]>
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/fault-inject.h> | |
30 | ||
31 | #define CREATE_TRACE_POINTS | |
32 | #include <trace/events/block.h> | |
33 | ||
34 | #include "blk.h" | |
35 | ||
36 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); | |
37 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); | |
38 | ||
39 | static int __make_request(struct request_queue *q, struct bio *bio); | |
40 | ||
41 | /* | |
42 | * For the allocated request tables | |
43 | */ | |
44 | static struct kmem_cache *request_cachep; | |
45 | ||
46 | /* | |
47 | * For queue allocation | |
48 | */ | |
49 | struct kmem_cache *blk_requestq_cachep; | |
50 | ||
51 | /* | |
52 | * Controlling structure to kblockd | |
53 | */ | |
54 | static struct workqueue_struct *kblockd_workqueue; | |
55 | ||
56 | static void drive_stat_acct(struct request *rq, int new_io) | |
57 | { | |
58 | struct hd_struct *part; | |
59 | int rw = rq_data_dir(rq); | |
60 | int cpu; | |
61 | ||
62 | if (!blk_do_io_stat(rq)) | |
63 | return; | |
64 | ||
65 | cpu = part_stat_lock(); | |
66 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
67 | ||
68 | if (!new_io) | |
69 | part_stat_inc(cpu, part, merges[rw]); | |
70 | else { | |
71 | part_round_stats(cpu, part); | |
72 | part_inc_in_flight(part, rw); | |
73 | } | |
74 | ||
75 | part_stat_unlock(); | |
76 | } | |
77 | ||
78 | void blk_queue_congestion_threshold(struct request_queue *q) | |
79 | { | |
80 | int nr; | |
81 | ||
82 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
83 | if (nr > q->nr_requests) | |
84 | nr = q->nr_requests; | |
85 | q->nr_congestion_on = nr; | |
86 | ||
87 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
88 | if (nr < 1) | |
89 | nr = 1; | |
90 | q->nr_congestion_off = nr; | |
91 | } | |
92 | ||
93 | /** | |
94 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
95 | * @bdev: device | |
96 | * | |
97 | * Locates the passed device's request queue and returns the address of its | |
98 | * backing_dev_info | |
99 | * | |
100 | * Will return NULL if the request queue cannot be located. | |
101 | */ | |
102 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
103 | { | |
104 | struct backing_dev_info *ret = NULL; | |
105 | struct request_queue *q = bdev_get_queue(bdev); | |
106 | ||
107 | if (q) | |
108 | ret = &q->backing_dev_info; | |
109 | return ret; | |
110 | } | |
111 | EXPORT_SYMBOL(blk_get_backing_dev_info); | |
112 | ||
113 | void blk_rq_init(struct request_queue *q, struct request *rq) | |
114 | { | |
115 | memset(rq, 0, sizeof(*rq)); | |
116 | ||
117 | INIT_LIST_HEAD(&rq->queuelist); | |
118 | INIT_LIST_HEAD(&rq->timeout_list); | |
119 | rq->cpu = -1; | |
120 | rq->q = q; | |
121 | rq->__sector = (sector_t) -1; | |
122 | INIT_HLIST_NODE(&rq->hash); | |
123 | RB_CLEAR_NODE(&rq->rb_node); | |
124 | rq->cmd = rq->__cmd; | |
125 | rq->cmd_len = BLK_MAX_CDB; | |
126 | rq->tag = -1; | |
127 | rq->ref_count = 1; | |
128 | rq->start_time = jiffies; | |
129 | } | |
130 | EXPORT_SYMBOL(blk_rq_init); | |
131 | ||
132 | static void req_bio_endio(struct request *rq, struct bio *bio, | |
133 | unsigned int nbytes, int error) | |
134 | { | |
135 | struct request_queue *q = rq->q; | |
136 | ||
137 | if (&q->bar_rq != rq) { | |
138 | if (error) | |
139 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
140 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
141 | error = -EIO; | |
142 | ||
143 | if (unlikely(nbytes > bio->bi_size)) { | |
144 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
145 | __func__, nbytes, bio->bi_size); | |
146 | nbytes = bio->bi_size; | |
147 | } | |
148 | ||
149 | if (unlikely(rq->cmd_flags & REQ_QUIET)) | |
150 | set_bit(BIO_QUIET, &bio->bi_flags); | |
151 | ||
152 | bio->bi_size -= nbytes; | |
153 | bio->bi_sector += (nbytes >> 9); | |
154 | ||
155 | if (bio_integrity(bio)) | |
156 | bio_integrity_advance(bio, nbytes); | |
157 | ||
158 | if (bio->bi_size == 0) | |
159 | bio_endio(bio, error); | |
160 | } else { | |
161 | ||
162 | /* | |
163 | * Okay, this is the barrier request in progress, just | |
164 | * record the error; | |
165 | */ | |
166 | if (error && !q->orderr) | |
167 | q->orderr = error; | |
168 | } | |
169 | } | |
170 | ||
171 | void blk_dump_rq_flags(struct request *rq, char *msg) | |
172 | { | |
173 | int bit; | |
174 | ||
175 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, | |
176 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
177 | rq->cmd_flags); | |
178 | ||
179 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", | |
180 | (unsigned long long)blk_rq_pos(rq), | |
181 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
182 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", | |
183 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); | |
184 | ||
185 | if (blk_pc_request(rq)) { | |
186 | printk(KERN_INFO " cdb: "); | |
187 | for (bit = 0; bit < BLK_MAX_CDB; bit++) | |
188 | printk("%02x ", rq->cmd[bit]); | |
189 | printk("\n"); | |
190 | } | |
191 | } | |
192 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
193 | ||
194 | /* | |
195 | * "plug" the device if there are no outstanding requests: this will | |
196 | * force the transfer to start only after we have put all the requests | |
197 | * on the list. | |
198 | * | |
199 | * This is called with interrupts off and no requests on the queue and | |
200 | * with the queue lock held. | |
201 | */ | |
202 | void blk_plug_device(struct request_queue *q) | |
203 | { | |
204 | WARN_ON(!irqs_disabled()); | |
205 | ||
206 | /* | |
207 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
208 | * which will restart the queueing | |
209 | */ | |
210 | if (blk_queue_stopped(q)) | |
211 | return; | |
212 | ||
213 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { | |
214 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | |
215 | trace_block_plug(q); | |
216 | } | |
217 | } | |
218 | EXPORT_SYMBOL(blk_plug_device); | |
219 | ||
220 | /** | |
221 | * blk_plug_device_unlocked - plug a device without queue lock held | |
222 | * @q: The &struct request_queue to plug | |
223 | * | |
224 | * Description: | |
225 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
226 | * interrupts. | |
227 | **/ | |
228 | void blk_plug_device_unlocked(struct request_queue *q) | |
229 | { | |
230 | unsigned long flags; | |
231 | ||
232 | spin_lock_irqsave(q->queue_lock, flags); | |
233 | blk_plug_device(q); | |
234 | spin_unlock_irqrestore(q->queue_lock, flags); | |
235 | } | |
236 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
237 | ||
238 | /* | |
239 | * remove the queue from the plugged list, if present. called with | |
240 | * queue lock held and interrupts disabled. | |
241 | */ | |
242 | int blk_remove_plug(struct request_queue *q) | |
243 | { | |
244 | WARN_ON(!irqs_disabled()); | |
245 | ||
246 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) | |
247 | return 0; | |
248 | ||
249 | del_timer(&q->unplug_timer); | |
250 | return 1; | |
251 | } | |
252 | EXPORT_SYMBOL(blk_remove_plug); | |
253 | ||
254 | /* | |
255 | * remove the plug and let it rip.. | |
256 | */ | |
257 | void __generic_unplug_device(struct request_queue *q) | |
258 | { | |
259 | if (unlikely(blk_queue_stopped(q))) | |
260 | return; | |
261 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) | |
262 | return; | |
263 | ||
264 | q->request_fn(q); | |
265 | } | |
266 | ||
267 | /** | |
