]>
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> |
55782138 LZ |
30 | |
31 | #define CREATE_TRACE_POINTS | |
32 | #include <trace/events/block.h> | |
1da177e4 | 33 | |
8324aa91 JA |
34 | #include "blk.h" |
35 | ||
d07335e5 | 36 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 37 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 38 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 39 | |
165125e1 | 40 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
41 | |
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; | |
156 | ||
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 | } |
143a87f4 TH |
162 | |
163 | if (unlikely(rq->cmd_flags & REQ_QUIET)) | |
164 | set_bit(BIO_QUIET, &bio->bi_flags); | |
165 | ||
166 | bio->bi_size -= nbytes; | |
167 | bio->bi_sector += (nbytes >> 9); | |
168 | ||
169 | if (bio_integrity(bio)) | |
170 | bio_integrity_advance(bio, nbytes); | |
171 | ||
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 | ||
1da177e4 LT |
200 | /* |
201 | * "plug" the device if there are no outstanding requests: this will | |
202 | * force the transfer to start only after we have put all the requests | |
203 | * on the list. | |
204 | * | |
205 | * This is called with interrupts off and no requests on the queue and | |
206 | * with the queue lock held. | |
207 | */ | |
165125e1 | 208 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
209 | { |
210 | WARN_ON(!irqs_disabled()); | |
211 | ||
212 | /* | |
213 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
214 | * which will restart the queueing | |
215 | */ | |
7daac490 | 216 | if (blk_queue_stopped(q)) |
1da177e4 LT |
217 | return; |
218 | ||
e48ec690 | 219 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 220 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 221 | trace_block_plug(q); |
2056a782 | 222 | } |
1da177e4 | 223 | } |
1da177e4 LT |
224 | EXPORT_SYMBOL(blk_plug_device); |
225 | ||
6c5e0c4d JA |
226 | /** |
227 | * blk_plug_device_unlocked - plug a device without queue lock held | |
228 | * @q: The &struct request_queue to plug | |
229 | * | |
230 | * Description: | |
231 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
232 | * interrupts. | |
233 | **/ | |
234 | void blk_plug_device_unlocked(struct request_queue *q) | |
235 | { | |
236 | unsigned long flags; | |
237 | ||
238 | spin_lock_irqsave(q->queue_lock, flags); | |
239 | blk_plug_device(q); | |
240 | spin_unlock_irqrestore(q->queue_lock, flags); | |
241 | } | |
242 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
243 | ||
1da177e4 LT |
244 | /* |
245 | * remove the queue from the plugged list, if present. called with | |
246 | * queue lock held and interrupts disabled. | |
247 | */ | |
165125e1 | 248 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
249 | { |
250 | WARN_ON(!irqs_disabled()); | |
251 | ||
e48ec690 | 252 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
253 | return 0; |
254 | ||
255 | del_timer(&q->unplug_timer); | |
256 | return 1; | |
257 | } | |
1da177e4 LT |
258 | EXPORT_SYMBOL(blk_remove_plug); |
259 | ||
260 | /* | |
261 | * remove the plug and let it rip.. | |
262 | */ | |
165125e1 | 263 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 264 | { |
7daac490 | 265 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 | 266 | return; |
a31a9738 | 267 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) |
1da177e4 LT |
268 | return; |
269 | ||
22e2c507 | 270 | q->request_fn(q); |
1da177e4 | 271 | } |
1da177e4 LT |
272 | |
273 | /** | |
274 | * generic_unplug_device - fire a request queue | |
165125e1 | 275 | * @q: The &struct request_queue in question |
1da177e4 LT |
276 | * |
277 | * Description: | |
278 | * Linux uses plugging to build bigger requests queues before letting | |
279 | * the device have at them. If a queue is plugged, the I/O scheduler | |
280 | * is still adding and merging requests on the queue. Once the queue | |
281 | * gets unplugged, the request_fn defined for the queue is invoked and | |
282 | * transfers started. | |
283 | **/ | |
165125e1 | 284 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 285 | { |
dbaf2c00 JA |
286 | if (blk_queue_plugged(q)) { |
287 | spin_lock_irq(q->queue_lock); | |
288 | __generic_unplug_device(q); | |
289 | spin_unlock_irq(q->queue_lock); | |
290 | } | |
1da177e4 LT |
291 | } |
292 | EXPORT_SYMBOL(generic_unplug_device); | |
293 | ||
294 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
295 | struct page *page) | |
296 | { | |
165125e1 | 297 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 298 | |
2ad8b1ef | 299 | blk_unplug(q); |
1da177e4 LT |
300 | } |
301 | ||
86db1e29 | 302 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 303 | { |
165125e1 JA |
304 | struct request_queue *q = |
305 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 306 | |
5f3ea37c | 307 | trace_block_unplug_io(q); |
1da177e4 LT |
308 | q->unplug_fn(q); |
309 | } | |
310 | ||
86db1e29 | 311 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 312 | { |
165125e1 | 313 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 314 | |
5f3ea37c | 315 | trace_block_unplug_timer(q); |
18887ad9 | 316 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
317 | } |
318 | ||
2ad8b1ef AB |
319 | void blk_unplug(struct request_queue *q) |
320 | { | |
321 | /* | |
322 | * devices don't necessarily have an ->unplug_fn defined | |
323 | */ | |
324 | if (q->unplug_fn) { | |
5f3ea37c | 325 | trace_block_unplug_io(q); |
2ad8b1ef AB |
326 | q->unplug_fn(q); |
327 | } | |
328 | } | |
329 | EXPORT_SYMBOL(blk_unplug); | |
330 | ||
1da177e4 LT |
331 | /** |
332 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 333 | * @q: The &struct request_queue in question |
1da177e4 LT |
334 | * |
335 | * Description: | |
336 | * blk_start_queue() will clear the stop flag on the queue, and call | |
337 | * the request_fn for the queue if it was in a stopped state when | |
338 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
339 | **/ | |
165125e1 | 340 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 341 | { |
a038e253 PBG |
342 | WARN_ON(!irqs_disabled()); |
343 | ||
75ad23bc | 344 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
1654e741 | 345 | __blk_run_queue(q, false); |
1da177e4 | 346 | } |
1da177e4 LT |
347 | EXPORT_SYMBOL(blk_start_queue); |
348 | ||
349 | /** | |
350 | * blk_stop_queue - stop a queue | |
165125e1 | 351 | * @q: The &struct request_queue in question |
1da177e4 LT |
352 | * |
353 | * Description: | |
354 | * The Linux block layer assumes that a block driver will consume all | |
355 | * entries on the request queue when the request_fn strategy is called. | |
356 | * Often this will not happen, because of hardware limitations (queue | |
357 | * depth settings). If a device driver gets a 'queue full' response, | |
358 | * or if it simply chooses not to queue more I/O at one point, it can | |
359 | * call this function to prevent the request_fn from being called until | |
360 | * the driver has signalled it's ready to go again. This happens by calling | |
361 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
362 | **/ | |
165125e1 | 363 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
364 | { |
365 | blk_remove_plug(q); | |
75ad23bc | 366 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
367 | } |
368 | EXPORT_SYMBOL(blk_stop_queue); | |
369 | ||
370 | /** | |
371 | * blk_sync_queue - cancel any pending callbacks on a queue | |
372 | * @q: the queue | |
373 | * | |
374 | * Description: | |
375 | * The block layer may perform asynchronous callback activity | |
376 | * on a queue, such as calling the unplug function after a timeout. | |
377 | * A block device may call blk_sync_queue to ensure that any | |
378 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 379 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
380 | * that its ->make_request_fn will not re-add plugging prior to calling |
381 | * this function. | |
382 | * | |
da527770 VG |
383 | * This function does not cancel any asynchronous activity arising |
384 | * out of elevator or throttling code. That would require elevaotor_exit() | |
385 | * and blk_throtl_exit() to be called with queue lock initialized. | |
386 | * | |
1da177e4 LT |
387 | */ |
388 | void blk_sync_queue(struct request_queue *q) | |
389 | { | |
390 | del_timer_sync(&q->unplug_timer); | |
70ed28b9 | 391 | del_timer_sync(&q->timeout); |
64d01dc9 | 392 | cancel_work_sync(&q->unplug_work); |
1da177e4 LT |
393 | } |
394 | EXPORT_SYMBOL(blk_sync_queue); | |
395 | ||
396 | /** | |
80a4b58e | 397 | * __blk_run_queue - run a single device queue |
1da177e4 | 398 | * @q: The queue to run |
1654e741 | 399 | * @force_kblockd: Don't run @q->request_fn directly. Use kblockd. |
80a4b58e JA |
400 | * |
401 | * Description: | |
402 | * See @blk_run_queue. This variant must be called with the queue lock | |
403 | * held and interrupts disabled. | |
404 | * | |
1da177e4 | 405 | */ |
1654e741 | 406 | void __blk_run_queue(struct request_queue *q, bool force_kblockd) |
1da177e4 | 407 | { |
1da177e4 | 408 | blk_remove_plug(q); |
dac07ec1 | 409 | |
a538cd03 TH |
410 | if (unlikely(blk_queue_stopped(q))) |
411 | return; | |
412 | ||
413 | if (elv_queue_empty(q)) | |
414 | return; | |
415 | ||
dac07ec1 JA |
416 | /* |
417 | * Only recurse once to avoid overrunning the stack, let the unplug | |
418 | * handling reinvoke the handler shortly if we already got there. | |
419 | */ | |
1654e741 | 420 | if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { |
a538cd03 TH |
421 | q->request_fn(q); |
422 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
423 | } else { | |
424 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
425 | kblockd_schedule_work(q, &q->unplug_work); | |
426 | } | |
75ad23bc NP |
427 | } |
428 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 429 | |
75ad23bc NP |
430 | /** |
431 | * blk_run_queue - run a single device queue | |
432 | * @q: The queue to run | |
80a4b58e JA |
433 | * |
434 | * Description: | |
435 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 436 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
437 | */ |
438 | void blk_run_queue(struct request_queue *q) | |
439 | { | |
440 | unsigned long flags; | |
441 | ||
442 | spin_lock_irqsave(q->queue_lock, flags); | |
1654e741 | 443 | __blk_run_queue(q, false); |
1da177e4 LT |
444 | spin_unlock_irqrestore(q->queue_lock, flags); |
445 | } | |
446 | EXPORT_SYMBOL(blk_run_queue); | |
447 | ||
165125e1 | 448 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
449 | { |
450 | kobject_put(&q->kobj); | |
451 | } | |
483f4afc | 452 | |
c94a96ac VG |
453 | /* |
454 | * Note: If a driver supplied the queue lock, it should not zap that lock | |
455 | * unexpectedly as some queue cleanup components like elevator_exit() and | |
456 | * blk_throtl_exit() need queue lock. | |
457 | */ | |
6728cb0e | 458 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 459 | { |
e3335de9 JA |
460 | /* |
461 | * We know we have process context here, so we can be a little | |
462 | * cautious and ensure that pending block actions on this device | |
463 | * are done before moving on. Going into this function, we should | |
464 | * not have processes doing IO to this device. | |
465 | */ | |
466 | blk_sync_queue(q); | |
467 | ||
31373d09 | 468 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); |
483f4afc | 469 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 470 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
471 | mutex_unlock(&q->sysfs_lock); |
472 | ||
473 | if (q->elevator) | |
474 | elevator_exit(q->elevator); | |
475 | ||
da527770 VG |
476 | blk_throtl_exit(q); |
477 | ||
483f4afc AV |
478 | blk_put_queue(q); |
479 | } | |
1da177e4 LT |
480 | EXPORT_SYMBOL(blk_cleanup_queue); |
481 | ||
165125e1 | 482 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
483 | { |
484 | struct request_list *rl = &q->rq; | |
