]>
Commit | Line | Data |
---|---|---|
1 | /* | |
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> | |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> | |
7 | * - July2000 | |
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 | */ | |
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> | |
24 | #include <linux/completion.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
28 | #include <linux/task_io_accounting_ops.h> | |
29 | #include <linux/blktrace_api.h> | |
30 | #include <linux/fault-inject.h> | |
31 | #include <trace/block.h> | |
32 | ||
33 | #include "blk.h" | |
34 | ||
35 | DEFINE_TRACE(block_plug); | |
36 | DEFINE_TRACE(block_unplug_io); | |
37 | DEFINE_TRACE(block_unplug_timer); | |
38 | DEFINE_TRACE(block_getrq); | |
39 | DEFINE_TRACE(block_sleeprq); | |
40 | DEFINE_TRACE(block_rq_requeue); | |
41 | DEFINE_TRACE(block_bio_backmerge); | |
42 | DEFINE_TRACE(block_bio_frontmerge); | |
43 | DEFINE_TRACE(block_bio_queue); | |
44 | DEFINE_TRACE(block_rq_complete); | |
45 | DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */ | |
46 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); | |
47 | ||
48 | static int __make_request(struct request_queue *q, struct bio *bio); | |
49 | ||
50 | /* | |
51 | * For the allocated request tables | |
52 | */ | |
53 | static struct kmem_cache *request_cachep; | |
54 | ||
55 | /* | |
56 | * For queue allocation | |
57 | */ | |
58 | struct kmem_cache *blk_requestq_cachep; | |
59 | ||
60 | /* | |
61 | * Controlling structure to kblockd | |
62 | */ | |
63 | static struct workqueue_struct *kblockd_workqueue; | |
64 | ||
65 | static void drive_stat_acct(struct request *rq, int new_io) | |
66 | { | |
67 | struct hd_struct *part; | |
68 | int rw = rq_data_dir(rq); | |
69 | int cpu; | |
70 | ||
71 | if (!blk_do_io_stat(rq)) | |
72 | return; | |
73 | ||
74 | cpu = part_stat_lock(); | |
75 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); | |
76 | ||
77 | if (!new_io) | |
78 | part_stat_inc(cpu, part, merges[rw]); | |
79 | else { | |
80 | part_round_stats(cpu, part); | |
81 | part_inc_in_flight(part); | |
82 | } | |
83 | ||
84 | part_stat_unlock(); | |
85 | } | |
86 | ||
87 | void blk_queue_congestion_threshold(struct request_queue *q) | |
88 | { | |
89 | int nr; | |
90 | ||
91 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
92 | if (nr > q->nr_requests) | |
93 | nr = q->nr_requests; | |
94 | q->nr_congestion_on = nr; | |
95 | ||
96 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
97 | if (nr < 1) | |
98 | nr = 1; | |
99 | q->nr_congestion_off = nr; | |
100 | } | |
101 | ||
102 | /** | |
103 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
104 | * @bdev: device | |
105 | * | |
106 | * Locates the passed device's request queue and returns the address of its | |
107 | * backing_dev_info | |
108 | * | |
109 | * Will return NULL if the request queue cannot be located. | |
110 | */ | |
111 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
112 | { | |
113 | struct backing_dev_info *ret = NULL; | |
114 | struct request_queue *q = bdev_get_queue(bdev); | |
115 | ||
116 | if (q) | |
117 | ret = &q->backing_dev_info; | |
118 | return ret; | |
119 | } | |
120 | EXPORT_SYMBOL(blk_get_backing_dev_info); | |
121 | ||
122 | void blk_rq_init(struct request_queue *q, struct request *rq) | |
123 | { | |
124 | memset(rq, 0, sizeof(*rq)); | |
125 | ||
126 | INIT_LIST_HEAD(&rq->queuelist); | |
127 | INIT_LIST_HEAD(&rq->timeout_list); | |
128 | rq->cpu = -1; | |
129 | rq->q = q; | |
130 | rq->sector = rq->hard_sector = (sector_t) -1; | |
131 | INIT_HLIST_NODE(&rq->hash); | |
132 | RB_CLEAR_NODE(&rq->rb_node); | |
133 | rq->cmd = rq->__cmd; | |
134 | rq->cmd_len = BLK_MAX_CDB; | |
135 | rq->tag = -1; | |
136 | rq->ref_count = 1; | |
137 | rq->start_time = jiffies; | |
138 | } | |
139 | EXPORT_SYMBOL(blk_rq_init); | |
140 | ||
141 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
142 | unsigned int nbytes, int error) | |
143 | { | |
144 | struct request_queue *q = rq->q; | |
145 | ||
146 | if (&q->bar_rq != rq) { | |
147 | if (error) | |
148 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
149 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
150 | error = -EIO; | |
151 | ||
152 | if (unlikely(nbytes > bio->bi_size)) { | |
153 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
154 | __func__, nbytes, bio->bi_size); | |
155 | nbytes = bio->bi_size; | |
156 | } | |
157 | ||
158 | if (unlikely(rq->cmd_flags & REQ_QUIET)) | |
159 | set_bit(BIO_QUIET, &bio->bi_flags); | |
160 | ||
161 | bio->bi_size -= nbytes; | |
162 | bio->bi_sector += (nbytes >> 9); | |
163 | ||
164 | if (bio_integrity(bio)) | |
165 | bio_integrity_advance(bio, nbytes); | |
166 | ||
167 | if (bio->bi_size == 0) | |
168 | bio_endio(bio, error); | |
169 | } else { | |
170 | ||
171 | /* | |
172 | * Okay, this is the barrier request in progress, just | |
173 | * record the error; | |
174 | */ | |
175 | if (error && !q->orderr) | |
176 | q->orderr = error; | |
177 | } | |
178 | } | |
179 | ||
180 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
181 | { | |
182 | int bit; | |
183 | ||
184 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, | |
185 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
186 | rq->cmd_flags); | |
187 | ||
188 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", | |
189 | (unsigned long long)rq->sector, | |
190 | rq->nr_sectors, | |
191 | rq->current_nr_sectors); | |
192 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", | |
193 | rq->bio, rq->biotail, | |
194 | rq->buffer, rq->data_len); | |
195 | ||
196 | if (blk_pc_request(rq)) { | |
197 | printk(KERN_INFO " cdb: "); | |
198 | for (bit = 0; bit < BLK_MAX_CDB; bit++) | |
199 | printk("%02x ", rq->cmd[bit]); | |
200 | printk("\n"); | |
201 | } | |
202 | } | |
203 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
204 | ||
205 | /* | |
206 | * "plug" the device if there are no outstanding requests: this will | |
207 | * force the transfer to start only after we have put all the requests | |
208 | * on the list. | |
209 | * | |
210 | * This is called with interrupts off and no requests on the queue and | |
211 | * with the queue lock held. | |
212 | */ | |
213 | void blk_plug_device(struct request_queue *q) | |
214 | { | |
215 | WARN_ON(!irqs_disabled()); | |
216 | ||
217 | /* | |
218 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
219 | * which will restart the queueing | |
220 | */ | |
221 | if (blk_queue_stopped(q)) | |
222 | return; | |
223 | ||
224 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { | |
225 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | |
226 | trace_block_plug(q); | |
227 | } | |
228 | } | |
229 | EXPORT_SYMBOL(blk_plug_device); | |
230 | ||
231 | /** | |
232 | * blk_plug_device_unlocked - plug a device without queue lock held | |
233 | * @q: The &struct request_queue to plug | |
234 | * | |
235 | * Description: | |
236 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
237 | * interrupts. | |
238 | **/ | |
239 | void blk_plug_device_unlocked(struct request_queue *q) | |
240 | { | |
241 | unsigned long flags; | |
242 | ||
243 | spin_lock_irqsave(q->queue_lock, flags); | |
244 | blk_plug_device(q); | |
245 | spin_unlock_irqrestore(q->queue_lock, flags); | |
246 | } | |
247 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
248 | ||
249 | /* | |
250 | * remove the queue from the plugged list, if present. called with | |
251 | * queue lock held and interrupts disabled. | |
252 | */ | |
253 | int blk_remove_plug(struct request_queue *q) | |
254 | { | |
255 | WARN_ON(!irqs_disabled()); | |
256 | ||
257 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) | |
258 | return 0; | |
259 | ||
260 | del_timer(&q->unplug_timer); | |
261 | return 1; | |
262 | } | |
263 | EXPORT_SYMBOL(blk_remove_plug); | |
264 | ||
265 | /* | |
266 | * remove the plug and let it rip.. | |
267 | */ | |
268 | void __generic_unplug_device(struct request_queue *q) | |
269 | { | |
270 | if (unlikely(blk_queue_stopped(q))) | |
271 | return; | |
272 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) | |
273 | return; | |
274 | ||
275 | q->request_fn(q); | |
276 | } | |
277 | ||
278 | /** | |
279 | * generic_unplug_device - fire a request queue | |
280 | * @q: The &struct request_queue in question | |
281 | * | |
282 | * Description: | |
283 | * Linux uses plugging to build bigger requests queues before letting | |
284 | * the device have at them. If a queue is plugged, the I/O scheduler | |
285 | * is still adding and merging requests on the queue. Once the queue | |
286 | * gets unplugged, the request_fn defined for the queue is invoked and | |
287 | * transfers started. | |
288 | **/ | |
289 | void generic_unplug_device(struct request_queue *q) | |
290 | { | |
291 | if (blk_queue_plugged(q)) { | |
292 | spin_lock_irq(q->queue_lock); | |
293 | __generic_unplug_device(q); | |
294 | spin_unlock_irq(q->queue_lock); | |
295 | } | |
296 | } | |
297 | EXPORT_SYMBOL(generic_unplug_device); | |
298 | ||
299 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
300 | struct page *page) | |
301 | { | |
