit. There will always | |
955 | * be some limit enforced by BLK_BATCH_TIME. | |
956 | */ | |
1da177e4 LT |
957 | if (ioc_batching(q, ioc)) |
958 | ioc->nr_batch_requests--; | |
6728cb0e | 959 | |
1faa16d2 | 960 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 961 | return rq; |
b679281a TH |
962 | |
963 | fail_alloc: | |
964 | /* | |
965 | * Allocation failed presumably due to memory. Undo anything we | |
966 | * might have messed up. | |
967 | * | |
968 | * Allocating task should really be put onto the front of the wait | |
969 | * queue, but this is pretty rare. | |
970 | */ | |
971 | spin_lock_irq(q->queue_lock); | |
972 | freed_request(q, rw_flags); | |
973 | ||
974 | /* | |
975 | * in the very unlikely event that allocation failed and no | |
976 | * requests for this direction was pending, mark us starved so that | |
977 | * freeing of a request in the other direction will notice | |
978 | * us. another possible fix would be to split the rq mempool into | |
979 | * READ and WRITE | |
980 | */ | |
981 | rq_starved: | |
982 | if (unlikely(rl->count[is_sync] == 0)) | |
983 | rl->starved[is_sync] = 1; | |
984 | return NULL; | |
1da177e4 LT |
985 | } |
986 | ||
da8303c6 TH |
987 | /** |
988 | * get_request_wait - get a free request with retry | |
989 | * @q: request_queue to allocate request from | |
990 | * @rw_flags: RW and SYNC flags | |
991 | * @bio: bio to allocate request for (can be %NULL) | |
992 | * | |
993 | * Get a free request from @q. This function keeps retrying under memory | |
994 | * pressure and fails iff @q is dead. | |
d6344532 | 995 | * |
da8303c6 TH |
996 | * Must be callled with @q->queue_lock held and, |
997 | * Returns %NULL on failure, with @q->queue_lock held. | |
998 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 999 | */ |
165125e1 | 1000 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 1001 | struct bio *bio) |
1da177e4 | 1002 | { |
1faa16d2 | 1003 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
1004 | struct request *rq; |
1005 | ||
7749a8d4 | 1006 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
1007 | while (!rq) { |
1008 | DEFINE_WAIT(wait); | |
1da177e4 LT |
1009 | struct request_list *rl = &q->rq; |
1010 | ||
34f6055c | 1011 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
1012 | return NULL; |
1013 | ||
1faa16d2 | 1014 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
1015 | TASK_UNINTERRUPTIBLE); |
1016 | ||
1faa16d2 | 1017 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 1018 | |
05caf8db ZY |
1019 | spin_unlock_irq(q->queue_lock); |
1020 | io_schedule(); | |
1da177e4 | 1021 | |
05caf8db ZY |
1022 | /* |
1023 | * After sleeping, we become a "batching" process and | |
1024 | * will be able to allocate at least one request, and | |
1025 | * up to a big batch of them for a small period time. | |
1026 | * See ioc_batching, ioc_set_batching | |
1027 | */ | |
24acfc34 | 1028 | create_io_context(GFP_NOIO, q->node); |
f2dbd76a | 1029 | ioc_set_batching(q, current->io_context); |
d6344532 | 1030 | |
05caf8db | 1031 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 1032 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
1033 | |
1034 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
1035 | }; | |
1da177e4 LT |
1036 | |
1037 | return rq; | |
1038 | } | |
1039 | ||
165125e1 | 1040 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
1041 | { |
1042 | struct request *rq; | |
1043 | ||
1044 | BUG_ON(rw != READ && rw != WRITE); | |
1045 | ||
d6344532 | 1046 | spin_lock_irq(q->queue_lock); |
da8303c6 | 1047 | if (gfp_mask & __GFP_WAIT) |
22e2c507 | 1048 | rq = get_request_wait(q, rw, NULL); |
da8303c6 | 1049 | else |
22e2c507 | 1050 | rq = get_request(q, rw, NULL, gfp_mask); |
da8303c6 TH |
1051 | if (!rq) |
1052 | spin_unlock_irq(q->queue_lock); | |
d6344532 | 1053 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
1054 | |
1055 | return rq; | |
1056 | } | |
1da177e4 LT |
1057 | EXPORT_SYMBOL(blk_get_request); |
1058 | ||
dc72ef4a | 1059 | /** |
79eb63e9 | 1060 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 1061 | * @q: target request queue |
79eb63e9 BH |
1062 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
1063 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 1064 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 1065 | * |
79eb63e9 BH |
1066 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
1067 | * type commands. Where the struct request needs to be farther initialized by | |
1068 | * the caller. It is passed a &struct bio, which describes the memory info of | |
1069 | * the I/O transfer. | |
dc72ef4a | 1070 | * |
79eb63e9 BH |
1071 | * The caller of blk_make_request must make sure that bi_io_vec |
1072 | * are set to describe the memory buffers. That bio_data_dir() will return | |
1073 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
1074 | * are properly set accordingly) | |
1075 | * | |
1076 | * If called under none-sleepable conditions, mapped bio buffers must not | |
1077 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
1078 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
1079 | * BUG. | |
53674ac5 JA |
1080 | * |
1081 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
1082 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
1083 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
1084 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
1085 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
1086 | * of bio_alloc(), as that avoids the mempool deadlock. | |
1087 | * If possible a big IO should be split into smaller parts when allocation | |
1088 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 1089 | */ |
79eb63e9 BH |
1090 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
1091 | gfp_t gfp_mask) | |
dc72ef4a | 1092 | { |
79eb63e9 BH |
1093 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1094 | ||
1095 | if (unlikely(!rq)) | |
1096 | return ERR_PTR(-ENOMEM); | |
1097 | ||
1098 | for_each_bio(bio) { | |
1099 | struct bio *bounce_bio = bio; | |
1100 | int ret; | |
1101 | ||
1102 | blk_queue_bounce(q, &bounce_bio); | |
1103 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1104 | if (unlikely(ret)) { | |
1105 | blk_put_request(rq); | |
1106 | return ERR_PTR(ret); | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | return rq; | |
dc72ef4a | 1111 | } |
79eb63e9 | 1112 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1113 | |
1da177e4 LT |
1114 | /** |
1115 | * blk_requeue_request - put a request back on queue | |
1116 | * @q: request queue where request should be inserted | |
1117 | * @rq: request to be inserted | |
1118 | * | |
1119 | * Description: | |
1120 | * Drivers often keep queueing requests until the hardware cannot accept | |
1121 | * more, when that condition happens we need to put the request back | |
1122 | * on the queue. Must be called with queue lock held. | |
1123 | */ | |
165125e1 | 1124 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1125 | { |
242f9dcb JA |
1126 | blk_delete_timer(rq); |
1127 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1128 | trace_block_rq_requeue(q, rq); |
2056a782 | 1129 | |
1da177e4 LT |
1130 | if (blk_rq_tagged(rq)) |
1131 | blk_queue_end_tag(q, rq); | |
1132 | ||
ba396a6c JB |
1133 | BUG_ON(blk_queued_rq(rq)); |
1134 | ||
1da177e4 LT |
1135 | elv_requeue_request(q, rq); |
1136 | } | |
1da177e4 LT |
1137 | EXPORT_SYMBOL(blk_requeue_request); |
1138 | ||
73c10101 JA |
1139 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1140 | int where) | |
1141 | { | |
1142 | drive_stat_acct(rq, 1); | |
7eaceacc | 1143 | __elv_add_request(q, rq, where); |
73c10101 JA |
1144 | } |
1145 | ||
074a7aca TH |
1146 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1147 | unsigned long now) | |
1148 | { | |
1149 | if (now == part->stamp) | |
1150 | return; | |
1151 | ||
316d315b | 1152 | if (part_in_flight(part)) { |
074a7aca | 1153 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1154 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1155 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1156 | } | |
1157 | part->stamp = now; | |
1158 | } | |
1159 | ||
1160 | /** | |
496aa8a9 RD |
1161 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1162 | * @cpu: cpu number for stats access | |
1163 | * @part: target partition | |
1da177e4 LT |
1164 | * |
1165 | * The average IO queue length and utilisation statistics are maintained | |
1166 | * by observing the current state of the queue length and the amount of | |
1167 | * time it has been in this state for. | |
1168 | * | |
1169 | * Normally, that accounting is done on IO completion, but that can result | |
1170 | * in more than a second's worth of IO being accounted for within any one | |
1171 | * second, leading to >100% utilisation. To deal with that, we call this | |
1172 | * function to do a round-off before returning the results when reading | |
1173 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1174 | * the current jiffies and restarts the counters again. | |
1175 | */ | |
c9959059 | 1176 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1177 | { |
1178 | unsigned long now = jiffies; | |
1179 | ||
074a7aca TH |
1180 | if (part->partno) |
1181 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1182 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1183 | } |
074a7aca | 1184 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1185 | |
1da177e4 LT |
1186 | /* |
1187 | * queue lock must be held | |
1188 | */ | |
165125e1 | 1189 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1190 | { |
