]>
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> |
1da177e4 | 31 | |
8324aa91 JA |
32 | #include "blk.h" |
33 | ||
165125e1 | 34 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
35 | |
36 | /* | |
37 | * For the allocated request tables | |
38 | */ | |
5ece6c52 | 39 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
40 | |
41 | /* | |
42 | * For queue allocation | |
43 | */ | |
6728cb0e | 44 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 45 | |
1da177e4 LT |
46 | /* |
47 | * Controlling structure to kblockd | |
48 | */ | |
ff856bad | 49 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 50 | |
26b8256e JA |
51 | static void drive_stat_acct(struct request *rq, int new_io) |
52 | { | |
28f13702 | 53 | struct hd_struct *part; |
26b8256e | 54 | int rw = rq_data_dir(rq); |
c9959059 | 55 | int cpu; |
26b8256e JA |
56 | |
57 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
58 | return; | |
59 | ||
074a7aca | 60 | cpu = part_stat_lock(); |
e71bf0d0 | 61 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
c9959059 | 62 | |
28f13702 | 63 | if (!new_io) |
074a7aca | 64 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 65 | else { |
074a7aca TH |
66 | part_round_stats(cpu, part); |
67 | part_inc_in_flight(part); | |
26b8256e | 68 | } |
e71bf0d0 | 69 | |
074a7aca | 70 | part_stat_unlock(); |
26b8256e JA |
71 | } |
72 | ||
8324aa91 | 73 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
74 | { |
75 | int nr; | |
76 | ||
77 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
78 | if (nr > q->nr_requests) | |
79 | nr = q->nr_requests; | |
80 | q->nr_congestion_on = nr; | |
81 | ||
82 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
83 | if (nr < 1) | |
84 | nr = 1; | |
85 | q->nr_congestion_off = nr; | |
86 | } | |
87 | ||
1da177e4 LT |
88 | /** |
89 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
90 | * @bdev: device | |
91 | * | |
92 | * Locates the passed device's request queue and returns the address of its | |
93 | * backing_dev_info | |
94 | * | |
95 | * Will return NULL if the request queue cannot be located. | |
96 | */ | |
97 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
98 | { | |
99 | struct backing_dev_info *ret = NULL; | |
165125e1 | 100 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
101 | |
102 | if (q) | |
103 | ret = &q->backing_dev_info; | |
104 | return ret; | |
105 | } | |
1da177e4 LT |
106 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
107 | ||
2a4aa30c | 108 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 109 | { |
1afb20f3 FT |
110 | memset(rq, 0, sizeof(*rq)); |
111 | ||
1da177e4 | 112 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 113 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 114 | rq->cpu = -1; |
63a71386 JA |
115 | rq->q = q; |
116 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
117 | INIT_HLIST_NODE(&rq->hash); |
118 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 119 | rq->cmd = rq->__cmd; |
63a71386 | 120 | rq->tag = -1; |
1da177e4 | 121 | rq->ref_count = 1; |
1da177e4 | 122 | } |
2a4aa30c | 123 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 124 | |
5bb23a68 N |
125 | static void req_bio_endio(struct request *rq, struct bio *bio, |
126 | unsigned int nbytes, int error) | |
1da177e4 | 127 | { |
165125e1 | 128 | struct request_queue *q = rq->q; |
797e7dbb | 129 | |
5bb23a68 N |
130 | if (&q->bar_rq != rq) { |
131 | if (error) | |
132 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
133 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
134 | error = -EIO; | |
797e7dbb | 135 | |
5bb23a68 | 136 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 137 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 138 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
139 | nbytes = bio->bi_size; |
140 | } | |
797e7dbb | 141 | |
5bb23a68 N |
142 | bio->bi_size -= nbytes; |
143 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
144 | |
145 | if (bio_integrity(bio)) | |
146 | bio_integrity_advance(bio, nbytes); | |
147 | ||
5bb23a68 | 148 | if (bio->bi_size == 0) |
6712ecf8 | 149 | bio_endio(bio, error); |
5bb23a68 N |
150 | } else { |
151 | ||
152 | /* | |
153 | * Okay, this is the barrier request in progress, just | |
154 | * record the error; | |
155 | */ | |
156 | if (error && !q->orderr) | |
157 | q->orderr = error; | |
158 | } | |
1da177e4 | 159 | } |
1da177e4 | 160 | |
1da177e4 LT |
161 | void blk_dump_rq_flags(struct request *rq, char *msg) |
162 | { | |
163 | int bit; | |
164 | ||
6728cb0e | 165 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
166 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
167 | rq->cmd_flags); | |
1da177e4 | 168 | |
6728cb0e JA |
169 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
170 | (unsigned long long)rq->sector, | |
171 | rq->nr_sectors, | |
172 | rq->current_nr_sectors); | |
173 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
174 | rq->bio, rq->biotail, | |
175 | rq->buffer, rq->data, | |
176 | rq->data_len); | |
1da177e4 | 177 | |
4aff5e23 | 178 | if (blk_pc_request(rq)) { |
6728cb0e | 179 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 180 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
181 | printk("%02x ", rq->cmd[bit]); |
182 | printk("\n"); | |
183 | } | |
184 | } | |
1da177e4 LT |
185 | EXPORT_SYMBOL(blk_dump_rq_flags); |
186 | ||
1da177e4 LT |
187 | /* |
188 | * "plug" the device if there are no outstanding requests: this will | |
189 | * force the transfer to start only after we have put all the requests | |
190 | * on the list. | |
191 | * | |
192 | * This is called with interrupts off and no requests on the queue and | |
193 | * with the queue lock held. | |
194 | */ | |
165125e1 | 195 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
196 | { |
197 | WARN_ON(!irqs_disabled()); | |
198 | ||
199 | /* | |
200 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
201 | * which will restart the queueing | |
202 | */ | |
7daac490 | 203 | if (blk_queue_stopped(q)) |
1da177e4 LT |
204 | return; |
205 | ||
e48ec690 | 206 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 207 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
208 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
209 | } | |
1da177e4 | 210 | } |
1da177e4 LT |
211 | EXPORT_SYMBOL(blk_plug_device); |
212 | ||
6c5e0c4d JA |
213 | /** |
214 | * blk_plug_device_unlocked - plug a device without queue lock held | |
215 | * @q: The &struct request_queue to plug | |
216 | * | |
217 | * Description: | |
218 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
219 | * interrupts. | |
220 | **/ | |
221 | void blk_plug_device_unlocked(struct request_queue *q) | |
222 | { | |
223 | unsigned long flags; | |
224 | ||
225 | spin_lock_irqsave(q->queue_lock, flags); | |
226 | blk_plug_device(q); | |
227 | spin_unlock_irqrestore(q->queue_lock, flags); | |
228 | } | |
229 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
230 | ||
1da177e4 LT |
231 | /* |
232 | * remove the queue from the plugged list, if present. called with | |
233 | * queue lock held and interrupts disabled. | |
234 | */ | |
165125e1 | 235 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
236 | { |
237 | WARN_ON(!irqs_disabled()); | |
238 | ||
e48ec690 | 239 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
240 | return 0; |
241 | ||
242 | del_timer(&q->unplug_timer); | |
243 | return 1; | |
244 | } | |
1da177e4 LT |
245 | EXPORT_SYMBOL(blk_remove_plug); |
246 | ||
247 | /* | |
248 | * remove the plug and let it rip.. | |
249 | */ | |
165125e1 | 250 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 251 | { |
7daac490 | 252 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
253 | return; |
254 | ||
255 | if (!blk_remove_plug(q)) | |
256 | return; | |
257 | ||
22e2c507 | 258 | q->request_fn(q); |
1da177e4 LT |
259 | } |
260 | EXPORT_SYMBOL(__generic_unplug_device); | |
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 | |
2056a782 JA |
296 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
297 | q->rq.count[READ] + q->rq.count[WRITE]); | |
298 | ||
1da177e4 LT |
299 | q->unplug_fn(q); |
300 | } | |
301 | ||
86db1e29 | 302 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 303 | { |
165125e1 | 304 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 305 | |
2056a782 JA |
306 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
307 | q->rq.count[READ] + q->rq.count[WRITE]); | |
308 | ||
18887ad9 | 309 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
310 | } |
311 | ||
2ad8b1ef AB |
312 | void blk_unplug(struct request_queue *q) |
313 | { | |
314 | /* | |
315 | * devices don't necessarily have an ->unplug_fn defined | |
316 | */ | |
317 | if (q->unplug_fn) { | |
318 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
319 | q->rq.count[READ] + q->rq.count[WRITE]); | |
320 | ||
321 | q->unplug_fn(q); | |
322 | } | |
323 | } | |
324 | EXPORT_SYMBOL(blk_unplug); | |
325 | ||
c7c22e4d JA |
326 | static void blk_invoke_request_fn(struct request_queue *q) |
327 | { | |
328 | /* | |
329 | * one level of recursion is ok and is much faster than kicking | |
330 | * the unplug handling | |
331 | */ | |
332 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
333 | q->request_fn(q); | |
334 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
335 | } else { | |
336 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
337 | kblockd_schedule_work(q, &q->unplug_work); | |
338 | } | |
339 | } | |
340 | ||
1da177e4 LT |
341 | /** |
342 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 343 | * @q: The &struct request_queue in question |
1da177e4 LT |
344 | * |
345 | * Description: | |
346 | * blk_start_queue() will clear the stop flag on the queue, and call | |
347 | * the request_fn for the queue if it was in a stopped state when | |