485 | ||
1abec4fd MS |
486 | if (unlikely(rl->rq_pool)) |
487 | return 0; | |
488 | ||
1faa16d2 JA |
489 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
490 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 491 | rl->elvpriv = 0; |
1faa16d2 JA |
492 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
493 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 494 | |
1946089a CL |
495 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
496 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
497 | |
498 | if (!rl->rq_pool) | |
499 | return -ENOMEM; | |
500 | ||
501 | return 0; | |
502 | } | |
503 | ||
165125e1 | 504 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 505 | { |
1946089a CL |
506 | return blk_alloc_queue_node(gfp_mask, -1); |
507 | } | |
508 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 509 | |
165125e1 | 510 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 511 | { |
165125e1 | 512 | struct request_queue *q; |
e0bf68dd | 513 | int err; |
1946089a | 514 | |
8324aa91 | 515 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 516 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
517 | if (!q) |
518 | return NULL; | |
519 | ||
e0bf68dd PZ |
520 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
521 | q->backing_dev_info.unplug_io_data = q; | |
0989a025 JA |
522 | q->backing_dev_info.ra_pages = |
523 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
524 | q->backing_dev_info.state = 0; | |
525 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 526 | q->backing_dev_info.name = "block"; |
0989a025 | 527 | |
e0bf68dd PZ |
528 | err = bdi_init(&q->backing_dev_info); |
529 | if (err) { | |
8324aa91 | 530 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
531 | return NULL; |
532 | } | |
533 | ||
e43473b7 VG |
534 | if (blk_throtl_init(q)) { |
535 | kmem_cache_free(blk_requestq_cachep, q); | |
536 | return NULL; | |
537 | } | |
538 | ||
31373d09 MG |
539 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
540 | laptop_mode_timer_fn, (unsigned long) q); | |
1da177e4 | 541 | init_timer(&q->unplug_timer); |
242f9dcb JA |
542 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
543 | INIT_LIST_HEAD(&q->timeout_list); | |
ae1b1539 TH |
544 | INIT_LIST_HEAD(&q->flush_queue[0]); |
545 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
546 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
713ada9b | 547 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 548 | |
8324aa91 | 549 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 550 | |
483f4afc | 551 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 552 | spin_lock_init(&q->__queue_lock); |
483f4afc | 553 | |
c94a96ac VG |
554 | /* |
555 | * By default initialize queue_lock to internal lock and driver can | |
556 | * override it later if need be. | |
557 | */ | |
558 | q->queue_lock = &q->__queue_lock; | |
559 | ||
1da177e4 LT |
560 | return q; |
561 | } | |
1946089a | 562 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
563 | |
564 | /** | |
565 | * blk_init_queue - prepare a request queue for use with a block device | |
566 | * @rfn: The function to be called to process requests that have been | |
567 | * placed on the queue. | |
568 | * @lock: Request queue spin lock | |
569 | * | |
570 | * Description: | |
571 | * If a block device wishes to use the standard request handling procedures, | |
572 | * which sorts requests and coalesces adjacent requests, then it must | |
573 | * call blk_init_queue(). The function @rfn will be called when there | |
574 | * are requests on the queue that need to be processed. If the device | |
575 | * supports plugging, then @rfn may not be called immediately when requests | |
576 | * are available on the queue, but may be called at some time later instead. | |
577 | * Plugged queues are generally unplugged when a buffer belonging to one | |
578 | * of the requests on the queue is needed, or due to memory pressure. | |
579 | * | |
580 | * @rfn is not required, or even expected, to remove all requests off the | |
581 | * queue, but only as many as it can handle at a time. If it does leave | |
582 | * requests on the queue, it is responsible for arranging that the requests | |
583 | * get dealt with eventually. | |
584 | * | |
585 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
586 | * request queue; this lock will be taken also from interrupt context, so irq |
587 | * disabling is needed for it. | |
1da177e4 | 588 | * |
710027a4 | 589 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
590 | * it didn't succeed. |
591 | * | |
592 | * Note: | |
593 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
594 | * when the block device is deactivated (such as at module unload). | |
595 | **/ | |
1946089a | 596 | |
165125e1 | 597 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 598 | { |
1946089a CL |
599 | return blk_init_queue_node(rfn, lock, -1); |
600 | } | |
601 | EXPORT_SYMBOL(blk_init_queue); | |
602 | ||
165125e1 | 603 | struct request_queue * |
1946089a CL |
604 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
605 | { | |
c86d1b8a | 606 | struct request_queue *uninit_q, *q; |
1da177e4 | 607 | |
c86d1b8a MS |
608 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
609 | if (!uninit_q) | |
610 | return NULL; | |
611 | ||
612 | q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id); | |
613 | if (!q) | |
614 | blk_cleanup_queue(uninit_q); | |
615 | ||
616 | return q; | |
01effb0d MS |
617 | } |
618 | EXPORT_SYMBOL(blk_init_queue_node); | |
619 | ||
620 | struct request_queue * | |
621 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
622 | spinlock_t *lock) | |
623 | { | |
624 | return blk_init_allocated_queue_node(q, rfn, lock, -1); | |
625 | } | |
626 | EXPORT_SYMBOL(blk_init_allocated_queue); | |
627 | ||
628 | struct request_queue * | |
629 | blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn, | |
630 | spinlock_t *lock, int node_id) | |
631 | { | |
1da177e4 LT |
632 | if (!q) |
633 | return NULL; | |
634 | ||
1946089a | 635 | q->node = node_id; |
c86d1b8a | 636 | if (blk_init_free_list(q)) |
8669aafd | 637 | return NULL; |
1da177e4 LT |
638 | |
639 | q->request_fn = rfn; | |
1da177e4 | 640 | q->prep_rq_fn = NULL; |
28018c24 | 641 | q->unprep_rq_fn = NULL; |
1da177e4 | 642 | q->unplug_fn = generic_unplug_device; |
bc58ba94 | 643 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
644 | |
645 | /* Override internal queue lock with supplied lock pointer */ | |
646 | if (lock) | |
647 | q->queue_lock = lock; | |
1da177e4 | 648 | |
f3b144aa JA |
649 | /* |
650 | * This also sets hw/phys segments, boundary and size | |
651 | */ | |
1da177e4 | 652 | blk_queue_make_request(q, __make_request); |
1da177e4 | 653 | |
44ec9542 AS |
654 | q->sg_reserved_size = INT_MAX; |
655 | ||
1da177e4 LT |
656 | /* |
657 | * all done | |
658 | */ | |
659 | if (!elevator_init(q, NULL)) { | |
660 | blk_queue_congestion_threshold(q); | |
661 | return q; | |
662 | } | |
663 | ||
1da177e4 LT |
664 | return NULL; |
665 | } | |
01effb0d | 666 | EXPORT_SYMBOL(blk_init_allocated_queue_node); |
1da177e4 | 667 | |
165125e1 | 668 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 669 | { |
fde6ad22 | 670 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 671 | kobject_get(&q->kobj); |
1da177e4 LT |
672 | return 0; |
673 | } | |
674 | ||
675 | return 1; | |
676 | } | |
1da177e4 | 677 | |
165125e1 | 678 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 679 | { |
4aff5e23 | 680 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 681 | elv_put_request(q, rq); |
1da177e4 LT |
682 | mempool_free(rq, q->rq.rq_pool); |
683 | } | |
684 | ||
1ea25ecb | 685 | static struct request * |
42dad764 | 686 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) |
1da177e4 LT |
687 | { |
688 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
689 | ||
690 | if (!rq) | |
691 | return NULL; | |
692 | ||
2a4aa30c | 693 | blk_rq_init(q, rq); |
1afb20f3 | 694 | |
42dad764 | 695 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 696 | |
cb98fc8b | 697 | if (priv) { |
cb78b285 | 698 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
699 | mempool_free(rq, q->rq.rq_pool); |
700 | return NULL; | |
701 | } | |
4aff5e23 | 702 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 703 | } |
1da177e4 | 704 | |
cb98fc8b | 705 | return rq; |
1da177e4 LT |
706 | } |
707 | ||
708 | /* | |
709 | * ioc_batching returns true if the ioc is a valid batching request and | |
710 | * should be given priority access to a request. | |
711 | */ | |
165125e1 | 712 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
713 | { |
714 | if (!ioc) | |
715 | return 0; | |
716 | ||
717 | /* | |
718 | * Make sure the process is able to allocate at least 1 request | |
719 | * even if the batch times out, otherwise we could theoretically | |
720 | * lose wakeups. | |
721 | */ | |
722 | return ioc->nr_batch_requests == q->nr_batching || | |
723 | (ioc->nr_batch_requests > 0 | |
724 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
725 | } | |
726 | ||
727 | /* | |
728 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
729 | * will cause the process to be a "batcher" on all queues in the system. This | |
730 | * is the behaviour we want though - once it gets a wakeup it should be given | |
731 | * a nice run. | |
732 | */ | |
165125e1 | 733 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
734 | { |
735 | if (!ioc || ioc_batching(q, ioc)) | |
736 | return; | |
737 | ||
738 | ioc->nr_batch_requests = q->nr_batching; | |
739 | ioc->last_waited = jiffies; | |
740 | } | |
741 | ||
1faa16d2 | 742 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
743 | { |
744 | struct request_list *rl = &q->rq; | |
745 | ||
1faa16d2 JA |
746 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
747 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 748 | |
1faa16d2 JA |
749 | if (rl->count[sync] + 1 <= q->nr_requests) { |
750 | if (waitqueue_active(&rl->wait[sync])) | |
751 | wake_up(&rl->wait[sync]); | |
1da177e4 | 752 | |
1faa16d2 | 753 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
754 | } |
755 | } | |
756 | ||
757 | /* | |
758 | * A request has just been released. Account for it, update the full and | |
759 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
760 | */ | |
1faa16d2 | 761 | static void freed_request(struct request_queue *q, int sync, int priv) |
1da177e4 LT |
762 | { |
763 | struct request_list *rl = &q->rq; | |
764 | ||
1faa16d2 | 765 | rl->count[sync]--; |
cb98fc8b TH |
766 | if (priv) |
767 | rl->elvpriv--; | |
1da177e4 | 768 | |
1faa16d2 | 769 | __freed_request(q, sync); |
1da177e4 | 770 | |
1faa16d2 JA |
771 | if (unlikely(rl->starved[sync ^ 1])) |
772 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
773 | } |
774 | ||
9d5a4e94 MS |
775 | /* |
776 | * Determine if elevator data should be initialized when allocating the | |
777 | * request associated with @bio. | |
778 | */ | |
779 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
780 | { | |
781 | if (!bio) | |
782 | return true; | |
783 | ||
784 | /* | |
785 | * Flush requests do not use the elevator so skip initialization. | |
786 | * This allows a request to share the flush and elevator data. | |
787 | */ | |
788 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
789 | return false; | |
790 | ||
791 | return true; | |
792 | } | |
793 | ||
1da177e4 | 794 | /* |
d6344532 NP |
795 | * Get a free request, queue_lock must be held. |
796 | * Returns NULL on failure, with queue_lock held. | |
797 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 798 | */ |
165125e1 | 799 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 800 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
801 | { |
802 | struct request *rq = NULL; | |
803 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 804 | struct io_context *ioc = NULL; |
