Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
c17bb495 | 29 | #include <linux/fault-inject.h> |
73c10101 | 30 | #include <linux/list_sort.h> |
e3c78ca5 | 31 | #include <linux/delay.h> |
55782138 LZ |
32 | |
33 | #define CREATE_TRACE_POINTS | |
34 | #include <trace/events/block.h> | |
1da177e4 | 35 | |
8324aa91 JA |
36 | #include "blk.h" |
37 | ||
d07335e5 | 38 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 40 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 41 | |
1da177e4 LT |
42 | /* |
43 | * For the allocated request tables | |
44 | */ | |
5ece6c52 | 45 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
46 | |
47 | /* | |
48 | * For queue allocation | |
49 | */ | |
6728cb0e | 50 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 51 | |
1da177e4 LT |
52 | /* |
53 | * Controlling structure to kblockd | |
54 | */ | |
ff856bad | 55 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 56 | |
26b8256e JA |
57 | static void drive_stat_acct(struct request *rq, int new_io) |
58 | { | |
28f13702 | 59 | struct hd_struct *part; |
26b8256e | 60 | int rw = rq_data_dir(rq); |
c9959059 | 61 | int cpu; |
26b8256e | 62 | |
c2553b58 | 63 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
64 | return; |
65 | ||
074a7aca | 66 | cpu = part_stat_lock(); |
c9959059 | 67 | |
09e099d4 JM |
68 | if (!new_io) { |
69 | part = rq->part; | |
074a7aca | 70 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
71 | } else { |
72 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 73 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
74 | /* |
75 | * The partition is already being removed, | |
76 | * the request will be accounted on the disk only | |
77 | * | |
78 | * We take a reference on disk->part0 although that | |
79 | * partition will never be deleted, so we can treat | |
80 | * it as any other partition. | |
81 | */ | |
82 | part = &rq->rq_disk->part0; | |
6c23a968 | 83 | hd_struct_get(part); |
09e099d4 | 84 | } |
074a7aca | 85 | part_round_stats(cpu, part); |
316d315b | 86 | part_inc_in_flight(part, rw); |
09e099d4 | 87 | rq->part = part; |
26b8256e | 88 | } |
e71bf0d0 | 89 | |
074a7aca | 90 | part_stat_unlock(); |
26b8256e JA |
91 | } |
92 | ||
8324aa91 | 93 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
94 | { |
95 | int nr; | |
96 | ||
97 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
98 | if (nr > q->nr_requests) | |
99 | nr = q->nr_requests; | |
100 | q->nr_congestion_on = nr; | |
101 | ||
102 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
103 | if (nr < 1) | |
104 | nr = 1; | |
105 | q->nr_congestion_off = nr; | |
106 | } | |
107 | ||
1da177e4 LT |
108 | /** |
109 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
110 | * @bdev: device | |
111 | * | |
112 | * Locates the passed device's request queue and returns the address of its | |
113 | * backing_dev_info | |
114 | * | |
115 | * Will return NULL if the request queue cannot be located. | |
116 | */ | |
117 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
118 | { | |
119 | struct backing_dev_info *ret = NULL; | |
165125e1 | 120 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
121 | |
122 | if (q) | |
123 | ret = &q->backing_dev_info; | |
124 | return ret; | |
125 | } | |
1da177e4 LT |
126 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
127 | ||
2a4aa30c | 128 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 129 | { |
1afb20f3 FT |
130 | memset(rq, 0, sizeof(*rq)); |
131 | ||
1da177e4 | 132 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 133 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 134 | rq->cpu = -1; |
63a71386 | 135 | rq->q = q; |
a2dec7b3 | 136 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
137 | INIT_HLIST_NODE(&rq->hash); |
138 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 139 | rq->cmd = rq->__cmd; |
e2494e1b | 140 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 141 | rq->tag = -1; |
1da177e4 | 142 | rq->ref_count = 1; |
b243ddcb | 143 | rq->start_time = jiffies; |
9195291e | 144 | set_start_time_ns(rq); |
09e099d4 | 145 | rq->part = NULL; |
1da177e4 | 146 | } |
2a4aa30c | 147 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 148 | |
5bb23a68 N |
149 | static void req_bio_endio(struct request *rq, struct bio *bio, |
150 | unsigned int nbytes, int error) | |
1da177e4 | 151 | { |
143a87f4 TH |
152 | if (error) |
153 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
154 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
155 | error = -EIO; | |
797e7dbb | 156 | |
143a87f4 TH |
157 | if (unlikely(nbytes > bio->bi_size)) { |
158 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
159 | __func__, nbytes, bio->bi_size); | |
160 | nbytes = bio->bi_size; | |
5bb23a68 | 161 | } |
797e7dbb | 162 | |
143a87f4 TH |
163 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
164 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 165 | |
143a87f4 TH |
166 | bio->bi_size -= nbytes; |
167 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 168 | |
143a87f4 TH |
169 | if (bio_integrity(bio)) |
170 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 171 | |
143a87f4 TH |
172 | /* don't actually finish bio if it's part of flush sequence */ |
173 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
174 | bio_endio(bio, error); | |
1da177e4 | 175 | } |
1da177e4 | 176 | |
1da177e4 LT |
177 | void blk_dump_rq_flags(struct request *rq, char *msg) |
178 | { | |
179 | int bit; | |
180 | ||
6728cb0e | 181 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
182 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
183 | rq->cmd_flags); | |
1da177e4 | 184 | |
83096ebf TH |
185 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
186 | (unsigned long long)blk_rq_pos(rq), | |
187 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 188 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 189 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 190 | |
33659ebb | 191 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 192 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 193 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
194 | printk("%02x ", rq->cmd[bit]); |
195 | printk("\n"); | |
196 | } | |
197 | } | |
1da177e4 LT |
198 | EXPORT_SYMBOL(blk_dump_rq_flags); |
199 | ||
3cca6dc1 | 200 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 201 | { |
3cca6dc1 | 202 | struct request_queue *q; |
1da177e4 | 203 | |
3cca6dc1 JA |
204 | q = container_of(work, struct request_queue, delay_work.work); |
205 | spin_lock_irq(q->queue_lock); | |
24ecfbe2 | 206 | __blk_run_queue(q); |
3cca6dc1 | 207 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 208 | } |
1da177e4 LT |
209 | |
210 | /** | |
3cca6dc1 JA |
211 | * blk_delay_queue - restart queueing after defined interval |
212 | * @q: The &struct request_queue in question | |
213 | * @msecs: Delay in msecs | |
1da177e4 LT |
214 | * |
215 | * Description: | |
3cca6dc1 JA |
216 | * Sometimes queueing needs to be postponed for a little while, to allow |
217 | * resources to come back. This function will make sure that queueing is | |
218 | * restarted around the specified time. | |
219 | */ | |
220 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 221 | { |
4521cc4e JA |
222 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
223 | msecs_to_jiffies(msecs)); | |
2ad8b1ef | 224 | } |
3cca6dc1 | 225 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 226 | |
1da177e4 LT |
227 | /** |
228 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 229 | * @q: The &struct request_queue in question |
1da177e4 LT |
230 | * |
231 | * Description: | |
232 | * blk_start_queue() will clear the stop flag on the queue, and call | |
233 | * the request_fn for the queue if it was in a stopped state when | |
234 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
235 | **/ | |
165125e1 | 236 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 237 | { |
a038e253 PBG |
238 | WARN_ON(!irqs_disabled()); |
239 | ||
75ad23bc | 240 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
24ecfbe2 | 241 | __blk_run_queue(q); |
1da177e4 | 242 | } |
1da177e4 LT |
243 | EXPORT_SYMBOL(blk_start_queue); |
244 | ||
245 | /** | |
246 | * blk_stop_queue - stop a queue | |
165125e1 | 247 | * @q: The &struct request_queue in question |
1da177e4 LT |
248 | * |
249 | * Description: | |
250 | * The Linux block layer assumes that a block driver will consume all | |
251 | * entries on the request queue when the request_fn strategy is called. | |
252 | * Often this will not happen, because of hardware limitations (queue | |
253 | * depth settings). If a device driver gets a 'queue full' response, | |
254 | * or if it simply chooses not to queue more I/O at one point, it can | |
255 | * call this function to prevent the request_fn from being called until | |
256 | * the driver has signalled it's ready to go again. This happens by calling | |
257 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
258 | **/ | |
165125e1 | 259 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 260 | { |
ad3d9d7e | 261 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 262 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
263 | } |
264 | EXPORT_SYMBOL(blk_stop_queue); | |
265 | ||
266 | /** | |
267 | * blk_sync_queue - cancel any pending callbacks on a queue | |
268 | * @q: the queue | |
269 | * | |
270 | * Description: | |
271 | * The block layer may perform asynchronous callback activity | |
272 | * on a queue, such as calling the unplug function after a timeout. | |
273 | * A block device may call blk_sync_queue to ensure that any | |
274 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 275 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
276 | * that its ->make_request_fn will not re-add plugging prior to calling |
277 | * this function. | |
278 | * | |
da527770 VG |
279 | * This function does not cancel any asynchronous activity arising |
280 | * out of elevator or throttling code. That would require elevaotor_exit() | |
281 | * and blk_throtl_exit() to be called with queue lock initialized. | |
282 | * | |
1da177e4 LT |
283 | */ |
284 | void blk_sync_queue(struct request_queue *q) | |
285 | { | |
70ed28b9 | 286 | del_timer_sync(&q->timeout); |
3cca6dc1 | 287 | cancel_delayed_work_sync(&q->delay_work); |
1da177e4 LT |
288 | } |
289 | EXPORT_SYMBOL(blk_sync_queue); | |
290 | ||
291 | /** | |
80a4b58e | 292 | * __blk_run_queue - run a single device queue |
1da177e4 | 293 | * @q: The queue to run |
80a4b58e JA |
294 | * |
295 | * Description: | |
296 | * See @blk_run_queue. This variant must be called with the queue lock | |
24ecfbe2 | 297 | * held and interrupts disabled. |
1da177e4 | 298 | */ |
24ecfbe2 | 299 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 300 | { |
a538cd03 TH |
301 | if (unlikely(blk_queue_stopped(q))) |
302 | return; | |
303 | ||
c21e6beb | 304 | q->request_fn(q); |
75ad23bc NP |
305 | } |
306 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 307 | |
24ecfbe2 CH |
308 | /** |
309 | * blk_run_queue_async - run a single device queue in workqueue context | |
310 | * @q: The queue to run | |
311 | * | |
312 | * Description: | |
313 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf | |
314 | * of us. | |
315 | */ | |
316 | void blk_run_queue_async(struct request_queue *q) | |
317 | { | |
3ec717b7 SL |
318 | if (likely(!blk_queue_stopped(q))) { |
319 | __cancel_delayed_work(&q->delay_work); | |
24ecfbe2 | 320 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
3ec717b7 | 321 | } |
24ecfbe2 | 322 | } |
c21e6beb | 323 | EXPORT_SYMBOL(blk_run_queue_async); |
24ecfbe2 | 324 | |
75ad23bc NP |
325 | /** |
326 | * blk_run_queue - run a single device queue | |
327 | * @q: The queue to run | |
80a4b58e JA |
328 | * |
329 | * Description: | |
330 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 331 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
332 | */ |
333 | void blk_run_queue(struct request_queue *q) | |
334 | { | |
335 | unsigned long flags; | |
336 | ||
337 | spin_lock_irqsave(q->queue_lock, flags); | |
24ecfbe2 | 338 | __blk_run_queue(q); |
1da177e4 LT |
339 | spin_unlock_irqrestore(q->queue_lock, flags); |
340 | } | |
341 | EXPORT_SYMBOL(blk_run_queue); | |
342 | ||
165125e1 | 343 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
344 | { |
345 | kobject_put(&q->kobj); | |
346 | } | |
d86e0e83 | 347 | EXPORT_SYMBOL(blk_put_queue); |
483f4afc | 348 | |
e3c78ca5 TH |
349 | /** |
350 | * blk_drain_queue - drain requests from request_queue | |
351 | * @q: queue to drain | |
c9a929dd | 352 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
e3c78ca5 | 353 | * |
c9a929dd TH |
354 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
355 | * If not, only ELVPRIV requests are drained. The caller is responsible | |
356 | * for ensuring that no new requests which need to be drained are queued. | |
e3c78ca5 | 357 | */ |
c9a929dd | 358 | void blk_drain_queue(struct request_queue *q, bool drain_all) |
e3c78ca5 TH |
359 | { |
360 | while (true) { | |
361 | int nr_rqs; | |
362 | ||
363 | spin_lock_irq(q->queue_lock); | |
364 | ||
365 | elv_drain_elevator(q); | |
c9a929dd TH |
366 | if (drain_all) |
367 | blk_throtl_drain(q); | |
e3c78ca5 TH |
368 | |
369 | __blk_run_queue(q); | |
c9a929dd TH |
370 | |
371 | if (drain_all) | |
372 | nr_rqs = q->rq.count[0] + q->rq.count[1]; | |
373 | else | |
374 | nr_rqs = q->rq.elvpriv; | |
e3c78ca5 TH |
375 | |
376 | spin_unlock_irq(q->queue_lock); | |
377 | ||
378 | if (!nr_rqs) | |
379 | break; | |
