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 JA |
70 | |
71 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
72 | return; | |
73 | ||
074a7aca | 74 | cpu = part_stat_lock(); |
e71bf0d0 | 75 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
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; |
63a71386 | 134 | rq->tag = -1; |
1da177e4 | 135 | rq->ref_count = 1; |
1da177e4 | 136 | } |
2a4aa30c | 137 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 138 | |
5bb23a68 N |
139 | static void req_bio_endio(struct request *rq, struct bio *bio, |
140 | unsigned int nbytes, int error) | |
1da177e4 | 141 | { |
165125e1 | 142 | struct request_queue *q = rq->q; |
797e7dbb | 143 | |
5bb23a68 N |
144 | if (&q->bar_rq != rq) { |
145 | if (error) | |
146 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
147 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
148 | error = -EIO; | |
797e7dbb | 149 | |
5bb23a68 | 150 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 151 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 152 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
153 | nbytes = bio->bi_size; |
154 | } | |
797e7dbb | 155 | |
5bb23a68 N |
156 | bio->bi_size -= nbytes; |
157 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
158 | |
159 | if (bio_integrity(bio)) | |
160 | bio_integrity_advance(bio, nbytes); | |
161 | ||
5bb23a68 | 162 | if (bio->bi_size == 0) |
6712ecf8 | 163 | bio_endio(bio, error); |
5bb23a68 N |
164 | } else { |
165 | ||
166 | /* | |
167 | * Okay, this is the barrier request in progress, just | |
168 | * record the error; | |
169 | */ | |
170 | if (error && !q->orderr) | |
171 | q->orderr = error; | |
172 | } | |
1da177e4 | 173 | } |
1da177e4 | 174 | |
1da177e4 LT |
175 | void blk_dump_rq_flags(struct request *rq, char *msg) |
176 | { | |
177 | int bit; | |
178 | ||
6728cb0e | 179 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
180 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
181 | rq->cmd_flags); | |
1da177e4 | 182 | |
6728cb0e JA |
183 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
184 | (unsigned long long)rq->sector, | |
185 | rq->nr_sectors, | |
186 | rq->current_nr_sectors); | |
187 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
188 | rq->bio, rq->biotail, | |
189 | rq->buffer, rq->data, | |
190 | rq->data_len); | |
1da177e4 | 191 | |
4aff5e23 | 192 | if (blk_pc_request(rq)) { |
6728cb0e | 193 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 194 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
195 | printk("%02x ", rq->cmd[bit]); |
196 | printk("\n"); | |
197 | } | |
198 | } | |
1da177e4 LT |
199 | EXPORT_SYMBOL(blk_dump_rq_flags); |
200 | ||
1da177e4 LT |
201 | /* |
202 | * "plug" the device if there are no outstanding requests: this will | |
203 | * force the transfer to start only after we have put all the requests | |
204 | * on the list. | |
205 | * | |
206 | * This is called with interrupts off and no requests on the queue and | |
207 | * with the queue lock held. | |
208 | */ | |
165125e1 | 209 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
210 | { |
211 | WARN_ON(!irqs_disabled()); | |
212 | ||
213 | /* | |
214 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
215 | * which will restart the queueing | |
216 | */ | |
7daac490 | 217 | if (blk_queue_stopped(q)) |
1da177e4 LT |
218 | return; |
219 | ||
e48ec690 | 220 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 221 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 222 | trace_block_plug(q); |
2056a782 | 223 | } |
1da177e4 | 224 | } |
1da177e4 LT |
225 | EXPORT_SYMBOL(blk_plug_device); |
226 | ||
6c5e0c4d JA |
227 | /** |
228 | * blk_plug_device_unlocked - plug a device without queue lock held | |
229 | * @q: The &struct request_queue to plug | |
230 | * | |
231 | * Description: | |
232 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
233 | * interrupts. | |
234 | **/ | |
235 | void blk_plug_device_unlocked(struct request_queue *q) | |
236 | { | |
237 | unsigned long flags; | |
238 | ||
239 | spin_lock_irqsave(q->queue_lock, flags); | |
240 | blk_plug_device(q); | |
241 | spin_unlock_irqrestore(q->queue_lock, flags); | |
242 | } | |
243 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
244 | ||
1da177e4 LT |
245 | /* |
246 | * remove the queue from the plugged list, if present. called with | |
247 | * queue lock held and interrupts disabled. | |
248 | */ | |
165125e1 | 249 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
250 | { |
251 | WARN_ON(!irqs_disabled()); | |
252 | ||
e48ec690 | 253 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
254 | return 0; |
255 | ||
256 | del_timer(&q->unplug_timer); | |
257 | return 1; | |
258 | } | |
1da177e4 LT |
259 | EXPORT_SYMBOL(blk_remove_plug); |
260 | ||
261 | /* | |
262 | * remove the plug and let it rip.. | |
263 | */ | |
165125e1 | 264 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 265 | { |
7daac490 | 266 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
267 | return; |
268 | ||
269 | if (!blk_remove_plug(q)) | |
270 | return; | |
271 | ||
22e2c507 | 272 | q->request_fn(q); |
1da177e4 | 273 | } |
1da177e4 LT |
274 | |
275 | /** | |
276 | * generic_unplug_device - fire a request queue | |
165125e1 | 277 | * @q: The &struct request_queue in question |
1da177e4 LT |
278 | * |
279 | * Description: | |
280 | * Linux uses plugging to build bigger requests queues before letting | |
281 | * the device have at them. If a queue is plugged, the I/O scheduler | |
282 | * is still adding and merging requests on the queue. Once the queue | |
283 | * gets unplugged, the request_fn defined for the queue is invoked and | |
284 | * transfers started. | |
285 | **/ | |
165125e1 | 286 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 287 | { |
dbaf2c00 JA |
288 | if (blk_queue_plugged(q)) { |
289 | spin_lock_irq(q->queue_lock); | |
290 | __generic_unplug_device(q); | |
291 | spin_unlock_irq(q->queue_lock); | |
292 | } | |
1da177e4 LT |
293 | } |
294 | EXPORT_SYMBOL(generic_unplug_device); | |
295 | ||
296 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
297 | struct page *page) | |
298 | { | |
165125e1 | 299 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 300 | |
2ad8b1ef | 301 | blk_unplug(q); |
1da177e4 LT |
302 | } |
303 | ||
86db1e29 | 304 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 305 | { |
165125e1 JA |
306 | struct request_queue *q = |
307 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 308 | |
5f3ea37c | 309 | trace_block_unplug_io(q); |
1da177e4 LT |
310 | q->unplug_fn(q); |
311 | } | |
312 | ||
86db1e29 | 313 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 314 | { |
165125e1 | 315 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 316 | |
5f3ea37c | 317 | trace_block_unplug_timer(q); |
18887ad9 | 318 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
319 | } |
320 | ||
2ad8b1ef AB |
321 | void blk_unplug(struct request_queue *q) |
322 | { | |
323 | /* | |
324 | * devices don't necessarily have an ->unplug_fn defined | |
325 | */ | |
326 | if (q->unplug_fn) { | |
5f3ea37c | 327 | trace_block_unplug_io(q); |
2ad8b1ef AB |
328 | q->unplug_fn(q); |
329 | } | |
330 | } | |
331 | EXPORT_SYMBOL(blk_unplug); | |
332 | ||
c7c22e4d JA |
333 | static void blk_invoke_request_fn(struct request_queue *q) |
334 | { | |
80a4b58e JA |
335 | if (unlikely(blk_queue_stopped(q))) |
336 | return; | |
337 | ||
c7c22e4d JA |
338 | /* |
339 | * one level of recursion is ok and is much faster than kicking | |
340 | * the unplug handling | |
341 | */ | |
342 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
343 | q->request_fn(q); | |
344 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
345 | } else { | |
346 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
347 | kblockd_schedule_work(q, &q->unplug_work); | |
348 | } | |
349 | } | |
350 | ||
1da177e4 LT |
351 | /** |
352 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 353 | * @q: The &struct request_queue in question |
1da177e4 LT |
354 | * |
355 | * Description: | |
356 | * blk_start_queue() will clear the stop flag on the queue, and call | |
357 | * the request_fn for the queue if it was in a stopped state when | |
358 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
359 | **/ | |
165125e1 | 360 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 361 | { |
a038e253 PBG |
362 | WARN_ON(!irqs_disabled()); |
363 | ||
75ad23bc | 364 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
c7c22e4d | 365 | blk_invoke_request_fn(q); |
1da177e4 | 366 | } |
1da177e4 LT |
367 | EXPORT_SYMBOL(blk_start_queue); |
368 | ||
369 | /** | |
370 | * blk_stop_queue - stop a queue | |
165125e1 | 371 | * @q: The &struct request_queue in question |
1da177e4 LT |
372 | * |
373 | * Description: | |
374 | * The Linux block layer assumes that a block driver will consume all | |
375 | * entries on the request queue when the request_fn strategy is called. | |
376 | * Often this will not happen, because of hardware limitations (queue | |
377 | * depth settings). If a device driver gets a 'queue full' response, | |
378 | * or if it simply chooses not to queue more I/O at one point, it can | |
379 | * call this function to prevent the request_fn from being called until | |
380 | * the driver has signalled it's ready to go again. This happens by calling | |
381 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
382 | **/ | |
165125e1 | 383 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
384 | { |
385 | blk_remove_plug(q); | |
75ad23bc | 386 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
387 | } |
388 | EXPORT_SYMBOL(blk_stop_queue); | |
389 | ||
390 | /** | |
391 | * blk_sync_queue - cancel any pending callbacks on a queue | |
