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 | ||
165125e1 | 35 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
36 | |
37 | /* | |
38 | * For the allocated request tables | |
39 | */ | |
5ece6c52 | 40 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
41 | |
42 | /* | |
43 | * For queue allocation | |
44 | */ | |
6728cb0e | 45 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 46 | |
1da177e4 LT |
47 | /* |
48 | * Controlling structure to kblockd | |
49 | */ | |
ff856bad | 50 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 51 | |
26b8256e JA |
52 | static void drive_stat_acct(struct request *rq, int new_io) |
53 | { | |
28f13702 | 54 | struct hd_struct *part; |
26b8256e | 55 | int rw = rq_data_dir(rq); |
c9959059 | 56 | int cpu; |
26b8256e JA |
57 | |
58 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
59 | return; | |
60 | ||
074a7aca | 61 | cpu = part_stat_lock(); |
e71bf0d0 | 62 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
c9959059 | 63 | |
28f13702 | 64 | if (!new_io) |
074a7aca | 65 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 66 | else { |
074a7aca TH |
67 | part_round_stats(cpu, part); |
68 | part_inc_in_flight(part); | |
26b8256e | 69 | } |
e71bf0d0 | 70 | |
074a7aca | 71 | part_stat_unlock(); |
26b8256e JA |
72 | } |
73 | ||
8324aa91 | 74 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
75 | { |
76 | int nr; | |
77 | ||
78 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
79 | if (nr > q->nr_requests) | |
80 | nr = q->nr_requests; | |
81 | q->nr_congestion_on = nr; | |
82 | ||
83 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
84 | if (nr < 1) | |
85 | nr = 1; | |
86 | q->nr_congestion_off = nr; | |
87 | } | |
88 | ||
1da177e4 LT |
89 | /** |
90 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
91 | * @bdev: device | |
92 | * | |
93 | * Locates the passed device's request queue and returns the address of its | |
94 | * backing_dev_info | |
95 | * | |
96 | * Will return NULL if the request queue cannot be located. | |
97 | */ | |
98 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
99 | { | |
100 | struct backing_dev_info *ret = NULL; | |
165125e1 | 101 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
102 | |
103 | if (q) | |
104 | ret = &q->backing_dev_info; | |
105 | return ret; | |
106 | } | |
1da177e4 LT |
107 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
108 | ||
2a4aa30c | 109 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 110 | { |
1afb20f3 FT |
111 | memset(rq, 0, sizeof(*rq)); |
112 | ||
1da177e4 | 113 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 114 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 115 | rq->cpu = -1; |
63a71386 JA |
116 | rq->q = q; |
117 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
118 | INIT_HLIST_NODE(&rq->hash); |
119 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 120 | rq->cmd = rq->__cmd; |
63a71386 | 121 | rq->tag = -1; |
1da177e4 | 122 | rq->ref_count = 1; |
1da177e4 | 123 | } |
2a4aa30c | 124 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 125 | |
5bb23a68 N |
126 | static void req_bio_endio(struct request *rq, struct bio *bio, |
127 | unsigned int nbytes, int error) | |
1da177e4 | 128 | { |
165125e1 | 129 | struct request_queue *q = rq->q; |
797e7dbb | 130 | |
5bb23a68 N |
131 | if (&q->bar_rq != rq) { |
132 | if (error) | |
133 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
134 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
135 | error = -EIO; | |
797e7dbb | 136 | |
5bb23a68 | 137 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 138 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 139 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
140 | nbytes = bio->bi_size; |
141 | } | |
797e7dbb | 142 | |
5bb23a68 N |
143 | bio->bi_size -= nbytes; |
144 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
145 | |
146 | if (bio_integrity(bio)) | |
147 | bio_integrity_advance(bio, nbytes); | |
148 | ||
5bb23a68 | 149 | if (bio->bi_size == 0) |
6712ecf8 | 150 | bio_endio(bio, error); |
5bb23a68 N |
151 | } else { |
152 | ||
153 | /* | |
154 | * Okay, this is the barrier request in progress, just | |
155 | * record the error; | |
156 | */ | |
157 | if (error && !q->orderr) | |
158 | q->orderr = error; | |
159 | } | |
1da177e4 | 160 | } |
1da177e4 | 161 | |
1da177e4 LT |
162 | void blk_dump_rq_flags(struct request *rq, char *msg) |
163 | { | |
164 | int bit; | |
165 | ||
6728cb0e | 166 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
167 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
168 | rq->cmd_flags); | |
1da177e4 | 169 | |
6728cb0e JA |
170 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
171 | (unsigned long long)rq->sector, | |
172 | rq->nr_sectors, | |
173 | rq->current_nr_sectors); | |
174 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
175 | rq->bio, rq->biotail, | |
176 | rq->buffer, rq->data, | |
177 | rq->data_len); | |
1da177e4 | 178 | |
4aff5e23 | 179 | if (blk_pc_request(rq)) { |
6728cb0e | 180 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 181 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
182 | printk("%02x ", rq->cmd[bit]); |
183 | printk("\n"); | |
184 | } | |
185 | } | |
1da177e4 LT |
186 | EXPORT_SYMBOL(blk_dump_rq_flags); |
187 | ||
1da177e4 LT |
188 | /* |
189 | * "plug" the device if there are no outstanding requests: this will | |
190 | * force the transfer to start only after we have put all the requests | |
191 | * on the list. | |
192 | * | |
193 | * This is called with interrupts off and no requests on the queue and | |
194 | * with the queue lock held. | |
195 | */ | |
165125e1 | 196 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
197 | { |
198 | WARN_ON(!irqs_disabled()); | |
199 | ||
200 | /* | |
201 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
202 | * which will restart the queueing | |
203 | */ | |
7daac490 | 204 | if (blk_queue_stopped(q)) |
1da177e4 LT |
205 | return; |
206 | ||
e48ec690 | 207 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 208 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 209 | trace_block_plug(q); |
2056a782 | 210 | } |
1da177e4 | 211 | } |
1da177e4 LT |
212 | EXPORT_SYMBOL(blk_plug_device); |
213 | ||
6c5e0c4d JA |
214 | /** |
215 | * blk_plug_device_unlocked - plug a device without queue lock held | |
216 | * @q: The &struct request_queue to plug | |
217 | * | |
218 | * Description: | |
219 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
220 | * interrupts. | |
221 | **/ | |
222 | void blk_plug_device_unlocked(struct request_queue *q) | |
223 | { | |
224 | unsigned long flags; | |
225 | ||
226 | spin_lock_irqsave(q->queue_lock, flags); | |
227 | blk_plug_device(q); | |
228 | spin_unlock_irqrestore(q->queue_lock, flags); | |
229 | } | |
230 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
231 | ||
1da177e4 LT |
232 | /* |
233 | * remove the queue from the plugged list, if present. called with | |
234 | * queue lock held and interrupts disabled. | |
235 | */ | |
165125e1 | 236 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
237 | { |
238 | WARN_ON(!irqs_disabled()); | |
239 | ||
e48ec690 | 240 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
241 | return 0; |
242 | ||
243 | del_timer(&q->unplug_timer); | |
244 | return 1; | |
245 | } | |
1da177e4 LT |
246 | EXPORT_SYMBOL(blk_remove_plug); |
247 | ||
248 | /* | |
249 | * remove the plug and let it rip.. | |
250 | */ | |
165125e1 | 251 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 252 | { |
7daac490 | 253 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
254 | return; |
255 | ||
256 | if (!blk_remove_plug(q)) | |
257 | return; | |
258 | ||
22e2c507 | 259 | q->request_fn(q); |
1da177e4 | 260 | } |
1da177e4 LT |
261 | |
262 | /** | |
263 | * generic_unplug_device - fire a request queue | |
165125e1 | 264 | * @q: The &struct request_queue in question |
1da177e4 LT |
265 | * |
266 | * Description: | |
267 | * Linux uses plugging to build bigger requests queues before letting | |
268 | * the device have at them. If a queue is plugged, the I/O scheduler | |
269 | * is still adding and merging requests on the queue. Once the queue | |
270 | * gets unplugged, the request_fn defined for the queue is invoked and | |
271 | * transfers started. | |
272 | **/ | |
165125e1 | 273 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 274 | { |
dbaf2c00 JA |
275 | if (blk_queue_plugged(q)) { |
276 | spin_lock_irq(q->queue_lock); | |
277 | __generic_unplug_device(q); | |
278 | spin_unlock_irq(q->queue_lock); | |
279 | } | |
1da177e4 LT |
280 | } |
281 | EXPORT_SYMBOL(generic_unplug_device); | |
282 | ||
283 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
284 | struct page *page) | |
285 | { | |
165125e1 | 286 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 287 | |
2ad8b1ef | 288 | blk_unplug(q); |
1da177e4 LT |
289 | } |
290 | ||
86db1e29 | 291 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 292 | { |
165125e1 JA |
293 | struct request_queue *q = |
294 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 295 | |
5f3ea37c | 296 | trace_block_unplug_io(q); |
1da177e4 LT |
297 | q->unplug_fn(q); |
298 | } | |
299 | ||
86db1e29 | 300 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 301 | { |
165125e1 | 302 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 303 | |
5f3ea37c | 304 | trace_block_unplug_timer(q); |
18887ad9 | 305 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
306 | } |
307 | ||
2ad8b1ef AB |
308 | void blk_unplug(struct request_queue *q) |
309 | { | |
310 | /* | |
311 | * devices don't necessarily have an ->unplug_fn defined | |
312 | */ | |
313 | if (q->unplug_fn) { | |
5f3ea37c | 314 | trace_block_unplug_io(q); |
2ad8b1ef AB |
315 | q->unplug_fn(q); |
316 | } | |
317 | } | |
318 | EXPORT_SYMBOL(blk_unplug); | |
319 | ||
c7c22e4d JA |
320 | static void blk_invoke_request_fn(struct request_queue *q) |
321 | { | |
80a4b58e JA |
322 | if (unlikely(blk_queue_stopped(q))) |
