Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 29 | #include <linux/fault-inject.h> |
73c10101 | 30 | #include <linux/list_sort.h> |
e3c78ca5 | 31 | #include <linux/delay.h> |
aaf7c680 | 32 | #include <linux/ratelimit.h> |
55782138 LZ |
33 | |
34 | #define CREATE_TRACE_POINTS | |
35 | #include <trace/events/block.h> | |
1da177e4 | 36 | |
8324aa91 | 37 | #include "blk.h" |
5efd6113 | 38 | #include "blk-cgroup.h" |
8324aa91 | 39 | |
d07335e5 | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 41 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 42 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 43 | |
a73f730d TH |
44 | DEFINE_IDA(blk_queue_ida); |
45 | ||
1da177e4 LT |
46 | /* |
47 | * For the allocated request tables | |
48 | */ | |
5ece6c52 | 49 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
50 | |
51 | /* | |
52 | * For queue allocation | |
53 | */ | |
6728cb0e | 54 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 55 | |
1da177e4 LT |
56 | /* |
57 | * Controlling structure to kblockd | |
58 | */ | |
ff856bad | 59 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 60 | |
26b8256e JA |
61 | static void drive_stat_acct(struct request *rq, int new_io) |
62 | { | |
28f13702 | 63 | struct hd_struct *part; |
26b8256e | 64 | int rw = rq_data_dir(rq); |
c9959059 | 65 | int cpu; |
26b8256e | 66 | |
c2553b58 | 67 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
68 | return; |
69 | ||
074a7aca | 70 | cpu = part_stat_lock(); |
c9959059 | 71 | |
09e099d4 JM |
72 | if (!new_io) { |
73 | part = rq->part; | |
074a7aca | 74 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
75 | } else { |
76 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 77 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
78 | /* |
79 | * The partition is already being removed, | |
80 | * the request will be accounted on the disk only | |
81 | * | |
82 | * We take a reference on disk->part0 although that | |
83 | * partition will never be deleted, so we can treat | |
84 | * it as any other partition. | |
85 | */ | |
86 | part = &rq->rq_disk->part0; | |
6c23a968 | 87 | hd_struct_get(part); |
09e099d4 | 88 | } |
074a7aca | 89 | part_round_stats(cpu, part); |
316d315b | 90 | part_inc_in_flight(part, rw); |
09e099d4 | 91 | rq->part = part; |
26b8256e | 92 | } |
e71bf0d0 | 93 | |
074a7aca | 94 | part_stat_unlock(); |
26b8256e JA |
95 | } |
96 | ||
8324aa91 | 97 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
98 | { |
99 | int nr; | |
100 | ||
101 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
102 | if (nr > q->nr_requests) | |
103 | nr = q->nr_requests; | |
104 | q->nr_congestion_on = nr; | |
105 | ||
106 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
107 | if (nr < 1) | |
108 | nr = 1; | |
109 | q->nr_congestion_off = nr; | |
110 | } | |
111 | ||
1da177e4 LT |
112 | /** |
113 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
114 | * @bdev: device | |
115 | * | |
116 | * Locates the passed device's request queue and returns the address of its | |
117 | * backing_dev_info | |
118 | * | |
119 | * Will return NULL if the request queue cannot be located. | |
120 | */ | |
121 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
122 | { | |
123 | struct backing_dev_info *ret = NULL; | |
165125e1 | 124 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
125 | |
126 | if (q) | |
127 | ret = &q->backing_dev_info; | |
128 | return ret; | |
129 | } | |
1da177e4 LT |
130 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
131 | ||
2a4aa30c | 132 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 133 | { |
1afb20f3 FT |
134 | memset(rq, 0, sizeof(*rq)); |
135 | ||
1da177e4 | 136 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 137 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 138 | rq->cpu = -1; |
63a71386 | 139 | rq->q = q; |
a2dec7b3 | 140 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
141 | INIT_HLIST_NODE(&rq->hash); |
142 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 143 | rq->cmd = rq->__cmd; |
e2494e1b | 144 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 145 | rq->tag = -1; |
1da177e4 | 146 | rq->ref_count = 1; |
b243ddcb | 147 | rq->start_time = jiffies; |
9195291e | 148 | set_start_time_ns(rq); |
09e099d4 | 149 | rq->part = NULL; |
1da177e4 | 150 | } |
2a4aa30c | 151 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 152 | |
5bb23a68 N |
153 | static void req_bio_endio(struct request *rq, struct bio *bio, |
154 | unsigned int nbytes, int error) | |
1da177e4 | 155 | { |
143a87f4 TH |
156 | if (error) |
157 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
158 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
159 | error = -EIO; | |
797e7dbb | 160 | |
143a87f4 TH |
161 | if (unlikely(nbytes > bio->bi_size)) { |
162 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
163 | __func__, nbytes, bio->bi_size); | |
164 | nbytes = bio->bi_size; | |
5bb23a68 | 165 | } |
797e7dbb | 166 | |
143a87f4 TH |
167 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
168 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 169 | |
143a87f4 TH |
170 | bio->bi_size -= nbytes; |
171 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 172 | |
143a87f4 TH |
173 | if (bio_integrity(bio)) |
174 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 175 | |
143a87f4 TH |
176 | /* don't actually finish bio if it's part of flush sequence */ |
177 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
178 | bio_endio(bio, error); | |
1da177e4 | 179 | } |
1da177e4 | 180 | |
1da177e4 LT |
181 | void blk_dump_rq_flags(struct request *rq, char *msg) |
182 | { | |
183 | int bit; | |
184 | ||
6728cb0e | 185 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
186 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
187 | rq->cmd_flags); | |
1da177e4 | 188 | |
83096ebf TH |
189 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
190 | (unsigned long long)blk_rq_pos(rq), | |
191 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 192 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 193 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 194 | |
33659ebb | 195 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 196 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 197 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
198 | printk("%02x ", rq->cmd[bit]); |
199 | printk("\n"); | |
200 | } | |
201 | } | |
1da177e4 LT |
202 | EXPORT_SYMBOL(blk_dump_rq_flags); |
203 | ||
3cca6dc1 | 204 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 205 | { |
3cca6dc1 | 206 | struct request_queue *q; |
1da177e4 | 207 | |
3cca6dc1 JA |
208 | q = container_of(work, struct request_queue, delay_work.work); |
209 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 210 | __blk_run_queue(q); |
3cca6dc1 | 211 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 212 | } |
1da177e4 LT |
213 | |
214 | /** | |
3cca6dc1 JA |
215 | * blk_delay_queue - restart queueing after defined interval |
216 | * @q: The &struct request_queue in question | |
217 | * @msecs: Delay in msecs | |
1da177e4 LT |
218 | * |
219 | * Description: | |
3cca6dc1 JA |
220 | * Sometimes queueing needs to be postponed for a little while, to allow |
221 | * resources to come back. This function will make sure that queueing is | |
222 | * restarted around the specified time. | |
223 | */ | |
224 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 225 | { |
4521cc4e JA |
226 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
227 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 228 | } |
3cca6dc1 | 229 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 230 | |
1da177e4 LT |
231 | /** |
232 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 233 | * @q: The &struct request_queue in question |
1da177e4 LT |
234 | * |
235 | * Description: | |
236 | * blk_start_queue() will clear the stop flag on the queue, and call | |
237 | * the request_fn for the queue if it was in a stopped state when | |
238 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
239 | **/ | |
165125e1 | 240 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 241 | { |
a038e253 PBG |
242 | WARN_ON(!irqs_disabled()); |
243 | ||
75ad23bc | 244 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 245 | __blk_run_queue(q); |
1da177e4 | 246 | } |
1da177e4 LT |
247 | EXPORT_SYMBOL(blk_start_queue); |
248 | ||
249 | /** | |
250 | * blk_stop_queue - stop a queue | |
165125e1 | 251 | * @q: The &struct request_queue in question |
1da177e4 LT |
252 | * |
253 | * Description: | |
254 | * The Linux block layer assumes that a block driver will consume all | |
255 | * entries on the request queue when the request_fn strategy is called. | |
256 | * Often this will not happen, because of hardware limitations (queue | |
257 | * depth settings). If a device driver gets a 'queue full' response, | |
258 | * or if it simply chooses not to queue more I/O at one point, it can | |
259 | * call this function to prevent the request_fn from being called until | |
260 | * the driver has signalled it's ready to go again. This happens by calling | |
261 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
262 | **/ | |
165125e1 | 263 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 264 | { |
ad3d9d7e | 265 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 266 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
267 | } |
268 | EXPORT_SYMBOL(blk_stop_queue); | |
269 | ||
270 | /** | |
271 | * blk_sync_queue - cancel any pending callbacks on a queue | |
272 | * @q: the queue | |
273 | * | |
274 | * Description: | |
275 | * The block layer may perform asynchronous callback activity | |
276 | * on a queue, such as calling the unplug function after a timeout. | |
277 | * A block device may call blk_sync_queue to ensure that any | |
278 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 279 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
280 | * that its ->make_request_fn will not re-add plugging prior to calling |
281 | * this function. | |
282 | * | |
da527770 VG |
283 | * This function does not cancel any asynchronous activity arising |
284 | * out of elevator or throttling code. That would require elevaotor_exit() | |
5efd6113 | 285 | * and blkcg_exit_queue() to be called with queue lock initialized. |
da527770 | 286 | * |
1da177e4 LT |
287 | */ |
288 | void blk_sync_queue(struct request_queue *q) | |
289 | { | |
70ed28b9 | 290 | del_timer_sync(&q->timeout); |
3cca6dc1 | 291 | cancel_delayed_work_sync(&q->delay_work); |
1da177e4 LT |
292 | } |
293 | EXPORT_SYMBOL(blk_sync_queue); | |
294 | ||
295 | /** | |
80a4b58e | 296 | * __blk_run_queue - run a single device queue |
1da177e4 | 297 | * @q: The queue to run |
80a4b58e JA |
298 | * |
299 | * Description: | |
300 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 301 | * held and interrupts disabled. |
1da177e4 | 302 | */ |
24ecfbe2 | 303 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 304 | { |
a538cd03 TH |
305 | if (unlikely(blk_queue_stopped(q))) |
306 | return; | |
307 | ||
c21e6beb | 308 | q->request_fn(q); |
75ad23bc NP |
309 | } |
310 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 311 | |
24ecfbe2 CH |
312 | /** |
313 | * blk_run_queue_async - run a single device queue in workqueue context | |
314 | * @q: The queue to run | |
315 | * | |
316 | * Description: | |
317 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
318 | * of us. | |
319 | */ | |
320 | void blk_run_queue_async(struct request_queue *q) | |
321 | { | |
3ec717b7 SL |
322 | if (likely(!blk_queue_stopped(q))) { |
323 | __cancel_delayed_work(&q->delay_work); | |
24ecfbe2 | 324 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
3ec717b7 | 325 | } |
24ecfbe2 | 326 | } |
c21e6beb | 327 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 328 | |
75ad23bc NP |
329 | /** |
330 | * blk_run_queue - run a single device queue | |
331 | * @q: The queue to run | |
80a4b58e JA |
332 | * |
333 | * Description: | |
334 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 335 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
336 | */ |
337 | void blk_run_queue(struct request_queue *q) | |
338 | { | |
339 | unsigned long flags; | |
340 | ||
341 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 342 | __blk_run_queue(q); |
1da177e4 LT |
343 | spin_unlock_irqrestore(q->queue_lock, flags); |
344 | } | |
345 | EXPORT_SYMBOL(blk_run_queue); | |
346 | ||
165125e1 | 347 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
348 | { |
349 | kobject_put(&q->kobj); | |
350 | } | |
d86e0e83 | 351 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 352 | |
e3c78ca5 TH |
353 | /** |
354 | * blk_drain_queue - drain requests from request_queue | |
355 | * @q: queue to drain | |
c9a929dd | 356 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 357 | * |
c9a929dd TH |
358 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
359 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
360 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 361 | */ |
c9a929dd | 362 | void blk_drain_queue(struct request_queue *q, bool drain_all) |
e3c78ca5 | 363 | { |
458f27a9 AH |
364 | int i; |
365 | ||
e3c78ca5 | 366 | while (true) { |
481a7d64 | 367 | bool drain = false; |
e3c78ca5 TH |
368 | |
369 | spin_lock_irq(q->queue_lock); | |
370 | ||
b855b04a TH |
371 | /* |
372 | * The caller might be trying to drain @q before its | |
373 | * elevator is initialized. | |
374 | */ | |
375 | if (q->elevator) | |
376 | elv_drain_elevator(q); | |
377 | ||
5efd6113 | 378 | blkcg_drain_queue(q); |
e3c78ca5 | 379 | |
4eabc941 TH |
380 | /* |
381 | * This function might be called on a queue which failed | |
b855b04a TH |
382 | * driver init after queue creation or is not yet fully |
383 | * active yet. Some drivers (e.g. fd and loop) get unhappy | |
384 | * in such cases. Kick queue iff dispatch queue has | |
385 | * something on it and @q has request_fn set. | |
4eabc941 | 386 | */ |
b855b04a | 387 | if (!list_empty(&q->queue_head) && q->request_fn) |
4eabc941 | 388 | __blk_run_queue(q); |
c9a929dd | 389 | |
8a5ecdd4 | 390 | drain |= q->nr_rqs_elvpriv; |
481a7d64 TH |
391 | |
392 | /* | |
393 | * Unfortunately, requests are queued at and tracked from | |
394 | * multiple places and there's no single counter which can | |
395 | * be drained. Check all the queues and counters. | |
396 | */ | |
397 | if (drain_all) { | |
398 | drain |= !list_empty(&q->queue_head); | |
399 | for (i = 0; i < 2; i++) { | |
8a5ecdd4 | 400 | drain |= q->nr_rqs[i]; |