380 | msleep(10); | |
381 | } | |
382 | } | |
383 | ||
c9a929dd TH |
384 | /** |
385 | * blk_cleanup_queue - shutdown a request queue | |
386 | * @q: request queue to shutdown | |
387 | * | |
388 | * Mark @q DEAD, drain all pending requests, destroy and put it. All | |
389 | * future requests will be failed immediately with -ENODEV. | |
c94a96ac | 390 | */ |
6728cb0e | 391 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 392 | { |
c9a929dd | 393 | spinlock_t *lock = q->queue_lock; |
e3335de9 | 394 | |
c9a929dd | 395 | /* mark @q DEAD, no new request or merges will be allowed afterwards */ |
483f4afc | 396 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 397 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
c9a929dd TH |
398 | |
399 | spin_lock_irq(lock); | |
400 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); | |
401 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); | |
402 | queue_flag_set(QUEUE_FLAG_DEAD, q); | |
483f4afc | 403 | |
777eb1bf HR |
404 | if (q->queue_lock != &q->__queue_lock) |
405 | q->queue_lock = &q->__queue_lock; | |
da527770 | 406 | |
c9a929dd TH |
407 | spin_unlock_irq(lock); |
408 | mutex_unlock(&q->sysfs_lock); | |
409 | ||
6dd9ad7d TH |
410 | /* |
411 | * Drain all requests queued before DEAD marking. The caller might | |
412 | * be trying to tear down @q before its elevator is initialized, in | |
413 | * which case we don't want to call into draining. | |
414 | */ | |
415 | if (q->elevator) | |
416 | blk_drain_queue(q, true); | |
c9a929dd TH |
417 | |
418 | /* @q won't process any more request, flush async actions */ | |
419 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); | |
420 | blk_sync_queue(q); | |
421 | ||
422 | /* @q is and will stay empty, shutdown and put */ | |
483f4afc AV |
423 | blk_put_queue(q); |
424 | } | |
1da177e4 LT |
425 | EXPORT_SYMBOL(blk_cleanup_queue); |
426 | ||
165125e1 | 427 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
428 | { |
429 | struct request_list *rl = &q->rq; | |
430 | ||
1abec4fd MS |
431 | if (unlikely(rl->rq_pool)) |
432 | return 0; | |
433 | ||
1faa16d2 JA |
434 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
435 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 436 | rl->elvpriv = 0; |
1faa16d2 JA |
437 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
438 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 439 | |
1946089a CL |
440 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
441 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
442 | |
443 | if (!rl->rq_pool) | |
444 | return -ENOMEM; | |
445 | ||
446 | return 0; | |
447 | } | |
448 | ||
165125e1 | 449 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 450 | { |
1946089a CL |
451 | return blk_alloc_queue_node(gfp_mask, -1); |
452 | } | |
453 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 454 | |
165125e1 | 455 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 456 | { |
165125e1 | 457 | struct request_queue *q; |
e0bf68dd | 458 | int err; |
1946089a | 459 | |
8324aa91 | 460 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 461 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
462 | if (!q) |
463 | return NULL; | |
464 | ||
0989a025 JA |
465 | q->backing_dev_info.ra_pages = |
466 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
467 | q->backing_dev_info.state = 0; | |
468 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 469 | q->backing_dev_info.name = "block"; |
0989a025 | 470 | |
e0bf68dd PZ |
471 | err = bdi_init(&q->backing_dev_info); |
472 | if (err) { | |
8324aa91 | 473 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
474 | return NULL; |
475 | } | |
476 | ||
e43473b7 VG |
477 | if (blk_throtl_init(q)) { |
478 | kmem_cache_free(blk_requestq_cachep, q); | |
479 | return NULL; | |
480 | } | |
481 | ||
31373d09 MG |
482 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
483 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb JA |
484 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
485 | INIT_LIST_HEAD(&q->timeout_list); | |
ae1b1539 TH |
486 | INIT_LIST_HEAD(&q->flush_queue[0]); |
487 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
488 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 489 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 490 | |
8324aa91 | 491 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 492 | |
483f4afc | 493 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 494 | spin_lock_init(&q->__queue_lock); |
483f4afc | 495 | |
c94a96ac VG |
496 | /* |
497 | * By default initialize queue_lock to internal lock and driver can | |
498 | * override it later if need be. | |
499 | */ | |
500 | q->queue_lock = &q->__queue_lock; | |
501 | ||
1da177e4 LT |
502 | return q; |
503 | } | |
1946089a | 504 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
505 | |
506 | /** | |
507 | * blk_init_queue - prepare a request queue for use with a block device | |
508 | * @rfn: The function to be called to process requests that have been | |
509 | * placed on the queue. | |
510 | * @lock: Request queue spin lock | |
511 | * | |
512 | * Description: | |
513 | * If a block device wishes to use the standard request handling procedures, | |
514 | * which sorts requests and coalesces adjacent requests, then it must | |
515 | * call blk_init_queue(). The function @rfn will be called when there | |
516 | * are requests on the queue that need to be processed. If the device | |
517 | * supports plugging, then @rfn may not be called immediately when requests | |
518 | * are available on the queue, but may be called at some time later instead. | |
519 | * Plugged queues are generally unplugged when a buffer belonging to one | |
520 | * of the requests on the queue is needed, or due to memory pressure. | |
521 | * | |
522 | * @rfn is not required, or even expected, to remove all requests off the | |
523 | * queue, but only as many as it can handle at a time. If it does leave | |
524 | * requests on the queue, it is responsible for arranging that the requests | |
525 | * get dealt with eventually. | |
526 | * | |
527 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
528 | * request queue; this lock will be taken also from interrupt context, so irq |
529 | * disabling is needed for it. | |
1da177e4 | 530 | * |
710027a4 | 531 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
532 | * it didn't succeed. |
533 | * | |
534 | * Note: | |
535 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
536 | * when the block device is deactivated (such as at module unload). | |
537 | **/ | |
1946089a | 538 | |
165125e1 | 539 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 540 | { |
1946089a CL |
541 | return blk_init_queue_node(rfn, lock, -1); |
542 | } | |
543 | EXPORT_SYMBOL(blk_init_queue); | |
544 | ||
165125e1 | 545 | struct request_queue * |
1946089a CL |
546 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
547 | { | |
c86d1b8a | 548 | struct request_queue *uninit_q, *q; |
1da177e4 | 549 | |
c86d1b8a MS |
550 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
551 | if (!uninit_q) | |
552 | return NULL; | |
553 | ||
554 | q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id); | |
555 | if (!q) | |
556 | blk_cleanup_queue(uninit_q); | |
557 | ||
558 | return q; | |
01effb0d MS |
559 | } |
560 | EXPORT_SYMBOL(blk_init_queue_node); | |
561 | ||
562 | struct request_queue * | |
563 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
564 | spinlock_t *lock) | |
565 | { | |
566 | return blk_init_allocated_queue_node(q, rfn, lock, -1); | |
567 | } | |
568 | EXPORT_SYMBOL(blk_init_allocated_queue); | |
569 | ||
570 | struct request_queue * | |
571 | blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn, | |
572 | spinlock_t *lock, int node_id) | |
573 | { | |
1da177e4 LT |
574 | if (!q) |
575 | return NULL; | |
576 | ||
1946089a | 577 | q->node = node_id; |
c86d1b8a | 578 | if (blk_init_free_list(q)) |
8669aafd | 579 | return NULL; |
1da177e4 LT |
580 | |
581 | q->request_fn = rfn; | |
1da177e4 | 582 | q->prep_rq_fn = NULL; |
28018c24 | 583 | q->unprep_rq_fn = NULL; |
bc58ba94 | 584 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
585 | |
586 | /* Override internal queue lock with supplied lock pointer */ | |
587 | if (lock) | |
588 | q->queue_lock = lock; | |
1da177e4 | 589 | |
f3b144aa JA |
590 | /* |
591 | * This also sets hw/phys segments, boundary and size | |
592 | */ | |
c20e8de2 | 593 | blk_queue_make_request(q, blk_queue_bio); |
1da177e4 | 594 | |
44ec9542 AS |
595 | q->sg_reserved_size = INT_MAX; |
596 | ||
1da177e4 LT |
597 | /* |
598 | * all done | |
599 | */ | |
600 | if (!elevator_init(q, NULL)) { | |
601 | blk_queue_congestion_threshold(q); | |
602 | return q; | |
603 | } | |
604 | ||
1da177e4 LT |
605 | return NULL; |
606 | } | |
01effb0d | 607 | EXPORT_SYMBOL(blk_init_allocated_queue_node); |
1da177e4 | 608 | |
165125e1 | 609 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 610 | { |
fde6ad22 | 611 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 612 | kobject_get(&q->kobj); |
1da177e4 LT |
613 | return 0; |
614 | } | |
615 | ||
616 | return 1; | |
617 | } | |
d86e0e83 | 618 | EXPORT_SYMBOL(blk_get_queue); |
1da177e4 | 619 | |
165125e1 | 620 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 621 | { |
4aff5e23 | 622 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 623 | elv_put_request(q, rq); |
1da177e4 LT |
624 | mempool_free(rq, q->rq.rq_pool); |
625 | } | |
626 | ||
1ea25ecb | 627 | static struct request * |
75eb6c37 | 628 | blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask) |
1da177e4 LT |
629 | { |
630 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
631 | ||
632 | if (!rq) | |
633 | return NULL; | |
634 | ||
2a4aa30c | 635 | blk_rq_init(q, rq); |
1afb20f3 | 636 | |
42dad764 | 637 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 638 | |
75eb6c37 TH |
639 | if ((flags & REQ_ELVPRIV) && |
640 | unlikely(elv_set_request(q, rq, gfp_mask))) { | |
641 | mempool_free(rq, q->rq.rq_pool); | |
642 | return NULL; | |
cb98fc8b | 643 | } |
1da177e4 | 644 | |
cb98fc8b | 645 | return rq; |
1da177e4 LT |
646 | } |
647 | ||
648 | /* | |
649 | * ioc_batching returns true if the ioc is a valid batching request and | |
650 | * should be given priority access to a request. | |
651 | */ | |
165125e1 | 652 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
653 | { |
654 | if (!ioc) | |
655 | return 0; | |
656 | ||
657 | /* | |
658 | * Make sure the process is able to allocate at least 1 request | |
659 | * even if the batch times out, otherwise we could theoretically | |
660 | * lose wakeups. | |
661 | */ | |
662 | return ioc->nr_batch_requests == q->nr_batching || | |
663 | (ioc->nr_batch_requests > 0 | |
664 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
665 | } | |
666 | ||
667 | /* | |
668 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
669 | * will cause the process to be a "batcher" on all queues in the system. This | |
670 | * is the behaviour we want though - once it gets a wakeup it should be given | |
671 | * a nice run. | |
672 | */ | |
165125e1 | 673 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
674 | { |
675 | if (!ioc || ioc_batching(q, ioc)) | |
676 | return; | |
677 | ||
678 | ioc->nr_batch_requests = q->nr_batching; | |
679 | ioc->last_waited = jiffies; | |
680 | } | |
681 | ||
1faa16d2 | 682 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
683 | { |
684 | struct request_list *rl = &q->rq; | |
685 | ||
1faa16d2 JA |
686 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
687 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 688 | |
1faa16d2 JA |
689 | if (rl->count[sync] + 1 <= q->nr_requests) { |
690 | if (waitqueue_active(&rl->wait[sync])) | |
691 | wake_up(&rl->wait[sync]); | |
1da177e4 | 692 | |
1faa16d2 | 693 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
694 | } |
695 | } | |
696 | ||
697 | /* | |
698 | * A request has just been released. Account for it, update the full and | |
699 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
700 | */ | |
75eb6c37 | 701 | static void freed_request(struct request_queue *q, unsigned int flags) |
1da177e4 LT |
702 | { |
703 | struct request_list *rl = &q->rq; | |
75eb6c37 | 704 | int sync = rw_is_sync(flags); |
1da177e4 | 705 | |
1faa16d2 | 706 | rl->count[sync]--; |
75eb6c37 | 707 | if (flags & REQ_ELVPRIV) |
cb98fc8b | 708 | rl->elvpriv--; |
1da177e4 | 709 | |
1faa16d2 | 710 | __freed_request(q, sync); |
1da177e4 | 711 | |
1faa16d2 JA |
712 | if (unlikely(rl->starved[sync ^ 1])) |
713 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
714 | } |
715 | ||
9d5a4e94 MS |
716 | /* |
717 | * Determine if elevator data should be initialized when allocating the | |
718 | * request associated with @bio. | |
719 | */ | |
720 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
721 | { | |
722 | if (!bio) | |
723 | return true; | |
724 | ||
725 | /* | |
726 | * Flush requests do not use the elevator so skip initialization. | |
727 | * This allows a request to share the flush and elevator data. | |
728 | */ | |
729 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
730 | return false; | |
731 | ||
732 | return true; | |
733 | } | |
734 | ||
da8303c6 TH |
735 | /** |
736 | * get_request - get a free request | |
737 | * @q: request_queue to allocate request from | |
738 | * @rw_flags: RW and SYNC flags | |
739 | * @bio: bio to allocate request for (can be %NULL) | |
740 | * @gfp_mask: allocation mask | |
741 | * | |
