]>
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 | ||
32 | #include "blk.h" | |
33 | ||
34 | static int __make_request(struct request_queue *q, struct bio *bio); | |
35 | ||
36 | /* | |
37 | * For the allocated request tables | |
38 | */ | |
39 | static struct kmem_cache *request_cachep; | |
40 | ||
41 | /* | |
42 | * For queue allocation | |
43 | */ | |
44 | struct kmem_cache *blk_requestq_cachep; | |
45 | ||
46 | /* | |
47 | * Controlling structure to kblockd | |
48 | */ | |
49 | static struct workqueue_struct *kblockd_workqueue; | |
50 | ||
51 | static void drive_stat_acct(struct request *rq, int new_io) | |
52 | { | |
53 | struct hd_struct *part; | |
54 | int rw = rq_data_dir(rq); | |
55 | int cpu; | |
56 | ||
57 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
58 | return; | |
59 | ||
60 | cpu = part_stat_lock(); | |
61 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); | |
62 | ||
63 | if (!new_io) | |
64 | part_stat_inc(cpu, part, merges[rw]); | |
65 | else { | |
66 | part_round_stats(cpu, part); | |
67 | part_inc_in_flight(part); | |
68 | } | |
69 | ||
70 | part_stat_unlock(); | |
71 | } | |
72 | ||
73 | void blk_queue_congestion_threshold(struct request_queue *q) | |
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 | ||
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; | |
100 | struct request_queue *q = bdev_get_queue(bdev); | |
101 | ||
102 | if (q) | |
103 | ret = &q->backing_dev_info; | |
104 | return ret; | |
105 | } | |
106 | EXPORT_SYMBOL(blk_get_backing_dev_info); | |
107 | ||
108 | void blk_rq_init(struct request_queue *q, struct request *rq) | |
109 | { | |
110 | memset(rq, 0, sizeof(*rq)); | |
111 | ||
112 | INIT_LIST_HEAD(&rq->queuelist); | |
113 | INIT_LIST_HEAD(&rq->timeout_list); | |
114 | rq->cpu = -1; | |
115 | rq->q = q; | |
116 | rq->sector = rq->hard_sector = (sector_t) -1; | |
117 | INIT_HLIST_NODE(&rq->hash); | |
118 | RB_CLEAR_NODE(&rq->rb_node); | |
119 | rq->cmd = rq->__cmd; | |
120 | rq->tag = -1; | |
121 | rq->ref_count = 1; | |
122 | } | |
123 | EXPORT_SYMBOL(blk_rq_init); | |
124 | ||
125 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
126 | unsigned int nbytes, int error) | |
127 | { | |
128 | struct request_queue *q = rq->q; | |
129 | ||
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; | |
135 | ||
136 | if (unlikely(nbytes > bio->bi_size)) { | |
137 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
138 | __func__, nbytes, bio->bi_size); | |
139 | nbytes = bio->bi_size; | |
140 | } | |
141 | ||
142 | bio->bi_size -= nbytes; | |
143 | bio->bi_sector += (nbytes >> 9); | |
144 | ||
145 | if (bio_integrity(bio)) | |
146 | bio_integrity_advance(bio, nbytes); | |
147 | ||
148 | if (bio->bi_size == 0) | |
149 | bio_endio(bio, error); | |
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 | } | |
159 | } | |
160 | ||
161 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
162 | { | |
163 | int bit; | |
164 | ||
165 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, | |
166 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
167 | rq->cmd_flags); | |
168 | ||
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); | |
177 | ||
178 | if (blk_pc_request(rq)) { | |
179 | printk(KERN_INFO " cdb: "); | |
180 | for (bit = 0; bit < BLK_MAX_CDB; bit++) | |
181 | printk("%02x ", rq->cmd[bit]); | |
182 | printk("\n"); | |
183 | } | |
184 | } | |
185 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
186 | ||
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 | */ | |
195 | void blk_plug_device(struct request_queue *q) | |
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 | */ | |
203 | if (blk_queue_stopped(q)) | |
204 | return; | |
205 | ||
206 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { | |
207 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | |
208 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); | |
209 | } | |
210 | } | |
211 | EXPORT_SYMBOL(blk_plug_device); | |
212 | ||
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 | ||
231 | /* | |
232 | * remove the queue from the plugged list, if present. called with | |
233 | * queue lock held and interrupts disabled. | |
234 | */ | |
235 | int blk_remove_plug(struct request_queue *q) | |
236 | { | |
237 | WARN_ON(!irqs_disabled()); | |
238 | ||
239 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) | |
240 | return 0; | |
241 | ||
242 | del_timer(&q->unplug_timer); | |
243 | return 1; | |
244 | } | |
245 | EXPORT_SYMBOL(blk_remove_plug); | |
246 | ||
247 | /* | |
248 | * remove the plug and let it rip.. | |
249 | */ | |
250 | void __generic_unplug_device(struct request_queue *q) | |
251 | { | |
252 | if (unlikely(blk_queue_stopped(q))) | |
253 | return; | |
254 | ||
255 | if (!blk_remove_plug(q)) | |
256 | return; | |
257 | ||
258 | q->request_fn(q); | |
259 | } | |
260 | EXPORT_SYMBOL(__generic_unplug_device); | |
261 | ||
262 | /** | |
263 | * generic_unplug_device - fire a request queue | |
264 | * @q: The &struct request_queue in question | |
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 | **/ | |
273 | void generic_unplug_device(struct request_queue *q) | |
274 | { | |
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 | } | |
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 | { | |
286 | struct request_queue *q = bdi->unplug_io_data; | |
287 | ||
288 | blk_unplug(q); | |
289 | } | |
290 | ||
291 | void blk_unplug_work(struct work_struct *work) | |
292 | { | |
293 | struct request_queue *q = | |
294 | container_of(work, struct request_queue, unplug_work); | |
295 | ||
296 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
297 | q->rq.count[READ] + q->rq.count[WRITE]); | |
298 | ||
299 | q->unplug_fn(q); | |
300 | } | |
301 | ||
302 | void blk_unplug_timeout(unsigned long data) | |
303 | { | |
304 | struct request_queue *q = (struct request_queue *)data; | |
305 | ||
306 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, | |
307 | q->rq.count[READ] + q->rq.count[WRITE]); | |
308 | ||
309 | kblockd_schedule_work(q, &q->unplug_work); | |
310 | } | |
311 | ||
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 | ||
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 | ||
341 | /** | |
342 | * blk_start_queue - restart a previously stopped queue | |
343 | * @q: The &struct request_queue in question | |
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 | **/ | |
350 | void blk_start_queue(struct request_queue *q) | |
351 | { | |
352 | WARN_ON(!irqs_disabled()); | |
353 | ||
354 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); | |
355 | blk_invoke_request_fn(q); | |
356 | } | |
357 | EXPORT_SYMBOL(blk_start_queue); | |
358 | ||
359 | /** | |
360 | * blk_stop_queue - stop a queue | |
361 | * @q: The &struct request_queue in question | |
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 | **/ | |
373 | void blk_stop_queue(struct request_queue *q) | |
374 | { | |
375 | blk_remove_plug(q); | |
376 | queue_flag_set(QUEUE_FLAG_STOPPED, q); | |
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 | |
389 | * that the callbacks might use. The caller must already have made sure | |
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); | |
397 | kblockd_flush_work(&q->unplug_work); | |
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 | */ | |
405 | void __blk_run_queue(struct request_queue *q) | |
406 | { | |
407 | blk_remove_plug(q); | |