348 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
349 | **/ | |
165125e1 | 350 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 351 | { |
a038e253 PBG |
352 | WARN_ON(!irqs_disabled()); |
353 | ||
75ad23bc | 354 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
c7c22e4d | 355 | blk_invoke_request_fn(q); |
1da177e4 | 356 | } |
1da177e4 LT |
357 | EXPORT_SYMBOL(blk_start_queue); |
358 | ||
359 | /** | |
360 | * blk_stop_queue - stop a queue | |
165125e1 | 361 | * @q: The &struct request_queue in question |
1da177e4 LT |
362 | * |
363 | * Description: | |
364 | * The Linux block layer assumes that a block driver will consume all | |
365 | * entries on the request queue when the request_fn strategy is called. | |
366 | * Often this will not happen, because of hardware limitations (queue | |
367 | * depth settings). If a device driver gets a 'queue full' response, | |
368 | * or if it simply chooses not to queue more I/O at one point, it can | |
369 | * call this function to prevent the request_fn from being called until | |
370 | * the driver has signalled it's ready to go again. This happens by calling | |
371 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
372 | **/ | |
165125e1 | 373 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
374 | { |
375 | blk_remove_plug(q); | |
75ad23bc | 376 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
377 | } |
378 | EXPORT_SYMBOL(blk_stop_queue); | |
379 | ||
380 | /** | |
381 | * blk_sync_queue - cancel any pending callbacks on a queue | |
382 | * @q: the queue | |
383 | * | |
384 | * Description: | |
385 | * The block layer may perform asynchronous callback activity | |
386 | * on a queue, such as calling the unplug function after a timeout. | |
387 | * A block device may call blk_sync_queue to ensure that any | |
388 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 389 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
390 | * that its ->make_request_fn will not re-add plugging prior to calling |
391 | * this function. | |
392 | * | |
393 | */ | |
394 | void blk_sync_queue(struct request_queue *q) | |
395 | { | |
396 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 397 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
398 | } |
399 | EXPORT_SYMBOL(blk_sync_queue); | |
400 | ||
401 | /** | |
402 | * blk_run_queue - run a single device queue | |
403 | * @q: The queue to run | |
404 | */ | |
75ad23bc | 405 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 406 | { |
1da177e4 | 407 | blk_remove_plug(q); |
dac07ec1 JA |
408 | |
409 | /* | |
410 | * Only recurse once to avoid overrunning the stack, let the unplug | |
411 | * handling reinvoke the handler shortly if we already got there. | |
412 | */ | |
c7c22e4d JA |
413 | if (!elv_queue_empty(q)) |
414 | blk_invoke_request_fn(q); | |
75ad23bc NP |
415 | } |
416 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 417 | |
75ad23bc NP |
418 | /** |
419 | * blk_run_queue - run a single device queue | |
420 | * @q: The queue to run | |
421 | */ | |
422 | void blk_run_queue(struct request_queue *q) | |
423 | { | |
424 | unsigned long flags; | |
425 | ||
426 | spin_lock_irqsave(q->queue_lock, flags); | |
427 | __blk_run_queue(q); | |
1da177e4 LT |
428 | spin_unlock_irqrestore(q->queue_lock, flags); |
429 | } | |
430 | EXPORT_SYMBOL(blk_run_queue); | |
431 | ||
165125e1 | 432 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
433 | { |
434 | kobject_put(&q->kobj); | |
435 | } | |
483f4afc | 436 | |
6728cb0e | 437 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 438 | { |
e3335de9 JA |
439 | /* |
440 | * We know we have process context here, so we can be a little | |
441 | * cautious and ensure that pending block actions on this device | |
442 | * are done before moving on. Going into this function, we should | |
443 | * not have processes doing IO to this device. | |
444 | */ | |
445 | blk_sync_queue(q); | |
446 | ||
483f4afc | 447 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 448 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
449 | mutex_unlock(&q->sysfs_lock); |
450 | ||
451 | if (q->elevator) | |
452 | elevator_exit(q->elevator); | |
453 | ||
454 | blk_put_queue(q); | |
455 | } | |
1da177e4 LT |
456 | EXPORT_SYMBOL(blk_cleanup_queue); |
457 | ||
165125e1 | 458 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
459 | { |
460 | struct request_list *rl = &q->rq; | |
461 | ||
462 | rl->count[READ] = rl->count[WRITE] = 0; | |
463 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 464 | rl->elvpriv = 0; |
1da177e4 LT |
465 | init_waitqueue_head(&rl->wait[READ]); |
466 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 467 | |
1946089a CL |
468 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
469 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
470 | |
471 | if (!rl->rq_pool) | |
472 | return -ENOMEM; | |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
165125e1 | 477 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 478 | { |
1946089a CL |
479 | return blk_alloc_queue_node(gfp_mask, -1); |
480 | } | |
481 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 482 | |
165125e1 | 483 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 484 | { |
165125e1 | 485 | struct request_queue *q; |
e0bf68dd | 486 | int err; |
1946089a | 487 | |
8324aa91 | 488 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 489 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
490 | if (!q) |
491 | return NULL; | |
492 | ||
e0bf68dd PZ |
493 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
494 | q->backing_dev_info.unplug_io_data = q; | |
495 | err = bdi_init(&q->backing_dev_info); | |
496 | if (err) { | |
8324aa91 | 497 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
498 | return NULL; |
499 | } | |
500 | ||
1da177e4 | 501 | init_timer(&q->unplug_timer); |
242f9dcb JA |
502 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
503 | INIT_LIST_HEAD(&q->timeout_list); | |
483f4afc | 504 | |
8324aa91 | 505 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 506 | |
483f4afc | 507 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 508 | spin_lock_init(&q->__queue_lock); |
483f4afc | 509 | |
1da177e4 LT |
510 | return q; |
511 | } | |
1946089a | 512 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
513 | |
514 | /** | |
515 | * blk_init_queue - prepare a request queue for use with a block device | |
516 | * @rfn: The function to be called to process requests that have been | |
517 | * placed on the queue. | |
518 | * @lock: Request queue spin lock | |
519 | * | |
520 | * Description: | |
521 | * If a block device wishes to use the standard request handling procedures, | |
522 | * which sorts requests and coalesces adjacent requests, then it must | |
523 | * call blk_init_queue(). The function @rfn will be called when there | |
524 | * are requests on the queue that need to be processed. If the device | |
525 | * supports plugging, then @rfn may not be called immediately when requests | |
526 | * are available on the queue, but may be called at some time later instead. | |
527 | * Plugged queues are generally unplugged when a buffer belonging to one | |
528 | * of the requests on the queue is needed, or due to memory pressure. | |
529 | * | |
530 | * @rfn is not required, or even expected, to remove all requests off the | |
531 | * queue, but only as many as it can handle at a time. If it does leave | |
532 | * requests on the queue, it is responsible for arranging that the requests | |
533 | * get dealt with eventually. | |
534 | * | |
535 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
536 | * request queue; this lock will be taken also from interrupt context, so irq |
537 | * disabling is needed for it. | |
1da177e4 | 538 | * |
710027a4 | 539 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
540 | * it didn't succeed. |
541 | * | |
542 | * Note: | |
543 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
544 | * when the block device is deactivated (such as at module unload). | |
545 | **/ | |
1946089a | 546 | |
165125e1 | 547 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 548 | { |
1946089a CL |
549 | return blk_init_queue_node(rfn, lock, -1); |
550 | } | |
551 | EXPORT_SYMBOL(blk_init_queue); | |
552 | ||
165125e1 | 553 | struct request_queue * |
1946089a CL |
554 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
555 | { | |
165125e1 | 556 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
557 | |
558 | if (!q) | |
559 | return NULL; | |
560 | ||
1946089a | 561 | q->node = node_id; |
8669aafd | 562 | if (blk_init_free_list(q)) { |
8324aa91 | 563 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
564 | return NULL; |
565 | } | |
1da177e4 | 566 | |
152587de JA |
567 | /* |
568 | * if caller didn't supply a lock, they get per-queue locking with | |
569 | * our embedded lock | |
570 | */ | |
e7e72bf6 | 571 | if (!lock) |
152587de | 572 | lock = &q->__queue_lock; |
152587de | 573 | |
1da177e4 | 574 | q->request_fn = rfn; |
1da177e4 LT |
575 | q->prep_rq_fn = NULL; |
576 | q->unplug_fn = generic_unplug_device; | |
4ee5eaf4 KU |
577 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | |
578 | 1 << QUEUE_FLAG_STACKABLE); | |
1da177e4 LT |
579 | q->queue_lock = lock; |
580 | ||
581 | blk_queue_segment_boundary(q, 0xffffffff); | |
582 | ||
583 | blk_queue_make_request(q, __make_request); | |
584 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
585 | ||
586 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
587 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
588 | ||
44ec9542 AS |
589 | q->sg_reserved_size = INT_MAX; |
590 | ||
abf54393 FT |
591 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
592 | ||
1da177e4 LT |
593 | /* |
594 | * all done | |
595 | */ | |
596 | if (!elevator_init(q, NULL)) { | |
