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