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 | */ | |
413 | if (!elv_queue_empty(q)) | |
414 | blk_invoke_request_fn(q); | |
415 | } | |
416 | EXPORT_SYMBOL(__blk_run_queue); | |
417 | ||
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); | |
428 | spin_unlock_irqrestore(q->queue_lock, flags); | |
429 | } | |
430 | EXPORT_SYMBOL(blk_run_queue); | |
431 | ||
432 | void blk_put_queue(struct request_queue *q) | |
433 | { | |
434 | kobject_put(&q->kobj); | |
435 | } | |
436 | ||
437 | void blk_cleanup_queue(struct request_queue *q) | |
438 | { | |
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 | ||
447 | mutex_lock(&q->sysfs_lock); | |
448 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); | |
449 | mutex_unlock(&q->sysfs_lock); | |
450 | ||
451 | if (q->elevator) | |
452 | elevator_exit(q->elevator); | |
453 | ||
454 | blk_put_queue(q); | |
455 | } | |
456 | EXPORT_SYMBOL(blk_cleanup_queue); | |
457 | ||
458 | static int blk_init_free_list(struct request_queue *q) | |
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; | |
464 | rl->elvpriv = 0; | |
465 | init_waitqueue_head(&rl->wait[READ]); | |
466 | init_waitqueue_head(&rl->wait[WRITE]); | |
467 | ||
468 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | |
469 | mempool_free_slab, request_cachep, q->node); | |
470 | ||
471 | if (!rl->rq_pool) | |
472 | return -ENOMEM; | |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
478 | { | |
479 | return blk_alloc_queue_node(gfp_mask, -1); | |
480 | } | |
481 | EXPORT_SYMBOL(blk_alloc_queue); | |
482 | ||
483 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
484 | { | |
485 | struct request_queue *q; | |
486 | int err; | |
487 | ||
488 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
489 | gfp_mask | __GFP_ZERO, node_id); | |
490 | if (!q) | |
491 | return NULL; | |
492 | ||
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) { | |
497 | kmem_cache_free(blk_requestq_cachep, q); | |
498 | return NULL; | |
499 | } | |
500 | ||
501 | init_timer(&q->unplug_timer); | |
502 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); | |
503 | INIT_LIST_HEAD(&q->timeout_list); | |
504 | ||
505 | kobject_init(&q->kobj, &blk_queue_ktype); | |
506 | ||
507 | mutex_init(&q->sysfs_lock); | |
508 | spin_lock_init(&q->__queue_lock); | |
509 | ||
510 | return q; | |
511 | } | |
512 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
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 | |
536 | * request queue; this lock will be taken also from interrupt context, so irq | |
537 | * disabling is needed for it. | |
538 | * | |
539 | * Function returns a pointer to the initialized request queue, or %NULL if | |
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 | **/ | |
546 | ||
547 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
548 | { | |
549 | return blk_init_queue_node(rfn, lock, -1); | |
550 | } | |
551 | EXPORT_SYMBOL(blk_init_queue); | |
552 | ||
553 | struct request_queue * | |
554 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
555 | { | |
556 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
557 | ||
558 | if (!q) | |
559 | return NULL; | |
560 | ||
561 | q->node = node_id; | |
562 | if (blk_init_free_list(q)) { | |
563 | kmem_cache_free(blk_requestq_cachep, q); | |
564 | return NULL; | |
565 | } | |
566 | ||
567 | /* | |
568 | * if caller didn't supply a lock, they get per-queue locking with | |
569 | * our embedded lock | |
570 | */ | |
571 | if (!lock) | |
572 | lock = &q->__queue_lock; | |
573 | ||
574 | q->request_fn = rfn; | |
575 | q->prep_rq_fn = NULL; | |
576 | q->unplug_fn = generic_unplug_device; | |
577 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | | |
578 | 1 << QUEUE_FLAG_STACKABLE); | |
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 | ||
589 | q->sg_reserved_size = INT_MAX; | |
590 | ||
591 | blk_set_cmd_filter_defaults(&q->cmd_filter); | |
592 | ||
593 | /* | |
594 | * all done | |
595 | */ | |
596 | if (!elevator_init(q, NULL)) { | |
597 | blk_queue_congestion_threshold(q); | |
598 | return q; | |
599 | } | |
600 | ||
601 | blk_put_queue(q); | |
602 | return NULL; | |
603 | } | |
604 | EXPORT_SYMBOL(blk_init_queue_node); | |
605 | ||
606 | int blk_get_queue(struct request_queue *q) | |
607 | { | |
608 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { | |
609 | kobject_get(&q->kobj); | |
610 | return 0; | |
611 | } | |
612 | ||
613 | return 1; | |
614 | } | |
615 | ||
616 | static inline void blk_free_request(struct request_queue *q, struct request *rq) | |
617 | { | |
618 | if (rq->cmd_flags & REQ_ELVPRIV) | |
619 | elv_put_request(q, rq); | |
620 | mempool_free(rq, q->rq.rq_pool); | |
621 | } | |
622 | ||
623 | static struct request * | |
624 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) | |
625 | { | |
626 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
627 | ||
628 | if (!rq) | |
629 | return NULL; | |
630 | ||
631 | blk_rq_init(q, rq); | |
632 | ||
633 | rq->cmd_flags = rw | REQ_ALLOCED; | |
634 | ||
635 | if (priv) { | |
636 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { | |
637 | mempool_free(rq, q->rq.rq_pool); | |
638 | return NULL; | |
639 | } | |
640 | rq->cmd_flags |= REQ_ELVPRIV; | |
641 | } | |
642 | ||
643 | return rq; | |
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 | */ | |
650 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
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 | */ | |
671 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
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 | ||
680 | static void __freed_request(struct request_queue *q, int rw) | |
681 | { | |
682 | struct request_list *rl = &q->rq; | |
683 | ||
684 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
685 | blk_clear_queue_congested(q, rw); | |
686 | ||
687 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
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 | */ | |
699 | static void freed_request(struct request_queue *q, int rw, int priv) | |
700 | { | |
701 | struct request_list *rl = &q->rq; | |
702 | ||
703 | rl->count[rw]--; | |
704 | if (priv) | |
705 | rl->elvpriv--; | |
706 | ||
707 | __freed_request(q, rw); | |
708 | ||
709 | if (unlikely(rl->starved[rw ^ 1])) | |
710 | __freed_request(q, rw ^ 1); | |
711 | } | |
712 | ||
713 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
714 | /* | |
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*. | |
718 | */ | |
719 | static struct request *get_request(struct request_queue *q, int rw_flags, | |
720 | struct bio *bio, gfp_t gfp_mask) | |
721 | { | |
722 | struct request *rq = NULL; | |
723 | struct request_list *rl = &q->rq; | |
724 | struct io_context *ioc = NULL; | |
725 | const int rw = rw_flags & 0x01; | |
726 | int may_queue, priv; | |
727 | ||
728 | may_queue = elv_may_queue(q, rw_flags); | |
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) { | |
734 | ioc = current_io_context(GFP_ATOMIC, q->node); | |
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 | } | |
755 | } | |
756 | blk_set_queue_congested(q, rw); | |
757 | } | |
758 | ||
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 | */ | |
764 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) | |
765 | goto out; | |
766 | ||
767 | rl->count[rw]++; | |
768 | rl->starved[rw] = 0; | |
769 | ||
770 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
771 | if (priv) | |
772 | rl->elvpriv++; | |
773 | ||