742 | * Get a free request from @q. This function may fail under memory | |
743 | * pressure or if @q is dead. | |
744 | * | |
745 | * Must be callled with @q->queue_lock held and, | |
746 | * Returns %NULL on failure, with @q->queue_lock held. | |
747 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 748 | */ |
165125e1 | 749 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 750 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
751 | { |
752 | struct request *rq = NULL; | |
753 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 754 | struct io_context *ioc = NULL; |
1faa16d2 | 755 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
75eb6c37 | 756 | int may_queue; |
88ee5ef1 | 757 | |
da8303c6 TH |
758 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
759 | return NULL; | |
760 | ||
7749a8d4 | 761 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
762 | if (may_queue == ELV_MQUEUE_NO) |
763 | goto rq_starved; | |
764 | ||
1faa16d2 JA |
765 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
766 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 767 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
768 | /* |
769 | * The queue will fill after this allocation, so set | |
770 | * it as full, and mark this process as "batching". | |
771 | * This process will be allowed to complete a batch of | |
772 | * requests, others will be blocked. | |
773 | */ | |
1faa16d2 | 774 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 775 | ioc_set_batching(q, ioc); |
1faa16d2 | 776 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
777 | } else { |
778 | if (may_queue != ELV_MQUEUE_MUST | |
779 | && !ioc_batching(q, ioc)) { | |
780 | /* | |
781 | * The queue is full and the allocating | |
782 | * process is not a "batcher", and not | |
783 | * exempted by the IO scheduler | |
784 | */ | |
785 | goto out; | |
786 | } | |
787 | } | |
1da177e4 | 788 | } |
1faa16d2 | 789 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
790 | } |
791 | ||
082cf69e JA |
792 | /* |
793 | * Only allow batching queuers to allocate up to 50% over the defined | |
794 | * limit of requests, otherwise we could have thousands of requests | |
795 | * allocated with any setting of ->nr_requests | |
796 | */ | |
1faa16d2 | 797 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 798 | goto out; |
fd782a4a | 799 | |
1faa16d2 JA |
800 | rl->count[is_sync]++; |
801 | rl->starved[is_sync] = 0; | |
cb98fc8b | 802 | |
75eb6c37 TH |
803 | if (blk_rq_should_init_elevator(bio) && |
804 | !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) { | |
805 | rw_flags |= REQ_ELVPRIV; | |
806 | rl->elvpriv++; | |
9d5a4e94 | 807 | } |
cb98fc8b | 808 | |
f253b86b JA |
809 | if (blk_queue_io_stat(q)) |
810 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
811 | spin_unlock_irq(q->queue_lock); |
812 | ||
75eb6c37 | 813 | rq = blk_alloc_request(q, rw_flags, gfp_mask); |
88ee5ef1 | 814 | if (unlikely(!rq)) { |
1da177e4 LT |
815 | /* |
816 | * Allocation failed presumably due to memory. Undo anything | |
817 | * we might have messed up. | |
818 | * | |
819 | * Allocating task should really be put onto the front of the | |
820 | * wait queue, but this is pretty rare. | |
821 | */ | |
822 | spin_lock_irq(q->queue_lock); | |
75eb6c37 | 823 | freed_request(q, rw_flags); |
1da177e4 LT |
824 | |
825 | /* | |
826 | * in the very unlikely event that allocation failed and no | |
827 | * requests for this direction was pending, mark us starved | |
828 | * so that freeing of a request in the other direction will | |
829 | * notice us. another possible fix would be to split the | |
830 | * rq mempool into READ and WRITE | |
831 | */ | |
832 | rq_starved: | |
1faa16d2 JA |
833 | if (unlikely(rl->count[is_sync] == 0)) |
834 | rl->starved[is_sync] = 1; | |
1da177e4 | 835 | |
1da177e4 LT |
836 | goto out; |
837 | } | |
838 | ||
88ee5ef1 JA |
839 | /* |
840 | * ioc may be NULL here, and ioc_batching will be false. That's | |
841 | * OK, if the queue is under the request limit then requests need | |
842 | * not count toward the nr_batch_requests limit. There will always | |
843 | * be some limit enforced by BLK_BATCH_TIME. | |
844 | */ | |
1da177e4 LT |
845 | if (ioc_batching(q, ioc)) |
846 | ioc->nr_batch_requests--; | |
6728cb0e | 847 | |
1faa16d2 | 848 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 849 | out: |
1da177e4 LT |
850 | return rq; |
851 | } | |
852 | ||
da8303c6 TH |
853 | /** |
854 | * get_request_wait - get a free request with retry | |
855 | * @q: request_queue to allocate request from | |
856 | * @rw_flags: RW and SYNC flags | |
857 | * @bio: bio to allocate request for (can be %NULL) | |
858 | * | |
859 | * Get a free request from @q. This function keeps retrying under memory | |
860 | * pressure and fails iff @q is dead. | |
d6344532 | 861 | * |
da8303c6 TH |
862 | * Must be callled with @q->queue_lock held and, |
863 | * Returns %NULL on failure, with @q->queue_lock held. | |
864 | * Returns !%NULL on success, with @q->queue_lock *not held*. | |
1da177e4 | 865 | */ |
165125e1 | 866 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 867 | struct bio *bio) |
1da177e4 | 868 | { |
1faa16d2 | 869 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
870 | struct request *rq; |
871 | ||
7749a8d4 | 872 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
873 | while (!rq) { |
874 | DEFINE_WAIT(wait); | |
05caf8db | 875 | struct io_context *ioc; |
1da177e4 LT |
876 | struct request_list *rl = &q->rq; |
877 | ||
da8303c6 TH |
878 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
879 | return NULL; | |
880 | ||
1faa16d2 | 881 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
882 | TASK_UNINTERRUPTIBLE); |
883 | ||
1faa16d2 | 884 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 885 | |
05caf8db ZY |
886 | spin_unlock_irq(q->queue_lock); |
887 | io_schedule(); | |
1da177e4 | 888 | |
05caf8db ZY |
889 | /* |
890 | * After sleeping, we become a "batching" process and | |
891 | * will be able to allocate at least one request, and | |
892 | * up to a big batch of them for a small period time. | |
893 | * See ioc_batching, ioc_set_batching | |
894 | */ | |
895 | ioc = current_io_context(GFP_NOIO, q->node); | |
896 | ioc_set_batching(q, ioc); | |
d6344532 | 897 | |
05caf8db | 898 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 899 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
900 | |
901 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
902 | }; | |
1da177e4 LT |
903 | |
904 | return rq; | |
905 | } | |
906 | ||
165125e1 | 907 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
908 | { |
909 | struct request *rq; | |
910 | ||
911 | BUG_ON(rw != READ && rw != WRITE); | |
912 | ||
d6344532 | 913 | spin_lock_irq(q->queue_lock); |
da8303c6 | 914 | if (gfp_mask & __GFP_WAIT) |
22e2c507 | 915 | rq = get_request_wait(q, rw, NULL); |
da8303c6 | 916 | else |
22e2c507 | 917 | rq = get_request(q, rw, NULL, gfp_mask); |
da8303c6 TH |
918 | if (!rq) |
919 | spin_unlock_irq(q->queue_lock); | |
d6344532 | 920 | /* q->queue_lock is unlocked at this point */ |
1da177e4 LT |
921 | |
922 | return rq; | |
923 | } | |
1da177e4 LT |
924 | EXPORT_SYMBOL(blk_get_request); |
925 | ||
dc72ef4a | 926 | /** |
79eb63e9 | 927 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 928 | * @q: target request queue |
79eb63e9 BH |
929 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
930 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 931 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 932 | * |
79eb63e9 BH |
933 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
934 | * type commands. Where the struct request needs to be farther initialized by | |
935 | * the caller. It is passed a &struct bio, which describes the memory info of | |
936 | * the I/O transfer. | |
dc72ef4a | 937 | * |
79eb63e9 BH |
938 | * The caller of blk_make_request must make sure that bi_io_vec |
939 | * are set to describe the memory buffers. That bio_data_dir() will return | |
940 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
941 | * are properly set accordingly) | |
942 | * | |
943 | * If called under none-sleepable conditions, mapped bio buffers must not | |
944 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
945 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
946 | * BUG. | |
53674ac5 JA |
947 | * |
948 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
949 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
950 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
951 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
952 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
953 | * of bio_alloc(), as that avoids the mempool deadlock. | |
954 | * If possible a big IO should be split into smaller parts when allocation | |
955 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 956 | */ |
79eb63e9 BH |
957 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
958 | gfp_t gfp_mask) | |
dc72ef4a | 959 | { |
79eb63e9 BH |
960 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
961 | ||
962 | if (unlikely(!rq)) | |
963 | return ERR_PTR(-ENOMEM); | |
964 | ||
965 | for_each_bio(bio) { | |
966 | struct bio *bounce_bio = bio; | |
967 | int ret; | |
968 | ||
969 | blk_queue_bounce(q, &bounce_bio); | |
970 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
971 | if (unlikely(ret)) { | |
972 | blk_put_request(rq); | |
973 | return ERR_PTR(ret); | |
974 | } | |
975 | } | |
976 | ||
977 | return rq; | |
dc72ef4a | 978 | } |
79eb63e9 | 979 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 980 | |
1da177e4 LT |
981 | /** |
982 | * blk_requeue_request - put a request back on queue | |
983 | * @q: request queue where request should be inserted | |
984 | * @rq: request to be inserted | |
985 | * | |
986 | * Description: | |
987 | * Drivers often keep queueing requests until the hardware cannot accept | |
988 | * more, when that condition happens we need to put the request back | |
989 | * on the queue. Must be called with queue lock held. | |
990 | */ | |
165125e1 | 991 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 992 | { |
242f9dcb JA |
993 | blk_delete_timer(rq); |
994 | blk_clear_rq_complete(rq); | |
5f3ea37c | 995 | trace_block_rq_requeue(q, rq); |
2056a782 | 996 | |
1da177e4 LT |
997 | if (blk_rq_tagged(rq)) |
998 | blk_queue_end_tag(q, rq); | |
999 | ||
ba396a6c JB |
1000 | BUG_ON(blk_queued_rq(rq)); |
1001 | ||
1da177e4 LT |
1002 | elv_requeue_request(q, rq); |
1003 | } | |
1da177e4 LT |
1004 | EXPORT_SYMBOL(blk_requeue_request); |
1005 | ||
73c10101 JA |
1006 | static void add_acct_request(struct request_queue *q, struct request *rq, |
1007 | int where) | |
1008 | { | |
1009 | drive_stat_acct(rq, 1); | |
7eaceacc | 1010 | __elv_add_request(q, rq, where); |
73c10101 JA |
1011 | } |
1012 | ||
1da177e4 | 1013 | /** |
710027a4 | 1014 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
1015 | * @q: request queue where request should be inserted |
1016 | * @rq: request to be inserted | |
1017 | * @at_head: insert request at head or tail of queue | |
1018 | * @data: private data | |
1da177e4 LT |
1019 | * |
1020 | * Description: | |
1021 | * Many block devices need to execute commands asynchronously, so they don't | |
1022 | * block the whole kernel from preemption during request execution. This is | |
1023 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
1024 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
1025 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
1026 | * |
1027 | * We have the option of inserting the head or the tail of the queue. | |
1028 | * Typically we use the tail for new ioctls and so forth. We use the head | |
1029 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
1030 | * host that is unable to accept a particular command. | |
1031 | */ | |
165125e1 | 1032 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 1033 | int at_head, void *data) |
1da177e4 | 1034 | { |
867d1191 | 1035 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
1036 | unsigned long flags; |
1037 | ||
1038 | /* | |
1039 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
1040 | * must not attempt merges on this) and that it acts as a soft | |
1041 | * barrier | |
1042 | */ | |
4aff5e23 | 1043 | rq->cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
1044 | |
1045 | rq->special = data; | |
1046 | ||
1047 | spin_lock_irqsave(q->queue_lock, flags); | |
1048 | ||
1049 | /* | |
1050 | * If command is tagged, release the tag | |
1051 | */ | |
867d1191 TH |
1052 | if (blk_rq_tagged(rq)) |
1053 | blk_queue_end_tag(q, rq); | |
1da177e4 | 1054 | |
73c10101 | 1055 | add_acct_request(q, rq, where); |
24ecfbe2 | 1056 | __blk_run_queue(q); |
1da177e4 LT |
1057 | spin_unlock_irqrestore(q->queue_lock, flags); |
1058 | } | |
1da177e4 LT |
1059 | EXPORT_SYMBOL(blk_insert_request); |
1060 | ||
074a7aca TH |
1061 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1062 | unsigned long now) | |
1063 | { | |
1064 | if (now == part->stamp) | |
1065 | return; | |
1066 | ||
316d315b | 1067 | if (part_in_flight(part)) { |
074a7aca | 1068 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1069 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1070 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1071 | } | |
1072 | part->stamp = now; | |
1073 | } | |
1074 | ||
1075 | /** | |
496aa8a9 RD |
1076 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1077 | * @cpu: cpu number for stats access | |
1078 | * @part: target partition | |
1da177e4 LT |
1079 | * |