1da177e4 LT |
1191 | if (unlikely(!q)) |
1192 | return; | |
1193 | if (unlikely(--req->ref_count)) | |
1194 | return; | |
1195 | ||
8922e16c TH |
1196 | elv_completed_request(q, req); |
1197 | ||
1cd96c24 BH |
1198 | /* this is a bio leak */ |
1199 | WARN_ON(req->bio != NULL); | |
1200 | ||
1da177e4 LT |
1201 | /* |
1202 | * Request may not have originated from ll_rw_blk. if not, | |
1203 | * it didn't come out of our reserved rq pools | |
1204 | */ | |
49171e5c | 1205 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1206 | unsigned int flags = req->cmd_flags; |
1da177e4 | 1207 | |
1da177e4 | 1208 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1209 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1210 | |
1211 | blk_free_request(q, req); | |
75eb6c37 | 1212 | freed_request(q, flags); |
1da177e4 LT |
1213 | } |
1214 | } | |
6e39b69e MC |
1215 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1216 | ||
1da177e4 LT |
1217 | void blk_put_request(struct request *req) |
1218 | { | |
8922e16c | 1219 | unsigned long flags; |
165125e1 | 1220 | struct request_queue *q = req->q; |
8922e16c | 1221 | |
52a93ba8 FT |
1222 | spin_lock_irqsave(q->queue_lock, flags); |
1223 | __blk_put_request(q, req); | |
1224 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1225 | } |
1da177e4 LT |
1226 | EXPORT_SYMBOL(blk_put_request); |
1227 | ||
66ac0280 CH |
1228 | /** |
1229 | * blk_add_request_payload - add a payload to a request | |
1230 | * @rq: request to update | |
1231 | * @page: page backing the payload | |
1232 | * @len: length of the payload. | |
1233 | * | |
1234 | * This allows to later add a payload to an already submitted request by | |
1235 | * a block driver. The driver needs to take care of freeing the payload | |
1236 | * itself. | |
1237 | * | |
1238 | * Note that this is a quite horrible hack and nothing but handling of | |
1239 | * discard requests should ever use it. | |
1240 | */ | |
1241 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1242 | unsigned int len) | |
1243 | { | |
1244 | struct bio *bio = rq->bio; | |
1245 | ||
1246 | bio->bi_io_vec->bv_page = page; | |
1247 | bio->bi_io_vec->bv_offset = 0; | |
1248 | bio->bi_io_vec->bv_len = len; | |
1249 | ||
1250 | bio->bi_size = len; | |
1251 | bio->bi_vcnt = 1; | |
1252 | bio->bi_phys_segments = 1; | |
1253 | ||
1254 | rq->__data_len = rq->resid_len = len; | |
1255 | rq->nr_phys_segments = 1; | |
1256 | rq->buffer = bio_data(bio); | |
1257 | } | |
1258 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1259 | ||
73c10101 JA |
1260 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1261 | struct bio *bio) | |
1262 | { | |
1263 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1264 | ||
73c10101 JA |
1265 | if (!ll_back_merge_fn(q, req, bio)) |
1266 | return false; | |
1267 | ||
1268 | trace_block_bio_backmerge(q, bio); | |
1269 | ||
1270 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1271 | blk_rq_set_mixed_merge(req); | |
1272 | ||
1273 | req->biotail->bi_next = bio; | |
1274 | req->biotail = bio; | |
1275 | req->__data_len += bio->bi_size; | |
1276 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1277 | ||
1278 | drive_stat_acct(req, 0); | |
1279 | return true; | |
1280 | } | |
1281 | ||
1282 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1283 | struct request *req, struct bio *bio) | |
1284 | { | |
1285 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1286 | |
73c10101 JA |
1287 | if (!ll_front_merge_fn(q, req, bio)) |
1288 | return false; | |
1289 | ||
1290 | trace_block_bio_frontmerge(q, bio); | |
1291 | ||
1292 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1293 | blk_rq_set_mixed_merge(req); | |
1294 | ||
73c10101 JA |
1295 | bio->bi_next = req->bio; |
1296 | req->bio = bio; | |
1297 | ||
1298 | /* | |
1299 | * may not be valid. if the low level driver said | |
1300 | * it didn't need a bounce buffer then it better | |
1301 | * not touch req->buffer either... | |
1302 | */ | |
1303 | req->buffer = bio_data(bio); | |
1304 | req->__sector = bio->bi_sector; | |
1305 | req->__data_len += bio->bi_size; | |
1306 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1307 | ||
1308 | drive_stat_acct(req, 0); | |
1309 | return true; | |
1310 | } | |
1311 | ||
bd87b589 TH |
1312 | /** |
1313 | * attempt_plug_merge - try to merge with %current's plugged list | |
1314 | * @q: request_queue new bio is being queued at | |
1315 | * @bio: new bio being queued | |
1316 | * @request_count: out parameter for number of traversed plugged requests | |
1317 | * | |
1318 | * Determine whether @bio being queued on @q can be merged with a request | |
1319 | * on %current's plugged list. Returns %true if merge was successful, | |
1320 | * otherwise %false. | |
1321 | * | |
07c2bd37 TH |
1322 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1323 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1324 | * than scheduling, and the request, while may have elvpriv data, is not | |
1325 | * added on the elevator at this point. In addition, we don't have | |
1326 | * reliable access to the elevator outside queue lock. Only check basic | |
1327 | * merging parameters without querying the elevator. | |
73c10101 | 1328 | */ |
bd87b589 TH |
1329 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1330 | unsigned int *request_count) | |
73c10101 JA |
1331 | { |
1332 | struct blk_plug *plug; | |
1333 | struct request *rq; | |
1334 | bool ret = false; | |
1335 | ||
bd87b589 | 1336 | plug = current->plug; |
73c10101 JA |
1337 | if (!plug) |
1338 | goto out; | |
56ebdaf2 | 1339 | *request_count = 0; |
73c10101 JA |
1340 | |
1341 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1342 | int el_ret; | |
1343 | ||
56ebdaf2 SL |
1344 | (*request_count)++; |
1345 | ||
07c2bd37 | 1346 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1347 | continue; |
1348 | ||
050c8ea8 | 1349 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1350 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1351 | ret = bio_attempt_back_merge(q, rq, bio); | |
1352 | if (ret) | |
1353 | break; | |
1354 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1355 | ret = bio_attempt_front_merge(q, rq, bio); | |
1356 | if (ret) | |
1357 | break; | |
1358 | } | |
1359 | } | |
1360 | out: | |
1361 | return ret; | |
1362 | } | |
1363 | ||
86db1e29 | 1364 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1365 | { |
4aff5e23 | 1366 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1367 | |
7b6d91da CH |
1368 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1369 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1370 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1371 | |
52d9e675 | 1372 | req->errors = 0; |
a2dec7b3 | 1373 | req->__sector = bio->bi_sector; |
52d9e675 | 1374 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1375 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1376 | } |
1377 | ||
5a7bbad2 | 1378 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1379 | { |
5e00d1b5 | 1380 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1381 | struct blk_plug *plug; |
1382 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1383 | struct request *req; | |
56ebdaf2 | 1384 | unsigned int request_count = 0; |
1da177e4 | 1385 | |
1da177e4 LT |
1386 | /* |
1387 | * low level driver can indicate that it wants pages above a | |
1388 | * certain limit bounced to low memory (ie for highmem, or even | |
1389 | * ISA dma in theory) | |
1390 | */ | |
1391 | blk_queue_bounce(q, &bio); | |
1392 | ||
4fed947c | 1393 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1394 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1395 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1396 | goto get_rq; |
1397 | } | |
1398 | ||
73c10101 JA |
1399 | /* |
1400 | * Check if we can merge with the plugged list before grabbing | |
1401 | * any locks. | |
1402 | */ | |
bd87b589 | 1403 | if (attempt_plug_merge(q, bio, &request_count)) |
5a7bbad2 | 1404 | return; |
1da177e4 | 1405 | |
73c10101 | 1406 | spin_lock_irq(q->queue_lock); |
2056a782 | 1407 | |
73c10101 JA |
1408 | el_ret = elv_merge(q, &req, bio); |
1409 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1410 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1411 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1412 | if (!attempt_back_merge(q, req)) |
1413 | elv_merged_request(q, req, el_ret); | |
1414 | goto out_unlock; | |
1415 | } | |
1416 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1417 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1418 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1419 | if (!attempt_front_merge(q, req)) |
1420 | elv_merged_request(q, req, el_ret); | |
1421 | goto out_unlock; | |
80a761fd | 1422 | } |
1da177e4 LT |
1423 | } |
1424 | ||
450991bc | 1425 | get_rq: |
7749a8d4 JA |
1426 | /* |
1427 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1428 | * but we need to set it earlier to expose the sync flag to the | |
1429 | * rq allocator and io schedulers. | |
1430 | */ | |
1431 | rw_flags = bio_data_dir(bio); | |
1432 | if (sync) | |
7b6d91da | 1433 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1434 | |
1da177e4 | 1435 | /* |
450991bc | 1436 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1437 | * Returns with the queue unlocked. |
450991bc | 1438 | */ |
7749a8d4 | 1439 | req = get_request_wait(q, rw_flags, bio); |
da8303c6 TH |
1440 | if (unlikely(!req)) { |
1441 | bio_endio(bio, -ENODEV); /* @q is dead */ | |