302 | struct request_queue *q = bdi->unplug_io_data; | |
303 | ||
304 | blk_unplug(q); | |
305 | } | |
306 | ||
307 | void blk_unplug_work(struct work_struct *work) | |
308 | { | |
309 | struct request_queue *q = | |
310 | container_of(work, struct request_queue, unplug_work); | |
311 | ||
312 | trace_block_unplug_io(q); | |
313 | q->unplug_fn(q); | |
314 | } | |
315 | ||
316 | void blk_unplug_timeout(unsigned long data) | |
317 | { | |
318 | struct request_queue *q = (struct request_queue *)data; | |
319 | ||
320 | trace_block_unplug_timer(q); | |
321 | kblockd_schedule_work(q, &q->unplug_work); | |
322 | } | |
323 | ||
324 | void blk_unplug(struct request_queue *q) | |
325 | { | |
326 | /* | |
327 | * devices don't necessarily have an ->unplug_fn defined | |
328 | */ | |
329 | if (q->unplug_fn) { | |
330 | trace_block_unplug_io(q); | |
331 | q->unplug_fn(q); | |
332 | } | |
333 | } | |
334 | EXPORT_SYMBOL(blk_unplug); | |
335 | ||
336 | /** | |
337 | * blk_start_queue - restart a previously stopped queue | |
338 | * @q: The &struct request_queue in question | |
339 | * | |
340 | * Description: | |
341 | * blk_start_queue() will clear the stop flag on the queue, and call | |
342 | * the request_fn for the queue if it was in a stopped state when | |
343 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
344 | **/ | |
345 | void blk_start_queue(struct request_queue *q) | |
346 | { | |
347 | WARN_ON(!irqs_disabled()); | |
348 | ||
349 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
350 | __blk_run_queue(q); | |
351 | } | |
352 | EXPORT_SYMBOL(blk_start_queue); | |
353 | ||
354 | /** | |
355 | * blk_stop_queue - stop a queue | |
356 | * @q: The &struct request_queue in question | |
357 | * | |
358 | * Description: | |
359 | * The Linux block layer assumes that a block driver will consume all | |
360 | * entries on the request queue when the request_fn strategy is called. | |
361 | * Often this will not happen, because of hardware limitations (queue | |
362 | * depth settings). If a device driver gets a 'queue full' response, | |
363 | * or if it simply chooses not to queue more I/O at one point, it can | |
364 | * call this function to prevent the request_fn from being called until | |
365 | * the driver has signalled it's ready to go again. This happens by calling | |
366 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
367 | **/ | |
368 | void blk_stop_queue(struct request_queue *q) | |
369 | { | |
370 | blk_remove_plug(q); | |
371 | queue_flag_set(QUEUE_FLAG_STOPPED, q); | |
372 | } | |
373 | EXPORT_SYMBOL(blk_stop_queue); | |
374 | ||
375 | /** | |
376 | * blk_sync_queue - cancel any pending callbacks on a queue | |
377 | * @q: the queue | |
378 | * | |
379 | * Description: | |
380 | * The block layer may perform asynchronous callback activity | |
381 | * on a queue, such as calling the unplug function after a timeout. | |
382 | * A block device may call blk_sync_queue to ensure that any | |
383 | * such activity is cancelled, thus allowing it to release resources | |
384 | * that the callbacks might use. The caller must already have made sure | |
385 | * that its ->make_request_fn will not re-add plugging prior to calling | |
386 | * this function. | |
387 | * | |
388 | */ | |
389 | void blk_sync_queue(struct request_queue *q) | |
390 | { | |
391 | del_timer_sync(&q->unplug_timer); | |
392 | del_timer_sync(&q->timeout); | |
393 | cancel_work_sync(&q->unplug_work); | |
394 | } | |
395 | EXPORT_SYMBOL(blk_sync_queue); | |
396 | ||
397 | /** | |
398 | * __blk_run_queue - run a single device queue | |
399 | * @q: The queue to run | |
400 | * | |
401 | * Description: | |
402 | * See @blk_run_queue. This variant must be called with the queue lock | |
403 | * held and interrupts disabled. | |
404 | * | |
405 | */ | |
406 | void __blk_run_queue(struct request_queue *q) | |
407 | { | |
408 | blk_remove_plug(q); | |
409 | ||
410 | if (unlikely(blk_queue_stopped(q))) | |
411 | return; | |
412 | ||
413 | if (elv_queue_empty(q)) | |
414 | return; | |
415 | ||
416 | /* | |
417 | * Only recurse once to avoid overrunning the stack, let the unplug | |
418 | * handling reinvoke the handler shortly if we already got there. | |
419 | */ | |
420 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
421 | q->request_fn(q); | |
422 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
423 | } else { | |
424 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
425 | kblockd_schedule_work(q, &q->unplug_work); | |
426 | } | |
427 | } | |
428 | EXPORT_SYMBOL(__blk_run_queue); | |
429 | ||
430 | /** | |
431 | * blk_run_queue - run a single device queue | |
432 | * @q: The queue to run | |
433 | * | |
434 | * Description: | |
435 | * Invoke request handling on this queue, if it has pending work to do. | |
436 | * May be used to restart queueing when a request has completed. | |
437 | */ | |
438 | void blk_run_queue(struct request_queue *q) | |
439 | { | |
440 | unsigned long flags; | |
441 | ||
442 | spin_lock_irqsave(q->queue_lock, flags); | |
443 | __blk_run_queue(q); | |
444 | spin_unlock_irqrestore(q->queue_lock, flags); | |
445 | } | |
446 | EXPORT_SYMBOL(blk_run_queue); | |
447 | ||
448 | void blk_put_queue(struct request_queue *q) | |
449 | { | |
450 | kobject_put(&q->kobj); | |
451 | } | |
452 | ||
453 | void blk_cleanup_queue(struct request_queue *q) | |
454 | { | |
455 | /* | |
456 | * We know we have process context here, so we can be a little | |
457 | * cautious and ensure that pending block actions on this device | |
458 | * are done before moving on. Going into this function, we should | |
459 | * not have processes doing IO to this device. | |
460 | */ | |
461 | blk_sync_queue(q); | |
462 | ||
463 | mutex_lock(&q->sysfs_lock); | |
464 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); | |
465 | mutex_unlock(&q->sysfs_lock); | |
466 | ||
467 | if (q->elevator) | |
468 | elevator_exit(q->elevator); | |
469 | ||
470 | blk_put_queue(q); | |
471 | } | |
472 | EXPORT_SYMBOL(blk_cleanup_queue); | |
473 | ||
474 | static int blk_init_free_list(struct request_queue *q) | |
475 | { | |
476 | struct request_list *rl = &q->rq; | |
477 | ||
478 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; | |
479 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
480 | rl->elvpriv = 0; | |
481 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); | |
482 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
483 | ||
484 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | |
485 | mempool_free_slab, request_cachep, q->node); | |
486 | ||
487 | if (!rl->rq_pool) | |
488 | return -ENOMEM; | |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
493 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
494 | { | |
495 | return blk_alloc_queue_node(gfp_mask, -1); | |
496 | } | |
497 | EXPORT_SYMBOL(blk_alloc_queue); | |
498 | ||
499 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
500 | { | |
501 | struct request_queue *q; | |
502 | int err; | |
503 | ||
504 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
505 | gfp_mask | __GFP_ZERO, node_id); | |
506 | if (!q) | |
507 | return NULL; | |
508 | ||
509 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
510 | q->backing_dev_info.unplug_io_data = q; | |
511 | err = bdi_init(&q->backing_dev_info); | |
512 | if (err) { | |
513 | kmem_cache_free(blk_requestq_cachep, q); | |
514 | return NULL; | |
515 | } | |
516 | ||
517 | init_timer(&q->unplug_timer); | |
518 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); | |
519 | INIT_LIST_HEAD(&q->timeout_list); | |
520 | INIT_WORK(&q->unplug_work, blk_unplug_work); | |
521 | ||
522 | kobject_init(&q->kobj, &blk_queue_ktype); | |
523 | ||
524 | mutex_init(&q->sysfs_lock); | |
525 | spin_lock_init(&q->__queue_lock); | |
526 | ||
527 | return q; | |
528 | } | |
529 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
530 | ||
531 | /** | |
532 | * blk_init_queue - prepare a request queue for use with a block device | |
533 | * @rfn: The function to be called to process requests that have been | |
534 | * placed on the queue. | |
535 | * @lock: Request queue spin lock | |
536 | * | |
537 | * Description: | |
538 | * If a block device wishes to use the standard request handling procedures, | |
539 | * which sorts requests and coalesces adjacent requests, then it must | |
540 | * call blk_init_queue(). The function @rfn will be called when there | |
541 | * are requests on the queue that need to be processed. If the device | |
542 | * supports plugging, then @rfn may not be called immediately when requests | |
543 | * are available on the queue, but may be called at some time later instead. | |
544 | * Plugged queues are generally unplugged when a buffer belonging to one | |
545 | * of the requests on the queue is needed, or due to memory pressure. | |
546 | * | |
547 | * @rfn is not required, or even expected, to remove all requests off the | |
548 | * queue, but only as many as it can handle at a time. If it does leave | |
549 | * requests on the queue, it is responsible for arranging that the requests | |
550 | * get dealt with eventually. | |