323 | return; | |
324 | ||
c7c22e4d JA |
325 | /* |
326 | * one level of recursion is ok and is much faster than kicking | |
327 | * the unplug handling | |
328 | */ | |
329 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
330 | q->request_fn(q); | |
331 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
332 | } else { | |
333 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
334 | kblockd_schedule_work(q, &q->unplug_work); | |
335 | } | |
336 | } | |
337 | ||
1da177e4 LT |
338 | /** |
339 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 340 | * @q: The &struct request_queue in question |
1da177e4 LT |
341 | * |
342 | * Description: | |
343 | * blk_start_queue() will clear the stop flag on the queue, and call | |
344 | * the request_fn for the queue if it was in a stopped state when | |
345 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
346 | **/ | |
165125e1 | 347 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 348 | { |
a038e253 PBG |
349 | WARN_ON(!irqs_disabled()); |
350 | ||
75ad23bc | 351 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
c7c22e4d | 352 | blk_invoke_request_fn(q); |
1da177e4 | 353 | } |
1da177e4 LT |
354 | EXPORT_SYMBOL(blk_start_queue); |
355 | ||
356 | /** | |
357 | * blk_stop_queue - stop a queue | |
165125e1 | 358 | * @q: The &struct request_queue in question |
1da177e4 LT |
359 | * |
360 | * Description: | |
361 | * The Linux block layer assumes that a block driver will consume all | |
362 | * entries on the request queue when the request_fn strategy is called. | |
363 | * Often this will not happen, because of hardware limitations (queue | |
364 | * depth settings). If a device driver gets a 'queue full' response, | |
365 | * or if it simply chooses not to queue more I/O at one point, it can | |
366 | * call this function to prevent the request_fn from being called until | |
367 | * the driver has signalled it's ready to go again. This happens by calling | |
368 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
369 | **/ | |
165125e1 | 370 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
371 | { |
372 | blk_remove_plug(q); | |
75ad23bc | 373 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
374 | } |
375 | EXPORT_SYMBOL(blk_stop_queue); | |
376 | ||
377 | /** | |
378 | * blk_sync_queue - cancel any pending callbacks on a queue | |
379 | * @q: the queue | |
380 | * | |
381 | * Description: | |
382 | * The block layer may perform asynchronous callback activity | |
383 | * on a queue, such as calling the unplug function after a timeout. | |
384 | * A block device may call blk_sync_queue to ensure that any | |
385 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 386 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
387 | * that its ->make_request_fn will not re-add plugging prior to calling |
388 | * this function. | |
389 | * | |
390 | */ | |
391 | void blk_sync_queue(struct request_queue *q) | |
392 | { | |
393 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 394 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
395 | } |
396 | EXPORT_SYMBOL(blk_sync_queue); | |
397 | ||
398 | /** | |
80a4b58e | 399 | * __blk_run_queue - run a single device queue |
1da177e4 | 400 | * @q: The queue to run |
80a4b58e JA |
401 | * |
402 | * Description: | |
403 | * See @blk_run_queue. This variant must be called with the queue lock | |
404 | * held and interrupts disabled. | |
405 | * | |
1da177e4 | 406 | */ |
75ad23bc | 407 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 408 | { |
1da177e4 | 409 | blk_remove_plug(q); |
dac07ec1 JA |
410 | |
411 | /* | |
412 | * Only recurse once to avoid overrunning the stack, let the unplug | |
413 | * handling reinvoke the handler shortly if we already got there. | |
414 | */ | |
c7c22e4d JA |
415 | if (!elv_queue_empty(q)) |
416 | blk_invoke_request_fn(q); | |
75ad23bc NP |
417 | } |
418 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 419 | |
75ad23bc NP |
420 | /** |
421 | * blk_run_queue - run a single device queue | |
422 | * @q: The queue to run | |
80a4b58e JA |
423 | * |
424 | * Description: | |
425 | * Invoke request handling on this queue, if it has pending work to do. | |
426 | * May be used to restart queueing when a request has completed. Also | |
427 | * See @blk_start_queueing. | |
428 | * | |
75ad23bc NP |
429 | */ |
430 | void blk_run_queue(struct request_queue *q) | |
431 | { | |
432 | unsigned long flags; | |
433 | ||
434 | spin_lock_irqsave(q->queue_lock, flags); | |
435 | __blk_run_queue(q); | |
1da177e4 LT |
436 | spin_unlock_irqrestore(q->queue_lock, flags); |
437 | } | |
438 | EXPORT_SYMBOL(blk_run_queue); | |
439 | ||
165125e1 | 440 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
441 | { |
442 | kobject_put(&q->kobj); | |
443 | } | |
483f4afc | 444 | |
6728cb0e | 445 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 446 | { |
e3335de9 JA |
447 | /* |
448 | * We know we have process context here, so we can be a little | |
449 | * cautious and ensure that pending block actions on this device | |
450 | * are done before moving on. Going into this function, we should | |
451 | * not have processes doing IO to this device. | |
452 | */ | |
453 | blk_sync_queue(q); | |
454 | ||
483f4afc | 455 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 456 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
457 | mutex_unlock(&q->sysfs_lock); |
458 | ||
459 | if (q->elevator) | |
460 | elevator_exit(q->elevator); | |
461 | ||
462 | blk_put_queue(q); | |
463 | } | |
1da177e4 LT |
464 | EXPORT_SYMBOL(blk_cleanup_queue); |
465 | ||
165125e1 | 466 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
467 | { |
468 | struct request_list *rl = &q->rq; | |
469 | ||
470 | rl->count[READ] = rl->count[WRITE] = 0; | |
471 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 472 | rl->elvpriv = 0; |
1da177e4 LT |
473 | init_waitqueue_head(&rl->wait[READ]); |
474 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 475 | |
1946089a CL |
476 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
477 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
478 | |
479 | if (!rl->rq_pool) | |
480 | return -ENOMEM; | |
481 | ||
482 | return 0; | |
483 | } | |
484 | ||
165125e1 | 485 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 486 | { |
1946089a CL |
487 | return blk_alloc_queue_node(gfp_mask, -1); |
488 | } | |
489 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 490 | |
165125e1 | 491 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 492 | { |
165125e1 | 493 | struct request_queue *q; |
e0bf68dd | 494 | int err; |
1946089a | 495 | |
8324aa91 | 496 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 497 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
498 | if (!q) |
499 | return NULL; | |
500 | ||
e0bf68dd PZ |
501 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
502 | q->backing_dev_info.unplug_io_data = q; | |
503 | err = bdi_init(&q->backing_dev_info); | |
504 | if (err) { | |
8324aa91 | 505 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
506 | return NULL; |
507 | } | |
508 | ||
1da177e4 | 509 | init_timer(&q->unplug_timer); |
242f9dcb JA |
510 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
511 | INIT_LIST_HEAD(&q->timeout_list); | |
713ada9b | 512 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 513 | |
8324aa91 | 514 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 515 | |
483f4afc | 516 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 517 | spin_lock_init(&q->__queue_lock); |
483f4afc | 518 | |
1da177e4 LT |
519 | return q; |
520 | } | |
1946089a | 521 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
522 | |
523 | /** | |
524 | * blk_init_queue - prepare a request queue for use with a block device | |
525 | * @rfn: The function to be called to process requests that have been | |
526 | * placed on the queue. | |
527 | * @lock: Request queue spin lock | |
528 | * | |
529 | * Description: | |
530 | * If a block device wishes to use the standard request handling procedures, | |
531 | * which sorts requests and coalesces adjacent requests, then it must | |
532 | * call blk_init_queue(). The function @rfn will be called when there | |
533 | * are requests on the queue that need to be processed. If the device | |
534 | * supports plugging, then @rfn may not be called immediately when requests | |
535 | * are available on the queue, but may be called at some time later instead. | |
536 | * Plugged queues are generally unplugged when a buffer belonging to one | |
537 | * of the requests on the queue is needed, or due to memory pressure. | |
538 | * | |
539 | * @rfn is not required, or even expected, to remove all requests off the | |
540 | * queue, but only as many as it can handle at a time. If it does leave | |
541 | * requests on the queue, it is responsible for arranging that the requests | |
542 | * get dealt with eventually. | |
543 | * | |
544 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
545 | * request queue; this lock will be taken also from interrupt context, so irq |
546 | * disabling is needed for it. | |
1da177e4 | 547 | * |
710027a4 | 548 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
549 | * it didn't succeed. |
550 | * | |
551 | * Note: | |
552 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
553 | * when the block device is deactivated (such as at module unload). | |
554 | **/ | |
1946089a | 555 | |
165125e1 | 556 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 557 | { |
1946089a CL |
558 | return blk_init_queue_node(rfn, lock, -1); |
559 | } | |
560 | EXPORT_SYMBOL(blk_init_queue); | |
561 | ||
165125e1 | 562 | struct request_queue * |
1946089a CL |
563 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
564 | { | |
165125e1 | 565 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
566 | |
567 | if (!q) | |
568 | return NULL; | |
569 | ||
1946089a | 570 | q->node = node_id; |
8669aafd | 571 | if (blk_init_free_list(q)) { |