774 | spin_unlock_irq(q->queue_lock); | |
775 | ||
776 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); | |
777 | if (unlikely(!rq)) { | |
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); | |
786 | freed_request(q, rw, priv); | |
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 | ||
799 | goto out; | |
800 | } | |
801 | ||
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 | */ | |
808 | if (ioc_batching(q, ioc)) | |
809 | ioc->nr_batch_requests--; | |
810 | ||
811 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
812 | out: | |
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. | |
819 | * | |
820 | * Called with q->queue_lock held, and returns with it unlocked. | |
821 | */ | |
822 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, | |
823 | struct bio *bio) | |
824 | { | |
825 | const int rw = rw_flags & 0x01; | |
826 | struct request *rq; | |
827 | ||
828 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
829 | while (!rq) { | |
830 | DEFINE_WAIT(wait); | |
831 | struct io_context *ioc; | |
832 | struct request_list *rl = &q->rq; | |
833 | ||
834 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
835 | TASK_UNINTERRUPTIBLE); | |
836 | ||
837 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); | |
838 | ||
839 | __generic_unplug_device(q); | |
840 | spin_unlock_irq(q->queue_lock); | |
841 | io_schedule(); | |
842 | ||
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); | |
851 | ||
852 | spin_lock_irq(q->queue_lock); | |
853 | finish_wait(&rl->wait[rw], &wait); | |
854 | ||
855 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
856 | }; | |
857 | ||
858 | return rq; | |
859 | } | |
860 | ||
861 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) | |
862 | { | |
863 | struct request *rq; | |
864 | ||
865 | BUG_ON(rw != READ && rw != WRITE); | |
866 | ||
867 | spin_lock_irq(q->queue_lock); | |
868 | if (gfp_mask & __GFP_WAIT) { | |
869 | rq = get_request_wait(q, rw, NULL); | |
870 | } else { | |
871 | rq = get_request(q, rw, NULL, gfp_mask); | |
872 | if (!rq) | |
873 | spin_unlock_irq(q->queue_lock); | |
874 | } | |
875 | /* q->queue_lock is unlocked at this point */ | |
876 | ||
877 | return rq; | |
878 | } | |
879 | EXPORT_SYMBOL(blk_get_request); | |
880 | ||
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 | */ | |
891 | void blk_start_queueing(struct request_queue *q) | |
892 | { | |
893 | if (!blk_queue_plugged(q)) | |
894 | q->request_fn(q); | |
895 | else | |
896 | __generic_unplug_device(q); | |
897 | } | |
898 | EXPORT_SYMBOL(blk_start_queueing); | |
899 | ||
900 | /** | |
901 | * blk_requeue_request - put a request back on queue | |
902 | * @q: request queue where request should be inserted | |
903 | * @rq: request to be inserted | |
904 | * | |
905 | * Description: | |
906 | * Drivers often keep queueing requests until the hardware cannot accept | |
907 | * more, when that condition happens we need to put the request back | |
908 | * on the queue. Must be called with queue lock held. | |
909 | */ | |
910 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
911 | { | |
912 | blk_delete_timer(rq); | |
913 | blk_clear_rq_complete(rq); | |
914 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); | |
915 | ||
916 | if (blk_rq_tagged(rq)) | |
917 | blk_queue_end_tag(q, rq); | |
918 | ||
919 | elv_requeue_request(q, rq); | |
920 | } | |
921 | EXPORT_SYMBOL(blk_requeue_request); | |
922 | ||
923 | /** | |
924 | * blk_insert_request - insert a special request into a request queue | |
925 | * @q: request queue where request should be inserted | |
926 | * @rq: request to be inserted | |
927 | * @at_head: insert request at head or tail of queue | |
928 | * @data: private data | |
929 | * | |
930 | * Description: | |
931 | * Many block devices need to execute commands asynchronously, so they don't | |
932 | * block the whole kernel from preemption during request execution. This is | |
933 | * accomplished normally by inserting aritficial requests tagged as | |
934 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them | |
935 | * be scheduled for actual execution by the request queue. | |
936 | * | |
937 | * We have the option of inserting the head or the tail of the queue. | |
938 | * Typically we use the tail for new ioctls and so forth. We use the head | |
939 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
940 | * host that is unable to accept a particular command. | |
941 | */ | |
942 | void blk_insert_request(struct request_queue *q, struct request *rq, | |
943 | int at_head, void *data) | |
944 | { | |
945 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
946 | unsigned long flags; | |
947 | ||
948 | /* | |
949 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
950 | * must not attempt merges on this) and that it acts as a soft | |
951 | * barrier | |
952 | */ | |
953 | rq->cmd_type = REQ_TYPE_SPECIAL; | |
954 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
955 | ||
956 | rq->special = data; | |
957 | ||
958 | spin_lock_irqsave(q->queue_lock, flags); | |
959 | ||
960 | /* | |
961 | * If command is tagged, release the tag | |
962 | */ | |
963 | if (blk_rq_tagged(rq)) | |
964 | blk_queue_end_tag(q, rq); | |
965 | ||
966 | drive_stat_acct(rq, 1); | |
967 | __elv_add_request(q, rq, where, 0); | |
968 | blk_start_queueing(q); | |
969 | spin_unlock_irqrestore(q->queue_lock, flags); | |
970 | } | |
971 | EXPORT_SYMBOL(blk_insert_request); | |
972 | ||
973 | /* | |
974 | * add-request adds a request to the linked list. | |
975 | * queue lock is held and interrupts disabled, as we muck with the | |
976 | * request queue list. | |
977 | */ | |
978 | static inline void add_request(struct request_queue *q, struct request *req) | |
979 | { | |
980 | drive_stat_acct(req, 1); | |
981 | ||
982 | /* | |
983 | * elevator indicated where it wants this request to be | |
984 | * inserted at elevator_merge time | |
985 | */ | |
986 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
987 | } | |
988 | ||
989 | static void part_round_stats_single(int cpu, struct hd_struct *part, | |
990 | unsigned long now) | |
991 | { | |
992 | if (now == part->stamp) | |
993 | return; | |
994 | ||
995 | if (part->in_flight) { | |
996 | __part_stat_add(cpu, part, time_in_queue, | |
997 | part->in_flight * (now - part->stamp)); | |
998 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
999 | } | |
1000 | part->stamp = now; | |
1001 | } | |
1002 | ||
1003 | /** | |
1004 | * part_round_stats() - Round off the performance stats on a struct | |
1005 | * disk_stats. | |
1006 | * | |
1007 | * The average IO queue length and utilisation statistics are maintained | |
1008 | * by observing the current state of the queue length and the amount of | |
1009 | * time it has been in this state for. | |
1010 | * | |
1011 | * Normally, that accounting is done on IO completion, but that can result | |
1012 | * in more than a second's worth of IO being accounted for within any one | |
1013 | * second, leading to >100% utilisation. To deal with that, we call this | |
1014 | * function to do a round-off before returning the results when reading | |
1015 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1016 | * the current jiffies and restarts the counters again. | |
1017 | */ | |
1018 | void part_round_stats(int cpu, struct hd_struct *part) | |
1019 | { | |
1020 | unsigned long now = jiffies; | |
1021 | ||
1022 | if (part->partno) | |