1442 | goto out_unlock; | |
1443 | } | |
d6344532 | 1444 | |
450991bc NP |
1445 | /* |
1446 | * After dropping the lock and possibly sleeping here, our request | |
1447 | * may now be mergeable after it had proven unmergeable (above). | |
1448 | * We don't worry about that case for efficiency. It won't happen | |
1449 | * often, and the elevators are able to handle it. | |
1da177e4 | 1450 | */ |
52d9e675 | 1451 | init_request_from_bio(req, bio); |
1da177e4 | 1452 | |
9562ad9a | 1453 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1454 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1455 | |
1456 | plug = current->plug; | |
721a9602 | 1457 | if (plug) { |
dc6d36c9 JA |
1458 | /* |
1459 | * If this is the first request added after a plug, fire | |
1460 | * of a plug trace. If others have been added before, check | |
1461 | * if we have multiple devices in this plug. If so, make a | |
1462 | * note to sort the list before dispatch. | |
1463 | */ | |
1464 | if (list_empty(&plug->list)) | |
1465 | trace_block_plug(q); | |
3540d5e8 SL |
1466 | else { |
1467 | if (!plug->should_sort) { | |
1468 | struct request *__rq; | |
73c10101 | 1469 | |
3540d5e8 SL |
1470 | __rq = list_entry_rq(plug->list.prev); |
1471 | if (__rq->q != q) | |
1472 | plug->should_sort = 1; | |
1473 | } | |
019ceb7d | 1474 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
3540d5e8 | 1475 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1476 | trace_block_plug(q); |
1477 | } | |
73c10101 | 1478 | } |
73c10101 JA |
1479 | list_add_tail(&req->queuelist, &plug->list); |
1480 | drive_stat_acct(req, 1); | |
1481 | } else { | |
1482 | spin_lock_irq(q->queue_lock); | |
1483 | add_acct_request(q, req, where); | |
24ecfbe2 | 1484 | __blk_run_queue(q); |
73c10101 JA |
1485 | out_unlock: |
1486 | spin_unlock_irq(q->queue_lock); | |
1487 | } | |
1da177e4 | 1488 | } |
c20e8de2 | 1489 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1490 | |
1491 | /* | |
1492 | * If bio->bi_dev is a partition, remap the location | |
1493 | */ | |
1494 | static inline void blk_partition_remap(struct bio *bio) | |
1495 | { | |
1496 | struct block_device *bdev = bio->bi_bdev; | |
1497 | ||
bf2de6f5 | 1498 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1499 | struct hd_struct *p = bdev->bd_part; |
1500 | ||
1da177e4 LT |
1501 | bio->bi_sector += p->start_sect; |
1502 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1503 | |
d07335e5 MS |
1504 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1505 | bdev->bd_dev, | |
1506 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1507 | } |
1508 | } | |
1509 | ||
1da177e4 LT |
1510 | static void handle_bad_sector(struct bio *bio) |
1511 | { | |
1512 | char b[BDEVNAME_SIZE]; | |
1513 | ||
1514 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1515 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1516 | bdevname(bio->bi_bdev, b), | |
1517 | bio->bi_rw, | |
1518 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1519 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1520 | |
1521 | set_bit(BIO_EOF, &bio->bi_flags); | |
1522 | } | |
1523 | ||
c17bb495 AM |
1524 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1525 | ||
1526 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1527 | ||
1528 | static int __init setup_fail_make_request(char *str) | |
1529 | { | |
1530 | return setup_fault_attr(&fail_make_request, str); | |
1531 | } | |
1532 | __setup("fail_make_request=", setup_fail_make_request); | |
1533 | ||
b2c9cd37 | 1534 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1535 | { |
b2c9cd37 | 1536 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1537 | } |
1538 | ||
1539 | static int __init fail_make_request_debugfs(void) | |
1540 | { | |
dd48c085 AM |
1541 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1542 | NULL, &fail_make_request); | |
1543 | ||
1544 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; | |
c17bb495 AM |
1545 | } |
1546 | ||
1547 | late_initcall(fail_make_request_debugfs); | |
1548 | ||
1549 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1550 | ||
b2c9cd37 AM |
1551 | static inline bool should_fail_request(struct hd_struct *part, |
1552 | unsigned int bytes) | |
c17bb495 | 1553 | { |
b2c9cd37 | 1554 | return false; |
c17bb495 AM |
1555 | } |
1556 | ||
1557 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1558 | ||
c07e2b41 JA |
1559 | /* |
1560 | * Check whether this bio extends beyond the end of the device. | |
1561 | */ | |
1562 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1563 | { | |
1564 | sector_t maxsector; | |
1565 | ||
1566 | if (!nr_sectors) | |
1567 | return 0; | |
1568 | ||
1569 | /* Test device or partition size, when known. */ | |
77304d2a | 1570 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1571 | if (maxsector) { |
1572 | sector_t sector = bio->bi_sector; | |
1573 | ||
1574 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1575 | /* | |
1576 | * This may well happen - the kernel calls bread() | |
1577 | * without checking the size of the device, e.g., when | |
1578 | * mounting a device. | |
1579 | */ | |
1580 | handle_bad_sector(bio); | |
1581 | return 1; | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | return 0; | |
1586 | } | |
1587 | ||
27a84d54 CH |
1588 | static noinline_for_stack bool |
1589 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1590 | { |
165125e1 | 1591 | struct request_queue *q; |
5a7bbad2 | 1592 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1593 | int err = -EIO; |
5a7bbad2 CH |
1594 | char b[BDEVNAME_SIZE]; |
1595 | struct hd_struct *part; | |
1da177e4 LT |
1596 | |
1597 | might_sleep(); | |
1da177e4 | 1598 | |
c07e2b41 JA |
1599 | if (bio_check_eod(bio, nr_sectors)) |
1600 | goto end_io; | |
1da177e4 | 1601 | |
5a7bbad2 CH |
1602 | q = bdev_get_queue(bio->bi_bdev); |
1603 | if (unlikely(!q)) { | |
1604 | printk(KERN_ERR | |
1605 | "generic_make_request: Trying to access " | |
1606 | "nonexistent block-device %s (%Lu)\n", | |
1607 | bdevname(bio->bi_bdev, b), | |
1608 | (long long) bio->bi_sector); | |
1609 | goto end_io; | |
1610 | } | |
c17bb495 | 1611 | |
5a7bbad2 CH |
1612 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
1613 | nr_sectors > queue_max_hw_sectors(q))) { | |
1614 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1615 | bdevname(bio->bi_bdev, b), | |
1616 | bio_sectors(bio), | |
1617 | queue_max_hw_sectors(q)); | |
1618 | goto end_io; | |
1619 | } | |
1da177e4 | 1620 | |
5a7bbad2 CH |
1621 | part = bio->bi_bdev->bd_part; |
1622 | if (should_fail_request(part, bio->bi_size) || | |
1623 | should_fail_request(&part_to_disk(part)->part0, | |
1624 | bio->bi_size)) | |
1625 | goto end_io; | |
2056a782 | 1626 | |
5a7bbad2 CH |
1627 | /* |
1628 | * If this device has partitions, remap block n | |
1629 | * of partition p to block n+start(p) of the disk. | |
1630 | */ | |
1631 | blk_partition_remap(bio); | |
2056a782 | 1632 | |
5a7bbad2 CH |
1633 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1634 | goto end_io; | |
a7384677 | 1635 | |
5a7bbad2 CH |
1636 | if (bio_check_eod(bio, nr_sectors)) |
1637 | goto end_io; | |
1e87901e | 1638 | |
5a7bbad2 CH |
1639 | /* |
1640 | * Filter flush bio's early so that make_request based | |
1641 | * drivers without flush support don't have to worry | |
1642 | * about them. | |
1643 | */ | |
1644 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1645 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1646 | if (!nr_sectors) { | |
1647 | err = 0; | |
51fd77bd JA |
1648 | goto end_io; |
1649 | } | |
5a7bbad2 | 1650 | } |
5ddfe969 | 1651 | |
5a7bbad2 CH |
1652 | if ((bio->bi_rw & REQ_DISCARD) && |
1653 | (!blk_queue_discard(q) || | |
1654 | ((bio->bi_rw & REQ_SECURE) && | |
1655 | !blk_queue_secdiscard(q)))) { | |
1656 | err = -EOPNOTSUPP; | |
1657 | goto end_io; | |
1658 | } | |
01edede4 | 1659 | |
bc16a4f9 TH |
1660 | if (blk_throtl_bio(q, bio)) |
1661 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1662 | |
5a7bbad2 | 1663 | trace_block_bio_queue(q, bio); |
27a84d54 | 1664 | return true; |
a7384677 TH |
1665 | |
1666 | end_io: | |
1667 | bio_endio(bio, err); | |
27a84d54 | 1668 | return false; |
1da177e4 LT |
1669 | } |
1670 | ||
27a84d54 CH |
1671 | /** |
1672 | * generic_make_request - hand a buffer to its device driver for I/O | |
1673 | * @bio: The bio describing the location in memory and on the device. | |
1674 | * | |
1675 | * generic_make_request() is used to make I/O requests of block | |
1676 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1677 | * to be done. | |
1678 | * | |
1679 | * generic_make_request() does not return any status. The | |
1680 | * success/failure status of the request, along with notification of | |
1681 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1682 | * function described (one day) else where. | |
1683 | * | |
1684 | * The caller of generic_make_request must make sure that bi_io_vec | |
1685 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1686 | * set to describe the device address, and the | |
1687 | * bi_end_io and optionally bi_private are set to describe how | |
1688 | * completion notification should be signaled. | |
1689 | * | |
1690 | * generic_make_request and the drivers it calls may use bi_next if this | |
1691 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1692 | * a lower device by calling into generic_make_request recursively, which | |
1693 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1694 | */ |
1695 | void generic_make_request(struct bio *bio) | |