268 | * generic_unplug_device - fire a request queue | |
269 | * @q: The &struct request_queue in question | |
270 | * | |
271 | * Description: | |
272 | * Linux uses plugging to build bigger requests queues before letting | |
273 | * the device have at them. If a queue is plugged, the I/O scheduler | |
274 | * is still adding and merging requests on the queue. Once the queue | |
275 | * gets unplugged, the request_fn defined for the queue is invoked and | |
276 | * transfers started. | |
277 | **/ | |
278 | void generic_unplug_device(struct request_queue *q) | |
279 | { | |
280 | if (blk_queue_plugged(q)) { | |
281 | spin_lock_irq(q->queue_lock); | |
282 | __generic_unplug_device(q); | |
283 | spin_unlock_irq(q->queue_lock); | |
284 | } | |
285 | } | |
286 | EXPORT_SYMBOL(generic_unplug_device); | |
287 | ||
288 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
289 | struct page *page) | |
290 | { | |
291 | struct request_queue *q = bdi->unplug_io_data; | |
292 | ||
293 | blk_unplug(q); | |
294 | } | |
295 | ||
296 | void blk_unplug_work(struct work_struct *work) | |
297 | { | |
298 | struct request_queue *q = | |
299 | container_of(work, struct request_queue, unplug_work); | |
300 | ||
301 | trace_block_unplug_io(q); | |
302 | q->unplug_fn(q); | |
303 | } | |
304 | ||
305 | void blk_unplug_timeout(unsigned long data) | |
306 | { | |
307 | struct request_queue *q = (struct request_queue *)data; | |
308 | ||
309 | trace_block_unplug_timer(q); | |
310 | kblockd_schedule_work(q, &q->unplug_work); | |
311 | } | |
312 | ||
313 | void blk_unplug(struct request_queue *q) | |
314 | { | |
315 | /* | |
316 | * devices don't necessarily have an ->unplug_fn defined | |
317 | */ | |
318 | if (q->unplug_fn) { | |
319 | trace_block_unplug_io(q); | |
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 | __blk_run_queue(q); | |
340 | } | |
341 | EXPORT_SYMBOL(blk_start_queue); | |
342 | ||
343 | /** | |
344 | * blk_stop_queue - stop a queue | |
345 | * @q: The &struct request_queue in question | |
346 | * | |
347 | * Description: | |
348 | * The Linux block layer assumes that a block driver will consume all | |
349 | * entries on the request queue when the request_fn strategy is called. | |
350 | * Often this will not happen, because of hardware limitations (queue | |
351 | * depth settings). If a device driver gets a 'queue full' response, | |
352 | * or if it simply chooses not to queue more I/O at one point, it can | |
353 | * call this function to prevent the request_fn from being called until | |
354 | * the driver has signalled it's ready to go again. This happens by calling | |
355 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
356 | **/ | |
357 | void blk_stop_queue(struct request_queue *q) | |
358 | { | |
359 | blk_remove_plug(q); | |
360 | queue_flag_set(QUEUE_FLAG_STOPPED, q); | |
361 | } | |
362 | EXPORT_SYMBOL(blk_stop_queue); | |
363 | ||
364 | /** | |
365 | * blk_sync_queue - cancel any pending callbacks on a queue | |
366 | * @q: the queue | |
367 | * | |
368 | * Description: | |
369 | * The block layer may perform asynchronous callback activity | |
370 | * on a queue, such as calling the unplug function after a timeout. | |
371 | * A block device may call blk_sync_queue to ensure that any | |
372 | * such activity is cancelled, thus allowing it to release resources | |
373 | * that the callbacks might use. The caller must already have made sure | |
374 | * that its ->make_request_fn will not re-add plugging prior to calling | |
375 | * this function. | |
376 | * | |
377 | */ | |
378 | void blk_sync_queue(struct request_queue *q) | |
379 | { | |
380 | del_timer_sync(&q->unplug_timer); | |
381 | del_timer_sync(&q->timeout); | |
382 | cancel_work_sync(&q->unplug_work); | |
383 | } | |
384 | EXPORT_SYMBOL(blk_sync_queue); | |
385 | ||
386 | /** | |
387 | * __blk_run_queue - run a single device queue | |
388 | * @q: The queue to run | |
389 | * | |
390 | * Description: | |
391 | * See @blk_run_queue. This variant must be called with the queue lock | |
392 | * held and interrupts disabled. | |
393 | * | |
394 | */ | |
395 | void __blk_run_queue(struct request_queue *q) | |
396 | { | |
397 | blk_remove_plug(q); | |
398 | ||
399 | if (unlikely(blk_queue_stopped(q))) | |
400 | return; | |
401 | ||
402 | if (elv_queue_empty(q)) | |
403 | return; | |
404 | ||
405 | /* | |
406 | * Only recurse once to avoid overrunning the stack, let the unplug | |
407 | * handling reinvoke the handler shortly if we already got there. | |
408 | */ | |
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 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
414 | kblockd_schedule_work(q, &q->unplug_work); | |
415 | } | |
416 | } | |
417 | EXPORT_SYMBOL(__blk_run_queue); | |
418 | ||
419 | /** | |
420 | * blk_run_queue - run a single device queue | |
421 | * @q: The queue to run | |
422 | * | |
423 | * Description: | |
424 | * Invoke request handling on this queue, if it has pending work to do. | |
425 | * May be used to restart queueing when a request has completed. | |
426 | */ | |
427 | void blk_run_queue(struct request_queue *q) | |
428 | { | |
429 | unsigned long flags; | |
430 | ||
431 | spin_lock_irqsave(q->queue_lock, flags); | |
432 | __blk_run_queue(q); | |
433 | spin_unlock_irqrestore(q->queue_lock, flags); | |
434 | } | |
435 | EXPORT_SYMBOL(blk_run_queue); | |
436 | ||
437 | void blk_put_queue(struct request_queue *q) | |
438 | { | |
439 | kobject_put(&q->kobj); | |
440 | } | |
441 | ||
442 | void blk_cleanup_queue(struct request_queue *q) | |
443 | { | |
444 | /* | |
445 | * We know we have process context here, so we can be a little | |
446 | * cautious and ensure that pending block actions on this device | |
447 | * are done before moving on. Going into this function, we should | |
448 | * not have processes doing IO to this device. | |
449 | */ | |
450 | blk_sync_queue(q); | |
451 | ||
452 | mutex_lock(&q->sysfs_lock); | |
453 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); | |
454 | mutex_unlock(&q->sysfs_lock); | |
455 | ||
456 | if (q->elevator) | |
457 | elevator_exit(q->elevator); | |
458 | ||
459 | blk_put_queue(q); | |
460 | } | |
461 | EXPORT_SYMBOL(blk_cleanup_queue); | |
462 | ||
463 | static int blk_init_free_list(struct request_queue *q) | |
464 | { | |
465 | struct request_list *rl = &q->rq; | |
466 | ||
467 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; | |
468 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
469 | rl->elvpriv = 0; | |
470 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); | |
471 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
472 | ||
473 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | |
474 | mempool_free_slab, request_cachep, q->node); | |
475 | ||
476 | if (!rl->rq_pool) | |
477 | return -ENOMEM; | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) | |
483 | { | |
484 | return blk_alloc_queue_node(gfp_mask, -1); | |
485 | } | |
486 | EXPORT_SYMBOL(blk_alloc_queue); | |
487 | ||
488 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | |
489 | { | |
490 | struct request_queue *q; | |
491 | int err; | |
492 | ||
493 | q = kmem_cache_alloc_node(blk_requestq_cachep, | |
494 | gfp_mask | __GFP_ZERO, node_id); | |
495 | if (!q) | |
496 | return NULL; | |
497 | ||
498 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
499 | q->backing_dev_info.unplug_io_data = q; | |
500 | q->backing_dev_info.ra_pages = | |