1080 | * The average IO queue length and utilisation statistics are maintained | |
1081 | * by observing the current state of the queue length and the amount of | |
1082 | * time it has been in this state for. | |
1083 | * | |
1084 | * Normally, that accounting is done on IO completion, but that can result | |
1085 | * in more than a second's worth of IO being accounted for within any one | |
1086 | * second, leading to >100% utilisation. To deal with that, we call this | |
1087 | * function to do a round-off before returning the results when reading | |
1088 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1089 | * the current jiffies and restarts the counters again. | |
1090 | */ | |
c9959059 | 1091 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1092 | { |
1093 | unsigned long now = jiffies; | |
1094 | ||
074a7aca TH |
1095 | if (part->partno) |
1096 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1097 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1098 | } |
074a7aca | 1099 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * queue lock must be held | |
1103 | */ | |
165125e1 | 1104 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1105 | { |
1da177e4 LT |
1106 | if (unlikely(!q)) |
1107 | return; | |
1108 | if (unlikely(--req->ref_count)) | |
1109 | return; | |
1110 | ||
8922e16c TH |
1111 | elv_completed_request(q, req); |
1112 | ||
1cd96c24 BH |
1113 | /* this is a bio leak */ |
1114 | WARN_ON(req->bio != NULL); | |
1115 | ||
1da177e4 LT |
1116 | /* |
1117 | * Request may not have originated from ll_rw_blk. if not, | |
1118 | * it didn't come out of our reserved rq pools | |
1119 | */ | |
49171e5c | 1120 | if (req->cmd_flags & REQ_ALLOCED) { |
75eb6c37 | 1121 | unsigned int flags = req->cmd_flags; |
1da177e4 | 1122 | |
1da177e4 | 1123 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1124 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1125 | |
1126 | blk_free_request(q, req); | |
75eb6c37 | 1127 | freed_request(q, flags); |
1da177e4 LT |
1128 | } |
1129 | } | |
6e39b69e MC |
1130 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1131 | ||
1da177e4 LT |
1132 | void blk_put_request(struct request *req) |
1133 | { | |
8922e16c | 1134 | unsigned long flags; |
165125e1 | 1135 | struct request_queue *q = req->q; |
8922e16c | 1136 | |
52a93ba8 FT |
1137 | spin_lock_irqsave(q->queue_lock, flags); |
1138 | __blk_put_request(q, req); | |
1139 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1140 | } |
1da177e4 LT |
1141 | EXPORT_SYMBOL(blk_put_request); |
1142 | ||
66ac0280 CH |
1143 | /** |
1144 | * blk_add_request_payload - add a payload to a request | |
1145 | * @rq: request to update | |
1146 | * @page: page backing the payload | |
1147 | * @len: length of the payload. | |
1148 | * | |
1149 | * This allows to later add a payload to an already submitted request by | |
1150 | * a block driver. The driver needs to take care of freeing the payload | |
1151 | * itself. | |
1152 | * | |
1153 | * Note that this is a quite horrible hack and nothing but handling of | |
1154 | * discard requests should ever use it. | |
1155 | */ | |
1156 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1157 | unsigned int len) | |
1158 | { | |
1159 | struct bio *bio = rq->bio; | |
1160 | ||
1161 | bio->bi_io_vec->bv_page = page; | |
1162 | bio->bi_io_vec->bv_offset = 0; | |
1163 | bio->bi_io_vec->bv_len = len; | |
1164 | ||
1165 | bio->bi_size = len; | |
1166 | bio->bi_vcnt = 1; | |
1167 | bio->bi_phys_segments = 1; | |
1168 | ||
1169 | rq->__data_len = rq->resid_len = len; | |
1170 | rq->nr_phys_segments = 1; | |
1171 | rq->buffer = bio_data(bio); | |
1172 | } | |
1173 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1174 | ||
73c10101 JA |
1175 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1176 | struct bio *bio) | |
1177 | { | |
1178 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1179 | ||
73c10101 JA |
1180 | if (!ll_back_merge_fn(q, req, bio)) |
1181 | return false; | |
1182 | ||
1183 | trace_block_bio_backmerge(q, bio); | |
1184 | ||
1185 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1186 | blk_rq_set_mixed_merge(req); | |
1187 | ||
1188 | req->biotail->bi_next = bio; | |
1189 | req->biotail = bio; | |
1190 | req->__data_len += bio->bi_size; | |
1191 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1192 | ||
1193 | drive_stat_acct(req, 0); | |
95cf3dd9 | 1194 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1195 | return true; |
1196 | } | |
1197 | ||
1198 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1199 | struct request *req, struct bio *bio) | |
1200 | { | |
1201 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
73c10101 | 1202 | |
73c10101 JA |
1203 | if (!ll_front_merge_fn(q, req, bio)) |
1204 | return false; | |
1205 | ||
1206 | trace_block_bio_frontmerge(q, bio); | |
1207 | ||
1208 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1209 | blk_rq_set_mixed_merge(req); | |
1210 | ||
73c10101 JA |
1211 | bio->bi_next = req->bio; |
1212 | req->bio = bio; | |
1213 | ||
1214 | /* | |
1215 | * may not be valid. if the low level driver said | |
1216 | * it didn't need a bounce buffer then it better | |
1217 | * not touch req->buffer either... | |
1218 | */ | |
1219 | req->buffer = bio_data(bio); | |
1220 | req->__sector = bio->bi_sector; | |
1221 | req->__data_len += bio->bi_size; | |
1222 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1223 | ||
1224 | drive_stat_acct(req, 0); | |
95cf3dd9 | 1225 | elv_bio_merged(q, req, bio); |
73c10101 JA |
1226 | return true; |
1227 | } | |
1228 | ||
bd87b589 TH |
1229 | /** |
1230 | * attempt_plug_merge - try to merge with %current's plugged list | |
1231 | * @q: request_queue new bio is being queued at | |
1232 | * @bio: new bio being queued | |
1233 | * @request_count: out parameter for number of traversed plugged requests | |
1234 | * | |
1235 | * Determine whether @bio being queued on @q can be merged with a request | |
1236 | * on %current's plugged list. Returns %true if merge was successful, | |
1237 | * otherwise %false. | |
1238 | * | |
1239 | * This function is called without @q->queue_lock; however, elevator is | |
1240 | * accessed iff there already are requests on the plugged list which in | |
1241 | * turn guarantees validity of the elevator. | |
1242 | * | |
1243 | * Note that, on successful merge, elevator operation | |
1244 | * elevator_bio_merged_fn() will be called without queue lock. Elevator | |
1245 | * must be ready for this. | |
73c10101 | 1246 | */ |
bd87b589 TH |
1247 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1248 | unsigned int *request_count) | |
73c10101 JA |
1249 | { |
1250 | struct blk_plug *plug; | |
1251 | struct request *rq; | |
1252 | bool ret = false; | |
1253 | ||
bd87b589 | 1254 | plug = current->plug; |
73c10101 JA |
1255 | if (!plug) |
1256 | goto out; | |
56ebdaf2 | 1257 | *request_count = 0; |
73c10101 JA |
1258 | |
1259 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1260 | int el_ret; | |
1261 | ||
56ebdaf2 SL |
1262 | (*request_count)++; |
1263 | ||
73c10101 JA |
1264 | if (rq->q != q) |
1265 | continue; | |
1266 | ||
1267 | el_ret = elv_try_merge(rq, bio); | |
1268 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
1269 | ret = bio_attempt_back_merge(q, rq, bio); | |
1270 | if (ret) | |
1271 | break; | |
1272 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1273 | ret = bio_attempt_front_merge(q, rq, bio); | |
1274 | if (ret) | |
1275 | break; | |
1276 | } | |
1277 | } | |
1278 | out: | |
1279 | return ret; | |
1280 | } | |
1281 | ||
86db1e29 | 1282 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1283 | { |
4aff5e23 | 1284 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1285 | |
7b6d91da CH |
1286 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1287 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1288 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1289 | |
52d9e675 | 1290 | req->errors = 0; |
a2dec7b3 | 1291 | req->__sector = bio->bi_sector; |
52d9e675 | 1292 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1293 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1294 | } |
1295 | ||
5a7bbad2 | 1296 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
1da177e4 | 1297 | { |
5e00d1b5 | 1298 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1299 | struct blk_plug *plug; |
1300 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1301 | struct request *req; | |
56ebdaf2 | 1302 | unsigned int request_count = 0; |
1da177e4 | 1303 | |
1da177e4 LT |
1304 | /* |
1305 | * low level driver can indicate that it wants pages above a | |
1306 | * certain limit bounced to low memory (ie for highmem, or even | |
1307 | * ISA dma in theory) | |
1308 | */ | |
1309 | blk_queue_bounce(q, &bio); | |
1310 | ||
4fed947c | 1311 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1312 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1313 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1314 | goto get_rq; |
1315 | } | |
1316 | ||
73c10101 JA |
1317 | /* |
1318 | * Check if we can merge with the plugged list before grabbing | |
1319 | * any locks. | |
1320 | */ | |
bd87b589 | 1321 | if (attempt_plug_merge(q, bio, &request_count)) |
5a7bbad2 | 1322 | return; |
1da177e4 | 1323 | |
73c10101 | 1324 | spin_lock_irq(q->queue_lock); |
2056a782 | 1325 | |
73c10101 JA |
1326 | el_ret = elv_merge(q, &req, bio); |
1327 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
73c10101 JA |
1328 | if (bio_attempt_back_merge(q, req, bio)) { |
1329 | if (!attempt_back_merge(q, req)) | |
1330 | elv_merged_request(q, req, el_ret); | |
1331 | goto out_unlock; | |
1332 | } | |
1333 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
73c10101 JA |
1334 | if (bio_attempt_front_merge(q, req, bio)) { |
1335 | if (!attempt_front_merge(q, req)) | |
1336 | elv_merged_request(q, req, el_ret); | |
1337 | goto out_unlock; | |
80a761fd | 1338 | } |
1da177e4 LT |
1339 | } |
1340 | ||
450991bc | 1341 | get_rq: |
7749a8d4 JA |
1342 | /* |
1343 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1344 | * but we need to set it earlier to expose the sync flag to the | |
1345 | * rq allocator and io schedulers. | |
1346 | */ | |
1347 | rw_flags = bio_data_dir(bio); | |
1348 | if (sync) | |
7b6d91da | 1349 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1350 | |
1da177e4 | 1351 | /* |
450991bc | 1352 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1353 | * Returns with the queue unlocked. |
450991bc | 1354 | */ |
7749a8d4 | 1355 | req = get_request_wait(q, rw_flags, bio); |
da8303c6 TH |
1356 | if (unlikely(!req)) { |
1357 | bio_endio(bio, -ENODEV); /* @q is dead */ | |
1358 | goto out_unlock; | |
1359 | } | |
d6344532 | 1360 | |
450991bc NP |
1361 | /* |
1362 | * After dropping the lock and possibly sleeping here, our request | |
1363 | * may now be mergeable after it had proven unmergeable (above). | |
1364 | * We don't worry about that case for efficiency. It won't happen | |
1365 | * often, and the elevators are able to handle it. | |
1da177e4 | 1366 | */ |
52d9e675 | 1367 | init_request_from_bio(req, bio); |
1da177e4 | 1368 | |
9562ad9a | 1369 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
11ccf116 | 1370 | req->cpu = raw_smp_processor_id(); |
73c10101 JA |
1371 | |
1372 | plug = current->plug; | |
721a9602 | 1373 | if (plug) { |
dc6d36c9 JA |
1374 | /* |
1375 | * If this is the first request added after a plug, fire | |
1376 | * of a plug trace. If others have been added before, check | |
1377 | * if we have multiple devices in this plug. If so, make a | |
1378 | * note to sort the list before dispatch. | |
1379 | */ | |
1380 | if (list_empty(&plug->list)) | |
1381 | trace_block_plug(q); | |
1382 | else if (!plug->should_sort) { | |
73c10101 JA |
1383 | struct request *__rq; |
1384 | ||
1385 | __rq = list_entry_rq(plug->list.prev); | |
1386 | if (__rq->q != q) | |
1387 | plug->should_sort = 1; | |
1388 | } | |
56ebdaf2 | 1389 | if (request_count >= BLK_MAX_REQUEST_COUNT) |
55c022bb | 1390 | blk_flush_plug_list(plug, false); |
73c10101 JA |
1391 | list_add_tail(&req->queuelist, &plug->list); |
1392 | drive_stat_acct(req, 1); | |
1393 | } else { | |
1394 | spin_lock_irq(q->queue_lock); | |
1395 | add_acct_request(q, req, where); | |
24ecfbe2 | 1396 | __blk_run_queue(q); |
73c10101 JA |
1397 | out_unlock: |
1398 | spin_unlock_irq(q->queue_lock); | |
1399 | } | |
1da177e4 | 1400 | } |
c20e8de2 | 1401 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
1da177e4 LT |
1402 | |
1403 | /* | |
1404 | * If bio->bi_dev is a partition, remap the location | |
1405 | */ | |
1406 | static inline void blk_partition_remap(struct bio *bio) | |
1407 | { | |
1408 | struct block_device *bdev = bio->bi_bdev; | |
1409 | ||
bf2de6f5 | 1410 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1411 | struct hd_struct *p = bdev->bd_part; |
1412 | ||
1da177e4 LT |
1413 | bio->bi_sector += p->start_sect; |
1414 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1415 | |
d07335e5 MS |
1416 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1417 | bdev->bd_dev, | |
1418 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1419 | } |
1420 | } | |
1421 | ||
1da177e4 LT |
1422 | static void handle_bad_sector(struct bio *bio) |
1423 | { | |
1424 | char b[BDEVNAME_SIZE]; | |
1425 | ||
1426 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1427 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1428 | bdevname(bio->bi_bdev, b), | |
1429 | bio->bi_rw, | |