1696 | { | |
bddd87c7 AM |
1697 | struct bio_list bio_list_on_stack; |
1698 | ||
27a84d54 CH |
1699 | if (!generic_make_request_checks(bio)) |
1700 | return; | |
1701 | ||
1702 | /* | |
1703 | * We only want one ->make_request_fn to be active at a time, else | |
1704 | * stack usage with stacked devices could be a problem. So use | |
1705 | * current->bio_list to keep a list of requests submited by a | |
1706 | * make_request_fn function. current->bio_list is also used as a | |
1707 | * flag to say if generic_make_request is currently active in this | |
1708 | * task or not. If it is NULL, then no make_request is active. If | |
1709 | * it is non-NULL, then a make_request is active, and new requests | |
1710 | * should be added at the tail | |
1711 | */ | |
bddd87c7 | 1712 | if (current->bio_list) { |
bddd87c7 | 1713 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1714 | return; |
1715 | } | |
27a84d54 | 1716 | |
d89d8796 NB |
1717 | /* following loop may be a bit non-obvious, and so deserves some |
1718 | * explanation. | |
1719 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1720 | * ensure that) so we have a list with a single bio. | |
1721 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1722 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1723 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1724 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1725 | * through a recursive call to generic_make_request. If it |
1726 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1727 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1728 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1729 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1730 | */ |
1731 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1732 | bio_list_init(&bio_list_on_stack); |
1733 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1734 | do { |
27a84d54 CH |
1735 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1736 | ||
1737 | q->make_request_fn(q, bio); | |
1738 | ||
bddd87c7 | 1739 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1740 | } while (bio); |
bddd87c7 | 1741 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1742 | } |
1da177e4 LT |
1743 | EXPORT_SYMBOL(generic_make_request); |
1744 | ||
1745 | /** | |
710027a4 | 1746 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1747 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1748 | * @bio: The &struct bio which describes the I/O | |
1749 | * | |
1750 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1751 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1752 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1753 | * |
1754 | */ | |
1755 | void submit_bio(int rw, struct bio *bio) | |
1756 | { | |
1757 | int count = bio_sectors(bio); | |
1758 | ||
22e2c507 | 1759 | bio->bi_rw |= rw; |
1da177e4 | 1760 | |
bf2de6f5 JA |
1761 | /* |
1762 | * If it's a regular read/write or a barrier with data attached, | |
1763 | * go through the normal accounting stuff before submission. | |
1764 | */ | |
3ffb52e7 | 1765 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1766 | if (rw & WRITE) { |
1767 | count_vm_events(PGPGOUT, count); | |
1768 | } else { | |
1769 | task_io_account_read(bio->bi_size); | |
1770 | count_vm_events(PGPGIN, count); | |
1771 | } | |
1772 | ||
1773 | if (unlikely(block_dump)) { | |
1774 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1775 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1776 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1777 | (rw & WRITE) ? "WRITE" : "READ", |
1778 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1779 | bdevname(bio->bi_bdev, b), |
1780 | count); | |
bf2de6f5 | 1781 | } |
1da177e4 LT |
1782 | } |
1783 | ||
1784 | generic_make_request(bio); | |
1785 | } | |
1da177e4 LT |
1786 | EXPORT_SYMBOL(submit_bio); |
1787 | ||
82124d60 KU |
1788 | /** |
1789 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1790 | * @q: the queue | |
1791 | * @rq: the request being checked | |
1792 | * | |
1793 | * Description: | |
1794 | * @rq may have been made based on weaker limitations of upper-level queues | |
1795 | * in request stacking drivers, and it may violate the limitation of @q. | |
1796 | * Since the block layer and the underlying device driver trust @rq | |
1797 | * after it is inserted to @q, it should be checked against @q before | |
1798 | * the insertion using this generic function. | |
1799 | * | |
1800 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1801 | * in some cases below, so export this function. |
82124d60 KU |
1802 | * Request stacking drivers like request-based dm may change the queue |
1803 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1804 | * Such request stacking drivers should check those requests agaist | |
1805 | * the new queue limits again when they dispatch those requests, | |
1806 | * although such checkings are also done against the old queue limits | |
1807 | * when submitting requests. | |
1808 | */ | |
1809 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1810 | { | |
3383977f S |
1811 | if (rq->cmd_flags & REQ_DISCARD) |
1812 | return 0; | |
1813 | ||
ae03bf63 MP |
1814 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1815 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1816 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1817 | return -EIO; | |
1818 | } | |
1819 | ||
1820 | /* | |
1821 | * queue's settings related to segment counting like q->bounce_pfn | |
1822 | * may differ from that of other stacking queues. | |
1823 | * Recalculate it to check the request correctly on this queue's | |
1824 | * limitation. | |
1825 | */ | |
1826 | blk_recalc_rq_segments(rq); | |
8a78362c | 1827 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1828 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1829 | return -EIO; | |
1830 | } | |
1831 | ||
1832 | return 0; | |
1833 | } | |
1834 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1835 | ||
1836 | /** | |
1837 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1838 | * @q: the queue to submit the request | |
1839 | * @rq: the request being queued | |
1840 | */ | |
1841 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1842 | { | |
1843 | unsigned long flags; | |
4853abaa | 1844 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
1845 | |
1846 | if (blk_rq_check_limits(q, rq)) | |
1847 | return -EIO; | |
1848 | ||
b2c9cd37 AM |
1849 | if (rq->rq_disk && |
1850 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 1851 | return -EIO; |
82124d60 KU |
1852 | |
1853 | spin_lock_irqsave(q->queue_lock, flags); | |
8ba61435 TH |
1854 | if (unlikely(blk_queue_dead(q))) { |
1855 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1856 | return -ENODEV; | |
1857 | } | |
82124d60 KU |
1858 | |
1859 | /* | |
1860 | * Submitting request must be dequeued before calling this function | |
1861 | * because it will be linked to another request_queue | |
1862 | */ | |
1863 | BUG_ON(blk_queued_rq(rq)); | |
1864 | ||
4853abaa JM |
1865 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
1866 | where = ELEVATOR_INSERT_FLUSH; | |
1867 | ||
1868 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
1869 | if (where == ELEVATOR_INSERT_FLUSH) |
1870 | __blk_run_queue(q); | |
82124d60 KU |
1871 | spin_unlock_irqrestore(q->queue_lock, flags); |
1872 | ||
1873 | return 0; | |
1874 | } | |
1875 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1876 | ||
80a761fd TH |
1877 | /** |
1878 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1879 | * @rq: request to examine | |
1880 | * | |
1881 | * Description: | |
1882 | * A request could be merge of IOs which require different failure | |
1883 | * handling. This function determines the number of bytes which | |
1884 | * can be failed from the beginning of the request without | |
1885 | * crossing into area which need to be retried further. | |
1886 | * | |
1887 | * Return: | |
1888 | * The number of bytes to fail. | |
1889 | * | |
1890 | * Context: | |
1891 | * queue_lock must be held. | |
1892 | */ | |
1893 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1894 | { | |
1895 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1896 | unsigned int bytes = 0; | |
1897 | struct bio *bio; | |
1898 | ||
1899 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1900 | return blk_rq_bytes(rq); | |
1901 | ||
1902 | /* | |
1903 | * Currently the only 'mixing' which can happen is between | |
1904 | * different fastfail types. We can safely fail portions | |
1905 | * which have all the failfast bits that the first one has - | |
1906 | * the ones which are at least as eager to fail as the first | |
1907 | * one. | |
1908 | */ | |
1909 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1910 | if ((bio->bi_rw & ff) != ff) | |
1911 | break; | |
1912 | bytes += bio->bi_size; | |
1913 | } | |
1914 | ||
1915 | /* this could lead to infinite loop */ | |
1916 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1917 | return bytes; | |
1918 | } | |
1919 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1920 | ||
bc58ba94 JA |
1921 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1922 | { | |
c2553b58 | 1923 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1924 | const int rw = rq_data_dir(req); |
1925 | struct hd_struct *part; | |
1926 | int cpu; | |
1927 | ||
1928 | cpu = part_stat_lock(); | |
09e099d4 | 1929 | part = req->part; |
bc58ba94 JA |
1930 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1931 | part_stat_unlock(); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | static void blk_account_io_done(struct request *req) | |