8324aa91 | 572 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
573 | return NULL; |
574 | } | |
1da177e4 | 575 | |
152587de | 576 | /* |
577 | * if caller didn't supply a lock, they get per-queue locking with | |
578 | * our embedded lock | |
579 | */ | |
e7e72bf6 | 580 | if (!lock) |
152587de | 581 | lock = &q->__queue_lock; |
152587de | 582 | |
1da177e4 | 583 | q->request_fn = rfn; |
1da177e4 LT |
584 | q->prep_rq_fn = NULL; |
585 | q->unplug_fn = generic_unplug_device; | |
4ee5eaf4 KU |
586 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | |
587 | 1 << QUEUE_FLAG_STACKABLE); | |
1da177e4 LT |
588 | q->queue_lock = lock; |
589 | ||
590 | blk_queue_segment_boundary(q, 0xffffffff); | |
591 | ||
592 | blk_queue_make_request(q, __make_request); | |
593 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
594 | ||
595 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
596 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
597 | ||
44ec9542 AS |
598 | q->sg_reserved_size = INT_MAX; |
599 | ||
abf54393 FT |
600 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
601 | ||
1da177e4 LT |
602 | /* |
603 | * all done | |
604 | */ | |
605 | if (!elevator_init(q, NULL)) { | |
606 | blk_queue_congestion_threshold(q); | |
607 | return q; | |
608 | } | |
609 | ||
8669aafd | 610 | blk_put_queue(q); |
1da177e4 LT |
611 | return NULL; |
612 | } | |
1946089a | 613 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 614 | |
165125e1 | 615 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 616 | { |
fde6ad22 | 617 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 618 | kobject_get(&q->kobj); |
1da177e4 LT |
619 | return 0; |
620 | } | |
621 | ||
622 | return 1; | |
623 | } | |
1da177e4 | 624 | |
165125e1 | 625 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 626 | { |
4aff5e23 | 627 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 628 | elv_put_request(q, rq); |
1da177e4 LT |
629 | mempool_free(rq, q->rq.rq_pool); |
630 | } | |
631 | ||
1ea25ecb | 632 | static struct request * |
165125e1 | 633 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
634 | { |
635 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
636 | ||
637 | if (!rq) | |
638 | return NULL; | |
639 | ||
2a4aa30c | 640 | blk_rq_init(q, rq); |
1afb20f3 | 641 | |
49171e5c | 642 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 643 | |
cb98fc8b | 644 | if (priv) { |
cb78b285 | 645 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
646 | mempool_free(rq, q->rq.rq_pool); |
647 | return NULL; | |
648 | } | |
4aff5e23 | 649 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 650 | } |
1da177e4 | 651 | |
cb98fc8b | 652 | return rq; |
1da177e4 LT |
653 | } |
654 | ||
655 | /* | |
656 | * ioc_batching returns true if the ioc is a valid batching request and | |
657 | * should be given priority access to a request. | |
658 | */ | |
165125e1 | 659 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
660 | { |
661 | if (!ioc) | |
662 | return 0; | |
663 | ||
664 | /* | |
665 | * Make sure the process is able to allocate at least 1 request | |
666 | * even if the batch times out, otherwise we could theoretically | |
667 | * lose wakeups. | |
668 | */ | |
669 | return ioc->nr_batch_requests == q->nr_batching || | |
670 | (ioc->nr_batch_requests > 0 | |
671 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
672 | } | |
673 | ||
674 | /* | |
675 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
676 | * will cause the process to be a "batcher" on all queues in the system. This | |
677 | * is the behaviour we want though - once it gets a wakeup it should be given | |
678 | * a nice run. | |
679 | */ | |
165125e1 | 680 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
681 | { |
682 | if (!ioc || ioc_batching(q, ioc)) | |
683 | return; | |
684 | ||
685 | ioc->nr_batch_requests = q->nr_batching; | |
686 | ioc->last_waited = jiffies; | |
687 | } | |
688 | ||
165125e1 | 689 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
690 | { |
691 | struct request_list *rl = &q->rq; | |
692 | ||
693 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 694 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
695 | |
696 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
697 | if (waitqueue_active(&rl->wait[rw])) |
698 | wake_up(&rl->wait[rw]); | |
699 | ||
700 | blk_clear_queue_full(q, rw); | |
701 | } | |
702 | } | |
703 | ||
704 | /* | |
705 | * A request has just been released. Account for it, update the full and | |
706 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
707 | */ | |
165125e1 | 708 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
709 | { |
710 | struct request_list *rl = &q->rq; | |
711 | ||
712 | rl->count[rw]--; | |
cb98fc8b TH |
713 | if (priv) |
714 | rl->elvpriv--; | |
1da177e4 LT |
715 | |
716 | __freed_request(q, rw); | |
717 | ||
718 | if (unlikely(rl->starved[rw ^ 1])) | |
719 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
720 | } |
721 | ||
722 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
723 | /* | |
d6344532 NP |
724 | * Get a free request, queue_lock must be held. |
725 | * Returns NULL on failure, with queue_lock held. | |
726 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 727 | */ |
165125e1 | 728 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 729 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
730 | { |
731 | struct request *rq = NULL; | |
732 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 733 | struct io_context *ioc = NULL; |
7749a8d4 | 734 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
735 | int may_queue, priv; |
736 | ||
7749a8d4 | 737 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
738 | if (may_queue == ELV_MQUEUE_NO) |
739 | goto rq_starved; | |
740 | ||
741 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
742 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 743 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
744 | /* |
745 | * The queue will fill after this allocation, so set | |
746 | * it as full, and mark this process as "batching". | |
747 | * This process will be allowed to complete a batch of | |
748 | * requests, others will be blocked. | |
749 | */ | |
750 | if (!blk_queue_full(q, rw)) { | |
751 | ioc_set_batching(q, ioc); | |
752 | blk_set_queue_full(q, rw); | |
753 | } else { | |
754 | if (may_queue != ELV_MQUEUE_MUST | |
755 | && !ioc_batching(q, ioc)) { | |
756 | /* | |
757 | * The queue is full and the allocating | |
758 | * process is not a "batcher", and not | |
759 | * exempted by the IO scheduler | |
760 | */ | |
761 | goto out; | |
762 | } | |
763 | } | |
1da177e4 | 764 | } |
79e2de4b | 765 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
766 | } |
767 | ||
082cf69e JA |
768 | /* |
769 | * Only allow batching queuers to allocate up to 50% over the defined | |
770 | * limit of requests, otherwise we could have thousands of requests | |
771 | * allocated with any setting of ->nr_requests | |
772 | */ | |
fd782a4a | 773 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 774 | goto out; |
fd782a4a | 775 | |
1da177e4 LT |
776 | rl->count[rw]++; |
777 | rl->starved[rw] = 0; | |
cb98fc8b | 778 | |
64521d1a | 779 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
780 | if (priv) |
781 | rl->elvpriv++; | |
782 | ||
1da177e4 LT |
783 | spin_unlock_irq(q->queue_lock); |
784 | ||
7749a8d4 | 785 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 786 | if (unlikely(!rq)) { |
1da177e4 LT |
787 | /* |
788 | * Allocation failed presumably due to memory. Undo anything | |
789 | * we might have messed up. | |
790 | * | |
791 | * Allocating task should really be put onto the front of the | |
792 | * wait queue, but this is pretty rare. | |
793 | */ | |
794 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 795 | freed_request(q, rw, priv); |
1da177e4 LT |
796 | |
797 | /* | |
798 | * in the very unlikely event that allocation failed and no | |
799 | * requests for this direction was pending, mark us starved | |
800 | * so that freeing of a request in the other direction will | |
801 | * notice us. another possible fix would be to split the | |
802 | * rq mempool into READ and WRITE | |
803 | */ | |
804 | rq_starved: | |
805 | if (unlikely(rl->count[rw] == 0)) | |
806 | rl->starved[rw] = 1; | |
807 | ||
1da177e4 LT |
808 | goto out; |
809 | } | |
810 | ||
88ee5ef1 JA |
811 | /* |
812 | * ioc may be NULL here, and ioc_batching will be false. That's | |
813 | * OK, if the queue is under the request limit then requests need | |
814 | * not count toward the nr_batch_requests limit. There will always | |
815 | * be some limit enforced by BLK_BATCH_TIME. | |
816 | */ | |
1da177e4 LT |
817 | if (ioc_batching(q, ioc)) |
818 | ioc->nr_batch_requests--; | |
6728cb0e | 819 | |
5f3ea37c | 820 | trace_block_getrq(q, bio, rw); |
1da177e4 | 821 | out: |
1da177e4 LT |
822 | return rq; |
823 | } | |
824 | ||
825 | /* | |
826 | * No available requests for this queue, unplug the device and wait for some | |
827 | * requests to become available. | |
d6344532 NP |
828 | * |
829 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 830 | */ |
165125e1 | 831 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 832 | struct bio *bio) |
1da177e4 | 833 | { |
7749a8d4 | 834 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
835 | struct request *rq; |
836 | ||
7749a8d4 | 837 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
838 | while (!rq) { |
839 | DEFINE_WAIT(wait); | |
05caf8db | 840 | struct io_context *ioc; |
1da177e4 LT |
841 | struct request_list *rl = &q->rq; |
842 | ||
843 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
844 | TASK_UNINTERRUPTIBLE); | |
845 | ||
5f3ea37c | 846 | trace_block_sleeprq(q, bio, rw); |
1da177e4 | 847 | |
05caf8db ZY |
848 | __generic_unplug_device(q); |
849 | spin_unlock_irq(q->queue_lock); | |
850 | io_schedule(); | |
1da177e4 | 851 | |
05caf8db ZY |
852 | /* |
853 | * After sleeping, we become a "batching" process and | |