1430 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1431 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1432 | |
1433 | set_bit(BIO_EOF, &bio->bi_flags); | |
1434 | } | |
1435 | ||
c17bb495 AM |
1436 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1437 | ||
1438 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1439 | ||
1440 | static int __init setup_fail_make_request(char *str) | |
1441 | { | |
1442 | return setup_fault_attr(&fail_make_request, str); | |
1443 | } | |
1444 | __setup("fail_make_request=", setup_fail_make_request); | |
1445 | ||
b2c9cd37 | 1446 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
c17bb495 | 1447 | { |
b2c9cd37 | 1448 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
c17bb495 AM |
1449 | } |
1450 | ||
1451 | static int __init fail_make_request_debugfs(void) | |
1452 | { | |
dd48c085 AM |
1453 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
1454 | NULL, &fail_make_request); | |
1455 | ||
1456 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; | |
c17bb495 AM |
1457 | } |
1458 | ||
1459 | late_initcall(fail_make_request_debugfs); | |
1460 | ||
1461 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1462 | ||
b2c9cd37 AM |
1463 | static inline bool should_fail_request(struct hd_struct *part, |
1464 | unsigned int bytes) | |
c17bb495 | 1465 | { |
b2c9cd37 | 1466 | return false; |
c17bb495 AM |
1467 | } |
1468 | ||
1469 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1470 | ||
c07e2b41 JA |
1471 | /* |
1472 | * Check whether this bio extends beyond the end of the device. | |
1473 | */ | |
1474 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1475 | { | |
1476 | sector_t maxsector; | |
1477 | ||
1478 | if (!nr_sectors) | |
1479 | return 0; | |
1480 | ||
1481 | /* Test device or partition size, when known. */ | |
77304d2a | 1482 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1483 | if (maxsector) { |
1484 | sector_t sector = bio->bi_sector; | |
1485 | ||
1486 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1487 | /* | |
1488 | * This may well happen - the kernel calls bread() | |
1489 | * without checking the size of the device, e.g., when | |
1490 | * mounting a device. | |
1491 | */ | |
1492 | handle_bad_sector(bio); | |
1493 | return 1; | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | return 0; | |
1498 | } | |
1499 | ||
27a84d54 CH |
1500 | static noinline_for_stack bool |
1501 | generic_make_request_checks(struct bio *bio) | |
1da177e4 | 1502 | { |
165125e1 | 1503 | struct request_queue *q; |
5a7bbad2 | 1504 | int nr_sectors = bio_sectors(bio); |
51fd77bd | 1505 | int err = -EIO; |
5a7bbad2 CH |
1506 | char b[BDEVNAME_SIZE]; |
1507 | struct hd_struct *part; | |
1da177e4 LT |
1508 | |
1509 | might_sleep(); | |
1da177e4 | 1510 | |
c07e2b41 JA |
1511 | if (bio_check_eod(bio, nr_sectors)) |
1512 | goto end_io; | |
1da177e4 | 1513 | |
5a7bbad2 CH |
1514 | q = bdev_get_queue(bio->bi_bdev); |
1515 | if (unlikely(!q)) { | |
1516 | printk(KERN_ERR | |
1517 | "generic_make_request: Trying to access " | |
1518 | "nonexistent block-device %s (%Lu)\n", | |
1519 | bdevname(bio->bi_bdev, b), | |
1520 | (long long) bio->bi_sector); | |
1521 | goto end_io; | |
1522 | } | |
c17bb495 | 1523 | |
5a7bbad2 CH |
1524 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
1525 | nr_sectors > queue_max_hw_sectors(q))) { | |
1526 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1527 | bdevname(bio->bi_bdev, b), | |
1528 | bio_sectors(bio), | |
1529 | queue_max_hw_sectors(q)); | |
1530 | goto end_io; | |
1531 | } | |
1da177e4 | 1532 | |
5a7bbad2 CH |
1533 | part = bio->bi_bdev->bd_part; |
1534 | if (should_fail_request(part, bio->bi_size) || | |
1535 | should_fail_request(&part_to_disk(part)->part0, | |
1536 | bio->bi_size)) | |
1537 | goto end_io; | |
2056a782 | 1538 | |
5a7bbad2 CH |
1539 | /* |
1540 | * If this device has partitions, remap block n | |
1541 | * of partition p to block n+start(p) of the disk. | |
1542 | */ | |
1543 | blk_partition_remap(bio); | |
2056a782 | 1544 | |
5a7bbad2 CH |
1545 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1546 | goto end_io; | |
a7384677 | 1547 | |
5a7bbad2 CH |
1548 | if (bio_check_eod(bio, nr_sectors)) |
1549 | goto end_io; | |
1e87901e | 1550 | |
5a7bbad2 CH |
1551 | /* |
1552 | * Filter flush bio's early so that make_request based | |
1553 | * drivers without flush support don't have to worry | |
1554 | * about them. | |
1555 | */ | |
1556 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1557 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1558 | if (!nr_sectors) { | |
1559 | err = 0; | |
51fd77bd JA |
1560 | goto end_io; |
1561 | } | |
5a7bbad2 | 1562 | } |
5ddfe969 | 1563 | |
5a7bbad2 CH |
1564 | if ((bio->bi_rw & REQ_DISCARD) && |
1565 | (!blk_queue_discard(q) || | |
1566 | ((bio->bi_rw & REQ_SECURE) && | |
1567 | !blk_queue_secdiscard(q)))) { | |
1568 | err = -EOPNOTSUPP; | |
1569 | goto end_io; | |
1570 | } | |
01edede4 | 1571 | |
bc16a4f9 TH |
1572 | if (blk_throtl_bio(q, bio)) |
1573 | return false; /* throttled, will be resubmitted later */ | |
27a84d54 | 1574 | |
5a7bbad2 | 1575 | trace_block_bio_queue(q, bio); |
27a84d54 | 1576 | return true; |
a7384677 TH |
1577 | |
1578 | end_io: | |
1579 | bio_endio(bio, err); | |
27a84d54 | 1580 | return false; |
1da177e4 LT |
1581 | } |
1582 | ||
27a84d54 CH |
1583 | /** |
1584 | * generic_make_request - hand a buffer to its device driver for I/O | |
1585 | * @bio: The bio describing the location in memory and on the device. | |
1586 | * | |
1587 | * generic_make_request() is used to make I/O requests of block | |
1588 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1589 | * to be done. | |
1590 | * | |
1591 | * generic_make_request() does not return any status. The | |
1592 | * success/failure status of the request, along with notification of | |
1593 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1594 | * function described (one day) else where. | |
1595 | * | |
1596 | * The caller of generic_make_request must make sure that bi_io_vec | |
1597 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1598 | * set to describe the device address, and the | |
1599 | * bi_end_io and optionally bi_private are set to describe how | |
1600 | * completion notification should be signaled. | |
1601 | * | |
1602 | * generic_make_request and the drivers it calls may use bi_next if this | |
1603 | * bio happens to be merged with someone else, and may resubmit the bio to | |
1604 | * a lower device by calling into generic_make_request recursively, which | |
1605 | * means the bio should NOT be touched after the call to ->make_request_fn. | |
d89d8796 NB |
1606 | */ |
1607 | void generic_make_request(struct bio *bio) | |
1608 | { | |
bddd87c7 AM |
1609 | struct bio_list bio_list_on_stack; |
1610 | ||
27a84d54 CH |
1611 | if (!generic_make_request_checks(bio)) |
1612 | return; | |
1613 | ||
1614 | /* | |
1615 | * We only want one ->make_request_fn to be active at a time, else | |
1616 | * stack usage with stacked devices could be a problem. So use | |
1617 | * current->bio_list to keep a list of requests submited by a | |
1618 | * make_request_fn function. current->bio_list is also used as a | |
1619 | * flag to say if generic_make_request is currently active in this | |
1620 | * task or not. If it is NULL, then no make_request is active. If | |
1621 | * it is non-NULL, then a make_request is active, and new requests | |
1622 | * should be added at the tail | |
1623 | */ | |
bddd87c7 | 1624 | if (current->bio_list) { |
bddd87c7 | 1625 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1626 | return; |
1627 | } | |
27a84d54 | 1628 | |
d89d8796 NB |
1629 | /* following loop may be a bit non-obvious, and so deserves some |
1630 | * explanation. | |
1631 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1632 | * ensure that) so we have a list with a single bio. | |
1633 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1634 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1635 | * thus initialising the bio_list of new bios to be | |
27a84d54 | 1636 | * added. ->make_request() may indeed add some more bios |
d89d8796 NB |
1637 | * through a recursive call to generic_make_request. If it |
1638 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1639 | * from the top. In this case we really did just take the bio | |
bddd87c7 | 1640 | * of the top of the list (no pretending) and so remove it from |
27a84d54 | 1641 | * bio_list, and call into ->make_request() again. |
d89d8796 NB |
1642 | */ |
1643 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1644 | bio_list_init(&bio_list_on_stack); |
1645 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1646 | do { |
27a84d54 CH |
1647 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
1648 | ||
1649 | q->make_request_fn(q, bio); | |
1650 | ||
bddd87c7 | 1651 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1652 | } while (bio); |
bddd87c7 | 1653 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1654 | } |
1da177e4 LT |
1655 | EXPORT_SYMBOL(generic_make_request); |
1656 | ||
1657 | /** | |
710027a4 | 1658 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1659 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1660 | * @bio: The &struct bio which describes the I/O | |
1661 | * | |
1662 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1663 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1664 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1665 | * |
1666 | */ | |
1667 | void submit_bio(int rw, struct bio *bio) | |
1668 | { | |
1669 | int count = bio_sectors(bio); | |
1670 | ||
22e2c507 | 1671 | bio->bi_rw |= rw; |
1da177e4 | 1672 | |
bf2de6f5 JA |
1673 | /* |
1674 | * If it's a regular read/write or a barrier with data attached, | |
1675 | * go through the normal accounting stuff before submission. | |
1676 | */ | |
3ffb52e7 | 1677 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1678 | if (rw & WRITE) { |
1679 | count_vm_events(PGPGOUT, count); | |
1680 | } else { | |
1681 | task_io_account_read(bio->bi_size); | |
1682 | count_vm_events(PGPGIN, count); | |
1683 | } | |
1684 | ||
1685 | if (unlikely(block_dump)) { | |
1686 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1687 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1688 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1689 | (rw & WRITE) ? "WRITE" : "READ", |
1690 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1691 | bdevname(bio->bi_bdev, b), |
1692 | count); | |
bf2de6f5 | 1693 | } |
1da177e4 LT |
1694 | } |
1695 | ||
1696 | generic_make_request(bio); | |
1697 | } | |
1da177e4 LT |
1698 | EXPORT_SYMBOL(submit_bio); |
1699 | ||
82124d60 KU |
1700 | /** |
1701 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1702 | * @q: the queue | |
1703 | * @rq: the request being checked | |
1704 | * | |
1705 | * Description: | |
1706 | * @rq may have been made based on weaker limitations of upper-level queues | |
1707 | * in request stacking drivers, and it may violate the limitation of @q. | |
1708 | * Since the block layer and the underlying device driver trust @rq | |
1709 | * after it is inserted to @q, it should be checked against @q before | |
1710 | * the insertion using this generic function. | |
1711 | * | |
1712 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1713 | * in some cases below, so export this function. |
82124d60 KU |
1714 | * Request stacking drivers like request-based dm may change the queue |
1715 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1716 | * Such request stacking drivers should check those requests agaist | |
1717 | * the new queue limits again when they dispatch those requests, | |
1718 | * although such checkings are also done against the old queue limits | |
1719 | * when submitting requests. | |
1720 | */ | |
1721 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1722 | { | |
3383977f S |
1723 | if (rq->cmd_flags & REQ_DISCARD) |
1724 | return 0; | |
1725 | ||
ae03bf63 MP |
1726 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1727 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1728 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1729 | return -EIO; | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * queue's settings related to segment counting like q->bounce_pfn | |
1734 | * may differ from that of other stacking queues. | |
1735 | * Recalculate it to check the request correctly on this queue's | |
1736 | * limitation. | |
1737 | */ | |
1738 | blk_recalc_rq_segments(rq); | |
8a78362c | 1739 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1740 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1741 | return -EIO; | |
1742 | } | |
1743 | ||
1744 | return 0; | |
1745 | } | |
1746 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1747 | ||
1748 | /** | |
1749 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1750 | * @q: the queue to submit the request | |
1751 | * @rq: the request being queued | |
1752 | */ | |
1753 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1754 | { | |
1755 | unsigned long flags; | |
4853abaa | 1756 | int where = ELEVATOR_INSERT_BACK; |
82124d60 KU |
1757 | |
1758 | if (blk_rq_check_limits(q, rq)) | |
1759 | return -EIO; | |
1760 | ||
b2c9cd37 AM |
1761 | if (rq->rq_disk && |
1762 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) | |
82124d60 | 1763 | return -EIO; |
82124d60 KU |
1764 | |
1765 | spin_lock_irqsave(q->queue_lock, flags); | |
1766 | ||
1767 | /* | |
1768 | * Submitting request must be dequeued before calling this function | |
