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> |
2056a782 | 29 | #include <linux/blktrace_api.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
5f3ea37c | 31 | #include <trace/block.h> |
1da177e4 | 32 | |
8324aa91 JA |
33 | #include "blk.h" |
34 | ||
0bfc2455 IM |
35 | DEFINE_TRACE(block_plug); |
36 | DEFINE_TRACE(block_unplug_io); | |
37 | DEFINE_TRACE(block_unplug_timer); | |
38 | DEFINE_TRACE(block_getrq); | |
39 | DEFINE_TRACE(block_sleeprq); | |
40 | DEFINE_TRACE(block_rq_requeue); | |
41 | DEFINE_TRACE(block_bio_backmerge); | |
42 | DEFINE_TRACE(block_bio_frontmerge); | |
43 | DEFINE_TRACE(block_bio_queue); | |
44 | DEFINE_TRACE(block_rq_complete); | |
45 | DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */ | |
46 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); | |
47 | ||
165125e1 | 48 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
49 | |
50 | /* | |
51 | * For the allocated request tables | |
52 | */ | |
5ece6c52 | 53 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
54 | |
55 | /* | |
56 | * For queue allocation | |
57 | */ | |
6728cb0e | 58 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Controlling structure to kblockd | |
62 | */ | |
ff856bad | 63 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 64 | |
26b8256e JA |
65 | static void drive_stat_acct(struct request *rq, int new_io) |
66 | { | |
28f13702 | 67 | struct hd_struct *part; |
26b8256e | 68 | int rw = rq_data_dir(rq); |
c9959059 | 69 | int cpu; |
26b8256e | 70 | |
c2553b58 | 71 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
72 | return; |
73 | ||
074a7aca | 74 | cpu = part_stat_lock(); |
83096ebf | 75 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); |
c9959059 | 76 | |
28f13702 | 77 | if (!new_io) |
074a7aca | 78 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 79 | else { |
074a7aca TH |
80 | part_round_stats(cpu, part); |
81 | part_inc_in_flight(part); | |
26b8256e | 82 | } |
e71bf0d0 | 83 | |
074a7aca | 84 | part_stat_unlock(); |
26b8256e JA |
85 | } |
86 | ||
8324aa91 | 87 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
88 | { |
89 | int nr; | |
90 | ||
91 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
92 | if (nr > q->nr_requests) | |
93 | nr = q->nr_requests; | |
94 | q->nr_congestion_on = nr; | |
95 | ||
96 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
97 | if (nr < 1) | |
98 | nr = 1; | |
99 | q->nr_congestion_off = nr; | |
100 | } | |
101 | ||
1da177e4 LT |
102 | /** |
103 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
104 | * @bdev: device | |
105 | * | |
106 | * Locates the passed device's request queue and returns the address of its | |
107 | * backing_dev_info | |
108 | * | |
109 | * Will return NULL if the request queue cannot be located. | |
110 | */ | |
111 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
112 | { | |
113 | struct backing_dev_info *ret = NULL; | |
165125e1 | 114 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
115 | |
116 | if (q) | |
117 | ret = &q->backing_dev_info; | |
118 | return ret; | |
119 | } | |
1da177e4 LT |
120 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
121 | ||
2a4aa30c | 122 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 123 | { |
1afb20f3 FT |
124 | memset(rq, 0, sizeof(*rq)); |
125 | ||
1da177e4 | 126 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 127 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 128 | rq->cpu = -1; |
63a71386 JA |
129 | rq->q = q; |
130 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
131 | INIT_HLIST_NODE(&rq->hash); |
132 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 133 | rq->cmd = rq->__cmd; |
e2494e1b | 134 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 135 | rq->tag = -1; |
1da177e4 | 136 | rq->ref_count = 1; |
b243ddcb | 137 | rq->start_time = jiffies; |
1da177e4 | 138 | } |
2a4aa30c | 139 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 140 | |
5bb23a68 N |
141 | static void req_bio_endio(struct request *rq, struct bio *bio, |
142 | unsigned int nbytes, int error) | |
1da177e4 | 143 | { |
165125e1 | 144 | struct request_queue *q = rq->q; |
797e7dbb | 145 | |
5bb23a68 N |
146 | if (&q->bar_rq != rq) { |
147 | if (error) | |
148 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
149 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
150 | error = -EIO; | |
797e7dbb | 151 | |
5bb23a68 | 152 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 153 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 154 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
155 | nbytes = bio->bi_size; |
156 | } | |
797e7dbb | 157 | |
08bafc03 KM |
158 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
159 | set_bit(BIO_QUIET, &bio->bi_flags); | |
160 | ||
5bb23a68 N |
161 | bio->bi_size -= nbytes; |
162 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
163 | |
164 | if (bio_integrity(bio)) | |
165 | bio_integrity_advance(bio, nbytes); | |
166 | ||
5bb23a68 | 167 | if (bio->bi_size == 0) |
6712ecf8 | 168 | bio_endio(bio, error); |
5bb23a68 N |
169 | } else { |
170 | ||
171 | /* | |
172 | * Okay, this is the barrier request in progress, just | |
173 | * record the error; | |
174 | */ | |
175 | if (error && !q->orderr) | |
176 | q->orderr = error; | |
177 | } | |
1da177e4 | 178 | } |
1da177e4 | 179 | |
1da177e4 LT |
180 | void blk_dump_rq_flags(struct request *rq, char *msg) |
181 | { | |
182 | int bit; | |
183 | ||
6728cb0e | 184 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
185 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
186 | rq->cmd_flags); | |
1da177e4 | 187 | |
83096ebf TH |
188 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
189 | (unsigned long long)blk_rq_pos(rq), | |
190 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 191 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
6728cb0e | 192 | rq->bio, rq->biotail, |
731ec497 | 193 | rq->buffer, rq->data_len); |
1da177e4 | 194 | |
4aff5e23 | 195 | if (blk_pc_request(rq)) { |
6728cb0e | 196 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 197 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
198 | printk("%02x ", rq->cmd[bit]); |
199 | printk("\n"); | |
200 | } | |
201 | } | |
1da177e4 LT |
202 | EXPORT_SYMBOL(blk_dump_rq_flags); |
203 | ||
1da177e4 LT |
204 | /* |
205 | * "plug" the device if there are no outstanding requests: this will | |
206 | * force the transfer to start only after we have put all the requests | |
207 | * on the list. | |
208 | * | |
209 | * This is called with interrupts off and no requests on the queue and | |
210 | * with the queue lock held. | |
211 | */ | |
165125e1 | 212 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
213 | { |
214 | WARN_ON(!irqs_disabled()); | |
215 | ||
216 | /* | |
217 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
218 | * which will restart the queueing | |
219 | */ | |
7daac490 | 220 | if (blk_queue_stopped(q)) |
1da177e4 LT |
221 | return; |
222 | ||
e48ec690 | 223 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 224 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 225 | trace_block_plug(q); |
2056a782 | 226 | } |
1da177e4 | 227 | } |
1da177e4 LT |
228 | EXPORT_SYMBOL(blk_plug_device); |
229 | ||
6c5e0c4d JA |
230 | /** |
231 | * blk_plug_device_unlocked - plug a device without queue lock held | |
232 | * @q: The &struct request_queue to plug | |
233 | * | |
234 | * Description: | |
235 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
236 | * interrupts. | |
237 | **/ | |
238 | void blk_plug_device_unlocked(struct request_queue *q) | |
239 | { | |
240 | unsigned long flags; | |
241 | ||
242 | spin_lock_irqsave(q->queue_lock, flags); | |
243 | blk_plug_device(q); | |
244 | spin_unlock_irqrestore(q->queue_lock, flags); | |
245 | } | |
246 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
247 | ||
1da177e4 LT |
248 | /* |
249 | * remove the queue from the plugged list, if present. called with | |
250 | * queue lock held and interrupts disabled. | |
251 | */ | |
165125e1 | 252 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
253 | { |
254 | WARN_ON(!irqs_disabled()); | |
255 | ||
e48ec690 | 256 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
257 | return 0; |
258 | ||
259 | del_timer(&q->unplug_timer); | |
260 | return 1; | |
261 | } | |
1da177e4 LT |
262 | EXPORT_SYMBOL(blk_remove_plug); |
263 | ||
264 | /* | |
265 | * remove the plug and let it rip.. | |
266 | */ | |
165125e1 | 267 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 268 | { |
7daac490 | 269 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 | 270 | return; |
a31a9738 | 271 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) |
1da177e4 LT |
272 | return; |
273 | ||
22e2c507 | 274 | q->request_fn(q); |
1da177e4 | 275 | } |
1da177e4 LT |
276 | |
277 | /** | |
278 | * generic_unplug_device - fire a request queue | |
165125e1 | 279 | * @q: The &struct request_queue in question |
1da177e4 LT |
280 | * |
281 | * Description: | |
282 | * Linux uses plugging to build bigger requests queues before letting | |
283 | * the device have at them. If a queue is plugged, the I/O scheduler | |
284 | * is still adding and merging requests on the queue. Once the queue | |
285 | * gets unplugged, the request_fn defined for the queue is invoked and | |
286 | * transfers started. | |
287 | **/ | |
165125e1 | 288 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 289 | { |
dbaf2c00 JA |
290 | if (blk_queue_plugged(q)) { |
291 | spin_lock_irq(q->queue_lock); | |
292 | __generic_unplug_device(q); | |
293 | spin_unlock_irq(q->queue_lock); | |
294 | } | |
1da177e4 LT |
295 | } |
296 | EXPORT_SYMBOL(generic_unplug_device); | |
297 | ||
298 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
299 | struct page *page) | |
300 | { | |
165125e1 | 301 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 302 | |