1023 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1024 | part_round_stats_single(cpu, part, now); | |
1025 | } | |
1026 | EXPORT_SYMBOL_GPL(part_round_stats); | |
1027 | ||
1028 | /* | |
1029 | * queue lock must be held | |
1030 | */ | |
1031 | void __blk_put_request(struct request_queue *q, struct request *req) | |
1032 | { | |
1033 | if (unlikely(!q)) | |
1034 | return; | |
1035 | if (unlikely(--req->ref_count)) | |
1036 | return; | |
1037 | ||
1038 | elv_completed_request(q, req); | |
1039 | ||
1040 | /* | |
1041 | * Request may not have originated from ll_rw_blk. if not, | |
1042 | * it didn't come out of our reserved rq pools | |
1043 | */ | |
1044 | if (req->cmd_flags & REQ_ALLOCED) { | |
1045 | int rw = rq_data_dir(req); | |
1046 | int priv = req->cmd_flags & REQ_ELVPRIV; | |
1047 | ||
1048 | BUG_ON(!list_empty(&req->queuelist)); | |
1049 | BUG_ON(!hlist_unhashed(&req->hash)); | |
1050 | ||
1051 | blk_free_request(q, req); | |
1052 | freed_request(q, rw, priv); | |
1053 | } | |
1054 | } | |
1055 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
1056 | ||
1057 | void blk_put_request(struct request *req) | |
1058 | { | |
1059 | unsigned long flags; | |
1060 | struct request_queue *q = req->q; | |
1061 | ||
1062 | spin_lock_irqsave(q->queue_lock, flags); | |
1063 | __blk_put_request(q, req); | |
1064 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1065 | } | |
1066 | EXPORT_SYMBOL(blk_put_request); | |
1067 | ||
1068 | void init_request_from_bio(struct request *req, struct bio *bio) | |
1069 | { | |
1070 | req->cpu = bio->bi_comp_cpu; | |
1071 | req->cmd_type = REQ_TYPE_FS; | |
1072 | ||
1073 | /* | |
1074 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1075 | */ | |
1076 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
1077 | req->cmd_flags |= REQ_FAILFAST; | |
1078 | ||
1079 | /* | |
1080 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1081 | */ | |
1082 | if (unlikely(bio_discard(bio))) { | |
1083 | req->cmd_flags |= REQ_DISCARD; | |
1084 | if (bio_barrier(bio)) | |
1085 | req->cmd_flags |= REQ_SOFTBARRIER; | |
1086 | req->q->prepare_discard_fn(req->q, req); | |
1087 | } else if (unlikely(bio_barrier(bio))) | |
1088 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
1089 | ||
1090 | if (bio_sync(bio)) | |
1091 | req->cmd_flags |= REQ_RW_SYNC; | |
1092 | if (bio_rw_meta(bio)) | |
1093 | req->cmd_flags |= REQ_RW_META; | |
1094 | ||
1095 | req->errors = 0; | |
1096 | req->hard_sector = req->sector = bio->bi_sector; | |
1097 | req->ioprio = bio_prio(bio); | |
1098 | req->start_time = jiffies; | |
1099 | blk_rq_bio_prep(req->q, req, bio); | |
1100 | } | |
1101 | ||
1102 | static int __make_request(struct request_queue *q, struct bio *bio) | |
1103 | { | |
1104 | struct request *req; | |
1105 | int el_ret, nr_sectors, barrier, discard, err; | |
1106 | const unsigned short prio = bio_prio(bio); | |
1107 | const int sync = bio_sync(bio); | |
1108 | int rw_flags; | |
1109 | ||
1110 | nr_sectors = bio_sectors(bio); | |
1111 | ||
1112 | /* | |
1113 | * low level driver can indicate that it wants pages above a | |
1114 | * certain limit bounced to low memory (ie for highmem, or even | |
1115 | * ISA dma in theory) | |
1116 | */ | |
1117 | blk_queue_bounce(q, &bio); | |
1118 | ||
1119 | barrier = bio_barrier(bio); | |
1120 | if (unlikely(barrier) && bio_has_data(bio) && | |
1121 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1122 | err = -EOPNOTSUPP; | |
1123 | goto end_io; | |
1124 | } | |
1125 | ||
1126 | discard = bio_discard(bio); | |
1127 | if (unlikely(discard) && !q->prepare_discard_fn) { | |
1128 | err = -EOPNOTSUPP; | |
1129 | goto end_io; | |
1130 | } | |
1131 | ||
1132 | spin_lock_irq(q->queue_lock); | |
1133 | ||
1134 | if (unlikely(barrier) || 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 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); | |
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 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); | |
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 (elv_queue_empty(q)) | |
1221 | blk_plug_device(q); | |
1222 | add_request(q, req); | |
1223 | out: | |
1224 | if (sync) | |
1225 | __generic_unplug_device(q); | |
1226 | spin_unlock_irq(q->queue_lock); | |
1227 | return 0; | |
1228 | ||
1229 | end_io: | |
1230 | bio_endio(bio, err); | |
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * If bio->bi_dev is a partition, remap the location | |
1236 | */ | |
1237 | static inline void blk_partition_remap(struct bio *bio) | |
1238 | { | |
1239 | struct block_device *bdev = bio->bi_bdev; | |
1240 | ||
1241 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { | |
1242 | struct hd_struct *p = bdev->bd_part; | |
1243 | ||
1244 | bio->bi_sector += p->start_sect; | |
1245 | bio->bi_bdev = bdev->bd_contains; | |
1246 | ||
1247 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1248 | bdev->bd_dev, bio->bi_sector, | |
1249 | bio->bi_sector - p->start_sect); | |
1250 | } | |
1251 | } | |
1252 | ||
1253 | static void handle_bad_sector(struct bio *bio) | |
1254 | { | |
1255 | char b[BDEVNAME_SIZE]; | |
1256 | ||
1257 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1258 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1259 | bdevname(bio->bi_bdev, b), | |
1260 | bio->bi_rw, | |
1261 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1262 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1263 | ||
1264 | set_bit(BIO_EOF, &bio->bi_flags); | |
1265 | } | |
1266 | ||
1267 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1268 | ||
1269 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1270 | ||
1271 | static int __init setup_fail_make_request(char *str) | |
1272 | { | |
1273 | return setup_fault_attr(&fail_make_request, str); | |
1274 | } | |
1275 | __setup("fail_make_request=", setup_fail_make_request); | |
1276 | ||
1277 | static int should_fail_request(struct bio *bio) | |
1278 | { | |
1279 | struct hd_struct *part = bio->bi_bdev->bd_part; | |
1280 | ||
1281 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
1282 | return should_fail(&fail_make_request, bio->bi_size); | |
1283 | ||
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | static int __init fail_make_request_debugfs(void) | |
1288 | { | |
1289 | return init_fault_attr_dentries(&fail_make_request, | |
1290 | "fail_make_request"); | |
1291 | } | |
1292 | ||
1293 | late_initcall(fail_make_request_debugfs); | |
1294 | ||
1295 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1296 | ||
1297 | static inline int should_fail_request(struct bio *bio) | |
1298 | { | |
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1303 | ||
1304 | /* | |
1305 | * Check whether this bio extends beyond the end of the device. | |
1306 | */ | |
1307 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1308 | { | |
1309 | sector_t maxsector; | |
1310 | ||
1311 | if (!nr_sectors) | |
1312 | return 0; | |
1313 | ||
1314 | /* Test device or partition size, when known. */ | |
1315 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1316 | if (maxsector) { | |
1317 | sector_t sector = bio->bi_sector; | |
1318 | ||
1319 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1320 | /* | |
1321 | * This may well happen - the kernel calls bread() | |
1322 | * without checking the size of the device, e.g., when | |
1323 | * mounting a device. | |
1324 | */ | |
1325 | handle_bad_sector(bio); | |