392 | * @q: the queue | |
393 | * | |
394 | * Description: | |
395 | * The block layer may perform asynchronous callback activity | |
396 | * on a queue, such as calling the unplug function after a timeout. | |
397 | * A block device may call blk_sync_queue to ensure that any | |
398 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 399 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
400 | * that its ->make_request_fn will not re-add plugging prior to calling |
401 | * this function. | |
402 | * | |
403 | */ | |
404 | void blk_sync_queue(struct request_queue *q) | |
405 | { | |
406 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 407 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
408 | } |
409 | EXPORT_SYMBOL(blk_sync_queue); | |
410 | ||
411 | /** | |
80a4b58e | 412 | * __blk_run_queue - run a single device queue |
1da177e4 | 413 | * @q: The queue to run |
80a4b58e JA |
414 | * |
415 | * Description: | |
416 | * See @blk_run_queue. This variant must be called with the queue lock | |
417 | * held and interrupts disabled. | |
418 | * | |
1da177e4 | 419 | */ |
75ad23bc | 420 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 421 | { |
1da177e4 | 422 | blk_remove_plug(q); |
dac07ec1 JA |
423 | |
424 | /* | |
425 | * Only recurse once to avoid overrunning the stack, let the unplug | |
426 | * handling reinvoke the handler shortly if we already got there. | |
427 | */ | |
c7c22e4d JA |
428 | if (!elv_queue_empty(q)) |
429 | blk_invoke_request_fn(q); | |
75ad23bc NP |
430 | } |
431 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 432 | |
75ad23bc NP |
433 | /** |
434 | * blk_run_queue - run a single device queue | |
435 | * @q: The queue to run | |
80a4b58e JA |
436 | * |
437 | * Description: | |
438 | * Invoke request handling on this queue, if it has pending work to do. | |
439 | * May be used to restart queueing when a request has completed. Also | |
440 | * See @blk_start_queueing. | |
441 | * | |
75ad23bc NP |
442 | */ |
443 | void blk_run_queue(struct request_queue *q) | |
444 | { | |
445 | unsigned long flags; | |
446 | ||
447 | spin_lock_irqsave(q->queue_lock, flags); | |
448 | __blk_run_queue(q); | |
1da177e4 LT |
449 | spin_unlock_irqrestore(q->queue_lock, flags); |
450 | } | |
451 | EXPORT_SYMBOL(blk_run_queue); | |
452 | ||
165125e1 | 453 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
454 | { |
455 | kobject_put(&q->kobj); | |
456 | } | |
483f4afc | 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 | ||
483f4afc | 468 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 469 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
470 | mutex_unlock(&q->sysfs_lock); |
471 | ||
472 | if (q->elevator) | |
473 | elevator_exit(q->elevator); | |
474 | ||
475 | blk_put_queue(q); | |
476 | } | |
1da177e4 LT |
477 | EXPORT_SYMBOL(blk_cleanup_queue); |
478 | ||
165125e1 | 479 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
480 | { |
481 | struct request_list *rl = &q->rq; | |
482 | ||
483 | rl->count[READ] = rl->count[WRITE] = 0; | |
484 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 485 | rl->elvpriv = 0; |
1da177e4 LT |
486 | init_waitqueue_head(&rl->wait[READ]); |
487 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 488 | |
1946089a CL |
489 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
490 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
491 | |
492 | if (!rl->rq_pool) | |
493 | return -ENOMEM; | |
494 | ||
495 | return 0; | |
496 | } | |
497 | ||
165125e1 | 498 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 499 | { |
1946089a CL |
500 | return blk_alloc_queue_node(gfp_mask, -1); |
501 | } | |
502 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 503 | |
165125e1 | 504 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 505 | { |
165125e1 | 506 | struct request_queue *q; |
e0bf68dd | 507 | int err; |
1946089a | 508 | |
8324aa91 | 509 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 510 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
511 | if (!q) |
512 | return NULL; | |
513 | ||
e0bf68dd PZ |
514 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
515 | q->backing_dev_info.unplug_io_data = q; | |
516 | err = bdi_init(&q->backing_dev_info); | |
517 | if (err) { | |
8324aa91 | 518 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
519 | return NULL; |
520 | } | |
521 | ||
1da177e4 | 522 | init_timer(&q->unplug_timer); |
242f9dcb JA |
523 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
524 | INIT_LIST_HEAD(&q->timeout_list); | |
713ada9b | 525 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 526 | |
8324aa91 | 527 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 528 | |
483f4afc | 529 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 530 | spin_lock_init(&q->__queue_lock); |
483f4afc | 531 | |
1da177e4 LT |
532 | return q; |
533 | } | |
1946089a | 534 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
535 | |
536 | /** | |
537 | * blk_init_queue - prepare a request queue for use with a block device | |
538 | * @rfn: The function to be called to process requests that have been | |
539 | * placed on the queue. | |
540 | * @lock: Request queue spin lock | |
541 | * | |
542 | * Description: | |
543 | * If a block device wishes to use the standard request handling procedures, | |
544 | * which sorts requests and coalesces adjacent requests, then it must | |
545 | * call blk_init_queue(). The function @rfn will be called when there | |
546 | * are requests on the queue that need to be processed. If the device | |
547 | * supports plugging, then @rfn may not be called immediately when requests | |
548 | * are available on the queue, but may be called at some time later instead. | |
549 | * Plugged queues are generally unplugged when a buffer belonging to one | |
550 | * of the requests on the queue is needed, or due to memory pressure. | |
551 | * | |
552 | * @rfn is not required, or even expected, to remove all requests off the | |
553 | * queue, but only as many as it can handle at a time. If it does leave | |
554 | * requests on the queue, it is responsible for arranging that the requests | |
555 | * get dealt with eventually. | |
556 | * | |
557 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
558 | * request queue; this lock will be taken also from interrupt context, so irq |
559 | * disabling is needed for it. | |
1da177e4 | 560 | * |
710027a4 | 561 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
562 | * it didn't succeed. |
563 | * | |
564 | * Note: | |
565 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
566 | * when the block device is deactivated (such as at module unload). | |
567 | **/ | |
1946089a | 568 | |
165125e1 | 569 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 570 | { |
1946089a CL |
571 | return blk_init_queue_node(rfn, lock, -1); |
572 | } | |
573 | EXPORT_SYMBOL(blk_init_queue); | |
574 | ||
165125e1 | 575 | struct request_queue * |
1946089a CL |
576 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
577 | { | |
165125e1 | 578 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
579 | |
580 | if (!q) | |
581 | return NULL; | |
582 | ||
1946089a | 583 | q->node = node_id; |
8669aafd | 584 | if (blk_init_free_list(q)) { |
8324aa91 | 585 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
586 | return NULL; |
587 | } | |
1da177e4 | 588 | |
152587de | 589 | /* |
590 | * if caller didn't supply a lock, they get per-queue locking with | |
591 | * our embedded lock | |
592 | */ | |
e7e72bf6 | 593 | if (!lock) |
152587de | 594 | lock = &q->__queue_lock; |
152587de | 595 | |
1da177e4 | 596 | q->request_fn = rfn; |
1da177e4 LT |
597 | q->prep_rq_fn = NULL; |
598 | q->unplug_fn = generic_unplug_device; | |
4ee5eaf4 KU |
599 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | |
600 | 1 << QUEUE_FLAG_STACKABLE); | |
1da177e4 LT |
601 | q->queue_lock = lock; |
602 | ||
0e435ac2 | 603 | blk_queue_segment_boundary(q, BLK_SEG_BOUNDARY_MASK); |
1da177e4 LT |
604 | |
605 | blk_queue_make_request(q, __make_request); | |
606 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
607 | ||
608 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
609 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
610 | ||
44ec9542 AS |
611 | q->sg_reserved_size = INT_MAX; |
612 | ||
abf54393 FT |
613 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
614 | ||
1da177e4 LT |
615 | /* |
616 | * all done | |
617 | */ | |
618 | if (!elevator_init(q, NULL)) { | |
619 | blk_queue_congestion_threshold(q); | |
620 | return q; | |
621 | } | |
622 | ||
8669aafd | 623 | blk_put_queue(q); |
1da177e4 LT |
624 | return NULL; |
625 | } | |
1946089a | 626 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 627 | |
165125e1 | 628 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 629 | { |
fde6ad22 | 630 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 631 | kobject_get(&q->kobj); |
1da177e4 LT |
632 | return 0; |
633 | } | |
634 | ||
635 | return 1; | |
636 | } | |
1da177e4 | 637 | |
165125e1 | 638 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 639 | { |
4aff5e23 | 640 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 641 | elv_put_request(q, rq); |
1da177e4 LT |
642 | mempool_free(rq, q->rq.rq_pool); |
643 | } | |
644 | ||
1ea25ecb | 645 | static struct request * |
165125e1 | 646 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
647 | { |
648 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
649 | ||
650 | if (!rq) | |
651 | return NULL; | |
652 | ||
2a4aa30c | 653 | blk_rq_init(q, rq); |
1afb20f3 | 654 | |
49171e5c | 655 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 656 | |
cb98fc8b | 657 | if (priv) { |