2ad8b1ef | 303 | blk_unplug(q); |
1da177e4 LT |
304 | } |
305 | ||
86db1e29 | 306 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 307 | { |
165125e1 JA |
308 | struct request_queue *q = |
309 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 310 | |
5f3ea37c | 311 | trace_block_unplug_io(q); |
1da177e4 LT |
312 | q->unplug_fn(q); |
313 | } | |
314 | ||
86db1e29 | 315 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 316 | { |
165125e1 | 317 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 318 | |
5f3ea37c | 319 | trace_block_unplug_timer(q); |
18887ad9 | 320 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
321 | } |
322 | ||
2ad8b1ef AB |
323 | void blk_unplug(struct request_queue *q) |
324 | { | |
325 | /* | |
326 | * devices don't necessarily have an ->unplug_fn defined | |
327 | */ | |
328 | if (q->unplug_fn) { | |
5f3ea37c | 329 | trace_block_unplug_io(q); |
2ad8b1ef AB |
330 | q->unplug_fn(q); |
331 | } | |
332 | } | |
333 | EXPORT_SYMBOL(blk_unplug); | |
334 | ||
1da177e4 LT |
335 | /** |
336 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 337 | * @q: The &struct request_queue in question |
1da177e4 LT |
338 | * |
339 | * Description: | |
340 | * blk_start_queue() will clear the stop flag on the queue, and call | |
341 | * the request_fn for the queue if it was in a stopped state when | |
342 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
343 | **/ | |
165125e1 | 344 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 345 | { |
a038e253 PBG |
346 | WARN_ON(!irqs_disabled()); |
347 | ||
75ad23bc | 348 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
a538cd03 | 349 | __blk_run_queue(q); |
1da177e4 | 350 | } |
1da177e4 LT |
351 | EXPORT_SYMBOL(blk_start_queue); |
352 | ||
353 | /** | |
354 | * blk_stop_queue - stop a queue | |
165125e1 | 355 | * @q: The &struct request_queue in question |
1da177e4 LT |
356 | * |
357 | * Description: | |
358 | * The Linux block layer assumes that a block driver will consume all | |
359 | * entries on the request queue when the request_fn strategy is called. | |
360 | * Often this will not happen, because of hardware limitations (queue | |
361 | * depth settings). If a device driver gets a 'queue full' response, | |
362 | * or if it simply chooses not to queue more I/O at one point, it can | |
363 | * call this function to prevent the request_fn from being called until | |
364 | * the driver has signalled it's ready to go again. This happens by calling | |
365 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
366 | **/ | |
165125e1 | 367 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
368 | { |
369 | blk_remove_plug(q); | |
75ad23bc | 370 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
371 | } |
372 | EXPORT_SYMBOL(blk_stop_queue); | |
373 | ||
374 | /** | |
375 | * blk_sync_queue - cancel any pending callbacks on a queue | |
376 | * @q: the queue | |
377 | * | |
378 | * Description: | |
379 | * The block layer may perform asynchronous callback activity | |
380 | * on a queue, such as calling the unplug function after a timeout. | |
381 | * A block device may call blk_sync_queue to ensure that any | |
382 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 383 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
384 | * that its ->make_request_fn will not re-add plugging prior to calling |
385 | * this function. | |
386 | * | |
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 |
80a4b58e JA |
399 | * |
400 | * Description: | |
401 | * See @blk_run_queue. This variant must be called with the queue lock | |
402 | * held and interrupts disabled. | |
403 | * | |
1da177e4 | 404 | */ |
75ad23bc | 405 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 406 | { |
1da177e4 | 407 | blk_remove_plug(q); |
dac07ec1 | 408 | |
a538cd03 TH |
409 | if (unlikely(blk_queue_stopped(q))) |
410 | return; | |
411 | ||
412 | if (elv_queue_empty(q)) | |
413 | return; | |
414 | ||
dac07ec1 JA |
415 | /* |
416 | * Only recurse once to avoid overrunning the stack, let the unplug | |
417 | * handling reinvoke the handler shortly if we already got there. | |
418 | */ | |
a538cd03 TH |
419 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { |
420 | q->request_fn(q); | |
421 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
422 | } else { | |
423 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
424 | kblockd_schedule_work(q, &q->unplug_work); | |
425 | } | |
75ad23bc NP |
426 | } |
427 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 428 | |
75ad23bc NP |
429 | /** |
430 | * blk_run_queue - run a single device queue | |
431 | * @q: The queue to run | |
80a4b58e JA |
432 | * |
433 | * Description: | |
434 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 435 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
436 | */ |
437 | void blk_run_queue(struct request_queue *q) | |
438 | { | |
439 | unsigned long flags; | |
440 | ||
441 | spin_lock_irqsave(q->queue_lock, flags); | |
442 | __blk_run_queue(q); | |
1da177e4 LT |
443 | spin_unlock_irqrestore(q->queue_lock, flags); |
444 | } | |
445 | EXPORT_SYMBOL(blk_run_queue); | |
446 | ||
165125e1 | 447 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
448 | { |
449 | kobject_put(&q->kobj); | |
450 | } | |
483f4afc | 451 | |
6728cb0e | 452 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 453 | { |
e3335de9 JA |
454 | /* |
455 | * We know we have process context here, so we can be a little | |
456 | * cautious and ensure that pending block actions on this device | |
457 | * are done before moving on. Going into this function, we should | |
458 | * not have processes doing IO to this device. | |
459 | */ | |
460 | blk_sync_queue(q); | |
461 | ||
483f4afc | 462 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 463 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
464 | mutex_unlock(&q->sysfs_lock); |
465 | ||
466 | if (q->elevator) | |
467 | elevator_exit(q->elevator); | |
468 | ||
469 | blk_put_queue(q); | |
470 | } | |
1da177e4 LT |
471 | EXPORT_SYMBOL(blk_cleanup_queue); |
472 | ||
165125e1 | 473 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
474 | { |
475 | struct request_list *rl = &q->rq; | |
476 | ||
1faa16d2 JA |
477 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
478 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 479 | rl->elvpriv = 0; |
1faa16d2 JA |
480 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
481 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 482 | |
1946089a CL |
483 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
484 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
485 | |
486 | if (!rl->rq_pool) | |
487 | return -ENOMEM; | |
488 | ||
489 | return 0; | |
490 | } | |
491 | ||
165125e1 | 492 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 493 | { |
1946089a CL |
494 | return blk_alloc_queue_node(gfp_mask, -1); |
495 | } | |
496 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 497 | |
165125e1 | 498 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 499 | { |
165125e1 | 500 | struct request_queue *q; |
e0bf68dd | 501 | int err; |
1946089a | 502 | |
8324aa91 | 503 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 504 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
505 | if (!q) |
506 | return NULL; | |
507 | ||
e0bf68dd PZ |
508 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
509 | q->backing_dev_info.unplug_io_data = q; | |
510 | err = bdi_init(&q->backing_dev_info); | |
511 | if (err) { | |
8324aa91 | 512 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
513 | return NULL; |
514 | } | |
515 | ||
1da177e4 | 516 | init_timer(&q->unplug_timer); |
242f9dcb JA |
517 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
518 | INIT_LIST_HEAD(&q->timeout_list); | |
713ada9b | 519 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 520 | |
8324aa91 | 521 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 522 | |
483f4afc | 523 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 524 | spin_lock_init(&q->__queue_lock); |
483f4afc | 525 | |
1da177e4 LT |
526 | return q; |
527 | } | |
1946089a | 528 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
529 | |
530 | /** | |
531 | * blk_init_queue - prepare a request queue for use with a block device | |
532 | * @rfn: The function to be called to process requests that have been | |
533 | * placed on the queue. | |
534 | * @lock: Request queue spin lock | |
535 | * | |
536 | * Description: | |
537 | * If a block device wishes to use the standard request handling procedures, | |
538 | * which sorts requests and coalesces adjacent requests, then it must | |
539 | * call blk_init_queue(). The function @rfn will be called when there | |
540 | * are requests on the queue that need to be processed. If the device | |
541 | * supports plugging, then @rfn may not be called immediately when requests | |
542 | * are available on the queue, but may be called at some time later instead. | |
543 | * Plugged queues are generally unplugged when a buffer belonging to one | |
544 | * of the requests on the queue is needed, or due to memory pressure. | |
545 | * | |
546 | * @rfn is not required, or even expected, to remove all requests off the | |
547 | * queue, but only as many as it can handle at a time. If it does leave | |
548 | * requests on the queue, it is responsible for arranging that the requests | |
549 | * get dealt with eventually. | |
550 | * | |
551 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
552 | * request queue; this lock will be taken also from interrupt context, so irq |
553 | * disabling is needed for it. | |
1da177e4 | 554 | * |
710027a4 | 555 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
556 | * it didn't succeed. |
557 | * | |
558 | * Note: | |
559 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