1326 | return 1; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | /** | |
1334 | * generic_make_request - hand a buffer to its device driver for I/O | |
1335 | * @bio: The bio describing the location in memory and on the device. | |
1336 | * | |
1337 | * generic_make_request() is used to make I/O requests of block | |
1338 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1339 | * to be done. | |
1340 | * | |
1341 | * generic_make_request() does not return any status. The | |
1342 | * success/failure status of the request, along with notification of | |
1343 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1344 | * function described (one day) else where. | |
1345 | * | |
1346 | * The caller of generic_make_request must make sure that bi_io_vec | |
1347 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1348 | * set to describe the device address, and the | |
1349 | * bi_end_io and optionally bi_private are set to describe how | |
1350 | * completion notification should be signaled. | |
1351 | * | |
1352 | * generic_make_request and the drivers it calls may use bi_next if this | |
1353 | * bio happens to be merged with someone else, and may change bi_dev and | |
1354 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1355 | * should NOT be depended on after the call to generic_make_request. | |
1356 | */ | |
1357 | static inline void __generic_make_request(struct bio *bio) | |
1358 | { | |
1359 | struct request_queue *q; | |
1360 | sector_t old_sector; | |
1361 | int ret, nr_sectors = bio_sectors(bio); | |
1362 | dev_t old_dev; | |
1363 | int err = -EIO; | |
1364 | ||
1365 | might_sleep(); | |
1366 | ||
1367 | if (bio_check_eod(bio, nr_sectors)) | |
1368 | goto end_io; | |
1369 | ||
1370 | /* | |
1371 | * Resolve the mapping until finished. (drivers are | |
1372 | * still free to implement/resolve their own stacking | |
1373 | * by explicitly returning 0) | |
1374 | * | |
1375 | * NOTE: we don't repeat the blk_size check for each new device. | |
1376 | * Stacking drivers are expected to know what they are doing. | |
1377 | */ | |
1378 | old_sector = -1; | |
1379 | old_dev = 0; | |
1380 | do { | |
1381 | char b[BDEVNAME_SIZE]; | |
1382 | ||
1383 | q = bdev_get_queue(bio->bi_bdev); | |
1384 | if (!q) { | |
1385 | printk(KERN_ERR | |
1386 | "generic_make_request: Trying to access " | |
1387 | "nonexistent block-device %s (%Lu)\n", | |
1388 | bdevname(bio->bi_bdev, b), | |
1389 | (long long) bio->bi_sector); | |
1390 | end_io: | |
1391 | bio_endio(bio, err); | |
1392 | break; | |
1393 | } | |
1394 | ||
1395 | if (unlikely(nr_sectors > q->max_hw_sectors)) { | |
1396 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1397 | bdevname(bio->bi_bdev, b), | |
1398 | bio_sectors(bio), | |
1399 | q->max_hw_sectors); | |
1400 | goto end_io; | |
1401 | } | |
1402 | ||
1403 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) | |
1404 | goto end_io; | |
1405 | ||
1406 | if (should_fail_request(bio)) | |
1407 | goto end_io; | |
1408 | ||
1409 | /* | |
1410 | * If this device has partitions, remap block n | |
1411 | * of partition p to block n+start(p) of the disk. | |
1412 | */ | |
1413 | blk_partition_remap(bio); | |
1414 | ||
1415 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) | |
1416 | goto end_io; | |
1417 | ||
1418 | if (old_sector != -1) | |
1419 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, | |
1420 | old_sector); | |
1421 | ||
1422 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
1423 | ||
1424 | old_sector = bio->bi_sector; | |
1425 | old_dev = bio->bi_bdev->bd_dev; | |
1426 | ||
1427 | if (bio_check_eod(bio, nr_sectors)) | |
1428 | goto end_io; | |
1429 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || | |
1430 | (bio_discard(bio) && !q->prepare_discard_fn)) { | |
1431 | err = -EOPNOTSUPP; | |
1432 | goto end_io; | |
1433 | } | |
1434 | ||
1435 | ret = q->make_request_fn(q, bio); | |
1436 | } while (ret); | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * We only want one ->make_request_fn to be active at a time, | |
1441 | * else stack usage with stacked devices could be a problem. | |
1442 | * So use current->bio_{list,tail} to keep a list of requests | |
1443 | * submited by a make_request_fn function. | |
1444 | * current->bio_tail is also used as a flag to say if | |
1445 | * generic_make_request is currently active in this task or not. | |
1446 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1447 | * then a make_request is active, and new requests should be added | |
1448 | * at the tail | |
1449 | */ | |
1450 | void generic_make_request(struct bio *bio) | |
1451 | { | |
1452 | if (current->bio_tail) { | |
1453 | /* make_request is active */ | |
1454 | *(current->bio_tail) = bio; | |
1455 | bio->bi_next = NULL; | |
1456 | current->bio_tail = &bio->bi_next; | |
1457 | return; | |
1458 | } | |
1459 | /* following loop may be a bit non-obvious, and so deserves some | |
1460 | * explanation. | |
1461 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1462 | * ensure that) so we have a list with a single bio. | |
1463 | * We pretend that we have just taken it off a longer list, so | |
1464 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1465 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1466 | * added. __generic_make_request may indeed add some more bios | |
1467 | * through a recursive call to generic_make_request. If it | |
1468 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1469 | * from the top. In this case we really did just take the bio | |
1470 | * of the top of the list (no pretending) and so fixup bio_list and | |
1471 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1472 | * | |
1473 | * The loop was structured like this to make only one call to | |
1474 | * __generic_make_request (which is important as it is large and | |
1475 | * inlined) and to keep the structure simple. | |
1476 | */ | |
1477 | BUG_ON(bio->bi_next); | |
1478 | do { | |
1479 | current->bio_list = bio->bi_next; | |
1480 | if (bio->bi_next == NULL) | |
1481 | current->bio_tail = ¤t->bio_list; | |
1482 | else | |
1483 | bio->bi_next = NULL; | |
1484 | __generic_make_request(bio); | |
1485 | bio = current->bio_list; | |
1486 | } while (bio); | |
1487 | current->bio_tail = NULL; /* deactivate */ | |
1488 | } | |
1489 | EXPORT_SYMBOL(generic_make_request); | |
1490 | ||
1491 | /** | |
1492 | * submit_bio - submit a bio to the block device layer for I/O | |
1493 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
1494 | * @bio: The &struct bio which describes the I/O | |
1495 | * | |
1496 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1497 | * uses that function to do most of the work. Both are fairly rough | |
1498 | * interfaces; @bio must be presetup and ready for I/O. | |
1499 | * | |
1500 | */ | |
1501 | void submit_bio(int rw, struct bio *bio) | |
1502 | { | |
1503 | int count = bio_sectors(bio); | |
1504 | ||
1505 | bio->bi_rw |= rw; | |
1506 | ||
1507 | /* | |
1508 | * If it's a regular read/write or a barrier with data attached, | |
1509 | * go through the normal accounting stuff before submission. | |
1510 | */ | |
1511 | if (bio_has_data(bio)) { | |
1512 | if (rw & WRITE) { | |
1513 | count_vm_events(PGPGOUT, count); | |
1514 | } else { | |
1515 | task_io_account_read(bio->bi_size); | |
1516 | count_vm_events(PGPGIN, count); | |
1517 | } | |
1518 | ||
1519 | if (unlikely(block_dump)) { | |
1520 | char b[BDEVNAME_SIZE]; | |