854 | * will be able to allocate at least one request, and | |
855 | * up to a big batch of them for a small period time. | |
856 | * See ioc_batching, ioc_set_batching | |
857 | */ | |
858 | ioc = current_io_context(GFP_NOIO, q->node); | |
859 | ioc_set_batching(q, ioc); | |
d6344532 | 860 | |
05caf8db | 861 | spin_lock_irq(q->queue_lock); |
1da177e4 | 862 | finish_wait(&rl->wait[rw], &wait); |
05caf8db ZY |
863 | |
864 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
865 | }; | |
1da177e4 LT |
866 | |
867 | return rq; | |
868 | } | |
869 | ||
165125e1 | 870 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
871 | { |
872 | struct request *rq; | |
873 | ||
874 | BUG_ON(rw != READ && rw != WRITE); | |
875 | ||
d6344532 NP |
876 | spin_lock_irq(q->queue_lock); |
877 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 878 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 879 | } else { |
22e2c507 | 880 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
881 | if (!rq) |
882 | spin_unlock_irq(q->queue_lock); | |
883 | } | |
884 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
885 | |
886 | return rq; | |
887 | } | |
1da177e4 LT |
888 | EXPORT_SYMBOL(blk_get_request); |
889 | ||
dc72ef4a JA |
890 | /** |
891 | * blk_start_queueing - initiate dispatch of requests to device | |
892 | * @q: request queue to kick into gear | |
893 | * | |
894 | * This is basically a helper to remove the need to know whether a queue | |
895 | * is plugged or not if someone just wants to initiate dispatch of requests | |
80a4b58e JA |
896 | * for this queue. Should be used to start queueing on a device outside |
897 | * of ->request_fn() context. Also see @blk_run_queue. | |
dc72ef4a JA |
898 | * |
899 | * The queue lock must be held with interrupts disabled. | |
900 | */ | |
165125e1 | 901 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a | 902 | { |
336c3d8c EO |
903 | if (!blk_queue_plugged(q)) { |
904 | if (unlikely(blk_queue_stopped(q))) | |
905 | return; | |
dc72ef4a | 906 | q->request_fn(q); |
336c3d8c | 907 | } else |
dc72ef4a JA |
908 | __generic_unplug_device(q); |
909 | } | |
910 | EXPORT_SYMBOL(blk_start_queueing); | |
911 | ||
1da177e4 LT |
912 | /** |
913 | * blk_requeue_request - put a request back on queue | |
914 | * @q: request queue where request should be inserted | |
915 | * @rq: request to be inserted | |
916 | * | |
917 | * Description: | |
918 | * Drivers often keep queueing requests until the hardware cannot accept | |
919 | * more, when that condition happens we need to put the request back | |
920 | * on the queue. Must be called with queue lock held. | |
921 | */ | |
165125e1 | 922 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 923 | { |
242f9dcb JA |
924 | blk_delete_timer(rq); |
925 | blk_clear_rq_complete(rq); | |
5f3ea37c | 926 | trace_block_rq_requeue(q, rq); |
2056a782 | 927 | |
1da177e4 LT |
928 | if (blk_rq_tagged(rq)) |
929 | blk_queue_end_tag(q, rq); | |
930 | ||
931 | elv_requeue_request(q, rq); | |
932 | } | |
1da177e4 LT |
933 | EXPORT_SYMBOL(blk_requeue_request); |
934 | ||
935 | /** | |
710027a4 | 936 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
937 | * @q: request queue where request should be inserted |
938 | * @rq: request to be inserted | |
939 | * @at_head: insert request at head or tail of queue | |
940 | * @data: private data | |
1da177e4 LT |
941 | * |
942 | * Description: | |
943 | * Many block devices need to execute commands asynchronously, so they don't | |
944 | * block the whole kernel from preemption during request execution. This is | |
945 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
946 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
947 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
948 | * |
949 | * We have the option of inserting the head or the tail of the queue. | |
950 | * Typically we use the tail for new ioctls and so forth. We use the head | |
951 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
952 | * host that is unable to accept a particular command. | |
953 | */ | |
165125e1 | 954 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 955 | int at_head, void *data) |
1da177e4 | 956 | { |
867d1191 | 957 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
958 | unsigned long flags; |
959 | ||
960 | /* | |
961 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
962 | * must not attempt merges on this) and that it acts as a soft | |
963 | * barrier | |
964 | */ | |
4aff5e23 JA |
965 | rq->cmd_type = REQ_TYPE_SPECIAL; |
966 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
967 | |
968 | rq->special = data; | |
969 | ||
970 | spin_lock_irqsave(q->queue_lock, flags); | |
971 | ||
972 | /* | |
973 | * If command is tagged, release the tag | |
974 | */ | |
867d1191 TH |
975 | if (blk_rq_tagged(rq)) |
976 | blk_queue_end_tag(q, rq); | |
1da177e4 | 977 | |
b238b3d4 | 978 | drive_stat_acct(rq, 1); |
867d1191 | 979 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 980 | blk_start_queueing(q); |
1da177e4 LT |
981 | spin_unlock_irqrestore(q->queue_lock, flags); |
982 | } | |
1da177e4 LT |
983 | EXPORT_SYMBOL(blk_insert_request); |
984 | ||
1da177e4 LT |
985 | /* |
986 | * add-request adds a request to the linked list. | |
987 | * queue lock is held and interrupts disabled, as we muck with the | |
988 | * request queue list. | |
989 | */ | |
6728cb0e | 990 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 991 | { |
b238b3d4 | 992 | drive_stat_acct(req, 1); |
1da177e4 | 993 | |
1da177e4 LT |
994 | /* |
995 | * elevator indicated where it wants this request to be | |
996 | * inserted at elevator_merge time | |
997 | */ | |
998 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
999 | } | |
6728cb0e | 1000 | |
074a7aca TH |
1001 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1002 | unsigned long now) | |
1003 | { | |
1004 | if (now == part->stamp) | |
1005 | return; | |
1006 | ||
1007 | if (part->in_flight) { | |
1008 | __part_stat_add(cpu, part, time_in_queue, | |
1009 | part->in_flight * (now - part->stamp)); | |
1010 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1011 | } | |
1012 | part->stamp = now; | |
1013 | } | |
1014 | ||
1015 | /** | |
496aa8a9 RD |
1016 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1017 | * @cpu: cpu number for stats access | |
1018 | * @part: target partition | |
1da177e4 LT |
1019 | * |
1020 | * The average IO queue length and utilisation statistics are maintained | |
1021 | * by observing the current state of the queue length and the amount of | |
1022 | * time it has been in this state for. | |
1023 | * | |
1024 | * Normally, that accounting is done on IO completion, but that can result | |
1025 | * in more than a second's worth of IO being accounted for within any one | |
1026 | * second, leading to >100% utilisation. To deal with that, we call this | |
1027 | * function to do a round-off before returning the results when reading | |
1028 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1029 | * the current jiffies and restarts the counters again. | |
1030 | */ | |
c9959059 | 1031 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1032 | { |
1033 | unsigned long now = jiffies; | |
1034 | ||
074a7aca TH |
1035 | if (part->partno) |
1036 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1037 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1038 | } |
074a7aca | 1039 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1040 | |
1da177e4 LT |
1041 | /* |
1042 | * queue lock must be held | |
1043 | */ | |
165125e1 | 1044 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1045 | { |
1da177e4 LT |
1046 | if (unlikely(!q)) |
1047 | return; | |
1048 | if (unlikely(--req->ref_count)) | |
1049 | return; | |
1050 | ||
8922e16c TH |
1051 | elv_completed_request(q, req); |
1052 | ||
1da177e4 LT |
1053 | /* |
1054 | * Request may not have originated from ll_rw_blk. if not, | |
1055 | * it didn't come out of our reserved rq pools | |
1056 | */ | |
49171e5c | 1057 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1058 | int rw = rq_data_dir(req); |
4aff5e23 | 1059 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1060 | |
1da177e4 | 1061 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1062 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1063 | |
1064 | blk_free_request(q, req); | |
cb98fc8b | 1065 | freed_request(q, rw, priv); |
1da177e4 LT |
1066 | } |
1067 | } | |
6e39b69e MC |
1068 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1069 | ||
1da177e4 LT |
1070 | void blk_put_request(struct request *req) |
1071 | { | |
8922e16c | 1072 | unsigned long flags; |
165125e1 | 1073 | struct request_queue *q = req->q; |
8922e16c | 1074 | |
52a93ba8 FT |
1075 | spin_lock_irqsave(q->queue_lock, flags); |
1076 | __blk_put_request(q, req); | |
1077 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1078 | } |
1da177e4 LT |
1079 | EXPORT_SYMBOL(blk_put_request); |
1080 | ||
86db1e29 | 1081 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1082 | { |
c7c22e4d | 1083 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1084 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1085 | |
1086 | /* | |
1087 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1088 | */ | |
6000a368 MC |
1089 | if (bio_rw_ahead(bio)) |
1090 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1091 | REQ_FAILFAST_DRIVER); | |
1092 | if (bio_failfast_dev(bio)) | |
1093 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1094 | if (bio_failfast_transport(bio)) | |
1095 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1096 | if (bio_failfast_driver(bio)) | |
1097 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
52d9e675 TH |
1098 | |
1099 | /* | |
1100 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1101 | */ | |
fb2dce86 | 1102 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1103 | req->cmd_flags |= REQ_DISCARD; |
1104 | if (bio_barrier(bio)) | |
1105 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1106 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 DW |