cb78b285 | 658 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
659 | mempool_free(rq, q->rq.rq_pool); |
660 | return NULL; | |
661 | } | |
4aff5e23 | 662 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 663 | } |
1da177e4 | 664 | |
cb98fc8b | 665 | return rq; |
1da177e4 LT |
666 | } |
667 | ||
668 | /* | |
669 | * ioc_batching returns true if the ioc is a valid batching request and | |
670 | * should be given priority access to a request. | |
671 | */ | |
165125e1 | 672 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
673 | { |
674 | if (!ioc) | |
675 | return 0; | |
676 | ||
677 | /* | |
678 | * Make sure the process is able to allocate at least 1 request | |
679 | * even if the batch times out, otherwise we could theoretically | |
680 | * lose wakeups. | |
681 | */ | |
682 | return ioc->nr_batch_requests == q->nr_batching || | |
683 | (ioc->nr_batch_requests > 0 | |
684 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
685 | } | |
686 | ||
687 | /* | |
688 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
689 | * will cause the process to be a "batcher" on all queues in the system. This | |
690 | * is the behaviour we want though - once it gets a wakeup it should be given | |
691 | * a nice run. | |
692 | */ | |
165125e1 | 693 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
694 | { |
695 | if (!ioc || ioc_batching(q, ioc)) | |
696 | return; | |
697 | ||
698 | ioc->nr_batch_requests = q->nr_batching; | |
699 | ioc->last_waited = jiffies; | |
700 | } | |
701 | ||
165125e1 | 702 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
703 | { |
704 | struct request_list *rl = &q->rq; | |
705 | ||
706 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 707 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
708 | |
709 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
710 | if (waitqueue_active(&rl->wait[rw])) |
711 | wake_up(&rl->wait[rw]); | |
712 | ||
713 | blk_clear_queue_full(q, rw); | |
714 | } | |
715 | } | |
716 | ||
717 | /* | |
718 | * A request has just been released. Account for it, update the full and | |
719 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
720 | */ | |
165125e1 | 721 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
722 | { |
723 | struct request_list *rl = &q->rq; | |
724 | ||
725 | rl->count[rw]--; | |
cb98fc8b TH |
726 | if (priv) |
727 | rl->elvpriv--; | |
1da177e4 LT |
728 | |
729 | __freed_request(q, rw); | |
730 | ||
731 | if (unlikely(rl->starved[rw ^ 1])) | |
732 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
733 | } |
734 | ||
735 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
736 | /* | |
d6344532 NP |
737 | * Get a free request, queue_lock must be held. |
738 | * Returns NULL on failure, with queue_lock held. | |
739 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 740 | */ |
165125e1 | 741 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 742 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
743 | { |
744 | struct request *rq = NULL; | |
745 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 746 | struct io_context *ioc = NULL; |
7749a8d4 | 747 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
748 | int may_queue, priv; |
749 | ||
7749a8d4 | 750 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
751 | if (may_queue == ELV_MQUEUE_NO) |
752 | goto rq_starved; | |
753 | ||
754 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
755 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 756 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
757 | /* |
758 | * The queue will fill after this allocation, so set | |
759 | * it as full, and mark this process as "batching". | |
760 | * This process will be allowed to complete a batch of | |
761 | * requests, others will be blocked. | |
762 | */ | |
763 | if (!blk_queue_full(q, rw)) { | |
764 | ioc_set_batching(q, ioc); | |
765 | blk_set_queue_full(q, rw); | |
766 | } else { | |
767 | if (may_queue != ELV_MQUEUE_MUST | |
768 | && !ioc_batching(q, ioc)) { | |
769 | /* | |
770 | * The queue is full and the allocating | |
771 | * process is not a "batcher", and not | |
772 | * exempted by the IO scheduler | |
773 | */ | |
774 | goto out; | |
775 | } | |
776 | } | |
1da177e4 | 777 | } |
79e2de4b | 778 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
779 | } |
780 | ||
082cf69e JA |
781 | /* |
782 | * Only allow batching queuers to allocate up to 50% over the defined | |
783 | * limit of requests, otherwise we could have thousands of requests | |
784 | * allocated with any setting of ->nr_requests | |
785 | */ | |
fd782a4a | 786 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 787 | goto out; |
fd782a4a | 788 | |
1da177e4 LT |
789 | rl->count[rw]++; |
790 | rl->starved[rw] = 0; | |
cb98fc8b | 791 | |
64521d1a | 792 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
793 | if (priv) |
794 | rl->elvpriv++; | |
795 | ||
1da177e4 LT |
796 | spin_unlock_irq(q->queue_lock); |
797 | ||
7749a8d4 | 798 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 799 | if (unlikely(!rq)) { |
1da177e4 LT |
800 | /* |
801 | * Allocation failed presumably due to memory. Undo anything | |
802 | * we might have messed up. | |
803 | * | |
804 | * Allocating task should really be put onto the front of the | |
805 | * wait queue, but this is pretty rare. | |
806 | */ | |
807 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 808 | freed_request(q, rw, priv); |
1da177e4 LT |
809 | |
810 | /* | |
811 | * in the very unlikely event that allocation failed and no | |
812 | * requests for this direction was pending, mark us starved | |
813 | * so that freeing of a request in the other direction will | |
814 | * notice us. another possible fix would be to split the | |
815 | * rq mempool into READ and WRITE | |
816 | */ | |
817 | rq_starved: | |
818 | if (unlikely(rl->count[rw] == 0)) | |
819 | rl->starved[rw] = 1; | |
820 | ||
1da177e4 LT |
821 | goto out; |
822 | } | |
823 | ||
88ee5ef1 JA |
824 | /* |
825 | * ioc may be NULL here, and ioc_batching will be false. That's | |
826 | * OK, if the queue is under the request limit then requests need | |
827 | * not count toward the nr_batch_requests limit. There will always | |
828 | * be some limit enforced by BLK_BATCH_TIME. | |
829 | */ | |
1da177e4 LT |
830 | if (ioc_batching(q, ioc)) |
831 | ioc->nr_batch_requests--; | |
6728cb0e | 832 | |
5f3ea37c | 833 | trace_block_getrq(q, bio, rw); |
1da177e4 | 834 | out: |
1da177e4 LT |
835 | return rq; |
836 | } | |
837 | ||
838 | /* | |
839 | * No available requests for this queue, unplug the device and wait for some | |
840 | * requests to become available. | |
d6344532 NP |
841 | * |
842 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 843 | */ |
165125e1 | 844 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 845 | struct bio *bio) |
1da177e4 | 846 | { |
7749a8d4 | 847 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
848 | struct request *rq; |
849 | ||
7749a8d4 | 850 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
851 | while (!rq) { |
852 | DEFINE_WAIT(wait); | |
05caf8db | 853 | struct io_context *ioc; |
1da177e4 LT |
854 | struct request_list *rl = &q->rq; |
855 | ||
856 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
857 | TASK_UNINTERRUPTIBLE); | |
858 | ||
5f3ea37c | 859 | trace_block_sleeprq(q, bio, rw); |
1da177e4 | 860 | |
05caf8db ZY |
861 | __generic_unplug_device(q); |
862 | spin_unlock_irq(q->queue_lock); | |
863 | io_schedule(); | |
1da177e4 | 864 | |
05caf8db ZY |
865 | /* |
866 | * After sleeping, we become a "batching" process and | |
867 | * will be able to allocate at least one request, and | |
868 | * up to a big batch of them for a small period time. | |
869 | * See ioc_batching, ioc_set_batching | |
870 | */ | |
871 | ioc = current_io_context(GFP_NOIO, q->node); | |
872 | ioc_set_batching(q, ioc); | |
d6344532 | 873 | |
05caf8db | 874 | spin_lock_irq(q->queue_lock); |
1da177e4 | 875 | finish_wait(&rl->wait[rw], &wait); |
05caf8db ZY |
876 | |
877 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
878 | }; | |
1da177e4 LT |
879 | |
880 | return rq; | |
881 | } | |
882 | ||
165125e1 | 883 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
884 | { |
885 | struct request *rq; | |
886 | ||
887 | BUG_ON(rw != READ && rw != WRITE); | |
888 | ||
d6344532 NP |
889 | spin_lock_irq(q->queue_lock); |
890 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 891 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 892 | } else { |
22e2c507 | 893 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
894 | if (!rq) |
895 | spin_unlock_irq(q->queue_lock); | |
896 | } | |
897 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
898 | |
899 | return rq; | |
900 | } | |
1da177e4 LT |
901 | EXPORT_SYMBOL(blk_get_request); |
902 | ||
dc72ef4a JA |
903 | /** |
904 | * blk_start_queueing - initiate dispatch of requests to device | |
905 | * @q: request queue to kick into gear | |
906 | * | |
907 | * This is basically a helper to remove the need to know whether a queue | |
908 | * is plugged or not if someone just wants to initiate dispatch of requests | |
80a4b58e JA |
909 | * for this queue. Should be used to start queueing on a device outside |
910 | * of ->request_fn() context. Also see @blk_run_queue. | |
dc72ef4a JA |
911 | * |
912 | * The queue lock must be held with interrupts disabled. | |
913 | */ | |
165125e1 | 914 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a | 915 | { |
336c3d8c EO |
916 | if (!blk_queue_plugged(q)) { |
917 | if (unlikely(blk_queue_stopped(q))) | |
918 | return; | |
dc72ef4a | 919 | q->request_fn(q); |
336c3d8c | 920 | } else |
dc72ef4a JA |
921 | __generic_unplug_device(q); |
922 | } | |
923 | EXPORT_SYMBOL(blk_start_queueing); | |