501 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
502 | q->backing_dev_info.state = 0; | |
503 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
504 | q->backing_dev_info.name = "block"; | |
505 | ||
506 | err = bdi_init(&q->backing_dev_info); | |
507 | if (err) { | |
508 | kmem_cache_free(blk_requestq_cachep, q); | |
509 | return NULL; | |
510 | } | |
511 | ||
512 | init_timer(&q->unplug_timer); | |
513 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); | |
514 | INIT_LIST_HEAD(&q->timeout_list); | |
515 | INIT_WORK(&q->unplug_work, blk_unplug_work); | |
516 | ||
517 | kobject_init(&q->kobj, &blk_queue_ktype); | |
518 | ||
519 | mutex_init(&q->sysfs_lock); | |
520 | spin_lock_init(&q->__queue_lock); | |
521 | ||
522 | return q; | |
523 | } | |
524 | EXPORT_SYMBOL(blk_alloc_queue_node); | |
525 | ||
526 | /** | |
527 | * blk_init_queue - prepare a request queue for use with a block device | |
528 | * @rfn: The function to be called to process requests that have been | |
529 | * placed on the queue. | |
530 | * @lock: Request queue spin lock | |
531 | * | |
532 | * Description: | |
533 | * If a block device wishes to use the standard request handling procedures, | |
534 | * which sorts requests and coalesces adjacent requests, then it must | |
535 | * call blk_init_queue(). The function @rfn will be called when there | |
536 | * are requests on the queue that need to be processed. If the device | |
537 | * supports plugging, then @rfn may not be called immediately when requests | |
538 | * are available on the queue, but may be called at some time later instead. | |
539 | * Plugged queues are generally unplugged when a buffer belonging to one | |
540 | * of the requests on the queue is needed, or due to memory pressure. | |
541 | * | |
542 | * @rfn is not required, or even expected, to remove all requests off the | |
543 | * queue, but only as many as it can handle at a time. If it does leave | |
544 | * requests on the queue, it is responsible for arranging that the requests | |
545 | * get dealt with eventually. | |
546 | * | |
547 | * The queue spin lock must be held while manipulating the requests on the | |
548 | * request queue; this lock will be taken also from interrupt context, so irq | |
549 | * disabling is needed for it. | |
550 | * | |
551 | * Function returns a pointer to the initialized request queue, or %NULL if | |
552 | * it didn't succeed. | |
553 | * | |
554 | * Note: | |
555 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
556 | * when the block device is deactivated (such as at module unload). | |
557 | **/ | |
558 | ||
559 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | |
560 | { | |
561 | return blk_init_queue_node(rfn, lock, -1); | |
562 | } | |
563 | EXPORT_SYMBOL(blk_init_queue); | |
564 | ||
565 | struct request_queue * | |
566 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
567 | { | |
568 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
569 | ||
570 | if (!q) | |
571 | return NULL; | |
572 | ||
573 | q->node = node_id; | |
574 | if (blk_init_free_list(q)) { | |
575 | kmem_cache_free(blk_requestq_cachep, q); | |
576 | return NULL; | |
577 | } | |
578 | ||
579 | q->request_fn = rfn; | |
580 | q->prep_rq_fn = NULL; | |
581 | q->unplug_fn = generic_unplug_device; | |
582 | q->queue_flags = QUEUE_FLAG_DEFAULT; | |
583 | q->queue_lock = lock; | |
584 | ||
585 | /* | |
586 | * This also sets hw/phys segments, boundary and size | |
587 | */ | |
588 | blk_queue_make_request(q, __make_request); | |
589 | ||
590 | q->sg_reserved_size = INT_MAX; | |
591 | ||
592 | /* | |
593 | * all done | |
594 | */ | |
595 | if (!elevator_init(q, NULL)) { | |
596 | blk_queue_congestion_threshold(q); | |
597 | return q; | |
598 | } | |
599 | ||
600 | blk_put_queue(q); | |
601 | return NULL; | |
602 | } | |
603 | EXPORT_SYMBOL(blk_init_queue_node); | |
604 | ||
605 | int blk_get_queue(struct request_queue *q) | |
606 | { | |
607 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { | |
608 | kobject_get(&q->kobj); | |
609 | return 0; | |
610 | } | |
611 | ||
612 | return 1; | |
613 | } | |
614 | ||
615 | static inline void blk_free_request(struct request_queue *q, struct request *rq) | |
616 | { | |
617 | if (rq->cmd_flags & REQ_ELVPRIV) | |
618 | elv_put_request(q, rq); | |
619 | mempool_free(rq, q->rq.rq_pool); | |
620 | } | |
621 | ||
622 | static struct request * | |
623 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) | |
624 | { | |
625 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
626 | ||
627 | if (!rq) | |
628 | return NULL; | |
629 | ||
630 | blk_rq_init(q, rq); | |
631 | ||
632 | rq->cmd_flags = flags | REQ_ALLOCED; | |
633 | ||
634 | if (priv) { | |
635 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { | |
636 | mempool_free(rq, q->rq.rq_pool); | |
637 | return NULL; | |
638 | } | |
639 | rq->cmd_flags |= REQ_ELVPRIV; | |
640 | } | |
641 | ||
642 | return rq; | |
643 | } | |
644 | ||
645 | /* | |
646 | * ioc_batching returns true if the ioc is a valid batching request and | |
647 | * should be given priority access to a request. | |
648 | */ | |
649 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) | |
650 | { | |
651 | if (!ioc) | |
652 | return 0; | |
653 | ||
654 | /* | |
655 | * Make sure the process is able to allocate at least 1 request | |
656 | * even if the batch times out, otherwise we could theoretically | |
657 | * lose wakeups. | |
658 | */ | |
659 | return ioc->nr_batch_requests == q->nr_batching || | |
660 | (ioc->nr_batch_requests > 0 | |
661 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
662 | } | |
663 | ||
664 | /* | |
665 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
666 | * will cause the process to be a "batcher" on all queues in the system. This | |
667 | * is the behaviour we want though - once it gets a wakeup it should be given | |
668 | * a nice run. | |
669 | */ | |
670 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) | |
671 | { | |
672 | if (!ioc || ioc_batching(q, ioc)) | |
673 | return; | |
674 | ||
675 | ioc->nr_batch_requests = q->nr_batching; | |
676 | ioc->last_waited = jiffies; | |
677 | } | |
678 | ||
679 | static void __freed_request(struct request_queue *q, int sync) | |
680 | { | |
681 | struct request_list *rl = &q->rq; | |
682 | ||
683 | if (rl->count[sync] < queue_congestion_off_threshold(q)) | |
684 | blk_clear_queue_congested(q, sync); | |
685 | ||
686 | if (rl->count[sync] + 1 <= q->nr_requests) { | |
687 | if (waitqueue_active(&rl->wait[sync])) | |
688 | wake_up(&rl->wait[sync]); | |
689 | ||
690 | blk_clear_queue_full(q, sync); | |
691 | } | |
692 | } | |
693 | ||
694 | /* | |
695 | * A request has just been released. Account for it, update the full and | |
696 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
697 | */ | |
698 | static void freed_request(struct request_queue *q, int sync, int priv) | |
699 | { | |
700 | struct request_list *rl = &q->rq; | |
701 | ||
702 | rl->count[sync]--; | |
703 | if (priv) | |
704 | rl->elvpriv--; | |
705 | ||
706 | __freed_request(q, sync); | |
707 | ||
708 | if (unlikely(rl->starved[sync ^ 1])) | |
709 | __freed_request(q, sync ^ 1); | |
710 | } | |
711 | ||
712 | /* | |
713 | * Get a free request, queue_lock must be held. | |
714 | * Returns NULL on failure, with queue_lock held. | |
715 | * Returns !NULL on success, with queue_lock *not held*. | |
716 | */ | |