1faa16d2 | 805 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
9d5a4e94 | 806 | int may_queue, priv = 0; |
88ee5ef1 | 807 | |
7749a8d4 | 808 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
809 | if (may_queue == ELV_MQUEUE_NO) |
810 | goto rq_starved; | |
811 | ||
1faa16d2 JA |
812 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
813 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 814 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
815 | /* |
816 | * The queue will fill after this allocation, so set | |
817 | * it as full, and mark this process as "batching". | |
818 | * This process will be allowed to complete a batch of | |
819 | * requests, others will be blocked. | |
820 | */ | |
1faa16d2 | 821 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 822 | ioc_set_batching(q, ioc); |
1faa16d2 | 823 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
824 | } else { |
825 | if (may_queue != ELV_MQUEUE_MUST | |
826 | && !ioc_batching(q, ioc)) { | |
827 | /* | |
828 | * The queue is full and the allocating | |
829 | * process is not a "batcher", and not | |
830 | * exempted by the IO scheduler | |
831 | */ | |
832 | goto out; | |
833 | } | |
834 | } | |
1da177e4 | 835 | } |
1faa16d2 | 836 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
837 | } |
838 | ||
082cf69e JA |
839 | /* |
840 | * Only allow batching queuers to allocate up to 50% over the defined | |
841 | * limit of requests, otherwise we could have thousands of requests | |
842 | * allocated with any setting of ->nr_requests | |
843 | */ | |
1faa16d2 | 844 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 845 | goto out; |
fd782a4a | 846 | |
1faa16d2 JA |
847 | rl->count[is_sync]++; |
848 | rl->starved[is_sync] = 0; | |
cb98fc8b | 849 | |
9d5a4e94 MS |
850 | if (blk_rq_should_init_elevator(bio)) { |
851 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
852 | if (priv) | |
853 | rl->elvpriv++; | |
854 | } | |
cb98fc8b | 855 | |
f253b86b JA |
856 | if (blk_queue_io_stat(q)) |
857 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
858 | spin_unlock_irq(q->queue_lock); |
859 | ||
7749a8d4 | 860 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 861 | if (unlikely(!rq)) { |
1da177e4 LT |
862 | /* |
863 | * Allocation failed presumably due to memory. Undo anything | |
864 | * we might have messed up. | |
865 | * | |
866 | * Allocating task should really be put onto the front of the | |
867 | * wait queue, but this is pretty rare. | |
868 | */ | |
869 | spin_lock_irq(q->queue_lock); | |
1faa16d2 | 870 | freed_request(q, is_sync, priv); |
1da177e4 LT |
871 | |
872 | /* | |
873 | * in the very unlikely event that allocation failed and no | |
874 | * requests for this direction was pending, mark us starved | |
875 | * so that freeing of a request in the other direction will | |
876 | * notice us. another possible fix would be to split the | |
877 | * rq mempool into READ and WRITE | |
878 | */ | |
879 | rq_starved: | |
1faa16d2 JA |
880 | if (unlikely(rl->count[is_sync] == 0)) |
881 | rl->starved[is_sync] = 1; | |
1da177e4 | 882 | |
1da177e4 LT |
883 | goto out; |
884 | } | |
885 | ||
88ee5ef1 JA |
886 | /* |
887 | * ioc may be NULL here, and ioc_batching will be false. That's | |
888 | * OK, if the queue is under the request limit then requests need | |
889 | * not count toward the nr_batch_requests limit. There will always | |
890 | * be some limit enforced by BLK_BATCH_TIME. | |
891 | */ | |
1da177e4 LT |
892 | if (ioc_batching(q, ioc)) |
893 | ioc->nr_batch_requests--; | |
6728cb0e | 894 | |
1faa16d2 | 895 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 896 | out: |
1da177e4 LT |
897 | return rq; |
898 | } | |
899 | ||
900 | /* | |
901 | * No available requests for this queue, unplug the device and wait for some | |
902 | * requests to become available. | |
d6344532 NP |
903 | * |
904 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 905 | */ |
165125e1 | 906 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 907 | struct bio *bio) |
1da177e4 | 908 | { |
1faa16d2 | 909 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
910 | struct request *rq; |
911 | ||
7749a8d4 | 912 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
913 | while (!rq) { |
914 | DEFINE_WAIT(wait); | |
05caf8db | 915 | struct io_context *ioc; |
1da177e4 LT |
916 | struct request_list *rl = &q->rq; |
917 | ||
1faa16d2 | 918 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
919 | TASK_UNINTERRUPTIBLE); |
920 | ||
1faa16d2 | 921 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 922 | |
05caf8db ZY |
923 | __generic_unplug_device(q); |
924 | spin_unlock_irq(q->queue_lock); | |
925 | io_schedule(); | |
1da177e4 | 926 | |
05caf8db ZY |
927 | /* |
928 | * After sleeping, we become a "batching" process and | |
929 | * will be able to allocate at least one request, and | |
930 | * up to a big batch of them for a small period time. | |
931 | * See ioc_batching, ioc_set_batching | |
932 | */ | |
933 | ioc = current_io_context(GFP_NOIO, q->node); | |
934 | ioc_set_batching(q, ioc); | |
d6344532 | 935 | |
05caf8db | 936 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 937 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
938 | |
939 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
940 | }; | |
1da177e4 LT |
941 | |
942 | return rq; | |
943 | } | |
944 | ||
165125e1 | 945 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
946 | { |
947 | struct request *rq; | |
948 | ||
949 | BUG_ON(rw != READ && rw != WRITE); | |
950 | ||
d6344532 NP |
951 | spin_lock_irq(q->queue_lock); |
952 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 953 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 954 | } else { |
22e2c507 | 955 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
956 | if (!rq) |
957 | spin_unlock_irq(q->queue_lock); | |
958 | } | |
959 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
960 | |
961 | return rq; | |
962 | } | |
1da177e4 LT |
963 | EXPORT_SYMBOL(blk_get_request); |
964 | ||
dc72ef4a | 965 | /** |
79eb63e9 | 966 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 967 | * @q: target request queue |
79eb63e9 BH |
968 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
969 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 970 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 971 | * |
79eb63e9 BH |
972 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
973 | * type commands. Where the struct request needs to be farther initialized by | |
974 | * the caller. It is passed a &struct bio, which describes the memory info of | |
975 | * the I/O transfer. | |
dc72ef4a | 976 | * |
79eb63e9 BH |
977 | * The caller of blk_make_request must make sure that bi_io_vec |
978 | * are set to describe the memory buffers. That bio_data_dir() will return | |
979 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
980 | * are properly set accordingly) | |
981 | * | |
982 | * If called under none-sleepable conditions, mapped bio buffers must not | |
983 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
984 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
985 | * BUG. | |
53674ac5 JA |
986 | * |
987 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
988 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
989 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
990 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
991 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
992 | * of bio_alloc(), as that avoids the mempool deadlock. | |
993 | * If possible a big IO should be split into smaller parts when allocation | |
994 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 995 | */ |
79eb63e9 BH |
996 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
997 | gfp_t gfp_mask) | |
dc72ef4a | 998 | { |
79eb63e9 BH |
999 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
1000 | ||
1001 | if (unlikely(!rq)) | |
1002 | return ERR_PTR(-ENOMEM); | |
1003 | ||
1004 | for_each_bio(bio) { | |
1005 | struct bio *bounce_bio = bio; | |
1006 | int ret; | |
1007 | ||
1008 | blk_queue_bounce(q, &bounce_bio); | |
1009 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
1010 | if (unlikely(ret)) { | |
1011 | blk_put_request(rq); | |
1012 | return ERR_PTR(ret); | |
1013 | } | |
1014 | } | |
1015 | ||
1016 | return rq; | |
dc72ef4a | 1017 | } |
79eb63e9 | 1018 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 1019 | |
1da177e4 LT |
1020 | /** |
1021 | * blk_requeue_request - put a request back on queue | |
1022 | * @q: request queue where request should be inserted | |
1023 | * @rq: request to be inserted | |
1024 | * | |
1025 | * Description: | |
1026 | * Drivers often keep queueing requests until the hardware cannot accept | |
1027 | * more, when that condition happens we need to put the request back | |
1028 | * on the queue. Must be called with queue lock held. | |
1029 | */ | |
165125e1 | 1030 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1031 | { |
242f9dcb JA |
1032 | blk_delete_timer(rq); |
1033 | blk_clear_rq_complete(rq); | |
5f3ea37c | 1034 | trace_block_rq_requeue(q, rq); |
2056a782 | 1035 | |
1da177e4 LT |
1036 | if (blk_rq_tagged(rq)) |
1037 | blk_queue_end_tag(q, rq); | |
1038 | ||
ba396a6c JB |
1039 | BUG_ON(blk_queued_rq(rq)); |
1040 | ||
1da177e4 LT |
1041 | elv_requeue_request(q, rq); |
1042 | } | |
1da177e4 LT |
1043 | EXPORT_SYMBOL(blk_requeue_request); |
1044 | ||
1045 | /** | |
710027a4 | 1046 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
1047 | * @q: request queue where request should be inserted |
1048 | * @rq: request to be inserted | |
1049 | * @at_head: insert request at head or tail of queue | |
1050 | * @data: private data | |
1da177e4 LT |
1051 | * |
1052 | * Description: | |
1053 | * Many block devices need to execute commands asynchronously, so they don't | |
1054 | * block the whole kernel from preemption during request execution. This is | |
1055 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
1056 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
1057 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
1058 | * |
1059 | * We have the option of inserting the head or the tail of the queue. | |
1060 | * Typically we use the tail for new ioctls and so forth. We use the head | |
1061 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
1062 | * host that is unable to accept a particular command. | |
1063 | */ | |
165125e1 | 1064 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 1065 | int at_head, void *data) |
1da177e4 | 1066 | { |
867d1191 | 1067 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
1068 | unsigned long flags; |
1069 | ||
1070 | /* | |
1071 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
1072 | * must not attempt merges on this) and that it acts as a soft | |
1073 | * barrier | |
1074 | */ | |
4aff5e23 | 1075 | rq->cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
1076 | |
1077 | rq->special = data; | |
1078 | ||
1079 | spin_lock_irqsave(q->queue_lock, flags); | |
1080 | ||
1081 | /* | |
1082 | * If command is tagged, release the tag | |
1083 | */ | |
867d1191 TH |
1084 | if (blk_rq_tagged(rq)) |
1085 | blk_queue_end_tag(q, rq); | |
1da177e4 | 1086 | |
b238b3d4 | 1087 | drive_stat_acct(rq, 1); |
867d1191 | 1088 | __elv_add_request(q, rq, where, 0); |
1654e741 | 1089 | __blk_run_queue(q, false); |
1da177e4 LT |
1090 | spin_unlock_irqrestore(q->queue_lock, flags); |
1091 | } | |
1da177e4 LT |
1092 | EXPORT_SYMBOL(blk_insert_request); |
1093 | ||
074a7aca TH |
1094 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1095 | unsigned long now) | |
1096 | { | |
1097 | if (now == part->stamp) | |
1098 | return; | |
1099 | ||
316d315b | 1100 | if (part_in_flight(part)) { |
074a7aca | 1101 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1102 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1103 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1104 | } | |