551 | * | |
552 | * The queue spin lock must be held while manipulating the requests on the | |
553 | * request queue; this lock will be taken also from interrupt context, so irq | |
554 | * disabling is needed for it. | |
555 | * | |
556 | * Function returns a pointer to the initialized request queue, or %NULL if | |
557 | * it didn't succeed. | |
558 | * | |
559 | * Note: | |
560 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
561 | * when the block device is deactivated (such as at module unload). | |
562 | **/ | |
563 | ||
564 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
565 | { | |
566 | return blk_init_queue_node(rfn, lock, -1); | |
567 | } | |
568 | EXPORT_SYMBOL(blk_init_queue); | |
569 | ||
570 | struct request_queue * | |
571 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
572 | { | |
573 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
574 | ||
575 | if (!q) | |
576 | return NULL; | |
577 | ||
578 | q->node = node_id; | |
579 | if (blk_init_free_list(q)) { | |
580 | kmem_cache_free(blk_requestq_cachep, q); | |
581 | return NULL; | |
582 | } | |
583 | ||
584 | /* | |
585 | * if caller didn't supply a lock, they get per-queue locking with | |
586 | * our embedded lock | |
587 | */ | |
588 | if (!lock) | |
589 | lock = &q->__queue_lock; | |
590 | ||
591 | q->request_fn = rfn; | |
592 | q->prep_rq_fn = NULL; | |
593 | q->unplug_fn = generic_unplug_device; | |
594 | q->queue_flags = QUEUE_FLAG_DEFAULT; | |
595 | q->queue_lock = lock; | |
596 | ||
597 | /* | |
598 | * This also sets hw/phys segments, boundary and size | |
599 | */ | |
600 | blk_queue_make_request(q, __make_request); | |
601 | ||
602 | q->sg_reserved_size = INT_MAX; | |
603 | ||
604 | blk_set_cmd_filter_defaults(&q->cmd_filter); | |
605 | ||
606 | /* | |
607 | * all done | |
608 | */ | |
609 | if (!elevator_init(q, NULL)) { | |
610 | blk_queue_congestion_threshold(q); | |
611 | return q; | |
612 | } | |
613 | ||
614 | blk_put_queue(q); | |
615 | return NULL; | |
616 | } | |
617 | EXPORT_SYMBOL(blk_init_queue_node); | |
618 | ||
619 | int blk_get_queue(struct request_queue *q) | |
620 | { | |
621 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { | |
622 | kobject_get(&q->kobj); | |
623 | return 0; | |
624 | } | |
625 | ||
626 | return 1; | |
627 | } | |
628 | ||
629 | static inline void blk_free_request(struct request_queue *q, struct request *rq) | |
630 | { | |
631 | if (rq->cmd_flags & REQ_ELVPRIV) | |
632 | elv_put_request(q, rq); | |
633 | mempool_free(rq, q->rq.rq_pool); | |
634 | } | |
635 | ||
636 | static struct request * | |
637 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) | |
638 | { | |
639 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
640 | ||
641 | if (!rq) | |
642 | return NULL; | |
643 | ||
644 | blk_rq_init(q, rq); | |
645 | ||
646 | rq->cmd_flags = flags | REQ_ALLOCED; | |
647 | ||
648 | if (priv) { | |
649 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { | |
650 | mempool_free(rq, q->rq.rq_pool); | |
651 | return NULL; | |
652 | } | |
653 | rq->cmd_flags |= REQ_ELVPRIV; | |
654 | } | |
655 | ||
656 | return rq; | |
657 | } | |
658 | ||
659 | /* | |
660 | * ioc_batching returns true if the ioc is a valid batching request and | |
661 | * should be given priority access to a request. | |
662 | */ | |
663 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
664 | { | |
665 | if (!ioc) | |
666 | return 0; | |
667 | ||
668 | /* | |
669 | * Make sure the process is able to allocate at least 1 request | |
670 | * even if the batch times out, otherwise we could theoretically | |
671 | * lose wakeups. | |
672 | */ | |
673 | return ioc->nr_batch_requests == q->nr_batching || | |
674 | (ioc->nr_batch_requests > 0 | |
675 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
676 | } | |
677 | ||
678 | /* | |
679 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
680 | * will cause the process to be a "batcher" on all queues in the system. This | |
681 | * is the behaviour we want though - once it gets a wakeup it should be given | |
682 | * a nice run. | |
683 | */ | |
684 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
685 | { | |
686 | if (!ioc || ioc_batching(q, ioc)) | |
687 | return; | |
688 | ||
689 | ioc->nr_batch_requests = q->nr_batching; | |
690 | ioc->last_waited = jiffies; | |
691 | } | |
692 | ||
693 | static void __freed_request(struct request_queue *q, int sync) | |
694 | { | |
695 | struct request_list *rl = &q->rq; | |
696 | ||
697 | if (rl->count[sync] < queue_congestion_off_threshold(q)) | |
698 | blk_clear_queue_congested(q, sync); | |
699 | ||
700 | if (rl->count[sync] + 1 <= q->nr_requests) { | |
701 | if (waitqueue_active(&rl->wait[sync])) | |
702 | wake_up(&rl->wait[sync]); | |
703 | ||
704 | blk_clear_queue_full(q, sync); | |
705 | } | |
706 | } | |
707 | ||
708 | /* | |
709 | * A request has just been released. Account for it, update the full and | |
710 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
711 | */ | |
712 | static void freed_request(struct request_queue *q, int sync, int priv) | |
713 | { | |
714 | struct request_list *rl = &q->rq; | |
715 | ||
716 | rl->count[sync]--; | |
717 | if (priv) | |
718 | rl->elvpriv--; | |
719 | ||
720 | __freed_request(q, sync); | |
721 | ||
722 | if (unlikely(rl->starved[sync ^ 1])) | |
723 | __freed_request(q, sync ^ 1); | |
724 | } | |
725 | ||
726 | /* | |
727 | * Get a free request, queue_lock must be held. | |
728 | * Returns NULL on failure, with queue_lock held. | |
729 | * Returns !NULL on success, with queue_lock *not held*. | |
730 | */ | |
731 | static struct request *get_request(struct request_queue *q, int rw_flags, | |
732 | struct bio *bio, gfp_t gfp_mask) | |
733 | { | |
734 | struct request *rq = NULL; | |
735 | struct request_list *rl = &q->rq; | |
736 | struct io_context *ioc = NULL; | |
737 | const bool is_sync = rw_is_sync(rw_flags) != 0; | |
738 | int may_queue, priv; | |
739 | ||
740 | may_queue = elv_may_queue(q, rw_flags); | |
741 | if (may_queue == ELV_MQUEUE_NO) | |
742 | goto rq_starved; | |
743 | ||
744 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { | |
745 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
746 | ioc = current_io_context(GFP_ATOMIC, q->node); | |
747 | /* | |
748 | * The queue will fill after this allocation, so set | |
749 | * it as full, and mark this process as "batching". | |
750 | * This process will be allowed to complete a batch of | |
751 | * requests, others will be blocked. | |
752 | */ | |
753 | if (!blk_queue_full(q, is_sync)) { | |
754 | ioc_set_batching(q, ioc); | |
755 | blk_set_queue_full(q, is_sync); | |
756 | } else { | |
757 | if (may_queue != ELV_MQUEUE_MUST | |
758 | && !ioc_batching(q, ioc)) { | |
759 | /* | |
760 | * The queue is full and the allocating | |
761 | * process is not a "batcher", and not | |
762 | * exempted by the IO scheduler | |
763 | */ | |
764 | goto out; | |
765 | } | |
766 | } | |
767 | } | |
768 | blk_set_queue_congested(q, is_sync); | |
769 | } | |
770 | ||
771 | /* | |
772 | * Only allow batching queuers to allocate up to 50% over the defined | |
773 | * limit of requests, otherwise we could have thousands of requests | |
774 | * allocated with any setting of ->nr_requests | |
775 | */ | |
776 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) | |
777 | goto out; | |
778 | ||
779 | rl->count[is_sync]++; | |
780 | rl->starved[is_sync] = 0; | |
781 | ||
782 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
783 | if (priv) | |
784 | rl->elvpriv++; | |
785 | ||
786 | if (blk_queue_io_stat(q)) | |
787 | rw_flags |= REQ_IO_STAT; | |
788 | spin_unlock_irq(q->queue_lock); | |
789 | ||
790 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); | |
791 | if (unlikely(!rq)) { | |
792 | /* | |
793 | * Allocation failed presumably due to memory. Undo anything | |
794 | * we might have messed up. | |
795 | * | |
796 | * Allocating task should really be put onto the front of the | |
797 | * wait queue, but this is pretty rare. | |
798 | */ | |
799 | spin_lock_irq(q->queue_lock); | |
800 | freed_request(q, is_sync, priv); | |
801 | ||
802 | /* | |
803 | * in the very unlikely event that allocation failed and no | |
804 | * requests for this direction was pending, mark us starved | |
805 | * so that freeing of a request in the other direction will | |
806 | * notice us. another possible fix would be to split the | |
807 | * rq mempool into READ and WRITE | |
808 | */ | |
809 | rq_starved: | |
810 | if (unlikely(rl->count[is_sync] == 0)) | |
811 | rl->starved[is_sync] = 1; | |
812 | ||
813 | goto out; | |
814 | } | |
815 | ||
816 | /* | |
817 | * ioc may be NULL here, and ioc_batching will be false. That's | |
818 | * OK, if the queue is under the request limit then requests need | |
819 | * not count toward the nr_batch_requests limit. There will always | |
820 | * be some limit enforced by BLK_BATCH_TIME. | |
821 | */ | |
822 | if (ioc_batching(q, ioc)) | |
823 | ioc->nr_batch_requests--; | |
824 | ||
825 | trace_block_getrq(q, bio, rw_flags & 1); | |
826 | out: | |
827 | return rq; | |
828 | } | |
829 | ||
830 | /* | |