1107 | } else if (unlikely(bio_barrier(bio))) |
1108 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
52d9e675 | 1109 | |
b31dc66a | 1110 | if (bio_sync(bio)) |
4aff5e23 | 1111 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1112 | if (bio_rw_meta(bio)) |
1113 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1114 | |
52d9e675 TH |
1115 | req->errors = 0; |
1116 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1117 | req->ioprio = bio_prio(bio); |
52d9e675 | 1118 | req->start_time = jiffies; |
bc1c56fd | 1119 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1120 | } |
1121 | ||
165125e1 | 1122 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1123 | { |
450991bc | 1124 | struct request *req; |
fb2dce86 | 1125 | int el_ret, nr_sectors, barrier, discard, err; |
51da90fc JA |
1126 | const unsigned short prio = bio_prio(bio); |
1127 | const int sync = bio_sync(bio); | |
7749a8d4 | 1128 | int rw_flags; |
1da177e4 | 1129 | |
1da177e4 | 1130 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1131 | |
1132 | /* | |
1133 | * low level driver can indicate that it wants pages above a | |
1134 | * certain limit bounced to low memory (ie for highmem, or even | |
1135 | * ISA dma in theory) | |
1136 | */ | |
1137 | blk_queue_bounce(q, &bio); | |
1138 | ||
1da177e4 | 1139 | barrier = bio_barrier(bio); |
e17fc0a1 DW |
1140 | if (unlikely(barrier) && bio_has_data(bio) && |
1141 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1da177e4 LT |
1142 | err = -EOPNOTSUPP; |
1143 | goto end_io; | |
1144 | } | |
1145 | ||
fb2dce86 DW |
1146 | discard = bio_discard(bio); |
1147 | if (unlikely(discard) && !q->prepare_discard_fn) { | |
1148 | err = -EOPNOTSUPP; | |
1149 | goto end_io; | |
1150 | } | |
1151 | ||
1da177e4 LT |
1152 | spin_lock_irq(q->queue_lock); |
1153 | ||
450991bc | 1154 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1155 | goto get_rq; |
1156 | ||
1157 | el_ret = elv_merge(q, &req, bio); | |
1158 | switch (el_ret) { | |
6728cb0e JA |
1159 | case ELEVATOR_BACK_MERGE: |
1160 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1161 | |
6728cb0e JA |
1162 | if (!ll_back_merge_fn(q, req, bio)) |
1163 | break; | |
1da177e4 | 1164 | |
5f3ea37c | 1165 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1166 | |
6728cb0e JA |
1167 | req->biotail->bi_next = bio; |
1168 | req->biotail = bio; | |
1169 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1170 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1171 | if (!blk_rq_cpu_valid(req)) |
1172 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1173 | drive_stat_acct(req, 0); |
1174 | if (!attempt_back_merge(q, req)) | |
1175 | elv_merged_request(q, req, el_ret); | |
1176 | goto out; | |
1da177e4 | 1177 | |
6728cb0e JA |
1178 | case ELEVATOR_FRONT_MERGE: |
1179 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1180 | |
6728cb0e JA |
1181 | if (!ll_front_merge_fn(q, req, bio)) |
1182 | break; | |
1da177e4 | 1183 | |
5f3ea37c | 1184 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1185 | |
6728cb0e JA |
1186 | bio->bi_next = req->bio; |
1187 | req->bio = bio; | |
1da177e4 | 1188 | |
6728cb0e JA |
1189 | /* |
1190 | * may not be valid. if the low level driver said | |
1191 | * it didn't need a bounce buffer then it better | |
1192 | * not touch req->buffer either... | |
1193 | */ | |
1194 | req->buffer = bio_data(bio); | |
1195 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1196 | req->hard_cur_sectors = req->current_nr_sectors; | |
1197 | req->sector = req->hard_sector = bio->bi_sector; | |
1198 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1199 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1200 | if (!blk_rq_cpu_valid(req)) |
1201 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1202 | drive_stat_acct(req, 0); |
1203 | if (!attempt_front_merge(q, req)) | |
1204 | elv_merged_request(q, req, el_ret); | |
1205 | goto out; | |
1206 | ||
1207 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1208 | default: | |
1209 | ; | |
1da177e4 LT |
1210 | } |
1211 | ||
450991bc | 1212 | get_rq: |
7749a8d4 JA |
1213 | /* |
1214 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1215 | * but we need to set it earlier to expose the sync flag to the | |
1216 | * rq allocator and io schedulers. | |
1217 | */ | |
1218 | rw_flags = bio_data_dir(bio); | |
1219 | if (sync) | |
1220 | rw_flags |= REQ_RW_SYNC; | |
1221 | ||
1da177e4 | 1222 | /* |
450991bc | 1223 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1224 | * Returns with the queue unlocked. |
450991bc | 1225 | */ |
7749a8d4 | 1226 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1227 | |
450991bc NP |
1228 | /* |
1229 | * After dropping the lock and possibly sleeping here, our request | |
1230 | * may now be mergeable after it had proven unmergeable (above). | |
1231 | * We don't worry about that case for efficiency. It won't happen | |
1232 | * often, and the elevators are able to handle it. | |
1da177e4 | 1233 | */ |
52d9e675 | 1234 | init_request_from_bio(req, bio); |
1da177e4 | 1235 | |
450991bc | 1236 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1237 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1238 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1239 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
450991bc NP |
1240 | if (elv_queue_empty(q)) |
1241 | blk_plug_device(q); | |
1da177e4 LT |
1242 | add_request(q, req); |
1243 | out: | |
4a534f93 | 1244 | if (sync) |
1da177e4 | 1245 | __generic_unplug_device(q); |
1da177e4 LT |
1246 | spin_unlock_irq(q->queue_lock); |
1247 | return 0; | |
1248 | ||
1249 | end_io: | |
6712ecf8 | 1250 | bio_endio(bio, err); |
1da177e4 LT |
1251 | return 0; |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * If bio->bi_dev is a partition, remap the location | |
1256 | */ | |
1257 | static inline void blk_partition_remap(struct bio *bio) | |
1258 | { | |
1259 | struct block_device *bdev = bio->bi_bdev; | |
1260 | ||
bf2de6f5 | 1261 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1262 | struct hd_struct *p = bdev->bd_part; |
1263 | ||
1da177e4 LT |
1264 | bio->bi_sector += p->start_sect; |
1265 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1266 | |
5f3ea37c | 1267 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, |
c7149d6b AB |
1268 | bdev->bd_dev, bio->bi_sector, |
1269 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1270 | } |
1271 | } | |
1272 | ||
1da177e4 LT |
1273 | static void handle_bad_sector(struct bio *bio) |
1274 | { | |
1275 | char b[BDEVNAME_SIZE]; | |
1276 | ||
1277 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1278 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1279 | bdevname(bio->bi_bdev, b), | |
1280 | bio->bi_rw, | |
1281 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1282 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1283 | ||
1284 | set_bit(BIO_EOF, &bio->bi_flags); | |
1285 | } | |
1286 | ||
c17bb495 AM |
1287 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1288 | ||
1289 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1290 | ||
1291 | static int __init setup_fail_make_request(char *str) | |
1292 | { | |
1293 | return setup_fault_attr(&fail_make_request, str); | |
1294 | } | |
1295 | __setup("fail_make_request=", setup_fail_make_request); | |
1296 | ||
1297 | static int should_fail_request(struct bio *bio) | |
1298 | { | |
eddb2e26 TH |
1299 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1300 | ||
1301 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1302 | return should_fail(&fail_make_request, bio->bi_size); |
1303 | ||
1304 | return 0; | |
1305 | } | |
1306 | ||
1307 | static int __init fail_make_request_debugfs(void) | |
1308 | { | |
1309 | return init_fault_attr_dentries(&fail_make_request, | |
1310 | "fail_make_request"); | |
1311 | } | |
1312 | ||
1313 | late_initcall(fail_make_request_debugfs); | |
1314 | ||
1315 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1316 | ||
1317 | static inline int should_fail_request(struct bio *bio) | |
1318 | { | |
1319 | return 0; | |
1320 | } | |
1321 | ||
1322 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1323 | ||
c07e2b41 JA |
1324 | /* |
1325 | * Check whether this bio extends beyond the end of the device. | |
1326 | */ | |
1327 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1328 | { | |
1329 | sector_t maxsector; | |
1330 | ||
1331 | if (!nr_sectors) | |
1332 | return 0; | |
1333 | ||
1334 | /* Test device or partition size, when known. */ | |
1335 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1336 | if (maxsector) { | |
1337 | sector_t sector = bio->bi_sector; | |
1338 | ||
1339 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1340 | /* | |
1341 | * This may well happen - the kernel calls bread() | |
1342 | * without checking the size of the device, e.g., when | |
1343 | * mounting a device. | |
1344 | */ | |
1345 | handle_bad_sector(bio); | |
1346 | return 1; | |
1347 | } | |
1348 | } | |
1349 | ||
1350 | return 0; | |
1351 | } | |
1352 | ||
1da177e4 | 1353 | /** |
710027a4 | 1354 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1355 | * @bio: The bio describing the location in memory and on the device. |
1356 | * | |
1357 | * generic_make_request() is used to make I/O requests of block | |
1358 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1359 | * to be done. | |
1360 | * | |
1361 | * generic_make_request() does not return any status. The | |
1362 | * success/failure status of the request, along with notification of | |
1363 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1364 | * function described (one day) else where. | |
1365 | * | |
1366 | * The caller of generic_make_request must make sure that bi_io_vec | |
1367 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1368 | * set to describe the device address, and the | |
1369 | * bi_end_io and optionally bi_private are set to describe how | |
1370 | * completion notification should be signaled. | |