597 | blk_queue_congestion_threshold(q); | |
598 | return q; | |
599 | } | |
600 | ||
8669aafd | 601 | blk_put_queue(q); |
1da177e4 LT |
602 | return NULL; |
603 | } | |
1946089a | 604 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 605 | |
165125e1 | 606 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 607 | { |
fde6ad22 | 608 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 609 | kobject_get(&q->kobj); |
1da177e4 LT |
610 | return 0; |
611 | } | |
612 | ||
613 | return 1; | |
614 | } | |
1da177e4 | 615 | |
165125e1 | 616 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 617 | { |
4aff5e23 | 618 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 619 | elv_put_request(q, rq); |
1da177e4 LT |
620 | mempool_free(rq, q->rq.rq_pool); |
621 | } | |
622 | ||
1ea25ecb | 623 | static struct request * |
165125e1 | 624 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
625 | { |
626 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
627 | ||
628 | if (!rq) | |
629 | return NULL; | |
630 | ||
2a4aa30c | 631 | blk_rq_init(q, rq); |
1afb20f3 | 632 | |
49171e5c | 633 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 634 | |
cb98fc8b | 635 | if (priv) { |
cb78b285 | 636 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
637 | mempool_free(rq, q->rq.rq_pool); |
638 | return NULL; | |
639 | } | |
4aff5e23 | 640 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 641 | } |
1da177e4 | 642 | |
cb98fc8b | 643 | return rq; |
1da177e4 LT |
644 | } |
645 | ||
646 | /* | |
647 | * ioc_batching returns true if the ioc is a valid batching request and | |
648 | * should be given priority access to a request. | |
649 | */ | |
165125e1 | 650 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
651 | { |
652 | if (!ioc) | |
653 | return 0; | |
654 | ||
655 | /* | |
656 | * Make sure the process is able to allocate at least 1 request | |
657 | * even if the batch times out, otherwise we could theoretically | |
658 | * lose wakeups. | |
659 | */ | |
660 | return ioc->nr_batch_requests == q->nr_batching || | |
661 | (ioc->nr_batch_requests > 0 | |
662 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
663 | } | |
664 | ||
665 | /* | |
666 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
667 | * will cause the process to be a "batcher" on all queues in the system. This | |
668 | * is the behaviour we want though - once it gets a wakeup it should be given | |
669 | * a nice run. | |
670 | */ | |
165125e1 | 671 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
672 | { |
673 | if (!ioc || ioc_batching(q, ioc)) | |
674 | return; | |
675 | ||
676 | ioc->nr_batch_requests = q->nr_batching; | |
677 | ioc->last_waited = jiffies; | |
678 | } | |
679 | ||
165125e1 | 680 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
681 | { |
682 | struct request_list *rl = &q->rq; | |
683 | ||
684 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 685 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
686 | |
687 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
688 | if (waitqueue_active(&rl->wait[rw])) |
689 | wake_up(&rl->wait[rw]); | |
690 | ||
691 | blk_clear_queue_full(q, rw); | |
692 | } | |
693 | } | |
694 | ||
695 | /* | |
696 | * A request has just been released. Account for it, update the full and | |
697 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
698 | */ | |
165125e1 | 699 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
700 | { |
701 | struct request_list *rl = &q->rq; | |
702 | ||
703 | rl->count[rw]--; | |
cb98fc8b TH |
704 | if (priv) |
705 | rl->elvpriv--; | |
1da177e4 LT |
706 | |
707 | __freed_request(q, rw); | |
708 | ||
709 | if (unlikely(rl->starved[rw ^ 1])) | |
710 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
711 | } |
712 | ||
713 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
714 | /* | |
d6344532 NP |
715 | * Get a free request, queue_lock must be held. |
716 | * Returns NULL on failure, with queue_lock held. | |
717 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 718 | */ |
165125e1 | 719 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 720 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
721 | { |
722 | struct request *rq = NULL; | |
723 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 724 | struct io_context *ioc = NULL; |
7749a8d4 | 725 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
726 | int may_queue, priv; |
727 | ||
7749a8d4 | 728 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
729 | if (may_queue == ELV_MQUEUE_NO) |
730 | goto rq_starved; | |
731 | ||
732 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
733 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 734 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
735 | /* |
736 | * The queue will fill after this allocation, so set | |
737 | * it as full, and mark this process as "batching". | |
738 | * This process will be allowed to complete a batch of | |
739 | * requests, others will be blocked. | |
740 | */ | |
741 | if (!blk_queue_full(q, rw)) { | |
742 | ioc_set_batching(q, ioc); | |
743 | blk_set_queue_full(q, rw); | |
744 | } else { | |
745 | if (may_queue != ELV_MQUEUE_MUST | |
746 | && !ioc_batching(q, ioc)) { | |
747 | /* | |
748 | * The queue is full and the allocating | |
749 | * process is not a "batcher", and not | |
750 | * exempted by the IO scheduler | |
751 | */ | |
752 | goto out; | |
753 | } | |
754 | } | |
1da177e4 | 755 | } |
79e2de4b | 756 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
757 | } |
758 | ||
082cf69e JA |
759 | /* |
760 | * Only allow batching queuers to allocate up to 50% over the defined | |
761 | * limit of requests, otherwise we could have thousands of requests | |
762 | * allocated with any setting of ->nr_requests | |
763 | */ | |
fd782a4a | 764 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 765 | goto out; |
fd782a4a | 766 | |
1da177e4 LT |
767 | rl->count[rw]++; |
768 | rl->starved[rw] = 0; | |
cb98fc8b | 769 | |
64521d1a | 770 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
771 | if (priv) |
772 | rl->elvpriv++; | |
773 | ||
1da177e4 LT |
774 | spin_unlock_irq(q->queue_lock); |
775 | ||
7749a8d4 | 776 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 777 | if (unlikely(!rq)) { |
1da177e4 LT |
778 | /* |
779 | * Allocation failed presumably due to memory. Undo anything | |
780 | * we might have messed up. | |
781 | * | |
782 | * Allocating task should really be put onto the front of the | |
783 | * wait queue, but this is pretty rare. | |
784 | */ | |
785 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 786 | freed_request(q, rw, priv); |
1da177e4 LT |
787 | |
788 | /* | |
789 | * in the very unlikely event that allocation failed and no | |
790 | * requests for this direction was pending, mark us starved | |
791 | * so that freeing of a request in the other direction will | |
792 | * notice us. another possible fix would be to split the | |
793 | * rq mempool into READ and WRITE | |
794 | */ | |
795 | rq_starved: | |
796 | if (unlikely(rl->count[rw] == 0)) | |
797 | rl->starved[rw] = 1; | |
798 | ||
1da177e4 LT |
799 | goto out; |
800 | } | |
801 | ||
88ee5ef1 JA |
802 | /* |
803 | * ioc may be NULL here, and ioc_batching will be false. That's | |
804 | * OK, if the queue is under the request limit then requests need | |
805 | * not count toward the nr_batch_requests limit. There will always | |
806 | * be some limit enforced by BLK_BATCH_TIME. | |
807 | */ | |
1da177e4 LT |
808 | if (ioc_batching(q, ioc)) |
809 | ioc->nr_batch_requests--; | |
6728cb0e | 810 | |
2056a782 | 811 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); |
1da177e4 | 812 | out: |
1da177e4 LT |
813 | return rq; |
814 | } | |
815 | ||
816 | /* | |
817 | * No available requests for this queue, unplug the device and wait for some | |
818 | * requests to become available. | |
d6344532 NP |
819 | * |
820 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 821 | */ |
165125e1 | 822 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 823 | struct bio *bio) |
1da177e4 | 824 | { |
7749a8d4 | 825 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
826 | struct request *rq; |
827 | ||
7749a8d4 | 828 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
829 | while (!rq) { |
830 | DEFINE_WAIT(wait); | |
05caf8db | 831 | struct io_context *ioc; |
1da177e4 LT |
832 | struct request_list *rl = &q->rq; |
833 | ||
834 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
835 | TASK_UNINTERRUPTIBLE); | |
836 | ||
05caf8db | 837 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
1da177e4 | 838 | |
05caf8db ZY |
839 | __generic_unplug_device(q); |
840 | spin_unlock_irq(q->queue_lock); | |
841 | io_schedule(); | |
1da177e4 | 842 | |
05caf8db ZY |
843 | /* |
844 | * After sleeping, we become a "batching" process and | |
845 | * will be able to allocate at least one request, and | |
846 | * up to a big batch of them for a small period time. | |
847 | * See ioc_batching, ioc_set_batching | |
848 | */ | |
849 | ioc = current_io_context(GFP_NOIO, q->node); | |
850 | ioc_set_batching(q, ioc); | |
d6344532 | 851 | |
05caf8db | 852 | spin_lock_irq(q->queue_lock); |
1da177e4 | 853 | finish_wait(&rl->wait[rw], &wait); |
05caf8db ZY |
854 | |
855 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
856 | }; | |
1da177e4 LT |
857 | |
858 | return rq; | |
859 | } | |
860 | ||
165125e1 | 861 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
862 | { |
863 | struct request *rq; | |
864 | ||
865 | BUG_ON(rw != READ && rw != WRITE); | |
866 | ||
d6344532 NP |
867 | spin_lock_irq(q->queue_lock); |
868 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 869 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 870 | } else { |
22e2c507 | 871 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