924 | ||
1da177e4 LT |
925 | /** |
926 | * blk_requeue_request - put a request back on queue | |
927 | * @q: request queue where request should be inserted | |
928 | * @rq: request to be inserted | |
929 | * | |
930 | * Description: | |
931 | * Drivers often keep queueing requests until the hardware cannot accept | |
932 | * more, when that condition happens we need to put the request back | |
933 | * on the queue. Must be called with queue lock held. | |
934 | */ | |
165125e1 | 935 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 936 | { |
242f9dcb JA |
937 | blk_delete_timer(rq); |
938 | blk_clear_rq_complete(rq); | |
5f3ea37c | 939 | trace_block_rq_requeue(q, rq); |
2056a782 | 940 | |
1da177e4 LT |
941 | if (blk_rq_tagged(rq)) |
942 | blk_queue_end_tag(q, rq); | |
943 | ||
944 | elv_requeue_request(q, rq); | |
945 | } | |
1da177e4 LT |
946 | EXPORT_SYMBOL(blk_requeue_request); |
947 | ||
948 | /** | |
710027a4 | 949 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
950 | * @q: request queue where request should be inserted |
951 | * @rq: request to be inserted | |
952 | * @at_head: insert request at head or tail of queue | |
953 | * @data: private data | |
1da177e4 LT |
954 | * |
955 | * Description: | |
956 | * Many block devices need to execute commands asynchronously, so they don't | |
957 | * block the whole kernel from preemption during request execution. This is | |
958 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
959 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
960 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
961 | * |
962 | * We have the option of inserting the head or the tail of the queue. | |
963 | * Typically we use the tail for new ioctls and so forth. We use the head | |
964 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
965 | * host that is unable to accept a particular command. | |
966 | */ | |
165125e1 | 967 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 968 | int at_head, void *data) |
1da177e4 | 969 | { |
867d1191 | 970 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
971 | unsigned long flags; |
972 | ||
973 | /* | |
974 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
975 | * must not attempt merges on this) and that it acts as a soft | |
976 | * barrier | |
977 | */ | |
4aff5e23 JA |
978 | rq->cmd_type = REQ_TYPE_SPECIAL; |
979 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
980 | |
981 | rq->special = data; | |
982 | ||
983 | spin_lock_irqsave(q->queue_lock, flags); | |
984 | ||
985 | /* | |
986 | * If command is tagged, release the tag | |
987 | */ | |
867d1191 TH |
988 | if (blk_rq_tagged(rq)) |
989 | blk_queue_end_tag(q, rq); | |
1da177e4 | 990 | |
b238b3d4 | 991 | drive_stat_acct(rq, 1); |
867d1191 | 992 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 993 | blk_start_queueing(q); |
1da177e4 LT |
994 | spin_unlock_irqrestore(q->queue_lock, flags); |
995 | } | |
1da177e4 LT |
996 | EXPORT_SYMBOL(blk_insert_request); |
997 | ||
1da177e4 LT |
998 | /* |
999 | * add-request adds a request to the linked list. | |
1000 | * queue lock is held and interrupts disabled, as we muck with the | |
1001 | * request queue list. | |
1002 | */ | |
6728cb0e | 1003 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 1004 | { |
b238b3d4 | 1005 | drive_stat_acct(req, 1); |
1da177e4 | 1006 | |
1da177e4 LT |
1007 | /* |
1008 | * elevator indicated where it wants this request to be | |
1009 | * inserted at elevator_merge time | |
1010 | */ | |
1011 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
1012 | } | |
6728cb0e | 1013 | |
074a7aca TH |
1014 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1015 | unsigned long now) | |
1016 | { | |
1017 | if (now == part->stamp) | |
1018 | return; | |
1019 | ||
1020 | if (part->in_flight) { | |
1021 | __part_stat_add(cpu, part, time_in_queue, | |
1022 | part->in_flight * (now - part->stamp)); | |
1023 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1024 | } | |
1025 | part->stamp = now; | |
1026 | } | |
1027 | ||
1028 | /** | |
496aa8a9 RD |
1029 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1030 | * @cpu: cpu number for stats access | |
1031 | * @part: target partition | |
1da177e4 LT |
1032 | * |
1033 | * The average IO queue length and utilisation statistics are maintained | |
1034 | * by observing the current state of the queue length and the amount of | |
1035 | * time it has been in this state for. | |
1036 | * | |
1037 | * Normally, that accounting is done on IO completion, but that can result | |
1038 | * in more than a second's worth of IO being accounted for within any one | |
1039 | * second, leading to >100% utilisation. To deal with that, we call this | |
1040 | * function to do a round-off before returning the results when reading | |
1041 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1042 | * the current jiffies and restarts the counters again. | |
1043 | */ | |
c9959059 | 1044 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1045 | { |
1046 | unsigned long now = jiffies; | |
1047 | ||
074a7aca TH |
1048 | if (part->partno) |
1049 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1050 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1051 | } |
074a7aca | 1052 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1053 | |
1da177e4 LT |
1054 | /* |
1055 | * queue lock must be held | |
1056 | */ | |
165125e1 | 1057 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1058 | { |
1da177e4 LT |
1059 | if (unlikely(!q)) |
1060 | return; | |
1061 | if (unlikely(--req->ref_count)) | |
1062 | return; | |
1063 | ||
8922e16c TH |
1064 | elv_completed_request(q, req); |
1065 | ||
1da177e4 LT |
1066 | /* |
1067 | * Request may not have originated from ll_rw_blk. if not, | |
1068 | * it didn't come out of our reserved rq pools | |
1069 | */ | |
49171e5c | 1070 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1071 | int rw = rq_data_dir(req); |
4aff5e23 | 1072 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1073 | |
1da177e4 | 1074 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1075 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1076 | |
1077 | blk_free_request(q, req); | |
cb98fc8b | 1078 | freed_request(q, rw, priv); |
1da177e4 LT |
1079 | } |
1080 | } | |
6e39b69e MC |
1081 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1082 | ||
1da177e4 LT |
1083 | void blk_put_request(struct request *req) |
1084 | { | |
8922e16c | 1085 | unsigned long flags; |
165125e1 | 1086 | struct request_queue *q = req->q; |
8922e16c | 1087 | |
52a93ba8 FT |
1088 | spin_lock_irqsave(q->queue_lock, flags); |
1089 | __blk_put_request(q, req); | |
1090 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1091 | } |
1da177e4 LT |
1092 | EXPORT_SYMBOL(blk_put_request); |
1093 | ||
86db1e29 | 1094 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1095 | { |
c7c22e4d | 1096 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1097 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1098 | |
1099 | /* | |
1100 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1101 | */ | |
6000a368 MC |
1102 | if (bio_rw_ahead(bio)) |
1103 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1104 | REQ_FAILFAST_DRIVER); | |
1105 | if (bio_failfast_dev(bio)) | |
1106 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1107 | if (bio_failfast_transport(bio)) | |
1108 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1109 | if (bio_failfast_driver(bio)) | |
1110 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
52d9e675 TH |
1111 | |
1112 | /* | |
1113 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1114 | */ | |
fb2dce86 | 1115 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1116 | req->cmd_flags |= REQ_DISCARD; |
1117 | if (bio_barrier(bio)) | |
1118 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1119 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 DW |
1120 | } else if (unlikely(bio_barrier(bio))) |
1121 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
52d9e675 | 1122 | |
b31dc66a | 1123 | if (bio_sync(bio)) |
4aff5e23 | 1124 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1125 | if (bio_rw_meta(bio)) |
1126 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1127 | |
52d9e675 TH |
1128 | req->errors = 0; |
1129 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1130 | req->ioprio = bio_prio(bio); |
52d9e675 | 1131 | req->start_time = jiffies; |
bc1c56fd | 1132 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1133 | } |
1134 | ||
165125e1 | 1135 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1136 | { |
450991bc | 1137 | struct request *req; |
fb2dce86 | 1138 | int el_ret, nr_sectors, barrier, discard, err; |
51da90fc JA |
1139 | const unsigned short prio = bio_prio(bio); |
1140 | const int sync = bio_sync(bio); | |
7749a8d4 | 1141 | int rw_flags; |
1da177e4 | 1142 | |
1da177e4 | 1143 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1144 | |
1145 | /* | |
1146 | * low level driver can indicate that it wants pages above a | |
1147 | * certain limit bounced to low memory (ie for highmem, or even | |
1148 | * ISA dma in theory) | |
1149 | */ | |
1150 | blk_queue_bounce(q, &bio); | |
1151 | ||
1da177e4 | 1152 | barrier = bio_barrier(bio); |
e17fc0a1 DW |
1153 | if (unlikely(barrier) && bio_has_data(bio) && |
1154 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1da177e4 LT |
1155 | err = -EOPNOTSUPP; |
1156 | goto end_io; | |
1157 | } | |
1158 | ||
fb2dce86 DW |
1159 | discard = bio_discard(bio); |
1160 | if (unlikely(discard) && !q->prepare_discard_fn) { | |
1161 | err = -EOPNOTSUPP; | |
1162 | goto end_io; | |
1163 | } | |
1164 | ||
1da177e4 LT |
1165 | spin_lock_irq(q->queue_lock); |
1166 | ||