481a7d64 TH |
401 | drain |= q->in_flight[i]; |
402 | drain |= !list_empty(&q->flush_queue[i]); | |
403 | } | |
404 | } | |
e3c78ca5 TH |
405 | |
406 | spin_unlock_irq(q->queue_lock); | |
407 | ||
481a7d64 | 408 | if (!drain) |
e3c78ca5 TH |
409 | break; |
410 | msleep(10); | |
411 | } | |
458f27a9 AH |
412 | |
413 | /* | |
414 | * With queue marked dead, any woken up waiter will fail the | |
415 | * allocation path, so the wakeup chaining is lost and we're | |
416 | * left with hung waiters. We need to wake up those waiters. | |
417 | */ | |
418 | if (q->request_fn) { | |
a051661c TH |
419 | struct request_list *rl; |
420 | ||
458f27a9 | 421 | spin_lock_irq(q->queue_lock); |
a051661c TH |
422 | |
423 | blk_queue_for_each_rl(rl, q) | |
424 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) | |
425 | wake_up_all(&rl->wait[i]); | |
426 | ||
458f27a9 AH |
427 | spin_unlock_irq(q->queue_lock); |
428 | } | |
e3c78ca5 TH |
429 | } |
430 | ||
d732580b TH |
431 | /** |
432 | * blk_queue_bypass_start - enter queue bypass mode | |
433 | * @q: queue of interest | |
434 | * | |
435 | * In bypass mode, only the dispatch FIFO queue of @q is used. This | |
436 | * function makes @q enter bypass mode and drains all requests which were | |
6ecf23af | 437 | * throttled or issued before. On return, it's guaranteed that no request |
80fd9979 TH |
438 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
439 | * inside queue or RCU read lock. | |
d732580b TH |
440 | */ |
441 | void blk_queue_bypass_start(struct request_queue *q) | |
442 | { | |
b82d4b19 TH |
443 | bool drain; |
444 | ||
d732580b | 445 | spin_lock_irq(q->queue_lock); |
b82d4b19 | 446 | drain = !q->bypass_depth++; |
d732580b TH |
447 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
448 | spin_unlock_irq(q->queue_lock); | |
449 | ||
b82d4b19 TH |
450 | if (drain) { |
451 | blk_drain_queue(q, false); | |
452 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ | |
453 | synchronize_rcu(); | |
454 | } | |
d732580b TH |
455 | } |
456 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); | |
457 | ||
458 | /** | |
459 | * blk_queue_bypass_end - leave queue bypass mode | |
460 | * @q: queue of interest | |
461 | * | |
462 | * Leave bypass mode and restore the normal queueing behavior. | |
463 | */ | |
464 | void blk_queue_bypass_end(struct request_queue *q) | |
465 | { | |
466 | spin_lock_irq(q->queue_lock); | |
467 | if (!--q->bypass_depth) | |
468 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); | |
469 | WARN_ON_ONCE(q->bypass_depth < 0); | |
470 | spin_unlock_irq(q->queue_lock); | |
471 | } | |
472 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); | |
473 | ||
c9a929dd TH |
474 | /** |
475 | * blk_cleanup_queue - shutdown a request queue | |
476 | * @q: request queue to shutdown | |
477 | * | |
478 | * Mark @q DEAD, drain all pending requests, destroy and put it. All | |
479 | * future requests will be failed immediately with -ENODEV. | |
c94a96ac | 480 | */ |
6728cb0e | 481 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 482 | { |
c9a929dd | 483 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 484 | |
c9a929dd | 485 | /* mark @q DEAD, no new request or merges will be allowed afterwards */ |
483f4afc | 486 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 487 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
c9a929dd | 488 | spin_lock_irq(lock); |
6ecf23af | 489 | |
80fd9979 TH |
490 | /* |
491 | * Dead queue is permanently in bypass mode till released. Note | |
492 | * that, unlike blk_queue_bypass_start(), we aren't performing | |
493 | * synchronize_rcu() after entering bypass mode to avoid the delay | |
494 | * as some drivers create and destroy a lot of queues while | |
495 | * probing. This is still safe because blk_release_queue() will be | |
496 | * called only after the queue refcnt drops to zero and nothing, | |
497 | * RCU or not, would be traversing the queue by then. | |
498 | */ | |
6ecf23af TH |
499 | q->bypass_depth++; |
500 | queue_flag_set(QUEUE_FLAG_BYPASS, q); | |
501 | ||
c9a929dd TH |
502 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
503 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
504 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
c9a929dd TH |
505 | spin_unlock_irq(lock); |
506 | mutex_unlock(&q->sysfs_lock); | |
507 | ||
b855b04a TH |
508 | /* drain all requests queued before DEAD marking */ |
509 | blk_drain_queue(q, true); | |
c9a929dd TH |
510 | |
511 | /* @q won't process any more request, flush async actions */ | |
512 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
513 | blk_sync_queue(q); | |
514 | ||
5e5cfac0 AH |
515 | spin_lock_irq(lock); |
516 | if (q->queue_lock != &q->__queue_lock) | |
517 | q->queue_lock = &q->__queue_lock; | |
518 | spin_unlock_irq(lock); | |
519 | ||
c9a929dd | 520 | /* @q is and will stay empty, shutdown and put */ |
483f4afc AV |
521 | blk_put_queue(q); |
522 | } | |
1da177e4 LT |
523 | EXPORT_SYMBOL(blk_cleanup_queue); |
524 | ||
5b788ce3 TH |
525 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
526 | gfp_t gfp_mask) | |
1da177e4 | 527 | { |
1abec4fd MS |
528 | if (unlikely(rl->rq_pool)) |
529 | return 0; | |
530 | ||
5b788ce3 | 531 | rl->q = q; |
1faa16d2 JA |
532 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
533 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
1faa16d2 JA |
534 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
535 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 536 | |
1946089a | 537 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
a91a5ac6 | 538 | mempool_free_slab, request_cachep, |
5b788ce3 | 539 | gfp_mask, q->node); |
1da177e4 LT |
540 | if (!rl->rq_pool) |
541 | return -ENOMEM; | |
542 | ||
543 | return 0; | |
544 | } | |
545 | ||
5b788ce3 TH |
546 | void blk_exit_rl(struct request_list *rl) |
547 | { | |
548 | if (rl->rq_pool) | |
549 | mempool_destroy(rl->rq_pool); | |
550 | } | |
551 | ||
165125e1 | 552 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 553 | { |
1946089a CL |
554 | return blk_alloc_queue_node(gfp_mask, -1); |
555 | } | |
556 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 557 | |
165125e1 | 558 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 559 | { |
165125e1 | 560 | struct request_queue *q; |
e0bf68dd | 561 | int err; |
1946089a | 562 | |
8324aa91 | 563 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 564 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
565 | if (!q) |
566 | return NULL; | |
567 | ||
00380a40 | 568 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
a73f730d TH |
569 | if (q->id < 0) |
570 | goto fail_q; | |
571 | ||
0989a025 JA |
572 | q->backing_dev_info.ra_pages = |
573 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
574 | q->backing_dev_info.state = 0; | |
575 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 576 | q->backing_dev_info.name = "block"; |
5151412d | 577 | q->node = node_id; |
0989a025 | 578 | |
e0bf68dd | 579 | err = bdi_init(&q->backing_dev_info); |
a73f730d TH |
580 | if (err) |
581 | goto fail_id; | |
e0bf68dd | 582 | |
31373d09 MG |
583 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
584 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb | 585 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
b855b04a | 586 | INIT_LIST_HEAD(&q->queue_head); |
242f9dcb | 587 | INIT_LIST_HEAD(&q->timeout_list); |
a612fddf | 588 | INIT_LIST_HEAD(&q->icq_list); |
4eef3049 | 589 | #ifdef CONFIG_BLK_CGROUP |
e8989fae | 590 | INIT_LIST_HEAD(&q->blkg_list); |
4eef3049 | 591 | #endif |
ae1b1539 TH |
592 | INIT_LIST_HEAD(&q->flush_queue[0]); |
593 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
594 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 595 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 596 | |
8324aa91 | 597 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 598 | |
483f4afc | 599 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 600 | spin_lock_init(&q->__queue_lock); |
483f4afc | 601 | |
c94a96ac VG |
602 | /* |
603 | * By default initialize queue_lock to internal lock and driver can | |
604 | * override it later if need be. | |
605 | */ | |
606 | q->queue_lock = &q->__queue_lock; | |
607 | ||
b82d4b19 TH |
608 | /* |
609 | * A queue starts its life with bypass turned on to avoid | |
610 | * unnecessary bypass on/off overhead and nasty surprises during | |
611 | * init. The initial bypass will be finished at the end of | |
612 | * blk_init_allocated_queue(). | |
613 | */ | |
614 | q->bypass_depth = 1; | |
615 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
616 | ||
5efd6113 | 617 | if (blkcg_init_queue(q)) |
f51b802c TH |
618 | goto fail_id; |
619 | ||
1da177e4 | 620 | return q; |
a73f730d TH |
621 | |
622 | fail_id: | |
623 | ida_simple_remove(&blk_queue_ida, q->id); | |
624 | fail_q: | |
625 | kmem_cache_free(blk_requestq_cachep, q); | |
626 | return NULL; | |
1da177e4 | 627 | } |
1946089a | 628 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
629 | |
630 | /** | |
631 | * blk_init_queue - prepare a request queue for use with a block device | |
632 | * @rfn: The function to be called to process requests that have been | |
633 | * placed on the queue. | |
634 | * @lock: Request queue spin lock | |
635 | * | |
636 | * Description: | |
637 | * If a block device wishes to use the standard request handling procedures, | |
638 | * which sorts requests and coalesces adjacent requests, then it must | |
639 | * call blk_init_queue(). The function @rfn will be called when there | |
640 | * are requests on the queue that need to be processed. If the device | |
641 | * supports plugging, then @rfn may not be called immediately when requests | |
642 | * are available on the queue, but may be called at some time later instead. | |
643 | * Plugged queues are generally unplugged when a buffer belonging to one | |
644 | * of the requests on the queue is needed, or due to memory pressure. | |
645 | * | |
646 | * @rfn is not required, or even expected, to remove all requests off the | |
647 | * queue, but only as many as it can handle at a time. If it does leave | |
648 | * requests on the queue, it is responsible for arranging that the requests | |
649 | * get dealt with eventually. | |
650 | * | |
651 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
652 | * request queue; this lock will be taken also from interrupt context, so irq |
653 | * disabling is needed for it. | |
1da177e4 | 654 | * |
710027a4 | 655 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
656 | * it didn't succeed. |
657 | * | |
658 | * Note: | |
659 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
660 | * when the block device is deactivated (such as at module unload). | |
661 | **/ | |
1946089a | 662 | |
165125e1 | 663 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 664 | { |
1946089a CL |
665 | return blk_init_queue_node(rfn, lock, -1); |
666 | } | |
667 | EXPORT_SYMBOL(blk_init_queue); | |
668 | ||
165125e1 | 669 | struct request_queue * |
1946089a CL |
670 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
671 | { | |
c86d1b8a | 672 | struct request_queue *uninit_q, *q; |
1da177e4 | 673 | |
c86d1b8a MS |
674 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
675 | if (!uninit_q) | |
676 | return NULL; | |
677 | ||
5151412d | 678 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
c86d1b8a MS |
679 | if (!q) |
680 | blk_cleanup_queue(uninit_q); | |
681 | ||
682 | return q; | |
01effb0d MS |
683 | } |
684 | EXPORT_SYMBOL(blk_init_queue_node); | |
685 | ||
686 | struct request_queue * | |
687 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
688 | spinlock_t *lock) | |
01effb0d | 689 | { |
1da177e4 LT |
690 | if (!q) |
691 | return NULL; | |
692 | ||
a051661c | 693 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
8669aafd | 694 | return NULL; |
1da177e4 LT |
695 | |
696 | q->request_fn = rfn; | |
1da177e4 | 697 | q->prep_rq_fn = NULL; |
28018c24 | 698 | q->unprep_rq_fn = NULL; |
bc58ba94 | 699 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
700 | |
701 | /* Override internal queue lock with supplied lock pointer */ | |
702 | if (lock) | |
703 | q->queue_lock = lock; | |
1da177e4 | 704 | |
f3b144aa JA |
705 | /* |
706 | * This also sets hw/phys segments, boundary and size | |
707 | */ | |
c20e8de2 | 708 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 709 | |
44ec9542 AS |
710 | q->sg_reserved_size = INT_MAX; |
711 | ||
b82d4b19 TH |
712 | /* init elevator */ |
713 | if (elevator_init(q, NULL)) | |
714 | return NULL; | |
1da177e4 | 715 | |
b82d4b19 TH |
716 | /* all done, end the initial bypass */ |
717 | blk_queue_bypass_end(q); | |
718 | return q; | |
1da177e4 | 719 | } |
5151412d | 720 | EXPORT_SYMBOL(blk_init_allocated_queue); |
1da177e4 | 721 | |
09ac46c4 | 722 | bool blk_get_queue(struct request_queue *q) |
1da177e4 | 723 | { |
34f6055c | 724 | if (likely(!blk_queue_dead(q))) { |
09ac46c4 TH |
725 | __blk_get_queue(q); |
726 | return true; | |
1da177e4 LT |
727 | } |
728 | ||
09ac46c4 | 729 | return false; |
1da177e4 | 730 | } |
d86e0e83 | 731 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 732 | |
5b788ce3 | 733 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
1da177e4 | 734 | { |
f1f8cc94 | 735 | if (rq->cmd_flags & REQ_ELVPRIV) { |
5b788ce3 | 736 | elv_put_request(rl->q, rq); |
f1f8cc94 | 737 | if (rq->elv.icq) |
11a3122f | 738 | put_io_context(rq->elv.icq->ioc); |
f1f8cc94 TH |
739 | } |
740 | ||
5b788ce3 | 741 | mempool_free(rq, rl->rq_pool); |
1da177e4 LT |
742 | } |
743 | ||
1da177e4 LT |
744 | /* |
745 | * ioc_batching returns true if the ioc is a valid batching request and | |
746 | * should be given priority access to a request. | |
747 | */ | |
165125e1 | 748 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
749 | { |
750 | if (!ioc) | |
751 | return 0; | |
752 | ||
753 | /* | |
754 | * Make sure the process is able to allocate at least 1 request | |
755 | * even if the batch times out, otherwise we could theoretically | |
756 | * lose wakeups. | |
757 | */ | |
758 | return ioc->nr_batch_requests == q->nr_batching || | |
759 | (ioc->nr_batch_requests > 0 | |
760 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
761 | } | |
762 | ||
763 | /* | |
764 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
765 | * will cause the process to be a "batcher" on all queues in the system. This | |
766 | * is the behaviour we want though - once it gets a wakeup it should be given | |
767 | * a nice run. | |
768 | */ | |