1105 | part->stamp = now; | |
1106 | } | |
1107 | ||
1108 | /** | |
496aa8a9 RD |
1109 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1110 | * @cpu: cpu number for stats access | |
1111 | * @part: target partition | |
1da177e4 LT |
1112 | * |
1113 | * The average IO queue length and utilisation statistics are maintained | |
1114 | * by observing the current state of the queue length and the amount of | |
1115 | * time it has been in this state for. | |
1116 | * | |
1117 | * Normally, that accounting is done on IO completion, but that can result | |
1118 | * in more than a second's worth of IO being accounted for within any one | |
1119 | * second, leading to >100% utilisation. To deal with that, we call this | |
1120 | * function to do a round-off before returning the results when reading | |
1121 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1122 | * the current jiffies and restarts the counters again. | |
1123 | */ | |
c9959059 | 1124 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1125 | { |
1126 | unsigned long now = jiffies; | |
1127 | ||
074a7aca TH |
1128 | if (part->partno) |
1129 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1130 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1131 | } |
074a7aca | 1132 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1133 | |
1da177e4 LT |
1134 | /* |
1135 | * queue lock must be held | |
1136 | */ | |
165125e1 | 1137 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1138 | { |
1da177e4 LT |
1139 | if (unlikely(!q)) |
1140 | return; | |
1141 | if (unlikely(--req->ref_count)) | |
1142 | return; | |
1143 | ||
8922e16c TH |
1144 | elv_completed_request(q, req); |
1145 | ||
1cd96c24 BH |
1146 | /* this is a bio leak */ |
1147 | WARN_ON(req->bio != NULL); | |
1148 | ||
1da177e4 LT |
1149 | /* |
1150 | * Request may not have originated from ll_rw_blk. if not, | |
1151 | * it didn't come out of our reserved rq pools | |
1152 | */ | |
49171e5c | 1153 | if (req->cmd_flags & REQ_ALLOCED) { |
1faa16d2 | 1154 | int is_sync = rq_is_sync(req) != 0; |
4aff5e23 | 1155 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1156 | |
1da177e4 | 1157 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1158 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1159 | |
1160 | blk_free_request(q, req); | |
1faa16d2 | 1161 | freed_request(q, is_sync, priv); |
1da177e4 LT |
1162 | } |
1163 | } | |
6e39b69e MC |
1164 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1165 | ||
1da177e4 LT |
1166 | void blk_put_request(struct request *req) |
1167 | { | |
8922e16c | 1168 | unsigned long flags; |
165125e1 | 1169 | struct request_queue *q = req->q; |
8922e16c | 1170 | |
52a93ba8 FT |
1171 | spin_lock_irqsave(q->queue_lock, flags); |
1172 | __blk_put_request(q, req); | |
1173 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1174 | } |
1da177e4 LT |
1175 | EXPORT_SYMBOL(blk_put_request); |
1176 | ||
66ac0280 CH |
1177 | /** |
1178 | * blk_add_request_payload - add a payload to a request | |
1179 | * @rq: request to update | |
1180 | * @page: page backing the payload | |
1181 | * @len: length of the payload. | |
1182 | * | |
1183 | * This allows to later add a payload to an already submitted request by | |
1184 | * a block driver. The driver needs to take care of freeing the payload | |
1185 | * itself. | |
1186 | * | |
1187 | * Note that this is a quite horrible hack and nothing but handling of | |
1188 | * discard requests should ever use it. | |
1189 | */ | |
1190 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1191 | unsigned int len) | |
1192 | { | |
1193 | struct bio *bio = rq->bio; | |
1194 | ||
1195 | bio->bi_io_vec->bv_page = page; | |
1196 | bio->bi_io_vec->bv_offset = 0; | |
1197 | bio->bi_io_vec->bv_len = len; | |
1198 | ||
1199 | bio->bi_size = len; | |
1200 | bio->bi_vcnt = 1; | |
1201 | bio->bi_phys_segments = 1; | |
1202 | ||
1203 | rq->__data_len = rq->resid_len = len; | |
1204 | rq->nr_phys_segments = 1; | |
1205 | rq->buffer = bio_data(bio); | |
1206 | } | |
1207 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1208 | ||
86db1e29 | 1209 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1210 | { |
c7c22e4d | 1211 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1212 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1213 | |
7b6d91da CH |
1214 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1215 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1216 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1217 | |
52d9e675 | 1218 | req->errors = 0; |
a2dec7b3 | 1219 | req->__sector = bio->bi_sector; |
52d9e675 | 1220 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1221 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1222 | } |
1223 | ||
644b2d99 JA |
1224 | /* |
1225 | * Only disabling plugging for non-rotational devices if it does tagging | |
1226 | * as well, otherwise we do need the proper merging | |
1227 | */ | |
1228 | static inline bool queue_should_plug(struct request_queue *q) | |
1229 | { | |
79da0644 | 1230 | return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); |
644b2d99 JA |
1231 | } |
1232 | ||
165125e1 | 1233 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1234 | { |
450991bc | 1235 | struct request *req; |
2e46e8b2 TH |
1236 | int el_ret; |
1237 | unsigned int bytes = bio->bi_size; | |
51da90fc | 1238 | const unsigned short prio = bio_prio(bio); |
5e00d1b5 JS |
1239 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
1240 | const bool unplug = !!(bio->bi_rw & REQ_UNPLUG); | |
1241 | const unsigned long ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
28e7d184 | 1242 | int where = ELEVATOR_INSERT_SORT; |
7749a8d4 | 1243 | int rw_flags; |
1da177e4 | 1244 | |
1da177e4 LT |
1245 | /* |
1246 | * low level driver can indicate that it wants pages above a | |
1247 | * certain limit bounced to low memory (ie for highmem, or even | |
1248 | * ISA dma in theory) | |
1249 | */ | |
1250 | blk_queue_bounce(q, &bio); | |
1251 | ||
1da177e4 LT |
1252 | spin_lock_irq(q->queue_lock); |
1253 | ||
4fed947c | 1254 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
ae1b1539 | 1255 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1256 | goto get_rq; |
1257 | } | |
1258 | ||
1259 | if (elv_queue_empty(q)) | |
1da177e4 LT |
1260 | goto get_rq; |
1261 | ||
1262 | el_ret = elv_merge(q, &req, bio); | |
1263 | switch (el_ret) { | |
6728cb0e JA |
1264 | case ELEVATOR_BACK_MERGE: |
1265 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1266 | |
6728cb0e JA |
1267 | if (!ll_back_merge_fn(q, req, bio)) |
1268 | break; | |
1da177e4 | 1269 | |
5f3ea37c | 1270 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1271 | |
80a761fd TH |
1272 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
1273 | blk_rq_set_mixed_merge(req); | |
1274 | ||
6728cb0e JA |
1275 | req->biotail->bi_next = bio; |
1276 | req->biotail = bio; | |
a2dec7b3 | 1277 | req->__data_len += bytes; |
6728cb0e | 1278 | req->ioprio = ioprio_best(req->ioprio, prio); |
ab780f1e JA |
1279 | if (!blk_rq_cpu_valid(req)) |
1280 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e | 1281 | drive_stat_acct(req, 0); |
812d4026 | 1282 | elv_bio_merged(q, req, bio); |
6728cb0e JA |
1283 | if (!attempt_back_merge(q, req)) |
1284 | elv_merged_request(q, req, el_ret); | |
1285 | goto out; | |
1da177e4 | 1286 | |
6728cb0e JA |
1287 | case ELEVATOR_FRONT_MERGE: |
1288 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1289 | |
6728cb0e JA |
1290 | if (!ll_front_merge_fn(q, req, bio)) |
1291 | break; | |
1da177e4 | 1292 | |
5f3ea37c | 1293 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1294 | |
80a761fd TH |
1295 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) { |
1296 | blk_rq_set_mixed_merge(req); | |
1297 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
1298 | req->cmd_flags |= ff; | |
1299 | } | |
1300 | ||
6728cb0e JA |
1301 | bio->bi_next = req->bio; |
1302 | req->bio = bio; | |
1da177e4 | 1303 | |
6728cb0e JA |
1304 | /* |
1305 | * may not be valid. if the low level driver said | |
1306 | * it didn't need a bounce buffer then it better | |
1307 | * not touch req->buffer either... | |
1308 | */ | |
1309 | req->buffer = bio_data(bio); | |
a2dec7b3 TH |
1310 | req->__sector = bio->bi_sector; |
1311 | req->__data_len += bytes; | |
6728cb0e | 1312 | req->ioprio = ioprio_best(req->ioprio, prio); |
ab780f1e JA |
1313 | if (!blk_rq_cpu_valid(req)) |
1314 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e | 1315 | drive_stat_acct(req, 0); |
812d4026 | 1316 | elv_bio_merged(q, req, bio); |
6728cb0e JA |
1317 | if (!attempt_front_merge(q, req)) |
1318 | elv_merged_request(q, req, el_ret); | |
1319 | goto out; | |
1320 | ||
1321 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1322 | default: | |
1323 | ; | |
1da177e4 LT |
1324 | } |
1325 | ||
450991bc | 1326 | get_rq: |
7749a8d4 JA |
1327 | /* |
1328 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1329 | * but we need to set it earlier to expose the sync flag to the | |
1330 | * rq allocator and io schedulers. | |
1331 | */ | |
1332 | rw_flags = bio_data_dir(bio); | |
1333 | if (sync) | |
7b6d91da | 1334 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1335 | |
1da177e4 | 1336 | /* |
450991bc | 1337 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1338 | * Returns with the queue unlocked. |
450991bc | 1339 | */ |
7749a8d4 | 1340 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1341 | |
450991bc NP |
1342 | /* |
1343 | * After dropping the lock and possibly sleeping here, our request | |
1344 | * may now be mergeable after it had proven unmergeable (above). | |
1345 | * We don't worry about that case for efficiency. It won't happen | |
1346 | * often, and the elevators are able to handle it. | |
1da177e4 | 1347 | */ |
52d9e675 | 1348 | init_request_from_bio(req, bio); |
1da177e4 | 1349 | |
450991bc | 1350 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1351 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1352 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1353 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
644b2d99 | 1354 | if (queue_should_plug(q) && elv_queue_empty(q)) |
450991bc | 1355 | blk_plug_device(q); |
dd831006 TH |
1356 | |
1357 | /* insert the request into the elevator */ | |
1358 | drive_stat_acct(req, 1); | |
28e7d184 | 1359 | __elv_add_request(q, req, where, 0); |
1da177e4 | 1360 | out: |
644b2d99 | 1361 | if (unplug || !queue_should_plug(q)) |
1da177e4 | 1362 | __generic_unplug_device(q); |
1da177e4 LT |
1363 | spin_unlock_irq(q->queue_lock); |
1364 | return 0; | |
1da177e4 LT |
1365 | } |
1366 | ||
1367 | /* | |
1368 | * If bio->bi_dev is a partition, remap the location | |
1369 | */ | |
1370 | static inline void blk_partition_remap(struct bio *bio) | |
1371 | { | |
1372 | struct block_device *bdev = bio->bi_bdev; | |
1373 | ||
bf2de6f5 | 1374 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1375 | struct hd_struct *p = bdev->bd_part; |
1376 | ||
1da177e4 LT |
1377 | bio->bi_sector += p->start_sect; |
1378 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1379 | |
d07335e5 MS |
1380 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1381 | bdev->bd_dev, | |
1382 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1383 | } |
1384 | } | |
1385 | ||
1da177e4 LT |
1386 | static void handle_bad_sector(struct bio *bio) |
1387 | { | |
1388 | char b[BDEVNAME_SIZE]; | |
1389 | ||
1390 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1391 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1392 | bdevname(bio->bi_bdev, b), | |
1393 | bio->bi_rw, | |
1394 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1395 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1396 | |
1397 | set_bit(BIO_EOF, &bio->bi_flags); | |
1398 | } | |
1399 | ||
c17bb495 AM |
1400 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1401 | ||