450991bc | 1167 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1168 | goto get_rq; |
1169 | ||
1170 | el_ret = elv_merge(q, &req, bio); | |
1171 | switch (el_ret) { | |
6728cb0e JA |
1172 | case ELEVATOR_BACK_MERGE: |
1173 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1174 | |
6728cb0e JA |
1175 | if (!ll_back_merge_fn(q, req, bio)) |
1176 | break; | |
1da177e4 | 1177 | |
5f3ea37c | 1178 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1179 | |
6728cb0e JA |
1180 | req->biotail->bi_next = bio; |
1181 | req->biotail = bio; | |
1182 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1183 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1184 | if (!blk_rq_cpu_valid(req)) |
1185 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1186 | drive_stat_acct(req, 0); |
1187 | if (!attempt_back_merge(q, req)) | |
1188 | elv_merged_request(q, req, el_ret); | |
1189 | goto out; | |
1da177e4 | 1190 | |
6728cb0e JA |
1191 | case ELEVATOR_FRONT_MERGE: |
1192 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1193 | |
6728cb0e JA |
1194 | if (!ll_front_merge_fn(q, req, bio)) |
1195 | break; | |
1da177e4 | 1196 | |
5f3ea37c | 1197 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1198 | |
6728cb0e JA |
1199 | bio->bi_next = req->bio; |
1200 | req->bio = bio; | |
1da177e4 | 1201 | |
6728cb0e JA |
1202 | /* |
1203 | * may not be valid. if the low level driver said | |
1204 | * it didn't need a bounce buffer then it better | |
1205 | * not touch req->buffer either... | |
1206 | */ | |
1207 | req->buffer = bio_data(bio); | |
1208 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1209 | req->hard_cur_sectors = req->current_nr_sectors; | |
1210 | req->sector = req->hard_sector = bio->bi_sector; | |
1211 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1212 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1213 | if (!blk_rq_cpu_valid(req)) |
1214 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1215 | drive_stat_acct(req, 0); |
1216 | if (!attempt_front_merge(q, req)) | |
1217 | elv_merged_request(q, req, el_ret); | |
1218 | goto out; | |
1219 | ||
1220 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1221 | default: | |
1222 | ; | |
1da177e4 LT |
1223 | } |
1224 | ||
450991bc | 1225 | get_rq: |
7749a8d4 JA |
1226 | /* |
1227 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1228 | * but we need to set it earlier to expose the sync flag to the | |
1229 | * rq allocator and io schedulers. | |
1230 | */ | |
1231 | rw_flags = bio_data_dir(bio); | |
1232 | if (sync) | |
1233 | rw_flags |= REQ_RW_SYNC; | |
1234 | ||
1da177e4 | 1235 | /* |
450991bc | 1236 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1237 | * Returns with the queue unlocked. |
450991bc | 1238 | */ |
7749a8d4 | 1239 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1240 | |
450991bc NP |
1241 | /* |
1242 | * After dropping the lock and possibly sleeping here, our request | |
1243 | * may now be mergeable after it had proven unmergeable (above). | |
1244 | * We don't worry about that case for efficiency. It won't happen | |
1245 | * often, and the elevators are able to handle it. | |
1da177e4 | 1246 | */ |
52d9e675 | 1247 | init_request_from_bio(req, bio); |
1da177e4 | 1248 | |
450991bc | 1249 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1250 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1251 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1252 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
450991bc NP |
1253 | if (elv_queue_empty(q)) |
1254 | blk_plug_device(q); | |
1da177e4 LT |
1255 | add_request(q, req); |
1256 | out: | |
4a534f93 | 1257 | if (sync) |
1da177e4 | 1258 | __generic_unplug_device(q); |
1da177e4 LT |
1259 | spin_unlock_irq(q->queue_lock); |
1260 | return 0; | |
1261 | ||
1262 | end_io: | |
6712ecf8 | 1263 | bio_endio(bio, err); |
1da177e4 LT |
1264 | return 0; |
1265 | } | |
1266 | ||
1267 | /* | |
1268 | * If bio->bi_dev is a partition, remap the location | |
1269 | */ | |
1270 | static inline void blk_partition_remap(struct bio *bio) | |
1271 | { | |
1272 | struct block_device *bdev = bio->bi_bdev; | |
1273 | ||
bf2de6f5 | 1274 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1275 | struct hd_struct *p = bdev->bd_part; |
1276 | ||
1da177e4 LT |
1277 | bio->bi_sector += p->start_sect; |
1278 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1279 | |
5f3ea37c | 1280 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, |
c7149d6b AB |
1281 | bdev->bd_dev, bio->bi_sector, |
1282 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1283 | } |
1284 | } | |
1285 | ||
1da177e4 LT |
1286 | static void handle_bad_sector(struct bio *bio) |
1287 | { | |
1288 | char b[BDEVNAME_SIZE]; | |
1289 | ||
1290 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1291 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1292 | bdevname(bio->bi_bdev, b), | |
1293 | bio->bi_rw, | |
1294 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1295 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1296 | ||
1297 | set_bit(BIO_EOF, &bio->bi_flags); | |
1298 | } | |
1299 | ||
c17bb495 AM |
1300 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1301 | ||
1302 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1303 | ||
1304 | static int __init setup_fail_make_request(char *str) | |
1305 | { | |
1306 | return setup_fault_attr(&fail_make_request, str); | |
1307 | } | |
1308 | __setup("fail_make_request=", setup_fail_make_request); | |
1309 | ||
1310 | static int should_fail_request(struct bio *bio) | |
1311 | { | |
eddb2e26 TH |
1312 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1313 | ||
1314 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1315 | return should_fail(&fail_make_request, bio->bi_size); |
1316 | ||
1317 | return 0; | |
1318 | } | |
1319 | ||
1320 | static int __init fail_make_request_debugfs(void) | |
1321 | { | |
1322 | return init_fault_attr_dentries(&fail_make_request, | |
1323 | "fail_make_request"); | |
1324 | } | |
1325 | ||
1326 | late_initcall(fail_make_request_debugfs); | |
1327 | ||
1328 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1329 | ||
1330 | static inline int should_fail_request(struct bio *bio) | |
1331 | { | |
1332 | return 0; | |
1333 | } | |
1334 | ||
1335 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1336 | ||
c07e2b41 JA |
1337 | /* |
1338 | * Check whether this bio extends beyond the end of the device. | |
1339 | */ | |
1340 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1341 | { | |
1342 | sector_t maxsector; | |
1343 | ||
1344 | if (!nr_sectors) | |
1345 | return 0; | |
1346 | ||
1347 | /* Test device or partition size, when known. */ | |
1348 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1349 | if (maxsector) { | |
1350 | sector_t sector = bio->bi_sector; | |
1351 | ||
1352 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1353 | /* | |
1354 | * This may well happen - the kernel calls bread() | |
1355 | * without checking the size of the device, e.g., when | |
1356 | * mounting a device. | |
1357 | */ | |
1358 | handle_bad_sector(bio); | |
1359 | return 1; | |
1360 | } | |
1361 | } | |
1362 | ||
1363 | return 0; | |
1364 | } | |
1365 | ||
1da177e4 | 1366 | /** |
710027a4 | 1367 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1368 | * @bio: The bio describing the location in memory and on the device. |
1369 | * | |
1370 | * generic_make_request() is used to make I/O requests of block | |
1371 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1372 | * to be done. | |
1373 | * | |
1374 | * generic_make_request() does not return any status. The | |
1375 | * success/failure status of the request, along with notification of | |
1376 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1377 | * function described (one day) else where. | |
1378 | * | |
1379 | * The caller of generic_make_request must make sure that bi_io_vec | |
1380 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1381 | * set to describe the device address, and the | |
1382 | * bi_end_io and optionally bi_private are set to describe how | |
1383 | * completion notification should be signaled. | |
1384 | * | |
1385 | * generic_make_request and the drivers it calls may use bi_next if this | |
1386 | * bio happens to be merged with someone else, and may change bi_dev and | |
1387 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1388 | * should NOT be depended on after the call to generic_make_request. | |
1389 | */ | |
d89d8796 | 1390 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1391 | { |
165125e1 | 1392 | struct request_queue *q; |
5ddfe969 | 1393 | sector_t old_sector; |
1da177e4 | 1394 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1395 | dev_t old_dev; |
51fd77bd | 1396 | int err = -EIO; |
1da177e4 LT |
1397 | |
1398 | might_sleep(); | |
1da177e4 | 1399 | |
c07e2b41 JA |
1400 | if (bio_check_eod(bio, nr_sectors)) |
1401 | goto end_io; | |
1da177e4 LT |
1402 | |
1403 | /* | |
1404 | * Resolve the mapping until finished. (drivers are | |
1405 | * still free to implement/resolve their own stacking | |
1406 | * by explicitly returning 0) | |
1407 | * | |
1408 | * NOTE: we don't repeat the blk_size check for each new device. | |
1409 | * Stacking drivers are expected to know what they are doing. | |
1410 | */ | |
5ddfe969 | 1411 | old_sector = -1; |
2056a782 | 1412 | old_dev = 0; |
1da177e4 LT |
1413 | do { |
1414 | char b[BDEVNAME_SIZE]; | |
1415 | ||
1416 | q = bdev_get_queue(bio->bi_bdev); | |
1417 | if (!q) { | |
1418 | printk(KERN_ERR | |
1419 | "generic_make_request: Trying to access " | |
1420 | "nonexistent block-device %s (%Lu)\n", | |
1421 | bdevname(bio->bi_bdev, b), | |
1422 | (long long) bio->bi_sector); | |
1423 | end_io: | |