1936 | { | |
bc58ba94 | 1937 | /* |
dd4c133f TH |
1938 | * Account IO completion. flush_rq isn't accounted as a |
1939 | * normal IO on queueing nor completion. Accounting the | |
1940 | * containing request is enough. | |
bc58ba94 | 1941 | */ |
414b4ff5 | 1942 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1943 | unsigned long duration = jiffies - req->start_time; |
1944 | const int rw = rq_data_dir(req); | |
1945 | struct hd_struct *part; | |
1946 | int cpu; | |
1947 | ||
1948 | cpu = part_stat_lock(); | |
09e099d4 | 1949 | part = req->part; |
bc58ba94 JA |
1950 | |
1951 | part_stat_inc(cpu, part, ios[rw]); | |
1952 | part_stat_add(cpu, part, ticks[rw], duration); | |
1953 | part_round_stats(cpu, part); | |
316d315b | 1954 | part_dec_in_flight(part, rw); |
bc58ba94 | 1955 | |
6c23a968 | 1956 | hd_struct_put(part); |
bc58ba94 JA |
1957 | part_stat_unlock(); |
1958 | } | |
1959 | } | |
1960 | ||
3bcddeac | 1961 | /** |
9934c8c0 TH |
1962 | * blk_peek_request - peek at the top of a request queue |
1963 | * @q: request queue to peek at | |
1964 | * | |
1965 | * Description: | |
1966 | * Return the request at the top of @q. The returned request | |
1967 | * should be started using blk_start_request() before LLD starts | |
1968 | * processing it. | |
1969 | * | |
1970 | * Return: | |
1971 | * Pointer to the request at the top of @q if available. Null | |
1972 | * otherwise. | |
1973 | * | |
1974 | * Context: | |
1975 | * queue_lock must be held. | |
1976 | */ | |
1977 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
1978 | { |
1979 | struct request *rq; | |
1980 | int ret; | |
1981 | ||
1982 | while ((rq = __elv_next_request(q)) != NULL) { | |
1983 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1984 | /* | |
1985 | * This is the first time the device driver | |
1986 | * sees this request (possibly after | |
1987 | * requeueing). Notify IO scheduler. | |
1988 | */ | |
33659ebb | 1989 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
1990 | elv_activate_rq(q, rq); |
1991 | ||
1992 | /* | |
1993 | * just mark as started even if we don't start | |
1994 | * it, a request that has been delayed should | |
1995 | * not be passed by new incoming requests | |
1996 | */ | |
1997 | rq->cmd_flags |= REQ_STARTED; | |
1998 | trace_block_rq_issue(q, rq); | |
1999 | } | |
2000 | ||
2001 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2002 | q->end_sector = rq_end_sector(rq); | |
2003 | q->boundary_rq = NULL; | |
2004 | } | |
2005 | ||
2006 | if (rq->cmd_flags & REQ_DONTPREP) | |
2007 | break; | |
2008 | ||
2e46e8b2 | 2009 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2010 | /* |
2011 | * make sure space for the drain appears we | |
2012 | * know we can do this because max_hw_segments | |
2013 | * has been adjusted to be one fewer than the | |
2014 | * device can handle | |
2015 | */ | |
2016 | rq->nr_phys_segments++; | |
2017 | } | |
2018 | ||
2019 | if (!q->prep_rq_fn) | |
2020 | break; | |
2021 | ||
2022 | ret = q->prep_rq_fn(q, rq); | |
2023 | if (ret == BLKPREP_OK) { | |
2024 | break; | |
2025 | } else if (ret == BLKPREP_DEFER) { | |
2026 | /* | |
2027 | * the request may have been (partially) prepped. | |
2028 | * we need to keep this request in the front to | |
2029 | * avoid resource deadlock. REQ_STARTED will | |
2030 | * prevent other fs requests from passing this one. | |
2031 | */ | |
2e46e8b2 | 2032 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
2033 | !(rq->cmd_flags & REQ_DONTPREP)) { |
2034 | /* | |
2035 | * remove the space for the drain we added | |
2036 | * so that we don't add it again | |
2037 | */ | |
2038 | --rq->nr_phys_segments; | |
2039 | } | |
2040 | ||
2041 | rq = NULL; | |
2042 | break; | |
2043 | } else if (ret == BLKPREP_KILL) { | |
2044 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
2045 | /* |
2046 | * Mark this request as started so we don't trigger | |
2047 | * any debug logic in the end I/O path. | |
2048 | */ | |
2049 | blk_start_request(rq); | |
40cbbb78 | 2050 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
2051 | } else { |
2052 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2053 | break; | |
2054 | } | |
2055 | } | |
2056 | ||
2057 | return rq; | |
2058 | } | |
9934c8c0 | 2059 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2060 | |
9934c8c0 | 2061 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2062 | { |
9934c8c0 TH |
2063 | struct request_queue *q = rq->q; |
2064 | ||
158dbda0 TH |
2065 | BUG_ON(list_empty(&rq->queuelist)); |
2066 | BUG_ON(ELV_ON_HASH(rq)); | |
2067 | ||
2068 | list_del_init(&rq->queuelist); | |
2069 | ||
2070 | /* | |
2071 | * the time frame between a request being removed from the lists | |
2072 | * and to it is freed is accounted as io that is in progress at | |
2073 | * the driver side. | |
2074 | */ | |
9195291e | 2075 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2076 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2077 | set_io_start_time_ns(rq); |
2078 | } | |
158dbda0 TH |
2079 | } |
2080 | ||
9934c8c0 TH |
2081 | /** |
2082 | * blk_start_request - start request processing on the driver | |
2083 | * @req: request to dequeue | |
2084 | * | |
2085 | * Description: | |
2086 | * Dequeue @req and start timeout timer on it. This hands off the | |
2087 | * request to the driver. | |
2088 | * | |
2089 | * Block internal functions which don't want to start timer should | |
2090 | * call blk_dequeue_request(). | |
2091 | * | |
2092 | * Context: | |
2093 | * queue_lock must be held. | |
2094 | */ | |
2095 | void blk_start_request(struct request *req) | |
2096 | { | |
2097 | blk_dequeue_request(req); | |
2098 | ||
2099 | /* | |
5f49f631 TH |
2100 | * We are now handing the request to the hardware, initialize |
2101 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2102 | */ |
5f49f631 | 2103 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2104 | if (unlikely(blk_bidi_rq(req))) |
2105 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2106 | ||
9934c8c0 TH |
2107 | blk_add_timer(req); |
2108 | } | |
2109 | EXPORT_SYMBOL(blk_start_request); | |
2110 | ||
2111 | /** | |
2112 | * blk_fetch_request - fetch a request from a request queue | |
2113 | * @q: request queue to fetch a request from | |
2114 | * | |
2115 | * Description: | |
2116 | * Return the request at the top of @q. The request is started on | |
2117 | * return and LLD can start processing it immediately. | |
2118 | * | |
2119 | * Return: | |
2120 | * Pointer to the request at the top of @q if available. Null | |
2121 | * otherwise. | |
2122 | * | |
2123 | * Context: | |
2124 | * queue_lock must be held. | |
2125 | */ | |
2126 | struct request *blk_fetch_request(struct request_queue *q) | |
2127 | { | |
2128 | struct request *rq; | |
2129 | ||
2130 | rq = blk_peek_request(q); | |
2131 | if (rq) | |
2132 | blk_start_request(rq); | |
2133 | return rq; | |
2134 | } | |
2135 | EXPORT_SYMBOL(blk_fetch_request); | |
2136 | ||
3bcddeac | 2137 | /** |
2e60e022 | 2138 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2139 | * @req: the request being processed |
710027a4 | 2140 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2141 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2142 | * |
2143 | * Description: | |
8ebf9756 RD |
2144 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2145 | * the request structure even if @req doesn't have leftover. | |
2146 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2147 | * |
2148 | * This special helper function is only for request stacking drivers | |
2149 | * (e.g. request-based dm) so that they can handle partial completion. | |
2150 | * Actual device drivers should use blk_end_request instead. | |
2151 | * | |
2152 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2153 | * %false return from this function. | |
3bcddeac KU |
2154 | * |
2155 | * Return: | |
2e60e022 TH |
2156 | * %false - this request doesn't have any more data |
2157 | * %true - this request has more data | |
3bcddeac | 2158 | **/ |
2e60e022 | 2159 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2160 | { |
5450d3e1 | 2161 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2162 | struct bio *bio; |
2163 | ||
2e60e022 TH |
2164 | if (!req->bio) |
2165 | return false; | |
2166 | ||
5f3ea37c | 2167 | trace_block_rq_complete(req->q, req); |
2056a782 | 2168 | |
1da177e4 | 2169 | /* |
6f41469c TH |
2170 | * For fs requests, rq is just carrier of independent bio's |
2171 | * and each partial completion should be handled separately. | |
2172 | * Reset per-request error on each partial completion. | |
2173 | * | |
2174 | * TODO: tj: This is too subtle. It would be better to let | |
2175 | * low level drivers do what they see fit. | |
1da177e4 | 2176 | */ |
33659ebb | 2177 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2178 | req->errors = 0; |
2179 | ||
33659ebb CH |
2180 | if (error && req->cmd_type == REQ_TYPE_FS && |
2181 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2182 | char *error_type; |
2183 | ||
2184 | switch (error) { | |
2185 | case -ENOLINK: | |
2186 | error_type = "recoverable transport"; | |
2187 | break; | |
2188 | case -EREMOTEIO: | |
2189 | error_type = "critical target"; | |
2190 | break; | |
2191 | case -EBADE: | |
2192 | error_type = "critical nexus"; | |
2193 | break; | |
2194 | case -EIO: | |
2195 | default: | |
2196 | error_type = "I/O"; | |
2197 | break; | |
2198 | } | |