560 | * when the block device is deactivated (such as at module unload). | |
561 | **/ | |
1946089a | 562 | |
165125e1 | 563 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 564 | { |
1946089a CL |
565 | return blk_init_queue_node(rfn, lock, -1); |
566 | } | |
567 | EXPORT_SYMBOL(blk_init_queue); | |
568 | ||
165125e1 | 569 | struct request_queue * |
1946089a CL |
570 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
571 | { | |
165125e1 | 572 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
573 | |
574 | if (!q) | |
575 | return NULL; | |
576 | ||
1946089a | 577 | q->node = node_id; |
8669aafd | 578 | if (blk_init_free_list(q)) { |
8324aa91 | 579 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
580 | return NULL; |
581 | } | |
1da177e4 | 582 | |
152587de | 583 | /* |
584 | * if caller didn't supply a lock, they get per-queue locking with | |
585 | * our embedded lock | |
586 | */ | |
e7e72bf6 | 587 | if (!lock) |
152587de | 588 | lock = &q->__queue_lock; |
152587de | 589 | |
1da177e4 | 590 | q->request_fn = rfn; |
1da177e4 LT |
591 | q->prep_rq_fn = NULL; |
592 | q->unplug_fn = generic_unplug_device; | |
bc58ba94 | 593 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
1da177e4 LT |
594 | q->queue_lock = lock; |
595 | ||
f3b144aa JA |
596 | /* |
597 | * This also sets hw/phys segments, boundary and size | |
598 | */ | |
1da177e4 | 599 | blk_queue_make_request(q, __make_request); |
1da177e4 | 600 | |
44ec9542 AS |
601 | q->sg_reserved_size = INT_MAX; |
602 | ||
abf54393 FT |
603 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
604 | ||
1da177e4 LT |
605 | /* |
606 | * all done | |
607 | */ | |
608 | if (!elevator_init(q, NULL)) { | |
609 | blk_queue_congestion_threshold(q); | |
610 | return q; | |
611 | } | |
612 | ||
8669aafd | 613 | blk_put_queue(q); |
1da177e4 LT |
614 | return NULL; |
615 | } | |
1946089a | 616 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 617 | |
165125e1 | 618 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 619 | { |
fde6ad22 | 620 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 621 | kobject_get(&q->kobj); |
1da177e4 LT |
622 | return 0; |
623 | } | |
624 | ||
625 | return 1; | |
626 | } | |
1da177e4 | 627 | |
165125e1 | 628 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 629 | { |
4aff5e23 | 630 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 631 | elv_put_request(q, rq); |
1da177e4 LT |
632 | mempool_free(rq, q->rq.rq_pool); |
633 | } | |
634 | ||
1ea25ecb | 635 | static struct request * |
42dad764 | 636 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) |
1da177e4 LT |
637 | { |
638 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
639 | ||
640 | if (!rq) | |
641 | return NULL; | |
642 | ||
2a4aa30c | 643 | blk_rq_init(q, rq); |
1afb20f3 | 644 | |
42dad764 | 645 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 646 | |
cb98fc8b | 647 | if (priv) { |
cb78b285 | 648 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
649 | mempool_free(rq, q->rq.rq_pool); |
650 | return NULL; | |
651 | } | |
4aff5e23 | 652 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 653 | } |
1da177e4 | 654 | |
cb98fc8b | 655 | return rq; |
1da177e4 LT |
656 | } |
657 | ||
658 | /* | |
659 | * ioc_batching returns true if the ioc is a valid batching request and | |
660 | * should be given priority access to a request. | |
661 | */ | |
165125e1 | 662 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
663 | { |
664 | if (!ioc) | |
665 | return 0; | |
666 | ||
667 | /* | |
668 | * Make sure the process is able to allocate at least 1 request | |
669 | * even if the batch times out, otherwise we could theoretically | |
670 | * lose wakeups. | |
671 | */ | |
672 | return ioc->nr_batch_requests == q->nr_batching || | |
673 | (ioc->nr_batch_requests > 0 | |
674 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
675 | } | |
676 | ||
677 | /* | |
678 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
679 | * will cause the process to be a "batcher" on all queues in the system. This | |
680 | * is the behaviour we want though - once it gets a wakeup it should be given | |
681 | * a nice run. | |
682 | */ | |
165125e1 | 683 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
684 | { |
685 | if (!ioc || ioc_batching(q, ioc)) | |
686 | return; | |
687 | ||
688 | ioc->nr_batch_requests = q->nr_batching; | |
689 | ioc->last_waited = jiffies; | |
690 | } | |
691 | ||
1faa16d2 | 692 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
693 | { |
694 | struct request_list *rl = &q->rq; | |
695 | ||
1faa16d2 JA |
696 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
697 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 698 | |
1faa16d2 JA |
699 | if (rl->count[sync] + 1 <= q->nr_requests) { |
700 | if (waitqueue_active(&rl->wait[sync])) | |
701 | wake_up(&rl->wait[sync]); | |
1da177e4 | 702 | |
1faa16d2 | 703 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
704 | } |
705 | } | |
706 | ||
707 | /* | |
708 | * A request has just been released. Account for it, update the full and | |
709 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
710 | */ | |
1faa16d2 | 711 | static void freed_request(struct request_queue *q, int sync, int priv) |
1da177e4 LT |
712 | { |
713 | struct request_list *rl = &q->rq; | |
714 | ||
1faa16d2 | 715 | rl->count[sync]--; |
cb98fc8b TH |
716 | if (priv) |
717 | rl->elvpriv--; | |
1da177e4 | 718 | |
1faa16d2 | 719 | __freed_request(q, sync); |
1da177e4 | 720 | |
1faa16d2 JA |
721 | if (unlikely(rl->starved[sync ^ 1])) |
722 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
723 | } |
724 | ||
1da177e4 | 725 | /* |
d6344532 NP |
726 | * Get a free request, queue_lock must be held. |
727 | * Returns NULL on failure, with queue_lock held. | |
728 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 729 | */ |
165125e1 | 730 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 731 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
732 | { |
733 | struct request *rq = NULL; | |
734 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 735 | struct io_context *ioc = NULL; |
1faa16d2 | 736 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
88ee5ef1 JA |
737 | int may_queue, priv; |
738 | ||
7749a8d4 | 739 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
740 | if (may_queue == ELV_MQUEUE_NO) |
741 | goto rq_starved; | |
742 | ||
1faa16d2 JA |
743 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
744 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 745 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
746 | /* |
747 | * The queue will fill after this allocation, so set | |
748 | * it as full, and mark this process as "batching". | |
749 | * This process will be allowed to complete a batch of | |
750 | * requests, others will be blocked. | |
751 | */ | |
1faa16d2 | 752 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 753 | ioc_set_batching(q, ioc); |
1faa16d2 | 754 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
755 | } else { |
756 | if (may_queue != ELV_MQUEUE_MUST | |
757 | && !ioc_batching(q, ioc)) { | |
758 | /* | |
759 | * The queue is full and the allocating | |
760 | * process is not a "batcher", and not | |
761 | * exempted by the IO scheduler | |
762 | */ | |
763 | goto out; | |
764 | } | |
765 | } | |
1da177e4 | 766 | } |
1faa16d2 | 767 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
768 | } |
769 | ||
082cf69e JA |
770 | /* |
771 | * Only allow batching queuers to allocate up to 50% over the defined | |
772 | * limit of requests, otherwise we could have thousands of requests | |
773 | * allocated with any setting of ->nr_requests | |
774 | */ | |
1faa16d2 | 775 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 776 | goto out; |
fd782a4a | 777 | |
1faa16d2 JA |
778 | rl->count[is_sync]++; |
779 | rl->starved[is_sync] = 0; | |
cb98fc8b | 780 | |
64521d1a | 781 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
782 | if (priv) |
783 | rl->elvpriv++; | |
784 | ||
42dad764 JM |
785 | if (blk_queue_io_stat(q)) |
786 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
787 | spin_unlock_irq(q->queue_lock); |
788 | ||
7749a8d4 | 789 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 790 | if (unlikely(!rq)) { |
1da177e4 LT |
791 | /* |
792 | * Allocation failed presumably due to memory. Undo anything | |
793 | * we might have messed up. | |
794 | * | |
795 | * Allocating task should really be put onto the front of the | |
796 | * wait queue, but this is pretty rare. | |
797 | */ | |
798 | spin_lock_irq(q->queue_lock); | |
1faa16d2 | 799 | freed_request(q, is_sync, priv); |
1da177e4 LT |
800 | |
801 | /* | |
802 | * in the very unlikely event that allocation failed and no | |
803 | * requests for this direction was pending, mark us starved | |
804 | * so that freeing of a request in the other direction will | |
805 | * notice us. another possible fix would be to split the | |
806 | * rq mempool into READ and WRITE | |
807 | */ | |
808 | rq_starved: | |
1faa16d2 JA |
809 | if (unlikely(rl->count[is_sync] == 0)) |
810 | rl->starved[is_sync] = 1; | |
1da177e4 | 811 | |
1da177e4 LT |
812 | goto out; |
813 | } | |
814 | ||
88ee5ef1 JA |
815 | /* |
816 | * ioc may be NULL here, and ioc_batching will be false. That's | |
817 | * OK, if the queue is under the request limit then requests need | |
818 | * not count toward the nr_batch_requests limit. There will always | |
819 | * be some limit enforced by BLK_BATCH_TIME. | |
820 | */ | |
1da177e4 LT |
821 | if (ioc_batching(q, ioc)) |
822 | ioc->nr_batch_requests--; | |
6728cb0e | 823 | |
1faa16d2 | 824 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 825 | out: |
1da177e4 LT |
826 | return rq; |
827 | } | |
828 | ||
829 | /* | |
830 | * No available requests for this queue, unplug the device and wait for some | |