872 | if (!rq) |
873 | spin_unlock_irq(q->queue_lock); | |
874 | } | |
875 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
876 | |
877 | return rq; | |
878 | } | |
1da177e4 LT |
879 | EXPORT_SYMBOL(blk_get_request); |
880 | ||
dc72ef4a JA |
881 | /** |
882 | * blk_start_queueing - initiate dispatch of requests to device | |
883 | * @q: request queue to kick into gear | |
884 | * | |
885 | * This is basically a helper to remove the need to know whether a queue | |
886 | * is plugged or not if someone just wants to initiate dispatch of requests | |
887 | * for this queue. | |
888 | * | |
889 | * The queue lock must be held with interrupts disabled. | |
890 | */ | |
165125e1 | 891 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a | 892 | { |
336c3d8c EO |
893 | if (!blk_queue_plugged(q)) { |
894 | if (unlikely(blk_queue_stopped(q))) | |
895 | return; | |
dc72ef4a | 896 | q->request_fn(q); |
336c3d8c | 897 | } else |
dc72ef4a JA |
898 | __generic_unplug_device(q); |
899 | } | |
900 | EXPORT_SYMBOL(blk_start_queueing); | |
901 | ||
1da177e4 LT |
902 | /** |
903 | * blk_requeue_request - put a request back on queue | |
904 | * @q: request queue where request should be inserted | |
905 | * @rq: request to be inserted | |
906 | * | |
907 | * Description: | |
908 | * Drivers often keep queueing requests until the hardware cannot accept | |
909 | * more, when that condition happens we need to put the request back | |
910 | * on the queue. Must be called with queue lock held. | |
911 | */ | |
165125e1 | 912 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 913 | { |
242f9dcb JA |
914 | blk_delete_timer(rq); |
915 | blk_clear_rq_complete(rq); | |
2056a782 JA |
916 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
917 | ||
1da177e4 LT |
918 | if (blk_rq_tagged(rq)) |
919 | blk_queue_end_tag(q, rq); | |
920 | ||
921 | elv_requeue_request(q, rq); | |
922 | } | |
1da177e4 LT |
923 | EXPORT_SYMBOL(blk_requeue_request); |
924 | ||
925 | /** | |
710027a4 | 926 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
927 | * @q: request queue where request should be inserted |
928 | * @rq: request to be inserted | |
929 | * @at_head: insert request at head or tail of queue | |
930 | * @data: private data | |
1da177e4 LT |
931 | * |
932 | * Description: | |
933 | * Many block devices need to execute commands asynchronously, so they don't | |
934 | * block the whole kernel from preemption during request execution. This is | |
935 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
936 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
937 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
938 | * |
939 | * We have the option of inserting the head or the tail of the queue. | |
940 | * Typically we use the tail for new ioctls and so forth. We use the head | |
941 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
942 | * host that is unable to accept a particular command. | |
943 | */ | |
165125e1 | 944 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 945 | int at_head, void *data) |
1da177e4 | 946 | { |
867d1191 | 947 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
948 | unsigned long flags; |
949 | ||
950 | /* | |
951 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
952 | * must not attempt merges on this) and that it acts as a soft | |
953 | * barrier | |
954 | */ | |
4aff5e23 JA |
955 | rq->cmd_type = REQ_TYPE_SPECIAL; |
956 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
957 | |
958 | rq->special = data; | |
959 | ||
960 | spin_lock_irqsave(q->queue_lock, flags); | |
961 | ||
962 | /* | |
963 | * If command is tagged, release the tag | |
964 | */ | |
867d1191 TH |
965 | if (blk_rq_tagged(rq)) |
966 | blk_queue_end_tag(q, rq); | |
1da177e4 | 967 | |
b238b3d4 | 968 | drive_stat_acct(rq, 1); |
867d1191 | 969 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 970 | blk_start_queueing(q); |
1da177e4 LT |
971 | spin_unlock_irqrestore(q->queue_lock, flags); |
972 | } | |
1da177e4 LT |
973 | EXPORT_SYMBOL(blk_insert_request); |
974 | ||
1da177e4 LT |
975 | /* |
976 | * add-request adds a request to the linked list. | |
977 | * queue lock is held and interrupts disabled, as we muck with the | |
978 | * request queue list. | |
979 | */ | |
6728cb0e | 980 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 981 | { |
b238b3d4 | 982 | drive_stat_acct(req, 1); |
1da177e4 | 983 | |
1da177e4 LT |
984 | /* |
985 | * elevator indicated where it wants this request to be | |
986 | * inserted at elevator_merge time | |
987 | */ | |
988 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
989 | } | |
6728cb0e | 990 | |
074a7aca TH |
991 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
992 | unsigned long now) | |
993 | { | |
994 | if (now == part->stamp) | |
995 | return; | |
996 | ||
997 | if (part->in_flight) { | |
998 | __part_stat_add(cpu, part, time_in_queue, | |
999 | part->in_flight * (now - part->stamp)); | |
1000 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1001 | } | |
1002 | part->stamp = now; | |
1003 | } | |
1004 | ||
1005 | /** | |
1006 | * part_round_stats() - Round off the performance stats on a struct | |
1da177e4 LT |
1007 | * disk_stats. |
1008 | * | |
1009 | * The average IO queue length and utilisation statistics are maintained | |
1010 | * by observing the current state of the queue length and the amount of | |
1011 | * time it has been in this state for. | |
1012 | * | |
1013 | * Normally, that accounting is done on IO completion, but that can result | |
1014 | * in more than a second's worth of IO being accounted for within any one | |
1015 | * second, leading to >100% utilisation. To deal with that, we call this | |
1016 | * function to do a round-off before returning the results when reading | |
1017 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1018 | * the current jiffies and restarts the counters again. | |
1019 | */ | |
c9959059 | 1020 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1021 | { |
1022 | unsigned long now = jiffies; | |
1023 | ||
074a7aca TH |
1024 | if (part->partno) |
1025 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1026 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1027 | } |
074a7aca | 1028 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1029 | |
1da177e4 LT |
1030 | /* |
1031 | * queue lock must be held | |
1032 | */ | |
165125e1 | 1033 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1034 | { |
1da177e4 LT |
1035 | if (unlikely(!q)) |
1036 | return; | |
1037 | if (unlikely(--req->ref_count)) | |
1038 | return; | |
1039 | ||
8922e16c TH |
1040 | elv_completed_request(q, req); |
1041 | ||
1da177e4 LT |
1042 | /* |
1043 | * Request may not have originated from ll_rw_blk. if not, | |
1044 | * it didn't come out of our reserved rq pools | |
1045 | */ | |
49171e5c | 1046 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1047 | int rw = rq_data_dir(req); |
4aff5e23 | 1048 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1049 | |
1da177e4 | 1050 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1051 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1052 | |
1053 | blk_free_request(q, req); | |
cb98fc8b | 1054 | freed_request(q, rw, priv); |
1da177e4 LT |
1055 | } |
1056 | } | |
6e39b69e MC |
1057 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1058 | ||
1da177e4 LT |
1059 | void blk_put_request(struct request *req) |
1060 | { | |
8922e16c | 1061 | unsigned long flags; |
165125e1 | 1062 | struct request_queue *q = req->q; |
8922e16c | 1063 | |
52a93ba8 FT |
1064 | spin_lock_irqsave(q->queue_lock, flags); |
1065 | __blk_put_request(q, req); | |
1066 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1067 | } |
1da177e4 LT |
1068 | EXPORT_SYMBOL(blk_put_request); |
1069 | ||
86db1e29 | 1070 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1071 | { |
c7c22e4d | 1072 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1073 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1074 | |
1075 | /* | |
1076 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1077 | */ | |
1078 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 1079 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
1080 | |
1081 | /* | |
1082 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1083 | */ | |
fb2dce86 | 1084 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1085 | req->cmd_flags |= REQ_DISCARD; |
1086 | if (bio_barrier(bio)) | |
1087 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1088 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 DW |
1089 | } else if (unlikely(bio_barrier(bio))) |
1090 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
52d9e675 | 1091 | |
b31dc66a | 1092 | if (bio_sync(bio)) |
4aff5e23 | 1093 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1094 | if (bio_rw_meta(bio)) |
1095 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1096 | |
52d9e675 TH |
1097 | req->errors = 0; |
1098 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1099 | req->ioprio = bio_prio(bio); |
52d9e675 | 1100 | req->start_time = jiffies; |
bc1c56fd | 1101 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1102 | } |
1103 | ||
165125e1 | 1104 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1105 | { |
450991bc | 1106 | struct request *req; |
fb2dce86 | 1107 | int el_ret, nr_sectors, barrier, discard, err; |
51da90fc JA |
1108 | const unsigned short prio = bio_prio(bio); |
1109 | const int sync = bio_sync(bio); | |
7749a8d4 | 1110 | int rw_flags; |
1da177e4 | 1111 | |
1da177e4 | 1112 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1113 | |
1114 | /* | |
1115 | * low level driver can indicate that it wants pages above a | |
1116 | * certain limit bounced to low memory (ie for highmem, or even | |
1117 | * ISA dma in theory) | |
1118 | */ | |
1119 | blk_queue_bounce(q, &bio); | |