1521 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
1522 | current->comm, task_pid_nr(current), | |
1523 | (rw & WRITE) ? "WRITE" : "READ", | |
1524 | (unsigned long long)bio->bi_sector, | |
1525 | bdevname(bio->bi_bdev, b)); | |
1526 | } | |
1527 | } | |
1528 | ||
1529 | generic_make_request(bio); | |
1530 | } | |
1531 | EXPORT_SYMBOL(submit_bio); | |
1532 | ||
1533 | /** | |
1534 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1535 | * @q: the queue | |
1536 | * @rq: the request being checked | |
1537 | * | |
1538 | * Description: | |
1539 | * @rq may have been made based on weaker limitations of upper-level queues | |
1540 | * in request stacking drivers, and it may violate the limitation of @q. | |
1541 | * Since the block layer and the underlying device driver trust @rq | |
1542 | * after it is inserted to @q, it should be checked against @q before | |
1543 | * the insertion using this generic function. | |
1544 | * | |
1545 | * This function should also be useful for request stacking drivers | |
1546 | * in some cases below, so export this fuction. | |
1547 | * Request stacking drivers like request-based dm may change the queue | |
1548 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1549 | * Such request stacking drivers should check those requests agaist | |
1550 | * the new queue limits again when they dispatch those requests, | |
1551 | * although such checkings are also done against the old queue limits | |
1552 | * when submitting requests. | |
1553 | */ | |
1554 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1555 | { | |
1556 | if (rq->nr_sectors > q->max_sectors || | |
1557 | rq->data_len > q->max_hw_sectors << 9) { | |
1558 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1559 | return -EIO; | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * queue's settings related to segment counting like q->bounce_pfn | |
1564 | * may differ from that of other stacking queues. | |
1565 | * Recalculate it to check the request correctly on this queue's | |
1566 | * limitation. | |
1567 | */ | |
1568 | blk_recalc_rq_segments(rq); | |
1569 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1570 | rq->nr_phys_segments > q->max_hw_segments) { | |
1571 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1572 | return -EIO; | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1578 | ||
1579 | /** | |
1580 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1581 | * @q: the queue to submit the request | |
1582 | * @rq: the request being queued | |
1583 | */ | |
1584 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1585 | { | |
1586 | unsigned long flags; | |
1587 | ||
1588 | if (blk_rq_check_limits(q, rq)) | |
1589 | return -EIO; | |
1590 | ||
1591 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1592 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1593 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1594 | return -EIO; | |
1595 | #endif | |
1596 | ||
1597 | spin_lock_irqsave(q->queue_lock, flags); | |
1598 | ||
1599 | /* | |
1600 | * Submitting request must be dequeued before calling this function | |
1601 | * because it will be linked to another request_queue | |
1602 | */ | |
1603 | BUG_ON(blk_queued_rq(rq)); | |
1604 | ||
1605 | drive_stat_acct(rq, 1); | |
1606 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1607 | ||
1608 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1609 | ||
1610 | return 0; | |
1611 | } | |
1612 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1613 | ||
1614 | /** | |
1615 | * __end_that_request_first - end I/O on a request | |
1616 | * @req: the request being processed | |
1617 | * @error: %0 for success, < %0 for error | |
1618 | * @nr_bytes: number of bytes to complete | |
1619 | * | |
1620 | * Description: | |
1621 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1622 | * for the next range of segments (if any) in the cluster. | |
1623 | * | |
1624 | * Return: | |
1625 | * %0 - we are done with this request, call end_that_request_last() | |
1626 | * %1 - still buffers pending for this request | |
1627 | **/ | |
1628 | static int __end_that_request_first(struct request *req, int error, | |
1629 | int nr_bytes) | |
1630 | { | |
1631 | int total_bytes, bio_nbytes, next_idx = 0; | |
1632 | struct bio *bio; | |
1633 | ||
1634 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); | |
1635 | ||
1636 | /* | |
1637 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual | |
1638 | * sense key with us all the way through | |
1639 | */ | |
1640 | if (!blk_pc_request(req)) | |
1641 | req->errors = 0; | |
1642 | ||
1643 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { | |
1644 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1645 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
1646 | (unsigned long long)req->sector); | |
1647 | } | |
1648 | ||
1649 | if (blk_fs_request(req) && req->rq_disk) { | |
1650 | const int rw = rq_data_dir(req); | |
1651 | struct hd_struct *part; | |
1652 | int cpu; | |
1653 | ||
1654 | cpu = part_stat_lock(); | |
1655 | part = disk_map_sector_rcu(req->rq_disk, req->sector); | |
1656 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); | |
1657 | part_stat_unlock(); | |
1658 | } | |
1659 | ||
1660 | total_bytes = bio_nbytes = 0; | |
1661 | while ((bio = req->bio) != NULL) { | |
1662 | int nbytes; | |
1663 | ||
1664 | /* | |
1665 | * For an empty barrier request, the low level driver must | |
1666 | * store a potential error location in ->sector. We pass | |
1667 | * that back up in ->bi_sector. | |
1668 | */ | |
1669 | if (blk_empty_barrier(req)) | |
1670 | bio->bi_sector = req->sector; | |
1671 | ||
1672 | if (nr_bytes >= bio->bi_size) { | |
1673 | req->bio = bio->bi_next; | |
1674 | nbytes = bio->bi_size; | |
1675 | req_bio_endio(req, bio, nbytes, error); | |
1676 | next_idx = 0; | |
1677 | bio_nbytes = 0; | |
1678 | } else { | |
1679 | int idx = bio->bi_idx + next_idx; | |
1680 | ||
1681 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1682 | blk_dump_rq_flags(req, "__end_that"); | |
1683 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", | |
1684 | __func__, bio->bi_idx, bio->bi_vcnt); | |
1685 | break; | |
1686 | } | |
1687 | ||
1688 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1689 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1690 | ||
1691 | /* | |
1692 | * not a complete bvec done | |
1693 | */ | |
1694 | if (unlikely(nbytes > nr_bytes)) { | |
1695 | bio_nbytes += nr_bytes; | |
1696 | total_bytes += nr_bytes; | |
1697 | break; | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * advance to the next vector | |
1702 | */ | |
1703 | next_idx++; | |
1704 | bio_nbytes += nbytes; | |
1705 | } | |
1706 | ||
1707 | total_bytes += nbytes; | |
1708 | nr_bytes -= nbytes; | |
1709 | ||
1710 | bio = req->bio; | |
1711 | if (bio) { | |
1712 | /* | |
1713 | * end more in this run, or just return 'not-done' | |
1714 | */ | |
1715 | if (unlikely(nr_bytes <= 0)) | |
1716 | break; | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * completely done | |
1722 | */ | |
1723 | if (!req->bio) | |
1724 | return 0; | |
1725 | ||
1726 | /* | |
1727 | * if the request wasn't completed, update state | |
1728 | */ | |
1729 | if (bio_nbytes) { | |
1730 | req_bio_endio(req, bio, bio_nbytes, error); | |
1731 | bio->bi_idx += next_idx; | |
1732 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1733 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1734 | } | |