831 | * No available requests for this queue, unplug the device and wait for some | |
832 | * requests to become available. | |
833 | * | |
834 | * Called with q->queue_lock held, and returns with it unlocked. | |
835 | */ | |
836 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, | |
837 | struct bio *bio) | |
838 | { | |
839 | const bool is_sync = rw_is_sync(rw_flags) != 0; | |
840 | struct request *rq; | |
841 | ||
842 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
843 | while (!rq) { | |
844 | DEFINE_WAIT(wait); | |
845 | struct io_context *ioc; | |
846 | struct request_list *rl = &q->rq; | |
847 | ||
848 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, | |
849 | TASK_UNINTERRUPTIBLE); | |
850 | ||
851 | trace_block_sleeprq(q, bio, rw_flags & 1); | |
852 | ||
853 | __generic_unplug_device(q); | |
854 | spin_unlock_irq(q->queue_lock); | |
855 | io_schedule(); | |
856 | ||
857 | /* | |
858 | * After sleeping, we become a "batching" process and | |
859 | * will be able to allocate at least one request, and | |
860 | * up to a big batch of them for a small period time. | |
861 | * See ioc_batching, ioc_set_batching | |
862 | */ | |
863 | ioc = current_io_context(GFP_NOIO, q->node); | |
864 | ioc_set_batching(q, ioc); | |
865 | ||
866 | spin_lock_irq(q->queue_lock); | |
867 | finish_wait(&rl->wait[is_sync], &wait); | |
868 | ||
869 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
870 | }; | |
871 | ||
872 | return rq; | |
873 | } | |
874 | ||
875 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
876 | { | |
877 | struct request *rq; | |
878 | ||
879 | BUG_ON(rw != READ && rw != WRITE); | |
880 | ||
881 | spin_lock_irq(q->queue_lock); | |
882 | if (gfp_mask & __GFP_WAIT) { | |
883 | rq = get_request_wait(q, rw, NULL); | |
884 | } else { | |
885 | rq = get_request(q, rw, NULL, gfp_mask); | |
886 | if (!rq) | |
887 | spin_unlock_irq(q->queue_lock); | |
888 | } | |
889 | /* q->queue_lock is unlocked at this point */ | |
890 | ||
891 | return rq; | |
892 | } | |
893 | EXPORT_SYMBOL(blk_get_request); | |
894 | ||
895 | /** | |
896 | * blk_requeue_request - put a request back on queue | |
897 | * @q: request queue where request should be inserted | |
898 | * @rq: request to be inserted | |
899 | * | |
900 | * Description: | |
901 | * Drivers often keep queueing requests until the hardware cannot accept | |
902 | * more, when that condition happens we need to put the request back | |
903 | * on the queue. Must be called with queue lock held. | |
904 | */ | |
905 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
906 | { | |
907 | blk_delete_timer(rq); | |
908 | blk_clear_rq_complete(rq); | |
909 | trace_block_rq_requeue(q, rq); | |
910 | ||
911 | if (blk_rq_tagged(rq)) | |
912 | blk_queue_end_tag(q, rq); | |
913 | ||
914 | elv_requeue_request(q, rq); | |
915 | } | |
916 | EXPORT_SYMBOL(blk_requeue_request); | |
917 | ||
918 | /** | |
919 | * blk_insert_request - insert a special request into a request queue | |
920 | * @q: request queue where request should be inserted | |
921 | * @rq: request to be inserted | |
922 | * @at_head: insert request at head or tail of queue | |
923 | * @data: private data | |
924 | * | |
925 | * Description: | |
926 | * Many block devices need to execute commands asynchronously, so they don't | |
927 | * block the whole kernel from preemption during request execution. This is | |
928 | * accomplished normally by inserting aritficial requests tagged as | |
929 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them | |
930 | * be scheduled for actual execution by the request queue. | |
931 | * | |
932 | * We have the option of inserting the head or the tail of the queue. | |
933 | * Typically we use the tail for new ioctls and so forth. We use the head | |
934 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
935 | * host that is unable to accept a particular command. | |
936 | */ | |
937 | void blk_insert_request(struct request_queue *q, struct request *rq, | |
938 | int at_head, void *data) | |
939 | { | |
940 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
941 | unsigned long flags; | |
942 | ||
943 | /* | |
944 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
945 | * must not attempt merges on this) and that it acts as a soft | |
946 | * barrier | |
947 | */ | |
948 | rq->cmd_type = REQ_TYPE_SPECIAL; | |
949 | ||
950 | rq->special = data; | |
951 | ||
952 | spin_lock_irqsave(q->queue_lock, flags); | |
953 | ||
954 | /* | |
955 | * If command is tagged, release the tag | |
956 | */ | |
957 | if (blk_rq_tagged(rq)) | |
958 | blk_queue_end_tag(q, rq); | |
959 | ||
960 | drive_stat_acct(rq, 1); | |
961 | __elv_add_request(q, rq, where, 0); | |
962 | __blk_run_queue(q); | |
963 | spin_unlock_irqrestore(q->queue_lock, flags); | |
964 | } | |
965 | EXPORT_SYMBOL(blk_insert_request); | |
966 | ||
967 | /* | |
968 | * add-request adds a request to the linked list. | |
969 | * queue lock is held and interrupts disabled, as we muck with the | |
970 | * request queue list. | |
971 | */ | |
972 | static inline void add_request(struct request_queue *q, struct request *req) | |
973 | { | |
974 | drive_stat_acct(req, 1); | |
975 | ||
976 | /* | |
977 | * elevator indicated where it wants this request to be | |
978 | * inserted at elevator_merge time | |
979 | */ | |
980 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
981 | } | |
982 | ||
983 | static void part_round_stats_single(int cpu, struct hd_struct *part, | |
984 | unsigned long now) | |
985 | { | |
986 | if (now == part->stamp) | |
987 | return; | |
988 | ||
989 | if (part->in_flight) { | |
990 | __part_stat_add(cpu, part, time_in_queue, | |
991 | part->in_flight * (now - part->stamp)); | |
992 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
993 | } | |
994 | part->stamp = now; | |
995 | } | |
996 | ||
997 | /** | |
998 | * part_round_stats() - Round off the performance stats on a struct disk_stats. | |
999 | * @cpu: cpu number for stats access | |
1000 | * @part: target partition | |
1001 | * | |
1002 | * The average IO queue length and utilisation statistics are maintained | |
1003 | * by observing the current state of the queue length and the amount of | |
1004 | * time it has been in this state for. | |
1005 | * | |
1006 | * Normally, that accounting is done on IO completion, but that can result | |
1007 | * in more than a second's worth of IO being accounted for within any one | |
1008 | * second, leading to >100% utilisation. To deal with that, we call this | |
1009 | * function to do a round-off before returning the results when reading | |
1010 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1011 | * the current jiffies and restarts the counters again. | |
1012 | */ | |
1013 | void part_round_stats(int cpu, struct hd_struct *part) | |
1014 | { | |
1015 | unsigned long now = jiffies; | |
1016 | ||
1017 | if (part->partno) | |
1018 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1019 | part_round_stats_single(cpu, part, now); | |
1020 | } | |
1021 | EXPORT_SYMBOL_GPL(part_round_stats); | |
1022 | ||
1023 | /* | |
1024 | * queue lock must be held | |
1025 | */ | |
1026 | void __blk_put_request(struct request_queue *q, struct request *req) | |
1027 | { | |
1028 | if (unlikely(!q)) | |
1029 | return; | |
1030 | if (unlikely(--req->ref_count)) | |
1031 | return; | |
1032 | ||
1033 | elv_completed_request(q, req); | |
1034 | ||
1035 | /* this is a bio leak */ | |
1036 | WARN_ON(req->bio != NULL); | |
1037 | ||
1038 | /* | |
1039 | * Request may not have originated from ll_rw_blk. if not, | |
1040 | * it didn't come out of our reserved rq pools | |
1041 | */ | |
1042 | if (req->cmd_flags & REQ_ALLOCED) { | |
1043 | int is_sync = rq_is_sync(req) != 0; | |
1044 | int priv = req->cmd_flags & REQ_ELVPRIV; | |
1045 | ||
1046 | BUG_ON(!list_empty(&req->queuelist)); | |
1047 | BUG_ON(!hlist_unhashed(&req->hash)); | |
1048 | ||
1049 | blk_free_request(q, req); | |
1050 | freed_request(q, is_sync, priv); | |
1051 | } | |
1052 | } | |
1053 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
1054 | ||
1055 | void blk_put_request(struct request *req) | |
1056 | { | |
1057 | unsigned long flags; | |
1058 | struct request_queue *q = req->q; | |
1059 | ||
1060 | spin_lock_irqsave(q->queue_lock, flags); | |
1061 | __blk_put_request(q, req); | |
1062 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1063 | } | |
1064 | EXPORT_SYMBOL(blk_put_request); | |
1065 | ||
1066 | void init_request_from_bio(struct request *req, struct bio *bio) | |
1067 | { | |
1068 | req->cpu = bio->bi_comp_cpu; | |
1069 | req->cmd_type = REQ_TYPE_FS; | |
1070 | ||
1071 | /* | |
1072 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1073 | */ | |
1074 | if (bio_rw_ahead(bio)) | |
1075 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1076 | REQ_FAILFAST_DRIVER); | |
1077 | if (bio_failfast_dev(bio)) | |
1078 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1079 | if (bio_failfast_transport(bio)) | |
1080 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1081 | if (bio_failfast_driver(bio)) | |
1082 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