1120 | ||
1da177e4 | 1121 | barrier = bio_barrier(bio); |
e17fc0a1 DW |
1122 | if (unlikely(barrier) && bio_has_data(bio) && |
1123 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1da177e4 LT |
1124 | err = -EOPNOTSUPP; |
1125 | goto end_io; | |
1126 | } | |
1127 | ||
fb2dce86 DW |
1128 | discard = bio_discard(bio); |
1129 | if (unlikely(discard) && !q->prepare_discard_fn) { | |
1130 | err = -EOPNOTSUPP; | |
1131 | goto end_io; | |
1132 | } | |
1133 | ||
1da177e4 LT |
1134 | spin_lock_irq(q->queue_lock); |
1135 | ||
450991bc | 1136 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1137 | goto get_rq; |
1138 | ||
1139 | el_ret = elv_merge(q, &req, bio); | |
1140 | switch (el_ret) { | |
6728cb0e JA |
1141 | case ELEVATOR_BACK_MERGE: |
1142 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1143 | |
6728cb0e JA |
1144 | if (!ll_back_merge_fn(q, req, bio)) |
1145 | break; | |
1da177e4 | 1146 | |
6728cb0e | 1147 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2056a782 | 1148 | |
6728cb0e JA |
1149 | req->biotail->bi_next = bio; |
1150 | req->biotail = bio; | |
1151 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1152 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1153 | if (!blk_rq_cpu_valid(req)) |
1154 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1155 | drive_stat_acct(req, 0); |
1156 | if (!attempt_back_merge(q, req)) | |
1157 | elv_merged_request(q, req, el_ret); | |
1158 | goto out; | |
1da177e4 | 1159 | |
6728cb0e JA |
1160 | case ELEVATOR_FRONT_MERGE: |
1161 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1162 | |
6728cb0e JA |
1163 | if (!ll_front_merge_fn(q, req, bio)) |
1164 | break; | |
1da177e4 | 1165 | |
6728cb0e | 1166 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2056a782 | 1167 | |
6728cb0e JA |
1168 | bio->bi_next = req->bio; |
1169 | req->bio = bio; | |
1da177e4 | 1170 | |
6728cb0e JA |
1171 | /* |
1172 | * may not be valid. if the low level driver said | |
1173 | * it didn't need a bounce buffer then it better | |
1174 | * not touch req->buffer either... | |
1175 | */ | |
1176 | req->buffer = bio_data(bio); | |
1177 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1178 | req->hard_cur_sectors = req->current_nr_sectors; | |
1179 | req->sector = req->hard_sector = bio->bi_sector; | |
1180 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1181 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1182 | if (!blk_rq_cpu_valid(req)) |
1183 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1184 | drive_stat_acct(req, 0); |
1185 | if (!attempt_front_merge(q, req)) | |
1186 | elv_merged_request(q, req, el_ret); | |
1187 | goto out; | |
1188 | ||
1189 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1190 | default: | |
1191 | ; | |
1da177e4 LT |
1192 | } |
1193 | ||
450991bc | 1194 | get_rq: |
7749a8d4 JA |
1195 | /* |
1196 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1197 | * but we need to set it earlier to expose the sync flag to the | |
1198 | * rq allocator and io schedulers. | |
1199 | */ | |
1200 | rw_flags = bio_data_dir(bio); | |
1201 | if (sync) | |
1202 | rw_flags |= REQ_RW_SYNC; | |
1203 | ||
1da177e4 | 1204 | /* |
450991bc | 1205 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1206 | * Returns with the queue unlocked. |
450991bc | 1207 | */ |
7749a8d4 | 1208 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1209 | |
450991bc NP |
1210 | /* |
1211 | * After dropping the lock and possibly sleeping here, our request | |
1212 | * may now be mergeable after it had proven unmergeable (above). | |
1213 | * We don't worry about that case for efficiency. It won't happen | |
1214 | * often, and the elevators are able to handle it. | |
1da177e4 | 1215 | */ |
52d9e675 | 1216 | init_request_from_bio(req, bio); |
1da177e4 | 1217 | |
450991bc | 1218 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1219 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1220 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1221 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
450991bc NP |
1222 | if (elv_queue_empty(q)) |
1223 | blk_plug_device(q); | |
1da177e4 LT |
1224 | add_request(q, req); |
1225 | out: | |
4a534f93 | 1226 | if (sync) |
1da177e4 | 1227 | __generic_unplug_device(q); |
1da177e4 LT |
1228 | spin_unlock_irq(q->queue_lock); |
1229 | return 0; | |
1230 | ||
1231 | end_io: | |
6712ecf8 | 1232 | bio_endio(bio, err); |
1da177e4 LT |
1233 | return 0; |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * If bio->bi_dev is a partition, remap the location | |
1238 | */ | |
1239 | static inline void blk_partition_remap(struct bio *bio) | |
1240 | { | |
1241 | struct block_device *bdev = bio->bi_bdev; | |
1242 | ||
bf2de6f5 | 1243 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1244 | struct hd_struct *p = bdev->bd_part; |
1245 | ||
1da177e4 LT |
1246 | bio->bi_sector += p->start_sect; |
1247 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b AB |
1248 | |
1249 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1250 | bdev->bd_dev, bio->bi_sector, | |
1251 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1252 | } |
1253 | } | |
1254 | ||
1da177e4 LT |
1255 | static void handle_bad_sector(struct bio *bio) |
1256 | { | |
1257 | char b[BDEVNAME_SIZE]; | |
1258 | ||
1259 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1260 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1261 | bdevname(bio->bi_bdev, b), | |
1262 | bio->bi_rw, | |
1263 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1264 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1265 | ||
1266 | set_bit(BIO_EOF, &bio->bi_flags); | |
1267 | } | |
1268 | ||
c17bb495 AM |
1269 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1270 | ||
1271 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1272 | ||
1273 | static int __init setup_fail_make_request(char *str) | |
1274 | { | |
1275 | return setup_fault_attr(&fail_make_request, str); | |
1276 | } | |
1277 | __setup("fail_make_request=", setup_fail_make_request); | |
1278 | ||
1279 | static int should_fail_request(struct bio *bio) | |
1280 | { | |
eddb2e26 TH |
1281 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1282 | ||
1283 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1284 | return should_fail(&fail_make_request, bio->bi_size); |
1285 | ||
1286 | return 0; | |
1287 | } | |
1288 | ||
1289 | static int __init fail_make_request_debugfs(void) | |
1290 | { | |
1291 | return init_fault_attr_dentries(&fail_make_request, | |
1292 | "fail_make_request"); | |
1293 | } | |
1294 | ||
1295 | late_initcall(fail_make_request_debugfs); | |
1296 | ||
1297 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1298 | ||
1299 | static inline int should_fail_request(struct bio *bio) | |
1300 | { | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1305 | ||
c07e2b41 JA |
1306 | /* |
1307 | * Check whether this bio extends beyond the end of the device. | |
1308 | */ | |
1309 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1310 | { | |
1311 | sector_t maxsector; | |
1312 | ||
1313 | if (!nr_sectors) | |
1314 | return 0; | |
1315 | ||
1316 | /* Test device or partition size, when known. */ | |
1317 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1318 | if (maxsector) { | |
1319 | sector_t sector = bio->bi_sector; | |
1320 | ||
1321 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1322 | /* | |
1323 | * This may well happen - the kernel calls bread() | |
1324 | * without checking the size of the device, e.g., when | |
1325 | * mounting a device. | |
1326 | */ | |
1327 | handle_bad_sector(bio); | |
1328 | return 1; | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | return 0; | |
1333 | } | |
1334 | ||
1da177e4 | 1335 | /** |
710027a4 | 1336 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1337 | * @bio: The bio describing the location in memory and on the device. |
1338 | * | |
1339 | * generic_make_request() is used to make I/O requests of block | |
1340 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1341 | * to be done. | |
1342 | * | |
1343 | * generic_make_request() does not return any status. The | |
1344 | * success/failure status of the request, along with notification of | |
1345 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1346 | * function described (one day) else where. | |
1347 | * | |
1348 | * The caller of generic_make_request must make sure that bi_io_vec | |
1349 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1350 | * set to describe the device address, and the | |
1351 | * bi_end_io and optionally bi_private are set to describe how | |
1352 | * completion notification should be signaled. | |
1353 | * | |
1354 | * generic_make_request and the drivers it calls may use bi_next if this | |
1355 | * bio happens to be merged with someone else, and may change bi_dev and | |
1356 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1357 | * should NOT be depended on after the call to generic_make_request. | |
1358 | */ | |
d89d8796 | 1359 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1360 | { |
165125e1 | 1361 | struct request_queue *q; |
5ddfe969 | 1362 | sector_t old_sector; |
1da177e4 | 1363 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1364 | dev_t old_dev; |
51fd77bd | 1365 | int err = -EIO; |
1da177e4 LT |
1366 | |
1367 | might_sleep(); | |
1da177e4 | 1368 | |
c07e2b41 JA |
1369 | if (bio_check_eod(bio, nr_sectors)) |
1370 | goto end_io; | |
1da177e4 LT |
1371 | |
1372 | /* | |
1373 | * Resolve the mapping until finished. (drivers are | |
1374 | * still free to implement/resolve their own stacking | |
1375 | * by explicitly returning 0) | |
1376 | * | |
1377 | * NOTE: we don't repeat the blk_size check for each new device. | |
1378 | * Stacking drivers are expected to know what they are doing. | |
1379 | */ | |