1735 | ||
1736 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1737 | blk_recalc_rq_segments(req); | |
1738 | return 1; | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * queue lock must be held | |
1743 | */ | |
1744 | static void end_that_request_last(struct request *req, int error) | |
1745 | { | |
1746 | struct gendisk *disk = req->rq_disk; | |
1747 | ||
1748 | blk_delete_timer(req); | |
1749 | ||
1750 | if (blk_rq_tagged(req)) | |
1751 | blk_queue_end_tag(req->q, req); | |
1752 | ||
1753 | if (blk_queued_rq(req)) | |
1754 | blkdev_dequeue_request(req); | |
1755 | ||
1756 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1757 | laptop_io_completion(); | |
1758 | ||
1759 | /* | |
1760 | * Account IO completion. bar_rq isn't accounted as a normal | |
1761 | * IO on queueing nor completion. Accounting the containing | |
1762 | * request is enough. | |
1763 | */ | |
1764 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1765 | unsigned long duration = jiffies - req->start_time; | |
1766 | const int rw = rq_data_dir(req); | |
1767 | struct hd_struct *part; | |
1768 | int cpu; | |
1769 | ||
1770 | cpu = part_stat_lock(); | |
1771 | part = disk_map_sector_rcu(disk, req->sector); | |
1772 | ||
1773 | part_stat_inc(cpu, part, ios[rw]); | |
1774 | part_stat_add(cpu, part, ticks[rw], duration); | |
1775 | part_round_stats(cpu, part); | |
1776 | part_dec_in_flight(part); | |
1777 | ||
1778 | part_stat_unlock(); | |
1779 | } | |
1780 | ||
1781 | if (req->end_io) | |
1782 | req->end_io(req, error); | |
1783 | else { | |
1784 | if (blk_bidi_rq(req)) | |
1785 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1786 | ||
1787 | __blk_put_request(req->q, req); | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | static inline void __end_request(struct request *rq, int uptodate, | |
1792 | unsigned int nr_bytes) | |
1793 | { | |
1794 | int error = 0; | |
1795 | ||
1796 | if (uptodate <= 0) | |
1797 | error = uptodate ? uptodate : -EIO; | |
1798 | ||
1799 | __blk_end_request(rq, error, nr_bytes); | |
1800 | } | |
1801 | ||
1802 | /** | |
1803 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
1804 | * @rq: the request being processed | |
1805 | **/ | |
1806 | unsigned int blk_rq_bytes(struct request *rq) | |
1807 | { | |
1808 | if (blk_fs_request(rq)) | |
1809 | return rq->hard_nr_sectors << 9; | |
1810 | ||
1811 | return rq->data_len; | |
1812 | } | |
1813 | EXPORT_SYMBOL_GPL(blk_rq_bytes); | |
1814 | ||
1815 | /** | |
1816 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
1817 | * @rq: the request being processed | |
1818 | **/ | |
1819 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1820 | { | |
1821 | if (blk_fs_request(rq)) | |
1822 | return rq->current_nr_sectors << 9; | |
1823 | ||
1824 | if (rq->bio) | |
1825 | return rq->bio->bi_size; | |
1826 | ||
1827 | return rq->data_len; | |
1828 | } | |
1829 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
1830 | ||
1831 | /** | |
1832 | * end_queued_request - end all I/O on a queued request | |
1833 | * @rq: the request being processed | |
1834 | * @uptodate: error value or %0/%1 uptodate flag | |
1835 | * | |
1836 | * Description: | |
1837 | * Ends all I/O on a request, and removes it from the block layer queues. | |
1838 | * Not suitable for normal I/O completion, unless the driver still has | |
1839 | * the request attached to the block layer. | |
1840 | * | |
1841 | **/ | |
1842 | void end_queued_request(struct request *rq, int uptodate) | |
1843 | { | |
1844 | __end_request(rq, uptodate, blk_rq_bytes(rq)); | |
1845 | } | |
1846 | EXPORT_SYMBOL(end_queued_request); | |
1847 | ||
1848 | /** | |
1849 | * end_dequeued_request - end all I/O on a dequeued request | |
1850 | * @rq: the request being processed | |
1851 | * @uptodate: error value or %0/%1 uptodate flag | |
1852 | * | |
1853 | * Description: | |
1854 | * Ends all I/O on a request. The request must already have been | |
1855 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
1856 | * for most drivers. | |
1857 | * | |
1858 | **/ | |
1859 | void end_dequeued_request(struct request *rq, int uptodate) | |
1860 | { | |
1861 | __end_request(rq, uptodate, blk_rq_bytes(rq)); | |
1862 | } | |
1863 | EXPORT_SYMBOL(end_dequeued_request); | |
1864 | ||
1865 | ||
1866 | /** | |
1867 | * end_request - end I/O on the current segment of the request | |
1868 | * @req: the request being processed | |
1869 | * @uptodate: error value or %0/%1 uptodate flag | |
1870 | * | |
1871 | * Description: | |
1872 | * Ends I/O on the current segment of a request. If that is the only | |
1873 | * remaining segment, the request is also completed and freed. | |
1874 | * | |
1875 | * This is a remnant of how older block drivers handled I/O completions. | |
1876 | * Modern drivers typically end I/O on the full request in one go, unless | |
1877 | * they have a residual value to account for. For that case this function | |
1878 | * isn't really useful, unless the residual just happens to be the | |
1879 | * full current segment. In other words, don't use this function in new | |
1880 | * code. Use blk_end_request() or __blk_end_request() to end partial parts | |
1881 | * of a request, or end_dequeued_request() and end_queued_request() to | |
1882 | * completely end IO on a dequeued/queued request. | |
1883 | * | |
1884 | **/ | |
1885 | void end_request(struct request *req, int uptodate) | |
1886 | { | |
1887 | __end_request(req, uptodate, req->hard_cur_sectors << 9); | |
1888 | } | |
1889 | EXPORT_SYMBOL(end_request); | |
1890 | ||
1891 | static int end_that_request_data(struct request *rq, int error, | |
1892 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1893 | { | |
1894 | if (rq->bio) { | |
1895 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1896 | return 1; | |
1897 | ||
1898 | /* Bidi request must be completed as a whole */ | |
1899 | if (blk_bidi_rq(rq) && | |
1900 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1901 | return 1; | |
1902 | } | |
1903 | ||
1904 | return 0; | |
1905 | } | |
1906 | ||
1907 | /** | |
1908 | * blk_end_io - Generic end_io function to complete a request. | |
1909 | * @rq: the request being processed | |
1910 | * @error: %0 for success, < %0 for error | |
1911 | * @nr_bytes: number of bytes to complete @rq | |
1912 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1913 | * @drv_callback: function called between completion of bios in the request | |
1914 | * and completion of the request. | |
1915 | * If the callback returns non %0, this helper returns without | |
1916 | * completion of the request. | |
1917 | * | |
1918 | * Description: | |
1919 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1920 | * If @rq has leftover, sets it up for the next range of segments. | |
1921 | * | |
1922 | * Return: | |
1923 | * %0 - we are done with this request | |
1924 | * %1 - this request is not freed yet, it still has pending buffers. | |
1925 | **/ | |
1926 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, | |
1927 | unsigned int bidi_bytes, | |
1928 | int (drv_callback)(struct request *)) | |
1929 | { | |
1930 | struct request_queue *q = rq->q; | |
1931 | unsigned long flags = 0UL; | |
1932 | ||
1933 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) | |
1934 | return 1; | |
1935 | ||
1936 | /* Special feature for tricky drivers */ | |
1937 | if (drv_callback && drv_callback(rq)) | |