1371 | * | |
1372 | * generic_make_request and the drivers it calls may use bi_next if this | |
1373 | * bio happens to be merged with someone else, and may change bi_dev and | |
1374 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1375 | * should NOT be depended on after the call to generic_make_request. | |
1376 | */ | |
d89d8796 | 1377 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1378 | { |
165125e1 | 1379 | struct request_queue *q; |
5ddfe969 | 1380 | sector_t old_sector; |
1da177e4 | 1381 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1382 | dev_t old_dev; |
51fd77bd | 1383 | int err = -EIO; |
1da177e4 LT |
1384 | |
1385 | might_sleep(); | |
1da177e4 | 1386 | |
c07e2b41 JA |
1387 | if (bio_check_eod(bio, nr_sectors)) |
1388 | goto end_io; | |
1da177e4 LT |
1389 | |
1390 | /* | |
1391 | * Resolve the mapping until finished. (drivers are | |
1392 | * still free to implement/resolve their own stacking | |
1393 | * by explicitly returning 0) | |
1394 | * | |
1395 | * NOTE: we don't repeat the blk_size check for each new device. | |
1396 | * Stacking drivers are expected to know what they are doing. | |
1397 | */ | |
5ddfe969 | 1398 | old_sector = -1; |
2056a782 | 1399 | old_dev = 0; |
1da177e4 LT |
1400 | do { |
1401 | char b[BDEVNAME_SIZE]; | |
1402 | ||
1403 | q = bdev_get_queue(bio->bi_bdev); | |
1404 | if (!q) { | |
1405 | printk(KERN_ERR | |
1406 | "generic_make_request: Trying to access " | |
1407 | "nonexistent block-device %s (%Lu)\n", | |
1408 | bdevname(bio->bi_bdev, b), | |
1409 | (long long) bio->bi_sector); | |
1410 | end_io: | |
51fd77bd | 1411 | bio_endio(bio, err); |
1da177e4 LT |
1412 | break; |
1413 | } | |
1414 | ||
4fa253f3 | 1415 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1416 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1417 | bdevname(bio->bi_bdev, b), |
1418 | bio_sectors(bio), | |
1419 | q->max_hw_sectors); | |
1420 | goto end_io; | |
1421 | } | |
1422 | ||
fde6ad22 | 1423 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1424 | goto end_io; |
1425 | ||
c17bb495 AM |
1426 | if (should_fail_request(bio)) |
1427 | goto end_io; | |
1428 | ||
1da177e4 LT |
1429 | /* |
1430 | * If this device has partitions, remap block n | |
1431 | * of partition p to block n+start(p) of the disk. | |
1432 | */ | |
1433 | blk_partition_remap(bio); | |
1434 | ||
7ba1ba12 MP |
1435 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1436 | goto end_io; | |
1437 | ||
5ddfe969 | 1438 | if (old_sector != -1) |
5f3ea37c | 1439 | trace_block_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1440 | old_sector); |
2056a782 | 1441 | |
5f3ea37c | 1442 | trace_block_bio_queue(q, bio); |
2056a782 | 1443 | |
5ddfe969 | 1444 | old_sector = bio->bi_sector; |
2056a782 JA |
1445 | old_dev = bio->bi_bdev->bd_dev; |
1446 | ||
c07e2b41 JA |
1447 | if (bio_check_eod(bio, nr_sectors)) |
1448 | goto end_io; | |
fb2dce86 DW |
1449 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || |
1450 | (bio_discard(bio) && !q->prepare_discard_fn)) { | |
51fd77bd JA |
1451 | err = -EOPNOTSUPP; |
1452 | goto end_io; | |
1453 | } | |
5ddfe969 | 1454 | |
1da177e4 LT |
1455 | ret = q->make_request_fn(q, bio); |
1456 | } while (ret); | |
1457 | } | |
1458 | ||
d89d8796 NB |
1459 | /* |
1460 | * We only want one ->make_request_fn to be active at a time, | |
1461 | * else stack usage with stacked devices could be a problem. | |
1462 | * So use current->bio_{list,tail} to keep a list of requests | |
1463 | * submited by a make_request_fn function. | |
1464 | * current->bio_tail is also used as a flag to say if | |
1465 | * generic_make_request is currently active in this task or not. | |
1466 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1467 | * then a make_request is active, and new requests should be added | |
1468 | * at the tail | |
1469 | */ | |
1470 | void generic_make_request(struct bio *bio) | |
1471 | { | |
1472 | if (current->bio_tail) { | |
1473 | /* make_request is active */ | |
1474 | *(current->bio_tail) = bio; | |
1475 | bio->bi_next = NULL; | |
1476 | current->bio_tail = &bio->bi_next; | |
1477 | return; | |
1478 | } | |
1479 | /* following loop may be a bit non-obvious, and so deserves some | |
1480 | * explanation. | |
1481 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1482 | * ensure that) so we have a list with a single bio. | |
1483 | * We pretend that we have just taken it off a longer list, so | |
1484 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1485 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1486 | * added. __generic_make_request may indeed add some more bios | |
1487 | * through a recursive call to generic_make_request. If it | |
1488 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1489 | * from the top. In this case we really did just take the bio | |
1490 | * of the top of the list (no pretending) and so fixup bio_list and | |
1491 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1492 | * | |
1493 | * The loop was structured like this to make only one call to | |
1494 | * __generic_make_request (which is important as it is large and | |
1495 | * inlined) and to keep the structure simple. | |
1496 | */ | |
1497 | BUG_ON(bio->bi_next); | |
1498 | do { | |
1499 | current->bio_list = bio->bi_next; | |
1500 | if (bio->bi_next == NULL) | |
1501 | current->bio_tail = ¤t->bio_list; | |
1502 | else | |
1503 | bio->bi_next = NULL; | |
1504 | __generic_make_request(bio); | |
1505 | bio = current->bio_list; | |
1506 | } while (bio); | |
1507 | current->bio_tail = NULL; /* deactivate */ | |
1508 | } | |
1da177e4 LT |
1509 | EXPORT_SYMBOL(generic_make_request); |
1510 | ||
1511 | /** | |
710027a4 | 1512 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1513 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1514 | * @bio: The &struct bio which describes the I/O | |
1515 | * | |
1516 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1517 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1518 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1519 | * |
1520 | */ | |
1521 | void submit_bio(int rw, struct bio *bio) | |
1522 | { | |
1523 | int count = bio_sectors(bio); | |
1524 | ||
22e2c507 | 1525 | bio->bi_rw |= rw; |
1da177e4 | 1526 | |
bf2de6f5 JA |
1527 | /* |
1528 | * If it's a regular read/write or a barrier with data attached, | |
1529 | * go through the normal accounting stuff before submission. | |
1530 | */ | |
a9c701e5 | 1531 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1532 | if (rw & WRITE) { |
1533 | count_vm_events(PGPGOUT, count); | |
1534 | } else { | |
1535 | task_io_account_read(bio->bi_size); | |
1536 | count_vm_events(PGPGIN, count); | |
1537 | } | |
1538 | ||
1539 | if (unlikely(block_dump)) { | |
1540 | char b[BDEVNAME_SIZE]; | |
1541 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1542 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1543 | (rw & WRITE) ? "WRITE" : "READ", |
1544 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1545 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1546 | } |
1da177e4 LT |
1547 | } |
1548 | ||
1549 | generic_make_request(bio); | |
1550 | } | |
1da177e4 LT |
1551 | EXPORT_SYMBOL(submit_bio); |
1552 | ||
82124d60 KU |
1553 | /** |
1554 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1555 | * @q: the queue | |
1556 | * @rq: the request being checked | |
1557 | * | |
1558 | * Description: | |
1559 | * @rq may have been made based on weaker limitations of upper-level queues | |
1560 | * in request stacking drivers, and it may violate the limitation of @q. | |
1561 | * Since the block layer and the underlying device driver trust @rq | |
1562 | * after it is inserted to @q, it should be checked against @q before | |
1563 | * the insertion using this generic function. | |
1564 | * | |
1565 | * This function should also be useful for request stacking drivers | |
1566 | * in some cases below, so export this fuction. | |
1567 | * Request stacking drivers like request-based dm may change the queue | |
1568 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1569 | * Such request stacking drivers should check those requests agaist | |
1570 | * the new queue limits again when they dispatch those requests, | |
1571 | * although such checkings are also done against the old queue limits | |
1572 | * when submitting requests. | |
1573 | */ | |
1574 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1575 | { | |
1576 | if (rq->nr_sectors > q->max_sectors || | |
1577 | rq->data_len > q->max_hw_sectors << 9) { | |
1578 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1579 | return -EIO; | |
1580 | } | |
1581 | ||
1582 | /* | |
1583 | * queue's settings related to segment counting like q->bounce_pfn | |
1584 | * may differ from that of other stacking queues. | |
1585 | * Recalculate it to check the request correctly on this queue's | |
1586 | * limitation. | |
1587 | */ | |
1588 | blk_recalc_rq_segments(rq); | |
1589 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1590 | rq->nr_phys_segments > q->max_hw_segments) { | |
1591 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1592 | return -EIO; | |
1593 | } | |
1594 | ||
1595 | return 0; | |
1596 | } | |
1597 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1598 | ||
1599 | /** | |
1600 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1601 | * @q: the queue to submit the request | |
1602 | * @rq: the request being queued | |
1603 | */ | |
1604 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1605 | { | |
1606 | unsigned long flags; | |
1607 | ||
1608 | if (blk_rq_check_limits(q, rq)) | |
1609 | return -EIO; | |
1610 | ||
1611 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1612 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1613 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1614 | return -EIO; | |
1615 | #endif | |
1616 | ||
1617 | spin_lock_irqsave(q->queue_lock, flags); | |
1618 | ||
1619 | /* | |