717 | static struct request *get_request(struct request_queue *q, int rw_flags, | |
718 | struct bio *bio, gfp_t gfp_mask) | |
719 | { | |
720 | struct request *rq = NULL; | |
721 | struct request_list *rl = &q->rq; | |
722 | struct io_context *ioc = NULL; | |
723 | const bool is_sync = rw_is_sync(rw_flags) != 0; | |
724 | int may_queue, priv; | |
725 | ||
726 | may_queue = elv_may_queue(q, rw_flags); | |
727 | if (may_queue == ELV_MQUEUE_NO) | |
728 | goto rq_starved; | |
729 | ||
730 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { | |
731 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
732 | ioc = current_io_context(GFP_ATOMIC, q->node); | |
733 | /* | |
734 | * The queue will fill after this allocation, so set | |
735 | * it as full, and mark this process as "batching". | |
736 | * This process will be allowed to complete a batch of | |
737 | * requests, others will be blocked. | |
738 | */ | |
739 | if (!blk_queue_full(q, is_sync)) { | |
740 | ioc_set_batching(q, ioc); | |
741 | blk_set_queue_full(q, is_sync); | |
742 | } else { | |
743 | if (may_queue != ELV_MQUEUE_MUST | |
744 | && !ioc_batching(q, ioc)) { | |
745 | /* | |
746 | * The queue is full and the allocating | |
747 | * process is not a "batcher", and not | |
748 | * exempted by the IO scheduler | |
749 | */ | |
750 | goto out; | |
751 | } | |
752 | } | |
753 | } | |
754 | blk_set_queue_congested(q, is_sync); | |
755 | } | |
756 | ||
757 | /* | |
758 | * Only allow batching queuers to allocate up to 50% over the defined | |
759 | * limit of requests, otherwise we could have thousands of requests | |
760 | * allocated with any setting of ->nr_requests | |
761 | */ | |
762 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) | |
763 | goto out; | |
764 | ||
765 | rl->count[is_sync]++; | |
766 | rl->starved[is_sync] = 0; | |
767 | ||
768 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
769 | if (priv) | |
770 | rl->elvpriv++; | |
771 | ||
772 | if (blk_queue_io_stat(q)) | |
773 | rw_flags |= REQ_IO_STAT; | |
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, is_sync, 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[is_sync] == 0)) | |
797 | rl->starved[is_sync] = 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 | trace_block_getrq(q, bio, rw_flags & 1); | |
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 bool is_sync = rw_is_sync(rw_flags) != 0; | |
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[is_sync], &wait, | |
835 | TASK_UNINTERRUPTIBLE); | |
836 | ||
837 | trace_block_sleeprq(q, bio, rw_flags & 1); | |
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[is_sync], &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_make_request - given a bio, allocate a corresponding struct request. | |
883 | * @q: target request queue | |
884 | * @bio: The bio describing the memory mappings that will be submitted for IO. | |
885 | * It may be a chained-bio properly constructed by block/bio layer. | |
886 | * @gfp_mask: gfp flags to be used for memory allocation | |
887 | * | |
888 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC | |
889 | * type commands. Where the struct request needs to be farther initialized by | |
890 | * the caller. It is passed a &struct bio, which describes the memory info of | |
891 | * the I/O transfer. | |
892 | * | |
893 | * The caller of blk_make_request must make sure that bi_io_vec | |
894 | * are set to describe the memory buffers. That bio_data_dir() will return | |
895 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
896 | * are properly set accordingly) | |
897 | * | |
898 | * If called under none-sleepable conditions, mapped bio buffers must not | |
899 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
900 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
901 | * BUG. | |
902 | * | |
903 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
904 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
905 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
906 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
907 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
908 | * of bio_alloc(), as that avoids the mempool deadlock. | |
909 | * If possible a big IO should be split into smaller parts when allocation | |
910 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
911 | */ | |
912 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, | |
913 | gfp_t gfp_mask) | |
914 | { | |
915 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); | |
916 | ||
917 | if (unlikely(!rq)) | |
918 | return ERR_PTR(-ENOMEM); | |
919 | ||
920 | for_each_bio(bio) { | |
921 | struct bio *bounce_bio = bio; | |
922 | int ret; | |
923 | ||
924 | blk_queue_bounce(q, &bounce_bio); | |
925 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
926 | if (unlikely(ret)) { | |
927 | blk_put_request(rq); | |
928 | return ERR_PTR(ret); | |
929 | } | |
930 | } | |
931 | ||
932 | return rq; | |
933 | } | |
934 | EXPORT_SYMBOL(blk_make_request); | |
935 | ||
936 | /** | |
937 | * blk_requeue_request - put a request back on queue | |
938 | * @q: request queue where request should be inserted | |
939 | * @rq: request to be inserted | |
940 | * | |
941 | * Description: | |
942 | * Drivers often keep queueing requests until the hardware cannot accept | |
943 | * more, when that condition happens we need to put the request back | |
944 | * on the queue. Must be called with queue lock held. | |
945 | */ | |
946 | void blk_requeue_request(struct request_queue *q, struct request *rq) | |
947 | { | |
948 | blk_delete_timer(rq); | |
949 | blk_clear_rq_complete(rq); | |
950 | trace_block_rq_requeue(q, rq); | |
951 | ||
952 | if (blk_rq_tagged(rq)) | |
953 | blk_queue_end_tag(q, rq); | |
954 | ||
955 | BUG_ON(blk_queued_rq(rq)); | |
956 | ||
957 | elv_requeue_request(q, rq); | |
958 | } | |
959 | EXPORT_SYMBOL(blk_requeue_request); | |
960 | ||
961 | /** | |
962 | * blk_insert_request - insert a special request into a request queue | |
963 | * @q: request queue where request should be inserted | |
964 | * @rq: request to be inserted | |
965 | * @at_head: insert request at head or tail of queue | |
966 | * @data: private data | |
967 | * | |
968 | * Description: | |
969 | * Many block devices need to execute commands asynchronously, so they don't | |
970 | * block the whole kernel from preemption during request execution. This is | |
971 | * accomplished normally by inserting aritficial requests tagged as | |
972 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them | |
973 | * be scheduled for actual execution by the request queue. | |
974 | * | |
975 | * We have the option of inserting the head or the tail of the queue. | |
976 | * Typically we use the tail for new ioctls and so forth. We use the head | |
977 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
978 | * host that is unable to accept a particular command. | |
979 | */ | |
980 | void blk_insert_request(struct request_queue *q, struct request *rq, | |
981 | int at_head, void *data) | |
982 | { | |
983 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
984 | unsigned long flags; | |
985 | ||
986 | /* | |
987 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
988 | * must not attempt merges on this) and that it acts as a soft | |
989 | * barrier | |
990 | */ | |
991 | rq->cmd_type = REQ_TYPE_SPECIAL; | |
992 | ||
993 | rq->special = data; | |
994 | ||