1083 | ||
1084 | if (unlikely(bio_discard(bio))) { | |
1085 | req->cmd_flags |= REQ_DISCARD; | |
1086 | if (bio_barrier(bio)) | |
1087 | req->cmd_flags |= REQ_SOFTBARRIER; | |
1088 | req->q->prepare_discard_fn(req->q, req); | |
1089 | } else if (unlikely(bio_barrier(bio))) | |
1090 | req->cmd_flags |= REQ_HARDBARRIER; | |
1091 | ||
1092 | if (bio_sync(bio)) | |
1093 | req->cmd_flags |= REQ_RW_SYNC; | |
1094 | if (bio_rw_meta(bio)) | |
1095 | req->cmd_flags |= REQ_RW_META; | |
1096 | if (bio_noidle(bio)) | |
1097 | req->cmd_flags |= REQ_NOIDLE; | |
1098 | ||
1099 | req->errors = 0; | |
1100 | req->hard_sector = req->sector = bio->bi_sector; | |
1101 | req->ioprio = bio_prio(bio); | |
1102 | blk_rq_bio_prep(req->q, req, bio); | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Only disabling plugging for non-rotational devices if it does tagging | |
1107 | * as well, otherwise we do need the proper merging | |
1108 | */ | |
1109 | static inline bool queue_should_plug(struct request_queue *q) | |
1110 | { | |
1111 | return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); | |
1112 | } | |
1113 | ||
1114 | static int __make_request(struct request_queue *q, struct bio *bio) | |
1115 | { | |
1116 | struct request *req; | |
1117 | int el_ret, nr_sectors; | |
1118 | const unsigned short prio = bio_prio(bio); | |
1119 | const int sync = bio_sync(bio); | |
1120 | const int unplug = bio_unplug(bio); | |
1121 | int rw_flags; | |
1122 | ||
1123 | nr_sectors = bio_sectors(bio); | |
1124 | ||
1125 | /* | |
1126 | * low level driver can indicate that it wants pages above a | |
1127 | * certain limit bounced to low memory (ie for highmem, or even | |
1128 | * ISA dma in theory) | |
1129 | */ | |
1130 | blk_queue_bounce(q, &bio); | |
1131 | ||
1132 | spin_lock_irq(q->queue_lock); | |
1133 | ||
1134 | if (unlikely(bio_barrier(bio)) || elv_queue_empty(q)) | |
1135 | goto get_rq; | |
1136 | ||
1137 | el_ret = elv_merge(q, &req, bio); | |
1138 | switch (el_ret) { | |
1139 | case ELEVATOR_BACK_MERGE: | |
1140 | BUG_ON(!rq_mergeable(req)); | |
1141 | ||
1142 | if (!ll_back_merge_fn(q, req, bio)) | |
1143 | break; | |
1144 | ||
1145 | trace_block_bio_backmerge(q, bio); | |
1146 | ||
1147 | req->biotail->bi_next = bio; | |
1148 | req->biotail = bio; | |
1149 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1150 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1151 | if (!blk_rq_cpu_valid(req)) | |
1152 | req->cpu = bio->bi_comp_cpu; | |
1153 | drive_stat_acct(req, 0); | |
1154 | if (!attempt_back_merge(q, req)) | |
1155 | elv_merged_request(q, req, el_ret); | |
1156 | goto out; | |
1157 | ||
1158 | case ELEVATOR_FRONT_MERGE: | |
1159 | BUG_ON(!rq_mergeable(req)); | |
1160 | ||
1161 | if (!ll_front_merge_fn(q, req, bio)) | |
1162 | break; | |
1163 | ||
1164 | trace_block_bio_frontmerge(q, bio); | |
1165 | ||
1166 | bio->bi_next = req->bio; | |
1167 | req->bio = bio; | |
1168 | ||
1169 | /* | |
1170 | * may not be valid. if the low level driver said | |
1171 | * it didn't need a bounce buffer then it better | |
1172 | * not touch req->buffer either... | |
1173 | */ | |
1174 | req->buffer = bio_data(bio); | |
1175 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1176 | req->hard_cur_sectors = req->current_nr_sectors; | |
1177 | req->sector = req->hard_sector = bio->bi_sector; | |
1178 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1179 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1180 | if (!blk_rq_cpu_valid(req)) | |
1181 | req->cpu = bio->bi_comp_cpu; | |
1182 | drive_stat_acct(req, 0); | |
1183 | if (!attempt_front_merge(q, req)) | |
1184 | elv_merged_request(q, req, el_ret); | |
1185 | goto out; | |
1186 | ||
1187 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1188 | default: | |
1189 | ; | |
1190 | } | |
1191 | ||
1192 | get_rq: | |
1193 | /* | |
1194 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1195 | * but we need to set it earlier to expose the sync flag to the | |
1196 | * rq allocator and io schedulers. | |
1197 | */ | |
1198 | rw_flags = bio_data_dir(bio); | |
1199 | if (sync) | |
1200 | rw_flags |= REQ_RW_SYNC; | |
1201 | ||
1202 | /* | |
1203 | * Grab a free request. This is might sleep but can not fail. | |
1204 | * Returns with the queue unlocked. | |
1205 | */ | |
1206 | req = get_request_wait(q, rw_flags, bio); | |
1207 | ||
1208 | /* | |
1209 | * After dropping the lock and possibly sleeping here, our request | |
1210 | * may now be mergeable after it had proven unmergeable (above). | |
1211 | * We don't worry about that case for efficiency. It won't happen | |
1212 | * often, and the elevators are able to handle it. | |
1213 | */ | |
1214 | init_request_from_bio(req, bio); | |
1215 | ||
1216 | spin_lock_irq(q->queue_lock); | |
1217 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || | |
1218 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1219 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
1220 | if (queue_should_plug(q) && elv_queue_empty(q)) | |
1221 | blk_plug_device(q); | |
1222 | add_request(q, req); | |
1223 | out: | |
1224 | if (unplug || !queue_should_plug(q)) | |
1225 | __generic_unplug_device(q); | |
1226 | spin_unlock_irq(q->queue_lock); | |
1227 | return 0; | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * If bio->bi_dev is a partition, remap the location | |
1232 | */ | |
1233 | static inline void blk_partition_remap(struct bio *bio) | |
1234 | { | |
1235 | struct block_device *bdev = bio->bi_bdev; | |
1236 | ||
1237 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { | |
1238 | struct hd_struct *p = bdev->bd_part; | |
1239 | ||
1240 | bio->bi_sector += p->start_sect; | |
1241 | bio->bi_bdev = bdev->bd_contains; | |
1242 | ||
1243 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1244 | bdev->bd_dev, bio->bi_sector, | |
1245 | bio->bi_sector - p->start_sect); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | static void handle_bad_sector(struct bio *bio) | |
1250 | { | |
1251 | char b[BDEVNAME_SIZE]; | |
1252 | ||
1253 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1254 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1255 | bdevname(bio->bi_bdev, b), | |
1256 | bio->bi_rw, | |
1257 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1258 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1259 | ||
1260 | set_bit(BIO_EOF, &bio->bi_flags); | |
1261 | } | |
1262 | ||
1263 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1264 | ||
1265 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1266 | ||
1267 | static int __init setup_fail_make_request(char *str) | |
1268 | { | |
1269 | return setup_fault_attr(&fail_make_request, str); | |
1270 | } | |
1271 | __setup("fail_make_request=", setup_fail_make_request); | |
1272 | ||
1273 | static int should_fail_request(struct bio *bio) | |
1274 | { | |
1275 | struct hd_struct *part = bio->bi_bdev->bd_part; | |
1276 | ||
1277 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
1278 | return should_fail(&fail_make_request, bio->bi_size); | |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | static int __init fail_make_request_debugfs(void) | |
1284 | { | |
1285 | return init_fault_attr_dentries(&fail_make_request, | |
1286 | "fail_make_request"); | |
1287 | } | |
1288 | ||
1289 | late_initcall(fail_make_request_debugfs); | |
1290 | ||
1291 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1292 | ||
1293 | static inline int should_fail_request(struct bio *bio) | |
1294 | { | |
1295 | return 0; | |
1296 | } | |
1297 | ||
1298 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1299 | ||
1300 | /* | |
1301 | * Check whether this bio extends beyond the end of the device. | |
1302 | */ | |
1303 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1304 | { | |
1305 | sector_t maxsector; | |
1306 | ||
1307 | if (!nr_sectors) | |
1308 | return 0; | |
1309 | ||
1310 | /* Test device or partition size, when known. */ | |
1311 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1312 | if (maxsector) { | |
1313 | sector_t sector = bio->bi_sector; | |
1314 | ||
1315 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1316 | /* | |
1317 | * This may well happen - the kernel calls bread() | |
1318 | * without checking the size of the device, e.g., when | |
1319 | * mounting a device. | |
1320 | */ | |
1321 | handle_bad_sector(bio); | |
1322 | return 1; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | return 0; | |
1327 | } | |
1328 | ||
1329 | /** | |
1330 | * generic_make_request - hand a buffer to its device driver for I/O | |
1331 | * @bio: The bio describing the location in memory and on the device. | |
1332 | * | |
1333 | * generic_make_request() is used to make I/O requests of block | |
1334 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1335 | * to be done. | |
1336 | * | |
1337 | * generic_make_request() does not return any status. The | |
1338 | * success/failure status of the request, along with notification of | |
1339 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1340 | * function described (one day) else where. | |
1341 | * | |
1342 | * The caller of generic_make_request must make sure that bi_io_vec | |