1938 | return 1; | |
1939 | ||
1940 | add_disk_randomness(rq->rq_disk); | |
1941 | ||
1942 | spin_lock_irqsave(q->queue_lock, flags); | |
1943 | end_that_request_last(rq, error); | |
1944 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1945 | ||
1946 | return 0; | |
1947 | } | |
1948 | ||
1949 | /** | |
1950 | * blk_end_request - Helper function for drivers to complete the request. | |
1951 | * @rq: the request being processed | |
1952 | * @error: %0 for success, < %0 for error | |
1953 | * @nr_bytes: number of bytes to complete | |
1954 | * | |
1955 | * Description: | |
1956 | * Ends I/O on a number of bytes attached to @rq. | |
1957 | * If @rq has leftover, sets it up for the next range of segments. | |
1958 | * | |
1959 | * Return: | |
1960 | * %0 - we are done with this request | |
1961 | * %1 - still buffers pending for this request | |
1962 | **/ | |
1963 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) | |
1964 | { | |
1965 | return blk_end_io(rq, error, nr_bytes, 0, NULL); | |
1966 | } | |
1967 | EXPORT_SYMBOL_GPL(blk_end_request); | |
1968 | ||
1969 | /** | |
1970 | * __blk_end_request - Helper function for drivers to complete the request. | |
1971 | * @rq: the request being processed | |
1972 | * @error: %0 for success, < %0 for error | |
1973 | * @nr_bytes: number of bytes to complete | |
1974 | * | |
1975 | * Description: | |
1976 | * Must be called with queue lock held unlike blk_end_request(). | |
1977 | * | |
1978 | * Return: | |
1979 | * %0 - we are done with this request | |
1980 | * %1 - still buffers pending for this request | |
1981 | **/ | |
1982 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) | |
1983 | { | |
1984 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) | |
1985 | return 1; | |
1986 | ||
1987 | add_disk_randomness(rq->rq_disk); | |
1988 | ||
1989 | end_that_request_last(rq, error); | |
1990 | ||
1991 | return 0; | |
1992 | } | |
1993 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1994 | ||
1995 | /** | |
1996 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1997 | * @rq: the bidi request being processed | |
1998 | * @error: %0 for success, < %0 for error | |
1999 | * @nr_bytes: number of bytes to complete @rq | |
2000 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2001 | * | |
2002 | * Description: | |
2003 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2004 | * | |
2005 | * Return: | |
2006 | * %0 - we are done with this request | |
2007 | * %1 - still buffers pending for this request | |
2008 | **/ | |
2009 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, | |
2010 | unsigned int bidi_bytes) | |
2011 | { | |
2012 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
2013 | } | |
2014 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2015 | ||
2016 | /** | |
2017 | * blk_update_request - Special helper function for request stacking drivers | |
2018 | * @rq: the request being processed | |
2019 | * @error: %0 for success, < %0 for error | |
2020 | * @nr_bytes: number of bytes to complete @rq | |
2021 | * | |
2022 | * Description: | |
2023 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2024 | * the request structure even if @rq doesn't have leftover. | |
2025 | * If @rq has leftover, sets it up for the next range of segments. | |
2026 | * | |
2027 | * This special helper function is only for request stacking drivers | |
2028 | * (e.g. request-based dm) so that they can handle partial completion. | |
2029 | * Actual device drivers should use blk_end_request instead. | |
2030 | */ | |
2031 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2032 | { | |
2033 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2034 | /* | |
2035 | * These members are not updated in end_that_request_data() | |
2036 | * when all bios are completed. | |
2037 | * Update them so that the request stacking driver can find | |
2038 | * how many bytes remain in the request later. | |
2039 | */ | |
2040 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2041 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2042 | } | |
2043 | } | |
2044 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2045 | ||
2046 | /** | |
2047 | * blk_end_request_callback - Special helper function for tricky drivers | |
2048 | * @rq: the request being processed | |
2049 | * @error: %0 for success, < %0 for error | |
2050 | * @nr_bytes: number of bytes to complete | |
2051 | * @drv_callback: function called between completion of bios in the request | |
2052 | * and completion of the request. | |
2053 | * If the callback returns non %0, this helper returns without | |
2054 | * completion of the request. | |
2055 | * | |
2056 | * Description: | |
2057 | * Ends I/O on a number of bytes attached to @rq. | |
2058 | * If @rq has leftover, sets it up for the next range of segments. | |
2059 | * | |
2060 | * This special helper function is used only for existing tricky drivers. | |
2061 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2062 | * This interface will be removed when such drivers are rewritten. | |
2063 | * Don't use this interface in other places anymore. | |
2064 | * | |
2065 | * Return: | |
2066 | * %0 - we are done with this request | |
2067 | * %1 - this request is not freed yet. | |
2068 | * this request still has pending buffers or | |
2069 | * the driver doesn't want to finish this request yet. | |
2070 | **/ | |
2071 | int blk_end_request_callback(struct request *rq, int error, | |
2072 | unsigned int nr_bytes, | |
2073 | int (drv_callback)(struct request *)) | |
2074 | { | |
2075 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); | |
2076 | } | |
2077 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2078 | ||
2079 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
2080 | struct bio *bio) | |
2081 | { | |
2082 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and | |
2083 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
2084 | rq->cmd_flags |= (bio->bi_rw & 3); | |
2085 | ||
2086 | if (bio_has_data(bio)) { | |
2087 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
2088 | rq->buffer = bio_data(bio); | |
2089 | } | |
2090 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
2091 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2092 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
2093 | rq->data_len = bio->bi_size; | |
2094 | ||
2095 | rq->bio = rq->biotail = bio; | |
2096 | ||
2097 | if (bio->bi_bdev) | |
2098 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2099 | } | |
2100 | ||
2101 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) | |
2102 | { | |
2103 | return queue_work(kblockd_workqueue, work); | |
2104 | } | |
2105 | EXPORT_SYMBOL(kblockd_schedule_work); | |
2106 | ||
2107 | void kblockd_flush_work(struct work_struct *work) | |
2108 | { | |
2109 | cancel_work_sync(work); | |
2110 | } | |
2111 | EXPORT_SYMBOL(kblockd_flush_work); | |
2112 | ||
2113 | int __init blk_dev_init(void) | |
2114 | { | |
2115 | kblockd_workqueue = create_workqueue("kblockd"); | |
2116 | if (!kblockd_workqueue) | |
2117 | panic("Failed to create kblockd\n"); | |
2118 | ||
2119 | request_cachep = kmem_cache_create("blkdev_requests", | |
2120 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
2121 | ||
2122 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", | |
2123 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
2124 | ||
2125 | return 0; | |
2126 | } | |
2127 |