995 | spin_lock_irqsave(q->queue_lock, flags); | |
996 | ||
997 | /* | |
998 | * If command is tagged, release the tag | |
999 | */ | |
1000 | if (blk_rq_tagged(rq)) | |
1001 | blk_queue_end_tag(q, rq); | |
1002 | ||
1003 | drive_stat_acct(rq, 1); | |
1004 | __elv_add_request(q, rq, where, 0); | |
1005 | __blk_run_queue(q); | |
1006 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1007 | } | |
1008 | EXPORT_SYMBOL(blk_insert_request); | |
1009 | ||
1010 | /* | |
1011 | * add-request adds a request to the linked list. | |
1012 | * queue lock is held and interrupts disabled, as we muck with the | |
1013 | * request queue list. | |
1014 | */ | |
1015 | static inline void add_request(struct request_queue *q, struct request *req) | |
1016 | { | |
1017 | drive_stat_acct(req, 1); | |
1018 | ||
1019 | /* | |
1020 | * elevator indicated where it wants this request to be | |
1021 | * inserted at elevator_merge time | |
1022 | */ | |
1023 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
1024 | } | |
1025 | ||
1026 | static void part_round_stats_single(int cpu, struct hd_struct *part, | |
1027 | unsigned long now) | |
1028 | { | |
1029 | if (now == part->stamp) | |
1030 | return; | |
1031 | ||
1032 | if (part->in_flight) { | |
1033 | __part_stat_add(cpu, part, time_in_queue, | |
1034 | part_in_flight(part) * (now - part->stamp)); | |
1035 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1036 | } | |
1037 | part->stamp = now; | |
1038 | } | |
1039 | ||
1040 | /** | |
1041 | * part_round_stats() - Round off the performance stats on a struct disk_stats. | |
1042 | * @cpu: cpu number for stats access | |
1043 | * @part: target partition | |
1044 | * | |
1045 | * The average IO queue length and utilisation statistics are maintained | |
1046 | * by observing the current state of the queue length and the amount of | |
1047 | * time it has been in this state for. | |
1048 | * | |
1049 | * Normally, that accounting is done on IO completion, but that can result | |
1050 | * in more than a second's worth of IO being accounted for within any one | |
1051 | * second, leading to >100% utilisation. To deal with that, we call this | |
1052 | * function to do a round-off before returning the results when reading | |
1053 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1054 | * the current jiffies and restarts the counters again. | |
1055 | */ | |
1056 | void part_round_stats(int cpu, struct hd_struct *part) | |
1057 | { | |
1058 | unsigned long now = jiffies; | |
1059 | ||
1060 | if (part->partno) | |
1061 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1062 | part_round_stats_single(cpu, part, now); | |
1063 | } | |
1064 | EXPORT_SYMBOL_GPL(part_round_stats); | |
1065 | ||
1066 | /* | |
1067 | * queue lock must be held | |
1068 | */ | |
1069 | void __blk_put_request(struct request_queue *q, struct request *req) | |
1070 | { | |
1071 | if (unlikely(!q)) | |
1072 | return; | |
1073 | if (unlikely(--req->ref_count)) | |
1074 | return; | |
1075 | ||
1076 | elv_completed_request(q, req); | |
1077 | ||
1078 | /* this is a bio leak */ | |
1079 | WARN_ON(req->bio != NULL); | |
1080 | ||
1081 | /* | |
1082 | * Request may not have originated from ll_rw_blk. if not, | |
1083 | * it didn't come out of our reserved rq pools | |
1084 | */ | |
1085 | if (req->cmd_flags & REQ_ALLOCED) { | |
1086 | int is_sync = rq_is_sync(req) != 0; | |
1087 | int priv = req->cmd_flags & REQ_ELVPRIV; | |
1088 | ||
1089 | BUG_ON(!list_empty(&req->queuelist)); | |
1090 | BUG_ON(!hlist_unhashed(&req->hash)); | |
1091 | ||
1092 | blk_free_request(q, req); | |
1093 | freed_request(q, is_sync, priv); | |
1094 | } | |
1095 | } | |
1096 | EXPORT_SYMBOL_GPL(__blk_put_request); | |
1097 | ||
1098 | void blk_put_request(struct request *req) | |
1099 | { | |
1100 | unsigned long flags; | |
1101 | struct request_queue *q = req->q; | |
1102 | ||
1103 | spin_lock_irqsave(q->queue_lock, flags); | |
1104 | __blk_put_request(q, req); | |
1105 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1106 | } | |
1107 | EXPORT_SYMBOL(blk_put_request); | |
1108 | ||
1109 | void init_request_from_bio(struct request *req, struct bio *bio) | |
1110 | { | |
1111 | req->cpu = bio->bi_comp_cpu; | |
1112 | req->cmd_type = REQ_TYPE_FS; | |
1113 | ||
1114 | /* | |
1115 | * Inherit FAILFAST from bio (for read-ahead, and explicit | |
1116 | * FAILFAST). FAILFAST flags are identical for req and bio. | |
1117 | */ | |
1118 | if (bio_rw_flagged(bio, BIO_RW_AHEAD)) | |
1119 | req->cmd_flags |= REQ_FAILFAST_MASK; | |
1120 | else | |
1121 | req->cmd_flags |= bio->bi_rw & REQ_FAILFAST_MASK; | |
1122 | ||
1123 | if (unlikely(bio_rw_flagged(bio, BIO_RW_DISCARD))) { | |
1124 | req->cmd_flags |= REQ_DISCARD; | |
1125 | if (bio_rw_flagged(bio, BIO_RW_BARRIER)) | |
1126 | req->cmd_flags |= REQ_SOFTBARRIER; | |
1127 | } else if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) | |
1128 | req->cmd_flags |= REQ_HARDBARRIER; | |
1129 | ||
1130 | if (bio_rw_flagged(bio, BIO_RW_SYNCIO)) | |
1131 | req->cmd_flags |= REQ_RW_SYNC; | |
1132 | if (bio_rw_flagged(bio, BIO_RW_META)) | |
1133 | req->cmd_flags |= REQ_RW_META; | |
1134 | if (bio_rw_flagged(bio, BIO_RW_NOIDLE)) | |
1135 | req->cmd_flags |= REQ_NOIDLE; | |
1136 | ||
1137 | req->errors = 0; | |
1138 | req->__sector = bio->bi_sector; | |
1139 | req->ioprio = bio_prio(bio); | |
1140 | blk_rq_bio_prep(req->q, req, bio); | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Only disabling plugging for non-rotational devices if it does tagging | |
1145 | * as well, otherwise we do need the proper merging | |
1146 | */ | |
1147 | static inline bool queue_should_plug(struct request_queue *q) | |
1148 | { | |
1149 | return !(blk_queue_nonrot(q) && blk_queue_queuing(q)); | |
1150 | } | |
1151 | ||
1152 | static int __make_request(struct request_queue *q, struct bio *bio) | |
1153 | { | |
1154 | struct request *req; | |
1155 | int el_ret; | |
1156 | unsigned int bytes = bio->bi_size; | |
1157 | const unsigned short prio = bio_prio(bio); | |
1158 | const bool sync = bio_rw_flagged(bio, BIO_RW_SYNCIO); | |
1159 | const bool unplug = bio_rw_flagged(bio, BIO_RW_UNPLUG); | |
1160 | const unsigned int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1161 | int rw_flags; | |
1162 | ||
1163 | if (bio_rw_flagged(bio, BIO_RW_BARRIER) && bio_has_data(bio) && | |
1164 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1165 | bio_endio(bio, -EOPNOTSUPP); | |
1166 | return 0; | |
1167 | } | |
1168 | /* | |
1169 | * low level driver can indicate that it wants pages above a | |
1170 | * certain limit bounced to low memory (ie for highmem, or even | |
1171 | * ISA dma in theory) | |
1172 | */ | |
1173 | blk_queue_bounce(q, &bio); | |
1174 | ||
1175 | spin_lock_irq(q->queue_lock); | |
1176 | ||
1177 | if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER)) || elv_queue_empty(q)) | |
1178 | goto get_rq; | |
1179 | ||
1180 | el_ret = elv_merge(q, &req, bio); | |
1181 | switch (el_ret) { | |
1182 | case ELEVATOR_BACK_MERGE: | |
1183 | BUG_ON(!rq_mergeable(req)); | |
1184 | ||
1185 | if (!ll_back_merge_fn(q, req, bio)) | |
1186 | break; | |
1187 | ||
1188 | trace_block_bio_backmerge(q, bio); | |
1189 | ||
1190 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1191 | blk_rq_set_mixed_merge(req); | |
1192 | ||
1193 | req->biotail->bi_next = bio; | |
1194 | req->biotail = bio; | |
1195 | req->__data_len += bytes; | |