831 | * requests to become available. | |
d6344532 NP |
832 | * |
833 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 834 | */ |
165125e1 | 835 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 836 | struct bio *bio) |
1da177e4 | 837 | { |
1faa16d2 | 838 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
839 | struct request *rq; |
840 | ||
7749a8d4 | 841 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
842 | while (!rq) { |
843 | DEFINE_WAIT(wait); | |
05caf8db | 844 | struct io_context *ioc; |
1da177e4 LT |
845 | struct request_list *rl = &q->rq; |
846 | ||
1faa16d2 | 847 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
848 | TASK_UNINTERRUPTIBLE); |
849 | ||
1faa16d2 | 850 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 851 | |
05caf8db ZY |
852 | __generic_unplug_device(q); |
853 | spin_unlock_irq(q->queue_lock); | |
854 | io_schedule(); | |
1da177e4 | 855 | |
05caf8db ZY |
856 | /* |
857 | * After sleeping, we become a "batching" process and | |
858 | * will be able to allocate at least one request, and | |
859 | * up to a big batch of them for a small period time. | |
860 | * See ioc_batching, ioc_set_batching | |
861 | */ | |
862 | ioc = current_io_context(GFP_NOIO, q->node); | |
863 | ioc_set_batching(q, ioc); | |
d6344532 | 864 | |
05caf8db | 865 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 866 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
867 | |
868 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
869 | }; | |
1da177e4 LT |
870 | |
871 | return rq; | |
872 | } | |
873 | ||
165125e1 | 874 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
875 | { |
876 | struct request *rq; | |
877 | ||
878 | BUG_ON(rw != READ && rw != WRITE); | |
879 | ||
d6344532 NP |
880 | spin_lock_irq(q->queue_lock); |
881 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 882 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 883 | } else { |
22e2c507 | 884 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
885 | if (!rq) |
886 | spin_unlock_irq(q->queue_lock); | |
887 | } | |
888 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
889 | |
890 | return rq; | |
891 | } | |
1da177e4 LT |
892 | EXPORT_SYMBOL(blk_get_request); |
893 | ||
894 | /** | |
895 | * blk_requeue_request - put a request back on queue | |
896 | * @q: request queue where request should be inserted | |
897 | * @rq: request to be inserted | |
898 | * | |
899 | * Description: | |
900 | * Drivers often keep queueing requests until the hardware cannot accept | |
901 | * more, when that condition happens we need to put the request back | |
902 | * on the queue. Must be called with queue lock held. | |
903 | */ | |
165125e1 | 904 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 905 | { |
242f9dcb JA |
906 | blk_delete_timer(rq); |
907 | blk_clear_rq_complete(rq); | |
5f3ea37c | 908 | trace_block_rq_requeue(q, rq); |
2056a782 | 909 | |
1da177e4 LT |
910 | if (blk_rq_tagged(rq)) |
911 | blk_queue_end_tag(q, rq); | |
912 | ||
913 | elv_requeue_request(q, rq); | |
914 | } | |
1da177e4 LT |
915 | EXPORT_SYMBOL(blk_requeue_request); |
916 | ||
917 | /** | |
710027a4 | 918 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
919 | * @q: request queue where request should be inserted |
920 | * @rq: request to be inserted | |
921 | * @at_head: insert request at head or tail of queue | |
922 | * @data: private data | |
1da177e4 LT |
923 | * |
924 | * Description: | |
925 | * Many block devices need to execute commands asynchronously, so they don't | |
926 | * block the whole kernel from preemption during request execution. This is | |
927 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
928 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
929 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
930 | * |
931 | * We have the option of inserting the head or the tail of the queue. | |
932 | * Typically we use the tail for new ioctls and so forth. We use the head | |
933 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
934 | * host that is unable to accept a particular command. | |
935 | */ | |
165125e1 | 936 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 937 | int at_head, void *data) |
1da177e4 | 938 | { |
867d1191 | 939 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
940 | unsigned long flags; |
941 | ||
942 | /* | |
943 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
944 | * must not attempt merges on this) and that it acts as a soft | |
945 | * barrier | |
946 | */ | |
4aff5e23 | 947 | rq->cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
948 | |
949 | rq->special = data; | |
950 | ||
951 | spin_lock_irqsave(q->queue_lock, flags); | |
952 | ||
953 | /* | |
954 | * If command is tagged, release the tag | |
955 | */ | |
867d1191 TH |
956 | if (blk_rq_tagged(rq)) |
957 | blk_queue_end_tag(q, rq); | |
1da177e4 | 958 | |
b238b3d4 | 959 | drive_stat_acct(rq, 1); |
867d1191 | 960 | __elv_add_request(q, rq, where, 0); |
a7f55792 | 961 | __blk_run_queue(q); |
1da177e4 LT |
962 | spin_unlock_irqrestore(q->queue_lock, flags); |
963 | } | |
1da177e4 LT |
964 | EXPORT_SYMBOL(blk_insert_request); |
965 | ||
1da177e4 LT |
966 | /* |
967 | * add-request adds a request to the linked list. | |
968 | * queue lock is held and interrupts disabled, as we muck with the | |
969 | * request queue list. | |
970 | */ | |
6728cb0e | 971 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 972 | { |
b238b3d4 | 973 | drive_stat_acct(req, 1); |
1da177e4 | 974 | |
1da177e4 LT |
975 | /* |
976 | * elevator indicated where it wants this request to be | |
977 | * inserted at elevator_merge time | |
978 | */ | |
979 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
980 | } | |
6728cb0e | 981 | |
074a7aca TH |
982 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
983 | unsigned long now) | |
984 | { | |
985 | if (now == part->stamp) | |
986 | return; | |
987 | ||
988 | if (part->in_flight) { | |
989 | __part_stat_add(cpu, part, time_in_queue, | |
990 | part->in_flight * (now - part->stamp)); | |
991 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
992 | } | |
993 | part->stamp = now; | |
994 | } | |
995 | ||
996 | /** | |
496aa8a9 RD |
997 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
998 | * @cpu: cpu number for stats access | |
999 | * @part: target partition | |
1da177e4 LT |
1000 | * |
1001 | * The average IO queue length and utilisation statistics are maintained | |
1002 | * by observing the current state of the queue length and the amount of | |
1003 | * time it has been in this state for. | |
1004 | * | |
1005 | * Normally, that accounting is done on IO completion, but that can result | |
1006 | * in more than a second's worth of IO being accounted for within any one | |
1007 | * second, leading to >100% utilisation. To deal with that, we call this | |
1008 | * function to do a round-off before returning the results when reading | |
1009 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1010 | * the current jiffies and restarts the counters again. | |
1011 | */ | |
c9959059 | 1012 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1013 | { |
1014 | unsigned long now = jiffies; | |
1015 | ||
074a7aca TH |
1016 | if (part->partno) |
1017 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1018 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1019 | } |
074a7aca | 1020 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1021 | |
1da177e4 LT |
1022 | /* |
1023 | * queue lock must be held | |
1024 | */ | |
165125e1 | 1025 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1026 | { |
1da177e4 LT |
1027 | if (unlikely(!q)) |
1028 | return; | |
1029 | if (unlikely(--req->ref_count)) | |
1030 | return; | |
1031 | ||
8922e16c TH |
1032 | elv_completed_request(q, req); |
1033 | ||
1cd96c24 BH |
1034 | /* this is a bio leak */ |
1035 | WARN_ON(req->bio != NULL); | |
1036 | ||
1da177e4 LT |
1037 | /* |
1038 | * Request may not have originated from ll_rw_blk. if not, | |
1039 | * it didn't come out of our reserved rq pools | |
1040 | */ | |
49171e5c | 1041 | if (req->cmd_flags & REQ_ALLOCED) { |
1faa16d2 | 1042 | int is_sync = rq_is_sync(req) != 0; |
4aff5e23 | 1043 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1044 | |
1da177e4 | 1045 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1046 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1047 | |
1048 | blk_free_request(q, req); | |
1faa16d2 | 1049 | freed_request(q, is_sync, priv); |
1da177e4 LT |
1050 | } |
1051 | } | |
6e39b69e MC |
1052 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1053 | ||
1da177e4 LT |
1054 | void blk_put_request(struct request *req) |
1055 | { | |
8922e16c | 1056 | unsigned long flags; |
165125e1 | 1057 | struct request_queue *q = req->q; |
8922e16c | 1058 | |
52a93ba8 FT |
1059 | spin_lock_irqsave(q->queue_lock, flags); |
1060 | __blk_put_request(q, req); | |
1061 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1062 | } |
1da177e4 LT |
1063 | EXPORT_SYMBOL(blk_put_request); |
1064 | ||
86db1e29 | 1065 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1066 | { |
c7c22e4d | 1067 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1068 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1069 | |
1070 | /* | |
1071 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1072 | */ | |
6000a368 MC |
1073 | if (bio_rw_ahead(bio)) |
1074 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1075 | REQ_FAILFAST_DRIVER); | |
1076 | if (bio_failfast_dev(bio)) | |
1077 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1078 | if (bio_failfast_transport(bio)) | |
1079 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1080 | if (bio_failfast_driver(bio)) | |
1081 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
52d9e675 | 1082 | |