1343 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1344 | * set to describe the device address, and the | |
1345 | * bi_end_io and optionally bi_private are set to describe how | |
1346 | * completion notification should be signaled. | |
1347 | * | |
1348 | * generic_make_request and the drivers it calls may use bi_next if this | |
1349 | * bio happens to be merged with someone else, and may change bi_dev and | |
1350 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1351 | * should NOT be depended on after the call to generic_make_request. | |
1352 | */ | |
1353 | static inline void __generic_make_request(struct bio *bio) | |
1354 | { | |
1355 | struct request_queue *q; | |
1356 | sector_t old_sector; | |
1357 | int ret, nr_sectors = bio_sectors(bio); | |
1358 | dev_t old_dev; | |
1359 | int err = -EIO; | |
1360 | ||
1361 | might_sleep(); | |
1362 | ||
1363 | if (bio_check_eod(bio, nr_sectors)) | |
1364 | goto end_io; | |
1365 | ||
1366 | /* | |
1367 | * Resolve the mapping until finished. (drivers are | |
1368 | * still free to implement/resolve their own stacking | |
1369 | * by explicitly returning 0) | |
1370 | * | |
1371 | * NOTE: we don't repeat the blk_size check for each new device. | |
1372 | * Stacking drivers are expected to know what they are doing. | |
1373 | */ | |
1374 | old_sector = -1; | |
1375 | old_dev = 0; | |
1376 | do { | |
1377 | char b[BDEVNAME_SIZE]; | |
1378 | ||
1379 | q = bdev_get_queue(bio->bi_bdev); | |
1380 | if (unlikely(!q)) { | |
1381 | printk(KERN_ERR | |
1382 | "generic_make_request: Trying to access " | |
1383 | "nonexistent block-device %s (%Lu)\n", | |
1384 | bdevname(bio->bi_bdev, b), | |
1385 | (long long) bio->bi_sector); | |
1386 | goto end_io; | |
1387 | } | |
1388 | ||
1389 | if (unlikely(nr_sectors > q->max_hw_sectors)) { | |
1390 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1391 | bdevname(bio->bi_bdev, b), | |
1392 | bio_sectors(bio), | |
1393 | q->max_hw_sectors); | |
1394 | goto end_io; | |
1395 | } | |
1396 | ||
1397 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) | |
1398 | goto end_io; | |
1399 | ||
1400 | if (should_fail_request(bio)) | |
1401 | goto end_io; | |
1402 | ||
1403 | /* | |
1404 | * If this device has partitions, remap block n | |
1405 | * of partition p to block n+start(p) of the disk. | |
1406 | */ | |
1407 | blk_partition_remap(bio); | |
1408 | ||
1409 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) | |
1410 | goto end_io; | |
1411 | ||
1412 | if (old_sector != -1) | |
1413 | trace_block_remap(q, bio, old_dev, bio->bi_sector, | |
1414 | old_sector); | |
1415 | ||
1416 | trace_block_bio_queue(q, bio); | |
1417 | ||
1418 | old_sector = bio->bi_sector; | |
1419 | old_dev = bio->bi_bdev->bd_dev; | |
1420 | ||
1421 | if (bio_check_eod(bio, nr_sectors)) | |
1422 | goto end_io; | |
1423 | ||
1424 | if (bio_discard(bio) && !q->prepare_discard_fn) { | |
1425 | err = -EOPNOTSUPP; | |
1426 | goto end_io; | |
1427 | } | |
1428 | if (bio_barrier(bio) && bio_has_data(bio) && | |
1429 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1430 | err = -EOPNOTSUPP; | |
1431 | goto end_io; | |
1432 | } | |
1433 | ||
1434 | ret = q->make_request_fn(q, bio); | |
1435 | } while (ret); | |
1436 | ||
1437 | return; | |
1438 | ||
1439 | end_io: | |
1440 | bio_endio(bio, err); | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * We only want one ->make_request_fn to be active at a time, | |
1445 | * else stack usage with stacked devices could be a problem. | |
1446 | * So use current->bio_{list,tail} to keep a list of requests | |
1447 | * submited by a make_request_fn function. | |
1448 | * current->bio_tail is also used as a flag to say if | |
1449 | * generic_make_request is currently active in this task or not. | |
1450 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1451 | * then a make_request is active, and new requests should be added | |
1452 | * at the tail | |
1453 | */ | |
1454 | void generic_make_request(struct bio *bio) | |
1455 | { | |
1456 | if (current->bio_tail) { | |
1457 | /* make_request is active */ | |
1458 | *(current->bio_tail) = bio; | |
1459 | bio->bi_next = NULL; | |
1460 | current->bio_tail = &bio->bi_next; | |
1461 | return; | |
1462 | } | |
1463 | /* following loop may be a bit non-obvious, and so deserves some | |
1464 | * explanation. | |
1465 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1466 | * ensure that) so we have a list with a single bio. | |
1467 | * We pretend that we have just taken it off a longer list, so | |
1468 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1469 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1470 | * added. __generic_make_request may indeed add some more bios | |
1471 | * through a recursive call to generic_make_request. If it | |
1472 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1473 | * from the top. In this case we really did just take the bio | |
1474 | * of the top of the list (no pretending) and so fixup bio_list and | |
1475 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1476 | * | |
1477 | * The loop was structured like this to make only one call to | |
1478 | * __generic_make_request (which is important as it is large and | |
1479 | * inlined) and to keep the structure simple. | |
1480 | */ | |
1481 | BUG_ON(bio->bi_next); | |
1482 | do { | |
1483 | current->bio_list = bio->bi_next; | |
1484 | if (bio->bi_next == NULL) | |
1485 | current->bio_tail = ¤t->bio_list; | |
1486 | else | |
1487 | bio->bi_next = NULL; | |
1488 | __generic_make_request(bio); | |
1489 | bio = current->bio_list; | |
1490 | } while (bio); | |
1491 | current->bio_tail = NULL; /* deactivate */ | |
1492 | } | |
1493 | EXPORT_SYMBOL(generic_make_request); | |
1494 | ||
1495 | /** | |
1496 | * submit_bio - submit a bio to the block device layer for I/O | |
1497 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
1498 | * @bio: The &struct bio which describes the I/O | |
1499 | * | |
1500 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1501 | * uses that function to do most of the work. Both are fairly rough | |
1502 | * interfaces; @bio must be presetup and ready for I/O. | |
1503 | * | |
1504 | */ | |
1505 | void submit_bio(int rw, struct bio *bio) | |
1506 | { | |
1507 | int count = bio_sectors(bio); | |
1508 | ||
1509 | bio->bi_rw |= rw; | |
1510 | ||
1511 | /* | |
1512 | * If it's a regular read/write or a barrier with data attached, | |
1513 | * go through the normal accounting stuff before submission. | |
1514 | */ | |
1515 | if (bio_has_data(bio)) { | |
1516 | if (rw & WRITE) { | |
1517 | count_vm_events(PGPGOUT, count); | |
1518 | } else { | |
1519 | task_io_account_read(bio->bi_size); | |
1520 | count_vm_events(PGPGIN, count); | |
1521 | } | |
1522 | ||
1523 | if (unlikely(block_dump)) { | |
1524 | char b[BDEVNAME_SIZE]; | |
1525 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
1526 | current->comm, task_pid_nr(current), | |
1527 | (rw & WRITE) ? "WRITE" : "READ", | |
1528 | (unsigned long long)bio->bi_sector, | |
1529 | bdevname(bio->bi_bdev, b)); | |
1530 | } | |
1531 | } | |
1532 | ||
1533 | generic_make_request(bio); | |
1534 | } | |
1535 | EXPORT_SYMBOL(submit_bio); | |
1536 | ||
1537 | /** | |
1538 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1539 | * @q: the queue | |
1540 | * @rq: the request being checked | |
1541 | * | |
1542 | * Description: | |
1543 | * @rq may have been made based on weaker limitations of upper-level queues | |
1544 | * in request stacking drivers, and it may violate the limitation of @q. | |
1545 | * Since the block layer and the underlying device driver trust @rq | |
1546 | * after it is inserted to @q, it should be checked against @q before | |
1547 | * the insertion using this generic function. | |
1548 | * | |
1549 | * This function should also be useful for request stacking drivers | |
1550 | * in some cases below, so export this fuction. | |
1551 | * Request stacking drivers like request-based dm may change the queue | |
1552 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1553 | * Such request stacking drivers should check those requests agaist | |
1554 | * the new queue limits again when they dispatch those requests, | |
1555 | * although such checkings are also done against the old queue limits | |
1556 | * when submitting requests. | |
1557 | */ | |
1558 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1559 | { | |
1560 | if (rq->nr_sectors > q->max_sectors || | |
1561 | rq->data_len > q->max_hw_sectors << 9) { | |
1562 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1563 | return -EIO; | |
1564 | } | |
1565 | ||
1566 | /* | |
1567 | * queue's settings related to segment counting like q->bounce_pfn | |
1568 | * may differ from that of other stacking queues. | |
1569 | * Recalculate it to check the request correctly on this queue's | |
1570 | * limitation. | |
1571 | */ | |
1572 | blk_recalc_rq_segments(rq); | |
1573 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1574 | rq->nr_phys_segments > q->max_hw_segments) { | |
1575 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1576 | return -EIO; | |