1196 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1197 | if (!blk_rq_cpu_valid(req)) | |
1198 | req->cpu = bio->bi_comp_cpu; | |
1199 | drive_stat_acct(req, 0); | |
1200 | if (!attempt_back_merge(q, req)) | |
1201 | elv_merged_request(q, req, el_ret); | |
1202 | goto out; | |
1203 | ||
1204 | case ELEVATOR_FRONT_MERGE: | |
1205 | BUG_ON(!rq_mergeable(req)); | |
1206 | ||
1207 | if (!ll_front_merge_fn(q, req, bio)) | |
1208 | break; | |
1209 | ||
1210 | trace_block_bio_frontmerge(q, bio); | |
1211 | ||
1212 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) { | |
1213 | blk_rq_set_mixed_merge(req); | |
1214 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
1215 | req->cmd_flags |= ff; | |
1216 | } | |
1217 | ||
1218 | bio->bi_next = req->bio; | |
1219 | req->bio = bio; | |
1220 | ||
1221 | /* | |
1222 | * may not be valid. if the low level driver said | |
1223 | * it didn't need a bounce buffer then it better | |
1224 | * not touch req->buffer either... | |
1225 | */ | |
1226 | req->buffer = bio_data(bio); | |
1227 | req->__sector = bio->bi_sector; | |
1228 | req->__data_len += bytes; | |
1229 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1230 | if (!blk_rq_cpu_valid(req)) | |
1231 | req->cpu = bio->bi_comp_cpu; | |
1232 | drive_stat_acct(req, 0); | |
1233 | if (!attempt_front_merge(q, req)) | |
1234 | elv_merged_request(q, req, el_ret); | |
1235 | goto out; | |
1236 | ||
1237 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1238 | default: | |
1239 | ; | |
1240 | } | |
1241 | ||
1242 | get_rq: | |
1243 | /* | |
1244 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1245 | * but we need to set it earlier to expose the sync flag to the | |
1246 | * rq allocator and io schedulers. | |
1247 | */ | |
1248 | rw_flags = bio_data_dir(bio); | |
1249 | if (sync) | |
1250 | rw_flags |= REQ_RW_SYNC; | |
1251 | ||
1252 | /* | |
1253 | * Grab a free request. This is might sleep but can not fail. | |
1254 | * Returns with the queue unlocked. | |
1255 | */ | |
1256 | req = get_request_wait(q, rw_flags, bio); | |
1257 | ||
1258 | /* | |
1259 | * After dropping the lock and possibly sleeping here, our request | |
1260 | * may now be mergeable after it had proven unmergeable (above). | |
1261 | * We don't worry about that case for efficiency. It won't happen | |
1262 | * often, and the elevators are able to handle it. | |
1263 | */ | |
1264 | init_request_from_bio(req, bio); | |
1265 | ||
1266 | spin_lock_irq(q->queue_lock); | |
1267 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || | |
1268 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1269 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
1270 | if (queue_should_plug(q) && elv_queue_empty(q)) | |
1271 | blk_plug_device(q); | |
1272 | add_request(q, req); | |
1273 | out: | |
1274 | if (unplug || !queue_should_plug(q)) | |
1275 | __generic_unplug_device(q); | |
1276 | spin_unlock_irq(q->queue_lock); | |
1277 | return 0; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * If bio->bi_dev is a partition, remap the location | |
1282 | */ | |
1283 | static inline void blk_partition_remap(struct bio *bio) | |
1284 | { | |
1285 | struct block_device *bdev = bio->bi_bdev; | |
1286 | ||
1287 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { | |
1288 | struct hd_struct *p = bdev->bd_part; | |
1289 | ||
1290 | bio->bi_sector += p->start_sect; | |
1291 | bio->bi_bdev = bdev->bd_contains; | |
1292 | ||
1293 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1294 | bdev->bd_dev, | |
1295 | bio->bi_sector - p->start_sect); | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | static void handle_bad_sector(struct bio *bio) | |
1300 | { | |
1301 | char b[BDEVNAME_SIZE]; | |
1302 | ||
1303 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1304 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1305 | bdevname(bio->bi_bdev, b), | |
1306 | bio->bi_rw, | |
1307 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1308 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1309 | ||
1310 | set_bit(BIO_EOF, &bio->bi_flags); | |
1311 | } | |
1312 | ||
1313 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1314 | ||
1315 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1316 | ||
1317 | static int __init setup_fail_make_request(char *str) | |
1318 | { | |
1319 | return setup_fault_attr(&fail_make_request, str); | |
1320 | } | |
1321 | __setup("fail_make_request=", setup_fail_make_request); | |
1322 | ||
1323 | static int should_fail_request(struct bio *bio) | |
1324 | { | |
1325 | struct hd_struct *part = bio->bi_bdev->bd_part; | |
1326 | ||
1327 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
1328 | return should_fail(&fail_make_request, bio->bi_size); | |
1329 | ||
1330 | return 0; | |
1331 | } | |
1332 | ||
1333 | static int __init fail_make_request_debugfs(void) | |
1334 | { | |
1335 | return init_fault_attr_dentries(&fail_make_request, | |
1336 | "fail_make_request"); | |
1337 | } | |
1338 | ||
1339 | late_initcall(fail_make_request_debugfs); | |
1340 | ||
1341 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1342 | ||
1343 | static inline int should_fail_request(struct bio *bio) | |
1344 | { | |
1345 | return 0; | |
1346 | } | |
1347 | ||
1348 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1349 | ||
1350 | /* | |
1351 | * Check whether this bio extends beyond the end of the device. | |
1352 | */ | |
1353 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1354 | { | |
1355 | sector_t maxsector; | |
1356 | ||
1357 | if (!nr_sectors) | |
1358 | return 0; | |
1359 | ||
1360 | /* Test device or partition size, when known. */ | |
1361 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1362 | if (maxsector) { | |
1363 | sector_t sector = bio->bi_sector; | |
1364 | ||
1365 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1366 | /* | |
1367 | * This may well happen - the kernel calls bread() | |
1368 | * without checking the size of the device, e.g., when | |
1369 | * mounting a device. | |
1370 | */ | |
1371 | handle_bad_sector(bio); | |
1372 | return 1; | |
1373 | } | |
1374 | } | |
1375 | ||
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | /** | |
1380 | * generic_make_request - hand a buffer to its device driver for I/O | |
1381 | * @bio: The bio describing the location in memory and on the device. | |
1382 | * | |
1383 | * generic_make_request() is used to make I/O requests of block | |
1384 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1385 | * to be done. | |
1386 | * | |
1387 | * generic_make_request() does not return any status. The | |
1388 | * success/failure status of the request, along with notification of | |
1389 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1390 | * function described (one day) else where. | |
1391 | * | |
1392 | * The caller of generic_make_request must make sure that bi_io_vec | |
1393 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1394 | * set to describe the device address, and the | |
1395 | * bi_end_io and optionally bi_private are set to describe how | |
1396 | * completion notification should be signaled. | |
1397 | * | |
1398 | * generic_make_request and the drivers it calls may use bi_next if this | |
1399 | * bio happens to be merged with someone else, and may change bi_dev and | |
1400 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1401 | * should NOT be depended on after the call to generic_make_request. | |