1769 | * because it will be linked to another request_queue | |
1770 | */ | |
1771 | BUG_ON(blk_queued_rq(rq)); | |
1772 | ||
4853abaa JM |
1773 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
1774 | where = ELEVATOR_INSERT_FLUSH; | |
1775 | ||
1776 | add_acct_request(q, rq, where); | |
e67b77c7 JM |
1777 | if (where == ELEVATOR_INSERT_FLUSH) |
1778 | __blk_run_queue(q); | |
82124d60 KU |
1779 | spin_unlock_irqrestore(q->queue_lock, flags); |
1780 | ||
1781 | return 0; | |
1782 | } | |
1783 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1784 | ||
80a761fd TH |
1785 | /** |
1786 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1787 | * @rq: request to examine | |
1788 | * | |
1789 | * Description: | |
1790 | * A request could be merge of IOs which require different failure | |
1791 | * handling. This function determines the number of bytes which | |
1792 | * can be failed from the beginning of the request without | |
1793 | * crossing into area which need to be retried further. | |
1794 | * | |
1795 | * Return: | |
1796 | * The number of bytes to fail. | |
1797 | * | |
1798 | * Context: | |
1799 | * queue_lock must be held. | |
1800 | */ | |
1801 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1802 | { | |
1803 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1804 | unsigned int bytes = 0; | |
1805 | struct bio *bio; | |
1806 | ||
1807 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1808 | return blk_rq_bytes(rq); | |
1809 | ||
1810 | /* | |
1811 | * Currently the only 'mixing' which can happen is between | |
1812 | * different fastfail types. We can safely fail portions | |
1813 | * which have all the failfast bits that the first one has - | |
1814 | * the ones which are at least as eager to fail as the first | |
1815 | * one. | |
1816 | */ | |
1817 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1818 | if ((bio->bi_rw & ff) != ff) | |
1819 | break; | |
1820 | bytes += bio->bi_size; | |
1821 | } | |
1822 | ||
1823 | /* this could lead to infinite loop */ | |
1824 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1825 | return bytes; | |
1826 | } | |
1827 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1828 | ||
bc58ba94 JA |
1829 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1830 | { | |
c2553b58 | 1831 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1832 | const int rw = rq_data_dir(req); |
1833 | struct hd_struct *part; | |
1834 | int cpu; | |
1835 | ||
1836 | cpu = part_stat_lock(); | |
09e099d4 | 1837 | part = req->part; |
bc58ba94 JA |
1838 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1839 | part_stat_unlock(); | |
1840 | } | |
1841 | } | |
1842 | ||
1843 | static void blk_account_io_done(struct request *req) | |
1844 | { | |
bc58ba94 | 1845 | /* |
dd4c133f TH |
1846 | * Account IO completion. flush_rq isn't accounted as a |
1847 | * normal IO on queueing nor completion. Accounting the | |
1848 | * containing request is enough. | |
bc58ba94 | 1849 | */ |
414b4ff5 | 1850 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1851 | unsigned long duration = jiffies - req->start_time; |
1852 | const int rw = rq_data_dir(req); | |
1853 | struct hd_struct *part; | |
1854 | int cpu; | |
1855 | ||
1856 | cpu = part_stat_lock(); | |
09e099d4 | 1857 | part = req->part; |
bc58ba94 JA |
1858 | |
1859 | part_stat_inc(cpu, part, ios[rw]); | |
1860 | part_stat_add(cpu, part, ticks[rw], duration); | |
1861 | part_round_stats(cpu, part); | |
316d315b | 1862 | part_dec_in_flight(part, rw); |
bc58ba94 | 1863 | |
6c23a968 | 1864 | hd_struct_put(part); |
bc58ba94 JA |
1865 | part_stat_unlock(); |
1866 | } | |
1867 | } | |
1868 | ||
3bcddeac | 1869 | /** |
9934c8c0 TH |
1870 | * blk_peek_request - peek at the top of a request queue |
1871 | * @q: request queue to peek at | |
1872 | * | |
1873 | * Description: | |
1874 | * Return the request at the top of @q. The returned request | |
1875 | * should be started using blk_start_request() before LLD starts | |
1876 | * processing it. | |
1877 | * | |
1878 | * Return: | |
1879 | * Pointer to the request at the top of @q if available. Null | |
1880 | * otherwise. | |
1881 | * | |
1882 | * Context: | |
1883 | * queue_lock must be held. | |
1884 | */ | |
1885 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
1886 | { |
1887 | struct request *rq; | |
1888 | int ret; | |
1889 | ||
1890 | while ((rq = __elv_next_request(q)) != NULL) { | |
1891 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1892 | /* | |
1893 | * This is the first time the device driver | |
1894 | * sees this request (possibly after | |
1895 | * requeueing). Notify IO scheduler. | |
1896 | */ | |
33659ebb | 1897 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
1898 | elv_activate_rq(q, rq); |
1899 | ||
1900 | /* | |
1901 | * just mark as started even if we don't start | |
1902 | * it, a request that has been delayed should | |
1903 | * not be passed by new incoming requests | |
1904 | */ | |
1905 | rq->cmd_flags |= REQ_STARTED; | |
1906 | trace_block_rq_issue(q, rq); | |
1907 | } | |
1908 | ||
1909 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1910 | q->end_sector = rq_end_sector(rq); | |
1911 | q->boundary_rq = NULL; | |
1912 | } | |
1913 | ||
1914 | if (rq->cmd_flags & REQ_DONTPREP) | |
1915 | break; | |
1916 | ||
2e46e8b2 | 1917 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
1918 | /* |
1919 | * make sure space for the drain appears we | |
1920 | * know we can do this because max_hw_segments | |
1921 | * has been adjusted to be one fewer than the | |
1922 | * device can handle | |
1923 | */ | |
1924 | rq->nr_phys_segments++; | |
1925 | } | |
1926 | ||
1927 | if (!q->prep_rq_fn) | |
1928 | break; | |
1929 | ||
1930 | ret = q->prep_rq_fn(q, rq); | |
1931 | if (ret == BLKPREP_OK) { | |
1932 | break; | |
1933 | } else if (ret == BLKPREP_DEFER) { | |
1934 | /* | |
1935 | * the request may have been (partially) prepped. | |
1936 | * we need to keep this request in the front to | |
1937 | * avoid resource deadlock. REQ_STARTED will | |
1938 | * prevent other fs requests from passing this one. | |
1939 | */ | |
2e46e8b2 | 1940 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
1941 | !(rq->cmd_flags & REQ_DONTPREP)) { |
1942 | /* | |
1943 | * remove the space for the drain we added | |
1944 | * so that we don't add it again | |
1945 | */ | |
1946 | --rq->nr_phys_segments; | |
1947 | } | |
1948 | ||
1949 | rq = NULL; | |
1950 | break; | |
1951 | } else if (ret == BLKPREP_KILL) { | |
1952 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
1953 | /* |
1954 | * Mark this request as started so we don't trigger | |
1955 | * any debug logic in the end I/O path. | |
1956 | */ | |
1957 | blk_start_request(rq); | |
40cbbb78 | 1958 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
1959 | } else { |
1960 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1961 | break; | |
1962 | } | |
1963 | } | |
1964 | ||
1965 | return rq; | |
1966 | } | |
9934c8c0 | 1967 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 1968 | |
9934c8c0 | 1969 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 1970 | { |
9934c8c0 TH |
1971 | struct request_queue *q = rq->q; |
1972 | ||
158dbda0 TH |
1973 | BUG_ON(list_empty(&rq->queuelist)); |
1974 | BUG_ON(ELV_ON_HASH(rq)); | |
1975 | ||
1976 | list_del_init(&rq->queuelist); | |
1977 | ||
1978 | /* | |
1979 | * the time frame between a request being removed from the lists | |
1980 | * and to it is freed is accounted as io that is in progress at | |
1981 | * the driver side. | |
1982 | */ | |
9195291e | 1983 | if (blk_account_rq(rq)) { |
0a7ae2ff | 1984 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
1985 | set_io_start_time_ns(rq); |
1986 | } | |
158dbda0 TH |
1987 | } |
1988 | ||
9934c8c0 TH |
1989 | /** |
1990 | * blk_start_request - start request processing on the driver | |
1991 | * @req: request to dequeue | |
1992 | * | |
1993 | * Description: | |
1994 | * Dequeue @req and start timeout timer on it. This hands off the | |
1995 | * request to the driver. | |
1996 | * | |
1997 | * Block internal functions which don't want to start timer should | |
1998 | * call blk_dequeue_request(). | |
1999 | * | |
2000 | * Context: | |
2001 | * queue_lock must be held. | |
2002 | */ | |
2003 | void blk_start_request(struct request *req) | |
2004 | { | |
2005 | blk_dequeue_request(req); | |
2006 | ||
2007 | /* | |
5f49f631 TH |
2008 | * We are now handing the request to the hardware, initialize |
2009 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2010 | */ |
5f49f631 | 2011 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2012 | if (unlikely(blk_bidi_rq(req))) |
2013 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2014 | ||
9934c8c0 TH |
2015 | blk_add_timer(req); |
2016 | } | |
2017 | EXPORT_SYMBOL(blk_start_request); | |
2018 | ||
2019 | /** | |
2020 | * blk_fetch_request - fetch a request from a request queue | |
2021 | * @q: request queue to fetch a request from | |
2022 | * | |
2023 | * Description: | |
2024 | * Return the request at the top of @q. The request is started on | |
2025 | * return and LLD can start processing it immediately. | |
2026 | * | |
2027 | * Return: | |
2028 | * Pointer to the request at the top of @q if available. Null | |
2029 | * otherwise. | |
2030 | * | |
2031 | * Context: | |
2032 | * queue_lock must be held. | |
2033 | */ | |
2034 | struct request *blk_fetch_request(struct request_queue *q) | |
2035 | { | |
2036 | struct request *rq; | |
2037 | ||
2038 | rq = blk_peek_request(q); | |
2039 | if (rq) | |
2040 | blk_start_request(rq); | |
2041 | return rq; | |
2042 | } | |
2043 | EXPORT_SYMBOL(blk_fetch_request); | |
2044 | ||
3bcddeac | 2045 | /** |
2e60e022 | 2046 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2047 | * @req: the request being processed |
710027a4 | 2048 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2049 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2050 | * |
2051 | * Description: | |
8ebf9756 RD |
2052 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2053 | * the request structure even if @req doesn't have leftover. | |
2054 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2055 | * |
2056 | * This special helper function is only for request stacking drivers | |
2057 | * (e.g. request-based dm) so that they can handle partial completion. | |
2058 | * Actual device drivers should use blk_end_request instead. | |
2059 | * | |
2060 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2061 | * %false return from this function. | |
3bcddeac KU |
2062 | * |
2063 | * Return: | |
2e60e022 TH |
2064 | * %false - this request doesn't have any more data |
2065 | * %true - this request has more data | |
3bcddeac | 2066 | **/ |
2e60e022 | 2067 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2068 | { |
5450d3e1 | 2069 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2070 | struct bio *bio; |
2071 | ||
2e60e022 TH |
2072 | if (!req->bio) |
2073 | return false; | |
2074 | ||
5f3ea37c | 2075 | trace_block_rq_complete(req->q, req); |
2056a782 | 2076 | |
1da177e4 | 2077 | /* |
6f41469c TH |
2078 | * For fs requests, rq is just carrier of independent bio's |
2079 | * and each partial completion should be handled separately. | |
2080 | * Reset per-request error on each partial completion. | |
2081 | * | |
2082 | * TODO: tj: This is too subtle. It would be better to let | |
2083 | * low level drivers do what they see fit. | |
1da177e4 | 2084 | */ |
33659ebb | 2085 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2086 | req->errors = 0; |
2087 | ||
33659ebb CH |
2088 | if (error && req->cmd_type == REQ_TYPE_FS && |
2089 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2090 | char *error_type; |
2091 | ||
2092 | switch (error) { | |
2093 | case -ENOLINK: | |
2094 | error_type = "recoverable transport"; | |
2095 | break; | |
2096 | case -EREMOTEIO: | |
2097 | error_type = "critical target"; | |
2098 | break; | |
2099 | case -EBADE: | |
2100 | error_type = "critical nexus"; | |
2101 | break; | |
2102 | case -EIO: | |
2103 | default: | |
2104 | error_type = "I/O"; | |
2105 | break; | |
2106 | } | |
2107 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2108 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2109 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2110 | } |
2111 | ||
bc58ba94 | 2112 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2113 | |
1da177e4 LT |
2114 | total_bytes = bio_nbytes = 0; |
2115 | while ((bio = req->bio) != NULL) { | |
2116 | int nbytes; | |
2117 | ||
2118 | if (nr_bytes >= bio->bi_size) { | |
2119 | req->bio = bio->bi_next; | |
2120 | nbytes = bio->bi_size; | |
5bb23a68 | 2121 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2122 | next_idx = 0; |
2123 | bio_nbytes = 0; | |
2124 | } else { | |
2125 | int idx = bio->bi_idx + next_idx; | |
2126 | ||
af498d7f | 2127 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2128 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2129 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2130 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2131 | break; |
2132 | } | |
2133 | ||
2134 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2135 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2136 | ||
2137 | /* | |
2138 | * not a complete bvec done | |
2139 | */ | |
2140 | if (unlikely(nbytes > nr_bytes)) { | |
2141 | bio_nbytes += nr_bytes; | |
2142 | total_bytes += nr_bytes; | |
2143 | break; | |
2144 | } | |
2145 | ||
2146 | /* | |