5ddfe969 | 1380 | old_sector = -1; |
2056a782 | 1381 | old_dev = 0; |
1da177e4 LT |
1382 | do { |
1383 | char b[BDEVNAME_SIZE]; | |
1384 | ||
1385 | q = bdev_get_queue(bio->bi_bdev); | |
1386 | if (!q) { | |
1387 | printk(KERN_ERR | |
1388 | "generic_make_request: Trying to access " | |
1389 | "nonexistent block-device %s (%Lu)\n", | |
1390 | bdevname(bio->bi_bdev, b), | |
1391 | (long long) bio->bi_sector); | |
1392 | end_io: | |
51fd77bd | 1393 | bio_endio(bio, err); |
1da177e4 LT |
1394 | break; |
1395 | } | |
1396 | ||
4fa253f3 | 1397 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1398 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1399 | bdevname(bio->bi_bdev, b), |
1400 | bio_sectors(bio), | |
1401 | q->max_hw_sectors); | |
1402 | goto end_io; | |
1403 | } | |
1404 | ||
fde6ad22 | 1405 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1406 | goto end_io; |
1407 | ||
c17bb495 AM |
1408 | if (should_fail_request(bio)) |
1409 | goto end_io; | |
1410 | ||
1da177e4 LT |
1411 | /* |
1412 | * If this device has partitions, remap block n | |
1413 | * of partition p to block n+start(p) of the disk. | |
1414 | */ | |
1415 | blk_partition_remap(bio); | |
1416 | ||
7ba1ba12 MP |
1417 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1418 | goto end_io; | |
1419 | ||
5ddfe969 | 1420 | if (old_sector != -1) |
4fa253f3 | 1421 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1422 | old_sector); |
2056a782 JA |
1423 | |
1424 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
1425 | ||
5ddfe969 | 1426 | old_sector = bio->bi_sector; |
2056a782 JA |
1427 | old_dev = bio->bi_bdev->bd_dev; |
1428 | ||
c07e2b41 JA |
1429 | if (bio_check_eod(bio, nr_sectors)) |
1430 | goto end_io; | |
fb2dce86 DW |
1431 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || |
1432 | (bio_discard(bio) && !q->prepare_discard_fn)) { | |
51fd77bd JA |
1433 | err = -EOPNOTSUPP; |
1434 | goto end_io; | |
1435 | } | |
5ddfe969 | 1436 | |
1da177e4 LT |
1437 | ret = q->make_request_fn(q, bio); |
1438 | } while (ret); | |
1439 | } | |
1440 | ||
d89d8796 NB |
1441 | /* |
1442 | * We only want one ->make_request_fn to be active at a time, | |
1443 | * else stack usage with stacked devices could be a problem. | |
1444 | * So use current->bio_{list,tail} to keep a list of requests | |
1445 | * submited by a make_request_fn function. | |
1446 | * current->bio_tail is also used as a flag to say if | |
1447 | * generic_make_request is currently active in this task or not. | |
1448 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1449 | * then a make_request is active, and new requests should be added | |
1450 | * at the tail | |
1451 | */ | |
1452 | void generic_make_request(struct bio *bio) | |
1453 | { | |
1454 | if (current->bio_tail) { | |
1455 | /* make_request is active */ | |
1456 | *(current->bio_tail) = bio; | |
1457 | bio->bi_next = NULL; | |
1458 | current->bio_tail = &bio->bi_next; | |
1459 | return; | |
1460 | } | |
1461 | /* following loop may be a bit non-obvious, and so deserves some | |
1462 | * explanation. | |
1463 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1464 | * ensure that) so we have a list with a single bio. | |
1465 | * We pretend that we have just taken it off a longer list, so | |
1466 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1467 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1468 | * added. __generic_make_request may indeed add some more bios | |
1469 | * through a recursive call to generic_make_request. If it | |
1470 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1471 | * from the top. In this case we really did just take the bio | |
1472 | * of the top of the list (no pretending) and so fixup bio_list and | |
1473 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1474 | * | |
1475 | * The loop was structured like this to make only one call to | |
1476 | * __generic_make_request (which is important as it is large and | |
1477 | * inlined) and to keep the structure simple. | |
1478 | */ | |
1479 | BUG_ON(bio->bi_next); | |
1480 | do { | |
1481 | current->bio_list = bio->bi_next; | |
1482 | if (bio->bi_next == NULL) | |
1483 | current->bio_tail = ¤t->bio_list; | |
1484 | else | |
1485 | bio->bi_next = NULL; | |
1486 | __generic_make_request(bio); | |
1487 | bio = current->bio_list; | |
1488 | } while (bio); | |
1489 | current->bio_tail = NULL; /* deactivate */ | |
1490 | } | |
1da177e4 LT |
1491 | EXPORT_SYMBOL(generic_make_request); |
1492 | ||
1493 | /** | |
710027a4 | 1494 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1495 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1496 | * @bio: The &struct bio which describes the I/O | |
1497 | * | |
1498 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1499 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1500 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1501 | * |
1502 | */ | |
1503 | void submit_bio(int rw, struct bio *bio) | |
1504 | { | |
1505 | int count = bio_sectors(bio); | |
1506 | ||
22e2c507 | 1507 | bio->bi_rw |= rw; |
1da177e4 | 1508 | |
bf2de6f5 JA |
1509 | /* |
1510 | * If it's a regular read/write or a barrier with data attached, | |
1511 | * go through the normal accounting stuff before submission. | |
1512 | */ | |
a9c701e5 | 1513 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1514 | if (rw & WRITE) { |
1515 | count_vm_events(PGPGOUT, count); | |
1516 | } else { | |
1517 | task_io_account_read(bio->bi_size); | |
1518 | count_vm_events(PGPGIN, count); | |
1519 | } | |
1520 | ||
1521 | if (unlikely(block_dump)) { | |
1522 | char b[BDEVNAME_SIZE]; | |
1523 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1524 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1525 | (rw & WRITE) ? "WRITE" : "READ", |
1526 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1527 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1528 | } |
1da177e4 LT |
1529 | } |
1530 | ||
1531 | generic_make_request(bio); | |
1532 | } | |
1da177e4 LT |
1533 | EXPORT_SYMBOL(submit_bio); |
1534 | ||
82124d60 KU |
1535 | /** |
1536 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1537 | * @q: the queue | |
1538 | * @rq: the request being checked | |
1539 | * | |
1540 | * Description: | |
1541 | * @rq may have been made based on weaker limitations of upper-level queues | |
1542 | * in request stacking drivers, and it may violate the limitation of @q. | |
1543 | * Since the block layer and the underlying device driver trust @rq | |
1544 | * after it is inserted to @q, it should be checked against @q before | |
1545 | * the insertion using this generic function. | |
1546 | * | |
1547 | * This function should also be useful for request stacking drivers | |
1548 | * in some cases below, so export this fuction. | |
1549 | * Request stacking drivers like request-based dm may change the queue | |
1550 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1551 | * Such request stacking drivers should check those requests agaist | |
1552 | * the new queue limits again when they dispatch those requests, | |
1553 | * although such checkings are also done against the old queue limits | |
1554 | * when submitting requests. | |
1555 | */ | |
1556 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1557 | { | |
1558 | if (rq->nr_sectors > q->max_sectors || | |
1559 | rq->data_len > q->max_hw_sectors << 9) { | |
1560 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1561 | return -EIO; | |
1562 | } | |
1563 | ||
1564 | /* | |
1565 | * queue's settings related to segment counting like q->bounce_pfn | |
1566 | * may differ from that of other stacking queues. | |
1567 | * Recalculate it to check the request correctly on this queue's | |
1568 | * limitation. | |
1569 | */ | |
1570 | blk_recalc_rq_segments(rq); | |
1571 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1572 | rq->nr_phys_segments > q->max_hw_segments) { | |
1573 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1574 | return -EIO; | |
1575 | } | |
1576 | ||
1577 | return 0; | |
1578 | } | |
1579 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1580 | ||
1581 | /** | |
1582 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1583 | * @q: the queue to submit the request | |
1584 | * @rq: the request being queued | |
1585 | */ | |
1586 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1587 | { | |
1588 | unsigned long flags; | |
1589 | ||
1590 | if (blk_rq_check_limits(q, rq)) | |
1591 | return -EIO; | |
1592 | ||
1593 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1594 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1595 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1596 | return -EIO; | |
1597 | #endif | |
1598 | ||
1599 | spin_lock_irqsave(q->queue_lock, flags); | |
1600 | ||
1601 | /* | |
1602 | * Submitting request must be dequeued before calling this function | |
1603 | * because it will be linked to another request_queue | |
1604 | */ | |
1605 | BUG_ON(blk_queued_rq(rq)); | |
1606 | ||
1607 | drive_stat_acct(rq, 1); | |
1608 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1609 | ||
1610 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1611 | ||
1612 | return 0; | |
1613 | } | |
1614 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1615 | ||
3bcddeac KU |
1616 | /** |
1617 | * __end_that_request_first - end I/O on a request | |
1618 | * @req: the request being processed | |
710027a4 | 1619 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
1620 | * @nr_bytes: number of bytes to complete |
1621 | * | |
1622 | * Description: | |
1623 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1624 | * for the next range of segments (if any) in the cluster. | |
1625 | * | |
1626 | * Return: | |
710027a4 RD |
1627 | * %0 - we are done with this request, call end_that_request_last() |