1402 | */ | |
1403 | static inline void __generic_make_request(struct bio *bio) | |
1404 | { | |
1405 | struct request_queue *q; | |
1406 | sector_t old_sector; | |
1407 | int ret, nr_sectors = bio_sectors(bio); | |
1408 | dev_t old_dev; | |
1409 | int err = -EIO; | |
1410 | ||
1411 | might_sleep(); | |
1412 | ||
1413 | if (bio_check_eod(bio, nr_sectors)) | |
1414 | goto end_io; | |
1415 | ||
1416 | /* | |
1417 | * Resolve the mapping until finished. (drivers are | |
1418 | * still free to implement/resolve their own stacking | |
1419 | * by explicitly returning 0) | |
1420 | * | |
1421 | * NOTE: we don't repeat the blk_size check for each new device. | |
1422 | * Stacking drivers are expected to know what they are doing. | |
1423 | */ | |
1424 | old_sector = -1; | |
1425 | old_dev = 0; | |
1426 | do { | |
1427 | char b[BDEVNAME_SIZE]; | |
1428 | ||
1429 | q = bdev_get_queue(bio->bi_bdev); | |
1430 | if (unlikely(!q)) { | |
1431 | printk(KERN_ERR | |
1432 | "generic_make_request: Trying to access " | |
1433 | "nonexistent block-device %s (%Lu)\n", | |
1434 | bdevname(bio->bi_bdev, b), | |
1435 | (long long) bio->bi_sector); | |
1436 | goto end_io; | |
1437 | } | |
1438 | ||
1439 | if (unlikely(nr_sectors > queue_max_hw_sectors(q))) { | |
1440 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", | |
1441 | bdevname(bio->bi_bdev, b), | |
1442 | bio_sectors(bio), | |
1443 | queue_max_hw_sectors(q)); | |
1444 | goto end_io; | |
1445 | } | |
1446 | ||
1447 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) | |
1448 | goto end_io; | |
1449 | ||
1450 | if (should_fail_request(bio)) | |
1451 | goto end_io; | |
1452 | ||
1453 | /* | |
1454 | * If this device has partitions, remap block n | |
1455 | * of partition p to block n+start(p) of the disk. | |
1456 | */ | |
1457 | blk_partition_remap(bio); | |
1458 | ||
1459 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) | |
1460 | goto end_io; | |
1461 | ||
1462 | if (old_sector != -1) | |
1463 | trace_block_remap(q, bio, old_dev, old_sector); | |
1464 | ||
1465 | old_sector = bio->bi_sector; | |
1466 | old_dev = bio->bi_bdev->bd_dev; | |
1467 | ||
1468 | if (bio_check_eod(bio, nr_sectors)) | |
1469 | goto end_io; | |
1470 | ||
1471 | if (bio_rw_flagged(bio, BIO_RW_DISCARD) && | |
1472 | !blk_queue_discard(q)) { | |
1473 | err = -EOPNOTSUPP; | |
1474 | goto end_io; | |
1475 | } | |
1476 | ||
1477 | trace_block_bio_queue(q, bio); | |
1478 | ||
1479 | ret = q->make_request_fn(q, bio); | |
1480 | } while (ret); | |
1481 | ||
1482 | return; | |
1483 | ||
1484 | end_io: | |
1485 | bio_endio(bio, err); | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * We only want one ->make_request_fn to be active at a time, | |
1490 | * else stack usage with stacked devices could be a problem. | |
1491 | * So use current->bio_{list,tail} to keep a list of requests | |
1492 | * submited by a make_request_fn function. | |
1493 | * current->bio_tail is also used as a flag to say if | |
1494 | * generic_make_request is currently active in this task or not. | |
1495 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1496 | * then a make_request is active, and new requests should be added | |
1497 | * at the tail | |
1498 | */ | |
1499 | void generic_make_request(struct bio *bio) | |
1500 | { | |
1501 | if (current->bio_tail) { | |
1502 | /* make_request is active */ | |
1503 | *(current->bio_tail) = bio; | |
1504 | bio->bi_next = NULL; | |
1505 | current->bio_tail = &bio->bi_next; | |
1506 | return; | |
1507 | } | |
1508 | /* following loop may be a bit non-obvious, and so deserves some | |
1509 | * explanation. | |
1510 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1511 | * ensure that) so we have a list with a single bio. | |
1512 | * We pretend that we have just taken it off a longer list, so | |
1513 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1514 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1515 | * added. __generic_make_request may indeed add some more bios | |
1516 | * through a recursive call to generic_make_request. If it | |
1517 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1518 | * from the top. In this case we really did just take the bio | |
1519 | * of the top of the list (no pretending) and so fixup bio_list and | |
1520 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1521 | * | |
1522 | * The loop was structured like this to make only one call to | |
1523 | * __generic_make_request (which is important as it is large and | |
1524 | * inlined) and to keep the structure simple. | |
1525 | */ | |
1526 | BUG_ON(bio->bi_next); | |
1527 | do { | |
1528 | current->bio_list = bio->bi_next; | |
1529 | if (bio->bi_next == NULL) | |
1530 | current->bio_tail = ¤t->bio_list; | |
1531 | else | |
1532 | bio->bi_next = NULL; | |
1533 | __generic_make_request(bio); | |
1534 | bio = current->bio_list; | |
1535 | } while (bio); | |
1536 | current->bio_tail = NULL; /* deactivate */ | |
1537 | } | |
1538 | EXPORT_SYMBOL(generic_make_request); | |
1539 | ||
1540 | /** | |
1541 | * submit_bio - submit a bio to the block device layer for I/O | |
1542 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
1543 | * @bio: The &struct bio which describes the I/O | |
1544 | * | |
1545 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1546 | * uses that function to do most of the work. Both are fairly rough | |
1547 | * interfaces; @bio must be presetup and ready for I/O. | |
1548 | * | |
1549 | */ | |
1550 | void submit_bio(int rw, struct bio *bio) | |
1551 | { | |
1552 | int count = bio_sectors(bio); | |
1553 | ||
1554 | bio->bi_rw |= rw; | |
1555 | ||
1556 | /* | |
1557 | * If it's a regular read/write or a barrier with data attached, | |
1558 | * go through the normal accounting stuff before submission. | |
1559 | */ | |
1560 | if (bio_has_data(bio)) { | |
1561 | if (rw & WRITE) { | |
1562 | count_vm_events(PGPGOUT, count); | |
1563 | } else { | |
1564 | task_io_account_read(bio->bi_size); | |
1565 | count_vm_events(PGPGIN, count); | |
1566 | } | |
1567 | ||
1568 | if (unlikely(block_dump)) { | |
1569 | char b[BDEVNAME_SIZE]; | |
1570 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
1571 | current->comm, task_pid_nr(current), | |
1572 | (rw & WRITE) ? "WRITE" : "READ", | |
1573 | (unsigned long long)bio->bi_sector, | |
1574 | bdevname(bio->bi_bdev, b)); | |
1575 | } | |
1576 | } | |
1577 | ||
1578 | generic_make_request(bio); | |
1579 | } | |
1580 | EXPORT_SYMBOL(submit_bio); | |
1581 | ||
1582 | /** | |
1583 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1584 | * @q: the queue | |
1585 | * @rq: the request being checked | |
1586 | * | |
1587 | * Description: | |
1588 | * @rq may have been made based on weaker limitations of upper-level queues | |
1589 | * in request stacking drivers, and it may violate the limitation of @q. | |
1590 | * Since the block layer and the underlying device driver trust @rq | |
1591 | * after it is inserted to @q, it should be checked against @q before | |
1592 | * the insertion using this generic function. | |
1593 | * | |
1594 | * This function should also be useful for request stacking drivers | |
1595 | * in some cases below, so export this fuction. | |
1596 | * Request stacking drivers like request-based dm may change the queue | |
1597 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1598 | * Such request stacking drivers should check those requests agaist | |