51fd77bd | 1424 | bio_endio(bio, err); |
1da177e4 LT |
1425 | break; |
1426 | } | |
1427 | ||
4fa253f3 | 1428 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1429 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1430 | bdevname(bio->bi_bdev, b), |
1431 | bio_sectors(bio), | |
1432 | q->max_hw_sectors); | |
1433 | goto end_io; | |
1434 | } | |
1435 | ||
fde6ad22 | 1436 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1437 | goto end_io; |
1438 | ||
c17bb495 AM |
1439 | if (should_fail_request(bio)) |
1440 | goto end_io; | |
1441 | ||
1da177e4 LT |
1442 | /* |
1443 | * If this device has partitions, remap block n | |
1444 | * of partition p to block n+start(p) of the disk. | |
1445 | */ | |
1446 | blk_partition_remap(bio); | |
1447 | ||
7ba1ba12 MP |
1448 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1449 | goto end_io; | |
1450 | ||
5ddfe969 | 1451 | if (old_sector != -1) |
5f3ea37c | 1452 | trace_block_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1453 | old_sector); |
2056a782 | 1454 | |
5f3ea37c | 1455 | trace_block_bio_queue(q, bio); |
2056a782 | 1456 | |
5ddfe969 | 1457 | old_sector = bio->bi_sector; |
2056a782 JA |
1458 | old_dev = bio->bi_bdev->bd_dev; |
1459 | ||
c07e2b41 JA |
1460 | if (bio_check_eod(bio, nr_sectors)) |
1461 | goto end_io; | |
fb2dce86 DW |
1462 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || |
1463 | (bio_discard(bio) && !q->prepare_discard_fn)) { | |
51fd77bd JA |
1464 | err = -EOPNOTSUPP; |
1465 | goto end_io; | |
1466 | } | |
5ddfe969 | 1467 | |
1da177e4 LT |
1468 | ret = q->make_request_fn(q, bio); |
1469 | } while (ret); | |
1470 | } | |
1471 | ||
d89d8796 NB |
1472 | /* |
1473 | * We only want one ->make_request_fn to be active at a time, | |
1474 | * else stack usage with stacked devices could be a problem. | |
1475 | * So use current->bio_{list,tail} to keep a list of requests | |
1476 | * submited by a make_request_fn function. | |
1477 | * current->bio_tail is also used as a flag to say if | |
1478 | * generic_make_request is currently active in this task or not. | |
1479 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1480 | * then a make_request is active, and new requests should be added | |
1481 | * at the tail | |
1482 | */ | |
1483 | void generic_make_request(struct bio *bio) | |
1484 | { | |
1485 | if (current->bio_tail) { | |
1486 | /* make_request is active */ | |
1487 | *(current->bio_tail) = bio; | |
1488 | bio->bi_next = NULL; | |
1489 | current->bio_tail = &bio->bi_next; | |
1490 | return; | |
1491 | } | |
1492 | /* following loop may be a bit non-obvious, and so deserves some | |
1493 | * explanation. | |
1494 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1495 | * ensure that) so we have a list with a single bio. | |
1496 | * We pretend that we have just taken it off a longer list, so | |
1497 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1498 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1499 | * added. __generic_make_request may indeed add some more bios | |
1500 | * through a recursive call to generic_make_request. If it | |
1501 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1502 | * from the top. In this case we really did just take the bio | |
1503 | * of the top of the list (no pretending) and so fixup bio_list and | |
1504 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1505 | * | |
1506 | * The loop was structured like this to make only one call to | |
1507 | * __generic_make_request (which is important as it is large and | |
1508 | * inlined) and to keep the structure simple. | |
1509 | */ | |
1510 | BUG_ON(bio->bi_next); | |
1511 | do { | |
1512 | current->bio_list = bio->bi_next; | |
1513 | if (bio->bi_next == NULL) | |
1514 | current->bio_tail = ¤t->bio_list; | |
1515 | else | |
1516 | bio->bi_next = NULL; | |
1517 | __generic_make_request(bio); | |
1518 | bio = current->bio_list; | |
1519 | } while (bio); | |
1520 | current->bio_tail = NULL; /* deactivate */ | |
1521 | } | |
1da177e4 LT |
1522 | EXPORT_SYMBOL(generic_make_request); |
1523 | ||
1524 | /** | |
710027a4 | 1525 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1526 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1527 | * @bio: The &struct bio which describes the I/O | |
1528 | * | |
1529 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1530 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1531 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1532 | * |
1533 | */ | |
1534 | void submit_bio(int rw, struct bio *bio) | |
1535 | { | |
1536 | int count = bio_sectors(bio); | |
1537 | ||
22e2c507 | 1538 | bio->bi_rw |= rw; |
1da177e4 | 1539 | |
bf2de6f5 JA |
1540 | /* |
1541 | * If it's a regular read/write or a barrier with data attached, | |
1542 | * go through the normal accounting stuff before submission. | |
1543 | */ | |
a9c701e5 | 1544 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1545 | if (rw & WRITE) { |
1546 | count_vm_events(PGPGOUT, count); | |
1547 | } else { | |
1548 | task_io_account_read(bio->bi_size); | |
1549 | count_vm_events(PGPGIN, count); | |
1550 | } | |
1551 | ||
1552 | if (unlikely(block_dump)) { | |
1553 | char b[BDEVNAME_SIZE]; | |
1554 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1555 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1556 | (rw & WRITE) ? "WRITE" : "READ", |
1557 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1558 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1559 | } |
1da177e4 LT |
1560 | } |
1561 | ||
1562 | generic_make_request(bio); | |
1563 | } | |
1da177e4 LT |
1564 | EXPORT_SYMBOL(submit_bio); |
1565 | ||
82124d60 KU |
1566 | /** |
1567 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1568 | * @q: the queue | |
1569 | * @rq: the request being checked | |
1570 | * | |
1571 | * Description: | |
1572 | * @rq may have been made based on weaker limitations of upper-level queues | |
1573 | * in request stacking drivers, and it may violate the limitation of @q. | |
1574 | * Since the block layer and the underlying device driver trust @rq | |
1575 | * after it is inserted to @q, it should be checked against @q before | |
1576 | * the insertion using this generic function. | |
1577 | * | |
1578 | * This function should also be useful for request stacking drivers | |
1579 | * in some cases below, so export this fuction. | |
1580 | * Request stacking drivers like request-based dm may change the queue | |
1581 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1582 | * Such request stacking drivers should check those requests agaist | |
1583 | * the new queue limits again when they dispatch those requests, | |
1584 | * although such checkings are also done against the old queue limits | |
1585 | * when submitting requests. | |
1586 | */ | |
1587 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1588 | { | |
1589 | if (rq->nr_sectors > q->max_sectors || | |
1590 | rq->data_len > q->max_hw_sectors << 9) { | |
1591 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1592 | return -EIO; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * queue's settings related to segment counting like q->bounce_pfn | |
1597 | * may differ from that of other stacking queues. | |
1598 | * Recalculate it to check the request correctly on this queue's | |
1599 | * limitation. | |
1600 | */ | |
1601 | blk_recalc_rq_segments(rq); | |
1602 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1603 | rq->nr_phys_segments > q->max_hw_segments) { | |
1604 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1605 | return -EIO; | |
1606 | } | |
1607 | ||
1608 | return 0; | |
1609 | } | |
1610 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1611 | ||
1612 | /** | |
1613 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1614 | * @q: the queue to submit the request | |
1615 | * @rq: the request being queued | |
1616 | */ | |
1617 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1618 | { | |
1619 | unsigned long flags; | |
1620 | ||
1621 | if (blk_rq_check_limits(q, rq)) | |
1622 | return -EIO; | |
1623 | ||
1624 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1625 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1626 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1627 | return -EIO; | |
1628 | #endif | |
1629 | ||
1630 | spin_lock_irqsave(q->queue_lock, flags); | |
1631 | ||
1632 | /* | |
1633 | * Submitting request must be dequeued before calling this function | |
1634 | * because it will be linked to another request_queue | |
1635 | */ | |
1636 | BUG_ON(blk_queued_rq(rq)); | |
1637 | ||
1638 | drive_stat_acct(rq, 1); | |
1639 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1640 | ||
1641 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1642 | ||
1643 | return 0; | |
1644 | } | |
1645 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1646 | ||
53a08807 TH |
1647 | /** |
1648 | * blkdev_dequeue_request - dequeue request and start timeout timer | |
1649 | * @req: request to dequeue | |
1650 | * | |
1651 | * Dequeue @req and start timeout timer on it. This hands off the | |
1652 | * request to the driver. | |
1653 | * | |
1654 | * Block internal functions which don't want to start timer should | |
1655 | * call elv_dequeue_request(). | |
1656 | */ | |
1657 | void blkdev_dequeue_request(struct request *req) | |
1658 | { | |
1659 | elv_dequeue_request(req->q, req); | |
1660 | ||
1661 | /* | |
1662 | * We are now handing the request to the hardware, add the | |
1663 | * timeout handler. | |
1664 | */ | |
1665 | blk_add_timer(req); | |
1666 | } | |
1667 | EXPORT_SYMBOL(blkdev_dequeue_request); | |
1668 | ||
3bcddeac KU |
1669 | /** |
1670 | * __end_that_request_first - end I/O on a request | |
1671 | * @req: the request being processed | |