1620 | * Submitting request must be dequeued before calling this function | |
1621 | * because it will be linked to another request_queue | |
1622 | */ | |
1623 | BUG_ON(blk_queued_rq(rq)); | |
1624 | ||
1625 | drive_stat_acct(rq, 1); | |
1626 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1627 | ||
1628 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1629 | ||
1630 | return 0; | |
1631 | } | |
1632 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1633 | ||
3bcddeac KU |
1634 | /** |
1635 | * __end_that_request_first - end I/O on a request | |
1636 | * @req: the request being processed | |
710027a4 | 1637 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
1638 | * @nr_bytes: number of bytes to complete |
1639 | * | |
1640 | * Description: | |
1641 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1642 | * for the next range of segments (if any) in the cluster. | |
1643 | * | |
1644 | * Return: | |
710027a4 RD |
1645 | * %0 - we are done with this request, call end_that_request_last() |
1646 | * %1 - still buffers pending for this request | |
3bcddeac | 1647 | **/ |
5450d3e1 | 1648 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1649 | int nr_bytes) |
1650 | { | |
5450d3e1 | 1651 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1652 | struct bio *bio; |
1653 | ||
5f3ea37c | 1654 | trace_block_rq_complete(req->q, req); |
2056a782 | 1655 | |
1da177e4 | 1656 | /* |
710027a4 | 1657 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual |
1da177e4 LT |
1658 | * sense key with us all the way through |
1659 | */ | |
1660 | if (!blk_pc_request(req)) | |
1661 | req->errors = 0; | |
1662 | ||
6728cb0e JA |
1663 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1664 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1665 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1666 | (unsigned long long)req->sector); | |
1667 | } | |
1668 | ||
d72d904a | 1669 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b | 1670 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1671 | struct hd_struct *part; |
c9959059 | 1672 | int cpu; |
a362357b | 1673 | |
074a7aca | 1674 | cpu = part_stat_lock(); |
e71bf0d0 | 1675 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
074a7aca TH |
1676 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); |
1677 | part_stat_unlock(); | |
d72d904a JA |
1678 | } |
1679 | ||
1da177e4 LT |
1680 | total_bytes = bio_nbytes = 0; |
1681 | while ((bio = req->bio) != NULL) { | |
1682 | int nbytes; | |
1683 | ||
bf2de6f5 JA |
1684 | /* |
1685 | * For an empty barrier request, the low level driver must | |
1686 | * store a potential error location in ->sector. We pass | |
1687 | * that back up in ->bi_sector. | |
1688 | */ | |
1689 | if (blk_empty_barrier(req)) | |
1690 | bio->bi_sector = req->sector; | |
1691 | ||
1da177e4 LT |
1692 | if (nr_bytes >= bio->bi_size) { |
1693 | req->bio = bio->bi_next; | |
1694 | nbytes = bio->bi_size; | |
5bb23a68 | 1695 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1696 | next_idx = 0; |
1697 | bio_nbytes = 0; | |
1698 | } else { | |
1699 | int idx = bio->bi_idx + next_idx; | |
1700 | ||
1701 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1702 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e | 1703 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1704 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
1705 | break; |
1706 | } | |
1707 | ||
1708 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1709 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1710 | ||
1711 | /* | |
1712 | * not a complete bvec done | |
1713 | */ | |
1714 | if (unlikely(nbytes > nr_bytes)) { | |
1715 | bio_nbytes += nr_bytes; | |
1716 | total_bytes += nr_bytes; | |
1717 | break; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * advance to the next vector | |
1722 | */ | |
1723 | next_idx++; | |
1724 | bio_nbytes += nbytes; | |
1725 | } | |
1726 | ||
1727 | total_bytes += nbytes; | |
1728 | nr_bytes -= nbytes; | |
1729 | ||
6728cb0e JA |
1730 | bio = req->bio; |
1731 | if (bio) { | |
1da177e4 LT |
1732 | /* |
1733 | * end more in this run, or just return 'not-done' | |
1734 | */ | |
1735 | if (unlikely(nr_bytes <= 0)) | |
1736 | break; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * completely done | |
1742 | */ | |
1743 | if (!req->bio) | |
1744 | return 0; | |
1745 | ||
1746 | /* | |
1747 | * if the request wasn't completed, update state | |
1748 | */ | |
1749 | if (bio_nbytes) { | |
5bb23a68 | 1750 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1751 | bio->bi_idx += next_idx; |
1752 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1753 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1754 | } | |
1755 | ||
1756 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1757 | blk_recalc_rq_segments(req); | |
1758 | return 1; | |
1759 | } | |
1760 | ||
1da177e4 LT |
1761 | /* |
1762 | * queue lock must be held | |
1763 | */ | |
5450d3e1 | 1764 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1765 | { |
1766 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1767 | |
b8286239 KU |
1768 | if (blk_rq_tagged(req)) |
1769 | blk_queue_end_tag(req->q, req); | |
1770 | ||
1771 | if (blk_queued_rq(req)) | |
1772 | blkdev_dequeue_request(req); | |
1da177e4 LT |
1773 | |
1774 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1775 | laptop_io_completion(); | |
1776 | ||
e78042e5 MA |
1777 | blk_delete_timer(req); |
1778 | ||
fd0ff8aa JA |
1779 | /* |
1780 | * Account IO completion. bar_rq isn't accounted as a normal | |
1781 | * IO on queueing nor completion. Accounting the containing | |
1782 | * request is enough. | |
1783 | */ | |
1784 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 1785 | unsigned long duration = jiffies - req->start_time; |
a362357b | 1786 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1787 | struct hd_struct *part; |
c9959059 | 1788 | int cpu; |
e71bf0d0 | 1789 | |
074a7aca | 1790 | cpu = part_stat_lock(); |
e71bf0d0 | 1791 | part = disk_map_sector_rcu(disk, req->sector); |
a362357b | 1792 | |
074a7aca TH |
1793 | part_stat_inc(cpu, part, ios[rw]); |
1794 | part_stat_add(cpu, part, ticks[rw], duration); | |
1795 | part_round_stats(cpu, part); | |
1796 | part_dec_in_flight(part); | |
e71bf0d0 | 1797 | |
074a7aca | 1798 | part_stat_unlock(); |
1da177e4 | 1799 | } |
b8286239 | 1800 | |
1da177e4 | 1801 | if (req->end_io) |
8ffdc655 | 1802 | req->end_io(req, error); |
b8286239 KU |
1803 | else { |
1804 | if (blk_bidi_rq(req)) | |
1805 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1806 | ||
1da177e4 | 1807 | __blk_put_request(req->q, req); |
b8286239 | 1808 | } |
1da177e4 LT |
1809 | } |
1810 | ||
3b11313a KU |
1811 | /** |
1812 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1813 | * @rq: the request being processed |
3b11313a KU |
1814 | **/ |
1815 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1816 | { |
1817 | if (blk_fs_request(rq)) | |
1818 | return rq->hard_nr_sectors << 9; | |
1819 | ||
1820 | return rq->data_len; | |
1821 | } | |
3b11313a KU |
1822 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1823 | ||
1824 | /** | |
1825 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1826 | * @rq: the request being processed |
3b11313a KU |
1827 | **/ |
1828 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1829 | { | |
1830 | if (blk_fs_request(rq)) | |
1831 | return rq->current_nr_sectors << 9; | |
1832 | ||
1833 | if (rq->bio) | |
1834 | return rq->bio->bi_size; | |
1835 | ||
1836 | return rq->data_len; | |
1837 | } | |
1838 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 | 1839 | |
a0cd1285 JA |
1840 | /** |
1841 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1842 | * @req: the request being processed |
710027a4 | 1843 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1844 | * |
1845 | * Description: | |
1846 | * Ends I/O on the current segment of a request. If that is the only | |
1847 | * remaining segment, the request is also completed and freed. | |
1848 | * | |
710027a4 RD |
1849 | * This is a remnant of how older block drivers handled I/O completions. |
1850 | * Modern drivers typically end I/O on the full request in one go, unless | |
a0cd1285 JA |
1851 | * they have a residual value to account for. For that case this function |
1852 | * isn't really useful, unless the residual just happens to be the | |
1853 | * full current segment. In other words, don't use this function in new | |
d00e29fd | 1854 | * code. Use blk_end_request() or __blk_end_request() to end a request. |
a0cd1285 JA |
1855 | **/ |
1856 | void end_request(struct request *req, int uptodate) | |
1857 | { | |
d00e29fd KU |
1858 | int error = 0; |
1859 | ||
1860 | if (uptodate <= 0) | |
1861 | error = uptodate ? uptodate : -EIO; | |
1862 | ||
1863 | __blk_end_request(req, error, req->hard_cur_sectors << 9); | |
a0cd1285 | 1864 | } |
1da177e4 LT |
1865 | EXPORT_SYMBOL(end_request); |
1866 | ||
32fab448 KU |
1867 | static int end_that_request_data(struct request *rq, int error, |
1868 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1869 | { | |
1870 | if (rq->bio) { | |
1871 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1872 | return 1; | |
1873 | ||
1874 | /* Bidi request must be completed as a whole */ | |
1875 | if (blk_bidi_rq(rq) && | |
1876 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1877 | return 1; | |
1878 | } | |
1879 | ||
1880 | return 0; | |
1881 | } | |
1882 | ||
336cdb40 | 1883 | /** |
e19a3ab0 KU |
1884 | * blk_end_io - Generic end_io function to complete a request. |
1885 | * @rq: the request being processed | |
710027a4 | 1886 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1887 | * @nr_bytes: number of bytes to complete @rq |
1888 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1889 | * @drv_callback: function called between completion of bios in the request |
1890 | * and completion of the request. | |
710027a4 | 1891 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1892 | * completion of the request. |
336cdb40 KU |