165125e1 | 769 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
770 | { |
771 | if (!ioc || ioc_batching(q, ioc)) | |
772 | return; | |
773 | ||
774 | ioc->nr_batch_requests = q->nr_batching; | |
775 | ioc->last_waited = jiffies; | |
776 | } | |
777 | ||
5b788ce3 | 778 | static void __freed_request(struct request_list *rl, int sync) |
1da177e4 | 779 | { |
5b788ce3 | 780 | struct request_queue *q = rl->q; |
1da177e4 | 781 | |
a051661c TH |
782 | /* |
783 | * bdi isn't aware of blkcg yet. As all async IOs end up root | |
784 | * blkcg anyway, just use root blkcg state. | |
785 | */ | |
786 | if (rl == &q->root_rl && | |
787 | rl->count[sync] < queue_congestion_off_threshold(q)) | |
1faa16d2 | 788 | blk_clear_queue_congested(q, sync); |
1da177e4 | 789 | |
1faa16d2 JA |
790 | if (rl->count[sync] + 1 <= q->nr_requests) { |
791 | if (waitqueue_active(&rl->wait[sync])) | |
792 | wake_up(&rl->wait[sync]); | |
1da177e4 | 793 | |
5b788ce3 | 794 | blk_clear_rl_full(rl, sync); |
1da177e4 LT |
795 | } |
796 | } | |
797 | ||
798 | /* | |
799 | * A request has just been released. Account for it, update the full and | |
800 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
801 | */ | |
5b788ce3 | 802 | static void freed_request(struct request_list *rl, unsigned int flags) |
1da177e4 | 803 | { |
5b788ce3 | 804 | struct request_queue *q = rl->q; |
75eb6c37 | 805 | int sync = rw_is_sync(flags); |
1da177e4 | 806 | |
8a5ecdd4 | 807 | q->nr_rqs[sync]--; |
1faa16d2 | 808 | rl->count[sync]--; |
75eb6c37 | 809 | if (flags & REQ_ELVPRIV) |
8a5ecdd4 | 810 | q->nr_rqs_elvpriv--; |
1da177e4 | 811 | |
5b788ce3 | 812 | __freed_request(rl, sync); |
1da177e4 | 813 | |
1faa16d2 | 814 | if (unlikely(rl->starved[sync ^ 1])) |
5b788ce3 | 815 | __freed_request(rl, sync ^ 1); |
1da177e4 LT |
816 | } |
817 | ||
9d5a4e94 MS |
818 | /* |
819 | * Determine if elevator data should be initialized when allocating the | |
820 | * request associated with @bio. | |
821 | */ | |
822 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
823 | { | |
824 | if (!bio) | |
825 | return true; | |
826 | ||
827 | /* | |
828 | * Flush requests do not use the elevator so skip initialization. | |
829 | * This allows a request to share the flush and elevator data. | |
830 | */ | |
831 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
832 | return false; | |
833 | ||
834 | return true; | |
835 | } | |
836 | ||
852c788f TH |
837 | /** |
838 | * rq_ioc - determine io_context for request allocation | |
839 | * @bio: request being allocated is for this bio (can be %NULL) | |
840 | * | |
841 | * Determine io_context to use for request allocation for @bio. May return | |
842 | * %NULL if %current->io_context doesn't exist. | |
843 | */ | |
844 | static struct io_context *rq_ioc(struct bio *bio) | |
845 | { | |
846 | #ifdef CONFIG_BLK_CGROUP | |
847 | if (bio && bio->bi_ioc) | |
848 | return bio->bi_ioc; | |
849 | #endif | |
850 | return current->io_context; | |
851 | } | |
852 | ||
da8303c6 | 853 | /** |
a06e05e6 | 854 | * __get_request - get a free request |
5b788ce3 | 855 | * @rl: request list to allocate from |
da8303c6 TH |
856 | * @rw_flags: RW and SYNC flags |
857 | * @bio: bio to allocate request for (can be %NULL) | |
858 | * @gfp_mask: allocation mask | |
859 | * | |
860 | * Get a free request from @q. This function may fail under memory | |
861 | * pressure or if @q is dead. | |
862 | * | |
863 | * Must be callled with @q->queue_lock held and, | |
864 | * Returns %NULL on failure, with @q->queue_lock held. | |
865 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 866 | */ |
5b788ce3 | 867 | static struct request *__get_request(struct request_list *rl, int rw_flags, |
a06e05e6 | 868 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 869 | { |
5b788ce3 | 870 | struct request_queue *q = rl->q; |
b679281a | 871 | struct request *rq; |
7f4b35d1 TH |
872 | struct elevator_type *et = q->elevator->type; |
873 | struct io_context *ioc = rq_ioc(bio); | |
f1f8cc94 | 874 | struct io_cq *icq = NULL; |
1faa16d2 | 875 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
75eb6c37 | 876 | int may_queue; |
88ee5ef1 | 877 | |
34f6055c | 878 | if (unlikely(blk_queue_dead(q))) |
da8303c6 TH |
879 | return NULL; |
880 | ||
7749a8d4 | 881 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
882 | if (may_queue == ELV_MQUEUE_NO) |
883 | goto rq_starved; | |
884 | ||
1faa16d2 JA |
885 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
886 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
88ee5ef1 JA |
887 | /* |
888 | * The queue will fill after this allocation, so set | |
889 | * it as full, and mark this process as "batching". | |
890 | * This process will be allowed to complete a batch of | |
891 | * requests, others will be blocked. | |
892 | */ | |
5b788ce3 | 893 | if (!blk_rl_full(rl, is_sync)) { |
88ee5ef1 | 894 | ioc_set_batching(q, ioc); |
5b788ce3 | 895 | blk_set_rl_full(rl, is_sync); |
88ee5ef1 JA |
896 | } else { |
897 | if (may_queue != ELV_MQUEUE_MUST | |
898 | && !ioc_batching(q, ioc)) { | |
899 | /* | |
900 | * The queue is full and the allocating | |
901 | * process is not a "batcher", and not | |
902 | * exempted by the IO scheduler | |
903 | */ | |
b679281a | 904 | return NULL; |
88ee5ef1 JA |
905 | } |
906 | } | |
1da177e4 | 907 | } |
a051661c TH |
908 | /* |
909 | * bdi isn't aware of blkcg yet. As all async IOs end up | |
910 | * root blkcg anyway, just use root blkcg state. | |
911 | */ | |
912 | if (rl == &q->root_rl) | |
913 | blk_set_queue_congested(q, is_sync); | |
1da177e4 LT |
914 | } |
915 | ||
082cf69e JA |
916 | /* |
917 | * Only allow batching queuers to allocate up to 50% over the defined | |
918 | * limit of requests, otherwise we could have thousands of requests | |
919 | * allocated with any setting of ->nr_requests | |
920 | */ | |
1faa16d2 | 921 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
b679281a | 922 | return NULL; |
fd782a4a | 923 | |
8a5ecdd4 | 924 | q->nr_rqs[is_sync]++; |
1faa16d2 JA |
925 | rl->count[is_sync]++; |
926 | rl->starved[is_sync] = 0; | |
cb98fc8b | 927 | |
f1f8cc94 TH |
928 | /* |
929 | * Decide whether the new request will be managed by elevator. If | |
930 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will | |
931 | * prevent the current elevator from being destroyed until the new | |
932 | * request is freed. This guarantees icq's won't be destroyed and | |
933 | * makes creating new ones safe. | |
934 | * | |
935 | * Also, lookup icq while holding queue_lock. If it doesn't exist, | |
936 | * it will be created after releasing queue_lock. | |
937 | */ | |
d732580b | 938 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
75eb6c37 | 939 | rw_flags |= REQ_ELVPRIV; |
8a5ecdd4 | 940 | q->nr_rqs_elvpriv++; |
f1f8cc94 TH |
941 | if (et->icq_cache && ioc) |
942 | icq = ioc_lookup_icq(ioc, q); | |
9d5a4e94 | 943 | } |
cb98fc8b | 944 | |
f253b86b JA |
945 | if (blk_queue_io_stat(q)) |
946 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
947 | spin_unlock_irq(q->queue_lock); |
948 | ||
29e2b09a | 949 | /* allocate and init request */ |
5b788ce3 | 950 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
29e2b09a | 951 | if (!rq) |
b679281a | 952 | goto fail_alloc; |
1da177e4 | 953 | |
29e2b09a | 954 | blk_rq_init(q, rq); |
a051661c | 955 | blk_rq_set_rl(rq, rl); |
29e2b09a TH |
956 | rq->cmd_flags = rw_flags | REQ_ALLOCED; |
957 | ||
aaf7c680 | 958 | /* init elvpriv */ |
29e2b09a | 959 | if (rw_flags & REQ_ELVPRIV) { |
aaf7c680 | 960 | if (unlikely(et->icq_cache && !icq)) { |
7f4b35d1 TH |
961 | if (ioc) |
962 | icq = ioc_create_icq(ioc, q, gfp_mask); | |
aaf7c680 TH |
963 | if (!icq) |
964 | goto fail_elvpriv; | |
29e2b09a | 965 | } |
aaf7c680 TH |
966 | |
967 | rq->elv.icq = icq; | |
968 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) | |
969 | goto fail_elvpriv; | |
970 | ||
971 | /* @rq->elv.icq holds io_context until @rq is freed */ | |
29e2b09a TH |
972 | if (icq) |
973 | get_io_context(icq->ioc); | |
974 | } | |
aaf7c680 | 975 | out: |
88ee5ef1 JA |
976 | /* |
977 | * ioc may be NULL here, and ioc_batching will be false. That's | |
978 | * OK, if the queue is under the request limit then requests need | |
979 | * not count toward the nr_batch_requests limit. There will always | |
980 | * be some limit enforced by BLK_BATCH_TIME. | |
981 | */ | |
1da177e4 LT |
982 | if (ioc_batching(q, ioc)) |
983 | ioc->nr_batch_requests--; | |
6728cb0e | 984 | |
1faa16d2 | 985 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 986 | return rq; |
b679281a | 987 | |
aaf7c680 TH |
988 | fail_elvpriv: |
989 | /* | |
990 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed | |
991 | * and may fail indefinitely under memory pressure and thus | |
992 | * shouldn't stall IO. Treat this request as !elvpriv. This will | |
993 | * disturb iosched and blkcg but weird is bettern than dead. | |
994 | */ | |
995 | printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n", | |
996 | dev_name(q->backing_dev_info.dev)); | |
997 | ||
998 | rq->cmd_flags &= ~REQ_ELVPRIV; | |
999 | rq->elv.icq = NULL; | |
1000 | ||
1001 | spin_lock_irq(q->queue_lock); | |
8a5ecdd4 | 1002 | q->nr_rqs_elvpriv--; |
aaf7c680 TH |
1003 | spin_unlock_irq(q->queue_lock); |
1004 | goto out; | |
1005 | ||
b679281a TH |
1006 | fail_alloc: |
1007 | /* | |
1008 | * Allocation failed presumably due to memory. Undo anything we | |
1009 | * might have messed up. | |
1010 | * | |
1011 | * Allocating task should really be put onto the front of the wait | |
1012 | * queue, but this is pretty rare. | |
1013 | */ | |
1014 | spin_lock_irq(q->queue_lock); | |
5b788ce3 | 1015 | freed_request(rl, rw_flags); |
b679281a TH |
1016 | |
1017 | /* | |
1018 | * in the very unlikely event that allocation failed and no | |
1019 | * requests for this direction was pending, mark us starved so that | |
1020 | * freeing of a request in the other direction will notice | |
1021 | * us. another possible fix would be to split the rq mempool into | |
1022 | * READ and WRITE | |
1023 | */ | |
1024 | rq_starved: | |
1025 | if (unlikely(rl->count[is_sync] == 0)) | |
1026 | rl->starved[is_sync] = 1; | |
1027 | return NULL; | |
1da177e4 LT |
1028 | } |
1029 | ||
da8303c6 | 1030 | /** |
a06e05e6 | 1031 | * get_request - get a free request |
da8303c6 TH |
1032 | * @q: request_queue to allocate request from |
1033 | * @rw_flags: RW and SYNC flags | |
1034 | * @bio: bio to allocate request for (can be %NULL) | |
a06e05e6 | 1035 | * @gfp_mask: allocation mask |
da8303c6 | 1036 | * |
a06e05e6 TH |
1037 | * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this |
1038 | * function keeps retrying under memory pressure and fails iff @q is dead. | |
d6344532 | 1039 | * |
da8303c6 TH |
1040 | * Must be callled with @q->queue_lock held and, |
1041 | * Returns %NULL on failure, with @q->queue_lock held. | |
1042 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 1043 | */ |
a06e05e6 TH |
1044 | static struct request *get_request(struct request_queue *q, int rw_flags, |
1045 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 | 1046 | { |
1faa16d2 | 1047 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
a06e05e6 | 1048 | DEFINE_WAIT(wait); |
a051661c | 1049 | struct request_list *rl; |
1da177e4 | 1050 | struct request *rq; |
a051661c TH |
1051 | |
1052 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ | |
a06e05e6 | 1053 | retry: |
a051661c | 1054 | rq = __get_request(rl, rw_flags, bio, gfp_mask); |
a06e05e6 TH |
1055 | if (rq) |
1056 | return rq; | |
1da177e4 | 1057 | |
a051661c TH |
1058 | if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dead(q))) { |
1059 | blk_put_rl(rl); | |
a06e05e6 | 1060 | return NULL; |
a051661c | 1061 | } |
1da177e4 | 1062 | |
a06e05e6 TH |
1063 | /* wait on @rl and retry */ |
1064 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
1065 | TASK_UNINTERRUPTIBLE); | |
1da177e4 | 1066 | |
a06e05e6 | 1067 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 1068 | |
a06e05e6 TH |
1069 | spin_unlock_irq(q->queue_lock); |
1070 | io_schedule(); | |
d6344532 | 1071 | |
a06e05e6 TH |
1072 | /* |
1073 | * After sleeping, we become a "batching" process and will be able | |
1074 | * to allocate at least one request, and up to a big batch of them | |
1075 | * for a small period time. See ioc_batching, ioc_set_batching | |
1076 | */ | |
a06e05e6 | 1077 | ioc_set_batching(q, current->io_context); |
05caf8db | 1078 | |
a06e05e6 TH |
1079 | spin_lock_irq(q->queue_lock); |
1080 | finish_wait(&rl->wait[is_sync], &wait); | |
1da177e4 | 1081 | |
a06e05e6 | 1082 | goto retry; |
1da177e4 LT |
1083 | } |
1084 | ||
165125e1 | 1085 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
1086 | { |
1087 | struct request *rq; | |
1088 | ||
1089 | BUG_ON(rw != READ && rw != WRITE); | |
1090 | ||
7f4b35d1 TH |
1091 | /* create ioc upfront */ |
1092 | create_io_context(gfp_mask, q->node); | |
1093 | ||
d6344532 | 1094 | spin_lock_irq(q->queue_lock); |
a06e05e6 | 1095 | rq = get_request(q, rw, NULL, gfp_mask); |
da8303c6 TH |
1096 | if (!rq) |
1097 | spin_unlock_irq(q->queue_lock); | |
d6344532 | 1098 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
1099 | |
1100 | return rq; | |
1101 | } | |
1da177e4 LT |
1102 | EXPORT_SYMBOL(blk_get_request); |
1103 | ||
dc72ef4a | 1104 | /** |
79eb63e9 | 1105 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 1106 | * @q: target request queue |
79eb63e9 BH |
1107 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
1108 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 1109 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 1110 | * |
79eb63e9 BH |
1111 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
1112 | * type commands. Where the struct request needs to be farther initialized by | |
1113 | * the caller. It is passed a &struct bio, which describes the memory info of | |
1114 | * the I/O transfer. | |
dc72ef4a | 1115 | * |
79eb63e9 BH |
1116 | * The caller of blk_make_request must make sure that bi_io_vec |
1117 | * are set to describe the memory buffers. That bio_data_dir() will return | |
1118 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
1119 | * are properly set accordingly) | |
1120 | * | |
1121 | * If called under none-sleepable conditions, mapped bio buffers must not | |
1122 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
1123 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
1124 | * BUG. | |
53674ac5 JA |
1125 | * |
1126 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