710027a4 | 1672 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
1673 | * @nr_bytes: number of bytes to complete |
1674 | * | |
1675 | * Description: | |
1676 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1677 | * for the next range of segments (if any) in the cluster. | |
1678 | * | |
1679 | * Return: | |
710027a4 RD |
1680 | * %0 - we are done with this request, call end_that_request_last() |
1681 | * %1 - still buffers pending for this request | |
3bcddeac | 1682 | **/ |
5450d3e1 | 1683 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1684 | int nr_bytes) |
1685 | { | |
5450d3e1 | 1686 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1687 | struct bio *bio; |
1688 | ||
5f3ea37c | 1689 | trace_block_rq_complete(req->q, req); |
2056a782 | 1690 | |
1da177e4 | 1691 | /* |
710027a4 | 1692 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual |
1da177e4 LT |
1693 | * sense key with us all the way through |
1694 | */ | |
1695 | if (!blk_pc_request(req)) | |
1696 | req->errors = 0; | |
1697 | ||
6728cb0e JA |
1698 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1699 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1700 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1701 | (unsigned long long)req->sector); | |
1702 | } | |
1703 | ||
d72d904a | 1704 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b | 1705 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1706 | struct hd_struct *part; |
c9959059 | 1707 | int cpu; |
a362357b | 1708 | |
074a7aca | 1709 | cpu = part_stat_lock(); |
e71bf0d0 | 1710 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
074a7aca TH |
1711 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); |
1712 | part_stat_unlock(); | |
d72d904a JA |
1713 | } |
1714 | ||
1da177e4 LT |
1715 | total_bytes = bio_nbytes = 0; |
1716 | while ((bio = req->bio) != NULL) { | |
1717 | int nbytes; | |
1718 | ||
bf2de6f5 JA |
1719 | /* |
1720 | * For an empty barrier request, the low level driver must | |
1721 | * store a potential error location in ->sector. We pass | |
1722 | * that back up in ->bi_sector. | |
1723 | */ | |
1724 | if (blk_empty_barrier(req)) | |
1725 | bio->bi_sector = req->sector; | |
1726 | ||
1da177e4 LT |
1727 | if (nr_bytes >= bio->bi_size) { |
1728 | req->bio = bio->bi_next; | |
1729 | nbytes = bio->bi_size; | |
5bb23a68 | 1730 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1731 | next_idx = 0; |
1732 | bio_nbytes = 0; | |
1733 | } else { | |
1734 | int idx = bio->bi_idx + next_idx; | |
1735 | ||
1736 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1737 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e | 1738 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1739 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
1740 | break; |
1741 | } | |
1742 | ||
1743 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1744 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1745 | ||
1746 | /* | |
1747 | * not a complete bvec done | |
1748 | */ | |
1749 | if (unlikely(nbytes > nr_bytes)) { | |
1750 | bio_nbytes += nr_bytes; | |
1751 | total_bytes += nr_bytes; | |
1752 | break; | |
1753 | } | |
1754 | ||
1755 | /* | |
1756 | * advance to the next vector | |
1757 | */ | |
1758 | next_idx++; | |
1759 | bio_nbytes += nbytes; | |
1760 | } | |
1761 | ||
1762 | total_bytes += nbytes; | |
1763 | nr_bytes -= nbytes; | |
1764 | ||
6728cb0e JA |
1765 | bio = req->bio; |
1766 | if (bio) { | |
1da177e4 LT |
1767 | /* |
1768 | * end more in this run, or just return 'not-done' | |
1769 | */ | |
1770 | if (unlikely(nr_bytes <= 0)) | |
1771 | break; | |
1772 | } | |
1773 | } | |
1774 | ||
1775 | /* | |
1776 | * completely done | |
1777 | */ | |
1778 | if (!req->bio) | |
1779 | return 0; | |
1780 | ||
1781 | /* | |
1782 | * if the request wasn't completed, update state | |
1783 | */ | |
1784 | if (bio_nbytes) { | |
5bb23a68 | 1785 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1786 | bio->bi_idx += next_idx; |
1787 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1788 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1789 | } | |
1790 | ||
1791 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1792 | blk_recalc_rq_segments(req); | |
1793 | return 1; | |
1794 | } | |
1795 | ||
1da177e4 LT |
1796 | /* |
1797 | * queue lock must be held | |
1798 | */ | |
5450d3e1 | 1799 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1800 | { |
1801 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1802 | |
b8286239 KU |
1803 | if (blk_rq_tagged(req)) |
1804 | blk_queue_end_tag(req->q, req); | |
1805 | ||
1806 | if (blk_queued_rq(req)) | |
53a08807 | 1807 | elv_dequeue_request(req->q, req); |
1da177e4 LT |
1808 | |
1809 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1810 | laptop_io_completion(); | |
1811 | ||
e78042e5 MA |
1812 | blk_delete_timer(req); |
1813 | ||
fd0ff8aa JA |
1814 | /* |
1815 | * Account IO completion. bar_rq isn't accounted as a normal | |
1816 | * IO on queueing nor completion. Accounting the containing | |
1817 | * request is enough. | |
1818 | */ | |
1819 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 1820 | unsigned long duration = jiffies - req->start_time; |
a362357b | 1821 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1822 | struct hd_struct *part; |
c9959059 | 1823 | int cpu; |
e71bf0d0 | 1824 | |
074a7aca | 1825 | cpu = part_stat_lock(); |
e71bf0d0 | 1826 | part = disk_map_sector_rcu(disk, req->sector); |
a362357b | 1827 | |
074a7aca TH |
1828 | part_stat_inc(cpu, part, ios[rw]); |
1829 | part_stat_add(cpu, part, ticks[rw], duration); | |
1830 | part_round_stats(cpu, part); | |
1831 | part_dec_in_flight(part); | |
e71bf0d0 | 1832 | |
074a7aca | 1833 | part_stat_unlock(); |
1da177e4 | 1834 | } |
b8286239 | 1835 | |
1da177e4 | 1836 | if (req->end_io) |
8ffdc655 | 1837 | req->end_io(req, error); |
b8286239 KU |
1838 | else { |
1839 | if (blk_bidi_rq(req)) | |
1840 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1841 | ||
1da177e4 | 1842 | __blk_put_request(req->q, req); |
b8286239 | 1843 | } |
1da177e4 LT |
1844 | } |
1845 | ||
3b11313a KU |
1846 | /** |
1847 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1848 | * @rq: the request being processed |
3b11313a KU |
1849 | **/ |
1850 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1851 | { |
1852 | if (blk_fs_request(rq)) | |
1853 | return rq->hard_nr_sectors << 9; | |
1854 | ||
1855 | return rq->data_len; | |
1856 | } | |
3b11313a KU |
1857 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1858 | ||
1859 | /** | |
1860 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1861 | * @rq: the request being processed |
3b11313a KU |
1862 | **/ |
1863 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1864 | { | |
1865 | if (blk_fs_request(rq)) | |
1866 | return rq->current_nr_sectors << 9; | |
1867 | ||
1868 | if (rq->bio) | |
1869 | return rq->bio->bi_size; | |
1870 | ||
1871 | return rq->data_len; | |
1872 | } | |
1873 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 | 1874 | |
a0cd1285 JA |
1875 | /** |
1876 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1877 | * @req: the request being processed |
710027a4 | 1878 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1879 | * |
1880 | * Description: | |
1881 | * Ends I/O on the current segment of a request. If that is the only | |
1882 | * remaining segment, the request is also completed and freed. | |
1883 | * | |
710027a4 RD |
1884 | * This is a remnant of how older block drivers handled I/O completions. |
1885 | * Modern drivers typically end I/O on the full request in one go, unless | |
a0cd1285 JA |
1886 | * they have a residual value to account for. For that case this function |
1887 | * isn't really useful, unless the residual just happens to be the | |
1888 | * full current segment. In other words, don't use this function in new | |
d00e29fd | 1889 | * code. Use blk_end_request() or __blk_end_request() to end a request. |
a0cd1285 JA |
1890 | **/ |
1891 | void end_request(struct request *req, int uptodate) | |
1892 | { | |
d00e29fd KU |
1893 | int error = 0; |
1894 | ||
1895 | if (uptodate <= 0) | |
1896 | error = uptodate ? uptodate : -EIO; | |
1897 | ||
1898 | __blk_end_request(req, error, req->hard_cur_sectors << 9); | |
a0cd1285 | 1899 | } |
1da177e4 LT |
1900 | EXPORT_SYMBOL(end_request); |
1901 | ||
32fab448 KU |
1902 | static int end_that_request_data(struct request *rq, int error, |
1903 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1904 | { | |
1905 | if (rq->bio) { | |
1906 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1907 | return 1; | |
1908 | ||
1909 | /* Bidi request must be completed as a whole */ | |
1910 | if (blk_bidi_rq(rq) && | |
1911 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1912 | return 1; | |
1913 | } | |
1914 | ||
1915 | return 0; | |
1916 | } | |
1917 | ||
336cdb40 | 1918 | /** |
e19a3ab0 KU |
1919 | * blk_end_io - Generic end_io function to complete a request. |
1920 | * @rq: the request being processed | |
710027a4 | 1921 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1922 | * @nr_bytes: number of bytes to complete @rq |
1923 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1924 | * @drv_callback: function called between completion of bios in the request |
1925 | * and completion of the request. | |
710027a4 | 1926 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1927 | * completion of the request. |
336cdb40 KU |
1928 | * |
1929 | * Description: | |
e3a04fe3 | 1930 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1931 | * If @rq has leftover, sets it up for the next range of segments. |
1932 | * | |
1933 | * Return: | |
710027a4 RD |
1934 | * %0 - we are done with this request |