1628 | * %1 - still buffers pending for this request | |
3bcddeac | 1629 | **/ |
5450d3e1 | 1630 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1631 | int nr_bytes) |
1632 | { | |
5450d3e1 | 1633 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1634 | struct bio *bio; |
1635 | ||
2056a782 JA |
1636 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
1637 | ||
1da177e4 | 1638 | /* |
710027a4 | 1639 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual |
1da177e4 LT |
1640 | * sense key with us all the way through |
1641 | */ | |
1642 | if (!blk_pc_request(req)) | |
1643 | req->errors = 0; | |
1644 | ||
6728cb0e JA |
1645 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1646 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1647 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1648 | (unsigned long long)req->sector); | |
1649 | } | |
1650 | ||
d72d904a | 1651 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b | 1652 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1653 | struct hd_struct *part; |
c9959059 | 1654 | int cpu; |
a362357b | 1655 | |
074a7aca | 1656 | cpu = part_stat_lock(); |
e71bf0d0 | 1657 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
074a7aca TH |
1658 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); |
1659 | part_stat_unlock(); | |
d72d904a JA |
1660 | } |
1661 | ||
1da177e4 LT |
1662 | total_bytes = bio_nbytes = 0; |
1663 | while ((bio = req->bio) != NULL) { | |
1664 | int nbytes; | |
1665 | ||
bf2de6f5 JA |
1666 | /* |
1667 | * For an empty barrier request, the low level driver must | |
1668 | * store a potential error location in ->sector. We pass | |
1669 | * that back up in ->bi_sector. | |
1670 | */ | |
1671 | if (blk_empty_barrier(req)) | |
1672 | bio->bi_sector = req->sector; | |
1673 | ||
1da177e4 LT |
1674 | if (nr_bytes >= bio->bi_size) { |
1675 | req->bio = bio->bi_next; | |
1676 | nbytes = bio->bi_size; | |
5bb23a68 | 1677 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1678 | next_idx = 0; |
1679 | bio_nbytes = 0; | |
1680 | } else { | |
1681 | int idx = bio->bi_idx + next_idx; | |
1682 | ||
1683 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1684 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e | 1685 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1686 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
1687 | break; |
1688 | } | |
1689 | ||
1690 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1691 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1692 | ||
1693 | /* | |
1694 | * not a complete bvec done | |
1695 | */ | |
1696 | if (unlikely(nbytes > nr_bytes)) { | |
1697 | bio_nbytes += nr_bytes; | |
1698 | total_bytes += nr_bytes; | |
1699 | break; | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * advance to the next vector | |
1704 | */ | |
1705 | next_idx++; | |
1706 | bio_nbytes += nbytes; | |
1707 | } | |
1708 | ||
1709 | total_bytes += nbytes; | |
1710 | nr_bytes -= nbytes; | |
1711 | ||
6728cb0e JA |
1712 | bio = req->bio; |
1713 | if (bio) { | |
1da177e4 LT |
1714 | /* |
1715 | * end more in this run, or just return 'not-done' | |
1716 | */ | |
1717 | if (unlikely(nr_bytes <= 0)) | |
1718 | break; | |
1719 | } | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * completely done | |
1724 | */ | |
1725 | if (!req->bio) | |
1726 | return 0; | |
1727 | ||
1728 | /* | |
1729 | * if the request wasn't completed, update state | |
1730 | */ | |
1731 | if (bio_nbytes) { | |
5bb23a68 | 1732 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1733 | bio->bi_idx += next_idx; |
1734 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1735 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1736 | } | |
1737 | ||
1738 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1739 | blk_recalc_rq_segments(req); | |
1740 | return 1; | |
1741 | } | |
1742 | ||
1da177e4 LT |
1743 | /* |
1744 | * queue lock must be held | |
1745 | */ | |
5450d3e1 | 1746 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1747 | { |
1748 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1749 | |
242f9dcb JA |
1750 | blk_delete_timer(req); |
1751 | ||
b8286239 KU |
1752 | if (blk_rq_tagged(req)) |
1753 | blk_queue_end_tag(req->q, req); | |
1754 | ||
1755 | if (blk_queued_rq(req)) | |
1756 | blkdev_dequeue_request(req); | |
1da177e4 LT |
1757 | |
1758 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1759 | laptop_io_completion(); | |
1760 | ||
fd0ff8aa JA |
1761 | /* |
1762 | * Account IO completion. bar_rq isn't accounted as a normal | |
1763 | * IO on queueing nor completion. Accounting the containing | |
1764 | * request is enough. | |
1765 | */ | |
1766 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 1767 | unsigned long duration = jiffies - req->start_time; |
a362357b | 1768 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1769 | struct hd_struct *part; |
c9959059 | 1770 | int cpu; |
e71bf0d0 | 1771 | |
074a7aca | 1772 | cpu = part_stat_lock(); |
e71bf0d0 | 1773 | part = disk_map_sector_rcu(disk, req->sector); |
a362357b | 1774 | |
074a7aca TH |
1775 | part_stat_inc(cpu, part, ios[rw]); |
1776 | part_stat_add(cpu, part, ticks[rw], duration); | |
1777 | part_round_stats(cpu, part); | |
1778 | part_dec_in_flight(part); | |
e71bf0d0 | 1779 | |
074a7aca | 1780 | part_stat_unlock(); |
1da177e4 | 1781 | } |
b8286239 | 1782 | |
1da177e4 | 1783 | if (req->end_io) |
8ffdc655 | 1784 | req->end_io(req, error); |
b8286239 KU |
1785 | else { |
1786 | if (blk_bidi_rq(req)) | |
1787 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1788 | ||
1da177e4 | 1789 | __blk_put_request(req->q, req); |
b8286239 | 1790 | } |
1da177e4 LT |
1791 | } |
1792 | ||
a0cd1285 | 1793 | static inline void __end_request(struct request *rq, int uptodate, |
9e6e39f2 | 1794 | unsigned int nr_bytes) |
1da177e4 | 1795 | { |
9e6e39f2 KU |
1796 | int error = 0; |
1797 | ||
1798 | if (uptodate <= 0) | |
1799 | error = uptodate ? uptodate : -EIO; | |
1800 | ||
1801 | __blk_end_request(rq, error, nr_bytes); | |
1da177e4 LT |
1802 | } |
1803 | ||
3b11313a KU |
1804 | /** |
1805 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1806 | * @rq: the request being processed |
3b11313a KU |
1807 | **/ |
1808 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1809 | { |
1810 | if (blk_fs_request(rq)) | |
1811 | return rq->hard_nr_sectors << 9; | |
1812 | ||
1813 | return rq->data_len; | |
1814 | } | |
3b11313a KU |
1815 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1816 | ||
1817 | /** | |
1818 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1819 | * @rq: the request being processed |
3b11313a KU |
1820 | **/ |
1821 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1822 | { | |
1823 | if (blk_fs_request(rq)) | |
1824 | return rq->current_nr_sectors << 9; | |
1825 | ||
1826 | if (rq->bio) | |
1827 | return rq->bio->bi_size; | |
1828 | ||
1829 | return rq->data_len; | |
1830 | } | |
1831 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 JA |
1832 | |
1833 | /** | |
1834 | * end_queued_request - end all I/O on a queued request | |
1835 | * @rq: the request being processed | |
710027a4 | 1836 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1837 | * |
1838 | * Description: | |
1839 | * Ends all I/O on a request, and removes it from the block layer queues. | |
710027a4 | 1840 | * Not suitable for normal I/O completion, unless the driver still has |
a0cd1285 JA |
1841 | * the request attached to the block layer. |
1842 | * | |
1843 | **/ | |
1844 | void end_queued_request(struct request *rq, int uptodate) | |
1845 | { | |
9e6e39f2 | 1846 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1847 | } |
1848 | EXPORT_SYMBOL(end_queued_request); | |
1849 | ||
1850 | /** | |
1851 | * end_dequeued_request - end all I/O on a dequeued request | |
1852 | * @rq: the request being processed | |
710027a4 | 1853 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1854 | * |
1855 | * Description: | |
1856 | * Ends all I/O on a request. The request must already have been | |
1857 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
1858 | * for most drivers. | |
1859 | * | |
1860 | **/ | |
1861 | void end_dequeued_request(struct request *rq, int uptodate) | |
1862 | { | |
9e6e39f2 | 1863 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1864 | } |
1865 | EXPORT_SYMBOL(end_dequeued_request); | |
1866 | ||
1867 | ||
1868 | /** | |
1869 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1870 | * @req: the request being processed |
710027a4 | 1871 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1872 | * |
1873 | * Description: | |
1874 | * Ends I/O on the current segment of a request. If that is the only | |
1875 | * remaining segment, the request is also completed and freed. | |
1876 | * | |
710027a4 RD |
1877 | * This is a remnant of how older block drivers handled I/O completions. |
1878 | * Modern drivers typically end I/O on the full request in one go, unless | |
a0cd1285 JA |
1879 | * they have a residual value to account for. For that case this function |
1880 | * isn't really useful, unless the residual just happens to be the | |
1881 | * full current segment. In other words, don't use this function in new | |
839e96af JA |
1882 | * code. Use blk_end_request() or __blk_end_request() to end partial parts |
1883 | * of a request, or end_dequeued_request() and end_queued_request() to | |
1884 | * completely end IO on a dequeued/queued request. | |
a0cd1285 JA |
1885 | * |
1886 | **/ | |
1887 | void end_request(struct request *req, int uptodate) | |
1888 | { | |
9e6e39f2 | 1889 | __end_request(req, uptodate, req->hard_cur_sectors << 9); |
a0cd1285 | 1890 | } |
1da177e4 LT |
1891 | EXPORT_SYMBOL(end_request); |
1892 | ||
32fab448 KU |
1893 | static int end_that_request_data(struct request *rq, int error, |