fb2dce86 | 1083 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1084 | req->cmd_flags |= REQ_DISCARD; |
1085 | if (bio_barrier(bio)) | |
1086 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1087 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 | 1088 | } else if (unlikely(bio_barrier(bio))) |
e4025f6c | 1089 | req->cmd_flags |= REQ_HARDBARRIER; |
52d9e675 | 1090 | |
b31dc66a | 1091 | if (bio_sync(bio)) |
4aff5e23 | 1092 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1093 | if (bio_rw_meta(bio)) |
1094 | req->cmd_flags |= REQ_RW_META; | |
aeb6fafb JA |
1095 | if (bio_noidle(bio)) |
1096 | req->cmd_flags |= REQ_NOIDLE; | |
b31dc66a | 1097 | |
52d9e675 TH |
1098 | req->errors = 0; |
1099 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1100 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1101 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1102 | } |
1103 | ||
644b2d99 JA |
1104 | /* |
1105 | * Only disabling plugging for non-rotational devices if it does tagging | |
1106 | * as well, otherwise we do need the proper merging | |
1107 | */ | |
1108 | static inline bool queue_should_plug(struct request_queue *q) | |
1109 | { | |
1110 | return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); | |
1111 | } | |
1112 | ||
165125e1 | 1113 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1114 | { |
450991bc | 1115 | struct request *req; |
a7384677 | 1116 | int el_ret, nr_sectors; |
51da90fc JA |
1117 | const unsigned short prio = bio_prio(bio); |
1118 | const int sync = bio_sync(bio); | |
213d9417 | 1119 | const int unplug = bio_unplug(bio); |
7749a8d4 | 1120 | int rw_flags; |
1da177e4 | 1121 | |
1da177e4 | 1122 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1123 | |
1124 | /* | |
1125 | * low level driver can indicate that it wants pages above a | |
1126 | * certain limit bounced to low memory (ie for highmem, or even | |
1127 | * ISA dma in theory) | |
1128 | */ | |
1129 | blk_queue_bounce(q, &bio); | |
1130 | ||
1da177e4 LT |
1131 | spin_lock_irq(q->queue_lock); |
1132 | ||
a7384677 | 1133 | if (unlikely(bio_barrier(bio)) || elv_queue_empty(q)) |
1da177e4 LT |
1134 | goto get_rq; |
1135 | ||
1136 | el_ret = elv_merge(q, &req, bio); | |
1137 | switch (el_ret) { | |
6728cb0e JA |
1138 | case ELEVATOR_BACK_MERGE: |
1139 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1140 | |
6728cb0e JA |
1141 | if (!ll_back_merge_fn(q, req, bio)) |
1142 | break; | |
1da177e4 | 1143 | |
5f3ea37c | 1144 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1145 | |
6728cb0e JA |
1146 | req->biotail->bi_next = bio; |
1147 | req->biotail = bio; | |
1148 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1149 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1150 | if (!blk_rq_cpu_valid(req)) |
1151 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1152 | drive_stat_acct(req, 0); |
1153 | if (!attempt_back_merge(q, req)) | |
1154 | elv_merged_request(q, req, el_ret); | |
1155 | goto out; | |
1da177e4 | 1156 | |
6728cb0e JA |
1157 | case ELEVATOR_FRONT_MERGE: |
1158 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1159 | |
6728cb0e JA |
1160 | if (!ll_front_merge_fn(q, req, bio)) |
1161 | break; | |
1da177e4 | 1162 | |
5f3ea37c | 1163 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1164 | |
6728cb0e JA |
1165 | bio->bi_next = req->bio; |
1166 | req->bio = bio; | |
1da177e4 | 1167 | |
6728cb0e JA |
1168 | /* |
1169 | * may not be valid. if the low level driver said | |
1170 | * it didn't need a bounce buffer then it better | |
1171 | * not touch req->buffer either... | |
1172 | */ | |
1173 | req->buffer = bio_data(bio); | |
1174 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1175 | req->hard_cur_sectors = req->current_nr_sectors; | |
1176 | req->sector = req->hard_sector = bio->bi_sector; | |
1177 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1178 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1179 | if (!blk_rq_cpu_valid(req)) |
1180 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1181 | drive_stat_acct(req, 0); |
1182 | if (!attempt_front_merge(q, req)) | |
1183 | elv_merged_request(q, req, el_ret); | |
1184 | goto out; | |
1185 | ||
1186 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1187 | default: | |
1188 | ; | |
1da177e4 LT |
1189 | } |
1190 | ||
450991bc | 1191 | get_rq: |
7749a8d4 JA |
1192 | /* |
1193 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1194 | * but we need to set it earlier to expose the sync flag to the | |
1195 | * rq allocator and io schedulers. | |
1196 | */ | |
1197 | rw_flags = bio_data_dir(bio); | |
1198 | if (sync) | |
1199 | rw_flags |= REQ_RW_SYNC; | |
1200 | ||
1da177e4 | 1201 | /* |
450991bc | 1202 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1203 | * Returns with the queue unlocked. |
450991bc | 1204 | */ |
7749a8d4 | 1205 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1206 | |
450991bc NP |
1207 | /* |
1208 | * After dropping the lock and possibly sleeping here, our request | |
1209 | * may now be mergeable after it had proven unmergeable (above). | |
1210 | * We don't worry about that case for efficiency. It won't happen | |
1211 | * often, and the elevators are able to handle it. | |
1da177e4 | 1212 | */ |
52d9e675 | 1213 | init_request_from_bio(req, bio); |
1da177e4 | 1214 | |
450991bc | 1215 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1216 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1217 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1218 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
644b2d99 | 1219 | if (queue_should_plug(q) && elv_queue_empty(q)) |
450991bc | 1220 | blk_plug_device(q); |
1da177e4 LT |
1221 | add_request(q, req); |
1222 | out: | |
644b2d99 | 1223 | if (unplug || !queue_should_plug(q)) |
1da177e4 | 1224 | __generic_unplug_device(q); |
1da177e4 LT |
1225 | spin_unlock_irq(q->queue_lock); |
1226 | return 0; | |
1da177e4 LT |
1227 | } |
1228 | ||
1229 | /* | |
1230 | * If bio->bi_dev is a partition, remap the location | |
1231 | */ | |
1232 | static inline void blk_partition_remap(struct bio *bio) | |
1233 | { | |
1234 | struct block_device *bdev = bio->bi_bdev; | |
1235 | ||
bf2de6f5 | 1236 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1237 | struct hd_struct *p = bdev->bd_part; |
1238 | ||
1da177e4 LT |
1239 | bio->bi_sector += p->start_sect; |
1240 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1241 | |
5f3ea37c | 1242 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, |
c7149d6b AB |
1243 | bdev->bd_dev, bio->bi_sector, |
1244 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1245 | } |
1246 | } | |
1247 | ||
1da177e4 LT |
1248 | static void handle_bad_sector(struct bio *bio) |
1249 | { | |
1250 | char b[BDEVNAME_SIZE]; | |
1251 | ||
1252 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1253 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1254 | bdevname(bio->bi_bdev, b), | |
1255 | bio->bi_rw, | |
1256 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1257 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1258 | ||
1259 | set_bit(BIO_EOF, &bio->bi_flags); | |
1260 | } | |
1261 | ||
c17bb495 AM |
1262 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1263 | ||
1264 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1265 | ||
1266 | static int __init setup_fail_make_request(char *str) | |
1267 | { | |
1268 | return setup_fault_attr(&fail_make_request, str); | |
1269 | } | |
1270 | __setup("fail_make_request=", setup_fail_make_request); | |
1271 | ||
1272 | static int should_fail_request(struct bio *bio) | |
1273 | { | |
eddb2e26 TH |
1274 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1275 | ||
1276 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1277 | return should_fail(&fail_make_request, bio->bi_size); |
1278 | ||
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static int __init fail_make_request_debugfs(void) | |
1283 | { | |
1284 | return init_fault_attr_dentries(&fail_make_request, | |
1285 | "fail_make_request"); | |
1286 | } | |
1287 | ||
1288 | late_initcall(fail_make_request_debugfs); | |
1289 | ||
1290 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1291 | ||
1292 | static inline int should_fail_request(struct bio *bio) | |
1293 | { | |
1294 | return 0; | |
1295 | } | |
1296 | ||
1297 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1298 | ||
c07e2b41 JA |
1299 | /* |
1300 | * Check whether this bio extends beyond the end of the device. | |
1301 | */ | |
1302 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1303 | { | |
1304 | sector_t maxsector; | |
1305 | ||
1306 | if (!nr_sectors) | |
1307 | return 0; | |
1308 | ||
1309 | /* Test device or partition size, when known. */ | |
1310 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1311 | if (maxsector) { | |
1312 | sector_t sector = bio->bi_sector; | |
1313 | ||
1314 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1315 | /* | |
1316 | * This may well happen - the kernel calls bread() | |
1317 | * without checking the size of the device, e.g., when | |
1318 | * mounting a device. | |
1319 | */ | |
1320 | handle_bad_sector(bio); | |
1321 | return 1; | |
1322 | } | |
1323 | } | |
1324 | ||
1325 | return 0; | |
1326 | } | |
1327 | ||
1da177e4 | 1328 | /** |
710027a4 | 1329 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1330 | * @bio: The bio describing the location in memory and on the device. |
1331 | * | |
1332 | * generic_make_request() is used to make I/O requests of block | |
1333 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1334 | * to be done. | |
1335 | * | |
1336 | * generic_make_request() does not return any status. The | |
1337 | * success/failure status of the request, along with notification of | |
1338 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1339 | * function described (one day) else where. | |