1402 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1403 | ||
1404 | static int __init setup_fail_make_request(char *str) | |
1405 | { | |
1406 | return setup_fault_attr(&fail_make_request, str); | |
1407 | } | |
1408 | __setup("fail_make_request=", setup_fail_make_request); | |
1409 | ||
1410 | static int should_fail_request(struct bio *bio) | |
1411 | { | |
eddb2e26 TH |
1412 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1413 | ||
1414 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1415 | return should_fail(&fail_make_request, bio->bi_size); |
1416 | ||
1417 | return 0; | |
1418 | } | |
1419 | ||
1420 | static int __init fail_make_request_debugfs(void) | |
1421 | { | |
1422 | return init_fault_attr_dentries(&fail_make_request, | |
1423 | "fail_make_request"); | |
1424 | } | |
1425 | ||
1426 | late_initcall(fail_make_request_debugfs); | |
1427 | ||
1428 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1429 | ||
1430 | static inline int should_fail_request(struct bio *bio) | |
1431 | { | |
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1436 | ||
c07e2b41 JA |
1437 | /* |
1438 | * Check whether this bio extends beyond the end of the device. | |
1439 | */ | |
1440 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1441 | { | |
1442 | sector_t maxsector; | |
1443 | ||
1444 | if (!nr_sectors) | |
1445 | return 0; | |
1446 | ||
1447 | /* Test device or partition size, when known. */ | |
77304d2a | 1448 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1449 | if (maxsector) { |
1450 | sector_t sector = bio->bi_sector; | |
1451 | ||
1452 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1453 | /* | |
1454 | * This may well happen - the kernel calls bread() | |
1455 | * without checking the size of the device, e.g., when | |
1456 | * mounting a device. | |
1457 | */ | |
1458 | handle_bad_sector(bio); | |
1459 | return 1; | |
1460 | } | |
1461 | } | |
1462 | ||
1463 | return 0; | |
1464 | } | |
1465 | ||
1da177e4 | 1466 | /** |
710027a4 | 1467 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1468 | * @bio: The bio describing the location in memory and on the device. |
1469 | * | |
1470 | * generic_make_request() is used to make I/O requests of block | |
1471 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1472 | * to be done. | |
1473 | * | |
1474 | * generic_make_request() does not return any status. The | |
1475 | * success/failure status of the request, along with notification of | |
1476 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1477 | * function described (one day) else where. | |
1478 | * | |
1479 | * The caller of generic_make_request must make sure that bi_io_vec | |
1480 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1481 | * set to describe the device address, and the | |
1482 | * bi_end_io and optionally bi_private are set to describe how | |
1483 | * completion notification should be signaled. | |
1484 | * | |
1485 | * generic_make_request and the drivers it calls may use bi_next if this | |
1486 | * bio happens to be merged with someone else, and may change bi_dev and | |
1487 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1488 | * should NOT be depended on after the call to generic_make_request. | |
1489 | */ | |
d89d8796 | 1490 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1491 | { |
165125e1 | 1492 | struct request_queue *q; |
5ddfe969 | 1493 | sector_t old_sector; |
1da177e4 | 1494 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1495 | dev_t old_dev; |
51fd77bd | 1496 | int err = -EIO; |
1da177e4 LT |
1497 | |
1498 | might_sleep(); | |
1da177e4 | 1499 | |
c07e2b41 JA |
1500 | if (bio_check_eod(bio, nr_sectors)) |
1501 | goto end_io; | |
1da177e4 LT |
1502 | |
1503 | /* | |
1504 | * Resolve the mapping until finished. (drivers are | |
1505 | * still free to implement/resolve their own stacking | |
1506 | * by explicitly returning 0) | |
1507 | * | |
1508 | * NOTE: we don't repeat the blk_size check for each new device. | |
1509 | * Stacking drivers are expected to know what they are doing. | |
1510 | */ | |
5ddfe969 | 1511 | old_sector = -1; |
2056a782 | 1512 | old_dev = 0; |
1da177e4 LT |
1513 | do { |
1514 | char b[BDEVNAME_SIZE]; | |
1515 | ||
1516 | q = bdev_get_queue(bio->bi_bdev); | |
a7384677 | 1517 | if (unlikely(!q)) { |
1da177e4 LT |
1518 | printk(KERN_ERR |
1519 | "generic_make_request: Trying to access " | |
1520 | "nonexistent block-device %s (%Lu)\n", | |
1521 | bdevname(bio->bi_bdev, b), | |
1522 | (long long) bio->bi_sector); | |
a7384677 | 1523 | goto end_io; |
1da177e4 LT |
1524 | } |
1525 | ||
7b6d91da | 1526 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
67efc925 | 1527 | nr_sectors > queue_max_hw_sectors(q))) { |
6728cb0e | 1528 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
ae03bf63 MP |
1529 | bdevname(bio->bi_bdev, b), |
1530 | bio_sectors(bio), | |
1531 | queue_max_hw_sectors(q)); | |
1da177e4 LT |
1532 | goto end_io; |
1533 | } | |
1534 | ||
fde6ad22 | 1535 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1536 | goto end_io; |
1537 | ||
c17bb495 AM |
1538 | if (should_fail_request(bio)) |
1539 | goto end_io; | |
1540 | ||
1da177e4 LT |
1541 | /* |
1542 | * If this device has partitions, remap block n | |
1543 | * of partition p to block n+start(p) of the disk. | |
1544 | */ | |
1545 | blk_partition_remap(bio); | |
1546 | ||
7ba1ba12 MP |
1547 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1548 | goto end_io; | |
1549 | ||
5ddfe969 | 1550 | if (old_sector != -1) |
d07335e5 | 1551 | trace_block_bio_remap(q, bio, old_dev, old_sector); |
2056a782 | 1552 | |
5ddfe969 | 1553 | old_sector = bio->bi_sector; |
2056a782 JA |
1554 | old_dev = bio->bi_bdev->bd_dev; |
1555 | ||
c07e2b41 JA |
1556 | if (bio_check_eod(bio, nr_sectors)) |
1557 | goto end_io; | |
a7384677 | 1558 | |
1e87901e TH |
1559 | /* |
1560 | * Filter flush bio's early so that make_request based | |
1561 | * drivers without flush support don't have to worry | |
1562 | * about them. | |
1563 | */ | |
1564 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1565 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1566 | if (!nr_sectors) { | |
1567 | err = 0; | |
1568 | goto end_io; | |
1569 | } | |
1570 | } | |
1571 | ||
8d57a98c AH |
1572 | if ((bio->bi_rw & REQ_DISCARD) && |
1573 | (!blk_queue_discard(q) || | |
1574 | ((bio->bi_rw & REQ_SECURE) && | |
1575 | !blk_queue_secdiscard(q)))) { | |
51fd77bd JA |
1576 | err = -EOPNOTSUPP; |
1577 | goto end_io; | |
1578 | } | |
5ddfe969 | 1579 | |
e43473b7 VG |
1580 | blk_throtl_bio(q, &bio); |
1581 | ||
1582 | /* | |
1583 | * If bio = NULL, bio has been throttled and will be submitted | |
1584 | * later. | |
1585 | */ | |
1586 | if (!bio) | |
1587 | break; | |
1588 | ||
01edede4 MK |
1589 | trace_block_bio_queue(q, bio); |
1590 | ||
1da177e4 LT |
1591 | ret = q->make_request_fn(q, bio); |
1592 | } while (ret); | |
a7384677 TH |
1593 | |
1594 | return; | |
1595 | ||
1596 | end_io: | |
1597 | bio_endio(bio, err); | |
1da177e4 LT |
1598 | } |
1599 | ||
d89d8796 NB |
1600 | /* |
1601 | * We only want one ->make_request_fn to be active at a time, | |
1602 | * else stack usage with stacked devices could be a problem. | |
bddd87c7 | 1603 | * So use current->bio_list to keep a list of requests |
d89d8796 | 1604 | * submited by a make_request_fn function. |
bddd87c7 | 1605 | * current->bio_list is also used as a flag to say if |
d89d8796 NB |
1606 | * generic_make_request is currently active in this task or not. |
1607 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1608 | * then a make_request is active, and new requests should be added | |
1609 | * at the tail | |
1610 | */ | |
1611 | void generic_make_request(struct bio *bio) | |
1612 | { | |
bddd87c7 AM |
1613 | struct bio_list bio_list_on_stack; |
1614 | ||
1615 | if (current->bio_list) { | |
d89d8796 | 1616 | /* make_request is active */ |
bddd87c7 | 1617 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1618 | return; |
1619 | } | |
1620 | /* following loop may be a bit non-obvious, and so deserves some | |
1621 | * explanation. | |
1622 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1623 | * ensure that) so we have a list with a single bio. | |
1624 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1625 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1626 | * thus initialising the bio_list of new bios to be | |
d89d8796 NB |
1627 | * added. __generic_make_request may indeed add some more bios |
1628 | * through a recursive call to generic_make_request. If it | |
1629 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1630 | * from the top. In this case we really did just take the bio | |
bddd87c7 AM |
1631 | * of the top of the list (no pretending) and so remove it from |
1632 | * bio_list, and call into __generic_make_request again. | |
d89d8796 NB |
1633 | * |
1634 | * The loop was structured like this to make only one call to | |
1635 | * __generic_make_request (which is important as it is large and | |
1636 | * inlined) and to keep the structure simple. | |
1637 | */ | |
1638 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1639 | bio_list_init(&bio_list_on_stack); |
1640 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1641 | do { |
d89d8796 | 1642 | __generic_make_request(bio); |
bddd87c7 | 1643 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1644 | } while (bio); |
bddd87c7 | 1645 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1646 | } |
1da177e4 LT |
1647 | EXPORT_SYMBOL(generic_make_request); |
1648 | ||
1649 | /** | |
710027a4 | 1650 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1651 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1652 | * @bio: The &struct bio which describes the I/O | |
1653 | * | |
1654 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1655 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1656 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1657 | * |
1658 | */ | |
1659 | void submit_bio(int rw, struct bio *bio) | |
1660 | { | |
1661 | int count = bio_sectors(bio); | |
1662 | ||
22e2c507 | 1663 | bio->bi_rw |= rw; |
1da177e4 | 1664 | |
bf2de6f5 JA |
1665 | /* |
1666 | * If it's a regular read/write or a barrier with data attached, | |
1667 | * go through the normal accounting stuff before submission. | |
1668 | */ | |
3ffb52e7 | 1669 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1670 | if (rw & WRITE) { |
1671 | count_vm_events(PGPGOUT, count); | |
1672 | } else { | |
1673 | task_io_account_read(bio->bi_size); | |
1674 | count_vm_events(PGPGIN, count); | |
1675 | } | |
1676 | ||
1677 | if (unlikely(block_dump)) { | |
1678 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1679 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1680 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1681 | (rw & WRITE) ? "WRITE" : "READ", |
1682 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1683 | bdevname(bio->bi_bdev, b), |
1684 | count); | |
bf2de6f5 | 1685 | } |
1da177e4 LT |
1686 | } |
1687 | ||
1688 | generic_make_request(bio); | |
1689 | } | |
1da177e4 LT |
1690 | EXPORT_SYMBOL(submit_bio); |
1691 | ||
82124d60 KU |
1692 | /** |
1693 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1694 | * @q: the queue | |
1695 | * @rq: the request being checked | |
1696 | * | |
1697 | * Description: | |
1698 | * @rq may have been made based on weaker limitations of upper-level queues | |
1699 | * in request stacking drivers, and it may violate the limitation of @q. | |
1700 | * Since the block layer and the underlying device driver trust @rq | |
1701 | * after it is inserted to @q, it should be checked against @q before | |
1702 | * the insertion using this generic function. | |
1703 | * | |
1704 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1705 | * in some cases below, so export this function. |
82124d60 KU |
1706 | * Request stacking drivers like request-based dm may change the queue |