1127 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
1128 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
1129 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
1130 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
1131 | * of bio_alloc(), as that avoids the mempool deadlock. | |
1132 | * If possible a big IO should be split into smaller parts when allocation | |
1133 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 1134 | */ |
79eb63e9 BH |
1135 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
1136 | gfp_t gfp_mask) | |
dc72ef4a | 1137 | { |
79eb63e9 BH |
1138 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1139 | ||
1140 | if (unlikely(!rq)) | |
1141 | return ERR_PTR(-ENOMEM); | |
1142 | ||
1143 | for_each_bio(bio) { | |
1144 | struct bio *bounce_bio = bio; | |
1145 | int ret; | |
1146 | ||
1147 | blk_queue_bounce(q, &bounce_bio); | |
1148 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1149 | if (unlikely(ret)) { | |
1150 | blk_put_request(rq); | |
1151 | return ERR_PTR(ret); | |
1152 | } | |
1153 | } | |
1154 | ||
1155 | return rq; | |
dc72ef4a | 1156 | } |
79eb63e9 | 1157 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1158 | |
1da177e4 LT |
1159 | /** |
1160 | * blk_requeue_request - put a request back on queue | |
1161 | * @q: request queue where request should be inserted | |
1162 | * @rq: request to be inserted | |
1163 | * | |
1164 | * Description: | |
1165 | * Drivers often keep queueing requests until the hardware cannot accept | |
1166 | * more, when that condition happens we need to put the request back | |
1167 | * on the queue. Must be called with queue lock held. | |
1168 | */ | |
165125e1 | 1169 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1170 | { |
242f9dcb JA |
1171 | blk_delete_timer(rq); |
1172 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1173 | trace_block_rq_requeue(q, rq); |
2056a782 | 1174 | |
1da177e4 LT |
1175 | if (blk_rq_tagged(rq)) |
1176 | blk_queue_end_tag(q, rq); | |
1177 | ||
ba396a6c JB |
1178 | BUG_ON(blk_queued_rq(rq)); |
1179 | ||
1da177e4 LT |
1180 | elv_requeue_request(q, rq); |
1181 | } | |
1da177e4 LT |
1182 | EXPORT_SYMBOL(blk_requeue_request); |
1183 | ||
73c10101 JA |
1184 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1185 | int where) | |
1186 | { | |
1187 | drive_stat_acct(rq, 1); | |
7eaceacc | 1188 | __elv_add_request(q, rq, where); |
73c10101 JA |
1189 | } |
1190 | ||
074a7aca TH |
1191 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1192 | unsigned long now) | |
1193 | { | |
1194 | if (now == part->stamp) | |
1195 | return; | |
1196 | ||
316d315b | 1197 | if (part_in_flight(part)) { |
074a7aca | 1198 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1199 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1200 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1201 | } | |
1202 | part->stamp = now; | |
1203 | } | |
1204 | ||
1205 | /** | |
496aa8a9 RD |
1206 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1207 | * @cpu: cpu number for stats access | |
1208 | * @part: target partition | |
1da177e4 LT |
1209 | * |
1210 | * The average IO queue length and utilisation statistics are maintained | |
1211 | * by observing the current state of the queue length and the amount of | |
1212 | * time it has been in this state for. | |
1213 | * | |
1214 | * Normally, that accounting is done on IO completion, but that can result | |
1215 | * in more than a second's worth of IO being accounted for within any one | |
1216 | * second, leading to >100% utilisation. To deal with that, we call this | |
1217 | * function to do a round-off before returning the results when reading | |
1218 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1219 | * the current jiffies and restarts the counters again. | |
1220 | */ | |
c9959059 | 1221 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1222 | { |
1223 | unsigned long now = jiffies; | |
1224 | ||
074a7aca TH |
1225 | if (part->partno) |
1226 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1227 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1228 | } |
074a7aca | 1229 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1230 | |
1da177e4 LT |
1231 | /* |
1232 | * queue lock must be held | |
1233 | */ | |
165125e1 | 1234 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1235 | { |
1da177e4 LT |
1236 | if (unlikely(!q)) |
1237 | return; | |
1238 | if (unlikely(--req->ref_count)) | |
1239 | return; | |
1240 | ||
8922e16c TH |
1241 | elv_completed_request(q, req); |
1242 | ||
1cd96c24 BH |
1243 | /* this is a bio leak */ |
1244 | WARN_ON(req->bio != NULL); | |
1245 | ||
1da177e4 LT |
1246 | /* |
1247 | * Request may not have originated from ll_rw_blk. if not, | |
1248 | * it didn't come out of our reserved rq pools | |
1249 | */ | |
49171e5c | 1250 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1251 | unsigned int flags = req->cmd_flags; |
a051661c | 1252 | struct request_list *rl = blk_rq_rl(req); |
1da177e4 | 1253 | |
1da177e4 | 1254 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1255 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 | 1256 | |
a051661c TH |
1257 | blk_free_request(rl, req); |
1258 | freed_request(rl, flags); | |
1259 | blk_put_rl(rl); | |
1da177e4 LT |
1260 | } |
1261 | } | |
6e39b69e MC |
1262 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1263 | ||
1da177e4 LT |
1264 | void blk_put_request(struct request *req) |
1265 | { | |
8922e16c | 1266 | unsigned long flags; |
165125e1 | 1267 | struct request_queue *q = req->q; |
8922e16c | 1268 | |
52a93ba8 FT |
1269 | spin_lock_irqsave(q->queue_lock, flags); |
1270 | __blk_put_request(q, req); | |
1271 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1272 | } |
1da177e4 LT |
1273 | EXPORT_SYMBOL(blk_put_request); |
1274 | ||
66ac0280 CH |
1275 | /** |
1276 | * blk_add_request_payload - add a payload to a request | |
1277 | * @rq: request to update | |
1278 | * @page: page backing the payload | |
1279 | * @len: length of the payload. | |
1280 | * | |
1281 | * This allows to later add a payload to an already submitted request by | |
1282 | * a block driver. The driver needs to take care of freeing the payload | |
1283 | * itself. | |
1284 | * | |
1285 | * Note that this is a quite horrible hack and nothing but handling of | |
1286 | * discard requests should ever use it. | |
1287 | */ | |
1288 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1289 | unsigned int len) | |
1290 | { | |
1291 | struct bio *bio = rq->bio; | |
1292 | ||
1293 | bio->bi_io_vec->bv_page = page; | |
1294 | bio->bi_io_vec->bv_offset = 0; | |
1295 | bio->bi_io_vec->bv_len = len; | |
1296 | ||
1297 | bio->bi_size = len; | |
1298 | bio->bi_vcnt = 1; | |
1299 | bio->bi_phys_segments = 1; | |
1300 | ||
1301 | rq->__data_len = rq->resid_len = len; | |
1302 | rq->nr_phys_segments = 1; | |
1303 | rq->buffer = bio_data(bio); | |
1304 | } | |
1305 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1306 | ||
73c10101 JA |
1307 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1308 | struct bio *bio) | |
1309 | { | |
1310 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1311 | ||
73c10101 JA |
1312 | if (!ll_back_merge_fn(q, req, bio)) |
1313 | return false; | |
1314 | ||
1315 | trace_block_bio_backmerge(q, bio); | |
1316 | ||
1317 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1318 | blk_rq_set_mixed_merge(req); | |
1319 | ||
1320 | req->biotail->bi_next = bio; | |
1321 | req->biotail = bio; | |
1322 | req->__data_len += bio->bi_size; | |
1323 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1324 | ||
1325 | drive_stat_acct(req, 0); | |
1326 | return true; | |
1327 | } | |
1328 | ||
1329 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1330 | struct request *req, struct bio *bio) | |
1331 | { | |
1332 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1333 | |
73c10101 JA |
1334 | if (!ll_front_merge_fn(q, req, bio)) |
1335 | return false; | |
1336 | ||
1337 | trace_block_bio_frontmerge(q, bio); | |
1338 | ||
1339 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1340 | blk_rq_set_mixed_merge(req); | |
1341 | ||
73c10101 JA |
1342 | bio->bi_next = req->bio; |
1343 | req->bio = bio; | |
1344 | ||
1345 | /* | |
1346 | * may not be valid. if the low level driver said | |
1347 | * it didn't need a bounce buffer then it better | |
1348 | * not touch req->buffer either... | |
1349 | */ | |
1350 | req->buffer = bio_data(bio); | |
1351 | req->__sector = bio->bi_sector; | |
1352 | req->__data_len += bio->bi_size; | |
1353 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1354 | ||
1355 | drive_stat_acct(req, 0); | |
1356 | return true; | |
1357 | } | |
1358 | ||
bd87b589 TH |
1359 | /** |
1360 | * attempt_plug_merge - try to merge with %current's plugged list | |
1361 | * @q: request_queue new bio is being queued at | |
1362 | * @bio: new bio being queued | |
1363 | * @request_count: out parameter for number of traversed plugged requests | |
1364 | * | |
1365 | * Determine whether @bio being queued on @q can be merged with a request | |
1366 | * on %current's plugged list. Returns %true if merge was successful, | |
1367 | * otherwise %false. | |
1368 | * | |
07c2bd37 TH |
1369 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1370 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1371 | * than scheduling, and the request, while may have elvpriv data, is not | |
1372 | * added on the elevator at this point. In addition, we don't have | |
1373 | * reliable access to the elevator outside queue lock. Only check basic | |
1374 | * merging parameters without querying the elevator. | |
73c10101 | 1375 | */ |
bd87b589 TH |
1376 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1377 | unsigned int *request_count) | |
73c10101 JA |
1378 | { |
1379 | struct blk_plug *plug; | |
1380 | struct request *rq; | |
1381 | bool ret = false; | |
1382 | ||
bd87b589 | 1383 | plug = current->plug; |
73c10101 JA |
1384 | if (!plug) |
1385 | goto out; | |
56ebdaf2 | 1386 | *request_count = 0; |
73c10101 JA |
1387 | |
1388 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1389 | int el_ret; | |
1390 | ||
1b2e19f1 SL |
1391 | if (rq->q == q) |
1392 | (*request_count)++; | |
56ebdaf2 | 1393 | |
07c2bd37 | 1394 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
73c10101 JA |
1395 | continue; |
1396 | ||
050c8ea8 | 1397 | el_ret = blk_try_merge(rq, bio); |
73c10101 JA |
1398 | if (el_ret == ELEVATOR_BACK_MERGE) { |
1399 | ret = bio_attempt_back_merge(q, rq, bio); | |
1400 | if (ret) | |
1401 | break; | |
1402 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1403 | ret = bio_attempt_front_merge(q, rq, bio); | |
1404 | if (ret) | |
1405 | break; | |
1406 | } | |
1407 | } | |
1408 | out: | |
1409 | return ret; | |
1410 | } | |
1411 | ||
86db1e29 | 1412 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1413 | { |
4aff5e23 | 1414 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1415 | |
7b6d91da CH |
1416 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1417 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1418 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1419 | |
52d9e675 | 1420 | req->errors = 0; |
a2dec7b3 | 1421 | req->__sector = bio->bi_sector; |
52d9e675 | 1422 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1423 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1424 | } |
1425 | ||
5a7bbad2 | 1426 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1427 | { |
5e00d1b5 | 1428 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1429 | struct blk_plug *plug; |
1430 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1431 | struct request *req; | |
56ebdaf2 | 1432 | unsigned int request_count = 0; |
1da177e4 | 1433 | |
1da177e4 LT |
1434 | /* |
1435 | * low level driver can indicate that it wants pages above a | |
1436 | * certain limit bounced to low memory (ie for highmem, or even | |
1437 | * ISA dma in theory) | |
1438 | */ | |
1439 | blk_queue_bounce(q, &bio); | |
1440 | ||
4fed947c | 1441 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1442 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1443 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1444 | goto get_rq; |
1445 | } | |
1446 | ||
73c10101 JA |
1447 | /* |
1448 | * Check if we can merge with the plugged list before grabbing | |
1449 | * any locks. | |
1450 | */ | |
bd87b589 | 1451 | if (attempt_plug_merge(q, bio, &request_count)) |
5a7bbad2 | 1452 | return; |
1da177e4 | 1453 | |
73c10101 | 1454 | spin_lock_irq(q->queue_lock); |
2056a782 | 1455 | |
73c10101 JA |
1456 | el_ret = elv_merge(q, &req, bio); |
1457 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 | 1458 | if (bio_attempt_back_merge(q, req, bio)) { |
07c2bd37 | 1459 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1460 | if (!attempt_back_merge(q, req)) |
1461 | elv_merged_request(q, req, el_ret); | |
1462 | goto out_unlock; | |
1463 | } | |
1464 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 | 1465 | if (bio_attempt_front_merge(q, req, bio)) { |
07c2bd37 | 1466 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1467 | if (!attempt_front_merge(q, req)) |
1468 | elv_merged_request(q, req, el_ret); | |
1469 | goto out_unlock; | |
80a761fd | 1470 | } |
1da177e4 LT |
1471 | } |
1472 | ||
450991bc | 1473 | get_rq: |
7749a8d4 JA |
1474 | /* |
1475 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1476 | * but we need to set it earlier to expose the sync flag to the | |
1477 | * rq allocator and io schedulers. | |
1478 | */ | |
1479 | rw_flags = bio_data_dir(bio); | |
1480 | if (sync) | |
7b6d91da | 1481 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1482 | |
1da177e4 | 1483 | /* |
450991bc | 1484 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1485 | * Returns with the queue unlocked. |
450991bc | 1486 | */ |
a06e05e6 | 1487 | req = get_request(q, rw_flags, bio, GFP_NOIO); |
da8303c6 TH |
1488 | if (unlikely(!req)) { |
1489 | bio_endio(bio, -ENODEV); /* @q is dead */ | |
1490 | goto out_unlock; | |
1491 | } | |
d6344532 | 1492 | |
450991bc NP |
1493 | /* |
1494 | * After dropping the lock and possibly sleeping here, our request | |
1495 | * may now be mergeable after it had proven unmergeable (above). | |
1496 | * We don't worry about that case for efficiency. It won't happen | |
1497 | * often, and the elevators are able to handle it. | |
1da177e4 | 1498 | */ |
52d9e675 | 1499 | init_request_from_bio(req, bio); |
1da177e4 | 1500 | |