1340 | * | |
1341 | * The caller of generic_make_request must make sure that bi_io_vec | |
1342 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1343 | * set to describe the device address, and the | |
1344 | * bi_end_io and optionally bi_private are set to describe how | |
1345 | * completion notification should be signaled. | |
1346 | * | |
1347 | * generic_make_request and the drivers it calls may use bi_next if this | |
1348 | * bio happens to be merged with someone else, and may change bi_dev and | |
1349 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1350 | * should NOT be depended on after the call to generic_make_request. | |
1351 | */ | |
d89d8796 | 1352 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1353 | { |
165125e1 | 1354 | struct request_queue *q; |
5ddfe969 | 1355 | sector_t old_sector; |
1da177e4 | 1356 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1357 | dev_t old_dev; |
51fd77bd | 1358 | int err = -EIO; |
1da177e4 LT |
1359 | |
1360 | might_sleep(); | |
1da177e4 | 1361 | |
c07e2b41 JA |
1362 | if (bio_check_eod(bio, nr_sectors)) |
1363 | goto end_io; | |
1da177e4 LT |
1364 | |
1365 | /* | |
1366 | * Resolve the mapping until finished. (drivers are | |
1367 | * still free to implement/resolve their own stacking | |
1368 | * by explicitly returning 0) | |
1369 | * | |
1370 | * NOTE: we don't repeat the blk_size check for each new device. | |
1371 | * Stacking drivers are expected to know what they are doing. | |
1372 | */ | |
5ddfe969 | 1373 | old_sector = -1; |
2056a782 | 1374 | old_dev = 0; |
1da177e4 LT |
1375 | do { |
1376 | char b[BDEVNAME_SIZE]; | |
1377 | ||
1378 | q = bdev_get_queue(bio->bi_bdev); | |
a7384677 | 1379 | if (unlikely(!q)) { |
1da177e4 LT |
1380 | printk(KERN_ERR |
1381 | "generic_make_request: Trying to access " | |
1382 | "nonexistent block-device %s (%Lu)\n", | |
1383 | bdevname(bio->bi_bdev, b), | |
1384 | (long long) bio->bi_sector); | |
a7384677 | 1385 | goto end_io; |
1da177e4 LT |
1386 | } |
1387 | ||
4fa253f3 | 1388 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1389 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1390 | bdevname(bio->bi_bdev, b), |
1391 | bio_sectors(bio), | |
1392 | q->max_hw_sectors); | |
1393 | goto end_io; | |
1394 | } | |
1395 | ||
fde6ad22 | 1396 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1397 | goto end_io; |
1398 | ||
c17bb495 AM |
1399 | if (should_fail_request(bio)) |
1400 | goto end_io; | |
1401 | ||
1da177e4 LT |
1402 | /* |
1403 | * If this device has partitions, remap block n | |
1404 | * of partition p to block n+start(p) of the disk. | |
1405 | */ | |
1406 | blk_partition_remap(bio); | |
1407 | ||
7ba1ba12 MP |
1408 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1409 | goto end_io; | |
1410 | ||
5ddfe969 | 1411 | if (old_sector != -1) |
5f3ea37c | 1412 | trace_block_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1413 | old_sector); |
2056a782 | 1414 | |
5f3ea37c | 1415 | trace_block_bio_queue(q, bio); |
2056a782 | 1416 | |
5ddfe969 | 1417 | old_sector = bio->bi_sector; |
2056a782 JA |
1418 | old_dev = bio->bi_bdev->bd_dev; |
1419 | ||
c07e2b41 JA |
1420 | if (bio_check_eod(bio, nr_sectors)) |
1421 | goto end_io; | |
a7384677 TH |
1422 | |
1423 | if (bio_discard(bio) && !q->prepare_discard_fn) { | |
51fd77bd JA |
1424 | err = -EOPNOTSUPP; |
1425 | goto end_io; | |
1426 | } | |
cec0707e JA |
1427 | if (bio_barrier(bio) && bio_has_data(bio) && |
1428 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1429 | err = -EOPNOTSUPP; | |
1430 | goto end_io; | |
1431 | } | |
5ddfe969 | 1432 | |
1da177e4 LT |
1433 | ret = q->make_request_fn(q, bio); |
1434 | } while (ret); | |
a7384677 TH |
1435 | |
1436 | return; | |
1437 | ||
1438 | end_io: | |
1439 | bio_endio(bio, err); | |
1da177e4 LT |
1440 | } |
1441 | ||
d89d8796 NB |
1442 | /* |
1443 | * We only want one ->make_request_fn to be active at a time, | |
1444 | * else stack usage with stacked devices could be a problem. | |
1445 | * So use current->bio_{list,tail} to keep a list of requests | |
1446 | * submited by a make_request_fn function. | |
1447 | * current->bio_tail is also used as a flag to say if | |
1448 | * generic_make_request is currently active in this task or not. | |
1449 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1450 | * then a make_request is active, and new requests should be added | |
1451 | * at the tail | |
1452 | */ | |
1453 | void generic_make_request(struct bio *bio) | |
1454 | { | |
1455 | if (current->bio_tail) { | |
1456 | /* make_request is active */ | |
1457 | *(current->bio_tail) = bio; | |
1458 | bio->bi_next = NULL; | |
1459 | current->bio_tail = &bio->bi_next; | |
1460 | return; | |
1461 | } | |
1462 | /* following loop may be a bit non-obvious, and so deserves some | |
1463 | * explanation. | |
1464 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1465 | * ensure that) so we have a list with a single bio. | |
1466 | * We pretend that we have just taken it off a longer list, so | |
1467 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1468 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1469 | * added. __generic_make_request may indeed add some more bios | |
1470 | * through a recursive call to generic_make_request. If it | |
1471 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1472 | * from the top. In this case we really did just take the bio | |
1473 | * of the top of the list (no pretending) and so fixup bio_list and | |
1474 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1475 | * | |
1476 | * The loop was structured like this to make only one call to | |
1477 | * __generic_make_request (which is important as it is large and | |
1478 | * inlined) and to keep the structure simple. | |
1479 | */ | |
1480 | BUG_ON(bio->bi_next); | |
1481 | do { | |
1482 | current->bio_list = bio->bi_next; | |
1483 | if (bio->bi_next == NULL) | |
1484 | current->bio_tail = ¤t->bio_list; | |
1485 | else | |
1486 | bio->bi_next = NULL; | |
1487 | __generic_make_request(bio); | |
1488 | bio = current->bio_list; | |
1489 | } while (bio); | |
1490 | current->bio_tail = NULL; /* deactivate */ | |
1491 | } | |
1da177e4 LT |
1492 | EXPORT_SYMBOL(generic_make_request); |
1493 | ||
1494 | /** | |
710027a4 | 1495 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1496 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1497 | * @bio: The &struct bio which describes the I/O | |
1498 | * | |
1499 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1500 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1501 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1502 | * |
1503 | */ | |
1504 | void submit_bio(int rw, struct bio *bio) | |
1505 | { | |
1506 | int count = bio_sectors(bio); | |
1507 | ||
22e2c507 | 1508 | bio->bi_rw |= rw; |
1da177e4 | 1509 | |
bf2de6f5 JA |
1510 | /* |
1511 | * If it's a regular read/write or a barrier with data attached, | |
1512 | * go through the normal accounting stuff before submission. | |
1513 | */ | |
a9c701e5 | 1514 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1515 | if (rw & WRITE) { |
1516 | count_vm_events(PGPGOUT, count); | |
1517 | } else { | |
1518 | task_io_account_read(bio->bi_size); | |
1519 | count_vm_events(PGPGIN, count); | |
1520 | } | |
1521 | ||
1522 | if (unlikely(block_dump)) { | |
1523 | char b[BDEVNAME_SIZE]; | |
1524 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1525 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1526 | (rw & WRITE) ? "WRITE" : "READ", |
1527 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1528 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1529 | } |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | generic_make_request(bio); | |
1533 | } | |
1da177e4 LT |
1534 | EXPORT_SYMBOL(submit_bio); |
1535 | ||
82124d60 KU |
1536 | /** |
1537 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1538 | * @q: the queue | |
1539 | * @rq: the request being checked | |
1540 | * | |
1541 | * Description: | |
1542 | * @rq may have been made based on weaker limitations of upper-level queues | |
1543 | * in request stacking drivers, and it may violate the limitation of @q. | |
1544 | * Since the block layer and the underlying device driver trust @rq | |
1545 | * after it is inserted to @q, it should be checked against @q before | |
1546 | * the insertion using this generic function. | |
1547 | * | |
1548 | * This function should also be useful for request stacking drivers | |
1549 | * in some cases below, so export this fuction. | |
1550 | * Request stacking drivers like request-based dm may change the queue | |
1551 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1552 | * Such request stacking drivers should check those requests agaist | |
1553 | * the new queue limits again when they dispatch those requests, | |
1554 | * although such checkings are also done against the old queue limits | |
1555 | * when submitting requests. | |
1556 | */ | |
1557 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1558 | { | |
83096ebf | 1559 | if (blk_rq_sectors(rq) > q->max_sectors || |
82124d60 KU |
1560 | rq->data_len > q->max_hw_sectors << 9) { |
1561 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1562 | return -EIO; | |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * queue's settings related to segment counting like q->bounce_pfn | |
1567 | * may differ from that of other stacking queues. | |
1568 | * Recalculate it to check the request correctly on this queue's | |
1569 | * limitation. | |
1570 | */ | |
1571 | blk_recalc_rq_segments(rq); | |
1572 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1573 | rq->nr_phys_segments > q->max_hw_segments) { | |
1574 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1575 | return -EIO; | |
1576 | } | |
1577 | ||
1578 | return 0; | |