1707 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1708 | * Such request stacking drivers should check those requests agaist | |
1709 | * the new queue limits again when they dispatch those requests, | |
1710 | * although such checkings are also done against the old queue limits | |
1711 | * when submitting requests. | |
1712 | */ | |
1713 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1714 | { | |
3383977f S |
1715 | if (rq->cmd_flags & REQ_DISCARD) |
1716 | return 0; | |
1717 | ||
ae03bf63 MP |
1718 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1719 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1720 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1721 | return -EIO; | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * queue's settings related to segment counting like q->bounce_pfn | |
1726 | * may differ from that of other stacking queues. | |
1727 | * Recalculate it to check the request correctly on this queue's | |
1728 | * limitation. | |
1729 | */ | |
1730 | blk_recalc_rq_segments(rq); | |
8a78362c | 1731 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1732 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1733 | return -EIO; | |
1734 | } | |
1735 | ||
1736 | return 0; | |
1737 | } | |
1738 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1739 | ||
1740 | /** | |
1741 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1742 | * @q: the queue to submit the request | |
1743 | * @rq: the request being queued | |
1744 | */ | |
1745 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1746 | { | |
1747 | unsigned long flags; | |
1748 | ||
1749 | if (blk_rq_check_limits(q, rq)) | |
1750 | return -EIO; | |
1751 | ||
1752 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1753 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1754 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1755 | return -EIO; | |
1756 | #endif | |
1757 | ||
1758 | spin_lock_irqsave(q->queue_lock, flags); | |
1759 | ||
1760 | /* | |
1761 | * Submitting request must be dequeued before calling this function | |
1762 | * because it will be linked to another request_queue | |
1763 | */ | |
1764 | BUG_ON(blk_queued_rq(rq)); | |
1765 | ||
1766 | drive_stat_acct(rq, 1); | |
1767 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1768 | ||
1769 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1770 | ||
1771 | return 0; | |
1772 | } | |
1773 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1774 | ||
80a761fd TH |
1775 | /** |
1776 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1777 | * @rq: request to examine | |
1778 | * | |
1779 | * Description: | |
1780 | * A request could be merge of IOs which require different failure | |
1781 | * handling. This function determines the number of bytes which | |
1782 | * can be failed from the beginning of the request without | |
1783 | * crossing into area which need to be retried further. | |
1784 | * | |
1785 | * Return: | |
1786 | * The number of bytes to fail. | |
1787 | * | |
1788 | * Context: | |
1789 | * queue_lock must be held. | |
1790 | */ | |
1791 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1792 | { | |
1793 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1794 | unsigned int bytes = 0; | |
1795 | struct bio *bio; | |
1796 | ||
1797 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1798 | return blk_rq_bytes(rq); | |
1799 | ||
1800 | /* | |
1801 | * Currently the only 'mixing' which can happen is between | |
1802 | * different fastfail types. We can safely fail portions | |
1803 | * which have all the failfast bits that the first one has - | |
1804 | * the ones which are at least as eager to fail as the first | |
1805 | * one. | |
1806 | */ | |
1807 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1808 | if ((bio->bi_rw & ff) != ff) | |
1809 | break; | |
1810 | bytes += bio->bi_size; | |
1811 | } | |
1812 | ||
1813 | /* this could lead to infinite loop */ | |
1814 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1815 | return bytes; | |
1816 | } | |
1817 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1818 | ||
bc58ba94 JA |
1819 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1820 | { | |
c2553b58 | 1821 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1822 | const int rw = rq_data_dir(req); |
1823 | struct hd_struct *part; | |
1824 | int cpu; | |
1825 | ||
1826 | cpu = part_stat_lock(); | |
09e099d4 | 1827 | part = req->part; |
bc58ba94 JA |
1828 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1829 | part_stat_unlock(); | |
1830 | } | |
1831 | } | |
1832 | ||
1833 | static void blk_account_io_done(struct request *req) | |
1834 | { | |
bc58ba94 | 1835 | /* |
dd4c133f TH |
1836 | * Account IO completion. flush_rq isn't accounted as a |
1837 | * normal IO on queueing nor completion. Accounting the | |
1838 | * containing request is enough. | |
bc58ba94 | 1839 | */ |
414b4ff5 | 1840 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1841 | unsigned long duration = jiffies - req->start_time; |
1842 | const int rw = rq_data_dir(req); | |
1843 | struct hd_struct *part; | |
1844 | int cpu; | |
1845 | ||
1846 | cpu = part_stat_lock(); | |
09e099d4 | 1847 | part = req->part; |
bc58ba94 JA |
1848 | |
1849 | part_stat_inc(cpu, part, ios[rw]); | |
1850 | part_stat_add(cpu, part, ticks[rw], duration); | |
1851 | part_round_stats(cpu, part); | |
316d315b | 1852 | part_dec_in_flight(part, rw); |
bc58ba94 | 1853 | |
6c23a968 | 1854 | hd_struct_put(part); |
bc58ba94 JA |
1855 | part_stat_unlock(); |
1856 | } | |
1857 | } | |
1858 | ||
3bcddeac | 1859 | /** |
9934c8c0 TH |
1860 | * blk_peek_request - peek at the top of a request queue |
1861 | * @q: request queue to peek at | |
1862 | * | |
1863 | * Description: | |
1864 | * Return the request at the top of @q. The returned request | |
1865 | * should be started using blk_start_request() before LLD starts | |
1866 | * processing it. | |
1867 | * | |
1868 | * Return: | |
1869 | * Pointer to the request at the top of @q if available. Null | |
1870 | * otherwise. | |
1871 | * | |
1872 | * Context: | |
1873 | * queue_lock must be held. | |
1874 | */ | |
1875 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
1876 | { |
1877 | struct request *rq; | |
1878 | int ret; | |
1879 | ||
1880 | while ((rq = __elv_next_request(q)) != NULL) { | |
1881 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1882 | /* | |
1883 | * This is the first time the device driver | |
1884 | * sees this request (possibly after | |
1885 | * requeueing). Notify IO scheduler. | |
1886 | */ | |
33659ebb | 1887 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
1888 | elv_activate_rq(q, rq); |
1889 | ||
1890 | /* | |
1891 | * just mark as started even if we don't start | |
1892 | * it, a request that has been delayed should | |
1893 | * not be passed by new incoming requests | |
1894 | */ | |
1895 | rq->cmd_flags |= REQ_STARTED; | |
1896 | trace_block_rq_issue(q, rq); | |
1897 | } | |
1898 | ||
1899 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1900 | q->end_sector = rq_end_sector(rq); | |
1901 | q->boundary_rq = NULL; | |
1902 | } | |
1903 | ||
1904 | if (rq->cmd_flags & REQ_DONTPREP) | |
1905 | break; | |
1906 | ||
2e46e8b2 | 1907 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
1908 | /* |
1909 | * make sure space for the drain appears we | |
1910 | * know we can do this because max_hw_segments | |
1911 | * has been adjusted to be one fewer than the | |
1912 | * device can handle | |
1913 | */ | |
1914 | rq->nr_phys_segments++; | |
1915 | } | |
1916 | ||
1917 | if (!q->prep_rq_fn) | |
1918 | break; | |
1919 | ||
1920 | ret = q->prep_rq_fn(q, rq); | |
1921 | if (ret == BLKPREP_OK) { | |
1922 | break; | |
1923 | } else if (ret == BLKPREP_DEFER) { | |
1924 | /* | |
1925 | * the request may have been (partially) prepped. | |
1926 | * we need to keep this request in the front to | |
1927 | * avoid resource deadlock. REQ_STARTED will | |
1928 | * prevent other fs requests from passing this one. | |
1929 | */ | |
2e46e8b2 | 1930 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
1931 | !(rq->cmd_flags & REQ_DONTPREP)) { |
1932 | /* | |
1933 | * remove the space for the drain we added | |
1934 | * so that we don't add it again | |
1935 | */ | |
1936 | --rq->nr_phys_segments; | |
1937 | } | |
1938 | ||
1939 | rq = NULL; | |
1940 | break; | |
1941 | } else if (ret == BLKPREP_KILL) { | |
1942 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
1943 | /* |
1944 | * Mark this request as started so we don't trigger | |
1945 | * any debug logic in the end I/O path. | |
1946 | */ | |
1947 | blk_start_request(rq); | |
40cbbb78 | 1948 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
1949 | } else { |
1950 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1951 | break; | |
1952 | } | |
1953 | } | |
1954 | ||
1955 | return rq; | |
1956 | } | |
9934c8c0 | 1957 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 1958 | |
9934c8c0 | 1959 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 1960 | { |
9934c8c0 TH |
1961 | struct request_queue *q = rq->q; |
1962 | ||
158dbda0 TH |
1963 | BUG_ON(list_empty(&rq->queuelist)); |
1964 | BUG_ON(ELV_ON_HASH(rq)); | |
1965 | ||
1966 | list_del_init(&rq->queuelist); | |
1967 | ||
1968 | /* | |
1969 | * the time frame between a request being removed from the lists | |
1970 | * and to it is freed is accounted as io that is in progress at | |
1971 | * the driver side. | |
1972 | */ | |
9195291e | 1973 | if (blk_account_rq(rq)) { |
0a7ae2ff | 1974 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
1975 | set_io_start_time_ns(rq); |
1976 | } | |
158dbda0 TH |
1977 | } |
1978 | ||
9934c8c0 TH |
1979 | /** |
1980 | * blk_start_request - start request processing on the driver | |
1981 | * @req: request to dequeue | |
1982 | * | |
1983 | * Description: | |
1984 | * Dequeue @req and start timeout timer on it. This hands off the | |
1985 | * request to the driver. | |
1986 | * | |
1987 | * Block internal functions which don't want to start timer should | |
1988 | * call blk_dequeue_request(). | |
1989 | * | |
1990 | * Context: | |
1991 | * queue_lock must be held. | |
1992 | */ | |
1993 | void blk_start_request(struct request *req) | |
1994 | { | |
1995 | blk_dequeue_request(req); | |
1996 | ||
1997 | /* | |
5f49f631 TH |
1998 | * We are now handing the request to the hardware, initialize |
1999 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 2000 | */ |
5f49f631 | 2001 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
2002 | if (unlikely(blk_bidi_rq(req))) |
2003 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
2004 | ||
9934c8c0 TH |
2005 | blk_add_timer(req); |
2006 | } | |
2007 | EXPORT_SYMBOL(blk_start_request); | |
2008 | ||
2009 | /** | |
2010 | * blk_fetch_request - fetch a request from a request queue | |
2011 | * @q: request queue to fetch a request from | |
2012 | * | |
2013 | * Description: | |
2014 | * Return the request at the top of @q. The request is started on | |
2015 | * return and LLD can start processing it immediately. | |
2016 | * | |
2017 | * Return: | |
2018 | * Pointer to the request at the top of @q if available. Null | |
2019 | * otherwise. | |
2020 | * | |
2021 | * Context: | |
2022 | * queue_lock must be held. | |
2023 | */ | |
2024 | struct request *blk_fetch_request(struct request_queue *q) | |
2025 | { | |
2026 | struct request *rq; | |
2027 | ||
2028 | rq = blk_peek_request(q); | |
2029 | if (rq) | |
2030 | blk_start_request(rq); | |
2031 | return rq; | |
2032 | } | |
2033 | EXPORT_SYMBOL(blk_fetch_request); | |
2034 | ||
3bcddeac | 2035 | /** |
2e60e022 | 2036 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2037 | * @req: the request being processed |
710027a4 | 2038 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2039 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2040 | * |
2041 | * Description: | |
8ebf9756 RD |