1893 | * |
1894 | * Description: | |
e3a04fe3 | 1895 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1896 | * If @rq has leftover, sets it up for the next range of segments. |
1897 | * | |
1898 | * Return: | |
710027a4 RD |
1899 | * %0 - we are done with this request |
1900 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1901 | **/ |
22b13210 JA |
1902 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1903 | unsigned int bidi_bytes, | |
1904 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1905 | { |
1906 | struct request_queue *q = rq->q; | |
1907 | unsigned long flags = 0UL; | |
336cdb40 | 1908 | |
32fab448 KU |
1909 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
1910 | return 1; | |
336cdb40 | 1911 | |
e19a3ab0 KU |
1912 | /* Special feature for tricky drivers */ |
1913 | if (drv_callback && drv_callback(rq)) | |
1914 | return 1; | |
1915 | ||
336cdb40 KU |
1916 | add_disk_randomness(rq->rq_disk); |
1917 | ||
1918 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1919 | end_that_request_last(rq, error); |
336cdb40 KU |
1920 | spin_unlock_irqrestore(q->queue_lock, flags); |
1921 | ||
1922 | return 0; | |
1923 | } | |
e19a3ab0 KU |
1924 | |
1925 | /** | |
1926 | * blk_end_request - Helper function for drivers to complete the request. | |
1927 | * @rq: the request being processed | |
710027a4 | 1928 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
1929 | * @nr_bytes: number of bytes to complete |
1930 | * | |
1931 | * Description: | |
1932 | * Ends I/O on a number of bytes attached to @rq. | |
1933 | * If @rq has leftover, sets it up for the next range of segments. | |
1934 | * | |
1935 | * Return: | |
710027a4 RD |
1936 | * %0 - we are done with this request |
1937 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1938 | **/ |
22b13210 | 1939 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1940 | { |
e3a04fe3 | 1941 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1942 | } |
336cdb40 KU |
1943 | EXPORT_SYMBOL_GPL(blk_end_request); |
1944 | ||
1945 | /** | |
1946 | * __blk_end_request - Helper function for drivers to complete the request. | |
1947 | * @rq: the request being processed | |
710027a4 | 1948 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
1949 | * @nr_bytes: number of bytes to complete |
1950 | * | |
1951 | * Description: | |
1952 | * Must be called with queue lock held unlike blk_end_request(). | |
1953 | * | |
1954 | * Return: | |
710027a4 RD |
1955 | * %0 - we are done with this request |
1956 | * %1 - still buffers pending for this request | |
336cdb40 | 1957 | **/ |
22b13210 | 1958 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1959 | { |
60540161 | 1960 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
051cc395 | 1961 | return 1; |
336cdb40 KU |
1962 | |
1963 | add_disk_randomness(rq->rq_disk); | |
1964 | ||
b8286239 | 1965 | end_that_request_last(rq, error); |
336cdb40 KU |
1966 | |
1967 | return 0; | |
1968 | } | |
1969 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1970 | ||
e3a04fe3 KU |
1971 | /** |
1972 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1973 | * @rq: the bidi request being processed | |
710027a4 | 1974 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1975 | * @nr_bytes: number of bytes to complete @rq |
1976 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1977 | * | |
1978 | * Description: | |
1979 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1980 | * | |
1981 | * Return: | |
710027a4 RD |
1982 | * %0 - we are done with this request |
1983 | * %1 - still buffers pending for this request | |
e3a04fe3 | 1984 | **/ |
22b13210 JA |
1985 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
1986 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
1987 | { |
1988 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
1989 | } | |
1990 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
1991 | ||
32fab448 KU |
1992 | /** |
1993 | * blk_update_request - Special helper function for request stacking drivers | |
1994 | * @rq: the request being processed | |
1995 | * @error: %0 for success, < %0 for error | |
1996 | * @nr_bytes: number of bytes to complete @rq | |
1997 | * | |
1998 | * Description: | |
1999 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2000 | * the request structure even if @rq doesn't have leftover. | |
2001 | * If @rq has leftover, sets it up for the next range of segments. | |
2002 | * | |
2003 | * This special helper function is only for request stacking drivers | |
2004 | * (e.g. request-based dm) so that they can handle partial completion. | |
2005 | * Actual device drivers should use blk_end_request instead. | |
2006 | */ | |
2007 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2008 | { | |
2009 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2010 | /* | |
2011 | * These members are not updated in end_that_request_data() | |
2012 | * when all bios are completed. | |
2013 | * Update them so that the request stacking driver can find | |
2014 | * how many bytes remain in the request later. | |
2015 | */ | |
2016 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2017 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2018 | } | |
2019 | } | |
2020 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2021 | ||
e19a3ab0 KU |
2022 | /** |
2023 | * blk_end_request_callback - Special helper function for tricky drivers | |
2024 | * @rq: the request being processed | |
710027a4 | 2025 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2026 | * @nr_bytes: number of bytes to complete |
2027 | * @drv_callback: function called between completion of bios in the request | |
2028 | * and completion of the request. | |
710027a4 | 2029 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
2030 | * completion of the request. |
2031 | * | |
2032 | * Description: | |
2033 | * Ends I/O on a number of bytes attached to @rq. | |
2034 | * If @rq has leftover, sets it up for the next range of segments. | |
2035 | * | |
2036 | * This special helper function is used only for existing tricky drivers. | |
2037 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2038 | * This interface will be removed when such drivers are rewritten. | |
2039 | * Don't use this interface in other places anymore. | |
2040 | * | |
2041 | * Return: | |
710027a4 RD |
2042 | * %0 - we are done with this request |
2043 | * %1 - this request is not freed yet. | |
2044 | * this request still has pending buffers or | |
2045 | * the driver doesn't want to finish this request yet. | |
e19a3ab0 | 2046 | **/ |
22b13210 JA |
2047 | int blk_end_request_callback(struct request *rq, int error, |
2048 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2049 | int (drv_callback)(struct request *)) |
2050 | { | |
e3a04fe3 | 2051 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2052 | } |
2053 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2054 | ||
86db1e29 JA |
2055 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2056 | struct bio *bio) | |
1da177e4 | 2057 | { |
d628eaef DW |
2058 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2059 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2060 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2061 | |
fb2dce86 DW |
2062 | if (bio_has_data(bio)) { |
2063 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2064 | rq->buffer = bio_data(bio); |
2065 | } | |
1da177e4 LT |
2066 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2067 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2068 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2069 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2070 | |
2071 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2072 | |
66846572 N |
2073 | if (bio->bi_bdev) |
2074 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2075 | } | |
1da177e4 | 2076 | |
ef9e3fac KU |
2077 | /** |
2078 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2079 | * @q : the queue of the device being checked | |
2080 | * | |
2081 | * Description: | |
2082 | * Check if underlying low-level drivers of a device are busy. | |
2083 | * If the drivers want to export their busy state, they must set own | |
2084 | * exporting function using blk_queue_lld_busy() first. | |
2085 | * | |
2086 | * Basically, this function is used only by request stacking drivers | |
2087 | * to stop dispatching requests to underlying devices when underlying | |
2088 | * devices are busy. This behavior helps more I/O merging on the queue | |
2089 | * of the request stacking driver and prevents I/O throughput regression | |
2090 | * on burst I/O load. | |
2091 | * | |
2092 | * Return: | |
2093 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2094 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2095 | */ | |
2096 | int blk_lld_busy(struct request_queue *q) | |
2097 | { | |
2098 | if (q->lld_busy_fn) | |
2099 | return q->lld_busy_fn(q); | |
2100 | ||
2101 | return 0; | |
2102 | } | |
2103 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2104 | ||
18887ad9 | 2105 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2106 | { |
2107 | return queue_work(kblockd_workqueue, work); | |
2108 | } | |
1da177e4 LT |
2109 | EXPORT_SYMBOL(kblockd_schedule_work); |
2110 | ||
19a75d83 | 2111 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 2112 | { |
28e53bdd | 2113 | cancel_work_sync(work); |
1da177e4 | 2114 | } |
19a75d83 | 2115 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
2116 | |
2117 | int __init blk_dev_init(void) | |
2118 | { | |
2119 | kblockd_workqueue = create_workqueue("kblockd"); | |
2120 | if (!kblockd_workqueue) | |
2121 | panic("Failed to create kblockd\n"); | |
2122 | ||
2123 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2124 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2125 | |
8324aa91 | 2126 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2127 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2128 | |
d38ecf93 | 2129 | return 0; |
1da177e4 | 2130 | } |
1da177e4 | 2131 |