9562ad9a | 1501 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1502 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1503 | |
1504 | plug = current->plug; | |
721a9602 | 1505 | if (plug) { |
dc6d36c9 JA |
1506 | /* |
1507 | * If this is the first request added after a plug, fire | |
1508 | * of a plug trace. If others have been added before, check | |
1509 | * if we have multiple devices in this plug. If so, make a | |
1510 | * note to sort the list before dispatch. | |
1511 | */ | |
1512 | if (list_empty(&plug->list)) | |
1513 | trace_block_plug(q); | |
3540d5e8 SL |
1514 | else { |
1515 | if (!plug->should_sort) { | |
1516 | struct request *__rq; | |
73c10101 | 1517 | |
3540d5e8 SL |
1518 | __rq = list_entry_rq(plug->list.prev); |
1519 | if (__rq->q != q) | |
1520 | plug->should_sort = 1; | |
1521 | } | |
019ceb7d | 1522 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
3540d5e8 | 1523 | blk_flush_plug_list(plug, false); |
019ceb7d SL |
1524 | trace_block_plug(q); |
1525 | } | |
73c10101 | 1526 | } |
73c10101 JA |
1527 | list_add_tail(&req->queuelist, &plug->list); |
1528 | drive_stat_acct(req, 1); | |
1529 | } else { | |
1530 | spin_lock_irq(q->queue_lock); | |
1531 | add_acct_request(q, req, where); | |
24ecfbe2 | 1532 | __blk_run_queue(q); |
73c10101 JA |
1533 | out_unlock: |
1534 | spin_unlock_irq(q->queue_lock); | |
1535 | } | |
1da177e4 | 1536 | } |
c20e8de2 | 1537 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1538 | |
1539 | /* | |
1540 | * If bio->bi_dev is a partition, remap the location | |
1541 | */ | |
1542 | static inline void blk_partition_remap(struct bio *bio) | |
1543 | { | |
1544 | struct block_device *bdev = bio->bi_bdev; | |
1545 | ||
bf2de6f5 | 1546 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1547 | struct hd_struct *p = bdev->bd_part; |
1548 | ||
1da177e4 LT |
1549 | bio->bi_sector += p->start_sect; |
1550 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1551 | |
d07335e5 MS |
1552 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1553 | bdev->bd_dev, | |
1554 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1555 | } |
1556 | } | |
1557 | ||
1da177e4 LT |
1558 | static void handle_bad_sector(struct bio *bio) |
1559 | { | |
1560 | char b[BDEVNAME_SIZE]; | |
1561 | ||
1562 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1563 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1564 | bdevname(bio->bi_bdev, b), | |
1565 | bio->bi_rw, | |
1566 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1567 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1568 | |
1569 | set_bit(BIO_EOF, &bio->bi_flags); | |
1570 | } | |
1571 | ||
c17bb495 AM |
1572 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1573 | ||
1574 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1575 | ||
1576 | static int __init setup_fail_make_request(char *str) | |
1577 | { | |
1578 | return setup_fault_attr(&fail_make_request, str); | |
1579 | } | |
1580 | __setup("fail_make_request=", setup_fail_make_request); | |
1581 | ||
b2c9cd37 | 1582 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1583 | { |
b2c9cd37 | 1584 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1585 | } |
1586 | ||
1587 | static int __init fail_make_request_debugfs(void) | |
1588 | { | |
dd48c085 AM |
1589 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1590 | NULL, &fail_make_request); | |
1591 | ||
1592 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; | |
c17bb495 AM |
1593 | } |
1594 | ||
1595 | late_initcall(fail_make_request_debugfs); | |
1596 | ||
1597 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1598 | ||
b2c9cd37 AM |
1599 | static inline bool should_fail_request(struct hd_struct *part, |
1600 | unsigned int bytes) | |
c17bb495 | 1601 | { |
b2c9cd37 | 1602 | return false; |
c17bb495 AM |
1603 | } |
1604 | ||
1605 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1606 | ||
c07e2b41 JA |
1607 | /* |
1608 | * Check whether this bio extends beyond the end of the device. | |
1609 | */ | |
1610 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1611 | { | |
1612 | sector_t maxsector; | |
1613 | ||
1614 | if (!nr_sectors) | |
1615 | return 0; | |
1616 | ||
1617 | /* Test device or partition size, when known. */ | |
77304d2a | 1618 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1619 | if (maxsector) { |
1620 | sector_t sector = bio->bi_sector; | |
1621 | ||
1622 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1623 | /* | |
1624 | * This may well happen - the kernel calls bread() | |
1625 | * without checking the size of the device, e.g., when | |
1626 | * mounting a device. | |
1627 | */ | |
1628 | handle_bad_sector(bio); | |
1629 | return 1; | |
1630 | } | |
1631 | } | |
1632 | ||
1633 | return 0; | |
1634 | } | |
1635 | ||
27a84d54 CH |
1636 | static noinline_for_stack bool |
1637 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1638 | { |
165125e1 | 1639 | struct request_queue *q; |
5a7bbad2 | 1640 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1641 | int err = -EIO; |
5a7bbad2 CH |
1642 | char b[BDEVNAME_SIZE]; |
1643 | struct hd_struct *part; | |
1da177e4 LT |
1644 | |
1645 | might_sleep(); | |
1da177e4 | 1646 | |
c07e2b41 JA |
1647 | if (bio_check_eod(bio, nr_sectors)) |
1648 | goto end_io; | |
1da177e4 | 1649 | |
5a7bbad2 CH |
1650 | q = bdev_get_queue(bio->bi_bdev); |
1651 | if (unlikely(!q)) { | |
1652 | printk(KERN_ERR | |
1653 | "generic_make_request: Trying to access " | |
1654 | "nonexistent block-device %s (%Lu)\n", | |
1655 | bdevname(bio->bi_bdev, b), | |
1656 | (long long) bio->bi_sector); | |
1657 | goto end_io; | |
1658 | } | |
c17bb495 | 1659 | |
e2a60da7 MP |
1660 | if (likely(bio_is_rw(bio) && |
1661 | nr_sectors > queue_max_hw_sectors(q))) { | |
5a7bbad2 CH |
1662 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1663 | bdevname(bio->bi_bdev, b), | |
1664 | bio_sectors(bio), | |
1665 | queue_max_hw_sectors(q)); | |
1666 | goto end_io; | |
1667 | } | |
1da177e4 | 1668 | |
5a7bbad2 CH |
1669 | part = bio->bi_bdev->bd_part; |
1670 | if (should_fail_request(part, bio->bi_size) || | |
1671 | should_fail_request(&part_to_disk(part)->part0, | |
1672 | bio->bi_size)) | |
1673 | goto end_io; | |
2056a782 | 1674 | |
5a7bbad2 CH |
1675 | /* |
1676 | * If this device has partitions, remap block n | |
1677 | * of partition p to block n+start(p) of the disk. | |
1678 | */ | |
1679 | blk_partition_remap(bio); | |
2056a782 | 1680 | |
5a7bbad2 CH |
1681 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1682 | goto end_io; | |
a7384677 | 1683 | |
5a7bbad2 CH |
1684 | if (bio_check_eod(bio, nr_sectors)) |
1685 | goto end_io; | |
1e87901e | 1686 | |
5a7bbad2 CH |
1687 | /* |
1688 | * Filter flush bio's early so that make_request based | |
1689 | * drivers without flush support don't have to worry | |
1690 | * about them. | |
1691 | */ | |
1692 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1693 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1694 | if (!nr_sectors) { | |
1695 | err = 0; | |
51fd77bd JA |
1696 | goto end_io; |
1697 | } | |
5a7bbad2 | 1698 | } |
5ddfe969 | 1699 | |
5a7bbad2 CH |
1700 | if ((bio->bi_rw & REQ_DISCARD) && |
1701 | (!blk_queue_discard(q) || | |
e2a60da7 | 1702 | ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) { |
5a7bbad2 CH |
1703 | err = -EOPNOTSUPP; |
1704 | goto end_io; | |
1705 | } | |
01edede4 | 1706 | |
4363ac7c MP |
1707 | if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) { |
1708 | err = -EOPNOTSUPP; | |
1709 | goto end_io; | |
1710 | } | |
1711 | ||
7f4b35d1 TH |
1712 | /* |
1713 | * Various block parts want %current->io_context and lazy ioc | |
1714 | * allocation ends up trading a lot of pain for a small amount of | |
1715 | * memory. Just allocate it upfront. This may fail and block | |
1716 | * layer knows how to live with it. | |
1717 | */ | |
1718 | create_io_context(GFP_ATOMIC, q->node); | |
1719 | ||
bc16a4f9 TH |
1720 | if (blk_throtl_bio(q, bio)) |
1721 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1722 | |
5a7bbad2 | 1723 | trace_block_bio_queue(q, bio); |
27a84d54 | 1724 | return true; |
a7384677 TH |
1725 | |
1726 | end_io: | |
1727 | bio_endio(bio, err); | |
27a84d54 | 1728 | return false; |
1da177e4 LT |
1729 | } |
1730 | ||
27a84d54 CH |
1731 | /** |
1732 | * generic_make_request - hand a buffer to its device driver for I/O | |
1733 | * @bio: The bio describing the location in memory and on the device. | |
1734 | * | |
1735 | * generic_make_request() is used to make I/O requests of block | |
1736 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1737 | * to be done. | |
1738 | * | |
1739 | * generic_make_request() does not return any status. The | |
1740 | * success/failure status of the request, along with notification of | |
1741 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1742 | * function described (one day) else where. | |
1743 | * | |
1744 | * The caller of generic_make_request must make sure that bi_io_vec | |
1745 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1746 | * set to describe the device address, and the | |
1747 | * bi_end_io and optionally bi_private are set to describe how | |
1748 | * completion notification should be signaled. | |
1749 | * | |
1750 | * generic_make_request and the drivers it calls may use bi_next if this | |
1751 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1752 | * a lower device by calling into generic_make_request recursively, which | |
1753 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1754 | */ |
1755 | void generic_make_request(struct bio *bio) | |
1756 | { | |
bddd87c7 AM |
1757 | struct bio_list bio_list_on_stack; |
1758 | ||
27a84d54 CH |
1759 | if (!generic_make_request_checks(bio)) |
1760 | return; | |
1761 | ||
1762 | /* | |
1763 | * We only want one ->make_request_fn to be active at a time, else | |
1764 | * stack usage with stacked devices could be a problem. So use | |
1765 | * current->bio_list to keep a list of requests submited by a | |
1766 | * make_request_fn function. current->bio_list is also used as a | |
1767 | * flag to say if generic_make_request is currently active in this | |
1768 | * task or not. If it is NULL, then no make_request is active. If | |
1769 | * it is non-NULL, then a make_request is active, and new requests | |
1770 | * should be added at the tail | |
1771 | */ | |
bddd87c7 | 1772 | if (current->bio_list) { |
bddd87c7 | 1773 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1774 | return; |
1775 | } | |
27a84d54 | 1776 | |
d89d8796 NB |
1777 | /* following loop may be a bit non-obvious, and so deserves some |
1778 | * explanation. | |
1779 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1780 | * ensure that) so we have a list with a single bio. | |
1781 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1782 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1783 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1784 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1785 | * through a recursive call to generic_make_request. If it |
1786 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1787 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1788 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1789 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1790 | */ |
1791 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1792 | bio_list_init(&bio_list_on_stack); |
1793 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1794 | do { |
27a84d54 CH |
1795 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1796 | ||
1797 | q->make_request_fn(q, bio); | |
1798 | ||
bddd87c7 | 1799 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1800 | } while (bio); |
bddd87c7 | 1801 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1802 | } |
1da177e4 LT |
1803 | EXPORT_SYMBOL(generic_make_request); |
1804 | ||
1805 | /** | |
710027a4 | 1806 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1807 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1808 | * @bio: The &struct bio which describes the I/O | |
1809 | * | |
1810 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1811 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1812 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1813 | * |
1814 | */ | |
1815 | void submit_bio(int rw, struct bio *bio) | |
1816 | { | |
22e2c507 | 1817 | bio->bi_rw |= rw; |
1da177e4 | 1818 | |
bf2de6f5 JA |
1819 | /* |
1820 | * If it's a regular read/write or a barrier with data attached, | |
1821 | * go through the normal accounting stuff before submission. | |
1822 | */ | |
e2a60da7 | 1823 | if (bio_has_data(bio)) { |
4363ac7c MP |
1824 | unsigned int count; |
1825 | ||
1826 | if (unlikely(rw & REQ_WRITE_SAME)) | |
1827 | count = bdev_logical_block_size(bio->bi_bdev) >> 9; | |
1828 | else | |
1829 | count = bio_sectors(bio); | |
1830 | ||
bf2de6f5 JA |
1831 | if (rw & WRITE) { |
1832 | count_vm_events(PGPGOUT, count); | |
1833 | } else { | |
1834 | task_io_account_read(bio->bi_size); | |
1835 | count_vm_events(PGPGIN, count); | |
1836 | } | |
1837 | ||
1838 | if (unlikely(block_dump)) { | |
1839 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1840 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1841 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1842 | (rw & WRITE) ? "WRITE" : "READ", |
1843 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1844 | bdevname(bio->bi_bdev, b), |
1845 | count); | |
bf2de6f5 | 1846 | } |
1da177e4 LT |
1847 | } |
1848 | ||
1849 | generic_make_request(bio); | |
1850 | } | |
1da177e4 LT |
1851 | EXPORT_SYMBOL(submit_bio); |
1852 | ||
82124d60 KU |
1853 | /** |
1854 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1855 | * @q: the queue | |
1856 | * @rq: the request being checked | |
1857 | * | |
1858 | * Description: | |
1859 | * @rq may have been made based on weaker limitations of upper-level queues | |
1860 | * in request stacking drivers, and it may violate the limitation of @q. | |
1861 | * Since the block layer and the underlying device driver trust @rq | |
1862 | * after it is inserted to @q, it should be checked against @q before | |
1863 | * the insertion using this generic function. | |
1864 | * | |
1865 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1866 | * in some cases below, so export this function. |
82124d60 KU |
1867 | * Request stacking drivers like request-based dm may change the queue |
1868 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1869 | * Such request stacking drivers should check those requests agaist | |