2147 | * advance to the next vector | |
2148 | */ | |
2149 | next_idx++; | |
2150 | bio_nbytes += nbytes; | |
2151 | } | |
2152 | ||
2153 | total_bytes += nbytes; | |
2154 | nr_bytes -= nbytes; | |
2155 | ||
6728cb0e JA |
2156 | bio = req->bio; |
2157 | if (bio) { | |
1da177e4 LT |
2158 | /* |
2159 | * end more in this run, or just return 'not-done' | |
2160 | */ | |
2161 | if (unlikely(nr_bytes <= 0)) | |
2162 | break; | |
2163 | } | |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * completely done | |
2168 | */ | |
2e60e022 TH |
2169 | if (!req->bio) { |
2170 | /* | |
2171 | * Reset counters so that the request stacking driver | |
2172 | * can find how many bytes remain in the request | |
2173 | * later. | |
2174 | */ | |
a2dec7b3 | 2175 | req->__data_len = 0; |
2e60e022 TH |
2176 | return false; |
2177 | } | |
1da177e4 LT |
2178 | |
2179 | /* | |
2180 | * if the request wasn't completed, update state | |
2181 | */ | |
2182 | if (bio_nbytes) { | |
5bb23a68 | 2183 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2184 | bio->bi_idx += next_idx; |
2185 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2186 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2187 | } | |
2188 | ||
a2dec7b3 | 2189 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2190 | req->buffer = bio_data(req->bio); |
2191 | ||
2192 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2193 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2194 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2195 | |
80a761fd TH |
2196 | /* mixed attributes always follow the first bio */ |
2197 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2198 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2199 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2200 | } | |
2201 | ||
2e46e8b2 TH |
2202 | /* |
2203 | * If total number of sectors is less than the first segment | |
2204 | * size, something has gone terribly wrong. | |
2205 | */ | |
2206 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2207 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2208 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2209 | } |
2210 | ||
2211 | /* recalculate the number of segments */ | |
1da177e4 | 2212 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2213 | |
2e60e022 | 2214 | return true; |
1da177e4 | 2215 | } |
2e60e022 | 2216 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2217 | |
2e60e022 TH |
2218 | static bool blk_update_bidi_request(struct request *rq, int error, |
2219 | unsigned int nr_bytes, | |
2220 | unsigned int bidi_bytes) | |
5efccd17 | 2221 | { |
2e60e022 TH |
2222 | if (blk_update_request(rq, error, nr_bytes)) |
2223 | return true; | |
5efccd17 | 2224 | |
2e60e022 TH |
2225 | /* Bidi request must be completed as a whole */ |
2226 | if (unlikely(blk_bidi_rq(rq)) && | |
2227 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2228 | return true; | |
5efccd17 | 2229 | |
e2e1a148 JA |
2230 | if (blk_queue_add_random(rq->q)) |
2231 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2232 | |
2233 | return false; | |
1da177e4 LT |
2234 | } |
2235 | ||
28018c24 JB |
2236 | /** |
2237 | * blk_unprep_request - unprepare a request | |
2238 | * @req: the request | |
2239 | * | |
2240 | * This function makes a request ready for complete resubmission (or | |
2241 | * completion). It happens only after all error handling is complete, | |
2242 | * so represents the appropriate moment to deallocate any resources | |
2243 | * that were allocated to the request in the prep_rq_fn. The queue | |
2244 | * lock is held when calling this. | |
2245 | */ | |
2246 | void blk_unprep_request(struct request *req) | |
2247 | { | |
2248 | struct request_queue *q = req->q; | |
2249 | ||
2250 | req->cmd_flags &= ~REQ_DONTPREP; | |
2251 | if (q->unprep_rq_fn) | |
2252 | q->unprep_rq_fn(q, req); | |
2253 | } | |
2254 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2255 | ||
1da177e4 LT |
2256 | /* |
2257 | * queue lock must be held | |
2258 | */ | |
2e60e022 | 2259 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2260 | { |
b8286239 KU |
2261 | if (blk_rq_tagged(req)) |
2262 | blk_queue_end_tag(req->q, req); | |
2263 | ||
ba396a6c | 2264 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2265 | |
33659ebb | 2266 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2267 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2268 | |
e78042e5 MA |
2269 | blk_delete_timer(req); |
2270 | ||
28018c24 JB |
2271 | if (req->cmd_flags & REQ_DONTPREP) |
2272 | blk_unprep_request(req); | |
2273 | ||
2274 | ||
bc58ba94 | 2275 | blk_account_io_done(req); |
b8286239 | 2276 | |
1da177e4 | 2277 | if (req->end_io) |
8ffdc655 | 2278 | req->end_io(req, error); |
b8286239 KU |
2279 | else { |
2280 | if (blk_bidi_rq(req)) | |
2281 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2282 | ||
1da177e4 | 2283 | __blk_put_request(req->q, req); |
b8286239 | 2284 | } |
1da177e4 LT |
2285 | } |
2286 | ||
3b11313a | 2287 | /** |
2e60e022 TH |
2288 | * blk_end_bidi_request - Complete a bidi request |
2289 | * @rq: the request to complete | |
2290 | * @error: %0 for success, < %0 for error | |
2291 | * @nr_bytes: number of bytes to complete @rq | |
2292 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2293 | * |
2294 | * Description: | |
e3a04fe3 | 2295 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2296 | * Drivers that supports bidi can safely call this member for any |
2297 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2298 | * just ignored. | |
336cdb40 KU |
2299 | * |
2300 | * Return: | |
2e60e022 TH |
2301 | * %false - we are done with this request |
2302 | * %true - still buffers pending for this request | |
a0cd1285 | 2303 | **/ |
b1f74493 | 2304 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2305 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2306 | { | |
336cdb40 | 2307 | struct request_queue *q = rq->q; |
2e60e022 | 2308 | unsigned long flags; |
32fab448 | 2309 | |
2e60e022 TH |
2310 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2311 | return true; | |
32fab448 | 2312 | |
336cdb40 | 2313 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2314 | blk_finish_request(rq, error); |
336cdb40 KU |
2315 | spin_unlock_irqrestore(q->queue_lock, flags); |
2316 | ||
2e60e022 | 2317 | return false; |
32fab448 KU |
2318 | } |
2319 | ||
336cdb40 | 2320 | /** |
2e60e022 TH |
2321 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2322 | * @rq: the request to complete | |
710027a4 | 2323 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2324 | * @nr_bytes: number of bytes to complete @rq |
2325 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2326 | * |
2327 | * Description: | |
2e60e022 TH |
2328 | * Identical to blk_end_bidi_request() except that queue lock is |
2329 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2330 | * |
2331 | * Return: | |
2e60e022 TH |
2332 | * %false - we are done with this request |
2333 | * %true - still buffers pending for this request | |
336cdb40 | 2334 | **/ |
4853abaa | 2335 | bool __blk_end_bidi_request(struct request *rq, int error, |
b1f74493 | 2336 | unsigned int nr_bytes, unsigned int bidi_bytes) |
336cdb40 | 2337 | { |
2e60e022 TH |
2338 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2339 | return true; | |
336cdb40 | 2340 | |
2e60e022 | 2341 | blk_finish_request(rq, error); |
336cdb40 | 2342 | |
2e60e022 | 2343 | return false; |
336cdb40 | 2344 | } |
e19a3ab0 KU |
2345 | |
2346 | /** | |
2347 | * blk_end_request - Helper function for drivers to complete the request. | |
2348 | * @rq: the request being processed | |
710027a4 | 2349 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2350 | * @nr_bytes: number of bytes to complete |
2351 | * | |
2352 | * Description: | |
2353 | * Ends I/O on a number of bytes attached to @rq. | |
2354 | * If @rq has leftover, sets it up for the next range of segments. | |
2355 | * | |
2356 | * Return: | |
b1f74493 FT |
2357 | * %false - we are done with this request |
2358 | * %true - still buffers pending for this request | |
e19a3ab0 | 2359 | **/ |
b1f74493 | 2360 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2361 | { |
b1f74493 | 2362 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2363 | } |
56ad1740 | 2364 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2365 | |
2366 | /** | |
b1f74493 FT |
2367 | * blk_end_request_all - Helper function for drives to finish the request. |
2368 | * @rq: the request to finish | |
8ebf9756 | 2369 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2370 | * |
2371 | * Description: | |
b1f74493 FT |
2372 | * Completely finish @rq. |
2373 | */ | |
2374 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2375 | { |
b1f74493 FT |
2376 | bool pending; |
2377 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2378 | |
b1f74493 FT |
2379 | if (unlikely(blk_bidi_rq(rq))) |
2380 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2381 | |
b1f74493 FT |
2382 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2383 | BUG_ON(pending); | |
2384 | } | |
56ad1740 | 2385 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2386 | |
b1f74493 FT |
2387 | /** |
2388 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2389 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2390 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2391 | * |
2392 | * Description: | |
2393 | * Complete the current consecutively mapped chunk from @rq. | |
2394 | * | |
2395 | * Return: | |
2396 | * %false - we are done with this request | |
2397 | * %true - still buffers pending for this request | |
2398 | */ | |
2399 | bool blk_end_request_cur(struct request *rq, int error) | |
2400 | { | |
2401 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2402 | } |
56ad1740 | 2403 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2404 | |
80a761fd TH |
2405 | /** |
2406 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2407 | * @rq: the request to finish till the next failure boundary for | |
2408 | * @error: must be negative errno | |
2409 | * | |
2410 | * Description: | |
2411 | * Complete @rq till the next failure boundary. | |
2412 | * | |
2413 | * Return: | |
2414 | * %false - we are done with this request | |
2415 | * %true - still buffers pending for this request | |
2416 | */ | |
2417 | bool blk_end_request_err(struct request *rq, int error) | |
2418 | { | |
2419 | WARN_ON(error >= 0); | |
2420 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2421 | } | |
2422 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2423 | ||
e3a04fe3 | 2424 | /** |
b1f74493 FT |
2425 | * __blk_end_request - Helper function for drivers to complete the request. |
2426 | * @rq: the request being processed | |
2427 | * @error: %0 for success, < %0 for error | |
2428 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2429 | * |
2430 | * Description: | |
b1f74493 | 2431 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2432 | * |
2433 | * Return: | |
b1f74493 FT |
2434 | * %false - we are done with this request |
2435 | * %true - still buffers pending for this request | |
e3a04fe3 | 2436 | **/ |
b1f74493 | 2437 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2438 | { |
b1f74493 | 2439 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2440 | } |
56ad1740 | 2441 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2442 | |
32fab448 | 2443 | /** |
b1f74493 FT |
2444 | * __blk_end_request_all - Helper function for drives to finish the request. |
2445 | * @rq: the request to finish | |
8ebf9756 | 2446 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2447 | * |
2448 | * Description: | |
b1f74493 | 2449 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2450 | */ |
b1f74493 | 2451 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2452 | { |
b1f74493 FT |
2453 | bool pending; |
2454 | unsigned int bidi_bytes = 0; | |
2455 | ||
2456 | if (unlikely(blk_bidi_rq(rq))) | |
2457 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2458 | ||
2459 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2460 | BUG_ON(pending); | |
32fab448 | 2461 | } |
56ad1740 | 2462 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2463 | |
e19a3ab0 | 2464 | /** |
b1f74493 FT |
2465 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2466 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2467 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2468 | * |
2469 | * Description: | |
b1f74493 FT |
2470 | * Complete the current consecutively mapped chunk from @rq. Must |
2471 | * be called with queue lock held. | |
e19a3ab0 KU |
2472 | * |
2473 | * Return: | |
b1f74493 FT |
2474 | * %false - we are done with this request |
2475 | * %true - still buffers pending for this request | |
2476 | */ | |
2477 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2478 | { |
b1f74493 | 2479 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2480 | } |
56ad1740 | 2481 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2482 | |
80a761fd TH |
2483 | /** |
2484 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2485 | * @rq: the request to finish till the next failure boundary for | |
2486 | * @error: must be negative errno | |
2487 | * | |
2488 | * Description: | |
2489 | * Complete @rq till the next failure boundary. Must be called | |
2490 | * with queue lock held. | |
2491 | * | |
2492 | * Return: | |
2493 | * %false - we are done with this request | |
2494 | * %true - still buffers pending for this request | |
2495 | */ | |
2496 | bool __blk_end_request_err(struct request *rq, int error) | |