2199 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2200 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2201 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2202 | } |
2203 | ||
bc58ba94 | 2204 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2205 | |
1da177e4 LT |
2206 | total_bytes = bio_nbytes = 0; |
2207 | while ((bio = req->bio) != NULL) { | |
2208 | int nbytes; | |
2209 | ||
2210 | if (nr_bytes >= bio->bi_size) { | |
2211 | req->bio = bio->bi_next; | |
2212 | nbytes = bio->bi_size; | |
5bb23a68 | 2213 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2214 | next_idx = 0; |
2215 | bio_nbytes = 0; | |
2216 | } else { | |
2217 | int idx = bio->bi_idx + next_idx; | |
2218 | ||
af498d7f | 2219 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2220 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2221 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2222 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2223 | break; |
2224 | } | |
2225 | ||
2226 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2227 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2228 | ||
2229 | /* | |
2230 | * not a complete bvec done | |
2231 | */ | |
2232 | if (unlikely(nbytes > nr_bytes)) { | |
2233 | bio_nbytes += nr_bytes; | |
2234 | total_bytes += nr_bytes; | |
2235 | break; | |
2236 | } | |
2237 | ||
2238 | /* | |
2239 | * advance to the next vector | |
2240 | */ | |
2241 | next_idx++; | |
2242 | bio_nbytes += nbytes; | |
2243 | } | |
2244 | ||
2245 | total_bytes += nbytes; | |
2246 | nr_bytes -= nbytes; | |
2247 | ||
6728cb0e JA |
2248 | bio = req->bio; |
2249 | if (bio) { | |
1da177e4 LT |
2250 | /* |
2251 | * end more in this run, or just return 'not-done' | |
2252 | */ | |
2253 | if (unlikely(nr_bytes <= 0)) | |
2254 | break; | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | /* | |
2259 | * completely done | |
2260 | */ | |
2e60e022 TH |
2261 | if (!req->bio) { |
2262 | /* | |
2263 | * Reset counters so that the request stacking driver | |
2264 | * can find how many bytes remain in the request | |
2265 | * later. | |
2266 | */ | |
a2dec7b3 | 2267 | req->__data_len = 0; |
2e60e022 TH |
2268 | return false; |
2269 | } | |
1da177e4 LT |
2270 | |
2271 | /* | |
2272 | * if the request wasn't completed, update state | |
2273 | */ | |
2274 | if (bio_nbytes) { | |
5bb23a68 | 2275 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2276 | bio->bi_idx += next_idx; |
2277 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2278 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2279 | } | |
2280 | ||
a2dec7b3 | 2281 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2282 | req->buffer = bio_data(req->bio); |
2283 | ||
2284 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2285 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2286 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2287 | |
80a761fd TH |
2288 | /* mixed attributes always follow the first bio */ |
2289 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2290 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2291 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2292 | } | |
2293 | ||
2e46e8b2 TH |
2294 | /* |
2295 | * If total number of sectors is less than the first segment | |
2296 | * size, something has gone terribly wrong. | |
2297 | */ | |
2298 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2299 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2300 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2301 | } |
2302 | ||
2303 | /* recalculate the number of segments */ | |
1da177e4 | 2304 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2305 | |
2e60e022 | 2306 | return true; |
1da177e4 | 2307 | } |
2e60e022 | 2308 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2309 | |
2e60e022 TH |
2310 | static bool blk_update_bidi_request(struct request *rq, int error, |
2311 | unsigned int nr_bytes, | |
2312 | unsigned int bidi_bytes) | |
5efccd17 | 2313 | { |
2e60e022 TH |
2314 | if (blk_update_request(rq, error, nr_bytes)) |
2315 | return true; | |
5efccd17 | 2316 | |
2e60e022 TH |
2317 | /* Bidi request must be completed as a whole */ |
2318 | if (unlikely(blk_bidi_rq(rq)) && | |
2319 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2320 | return true; | |
5efccd17 | 2321 | |
e2e1a148 JA |
2322 | if (blk_queue_add_random(rq->q)) |
2323 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2324 | |
2325 | return false; | |
1da177e4 LT |
2326 | } |
2327 | ||
28018c24 JB |
2328 | /** |
2329 | * blk_unprep_request - unprepare a request | |
2330 | * @req: the request | |
2331 | * | |
2332 | * This function makes a request ready for complete resubmission (or | |
2333 | * completion). It happens only after all error handling is complete, | |
2334 | * so represents the appropriate moment to deallocate any resources | |
2335 | * that were allocated to the request in the prep_rq_fn. The queue | |
2336 | * lock is held when calling this. | |
2337 | */ | |
2338 | void blk_unprep_request(struct request *req) | |
2339 | { | |
2340 | struct request_queue *q = req->q; | |
2341 | ||
2342 | req->cmd_flags &= ~REQ_DONTPREP; | |
2343 | if (q->unprep_rq_fn) | |
2344 | q->unprep_rq_fn(q, req); | |
2345 | } | |
2346 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2347 | ||
1da177e4 LT |
2348 | /* |
2349 | * queue lock must be held | |
2350 | */ | |
2e60e022 | 2351 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2352 | { |
b8286239 KU |
2353 | if (blk_rq_tagged(req)) |
2354 | blk_queue_end_tag(req->q, req); | |
2355 | ||
ba396a6c | 2356 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2357 | |
33659ebb | 2358 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2359 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2360 | |
e78042e5 MA |
2361 | blk_delete_timer(req); |
2362 | ||
28018c24 JB |
2363 | if (req->cmd_flags & REQ_DONTPREP) |
2364 | blk_unprep_request(req); | |
2365 | ||
2366 | ||
bc58ba94 | 2367 | blk_account_io_done(req); |
b8286239 | 2368 | |
1da177e4 | 2369 | if (req->end_io) |
8ffdc655 | 2370 | req->end_io(req, error); |
b8286239 KU |
2371 | else { |
2372 | if (blk_bidi_rq(req)) | |
2373 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2374 | ||
1da177e4 | 2375 | __blk_put_request(req->q, req); |
b8286239 | 2376 | } |
1da177e4 LT |
2377 | } |
2378 | ||
3b11313a | 2379 | /** |
2e60e022 TH |
2380 | * blk_end_bidi_request - Complete a bidi request |
2381 | * @rq: the request to complete | |
2382 | * @error: %0 for success, < %0 for error | |
2383 | * @nr_bytes: number of bytes to complete @rq | |
2384 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2385 | * |
2386 | * Description: | |
e3a04fe3 | 2387 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2388 | * Drivers that supports bidi can safely call this member for any |
2389 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2390 | * just ignored. | |
336cdb40 KU |
2391 | * |
2392 | * Return: | |
2e60e022 TH |
2393 | * %false - we are done with this request |
2394 | * %true - still buffers pending for this request | |
a0cd1285 | 2395 | **/ |
b1f74493 | 2396 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2397 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2398 | { | |
336cdb40 | 2399 | struct request_queue *q = rq->q; |
2e60e022 | 2400 | unsigned long flags; |
32fab448 | 2401 | |
2e60e022 TH |
2402 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2403 | return true; | |
32fab448 | 2404 | |
336cdb40 | 2405 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2406 | blk_finish_request(rq, error); |
336cdb40 KU |
2407 | spin_unlock_irqrestore(q->queue_lock, flags); |
2408 | ||
2e60e022 | 2409 | return false; |
32fab448 KU |
2410 | } |
2411 | ||
336cdb40 | 2412 | /** |
2e60e022 TH |
2413 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2414 | * @rq: the request to complete | |
710027a4 | 2415 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2416 | * @nr_bytes: number of bytes to complete @rq |
2417 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2418 | * |
2419 | * Description: | |
2e60e022 TH |
2420 | * Identical to blk_end_bidi_request() except that queue lock is |
2421 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2422 | * |
2423 | * Return: | |
2e60e022 TH |
2424 | * %false - we are done with this request |
2425 | * %true - still buffers pending for this request | |
336cdb40 | 2426 | **/ |
4853abaa | 2427 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2428 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2429 | { |
2e60e022 TH |
2430 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2431 | return true; | |
336cdb40 | 2432 | |
2e60e022 | 2433 | blk_finish_request(rq, error); |
336cdb40 | 2434 | |
2e60e022 | 2435 | return false; |
336cdb40 | 2436 | } |
e19a3ab0 KU |
2437 | |
2438 | /** | |
2439 | * blk_end_request - Helper function for drivers to complete the request. | |
2440 | * @rq: the request being processed | |
710027a4 | 2441 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2442 | * @nr_bytes: number of bytes to complete |
2443 | * | |
2444 | * Description: | |
2445 | * Ends I/O on a number of bytes attached to @rq. | |
2446 | * If @rq has leftover, sets it up for the next range of segments. | |
2447 | * | |
2448 | * Return: | |
b1f74493 FT |
2449 | * %false - we are done with this request |
2450 | * %true - still buffers pending for this request | |
e19a3ab0 | 2451 | **/ |
b1f74493 | 2452 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2453 | { |
b1f74493 | 2454 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2455 | } |