1935 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1936 | **/ |
22b13210 JA |
1937 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1938 | unsigned int bidi_bytes, | |
1939 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1940 | { |
1941 | struct request_queue *q = rq->q; | |
1942 | unsigned long flags = 0UL; | |
336cdb40 | 1943 | |
32fab448 KU |
1944 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
1945 | return 1; | |
336cdb40 | 1946 | |
e19a3ab0 KU |
1947 | /* Special feature for tricky drivers */ |
1948 | if (drv_callback && drv_callback(rq)) | |
1949 | return 1; | |
1950 | ||
336cdb40 KU |
1951 | add_disk_randomness(rq->rq_disk); |
1952 | ||
1953 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1954 | end_that_request_last(rq, error); |
336cdb40 KU |
1955 | spin_unlock_irqrestore(q->queue_lock, flags); |
1956 | ||
1957 | return 0; | |
1958 | } | |
e19a3ab0 KU |
1959 | |
1960 | /** | |
1961 | * blk_end_request - Helper function for drivers to complete the request. | |
1962 | * @rq: the request being processed | |
710027a4 | 1963 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
1964 | * @nr_bytes: number of bytes to complete |
1965 | * | |
1966 | * Description: | |
1967 | * Ends I/O on a number of bytes attached to @rq. | |
1968 | * If @rq has leftover, sets it up for the next range of segments. | |
1969 | * | |
1970 | * Return: | |
710027a4 RD |
1971 | * %0 - we are done with this request |
1972 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1973 | **/ |
22b13210 | 1974 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1975 | { |
e3a04fe3 | 1976 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1977 | } |
336cdb40 KU |
1978 | EXPORT_SYMBOL_GPL(blk_end_request); |
1979 | ||
1980 | /** | |
1981 | * __blk_end_request - Helper function for drivers to complete the request. | |
1982 | * @rq: the request being processed | |
710027a4 | 1983 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
1984 | * @nr_bytes: number of bytes to complete |
1985 | * | |
1986 | * Description: | |
1987 | * Must be called with queue lock held unlike blk_end_request(). | |
1988 | * | |
1989 | * Return: | |
710027a4 RD |
1990 | * %0 - we are done with this request |
1991 | * %1 - still buffers pending for this request | |
336cdb40 | 1992 | **/ |
22b13210 | 1993 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1994 | { |
60540161 | 1995 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
051cc395 | 1996 | return 1; |
336cdb40 KU |
1997 | |
1998 | add_disk_randomness(rq->rq_disk); | |
1999 | ||
b8286239 | 2000 | end_that_request_last(rq, error); |
336cdb40 KU |
2001 | |
2002 | return 0; | |
2003 | } | |
2004 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
2005 | ||
e3a04fe3 KU |
2006 | /** |
2007 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
2008 | * @rq: the bidi request being processed | |
710027a4 | 2009 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2010 | * @nr_bytes: number of bytes to complete @rq |
2011 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2012 | * | |
2013 | * Description: | |
2014 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2015 | * | |
2016 | * Return: | |
710027a4 RD |
2017 | * %0 - we are done with this request |
2018 | * %1 - still buffers pending for this request | |
e3a04fe3 | 2019 | **/ |
22b13210 JA |
2020 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
2021 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
2022 | { |
2023 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
2024 | } | |
2025 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2026 | ||
32fab448 KU |
2027 | /** |
2028 | * blk_update_request - Special helper function for request stacking drivers | |
2029 | * @rq: the request being processed | |
2030 | * @error: %0 for success, < %0 for error | |
2031 | * @nr_bytes: number of bytes to complete @rq | |
2032 | * | |
2033 | * Description: | |
2034 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2035 | * the request structure even if @rq doesn't have leftover. | |
2036 | * If @rq has leftover, sets it up for the next range of segments. | |
2037 | * | |
2038 | * This special helper function is only for request stacking drivers | |
2039 | * (e.g. request-based dm) so that they can handle partial completion. | |
2040 | * Actual device drivers should use blk_end_request instead. | |
2041 | */ | |
2042 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2043 | { | |
2044 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2045 | /* | |
2046 | * These members are not updated in end_that_request_data() | |
2047 | * when all bios are completed. | |
2048 | * Update them so that the request stacking driver can find | |
2049 | * how many bytes remain in the request later. | |
2050 | */ | |
2051 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2052 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2053 | } | |
2054 | } | |
2055 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2056 | ||
e19a3ab0 KU |
2057 | /** |
2058 | * blk_end_request_callback - Special helper function for tricky drivers | |
2059 | * @rq: the request being processed | |
710027a4 | 2060 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2061 | * @nr_bytes: number of bytes to complete |
2062 | * @drv_callback: function called between completion of bios in the request | |
2063 | * and completion of the request. | |
710027a4 | 2064 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
2065 | * completion of the request. |
2066 | * | |
2067 | * Description: | |
2068 | * Ends I/O on a number of bytes attached to @rq. | |
2069 | * If @rq has leftover, sets it up for the next range of segments. | |
2070 | * | |
2071 | * This special helper function is used only for existing tricky drivers. | |
2072 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2073 | * This interface will be removed when such drivers are rewritten. | |
2074 | * Don't use this interface in other places anymore. | |
2075 | * | |
2076 | * Return: | |
710027a4 RD |
2077 | * %0 - we are done with this request |
2078 | * %1 - this request is not freed yet. | |
2079 | * this request still has pending buffers or | |
2080 | * the driver doesn't want to finish this request yet. | |
e19a3ab0 | 2081 | **/ |
22b13210 JA |
2082 | int blk_end_request_callback(struct request *rq, int error, |
2083 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2084 | int (drv_callback)(struct request *)) |
2085 | { | |
e3a04fe3 | 2086 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2087 | } |
2088 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2089 | ||
86db1e29 JA |
2090 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2091 | struct bio *bio) | |
1da177e4 | 2092 | { |
d628eaef DW |
2093 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2094 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2095 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2096 | |
fb2dce86 DW |
2097 | if (bio_has_data(bio)) { |
2098 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2099 | rq->buffer = bio_data(bio); |
2100 | } | |
1da177e4 LT |
2101 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2102 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2103 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2104 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2105 | |
2106 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2107 | |
66846572 N |
2108 | if (bio->bi_bdev) |
2109 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2110 | } | |
1da177e4 | 2111 | |
ef9e3fac KU |
2112 | /** |
2113 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2114 | * @q : the queue of the device being checked | |
2115 | * | |
2116 | * Description: | |
2117 | * Check if underlying low-level drivers of a device are busy. | |
2118 | * If the drivers want to export their busy state, they must set own | |
2119 | * exporting function using blk_queue_lld_busy() first. | |
2120 | * | |
2121 | * Basically, this function is used only by request stacking drivers | |
2122 | * to stop dispatching requests to underlying devices when underlying | |
2123 | * devices are busy. This behavior helps more I/O merging on the queue | |
2124 | * of the request stacking driver and prevents I/O throughput regression | |
2125 | * on burst I/O load. | |
2126 | * | |
2127 | * Return: | |
2128 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2129 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2130 | */ | |
2131 | int blk_lld_busy(struct request_queue *q) | |
2132 | { | |
2133 | if (q->lld_busy_fn) | |
2134 | return q->lld_busy_fn(q); | |
2135 | ||
2136 | return 0; | |
2137 | } | |
2138 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2139 | ||
18887ad9 | 2140 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2141 | { |
2142 | return queue_work(kblockd_workqueue, work); | |
2143 | } | |
1da177e4 LT |
2144 | EXPORT_SYMBOL(kblockd_schedule_work); |
2145 | ||
19a75d83 | 2146 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 2147 | { |
28e53bdd | 2148 | cancel_work_sync(work); |
1da177e4 | 2149 | } |
19a75d83 | 2150 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
2151 | |
2152 | int __init blk_dev_init(void) | |
2153 | { | |
2154 | kblockd_workqueue = create_workqueue("kblockd"); | |
2155 | if (!kblockd_workqueue) | |
2156 | panic("Failed to create kblockd\n"); | |
2157 | ||
2158 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2159 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2160 | |
8324aa91 | 2161 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2162 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2163 | |
d38ecf93 | 2164 | return 0; |
1da177e4 | 2165 | } |
1da177e4 | 2166 |