1577 | } | |
1578 | ||
1579 | return 0; | |
1580 | } | |
1581 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1582 | ||
1583 | /** | |
1584 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1585 | * @q: the queue to submit the request | |
1586 | * @rq: the request being queued | |
1587 | */ | |
1588 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1589 | { | |
1590 | unsigned long flags; | |
1591 | ||
1592 | if (blk_rq_check_limits(q, rq)) | |
1593 | return -EIO; | |
1594 | ||
1595 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1596 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1597 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1598 | return -EIO; | |
1599 | #endif | |
1600 | ||
1601 | spin_lock_irqsave(q->queue_lock, flags); | |
1602 | ||
1603 | /* | |
1604 | * Submitting request must be dequeued before calling this function | |
1605 | * because it will be linked to another request_queue | |
1606 | */ | |
1607 | BUG_ON(blk_queued_rq(rq)); | |
1608 | ||
1609 | drive_stat_acct(rq, 1); | |
1610 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1611 | ||
1612 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1613 | ||
1614 | return 0; | |
1615 | } | |
1616 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1617 | ||
1618 | /** | |
1619 | * blkdev_dequeue_request - dequeue request and start timeout timer | |
1620 | * @req: request to dequeue | |
1621 | * | |
1622 | * Dequeue @req and start timeout timer on it. This hands off the | |
1623 | * request to the driver. | |
1624 | * | |
1625 | * Block internal functions which don't want to start timer should | |
1626 | * call elv_dequeue_request(). | |
1627 | */ | |
1628 | void blkdev_dequeue_request(struct request *req) | |
1629 | { | |
1630 | elv_dequeue_request(req->q, req); | |
1631 | ||
1632 | /* | |
1633 | * We are now handing the request to the hardware, add the | |
1634 | * timeout handler. | |
1635 | */ | |
1636 | blk_add_timer(req); | |
1637 | } | |
1638 | EXPORT_SYMBOL(blkdev_dequeue_request); | |
1639 | ||
1640 | static void blk_account_io_completion(struct request *req, unsigned int bytes) | |
1641 | { | |
1642 | if (blk_do_io_stat(req)) { | |
1643 | const int rw = rq_data_dir(req); | |
1644 | struct hd_struct *part; | |
1645 | int cpu; | |
1646 | ||
1647 | cpu = part_stat_lock(); | |
1648 | part = disk_map_sector_rcu(req->rq_disk, req->sector); | |
1649 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
1650 | part_stat_unlock(); | |
1651 | } | |
1652 | } | |
1653 | ||
1654 | static void blk_account_io_done(struct request *req) | |
1655 | { | |
1656 | /* | |
1657 | * Account IO completion. bar_rq isn't accounted as a normal | |
1658 | * IO on queueing nor completion. Accounting the containing | |
1659 | * request is enough. | |
1660 | */ | |
1661 | if (blk_do_io_stat(req) && req != &req->q->bar_rq) { | |
1662 | unsigned long duration = jiffies - req->start_time; | |
1663 | const int rw = rq_data_dir(req); | |
1664 | struct hd_struct *part; | |
1665 | int cpu; | |
1666 | ||
1667 | cpu = part_stat_lock(); | |
1668 | part = disk_map_sector_rcu(req->rq_disk, req->sector); | |
1669 | ||
1670 | part_stat_inc(cpu, part, ios[rw]); | |
1671 | part_stat_add(cpu, part, ticks[rw], duration); | |
1672 | part_round_stats(cpu, part); | |
1673 | part_dec_in_flight(part); | |
1674 | ||
1675 | part_stat_unlock(); | |
1676 | } | |
1677 | } | |
1678 | ||
1679 | /** | |
1680 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
1681 | * @rq: the request being processed | |
1682 | **/ | |
1683 | unsigned int blk_rq_bytes(struct request *rq) | |
1684 | { | |
1685 | if (blk_fs_request(rq)) | |
1686 | return rq->hard_nr_sectors << 9; | |
1687 | ||
1688 | return rq->data_len; | |
1689 | } | |
1690 | EXPORT_SYMBOL_GPL(blk_rq_bytes); | |
1691 | ||
1692 | /** | |
1693 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
1694 | * @rq: the request being processed | |
1695 | **/ | |
1696 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1697 | { | |
1698 | if (blk_fs_request(rq)) | |
1699 | return rq->current_nr_sectors << 9; | |
1700 | ||
1701 | if (rq->bio) | |
1702 | return rq->bio->bi_size; | |
1703 | ||
1704 | return rq->data_len; | |
1705 | } | |
1706 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
1707 | ||
1708 | struct request *elv_next_request(struct request_queue *q) | |
1709 | { | |
1710 | struct request *rq; | |
1711 | int ret; | |
1712 | ||
1713 | while ((rq = __elv_next_request(q)) != NULL) { | |
1714 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1715 | /* | |
1716 | * This is the first time the device driver | |
1717 | * sees this request (possibly after | |
1718 | * requeueing). Notify IO scheduler. | |
1719 | */ | |
1720 | if (blk_sorted_rq(rq)) | |
1721 | elv_activate_rq(q, rq); | |
1722 | ||
1723 | /* | |
1724 | * just mark as started even if we don't start | |
1725 | * it, a request that has been delayed should | |
1726 | * not be passed by new incoming requests | |
1727 | */ | |
1728 | rq->cmd_flags |= REQ_STARTED; | |
1729 | trace_block_rq_issue(q, rq); | |
1730 | } | |
1731 | ||
1732 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1733 | q->end_sector = rq_end_sector(rq); | |
1734 | q->boundary_rq = NULL; | |
1735 | } | |
1736 | ||
1737 | if (rq->cmd_flags & REQ_DONTPREP) | |
1738 | break; | |
1739 | ||
1740 | if (q->dma_drain_size && rq->data_len) { | |
1741 | /* | |
1742 | * make sure space for the drain appears we | |
1743 | * know we can do this because max_hw_segments | |
1744 | * has been adjusted to be one fewer than the | |
1745 | * device can handle | |
1746 | */ | |
1747 | rq->nr_phys_segments++; | |
1748 | } | |
1749 | ||
1750 | if (!q->prep_rq_fn) | |
1751 | break; | |
1752 | ||
1753 | ret = q->prep_rq_fn(q, rq); | |
1754 | if (ret == BLKPREP_OK) { | |
1755 | break; | |
1756 | } else if (ret == BLKPREP_DEFER) { | |
1757 | /* | |
1758 | * the request may have been (partially) prepped. | |
1759 | * we need to keep this request in the front to | |
1760 | * avoid resource deadlock. REQ_STARTED will | |
1761 | * prevent other fs requests from passing this one. | |
1762 | */ | |
1763 | if (q->dma_drain_size && rq->data_len && | |
1764 | !(rq->cmd_flags & REQ_DONTPREP)) { | |
1765 | /* | |
1766 | * remove the space for the drain we added | |
1767 | * so that we don't add it again | |
1768 | */ | |
1769 | --rq->nr_phys_segments; | |
1770 | } | |
1771 | ||
1772 | rq = NULL; | |
1773 | break; | |
1774 | } else if (ret == BLKPREP_KILL) { | |
1775 | rq->cmd_flags |= REQ_QUIET; | |
1776 | __blk_end_request_all(rq, -EIO); | |
1777 | } else { | |
1778 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1779 | break; | |
1780 | } | |
1781 | } | |
1782 | ||
1783 | return rq; | |
1784 | } | |
1785 | EXPORT_SYMBOL(elv_next_request); | |
1786 | ||
1787 | void elv_dequeue_request(struct request_queue *q, struct request *rq) | |
1788 | { | |
1789 | BUG_ON(list_empty(&rq->queuelist)); | |
1790 | BUG_ON(ELV_ON_HASH(rq)); | |
1791 | ||
1792 | list_del_init(&rq->queuelist); | |
1793 | ||
1794 | /* | |
1795 | * the time frame between a request being removed from the lists | |
1796 | * and to it is freed is accounted as io that is in progress at | |
1797 | * the driver side. | |
1798 | */ | |
1799 | if (blk_account_rq(rq)) | |
1800 | q->in_flight++; | |
1801 | } | |
1802 | ||
1803 | /** | |
1804 | * blk_update_request - Special helper function for request stacking drivers | |
1805 | * @rq: the request being processed | |
1806 | * @error: %0 for success, < %0 for error | |
1807 | * @nr_bytes: number of bytes to complete @rq | |
1808 | * | |
1809 | * Description: | |
1810 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
1811 | * the request structure even if @rq doesn't have leftover. | |
1812 | * If @rq has leftover, sets it up for the next range of segments. | |
1813 | * | |
1814 | * This special helper function is only for request stacking drivers | |
1815 | * (e.g. request-based dm) so that they can handle partial completion. | |
1816 | * Actual device drivers should use blk_end_request instead. | |
1817 | * | |
1818 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1819 | * %false return from this function. | |
1820 | * | |
1821 | * Return: | |
1822 | * %false - this request doesn't have any more data | |
1823 | * %true - this request has more data | |
1824 | **/ | |
1825 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) | |
1826 | { | |
1827 | int total_bytes, bio_nbytes, next_idx = 0; | |
1828 | struct bio *bio; | |
1829 | ||
1830 | if (!req->bio) | |
1831 | return false; | |
1832 | ||
1833 | trace_block_rq_complete(req->q, req); | |
1834 | ||
1835 | /* | |
1836 | * For fs requests, rq is just carrier of independent bio's | |
1837 | * and each partial completion should be handled separately. | |
1838 | * Reset per-request error on each partial completion. | |
1839 | * | |
1840 | * TODO: tj: This is too subtle. It would be better to let | |
1841 | * low level drivers do what they see fit. | |
1842 | */ | |
1843 | if (blk_fs_request(req)) | |
1844 | req->errors = 0; | |
1845 | ||