56ad1740 | 2456 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2457 | |
2458 | /** | |
b1f74493 FT |
2459 | * blk_end_request_all - Helper function for drives to finish the request. |
2460 | * @rq: the request to finish | |
8ebf9756 | 2461 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2462 | * |
2463 | * Description: | |
b1f74493 FT |
2464 | * Completely finish @rq. |
2465 | */ | |
2466 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2467 | { |
b1f74493 FT |
2468 | bool pending; |
2469 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2470 | |
b1f74493 FT |
2471 | if (unlikely(blk_bidi_rq(rq))) |
2472 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2473 | |
b1f74493 FT |
2474 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2475 | BUG_ON(pending); | |
2476 | } | |
56ad1740 | 2477 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2478 | |
b1f74493 FT |
2479 | /** |
2480 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2481 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2482 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2483 | * |
2484 | * Description: | |
2485 | * Complete the current consecutively mapped chunk from @rq. | |
2486 | * | |
2487 | * Return: | |
2488 | * %false - we are done with this request | |
2489 | * %true - still buffers pending for this request | |
2490 | */ | |
2491 | bool blk_end_request_cur(struct request *rq, int error) | |
2492 | { | |
2493 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2494 | } |
56ad1740 | 2495 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2496 | |
80a761fd TH |
2497 | /** |
2498 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2499 | * @rq: the request to finish till the next failure boundary for | |
2500 | * @error: must be negative errno | |
2501 | * | |
2502 | * Description: | |
2503 | * Complete @rq till the next failure boundary. | |
2504 | * | |
2505 | * Return: | |
2506 | * %false - we are done with this request | |
2507 | * %true - still buffers pending for this request | |
2508 | */ | |
2509 | bool blk_end_request_err(struct request *rq, int error) | |
2510 | { | |
2511 | WARN_ON(error >= 0); | |
2512 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2513 | } | |
2514 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2515 | ||
e3a04fe3 | 2516 | /** |
b1f74493 FT |
2517 | * __blk_end_request - Helper function for drivers to complete the request. |
2518 | * @rq: the request being processed | |
2519 | * @error: %0 for success, < %0 for error | |
2520 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2521 | * |
2522 | * Description: | |
b1f74493 | 2523 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2524 | * |
2525 | * Return: | |
b1f74493 FT |
2526 | * %false - we are done with this request |
2527 | * %true - still buffers pending for this request | |
e3a04fe3 | 2528 | **/ |
b1f74493 | 2529 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2530 | { |
b1f74493 | 2531 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2532 | } |
56ad1740 | 2533 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2534 | |
32fab448 | 2535 | /** |
b1f74493 FT |
2536 | * __blk_end_request_all - Helper function for drives to finish the request. |
2537 | * @rq: the request to finish | |
8ebf9756 | 2538 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2539 | * |
2540 | * Description: | |
b1f74493 | 2541 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2542 | */ |
b1f74493 | 2543 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2544 | { |
b1f74493 FT |
2545 | bool pending; |
2546 | unsigned int bidi_bytes = 0; | |
2547 | ||
2548 | if (unlikely(blk_bidi_rq(rq))) | |
2549 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2550 | ||
2551 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2552 | BUG_ON(pending); | |
32fab448 | 2553 | } |
56ad1740 | 2554 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2555 | |
e19a3ab0 | 2556 | /** |
b1f74493 FT |
2557 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2558 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2559 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2560 | * |
2561 | * Description: | |
b1f74493 FT |
2562 | * Complete the current consecutively mapped chunk from @rq. Must |
2563 | * be called with queue lock held. | |
e19a3ab0 KU |
2564 | * |
2565 | * Return: | |
b1f74493 FT |
2566 | * %false - we are done with this request |
2567 | * %true - still buffers pending for this request | |
2568 | */ | |
2569 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2570 | { |
b1f74493 | 2571 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2572 | } |
56ad1740 | 2573 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2574 | |
80a761fd TH |
2575 | /** |
2576 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2577 | * @rq: the request to finish till the next failure boundary for | |
2578 | * @error: must be negative errno | |
2579 | * | |
2580 | * Description: | |
2581 | * Complete @rq till the next failure boundary. Must be called | |
2582 | * with queue lock held. | |
2583 | * | |
2584 | * Return: | |
2585 | * %false - we are done with this request | |
2586 | * %true - still buffers pending for this request | |
2587 | */ | |
2588 | bool __blk_end_request_err(struct request *rq, int error) | |
2589 | { | |
2590 | WARN_ON(error >= 0); | |
2591 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2592 | } | |
2593 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2594 | ||
86db1e29 JA |
2595 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2596 | struct bio *bio) | |
1da177e4 | 2597 | { |
a82afdfc | 2598 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2599 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2600 | |
fb2dce86 DW |
2601 | if (bio_has_data(bio)) { |
2602 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2603 | rq->buffer = bio_data(bio); |
2604 | } | |
a2dec7b3 | 2605 | rq->__data_len = bio->bi_size; |
1da177e4 | 2606 | rq->bio = rq->biotail = bio; |
1da177e4 | 2607 | |
66846572 N |
2608 | if (bio->bi_bdev) |
2609 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2610 | } | |
1da177e4 | 2611 | |
2d4dc890 IL |
2612 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2613 | /** | |
2614 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2615 | * @rq: the request to be flushed | |
2616 | * | |
2617 | * Description: | |
2618 | * Flush all pages in @rq. | |
2619 | */ | |
2620 | void rq_flush_dcache_pages(struct request *rq) | |
2621 | { | |
2622 | struct req_iterator iter; | |
2623 | struct bio_vec *bvec; | |
2624 | ||
2625 | rq_for_each_segment(bvec, rq, iter) | |
2626 | flush_dcache_page(bvec->bv_page); | |
2627 | } | |
2628 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2629 | #endif | |
2630 | ||
ef9e3fac KU |
2631 | /** |
2632 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2633 | * @q : the queue of the device being checked | |
2634 | * | |
2635 | * Description: | |
2636 | * Check if underlying low-level drivers of a device are busy. | |
2637 | * If the drivers want to export their busy state, they must set own | |
2638 | * exporting function using blk_queue_lld_busy() first. | |
2639 | * | |
2640 | * Basically, this function is used only by request stacking drivers | |
2641 | * to stop dispatching requests to underlying devices when underlying | |
2642 | * devices are busy. This behavior helps more I/O merging on the queue | |
2643 | * of the request stacking driver and prevents I/O throughput regression | |
2644 | * on burst I/O load. | |
2645 | * | |
2646 | * Return: | |
2647 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2648 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2649 | */ | |
2650 | int blk_lld_busy(struct request_queue *q) | |
2651 | { | |
2652 | if (q->lld_busy_fn) | |
2653 | return q->lld_busy_fn(q); | |
2654 | ||
2655 | return 0; | |
2656 | } | |
2657 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2658 | ||
b0fd271d KU |
2659 | /** |
2660 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2661 | * @rq: the clone request to be cleaned up | |
2662 | * | |
2663 | * Description: | |
2664 | * Free all bios in @rq for a cloned request. | |
2665 | */ | |
2666 | void blk_rq_unprep_clone(struct request *rq) | |
2667 | { | |
2668 | struct bio *bio; | |
2669 | ||
2670 | while ((bio = rq->bio) != NULL) { | |
2671 | rq->bio = bio->bi_next; | |
2672 | ||
2673 | bio_put(bio); | |
2674 | } | |
2675 | } | |
2676 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2677 | ||
2678 | /* | |
2679 | * Copy attributes of the original request to the clone request. | |
2680 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2681 | */ | |
2682 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2683 | { | |
2684 | dst->cpu = src->cpu; | |
3a2edd0d | 2685 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2686 | dst->cmd_type = src->cmd_type; |
2687 | dst->__sector = blk_rq_pos(src); | |
2688 | dst->__data_len = blk_rq_bytes(src); | |
2689 | dst->nr_phys_segments = src->nr_phys_segments; | |
2690 | dst->ioprio = src->ioprio; | |
2691 | dst->extra_len = src->extra_len; | |
2692 | } | |
2693 | ||
2694 | /** | |
2695 | * blk_rq_prep_clone - Helper function to setup clone request | |
2696 | * @rq: the request to be setup | |
2697 | * @rq_src: original request to be cloned | |
2698 | * @bs: bio_set that bios for clone are allocated from | |
2699 | * @gfp_mask: memory allocation mask for bio | |
2700 | * @bio_ctr: setup function to be called for each clone bio. | |