1579 | } | |
1580 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1581 | ||
1582 | /** | |
1583 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1584 | * @q: the queue to submit the request | |
1585 | * @rq: the request being queued | |
1586 | */ | |
1587 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1588 | { | |
1589 | unsigned long flags; | |
1590 | ||
1591 | if (blk_rq_check_limits(q, rq)) | |
1592 | return -EIO; | |
1593 | ||
1594 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1595 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1596 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1597 | return -EIO; | |
1598 | #endif | |
1599 | ||
1600 | spin_lock_irqsave(q->queue_lock, flags); | |
1601 | ||
1602 | /* | |
1603 | * Submitting request must be dequeued before calling this function | |
1604 | * because it will be linked to another request_queue | |
1605 | */ | |
1606 | BUG_ON(blk_queued_rq(rq)); | |
1607 | ||
1608 | drive_stat_acct(rq, 1); | |
1609 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1610 | ||
1611 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1612 | ||
1613 | return 0; | |
1614 | } | |
1615 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1616 | ||
53a08807 TH |
1617 | /** |
1618 | * blkdev_dequeue_request - dequeue request and start timeout timer | |
1619 | * @req: request to dequeue | |
1620 | * | |
1621 | * Dequeue @req and start timeout timer on it. This hands off the | |
1622 | * request to the driver. | |
1623 | * | |
1624 | * Block internal functions which don't want to start timer should | |
1625 | * call elv_dequeue_request(). | |
1626 | */ | |
1627 | void blkdev_dequeue_request(struct request *req) | |
1628 | { | |
1629 | elv_dequeue_request(req->q, req); | |
1630 | ||
1631 | /* | |
1632 | * We are now handing the request to the hardware, add the | |
1633 | * timeout handler. | |
1634 | */ | |
1635 | blk_add_timer(req); | |
1636 | } | |
1637 | EXPORT_SYMBOL(blkdev_dequeue_request); | |
1638 | ||
bc58ba94 JA |
1639 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1640 | { | |
c2553b58 | 1641 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1642 | const int rw = rq_data_dir(req); |
1643 | struct hd_struct *part; | |
1644 | int cpu; | |
1645 | ||
1646 | cpu = part_stat_lock(); | |
83096ebf | 1647 | part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); |
bc58ba94 JA |
1648 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1649 | part_stat_unlock(); | |
1650 | } | |
1651 | } | |
1652 | ||
1653 | static void blk_account_io_done(struct request *req) | |
1654 | { | |
bc58ba94 JA |
1655 | /* |
1656 | * Account IO completion. bar_rq isn't accounted as a normal | |
1657 | * IO on queueing nor completion. Accounting the containing | |
1658 | * request is enough. | |
1659 | */ | |
c2553b58 | 1660 | if (blk_do_io_stat(req) && req != &req->q->bar_rq) { |
bc58ba94 JA |
1661 | unsigned long duration = jiffies - req->start_time; |
1662 | const int rw = rq_data_dir(req); | |
1663 | struct hd_struct *part; | |
1664 | int cpu; | |
1665 | ||
1666 | cpu = part_stat_lock(); | |
83096ebf | 1667 | part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); |
bc58ba94 JA |
1668 | |
1669 | part_stat_inc(cpu, part, ios[rw]); | |
1670 | part_stat_add(cpu, part, ticks[rw], duration); | |
1671 | part_round_stats(cpu, part); | |
1672 | part_dec_in_flight(part); | |
1673 | ||
1674 | part_stat_unlock(); | |
1675 | } | |
1676 | } | |
1677 | ||
5efccd17 TH |
1678 | /** |
1679 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
1680 | * @rq: the request being processed | |
1681 | **/ | |
1682 | unsigned int blk_rq_bytes(struct request *rq) | |
1683 | { | |
1684 | if (blk_fs_request(rq)) | |
5b93629b | 1685 | return blk_rq_sectors(rq) << 9; |
5efccd17 TH |
1686 | |
1687 | return rq->data_len; | |
1688 | } | |
1689 | EXPORT_SYMBOL_GPL(blk_rq_bytes); | |
1690 | ||
1691 | /** | |
1692 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
1693 | * @rq: the request being processed | |
1694 | **/ | |
1695 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1696 | { | |
1697 | if (blk_fs_request(rq)) | |
1698 | return rq->current_nr_sectors << 9; | |
1699 | ||
1700 | if (rq->bio) | |
1701 | return rq->bio->bi_size; | |
1702 | ||
1703 | return rq->data_len; | |
1704 | } | |
1705 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
1706 | ||
158dbda0 TH |
1707 | struct request *elv_next_request(struct request_queue *q) |
1708 | { | |
1709 | struct request *rq; | |
1710 | int ret; | |
1711 | ||
1712 | while ((rq = __elv_next_request(q)) != NULL) { | |
1713 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1714 | /* | |
1715 | * This is the first time the device driver | |
1716 | * sees this request (possibly after | |
1717 | * requeueing). Notify IO scheduler. | |
1718 | */ | |
1719 | if (blk_sorted_rq(rq)) | |
1720 | elv_activate_rq(q, rq); | |
1721 | ||
1722 | /* | |
1723 | * just mark as started even if we don't start | |
1724 | * it, a request that has been delayed should | |
1725 | * not be passed by new incoming requests | |
1726 | */ | |
1727 | rq->cmd_flags |= REQ_STARTED; | |
1728 | trace_block_rq_issue(q, rq); | |
1729 | } | |
1730 | ||
1731 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1732 | q->end_sector = rq_end_sector(rq); | |
1733 | q->boundary_rq = NULL; | |
1734 | } | |
1735 | ||
1736 | if (rq->cmd_flags & REQ_DONTPREP) | |
1737 | break; | |
1738 | ||
1739 | if (q->dma_drain_size && rq->data_len) { | |
1740 | /* | |
1741 | * make sure space for the drain appears we | |
1742 | * know we can do this because max_hw_segments | |
1743 | * has been adjusted to be one fewer than the | |
1744 | * device can handle | |
1745 | */ | |
1746 | rq->nr_phys_segments++; | |
1747 | } | |
1748 | ||
1749 | if (!q->prep_rq_fn) | |
1750 | break; | |
1751 | ||
1752 | ret = q->prep_rq_fn(q, rq); | |
1753 | if (ret == BLKPREP_OK) { | |
1754 | break; | |
1755 | } else if (ret == BLKPREP_DEFER) { | |
1756 | /* | |
1757 | * the request may have been (partially) prepped. | |
1758 | * we need to keep this request in the front to | |
1759 | * avoid resource deadlock. REQ_STARTED will | |
1760 | * prevent other fs requests from passing this one. | |
1761 | */ | |
1762 | if (q->dma_drain_size && rq->data_len && | |
1763 | !(rq->cmd_flags & REQ_DONTPREP)) { | |
1764 | /* | |
1765 | * remove the space for the drain we added | |
1766 | * so that we don't add it again | |
1767 | */ | |
1768 | --rq->nr_phys_segments; | |
1769 | } | |
1770 | ||
1771 | rq = NULL; | |
1772 | break; | |
1773 | } else if (ret == BLKPREP_KILL) { | |
1774 | rq->cmd_flags |= REQ_QUIET; | |
40cbbb78 | 1775 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
1776 | } else { |
1777 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1778 | break; | |
1779 | } | |
1780 | } | |
1781 | ||
1782 | return rq; | |
1783 | } | |
1784 | EXPORT_SYMBOL(elv_next_request); | |
1785 | ||
1786 | void elv_dequeue_request(struct request_queue *q, struct request *rq) | |
1787 | { | |
1788 | BUG_ON(list_empty(&rq->queuelist)); | |
1789 | BUG_ON(ELV_ON_HASH(rq)); | |
1790 | ||
1791 | list_del_init(&rq->queuelist); | |
1792 | ||
1793 | /* | |
1794 | * the time frame between a request being removed from the lists | |
1795 | * and to it is freed is accounted as io that is in progress at | |
1796 | * the driver side. | |
1797 | */ | |
1798 | if (blk_account_rq(rq)) | |
1799 | q->in_flight++; | |
1800 | } | |
1801 | ||
3bcddeac | 1802 | /** |
2e60e022 TH |
1803 | * blk_update_request - Special helper function for request stacking drivers |
1804 | * @rq: the request being processed | |
710027a4 | 1805 | * @error: %0 for success, < %0 for error |
2e60e022 | 1806 | * @nr_bytes: number of bytes to complete @rq |
3bcddeac KU |
1807 | * |
1808 | * Description: | |
2e60e022 TH |
1809 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete |
1810 | * the request structure even if @rq doesn't have leftover. | |
1811 | * If @rq has leftover, sets it up for the next range of segments. | |
1812 | * | |
1813 | * This special helper function is only for request stacking drivers | |
1814 | * (e.g. request-based dm) so that they can handle partial completion. | |
1815 | * Actual device drivers should use blk_end_request instead. | |
1816 | * | |
1817 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1818 | * %false return from this function. | |
3bcddeac KU |
1819 | * |
1820 | * Return: | |
2e60e022 TH |
1821 | * %false - this request doesn't have any more data |
1822 | * %true - this request has more data | |
3bcddeac | 1823 | **/ |
2e60e022 | 1824 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 1825 | { |
5450d3e1 | 1826 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1827 | struct bio *bio; |
1828 | ||
2e60e022 TH |
1829 | if (!req->bio) |
1830 | return false; | |
1831 | ||
5f3ea37c | 1832 | trace_block_rq_complete(req->q, req); |
2056a782 | 1833 | |
1da177e4 | 1834 | /* |
924cec77 TH |
1835 | * For fs requests, rq is just carrier of independent bio's |
1836 | * and each partial completion should be handled separately. | |
1837 | * Reset per-request error on each partial completion. | |
1838 | * | |
1839 | * TODO: tj: This is too subtle. It would be better to let | |
1840 | * low level drivers do what they see fit. | |
1da177e4 | 1841 | */ |
924cec77 | 1842 | if (blk_fs_request(req)) |
1da177e4 LT |
1843 | req->errors = 0; |
1844 | ||
6728cb0e JA |
1845 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1846 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 | 1847 | req->rq_disk ? req->rq_disk->disk_name : "?", |
83096ebf | 1848 | (unsigned long long)blk_rq_pos(req)); |
1da177e4 LT |
1849 | } |
1850 | ||
bc58ba94 | 1851 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1852 | |
1da177e4 LT |
1853 | total_bytes = bio_nbytes = 0; |