2042 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2043 | * the request structure even if @req doesn't have leftover. | |
2044 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2045 | * |
2046 | * This special helper function is only for request stacking drivers | |
2047 | * (e.g. request-based dm) so that they can handle partial completion. | |
2048 | * Actual device drivers should use blk_end_request instead. | |
2049 | * | |
2050 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2051 | * %false return from this function. | |
3bcddeac KU |
2052 | * |
2053 | * Return: | |
2e60e022 TH |
2054 | * %false - this request doesn't have any more data |
2055 | * %true - this request has more data | |
3bcddeac | 2056 | **/ |
2e60e022 | 2057 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2058 | { |
5450d3e1 | 2059 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2060 | struct bio *bio; |
2061 | ||
2e60e022 TH |
2062 | if (!req->bio) |
2063 | return false; | |
2064 | ||
5f3ea37c | 2065 | trace_block_rq_complete(req->q, req); |
2056a782 | 2066 | |
1da177e4 | 2067 | /* |
6f41469c TH |
2068 | * For fs requests, rq is just carrier of independent bio's |
2069 | * and each partial completion should be handled separately. | |
2070 | * Reset per-request error on each partial completion. | |
2071 | * | |
2072 | * TODO: tj: This is too subtle. It would be better to let | |
2073 | * low level drivers do what they see fit. | |
1da177e4 | 2074 | */ |
33659ebb | 2075 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2076 | req->errors = 0; |
2077 | ||
33659ebb CH |
2078 | if (error && req->cmd_type == REQ_TYPE_FS && |
2079 | !(req->cmd_flags & REQ_QUIET)) { | |
6728cb0e | 2080 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", |
1da177e4 | 2081 | req->rq_disk ? req->rq_disk->disk_name : "?", |
83096ebf | 2082 | (unsigned long long)blk_rq_pos(req)); |
1da177e4 LT |
2083 | } |
2084 | ||
bc58ba94 | 2085 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2086 | |
1da177e4 LT |
2087 | total_bytes = bio_nbytes = 0; |
2088 | while ((bio = req->bio) != NULL) { | |
2089 | int nbytes; | |
2090 | ||
2091 | if (nr_bytes >= bio->bi_size) { | |
2092 | req->bio = bio->bi_next; | |
2093 | nbytes = bio->bi_size; | |
5bb23a68 | 2094 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2095 | next_idx = 0; |
2096 | bio_nbytes = 0; | |
2097 | } else { | |
2098 | int idx = bio->bi_idx + next_idx; | |
2099 | ||
af498d7f | 2100 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2101 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2102 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2103 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2104 | break; |
2105 | } | |
2106 | ||
2107 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2108 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2109 | ||
2110 | /* | |
2111 | * not a complete bvec done | |
2112 | */ | |
2113 | if (unlikely(nbytes > nr_bytes)) { | |
2114 | bio_nbytes += nr_bytes; | |
2115 | total_bytes += nr_bytes; | |
2116 | break; | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * advance to the next vector | |
2121 | */ | |
2122 | next_idx++; | |
2123 | bio_nbytes += nbytes; | |
2124 | } | |
2125 | ||
2126 | total_bytes += nbytes; | |
2127 | nr_bytes -= nbytes; | |
2128 | ||
6728cb0e JA |
2129 | bio = req->bio; |
2130 | if (bio) { | |
1da177e4 LT |
2131 | /* |
2132 | * end more in this run, or just return 'not-done' | |
2133 | */ | |
2134 | if (unlikely(nr_bytes <= 0)) | |
2135 | break; | |
2136 | } | |
2137 | } | |
2138 | ||
2139 | /* | |
2140 | * completely done | |
2141 | */ | |
2e60e022 TH |
2142 | if (!req->bio) { |
2143 | /* | |
2144 | * Reset counters so that the request stacking driver | |
2145 | * can find how many bytes remain in the request | |
2146 | * later. | |
2147 | */ | |
a2dec7b3 | 2148 | req->__data_len = 0; |
2e60e022 TH |
2149 | return false; |
2150 | } | |
1da177e4 LT |
2151 | |
2152 | /* | |
2153 | * if the request wasn't completed, update state | |
2154 | */ | |
2155 | if (bio_nbytes) { | |
5bb23a68 | 2156 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2157 | bio->bi_idx += next_idx; |
2158 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2159 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2160 | } | |
2161 | ||
a2dec7b3 | 2162 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2163 | req->buffer = bio_data(req->bio); |
2164 | ||
2165 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2166 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2167 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2168 | |
80a761fd TH |
2169 | /* mixed attributes always follow the first bio */ |
2170 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2171 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2172 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2173 | } | |
2174 | ||
2e46e8b2 TH |
2175 | /* |
2176 | * If total number of sectors is less than the first segment | |
2177 | * size, something has gone terribly wrong. | |
2178 | */ | |
2179 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2180 | printk(KERN_ERR "blk: request botched\n"); | |
a2dec7b3 | 2181 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2182 | } |
2183 | ||
2184 | /* recalculate the number of segments */ | |
1da177e4 | 2185 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2186 | |
2e60e022 | 2187 | return true; |
1da177e4 | 2188 | } |
2e60e022 | 2189 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2190 | |
2e60e022 TH |
2191 | static bool blk_update_bidi_request(struct request *rq, int error, |
2192 | unsigned int nr_bytes, | |
2193 | unsigned int bidi_bytes) | |
5efccd17 | 2194 | { |
2e60e022 TH |
2195 | if (blk_update_request(rq, error, nr_bytes)) |
2196 | return true; | |
5efccd17 | 2197 | |
2e60e022 TH |
2198 | /* Bidi request must be completed as a whole */ |
2199 | if (unlikely(blk_bidi_rq(rq)) && | |
2200 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2201 | return true; | |
5efccd17 | 2202 | |
e2e1a148 JA |
2203 | if (blk_queue_add_random(rq->q)) |
2204 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2205 | |
2206 | return false; | |
1da177e4 LT |
2207 | } |
2208 | ||
28018c24 JB |
2209 | /** |
2210 | * blk_unprep_request - unprepare a request | |
2211 | * @req: the request | |
2212 | * | |
2213 | * This function makes a request ready for complete resubmission (or | |
2214 | * completion). It happens only after all error handling is complete, | |
2215 | * so represents the appropriate moment to deallocate any resources | |
2216 | * that were allocated to the request in the prep_rq_fn. The queue | |
2217 | * lock is held when calling this. | |
2218 | */ | |
2219 | void blk_unprep_request(struct request *req) | |
2220 | { | |
2221 | struct request_queue *q = req->q; | |
2222 | ||
2223 | req->cmd_flags &= ~REQ_DONTPREP; | |
2224 | if (q->unprep_rq_fn) | |
2225 | q->unprep_rq_fn(q, req); | |
2226 | } | |
2227 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2228 | ||
1da177e4 LT |
2229 | /* |
2230 | * queue lock must be held | |
2231 | */ | |
2e60e022 | 2232 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2233 | { |
b8286239 KU |
2234 | if (blk_rq_tagged(req)) |
2235 | blk_queue_end_tag(req->q, req); | |
2236 | ||
ba396a6c | 2237 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2238 | |
33659ebb | 2239 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2240 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2241 | |
e78042e5 MA |
2242 | blk_delete_timer(req); |
2243 | ||
28018c24 JB |
2244 | if (req->cmd_flags & REQ_DONTPREP) |
2245 | blk_unprep_request(req); | |
2246 | ||
2247 | ||
bc58ba94 | 2248 | blk_account_io_done(req); |
b8286239 | 2249 | |
1da177e4 | 2250 | if (req->end_io) |
8ffdc655 | 2251 | req->end_io(req, error); |
b8286239 KU |
2252 | else { |
2253 | if (blk_bidi_rq(req)) | |
2254 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2255 | ||
1da177e4 | 2256 | __blk_put_request(req->q, req); |
b8286239 | 2257 | } |
1da177e4 LT |
2258 | } |
2259 | ||
3b11313a | 2260 | /** |
2e60e022 TH |
2261 | * blk_end_bidi_request - Complete a bidi request |
2262 | * @rq: the request to complete | |
2263 | * @error: %0 for success, < %0 for error | |
2264 | * @nr_bytes: number of bytes to complete @rq | |
2265 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2266 | * |
2267 | * Description: | |
e3a04fe3 | 2268 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2269 | * Drivers that supports bidi can safely call this member for any |
2270 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2271 | * just ignored. | |
336cdb40 KU |
2272 | * |
2273 | * Return: | |
2e60e022 TH |
2274 | * %false - we are done with this request |
2275 | * %true - still buffers pending for this request | |
a0cd1285 | 2276 | **/ |
b1f74493 | 2277 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2278 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2279 | { | |
336cdb40 | 2280 | struct request_queue *q = rq->q; |
2e60e022 | 2281 | unsigned long flags; |
32fab448 | 2282 | |
2e60e022 TH |
2283 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2284 | return true; | |
32fab448 | 2285 | |
336cdb40 | 2286 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2287 | blk_finish_request(rq, error); |
336cdb40 KU |
2288 | spin_unlock_irqrestore(q->queue_lock, flags); |
2289 | ||
2e60e022 | 2290 | return false; |
32fab448 KU |
2291 | } |
2292 | ||
336cdb40 | 2293 | /** |
2e60e022 TH |
2294 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2295 | * @rq: the request to complete | |
710027a4 | 2296 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2297 | * @nr_bytes: number of bytes to complete @rq |
2298 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2299 | * |
2300 | * Description: | |
2e60e022 TH |
2301 | * Identical to blk_end_bidi_request() except that queue lock is |
2302 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2303 | * |
2304 | * Return: | |
2e60e022 TH |
2305 | * %false - we are done with this request |
2306 | * %true - still buffers pending for this request | |
336cdb40 | 2307 | **/ |
b1f74493 FT |
2308 | static bool __blk_end_bidi_request(struct request *rq, int error, |
2309 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
336cdb40 | 2310 | { |
2e60e022 TH |
2311 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2312 | return true; | |
336cdb40 | 2313 | |
2e60e022 | 2314 | blk_finish_request(rq, error); |
336cdb40 | 2315 | |
2e60e022 | 2316 | return false; |
336cdb40 | 2317 | } |
e19a3ab0 KU |
2318 | |
2319 | /** | |
2320 | * blk_end_request - Helper function for drivers to complete the request. | |
2321 | * @rq: the request being processed | |
710027a4 | 2322 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2323 | * @nr_bytes: number of bytes to complete |
2324 | * | |
2325 | * Description: | |
2326 | * Ends I/O on a number of bytes attached to @rq. | |
2327 | * If @rq has leftover, sets it up for the next range of segments. | |
2328 | * | |
2329 | * Return: | |
b1f74493 FT |
2330 | * %false - we are done with this request |
2331 | * %true - still buffers pending for this request | |
e19a3ab0 | 2332 | **/ |
b1f74493 | 2333 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2334 | { |
b1f74493 | 2335 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2336 | } |
56ad1740 | 2337 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2338 | |
2339 | /** | |
b1f74493 FT |
2340 | * blk_end_request_all - Helper function for drives to finish the request. |
2341 | * @rq: the request to finish | |
8ebf9756 | 2342 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2343 | * |
2344 | * Description: | |
b1f74493 FT |
2345 | * Completely finish @rq. |
2346 | */ | |
2347 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2348 | { |
b1f74493 FT |
2349 | bool pending; |
2350 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2351 | |