1894 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1895 | { | |
1896 | if (rq->bio) { | |
1897 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1898 | return 1; | |
1899 | ||
1900 | /* Bidi request must be completed as a whole */ | |
1901 | if (blk_bidi_rq(rq) && | |
1902 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1903 | return 1; | |
1904 | } | |
1905 | ||
1906 | return 0; | |
1907 | } | |
1908 | ||
336cdb40 | 1909 | /** |
e19a3ab0 KU |
1910 | * blk_end_io - Generic end_io function to complete a request. |
1911 | * @rq: the request being processed | |
710027a4 | 1912 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1913 | * @nr_bytes: number of bytes to complete @rq |
1914 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1915 | * @drv_callback: function called between completion of bios in the request |
1916 | * and completion of the request. | |
710027a4 | 1917 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1918 | * completion of the request. |
336cdb40 KU |
1919 | * |
1920 | * Description: | |
e3a04fe3 | 1921 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1922 | * If @rq has leftover, sets it up for the next range of segments. |
1923 | * | |
1924 | * Return: | |
710027a4 RD |
1925 | * %0 - we are done with this request |
1926 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1927 | **/ |
22b13210 JA |
1928 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1929 | unsigned int bidi_bytes, | |
1930 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1931 | { |
1932 | struct request_queue *q = rq->q; | |
1933 | unsigned long flags = 0UL; | |
336cdb40 | 1934 | |
32fab448 KU |
1935 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
1936 | return 1; | |
336cdb40 | 1937 | |
e19a3ab0 KU |
1938 | /* Special feature for tricky drivers */ |
1939 | if (drv_callback && drv_callback(rq)) | |
1940 | return 1; | |
1941 | ||
336cdb40 KU |
1942 | add_disk_randomness(rq->rq_disk); |
1943 | ||
1944 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1945 | end_that_request_last(rq, error); |
336cdb40 KU |
1946 | spin_unlock_irqrestore(q->queue_lock, flags); |
1947 | ||
1948 | return 0; | |
1949 | } | |
e19a3ab0 KU |
1950 | |
1951 | /** | |
1952 | * blk_end_request - Helper function for drivers to complete the request. | |
1953 | * @rq: the request being processed | |
710027a4 | 1954 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
1955 | * @nr_bytes: number of bytes to complete |
1956 | * | |
1957 | * Description: | |
1958 | * Ends I/O on a number of bytes attached to @rq. | |
1959 | * If @rq has leftover, sets it up for the next range of segments. | |
1960 | * | |
1961 | * Return: | |
710027a4 RD |
1962 | * %0 - we are done with this request |
1963 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1964 | **/ |
22b13210 | 1965 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1966 | { |
e3a04fe3 | 1967 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1968 | } |
336cdb40 KU |
1969 | EXPORT_SYMBOL_GPL(blk_end_request); |
1970 | ||
1971 | /** | |
1972 | * __blk_end_request - Helper function for drivers to complete the request. | |
1973 | * @rq: the request being processed | |
710027a4 | 1974 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
1975 | * @nr_bytes: number of bytes to complete |
1976 | * | |
1977 | * Description: | |
1978 | * Must be called with queue lock held unlike blk_end_request(). | |
1979 | * | |
1980 | * Return: | |
710027a4 RD |
1981 | * %0 - we are done with this request |
1982 | * %1 - still buffers pending for this request | |
336cdb40 | 1983 | **/ |
22b13210 | 1984 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1985 | { |
60540161 | 1986 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
051cc395 | 1987 | return 1; |
336cdb40 KU |
1988 | |
1989 | add_disk_randomness(rq->rq_disk); | |
1990 | ||
b8286239 | 1991 | end_that_request_last(rq, error); |
336cdb40 KU |
1992 | |
1993 | return 0; | |
1994 | } | |
1995 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1996 | ||
e3a04fe3 KU |
1997 | /** |
1998 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1999 | * @rq: the bidi request being processed | |
710027a4 | 2000 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2001 | * @nr_bytes: number of bytes to complete @rq |
2002 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2003 | * | |
2004 | * Description: | |
2005 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2006 | * | |
2007 | * Return: | |
710027a4 RD |
2008 | * %0 - we are done with this request |
2009 | * %1 - still buffers pending for this request | |
e3a04fe3 | 2010 | **/ |
22b13210 JA |
2011 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
2012 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
2013 | { |
2014 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
2015 | } | |
2016 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2017 | ||
32fab448 KU |
2018 | /** |
2019 | * blk_update_request - Special helper function for request stacking drivers | |
2020 | * @rq: the request being processed | |
2021 | * @error: %0 for success, < %0 for error | |
2022 | * @nr_bytes: number of bytes to complete @rq | |
2023 | * | |
2024 | * Description: | |
2025 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2026 | * the request structure even if @rq doesn't have leftover. | |
2027 | * If @rq has leftover, sets it up for the next range of segments. | |
2028 | * | |
2029 | * This special helper function is only for request stacking drivers | |
2030 | * (e.g. request-based dm) so that they can handle partial completion. | |
2031 | * Actual device drivers should use blk_end_request instead. | |
2032 | */ | |
2033 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2034 | { | |
2035 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2036 | /* | |
2037 | * These members are not updated in end_that_request_data() | |
2038 | * when all bios are completed. | |
2039 | * Update them so that the request stacking driver can find | |
2040 | * how many bytes remain in the request later. | |
2041 | */ | |
2042 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2043 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2044 | } | |
2045 | } | |
2046 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2047 | ||
e19a3ab0 KU |
2048 | /** |
2049 | * blk_end_request_callback - Special helper function for tricky drivers | |
2050 | * @rq: the request being processed | |
710027a4 | 2051 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2052 | * @nr_bytes: number of bytes to complete |
2053 | * @drv_callback: function called between completion of bios in the request | |
2054 | * and completion of the request. | |
710027a4 | 2055 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
2056 | * completion of the request. |
2057 | * | |
2058 | * Description: | |
2059 | * Ends I/O on a number of bytes attached to @rq. | |
2060 | * If @rq has leftover, sets it up for the next range of segments. | |
2061 | * | |
2062 | * This special helper function is used only for existing tricky drivers. | |
2063 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2064 | * This interface will be removed when such drivers are rewritten. | |
2065 | * Don't use this interface in other places anymore. | |
2066 | * | |
2067 | * Return: | |
710027a4 RD |
2068 | * %0 - we are done with this request |
2069 | * %1 - this request is not freed yet. | |
2070 | * this request still has pending buffers or | |
2071 | * the driver doesn't want to finish this request yet. | |
e19a3ab0 | 2072 | **/ |
22b13210 JA |
2073 | int blk_end_request_callback(struct request *rq, int error, |
2074 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2075 | int (drv_callback)(struct request *)) |
2076 | { | |
e3a04fe3 | 2077 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2078 | } |
2079 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2080 | ||
86db1e29 JA |
2081 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2082 | struct bio *bio) | |
1da177e4 | 2083 | { |
d628eaef DW |
2084 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2085 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2086 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2087 | |
fb2dce86 DW |
2088 | if (bio_has_data(bio)) { |
2089 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2090 | rq->buffer = bio_data(bio); |
2091 | } | |
1da177e4 LT |
2092 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2093 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2094 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2095 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2096 | |
2097 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2098 | |
66846572 N |
2099 | if (bio->bi_bdev) |
2100 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2101 | } | |
1da177e4 | 2102 | |
18887ad9 | 2103 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2104 | { |
2105 | return queue_work(kblockd_workqueue, work); | |
2106 | } | |
1da177e4 LT |
2107 | EXPORT_SYMBOL(kblockd_schedule_work); |
2108 | ||
19a75d83 | 2109 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 2110 | { |
28e53bdd | 2111 | cancel_work_sync(work); |
1da177e4 | 2112 | } |
19a75d83 | 2113 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
2114 | |
2115 | int __init blk_dev_init(void) | |
2116 | { | |
2117 | kblockd_workqueue = create_workqueue("kblockd"); | |
2118 | if (!kblockd_workqueue) | |
2119 | panic("Failed to create kblockd\n"); | |
2120 | ||
2121 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2122 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2123 | |
8324aa91 | 2124 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2125 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2126 | |
d38ecf93 | 2127 | return 0; |
1da177e4 | 2128 | } |
1da177e4 | 2129 |