1870 | * the new queue limits again when they dispatch those requests, | |
1871 | * although such checkings are also done against the old queue limits | |
1872 | * when submitting requests. | |
1873 | */ | |
1874 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1875 | { | |
e2a60da7 | 1876 | if (!rq_mergeable(rq)) |
3383977f S |
1877 | return 0; |
1878 | ||
f31dc1cd | 1879 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) { |
82124d60 KU |
1880 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1881 | return -EIO; | |
1882 | } | |
1883 | ||
1884 | /* | |
1885 | * queue's settings related to segment counting like q->bounce_pfn | |
1886 | * may differ from that of other stacking queues. | |
1887 | * Recalculate it to check the request correctly on this queue's | |
1888 | * limitation. | |
1889 | */ | |
1890 | blk_recalc_rq_segments(rq); | |
8a78362c | 1891 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1892 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1893 | return -EIO; | |
1894 | } | |
1895 | ||
1896 | return 0; | |
1897 | } | |
1898 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1899 | ||
1900 | /** | |
1901 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1902 | * @q: the queue to submit the request | |
1903 | * @rq: the request being queued | |
1904 | */ | |
1905 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1906 | { | |
1907 | unsigned long flags; | |
4853abaa | 1908 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
1909 | |
1910 | if (blk_rq_check_limits(q, rq)) | |
1911 | return -EIO; | |
1912 | ||
b2c9cd37 AM |
1913 | if (rq->rq_disk && |
1914 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 1915 | return -EIO; |
82124d60 KU |
1916 | |
1917 | spin_lock_irqsave(q->queue_lock, flags); | |
8ba61435 TH |
1918 | if (unlikely(blk_queue_dead(q))) { |
1919 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1920 | return -ENODEV; | |
1921 | } | |
82124d60 KU |
1922 | |
1923 | /* | |
1924 | * Submitting request must be dequeued before calling this function | |
1925 | * because it will be linked to another request_queue | |
1926 | */ | |
1927 | BUG_ON(blk_queued_rq(rq)); | |
1928 | ||
4853abaa JM |
1929 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
1930 | where = ELEVATOR_INSERT_FLUSH; | |
1931 | ||
1932 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
1933 | if (where == ELEVATOR_INSERT_FLUSH) |
1934 | __blk_run_queue(q); | |
82124d60 KU |
1935 | spin_unlock_irqrestore(q->queue_lock, flags); |
1936 | ||
1937 | return 0; | |
1938 | } | |
1939 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1940 | ||
80a761fd TH |
1941 | /** |
1942 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1943 | * @rq: request to examine | |
1944 | * | |
1945 | * Description: | |
1946 | * A request could be merge of IOs which require different failure | |
1947 | * handling. This function determines the number of bytes which | |
1948 | * can be failed from the beginning of the request without | |
1949 | * crossing into area which need to be retried further. | |
1950 | * | |
1951 | * Return: | |
1952 | * The number of bytes to fail. | |
1953 | * | |
1954 | * Context: | |
1955 | * queue_lock must be held. | |
1956 | */ | |
1957 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1958 | { | |
1959 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1960 | unsigned int bytes = 0; | |
1961 | struct bio *bio; | |
1962 | ||
1963 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1964 | return blk_rq_bytes(rq); | |
1965 | ||
1966 | /* | |
1967 | * Currently the only 'mixing' which can happen is between | |
1968 | * different fastfail types. We can safely fail portions | |
1969 | * which have all the failfast bits that the first one has - | |
1970 | * the ones which are at least as eager to fail as the first | |
1971 | * one. | |
1972 | */ | |
1973 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1974 | if ((bio->bi_rw & ff) != ff) | |
1975 | break; | |
1976 | bytes += bio->bi_size; | |
1977 | } | |
1978 | ||
1979 | /* this could lead to infinite loop */ | |
1980 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1981 | return bytes; | |
1982 | } | |
1983 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1984 | ||
bc58ba94 JA |
1985 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1986 | { | |
c2553b58 | 1987 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1988 | const int rw = rq_data_dir(req); |
1989 | struct hd_struct *part; | |
1990 | int cpu; | |
1991 | ||
1992 | cpu = part_stat_lock(); | |
09e099d4 | 1993 | part = req->part; |
bc58ba94 JA |
1994 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1995 | part_stat_unlock(); | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | static void blk_account_io_done(struct request *req) | |
2000 | { | |
bc58ba94 | 2001 | /* |
dd4c133f TH |
2002 | * Account IO completion. flush_rq isn't accounted as a |
2003 | * normal IO on queueing nor completion. Accounting the | |
2004 | * containing request is enough. | |
bc58ba94 | 2005 | */ |
414b4ff5 | 2006 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
2007 | unsigned long duration = jiffies - req->start_time; |
2008 | const int rw = rq_data_dir(req); | |
2009 | struct hd_struct *part; | |
2010 | int cpu; | |
2011 | ||
2012 | cpu = part_stat_lock(); | |
09e099d4 | 2013 | part = req->part; |
bc58ba94 JA |
2014 | |
2015 | part_stat_inc(cpu, part, ios[rw]); | |
2016 | part_stat_add(cpu, part, ticks[rw], duration); | |
2017 | part_round_stats(cpu, part); | |
316d315b | 2018 | part_dec_in_flight(part, rw); |
bc58ba94 | 2019 | |
6c23a968 | 2020 | hd_struct_put(part); |
bc58ba94 JA |
2021 | part_stat_unlock(); |
2022 | } | |
2023 | } | |
2024 | ||
3bcddeac | 2025 | /** |
9934c8c0 TH |
2026 | * blk_peek_request - peek at the top of a request queue |
2027 | * @q: request queue to peek at | |
2028 | * | |
2029 | * Description: | |
2030 | * Return the request at the top of @q. The returned request | |
2031 | * should be started using blk_start_request() before LLD starts | |
2032 | * processing it. | |
2033 | * | |
2034 | * Return: | |
2035 | * Pointer to the request at the top of @q if available. Null | |
2036 | * otherwise. | |
2037 | * | |
2038 | * Context: | |
2039 | * queue_lock must be held. | |
2040 | */ | |
2041 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
2042 | { |
2043 | struct request *rq; | |
2044 | int ret; | |
2045 | ||
2046 | while ((rq = __elv_next_request(q)) != NULL) { | |
2047 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
2048 | /* | |
2049 | * This is the first time the device driver | |
2050 | * sees this request (possibly after | |
2051 | * requeueing). Notify IO scheduler. | |
2052 | */ | |
33659ebb | 2053 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
2054 | elv_activate_rq(q, rq); |
2055 | ||
2056 | /* | |
2057 | * just mark as started even if we don't start | |
2058 | * it, a request that has been delayed should | |
2059 | * not be passed by new incoming requests | |
2060 | */ | |
2061 | rq->cmd_flags |= REQ_STARTED; | |
2062 | trace_block_rq_issue(q, rq); | |
2063 | } | |
2064 | ||
2065 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
2066 | q->end_sector = rq_end_sector(rq); | |
2067 | q->boundary_rq = NULL; | |
2068 | } | |
2069 | ||
2070 | if (rq->cmd_flags & REQ_DONTPREP) | |
2071 | break; | |
2072 | ||
2e46e8b2 | 2073 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
2074 | /* |
2075 | * make sure space for the drain appears we | |
2076 | * know we can do this because max_hw_segments | |
2077 | * has been adjusted to be one fewer than the | |
2078 | * device can handle | |
2079 | */ | |
2080 | rq->nr_phys_segments++; | |
2081 | } | |
2082 | ||
2083 | if (!q->prep_rq_fn) | |
2084 | break; | |
2085 | ||
2086 | ret = q->prep_rq_fn(q, rq); | |
2087 | if (ret == BLKPREP_OK) { | |
2088 | break; | |
2089 | } else if (ret == BLKPREP_DEFER) { | |
2090 | /* | |
2091 | * the request may have been (partially) prepped. | |
2092 | * we need to keep this request in the front to | |
2093 | * avoid resource deadlock. REQ_STARTED will | |
2094 | * prevent other fs requests from passing this one. | |
2095 | */ | |
2e46e8b2 | 2096 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
2097 | !(rq->cmd_flags & REQ_DONTPREP)) { |
2098 | /* | |
2099 | * remove the space for the drain we added | |
2100 | * so that we don't add it again | |
2101 | */ | |
2102 | --rq->nr_phys_segments; | |
2103 | } | |
2104 | ||
2105 | rq = NULL; | |
2106 | break; | |
2107 | } else if (ret == BLKPREP_KILL) { | |
2108 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
2109 | /* |
2110 | * Mark this request as started so we don't trigger | |
2111 | * any debug logic in the end I/O path. | |
2112 | */ | |
2113 | blk_start_request(rq); | |
40cbbb78 | 2114 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
2115 | } else { |
2116 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
2117 | break; | |
2118 | } | |
2119 | } | |
2120 | ||
2121 | return rq; | |
2122 | } | |
9934c8c0 | 2123 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 2124 | |
9934c8c0 | 2125 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 2126 | { |
9934c8c0 TH |
2127 | struct request_queue *q = rq->q; |
2128 | ||
158dbda0 TH |
2129 | BUG_ON(list_empty(&rq->queuelist)); |
2130 | BUG_ON(ELV_ON_HASH(rq)); | |
2131 | ||
2132 | list_del_init(&rq->queuelist); | |
2133 | ||
2134 | /* | |
2135 | * the time frame between a request being removed from the lists | |
2136 | * and to it is freed is accounted as io that is in progress at | |
2137 | * the driver side. | |
2138 | */ | |
9195291e | 2139 | if (blk_account_rq(rq)) { |
0a7ae2ff | 2140 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
2141 | set_io_start_time_ns(rq); |
2142 | } | |
158dbda0 TH |
2143 | } |
2144 | ||
9934c8c0 TH |
2145 | /** |
2146 | * blk_start_request - start request processing on the driver | |
2147 | * @req: request to dequeue | |
2148 | * | |
2149 | * Description: | |
2150 | * Dequeue @req and start timeout timer on it. This hands off the | |
2151 | * request to the driver. | |
2152 | * | |
2153 | * Block internal functions which don't want to start timer should | |
2154 | * call blk_dequeue_request(). | |
2155 | * | |
2156 | * Context: | |
2157 | * queue_lock must be held. | |
2158 | */ | |
2159 | void blk_start_request(struct request *req) | |
2160 | { | |
2161 | blk_dequeue_request(req); | |
2162 | ||
2163 | /* | |
5f49f631 TH |
2164 | * We are now handing the request to the hardware, initialize |
2165 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2166 | */ |
5f49f631 | 2167 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2168 | if (unlikely(blk_bidi_rq(req))) |
2169 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2170 | ||
9934c8c0 TH |
2171 | blk_add_timer(req); |
2172 | } | |
2173 | EXPORT_SYMBOL(blk_start_request); | |
2174 | ||
2175 | /** | |
2176 | * blk_fetch_request - fetch a request from a request queue | |
2177 | * @q: request queue to fetch a request from | |
2178 | * | |
2179 | * Description: | |
2180 | * Return the request at the top of @q. The request is started on | |
2181 | * return and LLD can start processing it immediately. | |
2182 | * | |
2183 | * Return: | |
2184 | * Pointer to the request at the top of @q if available. Null | |
2185 | * otherwise. | |
2186 | * | |
2187 | * Context: | |
2188 | * queue_lock must be held. | |
2189 | */ | |
2190 | struct request *blk_fetch_request(struct request_queue *q) | |
2191 | { | |
2192 | struct request *rq; | |
2193 | ||
2194 | rq = blk_peek_request(q); | |
2195 | if (rq) | |
2196 | blk_start_request(rq); | |
2197 | return rq; | |
2198 | } | |
2199 | EXPORT_SYMBOL(blk_fetch_request); | |
2200 | ||
3bcddeac | 2201 | /** |
2e60e022 | 2202 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2203 | * @req: the request being processed |
710027a4 | 2204 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2205 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2206 | * |
2207 | * Description: | |
8ebf9756 RD |
2208 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2209 | * the request structure even if @req doesn't have leftover. | |
2210 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2211 | * |
2212 | * This special helper function is only for request stacking drivers | |
2213 | * (e.g. request-based dm) so that they can handle partial completion. | |
2214 | * Actual device drivers should use blk_end_request instead. | |
2215 | * | |
2216 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2217 | * %false return from this function. | |
3bcddeac KU |
2218 | * |
2219 | * Return: | |
2e60e022 TH |
2220 | * %false - this request doesn't have any more data |
2221 | * %true - this request has more data | |
3bcddeac | 2222 | **/ |
2e60e022 | 2223 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2224 | { |
5450d3e1 | 2225 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2226 | struct bio *bio; |
2227 | ||
2e60e022 TH |
2228 | if (!req->bio) |
2229 | return false; | |
2230 | ||
5f3ea37c | 2231 | trace_block_rq_complete(req->q, req); |
2056a782 | 2232 | |
1da177e4 | 2233 | /* |
6f41469c TH |
2234 | * For fs requests, rq is just carrier of independent bio's |
2235 | * and each partial completion should be handled separately. | |
2236 | * Reset per-request error on each partial completion. | |
2237 | * | |
2238 | * TODO: tj: This is too subtle. It would be better to let | |
2239 | * low level drivers do what they see fit. | |
1da177e4 | 2240 | */ |
33659ebb | 2241 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2242 | req->errors = 0; |
2243 | ||
33659ebb CH |
2244 | if (error && req->cmd_type == REQ_TYPE_FS && |
2245 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2246 | char *error_type; |
2247 | ||
2248 | switch (error) { | |
2249 | case -ENOLINK: | |
2250 | error_type = "recoverable transport"; | |
2251 | break; | |
2252 | case -EREMOTEIO: | |
2253 | error_type = "critical target"; | |
2254 | break; | |
2255 | case -EBADE: | |
2256 | error_type = "critical nexus"; | |
2257 | break; | |
2258 | case -EIO: | |
2259 | default: | |
2260 | error_type = "I/O"; | |
2261 | break; | |
2262 | } | |
2263 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2264 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2265 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2266 | } |
2267 | ||
bc58ba94 | 2268 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2269 | |
1da177e4 LT |
2270 | total_bytes = bio_nbytes = 0; |