b1f74493 FT |
2352 | if (unlikely(blk_bidi_rq(rq))) |
2353 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2354 | |
b1f74493 FT |
2355 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2356 | BUG_ON(pending); | |
2357 | } | |
56ad1740 | 2358 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2359 | |
b1f74493 FT |
2360 | /** |
2361 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2362 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2363 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2364 | * |
2365 | * Description: | |
2366 | * Complete the current consecutively mapped chunk from @rq. | |
2367 | * | |
2368 | * Return: | |
2369 | * %false - we are done with this request | |
2370 | * %true - still buffers pending for this request | |
2371 | */ | |
2372 | bool blk_end_request_cur(struct request *rq, int error) | |
2373 | { | |
2374 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2375 | } |
56ad1740 | 2376 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2377 | |
80a761fd TH |
2378 | /** |
2379 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2380 | * @rq: the request to finish till the next failure boundary for | |
2381 | * @error: must be negative errno | |
2382 | * | |
2383 | * Description: | |
2384 | * Complete @rq till the next failure boundary. | |
2385 | * | |
2386 | * Return: | |
2387 | * %false - we are done with this request | |
2388 | * %true - still buffers pending for this request | |
2389 | */ | |
2390 | bool blk_end_request_err(struct request *rq, int error) | |
2391 | { | |
2392 | WARN_ON(error >= 0); | |
2393 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2394 | } | |
2395 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2396 | ||
e3a04fe3 | 2397 | /** |
b1f74493 FT |
2398 | * __blk_end_request - Helper function for drivers to complete the request. |
2399 | * @rq: the request being processed | |
2400 | * @error: %0 for success, < %0 for error | |
2401 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2402 | * |
2403 | * Description: | |
b1f74493 | 2404 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2405 | * |
2406 | * Return: | |
b1f74493 FT |
2407 | * %false - we are done with this request |
2408 | * %true - still buffers pending for this request | |
e3a04fe3 | 2409 | **/ |
b1f74493 | 2410 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2411 | { |
b1f74493 | 2412 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2413 | } |
56ad1740 | 2414 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2415 | |
32fab448 | 2416 | /** |
b1f74493 FT |
2417 | * __blk_end_request_all - Helper function for drives to finish the request. |
2418 | * @rq: the request to finish | |
8ebf9756 | 2419 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2420 | * |
2421 | * Description: | |
b1f74493 | 2422 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2423 | */ |
b1f74493 | 2424 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2425 | { |
b1f74493 FT |
2426 | bool pending; |
2427 | unsigned int bidi_bytes = 0; | |
2428 | ||
2429 | if (unlikely(blk_bidi_rq(rq))) | |
2430 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2431 | ||
2432 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2433 | BUG_ON(pending); | |
32fab448 | 2434 | } |
56ad1740 | 2435 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2436 | |
e19a3ab0 | 2437 | /** |
b1f74493 FT |
2438 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2439 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2440 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2441 | * |
2442 | * Description: | |
b1f74493 FT |
2443 | * Complete the current consecutively mapped chunk from @rq. Must |
2444 | * be called with queue lock held. | |
e19a3ab0 KU |
2445 | * |
2446 | * Return: | |
b1f74493 FT |
2447 | * %false - we are done with this request |
2448 | * %true - still buffers pending for this request | |
2449 | */ | |
2450 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2451 | { |
b1f74493 | 2452 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2453 | } |
56ad1740 | 2454 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2455 | |
80a761fd TH |
2456 | /** |
2457 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2458 | * @rq: the request to finish till the next failure boundary for | |
2459 | * @error: must be negative errno | |
2460 | * | |
2461 | * Description: | |
2462 | * Complete @rq till the next failure boundary. Must be called | |
2463 | * with queue lock held. | |
2464 | * | |
2465 | * Return: | |
2466 | * %false - we are done with this request | |
2467 | * %true - still buffers pending for this request | |
2468 | */ | |
2469 | bool __blk_end_request_err(struct request *rq, int error) | |
2470 | { | |
2471 | WARN_ON(error >= 0); | |
2472 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2473 | } | |
2474 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2475 | ||
86db1e29 JA |
2476 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2477 | struct bio *bio) | |
1da177e4 | 2478 | { |
a82afdfc | 2479 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2480 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2481 | |
fb2dce86 DW |
2482 | if (bio_has_data(bio)) { |
2483 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2484 | rq->buffer = bio_data(bio); |
2485 | } | |
a2dec7b3 | 2486 | rq->__data_len = bio->bi_size; |
1da177e4 | 2487 | rq->bio = rq->biotail = bio; |
1da177e4 | 2488 | |
66846572 N |
2489 | if (bio->bi_bdev) |
2490 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2491 | } | |
1da177e4 | 2492 | |
2d4dc890 IL |
2493 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2494 | /** | |
2495 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2496 | * @rq: the request to be flushed | |
2497 | * | |
2498 | * Description: | |
2499 | * Flush all pages in @rq. | |
2500 | */ | |
2501 | void rq_flush_dcache_pages(struct request *rq) | |
2502 | { | |
2503 | struct req_iterator iter; | |
2504 | struct bio_vec *bvec; | |
2505 | ||
2506 | rq_for_each_segment(bvec, rq, iter) | |
2507 | flush_dcache_page(bvec->bv_page); | |
2508 | } | |
2509 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2510 | #endif | |
2511 | ||
ef9e3fac KU |
2512 | /** |
2513 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2514 | * @q : the queue of the device being checked | |
2515 | * | |
2516 | * Description: | |
2517 | * Check if underlying low-level drivers of a device are busy. | |
2518 | * If the drivers want to export their busy state, they must set own | |
2519 | * exporting function using blk_queue_lld_busy() first. | |
2520 | * | |
2521 | * Basically, this function is used only by request stacking drivers | |
2522 | * to stop dispatching requests to underlying devices when underlying | |
2523 | * devices are busy. This behavior helps more I/O merging on the queue | |
2524 | * of the request stacking driver and prevents I/O throughput regression | |
2525 | * on burst I/O load. | |
2526 | * | |
2527 | * Return: | |
2528 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2529 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2530 | */ | |
2531 | int blk_lld_busy(struct request_queue *q) | |
2532 | { | |
2533 | if (q->lld_busy_fn) | |
2534 | return q->lld_busy_fn(q); | |
2535 | ||
2536 | return 0; | |
2537 | } | |
2538 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2539 | ||
b0fd271d KU |
2540 | /** |
2541 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2542 | * @rq: the clone request to be cleaned up | |
2543 | * | |
2544 | * Description: | |
2545 | * Free all bios in @rq for a cloned request. | |
2546 | */ | |
2547 | void blk_rq_unprep_clone(struct request *rq) | |
2548 | { | |
2549 | struct bio *bio; | |
2550 | ||
2551 | while ((bio = rq->bio) != NULL) { | |
2552 | rq->bio = bio->bi_next; | |
2553 | ||
2554 | bio_put(bio); | |
2555 | } | |
2556 | } | |
2557 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2558 | ||
2559 | /* | |
2560 | * Copy attributes of the original request to the clone request. | |
2561 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2562 | */ | |
2563 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2564 | { | |
2565 | dst->cpu = src->cpu; | |
3a2edd0d | 2566 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2567 | dst->cmd_type = src->cmd_type; |
2568 | dst->__sector = blk_rq_pos(src); | |
2569 | dst->__data_len = blk_rq_bytes(src); | |
2570 | dst->nr_phys_segments = src->nr_phys_segments; | |
2571 | dst->ioprio = src->ioprio; | |
2572 | dst->extra_len = src->extra_len; | |
2573 | } | |
2574 | ||
2575 | /** | |
2576 | * blk_rq_prep_clone - Helper function to setup clone request | |
2577 | * @rq: the request to be setup | |
2578 | * @rq_src: original request to be cloned | |
2579 | * @bs: bio_set that bios for clone are allocated from | |
2580 | * @gfp_mask: memory allocation mask for bio | |
2581 | * @bio_ctr: setup function to be called for each clone bio. | |
2582 | * Returns %0 for success, non %0 for failure. | |
2583 | * @data: private data to be passed to @bio_ctr | |
2584 | * | |
2585 | * Description: | |
2586 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2587 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2588 | * are not copied, and copying such parts is the caller's responsibility. | |
2589 | * Also, pages which the original bios are pointing to are not copied | |
2590 | * and the cloned bios just point same pages. | |
2591 | * So cloned bios must be completed before original bios, which means | |
2592 | * the caller must complete @rq before @rq_src. | |
2593 | */ | |
2594 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2595 | struct bio_set *bs, gfp_t gfp_mask, | |
2596 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2597 | void *data) | |
2598 | { | |
2599 | struct bio *bio, *bio_src; | |
2600 | ||
2601 | if (!bs) | |
2602 | bs = fs_bio_set; | |
2603 | ||
2604 | blk_rq_init(NULL, rq); | |
2605 | ||
2606 | __rq_for_each_bio(bio_src, rq_src) { | |
2607 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2608 | if (!bio) | |
2609 | goto free_and_out; | |
2610 | ||
2611 | __bio_clone(bio, bio_src); | |
2612 | ||
2613 | if (bio_integrity(bio_src) && | |
7878cba9 | 2614 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2615 | goto free_and_out; |
2616 | ||
2617 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2618 | goto free_and_out; | |
2619 | ||
2620 | if (rq->bio) { | |
2621 | rq->biotail->bi_next = bio; | |
2622 | rq->biotail = bio; | |
2623 | } else | |
2624 | rq->bio = rq->biotail = bio; | |
2625 | } | |
2626 | ||
2627 | __blk_rq_prep_clone(rq, rq_src); | |
2628 | ||
2629 | return 0; | |
2630 | ||
2631 | free_and_out: | |
2632 | if (bio) | |
2633 | bio_free(bio, bs); | |
2634 | blk_rq_unprep_clone(rq); | |
2635 | ||
2636 | return -ENOMEM; | |
2637 | } | |
2638 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2639 | ||
18887ad9 | 2640 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2641 | { |
2642 | return queue_work(kblockd_workqueue, work); | |
2643 | } | |
1da177e4 LT |
2644 | EXPORT_SYMBOL(kblockd_schedule_work); |
2645 | ||
1da177e4 LT |
2646 | int __init blk_dev_init(void) |
2647 | { | |
9eb55b03 NK |
2648 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2649 | sizeof(((struct request *)0)->cmd_flags)); | |
2650 | ||
89b90be2 TH |
2651 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2652 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2653 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2654 | if (!kblockd_workqueue) |
2655 | panic("Failed to create kblockd\n"); | |
2656 | ||
2657 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2658 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2659 | |
8324aa91 | 2660 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2661 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2662 | |
d38ecf93 | 2663 | return 0; |
1da177e4 | 2664 | } |