1846 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { | |
1847 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1848 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
1849 | (unsigned long long)req->sector); | |
1850 | } | |
1851 | ||
1852 | blk_account_io_completion(req, nr_bytes); | |
1853 | ||
1854 | total_bytes = bio_nbytes = 0; | |
1855 | while ((bio = req->bio) != NULL) { | |
1856 | int nbytes; | |
1857 | ||
1858 | if (nr_bytes >= bio->bi_size) { | |
1859 | req->bio = bio->bi_next; | |
1860 | nbytes = bio->bi_size; | |
1861 | req_bio_endio(req, bio, nbytes, error); | |
1862 | next_idx = 0; | |
1863 | bio_nbytes = 0; | |
1864 | } else { | |
1865 | int idx = bio->bi_idx + next_idx; | |
1866 | ||
1867 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1868 | blk_dump_rq_flags(req, "__end_that"); | |
1869 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", | |
1870 | __func__, bio->bi_idx, bio->bi_vcnt); | |
1871 | break; | |
1872 | } | |
1873 | ||
1874 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1875 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1876 | ||
1877 | /* | |
1878 | * not a complete bvec done | |
1879 | */ | |
1880 | if (unlikely(nbytes > nr_bytes)) { | |
1881 | bio_nbytes += nr_bytes; | |
1882 | total_bytes += nr_bytes; | |
1883 | break; | |
1884 | } | |
1885 | ||
1886 | /* | |
1887 | * advance to the next vector | |
1888 | */ | |
1889 | next_idx++; | |
1890 | bio_nbytes += nbytes; | |
1891 | } | |
1892 | ||
1893 | total_bytes += nbytes; | |
1894 | nr_bytes -= nbytes; | |
1895 | ||
1896 | bio = req->bio; | |
1897 | if (bio) { | |
1898 | /* | |
1899 | * end more in this run, or just return 'not-done' | |
1900 | */ | |
1901 | if (unlikely(nr_bytes <= 0)) | |
1902 | break; | |
1903 | } | |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * completely done | |
1908 | */ | |
1909 | if (!req->bio) { | |
1910 | /* | |
1911 | * Reset counters so that the request stacking driver | |
1912 | * can find how many bytes remain in the request | |
1913 | * later. | |
1914 | */ | |
1915 | req->nr_sectors = req->hard_nr_sectors = 0; | |
1916 | req->current_nr_sectors = req->hard_cur_sectors = 0; | |
1917 | return false; | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * if the request wasn't completed, update state | |
1922 | */ | |
1923 | if (bio_nbytes) { | |
1924 | req_bio_endio(req, bio, bio_nbytes, error); | |
1925 | bio->bi_idx += next_idx; | |
1926 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1927 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1928 | } | |
1929 | ||
1930 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1931 | blk_recalc_rq_segments(req); | |
1932 | return true; | |
1933 | } | |
1934 | EXPORT_SYMBOL_GPL(blk_update_request); | |
1935 | ||
1936 | static bool blk_update_bidi_request(struct request *rq, int error, | |
1937 | unsigned int nr_bytes, | |
1938 | unsigned int bidi_bytes) | |
1939 | { | |
1940 | if (blk_update_request(rq, error, nr_bytes)) | |
1941 | return true; | |
1942 | ||
1943 | /* Bidi request must be completed as a whole */ | |
1944 | if (unlikely(blk_bidi_rq(rq)) && | |
1945 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
1946 | return true; | |
1947 | ||
1948 | add_disk_randomness(rq->rq_disk); | |
1949 | ||
1950 | return false; | |
1951 | } | |
1952 | ||
1953 | /* | |
1954 | * queue lock must be held | |
1955 | */ | |
1956 | static void blk_finish_request(struct request *req, int error) | |
1957 | { | |
1958 | if (blk_rq_tagged(req)) | |
1959 | blk_queue_end_tag(req->q, req); | |
1960 | ||
1961 | if (blk_queued_rq(req)) | |
1962 | elv_dequeue_request(req->q, req); | |
1963 | ||
1964 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1965 | laptop_io_completion(); | |
1966 | ||
1967 | blk_delete_timer(req); | |
1968 | ||
1969 | blk_account_io_done(req); | |
1970 | ||
1971 | if (req->end_io) | |
1972 | req->end_io(req, error); | |
1973 | else { | |
1974 | if (blk_bidi_rq(req)) | |
1975 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1976 | ||
1977 | __blk_put_request(req->q, req); | |
1978 | } | |
1979 | } | |
1980 | ||
1981 | /** | |
1982 | * blk_end_bidi_request - Complete a bidi request | |
1983 | * @rq: the request to complete | |
1984 | * @error: %0 for success, < %0 for error | |
1985 | * @nr_bytes: number of bytes to complete @rq | |
1986 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1987 | * | |
1988 | * Description: | |
1989 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1990 | * Drivers that supports bidi can safely call this member for any | |
1991 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
1992 | * just ignored. | |
1993 | * | |
1994 | * Return: | |
1995 | * %false - we are done with this request | |
1996 | * %true - still buffers pending for this request | |
1997 | **/ | |
1998 | bool blk_end_bidi_request(struct request *rq, int error, | |
1999 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
2000 | { | |
2001 | struct request_queue *q = rq->q; | |
2002 | unsigned long flags; | |
2003 | ||
2004 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
2005 | return true; | |
2006 | ||
2007 | spin_lock_irqsave(q->queue_lock, flags); | |
2008 | blk_finish_request(rq, error); | |
2009 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2010 | ||
2011 | return false; | |
2012 | } | |
2013 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2014 | ||
2015 | /** | |
2016 | * __blk_end_bidi_request - Complete a bidi request with queue lock held | |
2017 | * @rq: the request to complete | |
2018 | * @error: %0 for success, < %0 for error | |
2019 | * @nr_bytes: number of bytes to complete @rq | |
2020 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2021 | * | |
2022 | * Description: | |
2023 | * Identical to blk_end_bidi_request() except that queue lock is | |
2024 | * assumed to be locked on entry and remains so on return. | |
2025 | * | |
2026 | * Return: | |
2027 | * %false - we are done with this request | |
2028 | * %true - still buffers pending for this request | |
2029 | **/ | |
2030 | bool __blk_end_bidi_request(struct request *rq, int error, | |
2031 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
2032 | { | |
2033 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
2034 | return true; | |
2035 | ||
2036 | blk_finish_request(rq, error); | |
2037 | ||
2038 | return false; | |
2039 | } | |
2040 | EXPORT_SYMBOL_GPL(__blk_end_bidi_request); | |
2041 | ||
2042 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
2043 | struct bio *bio) | |
2044 | { | |
2045 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and | |
2046 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
2047 | rq->cmd_flags |= (bio->bi_rw & 3); | |
2048 | ||
2049 | if (bio_has_data(bio)) { | |
2050 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
2051 | rq->buffer = bio_data(bio); | |
2052 | } | |
2053 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
2054 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2055 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
2056 | rq->data_len = bio->bi_size; | |
2057 | ||
2058 | rq->bio = rq->biotail = bio; | |
2059 | ||
2060 | if (bio->bi_bdev) | |
2061 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2062 | } | |
2063 | ||
2064 | /** | |
2065 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2066 | * @q : the queue of the device being checked | |
2067 | * | |
2068 | * Description: | |
2069 | * Check if underlying low-level drivers of a device are busy. | |
2070 | * If the drivers want to export their busy state, they must set own | |
2071 | * exporting function using blk_queue_lld_busy() first. | |
2072 | * | |
2073 | * Basically, this function is used only by request stacking drivers | |
2074 | * to stop dispatching requests to underlying devices when underlying | |
2075 | * devices are busy. This behavior helps more I/O merging on the queue | |
2076 | * of the request stacking driver and prevents I/O throughput regression | |
2077 | * on burst I/O load. | |
2078 | * | |
2079 | * Return: | |
2080 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2081 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2082 | */ | |
2083 | int blk_lld_busy(struct request_queue *q) | |
2084 | { | |
2085 | if (q->lld_busy_fn) | |
2086 | return q->lld_busy_fn(q); | |
2087 | ||
2088 | return 0; | |
2089 | } | |
2090 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2091 | ||
2092 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) | |
2093 | { | |
2094 | return queue_work(kblockd_workqueue, work); | |
2095 | } | |
2096 | EXPORT_SYMBOL(kblockd_schedule_work); | |
2097 | ||
2098 | int __init blk_dev_init(void) | |
2099 | { | |
2100 | kblockd_workqueue = create_workqueue("kblockd"); | |
2101 | if (!kblockd_workqueue) | |
2102 | panic("Failed to create kblockd\n"); | |
2103 | ||
2104 | request_cachep = kmem_cache_create("blkdev_requests", | |
2105 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
2106 | ||
2107 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", | |
2108 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
2109 | ||
2110 | return 0; | |
2111 | } | |
2112 |