2271 | while ((bio = req->bio) != NULL) { | |
2272 | int nbytes; | |
2273 | ||
2274 | if (nr_bytes >= bio->bi_size) { | |
2275 | req->bio = bio->bi_next; | |
2276 | nbytes = bio->bi_size; | |
5bb23a68 | 2277 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2278 | next_idx = 0; |
2279 | bio_nbytes = 0; | |
2280 | } else { | |
2281 | int idx = bio->bi_idx + next_idx; | |
2282 | ||
af498d7f | 2283 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2284 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2285 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2286 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2287 | break; |
2288 | } | |
2289 | ||
2290 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2291 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2292 | ||
2293 | /* | |
2294 | * not a complete bvec done | |
2295 | */ | |
2296 | if (unlikely(nbytes > nr_bytes)) { | |
2297 | bio_nbytes += nr_bytes; | |
2298 | total_bytes += nr_bytes; | |
2299 | break; | |
2300 | } | |
2301 | ||
2302 | /* | |
2303 | * advance to the next vector | |
2304 | */ | |
2305 | next_idx++; | |
2306 | bio_nbytes += nbytes; | |
2307 | } | |
2308 | ||
2309 | total_bytes += nbytes; | |
2310 | nr_bytes -= nbytes; | |
2311 | ||
6728cb0e JA |
2312 | bio = req->bio; |
2313 | if (bio) { | |
1da177e4 LT |
2314 | /* |
2315 | * end more in this run, or just return 'not-done' | |
2316 | */ | |
2317 | if (unlikely(nr_bytes <= 0)) | |
2318 | break; | |
2319 | } | |
2320 | } | |
2321 | ||
2322 | /* | |
2323 | * completely done | |
2324 | */ | |
2e60e022 TH |
2325 | if (!req->bio) { |
2326 | /* | |
2327 | * Reset counters so that the request stacking driver | |
2328 | * can find how many bytes remain in the request | |
2329 | * later. | |
2330 | */ | |
a2dec7b3 | 2331 | req->__data_len = 0; |
2e60e022 TH |
2332 | return false; |
2333 | } | |
1da177e4 LT |
2334 | |
2335 | /* | |
2336 | * if the request wasn't completed, update state | |
2337 | */ | |
2338 | if (bio_nbytes) { | |
5bb23a68 | 2339 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2340 | bio->bi_idx += next_idx; |
2341 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2342 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2343 | } | |
2344 | ||
a2dec7b3 | 2345 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2346 | req->buffer = bio_data(req->bio); |
2347 | ||
2348 | /* update sector only for requests with clear definition of sector */ | |
e2a60da7 | 2349 | if (req->cmd_type == REQ_TYPE_FS) |
a2dec7b3 | 2350 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2351 | |
80a761fd TH |
2352 | /* mixed attributes always follow the first bio */ |
2353 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2354 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2355 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2356 | } | |
2357 | ||
2e46e8b2 TH |
2358 | /* |
2359 | * If total number of sectors is less than the first segment | |
2360 | * size, something has gone terribly wrong. | |
2361 | */ | |
2362 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2363 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2364 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2365 | } |
2366 | ||
2367 | /* recalculate the number of segments */ | |
1da177e4 | 2368 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2369 | |
2e60e022 | 2370 | return true; |
1da177e4 | 2371 | } |
2e60e022 | 2372 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2373 | |
2e60e022 TH |
2374 | static bool blk_update_bidi_request(struct request *rq, int error, |
2375 | unsigned int nr_bytes, | |
2376 | unsigned int bidi_bytes) | |
5efccd17 | 2377 | { |
2e60e022 TH |
2378 | if (blk_update_request(rq, error, nr_bytes)) |
2379 | return true; | |
5efccd17 | 2380 | |
2e60e022 TH |
2381 | /* Bidi request must be completed as a whole */ |
2382 | if (unlikely(blk_bidi_rq(rq)) && | |
2383 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2384 | return true; | |
5efccd17 | 2385 | |
e2e1a148 JA |
2386 | if (blk_queue_add_random(rq->q)) |
2387 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2388 | |
2389 | return false; | |
1da177e4 LT |
2390 | } |
2391 | ||
28018c24 JB |
2392 | /** |
2393 | * blk_unprep_request - unprepare a request | |
2394 | * @req: the request | |
2395 | * | |
2396 | * This function makes a request ready for complete resubmission (or | |
2397 | * completion). It happens only after all error handling is complete, | |
2398 | * so represents the appropriate moment to deallocate any resources | |
2399 | * that were allocated to the request in the prep_rq_fn. The queue | |
2400 | * lock is held when calling this. | |
2401 | */ | |
2402 | void blk_unprep_request(struct request *req) | |
2403 | { | |
2404 | struct request_queue *q = req->q; | |
2405 | ||
2406 | req->cmd_flags &= ~REQ_DONTPREP; | |
2407 | if (q->unprep_rq_fn) | |
2408 | q->unprep_rq_fn(q, req); | |
2409 | } | |
2410 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2411 | ||
1da177e4 LT |
2412 | /* |
2413 | * queue lock must be held | |
2414 | */ | |
2e60e022 | 2415 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2416 | { |
b8286239 KU |
2417 | if (blk_rq_tagged(req)) |
2418 | blk_queue_end_tag(req->q, req); | |
2419 | ||
ba396a6c | 2420 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2421 | |
33659ebb | 2422 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2423 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2424 | |
e78042e5 MA |
2425 | blk_delete_timer(req); |
2426 | ||
28018c24 JB |
2427 | if (req->cmd_flags & REQ_DONTPREP) |
2428 | blk_unprep_request(req); | |
2429 | ||
2430 | ||
bc58ba94 | 2431 | blk_account_io_done(req); |
b8286239 | 2432 | |
1da177e4 | 2433 | if (req->end_io) |
8ffdc655 | 2434 | req->end_io(req, error); |
b8286239 KU |
2435 | else { |
2436 | if (blk_bidi_rq(req)) | |
2437 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2438 | ||
1da177e4 | 2439 | __blk_put_request(req->q, req); |
b8286239 | 2440 | } |
1da177e4 LT |
2441 | } |
2442 | ||
3b11313a | 2443 | /** |
2e60e022 TH |
2444 | * blk_end_bidi_request - Complete a bidi request |
2445 | * @rq: the request to complete | |
2446 | * @error: %0 for success, < %0 for error | |
2447 | * @nr_bytes: number of bytes to complete @rq | |
2448 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2449 | * |
2450 | * Description: | |
e3a04fe3 | 2451 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2452 | * Drivers that supports bidi can safely call this member for any |
2453 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2454 | * just ignored. | |
336cdb40 KU |
2455 | * |
2456 | * Return: | |
2e60e022 TH |
2457 | * %false - we are done with this request |
2458 | * %true - still buffers pending for this request | |
a0cd1285 | 2459 | **/ |
b1f74493 | 2460 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2461 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2462 | { | |
336cdb40 | 2463 | struct request_queue *q = rq->q; |
2e60e022 | 2464 | unsigned long flags; |
32fab448 | 2465 | |
2e60e022 TH |
2466 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2467 | return true; | |
32fab448 | 2468 | |
336cdb40 | 2469 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2470 | blk_finish_request(rq, error); |
336cdb40 KU |
2471 | spin_unlock_irqrestore(q->queue_lock, flags); |
2472 | ||
2e60e022 | 2473 | return false; |
32fab448 KU |
2474 | } |
2475 | ||
336cdb40 | 2476 | /** |
2e60e022 TH |
2477 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2478 | * @rq: the request to complete | |
710027a4 | 2479 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2480 | * @nr_bytes: number of bytes to complete @rq |
2481 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2482 | * |
2483 | * Description: | |
2e60e022 TH |
2484 | * Identical to blk_end_bidi_request() except that queue lock is |
2485 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2486 | * |
2487 | * Return: | |
2e60e022 TH |
2488 | * %false - we are done with this request |
2489 | * %true - still buffers pending for this request | |
336cdb40 | 2490 | **/ |
4853abaa | 2491 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2492 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2493 | { |
2e60e022 TH |
2494 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2495 | return true; | |
336cdb40 | 2496 | |
2e60e022 | 2497 | blk_finish_request(rq, error); |
336cdb40 | 2498 | |
2e60e022 | 2499 | return false; |
336cdb40 | 2500 | } |
e19a3ab0 KU |
2501 | |
2502 | /** | |
2503 | * blk_end_request - Helper function for drivers to complete the request. | |
2504 | * @rq: the request being processed | |
710027a4 | 2505 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2506 | * @nr_bytes: number of bytes to complete |
2507 | * | |
2508 | * Description: | |
2509 | * Ends I/O on a number of bytes attached to @rq. | |
2510 | * If @rq has leftover, sets it up for the next range of segments. | |
2511 | * | |
2512 | * Return: | |
b1f74493 FT |
2513 | * %false - we are done with this request |
2514 | * %true - still buffers pending for this request | |
e19a3ab0 | 2515 | **/ |
b1f74493 | 2516 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2517 | { |
b1f74493 | 2518 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2519 | } |
56ad1740 | 2520 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2521 | |
2522 | /** | |
b1f74493 FT |
2523 | * blk_end_request_all - Helper function for drives to finish the request. |
2524 | * @rq: the request to finish | |
8ebf9756 | 2525 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2526 | * |
2527 | * Description: | |
b1f74493 FT |
2528 | * Completely finish @rq. |
2529 | */ | |
2530 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2531 | { |
b1f74493 FT |
2532 | bool pending; |
2533 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2534 | |
b1f74493 FT |
2535 | if (unlikely(blk_bidi_rq(rq))) |
2536 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2537 | |
b1f74493 FT |
2538 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2539 | BUG_ON(pending); | |
2540 | } | |
56ad1740 | 2541 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2542 | |
b1f74493 FT |
2543 | /** |
2544 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2545 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2546 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2547 | * |
2548 | * Description: | |
2549 | * Complete the current consecutively mapped chunk from @rq. | |
2550 | * | |
2551 | * Return: | |
2552 | * %false - we are done with this request | |
2553 | * %true - still buffers pending for this request | |
2554 | */ | |
2555 | bool blk_end_request_cur(struct request *rq, int error) | |
2556 | { | |
2557 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2558 | } |
56ad1740 | 2559 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2560 | |
80a761fd TH |
2561 | /** |
2562 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2563 | * @rq: the request to finish till the next failure boundary for | |
2564 | * @error: must be negative errno | |
2565 | * | |
2566 | * Description: | |
2567 | * Complete @rq till the next failure boundary. | |
2568 | * | |
2569 | * Return: | |
2570 | * %false - we are done with this request | |
2571 | * %true - still buffers pending for this request | |
2572 | */ | |
2573 | bool blk_end_request_err(struct request *rq, int error) | |
2574 | { | |
2575 | WARN_ON(error >= 0); | |
2576 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2577 | } | |
2578 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2579 | ||
e3a04fe3 | 2580 | /** |
b1f74493 FT |
2581 | * __blk_end_request - Helper function for drivers to complete the request. |
2582 | * @rq: the request being processed | |
2583 | * @error: %0 for success, < %0 for error | |
2584 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2585 | * |
2586 | * Description: | |
b1f74493 | 2587 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2588 | * |
2589 | * Return: | |
b1f74493 FT |
2590 | * %false - we are done with this request |
2591 | * %true - still buffers pending for this request | |
e3a04fe3 | 2592 | **/ |
b1f74493 | 2593 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2594 | { |
b1f74493 | 2595 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2596 | } |
56ad1740 | 2597 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2598 | |
32fab448 | 2599 | /** |
b1f74493 FT |
2600 | * __blk_end_request_all - Helper function for drives to finish the request. |
2601 | * @rq: the request to finish | |
8ebf9756 | 2602 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2603 | * |
2604 | * Description: | |
b1f74493 | 2605 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2606 | */ |
b1f74493 | 2607 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2608 | { |
b1f74493 FT |
2609 | bool pending; |
2610 | unsigned int bidi_bytes = 0; | |
2611 | ||
2612 | if (unlikely(blk_bidi_rq(rq))) | |
2613 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2614 | ||
2615 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2616 | BUG_ON(pending); | |
32fab448 | 2617 | } |
56ad1740 | 2618 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2619 | |
e19a3ab0 | 2620 | /** |
b1f74493 FT |
2621 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2622 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2623 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2624 | * |
2625 | * Description: | |
b1f74493 FT |
2626 | * Complete the current consecutively mapped chunk from @rq. Must |
2627 | * be called with queue lock held. | |
e19a3ab0 KU |
2628 | * |
2629 | * Return: | |
b1f74493 FT |
2630 | * %false - we are done with this request |
2631 | * %true - still buffers pending for this request | |
2632 | */ | |
2633 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2634 | { |
b1f74493 | 2635 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2636 | } |
56ad1740 | 2637 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2638 | |
80a761fd TH |
2639 | /** |
2640 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2641 | * @rq: the request to finish till the next failure boundary for | |
2642 | * @error: must be negative errno | |
2643 | * | |
2644 | * Description: | |
2645 | * Complete @rq till the next failure boundary. Must be called | |
2646 | * with queue lock held. | |
2647 | * | |
2648 | * Return: | |
2649 | * %false - we are done with this request | |
2650 | * %true - still buffers pending for this request | |
2651 | */ | |
2652 | bool __blk_end_request_err(struct request *rq, int error) | |