1854 | while ((bio = req->bio) != NULL) { | |
1855 | int nbytes; | |
1856 | ||
1857 | if (nr_bytes >= bio->bi_size) { | |
1858 | req->bio = bio->bi_next; | |
1859 | nbytes = bio->bi_size; | |
5bb23a68 | 1860 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1861 | next_idx = 0; |
1862 | bio_nbytes = 0; | |
1863 | } else { | |
1864 | int idx = bio->bi_idx + next_idx; | |
1865 | ||
1866 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1867 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e | 1868 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1869 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
1870 | break; |
1871 | } | |
1872 | ||
1873 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1874 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1875 | ||
1876 | /* | |
1877 | * not a complete bvec done | |
1878 | */ | |
1879 | if (unlikely(nbytes > nr_bytes)) { | |
1880 | bio_nbytes += nr_bytes; | |
1881 | total_bytes += nr_bytes; | |
1882 | break; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * advance to the next vector | |
1887 | */ | |
1888 | next_idx++; | |
1889 | bio_nbytes += nbytes; | |
1890 | } | |
1891 | ||
1892 | total_bytes += nbytes; | |
1893 | nr_bytes -= nbytes; | |
1894 | ||
6728cb0e JA |
1895 | bio = req->bio; |
1896 | if (bio) { | |
1da177e4 LT |
1897 | /* |
1898 | * end more in this run, or just return 'not-done' | |
1899 | */ | |
1900 | if (unlikely(nr_bytes <= 0)) | |
1901 | break; | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * completely done | |
1907 | */ | |
2e60e022 TH |
1908 | if (!req->bio) { |
1909 | /* | |
1910 | * Reset counters so that the request stacking driver | |
1911 | * can find how many bytes remain in the request | |
1912 | * later. | |
1913 | */ | |
1914 | req->nr_sectors = req->hard_nr_sectors = 0; | |
1915 | req->current_nr_sectors = req->hard_cur_sectors = 0; | |
1916 | return false; | |
1917 | } | |
1da177e4 LT |
1918 | |
1919 | /* | |
1920 | * if the request wasn't completed, update state | |
1921 | */ | |
1922 | if (bio_nbytes) { | |
5bb23a68 | 1923 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1924 | bio->bi_idx += next_idx; |
1925 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1926 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1927 | } | |
1928 | ||
1929 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1930 | blk_recalc_rq_segments(req); | |
2e60e022 | 1931 | return true; |
1da177e4 | 1932 | } |
2e60e022 | 1933 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 1934 | |
2e60e022 TH |
1935 | static bool blk_update_bidi_request(struct request *rq, int error, |
1936 | unsigned int nr_bytes, | |
1937 | unsigned int bidi_bytes) | |
5efccd17 | 1938 | { |
2e60e022 TH |
1939 | if (blk_update_request(rq, error, nr_bytes)) |
1940 | return true; | |
5efccd17 | 1941 | |
2e60e022 TH |
1942 | /* Bidi request must be completed as a whole */ |
1943 | if (unlikely(blk_bidi_rq(rq)) && | |
1944 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
1945 | return true; | |
5efccd17 | 1946 | |
2e60e022 TH |
1947 | add_disk_randomness(rq->rq_disk); |
1948 | ||
1949 | return false; | |
5efccd17 TH |
1950 | } |
1951 | ||
1da177e4 LT |
1952 | /* |
1953 | * queue lock must be held | |
1954 | */ | |
2e60e022 | 1955 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 1956 | { |
b8286239 KU |
1957 | if (blk_rq_tagged(req)) |
1958 | blk_queue_end_tag(req->q, req); | |
1959 | ||
1960 | if (blk_queued_rq(req)) | |
53a08807 | 1961 | elv_dequeue_request(req->q, req); |
1da177e4 LT |
1962 | |
1963 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1964 | laptop_io_completion(); | |
1965 | ||
e78042e5 MA |
1966 | blk_delete_timer(req); |
1967 | ||
bc58ba94 | 1968 | blk_account_io_done(req); |
b8286239 | 1969 | |
1da177e4 | 1970 | if (req->end_io) |
8ffdc655 | 1971 | req->end_io(req, error); |
b8286239 KU |
1972 | else { |
1973 | if (blk_bidi_rq(req)) | |
1974 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1975 | ||
1da177e4 | 1976 | __blk_put_request(req->q, req); |
b8286239 | 1977 | } |
1da177e4 LT |
1978 | } |
1979 | ||
336cdb40 | 1980 | /** |
2e60e022 TH |
1981 | * blk_end_bidi_request - Complete a bidi request |
1982 | * @rq: the request to complete | |
1983 | * @error: %0 for success, < %0 for error | |
1984 | * @nr_bytes: number of bytes to complete @rq | |
1985 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
1986 | * |
1987 | * Description: | |
e3a04fe3 | 1988 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
1989 | * Drivers that supports bidi can safely call this member for any |
1990 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
1991 | * just ignored. | |
336cdb40 KU |
1992 | * |
1993 | * Return: | |
2e60e022 TH |
1994 | * %false - we are done with this request |
1995 | * %true - still buffers pending for this request | |
336cdb40 | 1996 | **/ |
2e60e022 TH |
1997 | bool blk_end_bidi_request(struct request *rq, int error, |
1998 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
336cdb40 KU |
1999 | { |
2000 | struct request_queue *q = rq->q; | |
2e60e022 | 2001 | unsigned long flags; |
336cdb40 | 2002 | |
2e60e022 TH |
2003 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2004 | return true; | |
336cdb40 KU |
2005 | |
2006 | spin_lock_irqsave(q->queue_lock, flags); | |
2e60e022 | 2007 | blk_finish_request(rq, error); |
336cdb40 KU |
2008 | spin_unlock_irqrestore(q->queue_lock, flags); |
2009 | ||
2e60e022 | 2010 | return false; |
336cdb40 | 2011 | } |
2e60e022 | 2012 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); |
336cdb40 | 2013 | |
e3a04fe3 | 2014 | /** |
2e60e022 TH |
2015 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2016 | * @rq: the request to complete | |
710027a4 | 2017 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2018 | * @nr_bytes: number of bytes to complete @rq |
2019 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2020 | * | |
2021 | * Description: | |
2e60e022 TH |
2022 | * Identical to blk_end_bidi_request() except that queue lock is |
2023 | * assumed to be locked on entry and remains so on return. | |
e3a04fe3 KU |
2024 | * |
2025 | * Return: | |
2e60e022 TH |
2026 | * %false - we are done with this request |
2027 | * %true - still buffers pending for this request | |
5efccd17 | 2028 | **/ |
2e60e022 TH |
2029 | bool __blk_end_bidi_request(struct request *rq, int error, |
2030 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
5efccd17 | 2031 | { |
2e60e022 TH |
2032 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2033 | return true; | |
5efccd17 | 2034 | |
2e60e022 | 2035 | blk_finish_request(rq, error); |
5efccd17 | 2036 | |
2e60e022 | 2037 | return false; |
32fab448 | 2038 | } |
2e60e022 | 2039 | EXPORT_SYMBOL_GPL(__blk_end_bidi_request); |
32fab448 | 2040 | |
86db1e29 JA |
2041 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2042 | struct bio *bio) | |
1da177e4 | 2043 | { |
d628eaef DW |
2044 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2045 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2046 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2047 | |
fb2dce86 DW |
2048 | if (bio_has_data(bio)) { |
2049 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2050 | rq->buffer = bio_data(bio); |
2051 | } | |
1da177e4 LT |
2052 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2053 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2054 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2055 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2056 | |
2057 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2058 | |
66846572 N |
2059 | if (bio->bi_bdev) |
2060 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2061 | } | |
1da177e4 | 2062 | |
ef9e3fac KU |
2063 | /** |
2064 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2065 | * @q : the queue of the device being checked | |
2066 | * | |
2067 | * Description: | |
2068 | * Check if underlying low-level drivers of a device are busy. | |
2069 | * If the drivers want to export their busy state, they must set own | |
2070 | * exporting function using blk_queue_lld_busy() first. | |
2071 | * | |
2072 | * Basically, this function is used only by request stacking drivers | |
2073 | * to stop dispatching requests to underlying devices when underlying | |
2074 | * devices are busy. This behavior helps more I/O merging on the queue | |
2075 | * of the request stacking driver and prevents I/O throughput regression | |
2076 | * on burst I/O load. | |
2077 | * | |
2078 | * Return: | |
2079 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2080 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2081 | */ | |
2082 | int blk_lld_busy(struct request_queue *q) | |
2083 | { | |
2084 | if (q->lld_busy_fn) | |
2085 | return q->lld_busy_fn(q); | |
2086 | ||
2087 | return 0; | |
2088 | } | |
2089 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2090 | ||
18887ad9 | 2091 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2092 | { |
2093 | return queue_work(kblockd_workqueue, work); | |
2094 | } | |
1da177e4 LT |
2095 | EXPORT_SYMBOL(kblockd_schedule_work); |
2096 | ||
1da177e4 LT |
2097 | int __init blk_dev_init(void) |
2098 | { | |
9eb55b03 NK |
2099 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2100 | sizeof(((struct request *)0)->cmd_flags)); | |
2101 | ||
1da177e4 LT |
2102 | kblockd_workqueue = create_workqueue("kblockd"); |
2103 | if (!kblockd_workqueue) | |
2104 | panic("Failed to create kblockd\n"); | |
2105 | ||
2106 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2107 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2108 | |
8324aa91 | 2109 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2110 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2111 | |
d38ecf93 | 2112 | return 0; |
1da177e4 | 2113 | } |
1da177e4 | 2114 |