2497 | { | |
2498 | WARN_ON(error >= 0); | |
2499 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2500 | } | |
2501 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2502 | ||
86db1e29 JA |
2503 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2504 | struct bio *bio) | |
1da177e4 | 2505 | { |
a82afdfc | 2506 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2507 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2508 | |
fb2dce86 DW |
2509 | if (bio_has_data(bio)) { |
2510 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2511 | rq->buffer = bio_data(bio); |
2512 | } | |
a2dec7b3 | 2513 | rq->__data_len = bio->bi_size; |
1da177e4 | 2514 | rq->bio = rq->biotail = bio; |
1da177e4 | 2515 | |
66846572 N |
2516 | if (bio->bi_bdev) |
2517 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2518 | } | |
1da177e4 | 2519 | |
2d4dc890 IL |
2520 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2521 | /** | |
2522 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2523 | * @rq: the request to be flushed | |
2524 | * | |
2525 | * Description: | |
2526 | * Flush all pages in @rq. | |
2527 | */ | |
2528 | void rq_flush_dcache_pages(struct request *rq) | |
2529 | { | |
2530 | struct req_iterator iter; | |
2531 | struct bio_vec *bvec; | |
2532 | ||
2533 | rq_for_each_segment(bvec, rq, iter) | |
2534 | flush_dcache_page(bvec->bv_page); | |
2535 | } | |
2536 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2537 | #endif | |
2538 | ||
ef9e3fac KU |
2539 | /** |
2540 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2541 | * @q : the queue of the device being checked | |
2542 | * | |
2543 | * Description: | |
2544 | * Check if underlying low-level drivers of a device are busy. | |
2545 | * If the drivers want to export their busy state, they must set own | |
2546 | * exporting function using blk_queue_lld_busy() first. | |
2547 | * | |
2548 | * Basically, this function is used only by request stacking drivers | |
2549 | * to stop dispatching requests to underlying devices when underlying | |
2550 | * devices are busy. This behavior helps more I/O merging on the queue | |
2551 | * of the request stacking driver and prevents I/O throughput regression | |
2552 | * on burst I/O load. | |
2553 | * | |
2554 | * Return: | |
2555 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2556 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2557 | */ | |
2558 | int blk_lld_busy(struct request_queue *q) | |
2559 | { | |
2560 | if (q->lld_busy_fn) | |
2561 | return q->lld_busy_fn(q); | |
2562 | ||
2563 | return 0; | |
2564 | } | |
2565 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2566 | ||
b0fd271d KU |
2567 | /** |
2568 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2569 | * @rq: the clone request to be cleaned up | |
2570 | * | |
2571 | * Description: | |
2572 | * Free all bios in @rq for a cloned request. | |
2573 | */ | |
2574 | void blk_rq_unprep_clone(struct request *rq) | |
2575 | { | |
2576 | struct bio *bio; | |
2577 | ||
2578 | while ((bio = rq->bio) != NULL) { | |
2579 | rq->bio = bio->bi_next; | |
2580 | ||
2581 | bio_put(bio); | |
2582 | } | |
2583 | } | |
2584 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2585 | ||
2586 | /* | |
2587 | * Copy attributes of the original request to the clone request. | |
2588 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2589 | */ | |
2590 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2591 | { | |
2592 | dst->cpu = src->cpu; | |
3a2edd0d | 2593 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2594 | dst->cmd_type = src->cmd_type; |
2595 | dst->__sector = blk_rq_pos(src); | |
2596 | dst->__data_len = blk_rq_bytes(src); | |
2597 | dst->nr_phys_segments = src->nr_phys_segments; | |
2598 | dst->ioprio = src->ioprio; | |
2599 | dst->extra_len = src->extra_len; | |
2600 | } | |
2601 | ||
2602 | /** | |
2603 | * blk_rq_prep_clone - Helper function to setup clone request | |
2604 | * @rq: the request to be setup | |
2605 | * @rq_src: original request to be cloned | |
2606 | * @bs: bio_set that bios for clone are allocated from | |
2607 | * @gfp_mask: memory allocation mask for bio | |
2608 | * @bio_ctr: setup function to be called for each clone bio. | |
2609 | * Returns %0 for success, non %0 for failure. | |
2610 | * @data: private data to be passed to @bio_ctr | |
2611 | * | |
2612 | * Description: | |
2613 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2614 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2615 | * are not copied, and copying such parts is the caller's responsibility. | |
2616 | * Also, pages which the original bios are pointing to are not copied | |
2617 | * and the cloned bios just point same pages. | |
2618 | * So cloned bios must be completed before original bios, which means | |
2619 | * the caller must complete @rq before @rq_src. | |
2620 | */ | |
2621 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2622 | struct bio_set *bs, gfp_t gfp_mask, | |
2623 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2624 | void *data) | |
2625 | { | |
2626 | struct bio *bio, *bio_src; | |
2627 | ||
2628 | if (!bs) | |
2629 | bs = fs_bio_set; | |
2630 | ||
2631 | blk_rq_init(NULL, rq); | |
2632 | ||
2633 | __rq_for_each_bio(bio_src, rq_src) { | |
2634 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2635 | if (!bio) | |
2636 | goto free_and_out; | |
2637 | ||
2638 | __bio_clone(bio, bio_src); | |
2639 | ||
2640 | if (bio_integrity(bio_src) && | |
7878cba9 | 2641 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2642 | goto free_and_out; |
2643 | ||
2644 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2645 | goto free_and_out; | |
2646 | ||
2647 | if (rq->bio) { | |
2648 | rq->biotail->bi_next = bio; | |
2649 | rq->biotail = bio; | |
2650 | } else | |
2651 | rq->bio = rq->biotail = bio; | |
2652 | } | |
2653 | ||
2654 | __blk_rq_prep_clone(rq, rq_src); | |
2655 | ||
2656 | return 0; | |
2657 | ||
2658 | free_and_out: | |
2659 | if (bio) | |
2660 | bio_free(bio, bs); | |
2661 | blk_rq_unprep_clone(rq); | |
2662 | ||
2663 | return -ENOMEM; | |
2664 | } | |
2665 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2666 | ||
18887ad9 | 2667 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2668 | { |
2669 | return queue_work(kblockd_workqueue, work); | |
2670 | } | |
1da177e4 LT |
2671 | EXPORT_SYMBOL(kblockd_schedule_work); |
2672 | ||
e43473b7 VG |
2673 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2674 | struct delayed_work *dwork, unsigned long delay) | |
2675 | { | |
2676 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2677 | } | |
2678 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2679 | ||
73c10101 JA |
2680 | #define PLUG_MAGIC 0x91827364 |
2681 | ||
75df7136 SJ |
2682 | /** |
2683 | * blk_start_plug - initialize blk_plug and track it inside the task_struct | |
2684 | * @plug: The &struct blk_plug that needs to be initialized | |
2685 | * | |
2686 | * Description: | |
2687 | * Tracking blk_plug inside the task_struct will help with auto-flushing the | |
2688 | * pending I/O should the task end up blocking between blk_start_plug() and | |
2689 | * blk_finish_plug(). This is important from a performance perspective, but | |
2690 | * also ensures that we don't deadlock. For instance, if the task is blocking | |
2691 | * for a memory allocation, memory reclaim could end up wanting to free a | |
2692 | * page belonging to that request that is currently residing in our private | |
2693 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid | |
2694 | * this kind of deadlock. | |
2695 | */ | |
73c10101 JA |
2696 | void blk_start_plug(struct blk_plug *plug) |
2697 | { | |
2698 | struct task_struct *tsk = current; | |
2699 | ||
2700 | plug->magic = PLUG_MAGIC; | |
2701 | INIT_LIST_HEAD(&plug->list); | |
048c9374 | 2702 | INIT_LIST_HEAD(&plug->cb_list); |
73c10101 JA |
2703 | plug->should_sort = 0; |
2704 | ||
2705 | /* | |
2706 | * If this is a nested plug, don't actually assign it. It will be | |
2707 | * flushed on its own. | |
2708 | */ | |
2709 | if (!tsk->plug) { | |
2710 | /* | |
2711 | * Store ordering should not be needed here, since a potential | |
2712 | * preempt will imply a full memory barrier | |
2713 | */ | |
2714 | tsk->plug = plug; | |
2715 | } | |
2716 | } | |
2717 | EXPORT_SYMBOL(blk_start_plug); | |
2718 | ||
2719 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2720 | { | |
2721 | struct request *rqa = container_of(a, struct request, queuelist); | |
2722 | struct request *rqb = container_of(b, struct request, queuelist); | |
2723 | ||
f83e8261 | 2724 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2725 | } |
2726 | ||
49cac01e JA |
2727 | /* |
2728 | * If 'from_schedule' is true, then postpone the dispatch of requests | |
2729 | * until a safe kblockd context. We due this to avoid accidental big | |
2730 | * additional stack usage in driver dispatch, in places where the originally | |
2731 | * plugger did not intend it. | |
2732 | */ | |
f6603783 | 2733 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
49cac01e | 2734 | bool from_schedule) |
99e22598 | 2735 | __releases(q->queue_lock) |
94b5eb28 | 2736 | { |
49cac01e | 2737 | trace_block_unplug(q, depth, !from_schedule); |
99e22598 JA |
2738 | |
2739 | /* | |
2740 | * If we are punting this to kblockd, then we can safely drop | |
2741 | * the queue_lock before waking kblockd (which needs to take | |
2742 | * this lock). | |
2743 | */ | |
2744 | if (from_schedule) { | |
2745 | spin_unlock(q->queue_lock); | |
24ecfbe2 | 2746 | blk_run_queue_async(q); |
99e22598 | 2747 | } else { |
24ecfbe2 | 2748 | __blk_run_queue(q); |
99e22598 JA |
2749 | spin_unlock(q->queue_lock); |
2750 | } | |
2751 | ||
94b5eb28 JA |
2752 | } |
2753 | ||
048c9374 N |
2754 | static void flush_plug_callbacks(struct blk_plug *plug) |
2755 | { | |
2756 | LIST_HEAD(callbacks); | |
2757 | ||
2758 | if (list_empty(&plug->cb_list)) | |
2759 | return; | |
2760 | ||
2761 | list_splice_init(&plug->cb_list, &callbacks); | |
2762 | ||
2763 | while (!list_empty(&callbacks)) { | |
2764 | struct blk_plug_cb *cb = list_first_entry(&callbacks, | |
2765 | struct blk_plug_cb, | |
2766 | list); | |
2767 | list_del(&cb->list); | |
2768 | cb->callback(cb); | |
2769 | } | |
2770 | } | |
2771 | ||
49cac01e | 2772 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
73c10101 JA |
2773 | { |
2774 | struct request_queue *q; | |
2775 | unsigned long flags; | |
2776 | struct request *rq; | |
109b8129 | 2777 | LIST_HEAD(list); |
94b5eb28 | 2778 | unsigned int depth; |
73c10101 JA |
2779 | |
2780 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2781 | ||
048c9374 | 2782 | flush_plug_callbacks(plug); |
73c10101 JA |
2783 | if (list_empty(&plug->list)) |
2784 | return; | |
2785 | ||
109b8129 N |
2786 | list_splice_init(&plug->list, &list); |
2787 | ||
2788 | if (plug->should_sort) { | |
2789 | list_sort(NULL, &list, plug_rq_cmp); | |
2790 | plug->should_sort = 0; | |
2791 | } | |
73c10101 JA |
2792 | |
2793 | q = NULL; | |
94b5eb28 | 2794 | depth = 0; |
18811272 JA |
2795 | |
2796 | /* | |
2797 | * Save and disable interrupts here, to avoid doing it for every | |
2798 | * queue lock we have to take. | |
2799 | */ | |
73c10101 | 2800 | local_irq_save(flags); |
109b8129 N |
2801 | while (!list_empty(&list)) { |
2802 | rq = list_entry_rq(list.next); | |
73c10101 | 2803 | list_del_init(&rq->queuelist); |
73c10101 JA |
2804 | BUG_ON(!rq->q); |
2805 | if (rq->q != q) { | |
99e22598 JA |
2806 | /* |
2807 | * This drops the queue lock | |
2808 | */ | |
2809 | if (q) | |
49cac01e | 2810 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2811 | q = rq->q; |
94b5eb28 | 2812 | depth = 0; |
73c10101 JA |
2813 | spin_lock(q->queue_lock); |
2814 | } | |
73c10101 JA |
2815 | /* |
2816 | * rq is already accounted, so use raw insert | |
2817 | */ | |
401a18e9 JA |
2818 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
2819 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
2820 | else | |
2821 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
94b5eb28 JA |
2822 | |
2823 | depth++; | |
73c10101 JA |
2824 | } |
2825 | ||
99e22598 JA |
2826 | /* |
2827 | * This drops the queue lock | |
2828 | */ | |
2829 | if (q) | |
49cac01e | 2830 | queue_unplugged(q, depth, from_schedule); |
73c10101 | 2831 | |
73c10101 JA |
2832 | local_irq_restore(flags); |
2833 | } | |
73c10101 JA |
2834 | |
2835 | void blk_finish_plug(struct blk_plug *plug) | |
2836 | { | |
f6603783 | 2837 | blk_flush_plug_list(plug, false); |
73c10101 | 2838 | |
88b996cd CH |
2839 | if (plug == current->plug) |
2840 | current->plug = NULL; | |
73c10101 | 2841 | } |
88b996cd | 2842 | EXPORT_SYMBOL(blk_finish_plug); |
73c10101 | 2843 | |
1da177e4 LT |
2844 | int __init blk_dev_init(void) |
2845 | { | |
9eb55b03 NK |
2846 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2847 | sizeof(((struct request *)0)->cmd_flags)); | |
2848 | ||
89b90be2 TH |
2849 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2850 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2851 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2852 | if (!kblockd_workqueue) |
2853 | panic("Failed to create kblockd\n"); | |
2854 | ||
2855 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2856 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2857 | |
8324aa91 | 2858 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2859 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2860 | |
d38ecf93 | 2861 | return 0; |
1da177e4 | 2862 | } |