2653 | { | |
2654 | WARN_ON(error >= 0); | |
2655 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2656 | } | |
2657 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2658 | ||
86db1e29 JA |
2659 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2660 | struct bio *bio) | |
1da177e4 | 2661 | { |
a82afdfc | 2662 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2663 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2664 | |
fb2dce86 DW |
2665 | if (bio_has_data(bio)) { |
2666 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2667 | rq->buffer = bio_data(bio); |
2668 | } | |
a2dec7b3 | 2669 | rq->__data_len = bio->bi_size; |
1da177e4 | 2670 | rq->bio = rq->biotail = bio; |
1da177e4 | 2671 | |
66846572 N |
2672 | if (bio->bi_bdev) |
2673 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2674 | } | |
1da177e4 | 2675 | |
2d4dc890 IL |
2676 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2677 | /** | |
2678 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2679 | * @rq: the request to be flushed | |
2680 | * | |
2681 | * Description: | |
2682 | * Flush all pages in @rq. | |
2683 | */ | |
2684 | void rq_flush_dcache_pages(struct request *rq) | |
2685 | { | |
2686 | struct req_iterator iter; | |
2687 | struct bio_vec *bvec; | |
2688 | ||
2689 | rq_for_each_segment(bvec, rq, iter) | |
2690 | flush_dcache_page(bvec->bv_page); | |
2691 | } | |
2692 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2693 | #endif | |
2694 | ||
ef9e3fac KU |
2695 | /** |
2696 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2697 | * @q : the queue of the device being checked | |
2698 | * | |
2699 | * Description: | |
2700 | * Check if underlying low-level drivers of a device are busy. | |
2701 | * If the drivers want to export their busy state, they must set own | |
2702 | * exporting function using blk_queue_lld_busy() first. | |
2703 | * | |
2704 | * Basically, this function is used only by request stacking drivers | |
2705 | * to stop dispatching requests to underlying devices when underlying | |
2706 | * devices are busy. This behavior helps more I/O merging on the queue | |
2707 | * of the request stacking driver and prevents I/O throughput regression | |
2708 | * on burst I/O load. | |
2709 | * | |
2710 | * Return: | |
2711 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2712 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2713 | */ | |
2714 | int blk_lld_busy(struct request_queue *q) | |
2715 | { | |
2716 | if (q->lld_busy_fn) | |
2717 | return q->lld_busy_fn(q); | |
2718 | ||
2719 | return 0; | |
2720 | } | |
2721 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2722 | ||
b0fd271d KU |
2723 | /** |
2724 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2725 | * @rq: the clone request to be cleaned up | |
2726 | * | |
2727 | * Description: | |
2728 | * Free all bios in @rq for a cloned request. | |
2729 | */ | |
2730 | void blk_rq_unprep_clone(struct request *rq) | |
2731 | { | |
2732 | struct bio *bio; | |
2733 | ||
2734 | while ((bio = rq->bio) != NULL) { | |
2735 | rq->bio = bio->bi_next; | |
2736 | ||
2737 | bio_put(bio); | |
2738 | } | |
2739 | } | |
2740 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2741 | ||
2742 | /* | |
2743 | * Copy attributes of the original request to the clone request. | |
2744 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2745 | */ | |
2746 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2747 | { | |
2748 | dst->cpu = src->cpu; | |
3a2edd0d | 2749 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2750 | dst->cmd_type = src->cmd_type; |
2751 | dst->__sector = blk_rq_pos(src); | |
2752 | dst->__data_len = blk_rq_bytes(src); | |
2753 | dst->nr_phys_segments = src->nr_phys_segments; | |
2754 | dst->ioprio = src->ioprio; | |
2755 | dst->extra_len = src->extra_len; | |
2756 | } | |
2757 | ||
2758 | /** | |
2759 | * blk_rq_prep_clone - Helper function to setup clone request | |
2760 | * @rq: the request to be setup | |
2761 | * @rq_src: original request to be cloned | |
2762 | * @bs: bio_set that bios for clone are allocated from | |
2763 | * @gfp_mask: memory allocation mask for bio | |
2764 | * @bio_ctr: setup function to be called for each clone bio. | |
2765 | * Returns %0 for success, non %0 for failure. | |
2766 | * @data: private data to be passed to @bio_ctr | |
2767 | * | |
2768 | * Description: | |
2769 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2770 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2771 | * are not copied, and copying such parts is the caller's responsibility. | |
2772 | * Also, pages which the original bios are pointing to are not copied | |
2773 | * and the cloned bios just point same pages. | |
2774 | * So cloned bios must be completed before original bios, which means | |
2775 | * the caller must complete @rq before @rq_src. | |
2776 | */ | |
2777 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2778 | struct bio_set *bs, gfp_t gfp_mask, | |
2779 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2780 | void *data) | |
2781 | { | |
2782 | struct bio *bio, *bio_src; | |
2783 | ||
2784 | if (!bs) | |
2785 | bs = fs_bio_set; | |
2786 | ||
2787 | blk_rq_init(NULL, rq); | |
2788 | ||
2789 | __rq_for_each_bio(bio_src, rq_src) { | |
bf800ef1 | 2790 | bio = bio_clone_bioset(bio_src, gfp_mask, bs); |
b0fd271d KU |
2791 | if (!bio) |
2792 | goto free_and_out; | |
2793 | ||
b0fd271d KU |
2794 | if (bio_ctr && bio_ctr(bio, bio_src, data)) |
2795 | goto free_and_out; | |
2796 | ||
2797 | if (rq->bio) { | |
2798 | rq->biotail->bi_next = bio; | |
2799 | rq->biotail = bio; | |
2800 | } else | |
2801 | rq->bio = rq->biotail = bio; | |
2802 | } | |
2803 | ||
2804 | __blk_rq_prep_clone(rq, rq_src); | |
2805 | ||
2806 | return 0; | |
2807 | ||
2808 | free_and_out: | |
2809 | if (bio) | |
4254bba1 | 2810 | bio_put(bio); |
b0fd271d KU |
2811 | blk_rq_unprep_clone(rq); |
2812 | ||
2813 | return -ENOMEM; | |
2814 | } | |
2815 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2816 | ||
18887ad9 | 2817 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2818 | { |
2819 | return queue_work(kblockd_workqueue, work); | |
2820 | } | |
1da177e4 LT |
2821 | EXPORT_SYMBOL(kblockd_schedule_work); |
2822 | ||
e43473b7 VG |
2823 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2824 | struct delayed_work *dwork, unsigned long delay) | |
2825 | { | |
2826 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2827 | } | |
2828 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2829 | ||
73c10101 JA |
2830 | #define PLUG_MAGIC 0x91827364 |
2831 | ||
75df7136 SJ |
2832 | /** |
2833 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2834 | * @plug: The &struct blk_plug that needs to be initialized | |
2835 | * | |
2836 | * Description: | |
2837 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2838 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2839 | * blk_finish_plug(). This is important from a performance perspective, but | |
2840 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2841 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2842 | * page belonging to that request that is currently residing in our private | |
2843 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2844 | * this kind of deadlock. | |
2845 | */ | |
73c10101 JA |
2846 | void blk_start_plug(struct blk_plug *plug) |
2847 | { | |
2848 | struct task_struct *tsk = current; | |
2849 | ||
2850 | plug->magic = PLUG_MAGIC; | |
2851 | INIT_LIST_HEAD(&plug->list); | |
048c9374 | 2852 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 JA |
2853 | plug->should_sort = 0; |
2854 | ||
2855 | /* | |
2856 | * If this is a nested plug, don't actually assign it. It will be | |
2857 | * flushed on its own. | |
2858 | */ | |
2859 | if (!tsk->plug) { | |
2860 | /* | |
2861 | * Store ordering should not be needed here, since a potential | |
2862 | * preempt will imply a full memory barrier | |
2863 | */ | |
2864 | tsk->plug = plug; | |
2865 | } | |
2866 | } | |
2867 | EXPORT_SYMBOL(blk_start_plug); | |
2868 | ||
2869 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2870 | { | |
2871 | struct request *rqa = container_of(a, struct request, queuelist); | |
2872 | struct request *rqb = container_of(b, struct request, queuelist); | |
2873 | ||
f83e8261 | 2874 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2875 | } |
2876 | ||
49cac01e JA |
2877 | /* |
2878 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2879 | * until a safe kblockd context. We due this to avoid accidental big | |
2880 | * additional stack usage in driver dispatch, in places where the originally | |
2881 | * plugger did not intend it. | |
2882 | */ | |
f6603783 | 2883 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 2884 | bool from_schedule) |
99e22598 | 2885 | __releases(q->queue_lock) |
94b5eb28 | 2886 | { |
49cac01e | 2887 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 | 2888 | |
8ba61435 TH |
2889 | /* |
2890 | * Don't mess with dead queue. | |
2891 | */ | |
2892 | if (unlikely(blk_queue_dead(q))) { | |
2893 | spin_unlock(q->queue_lock); | |
2894 | return; | |
2895 | } | |
2896 | ||
99e22598 JA |
2897 | /* |
2898 | * If we are punting this to kblockd, then we can safely drop | |
2899 | * the queue_lock before waking kblockd (which needs to take | |
2900 | * this lock). | |
2901 | */ | |
2902 | if (from_schedule) { | |
2903 | spin_unlock(q->queue_lock); | |
24ecfbe2 | 2904 | blk_run_queue_async(q); |
99e22598 | 2905 | } else { |
24ecfbe2 | 2906 | __blk_run_queue(q); |
99e22598 JA |
2907 | spin_unlock(q->queue_lock); |
2908 | } | |
2909 | ||
94b5eb28 JA |
2910 | } |
2911 | ||
74018dc3 | 2912 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
048c9374 N |
2913 | { |
2914 | LIST_HEAD(callbacks); | |
2915 | ||
2a7d5559 SL |
2916 | while (!list_empty(&plug->cb_list)) { |
2917 | list_splice_init(&plug->cb_list, &callbacks); | |
048c9374 | 2918 | |
2a7d5559 SL |
2919 | while (!list_empty(&callbacks)) { |
2920 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
048c9374 N |
2921 | struct blk_plug_cb, |
2922 | list); | |
2a7d5559 | 2923 | list_del(&cb->list); |
74018dc3 | 2924 | cb->callback(cb, from_schedule); |
2a7d5559 | 2925 | } |
048c9374 N |
2926 | } |
2927 | } | |
2928 | ||
9cbb1750 N |
2929 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
2930 | int size) | |
2931 | { | |
2932 | struct blk_plug *plug = current->plug; | |
2933 | struct blk_plug_cb *cb; | |
2934 | ||
2935 | if (!plug) | |
2936 | return NULL; | |
2937 | ||
2938 | list_for_each_entry(cb, &plug->cb_list, list) | |
2939 | if (cb->callback == unplug && cb->data == data) | |
2940 | return cb; | |
2941 | ||
2942 | /* Not currently on the callback list */ | |
2943 | BUG_ON(size < sizeof(*cb)); | |
2944 | cb = kzalloc(size, GFP_ATOMIC); | |
2945 | if (cb) { | |
2946 | cb->data = data; | |
2947 | cb->callback = unplug; | |
2948 | list_add(&cb->list, &plug->cb_list); | |
2949 | } | |
2950 | return cb; | |
2951 | } | |
2952 | EXPORT_SYMBOL(blk_check_plugged); | |
2953 | ||
49cac01e | 2954 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
2955 | { |
2956 | struct request_queue *q; | |
2957 | unsigned long flags; | |
2958 | struct request *rq; | |
109b8129 | 2959 | LIST_HEAD(list); |
94b5eb28 | 2960 | unsigned int depth; |
73c10101 JA |
2961 | |
2962 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2963 | ||
74018dc3 | 2964 | flush_plug_callbacks(plug, from_schedule); |
73c10101 JA |
2965 | if (list_empty(&plug->list)) |
2966 | return; | |
2967 | ||
109b8129 N |
2968 | list_splice_init(&plug->list, &list); |
2969 | ||
2970 | if (plug->should_sort) { | |
2971 | list_sort(NULL, &list, plug_rq_cmp); | |
2972 | plug->should_sort = 0; | |
2973 | } | |
73c10101 JA |
2974 | |
2975 | q = NULL; | |
94b5eb28 | 2976 | depth = 0; |
18811272 JA |
2977 | |
2978 | /* | |
2979 | * Save and disable interrupts here, to avoid doing it for every | |
2980 | * queue lock we have to take. | |
2981 | */ | |
73c10101 | 2982 | local_irq_save(flags); |
109b8129 N |
2983 | while (!list_empty(&list)) { |
2984 | rq = list_entry_rq(list.next); | |
73c10101 | 2985 | list_del_init(&rq->queuelist); |
73c10101 JA |
2986 | BUG_ON(!rq->q); |
2987 | if (rq->q != q) { | |
99e22598 JA |
2988 | /* |
2989 | * This drops the queue lock | |
2990 | */ | |
2991 | if (q) | |
49cac01e | 2992 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2993 | q = rq->q; |
94b5eb28 | 2994 | depth = 0; |
73c10101 JA |
2995 | spin_lock(q->queue_lock); |
2996 | } | |
8ba61435 TH |
2997 | |
2998 | /* | |
2999 | * Short-circuit if @q is dead | |
3000 | */ | |
3001 | if (unlikely(blk_queue_dead(q))) { | |
3002 | __blk_end_request_all(rq, -ENODEV); | |
3003 | continue; | |
3004 | } | |
3005 | ||
73c10101 JA |
3006 | /* |
3007 | * rq is already accounted, so use raw insert | |
3008 | */ | |
401a18e9 JA |
3009 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
3010 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
3011 | else | |
3012 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
3013 | |
3014 | depth++; | |
73c10101 JA |
3015 | } |
3016 | ||
99e22598 JA |
3017 | /* |
3018 | * This drops the queue lock | |
3019 | */ | |
3020 | if (q) | |
49cac01e | 3021 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 3022 | |
73c10101 JA |
3023 | local_irq_restore(flags); |
3024 | } | |
73c10101 JA |
3025 | |
3026 | void blk_finish_plug(struct blk_plug *plug) | |
3027 | { | |
f6603783 | 3028 | blk_flush_plug_list(plug, false); |
73c10101 | 3029 | |
88b996cd CH |
3030 | if (plug == current->plug) |
3031 | current->plug = NULL; | |
73c10101 | 3032 | } |
88b996cd | 3033 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 3034 | |
1da177e4 LT |
3035 | int __init blk_dev_init(void) |
3036 | { | |
9eb55b03 NK |
3037 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
3038 | sizeof(((struct request *)0)->cmd_flags)); | |
3039 | ||
89b90be2 TH |
3040 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
3041 | kblockd_workqueue = alloc_workqueue("kblockd", | |
3042 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
3043 | if (!kblockd_workqueue) |
3044 | panic("Failed to create kblockd\n"); | |
3045 | ||
3046 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 3047 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 3048 | |
8324aa91 | 3049 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 3050 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 3051 | |
d38ecf93 | 3052 | return 0; |
1da177e4 | 3053 | } |