2701 | * Returns %0 for success, non %0 for failure. | |
2702 | * @data: private data to be passed to @bio_ctr | |
2703 | * | |
2704 | * Description: | |
2705 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2706 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2707 | * are not copied, and copying such parts is the caller's responsibility. | |
2708 | * Also, pages which the original bios are pointing to are not copied | |
2709 | * and the cloned bios just point same pages. | |
2710 | * So cloned bios must be completed before original bios, which means | |
2711 | * the caller must complete @rq before @rq_src. | |
2712 | */ | |
2713 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2714 | struct bio_set *bs, gfp_t gfp_mask, | |
2715 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2716 | void *data) | |
2717 | { | |
2718 | struct bio *bio, *bio_src; | |
2719 | ||
2720 | if (!bs) | |
2721 | bs = fs_bio_set; | |
2722 | ||
2723 | blk_rq_init(NULL, rq); | |
2724 | ||
2725 | __rq_for_each_bio(bio_src, rq_src) { | |
2726 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2727 | if (!bio) | |
2728 | goto free_and_out; | |
2729 | ||
2730 | __bio_clone(bio, bio_src); | |
2731 | ||
2732 | if (bio_integrity(bio_src) && | |
7878cba9 | 2733 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2734 | goto free_and_out; |
2735 | ||
2736 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2737 | goto free_and_out; | |
2738 | ||
2739 | if (rq->bio) { | |
2740 | rq->biotail->bi_next = bio; | |
2741 | rq->biotail = bio; | |
2742 | } else | |
2743 | rq->bio = rq->biotail = bio; | |
2744 | } | |
2745 | ||
2746 | __blk_rq_prep_clone(rq, rq_src); | |
2747 | ||
2748 | return 0; | |
2749 | ||
2750 | free_and_out: | |
2751 | if (bio) | |
2752 | bio_free(bio, bs); | |
2753 | blk_rq_unprep_clone(rq); | |
2754 | ||
2755 | return -ENOMEM; | |
2756 | } | |
2757 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2758 | ||
18887ad9 | 2759 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2760 | { |
2761 | return queue_work(kblockd_workqueue, work); | |
2762 | } | |
1da177e4 LT |
2763 | EXPORT_SYMBOL(kblockd_schedule_work); |
2764 | ||
e43473b7 VG |
2765 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2766 | struct delayed_work *dwork, unsigned long delay) | |
2767 | { | |
2768 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2769 | } | |
2770 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2771 | ||
73c10101 JA |
2772 | #define PLUG_MAGIC 0x91827364 |
2773 | ||
75df7136 SJ |
2774 | /** |
2775 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2776 | * @plug: The &struct blk_plug that needs to be initialized | |
2777 | * | |
2778 | * Description: | |
2779 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2780 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2781 | * blk_finish_plug(). This is important from a performance perspective, but | |
2782 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2783 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2784 | * page belonging to that request that is currently residing in our private | |
2785 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2786 | * this kind of deadlock. | |
2787 | */ | |
73c10101 JA |
2788 | void blk_start_plug(struct blk_plug *plug) |
2789 | { | |
2790 | struct task_struct *tsk = current; | |
2791 | ||
2792 | plug->magic = PLUG_MAGIC; | |
2793 | INIT_LIST_HEAD(&plug->list); | |
048c9374 | 2794 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 JA |
2795 | plug->should_sort = 0; |
2796 | ||
2797 | /* | |
2798 | * If this is a nested plug, don't actually assign it. It will be | |
2799 | * flushed on its own. | |
2800 | */ | |
2801 | if (!tsk->plug) { | |
2802 | /* | |
2803 | * Store ordering should not be needed here, since a potential | |
2804 | * preempt will imply a full memory barrier | |
2805 | */ | |
2806 | tsk->plug = plug; | |
2807 | } | |
2808 | } | |
2809 | EXPORT_SYMBOL(blk_start_plug); | |
2810 | ||
2811 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2812 | { | |
2813 | struct request *rqa = container_of(a, struct request, queuelist); | |
2814 | struct request *rqb = container_of(b, struct request, queuelist); | |
2815 | ||
f83e8261 | 2816 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2817 | } |
2818 | ||
49cac01e JA |
2819 | /* |
2820 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2821 | * until a safe kblockd context. We due this to avoid accidental big | |
2822 | * additional stack usage in driver dispatch, in places where the originally | |
2823 | * plugger did not intend it. | |
2824 | */ | |
f6603783 | 2825 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 2826 | bool from_schedule) |
99e22598 | 2827 | __releases(q->queue_lock) |
94b5eb28 | 2828 | { |
49cac01e | 2829 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 2830 | |
8ba61435 TH |
2831 | /* |
2832 | * Don't mess with dead queue. | |
2833 | */ | |
2834 | if (unlikely(blk_queue_dead(q))) { | |
2835 | spin_unlock(q->queue_lock); | |
2836 | return; | |
2837 | } | |
2838 | ||
99e22598 JA |
2839 | /* |
2840 | * If we are punting this to kblockd, then we can safely drop | |
2841 | * the queue_lock before waking kblockd (which needs to take | |
2842 | * this lock). | |
2843 | */ | |
2844 | if (from_schedule) { | |
2845 | spin_unlock(q->queue_lock); | |
24ecfbe2 | 2846 | blk_run_queue_async(q); |
99e22598 | 2847 | } else { |
24ecfbe2 | 2848 | __blk_run_queue(q); |
99e22598 JA |
2849 | spin_unlock(q->queue_lock); |
2850 | } | |
2851 | ||
94b5eb28 JA |
2852 | } |
2853 | ||
048c9374 N |
2854 | static void flush_plug_callbacks(struct blk_plug *plug) |
2855 | { | |
2856 | LIST_HEAD(callbacks); | |
2857 | ||
2858 | if (list_empty(&plug->cb_list)) | |
2859 | return; | |
2860 | ||
2861 | list_splice_init(&plug->cb_list, &callbacks); | |
2862 | ||
2863 | while (!list_empty(&callbacks)) { | |
2864 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
2865 | struct blk_plug_cb, | |
2866 | list); | |
2867 | list_del(&cb->list); | |
2868 | cb->callback(cb); | |
2869 | } | |
2870 | } | |
2871 | ||
49cac01e | 2872 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
2873 | { |
2874 | struct request_queue *q; | |
2875 | unsigned long flags; | |
2876 | struct request *rq; | |
109b8129 | 2877 | LIST_HEAD(list); |
94b5eb28 | 2878 | unsigned int depth; |
73c10101 JA |
2879 | |
2880 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2881 | ||
048c9374 | 2882 | flush_plug_callbacks(plug); |
73c10101 JA |
2883 | if (list_empty(&plug->list)) |
2884 | return; | |
2885 | ||
109b8129 N |
2886 | list_splice_init(&plug->list, &list); |
2887 | ||
2888 | if (plug->should_sort) { | |
2889 | list_sort(NULL, &list, plug_rq_cmp); | |
2890 | plug->should_sort = 0; | |
2891 | } | |
73c10101 JA |
2892 | |
2893 | q = NULL; | |
94b5eb28 | 2894 | depth = 0; |
18811272 JA |
2895 | |
2896 | /* | |
2897 | * Save and disable interrupts here, to avoid doing it for every | |
2898 | * queue lock we have to take. | |
2899 | */ | |
73c10101 | 2900 | local_irq_save(flags); |
109b8129 N |
2901 | while (!list_empty(&list)) { |
2902 | rq = list_entry_rq(list.next); | |
73c10101 | 2903 | list_del_init(&rq->queuelist); |
73c10101 JA |
2904 | BUG_ON(!rq->q); |
2905 | if (rq->q != q) { | |
99e22598 JA |
2906 | /* |
2907 | * This drops the queue lock | |
2908 | */ | |
2909 | if (q) | |
49cac01e | 2910 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2911 | q = rq->q; |
94b5eb28 | 2912 | depth = 0; |
73c10101 JA |
2913 | spin_lock(q->queue_lock); |
2914 | } | |
8ba61435 TH |
2915 | |
2916 | /* | |
2917 | * Short-circuit if @q is dead | |
2918 | */ | |
2919 | if (unlikely(blk_queue_dead(q))) { | |
2920 | __blk_end_request_all(rq, -ENODEV); | |
2921 | continue; | |
2922 | } | |
2923 | ||
73c10101 JA |
2924 | /* |
2925 | * rq is already accounted, so use raw insert | |
2926 | */ | |
401a18e9 JA |
2927 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
2928 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
2929 | else | |
2930 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
2931 | |
2932 | depth++; | |
73c10101 JA |
2933 | } |
2934 | ||
99e22598 JA |
2935 | /* |
2936 | * This drops the queue lock | |
2937 | */ | |
2938 | if (q) | |
49cac01e | 2939 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2940 | |
73c10101 JA |
2941 | local_irq_restore(flags); |
2942 | } | |
73c10101 JA |
2943 | |
2944 | void blk_finish_plug(struct blk_plug *plug) | |
2945 | { | |
f6603783 | 2946 | blk_flush_plug_list(plug, false); |
73c10101 | 2947 | |
88b996cd CH |
2948 | if (plug == current->plug) |
2949 | current->plug = NULL; | |
73c10101 | 2950 | } |
88b996cd | 2951 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 2952 | |
1da177e4 LT |
2953 | int __init blk_dev_init(void) |
2954 | { | |
9eb55b03 NK |
2955 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2956 | sizeof(((struct request *)0)->cmd_flags)); | |
2957 | ||
89b90be2 TH |
2958 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2959 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2960 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2961 | if (!kblockd_workqueue) |
2962 | panic("Failed to create kblockd\n"); | |
2963 | ||
2964 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2965 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2966 | |
8324aa91 | 2967 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2968 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2969 | |
d38ecf93 | 2970 | return 0; |
1da177e4 | 2971 | } |