1599 | * the new queue limits again when they dispatch those requests, | |
1600 | * although such checkings are also done against the old queue limits | |
1601 | * when submitting requests. | |
1602 | */ | |
1603 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1604 | { | |
1605 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || | |
1606 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
1607 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1608 | return -EIO; | |
1609 | } | |
1610 | ||
1611 | /* | |
1612 | * queue's settings related to segment counting like q->bounce_pfn | |
1613 | * may differ from that of other stacking queues. | |
1614 | * Recalculate it to check the request correctly on this queue's | |
1615 | * limitation. | |
1616 | */ | |
1617 | blk_recalc_rq_segments(rq); | |
1618 | if (rq->nr_phys_segments > queue_max_phys_segments(q) || | |
1619 | rq->nr_phys_segments > queue_max_hw_segments(q)) { | |
1620 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1621 | return -EIO; | |
1622 | } | |
1623 | ||
1624 | return 0; | |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1627 | ||
1628 | /** | |
1629 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1630 | * @q: the queue to submit the request | |
1631 | * @rq: the request being queued | |
1632 | */ | |
1633 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1634 | { | |
1635 | unsigned long flags; | |
1636 | ||
1637 | if (blk_rq_check_limits(q, rq)) | |
1638 | return -EIO; | |
1639 | ||
1640 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1641 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1642 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1643 | return -EIO; | |
1644 | #endif | |
1645 | ||
1646 | spin_lock_irqsave(q->queue_lock, flags); | |
1647 | ||
1648 | /* | |
1649 | * Submitting request must be dequeued before calling this function | |
1650 | * because it will be linked to another request_queue | |
1651 | */ | |
1652 | BUG_ON(blk_queued_rq(rq)); | |
1653 | ||
1654 | drive_stat_acct(rq, 1); | |
1655 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1656 | ||
1657 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1658 | ||
1659 | return 0; | |
1660 | } | |
1661 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1662 | ||
1663 | /** | |
1664 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1665 | * @rq: request to examine | |
1666 | * | |
1667 | * Description: | |
1668 | * A request could be merge of IOs which require different failure | |
1669 | * handling. This function determines the number of bytes which | |
1670 | * can be failed from the beginning of the request without | |
1671 | * crossing into area which need to be retried further. | |
1672 | * | |
1673 | * Return: | |
1674 | * The number of bytes to fail. | |
1675 | * | |
1676 | * Context: | |
1677 | * queue_lock must be held. | |
1678 | */ | |
1679 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1680 | { | |
1681 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1682 | unsigned int bytes = 0; | |
1683 | struct bio *bio; | |
1684 | ||
1685 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1686 | return blk_rq_bytes(rq); | |
1687 | ||
1688 | /* | |
1689 | * Currently the only 'mixing' which can happen is between | |
1690 | * different fastfail types. We can safely fail portions | |
1691 | * which have all the failfast bits that the first one has - | |
1692 | * the ones which are at least as eager to fail as the first | |
1693 | * one. | |
1694 | */ | |
1695 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1696 | if ((bio->bi_rw & ff) != ff) | |
1697 | break; | |
1698 | bytes += bio->bi_size; | |
1699 | } | |
1700 | ||
1701 | /* this could lead to infinite loop */ | |
1702 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1703 | return bytes; | |
1704 | } | |
1705 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1706 | ||
1707 | static void blk_account_io_completion(struct request *req, unsigned int bytes) | |
1708 | { | |
1709 | if (blk_do_io_stat(req)) { | |
1710 | const int rw = rq_data_dir(req); | |
1711 | struct hd_struct *part; | |
1712 | int cpu; | |
1713 | ||
1714 | cpu = part_stat_lock(); | |
1715 | part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); | |
1716 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
1717 | part_stat_unlock(); | |
1718 | } | |
1719 | } | |
1720 | ||
1721 | static void blk_account_io_done(struct request *req) | |
1722 | { | |
1723 | /* | |
1724 | * Account IO completion. bar_rq isn't accounted as a normal | |
1725 | * IO on queueing nor completion. Accounting the containing | |
1726 | * request is enough. | |
1727 | */ | |
1728 | if (blk_do_io_stat(req) && req != &req->q->bar_rq) { | |
1729 | unsigned long duration = jiffies - req->start_time; | |
1730 | const int rw = rq_data_dir(req); | |
1731 | struct hd_struct *part; | |
1732 | int cpu; | |
1733 | ||
1734 | cpu = part_stat_lock(); | |
1735 | part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req)); | |
1736 | ||
1737 | part_stat_inc(cpu, part, ios[rw]); | |
1738 | part_stat_add(cpu, part, ticks[rw], duration); | |
1739 | part_round_stats(cpu, part); | |
1740 | part_dec_in_flight(part, rw); | |
1741 | ||
1742 | part_stat_unlock(); | |
1743 | } | |
1744 | } | |
1745 | ||
1746 | /** | |
1747 | * blk_peek_request - peek at the top of a request queue | |
1748 | * @q: request queue to peek at | |
1749 | * | |
1750 | * Description: | |
1751 | * Return the request at the top of @q. The returned request | |
1752 | * should be started using blk_start_request() before LLD starts | |
1753 | * processing it. | |
1754 | * | |
1755 | * Return: | |
1756 | * Pointer to the request at the top of @q if available. Null | |
1757 | * otherwise. | |
1758 | * | |
1759 | * Context: | |
1760 | * queue_lock must be held. | |
1761 | */ | |
1762 | struct request *blk_peek_request(struct request_queue *q) | |
1763 | { | |
1764 | struct request *rq; | |
1765 | int ret; | |
1766 | ||
1767 | while ((rq = __elv_next_request(q)) != NULL) { | |
1768 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1769 | /* | |
1770 | * This is the first time the device driver | |
1771 | * sees this request (possibly after | |
1772 | * requeueing). Notify IO scheduler. | |
1773 | */ | |
1774 | if (blk_sorted_rq(rq)) | |
1775 | elv_activate_rq(q, rq); | |
1776 | ||
1777 | /* | |
1778 | * just mark as started even if we don't start | |
1779 | * it, a request that has been delayed should | |
1780 | * not be passed by new incoming requests | |
1781 | */ | |
1782 | rq->cmd_flags |= REQ_STARTED; | |
1783 | trace_block_rq_issue(q, rq); | |
1784 | } | |
1785 | ||
1786 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1787 | q->end_sector = rq_end_sector(rq); | |
1788 | q->boundary_rq = NULL; | |
1789 | } | |
1790 | ||
1791 | if (rq->cmd_flags & REQ_DONTPREP) | |
1792 | break; | |
1793 | ||
1794 | if (q->dma_drain_size && blk_rq_bytes(rq)) { | |
1795 | /* | |
1796 | * make sure space for the drain appears we | |
1797 | * know we can do this because max_hw_segments | |
1798 | * has been adjusted to be one fewer than the | |
1799 | * device can handle | |
1800 | */ | |
1801 | rq->nr_phys_segments++; | |
1802 | } | |
1803 | ||
1804 | if (!q->prep_rq_fn) | |
1805 | break; | |
1806 | ||
1807 | ret = q->prep_rq_fn(q, rq); | |
1808 | if (ret == BLKPREP_OK) { | |
1809 | break; | |
1810 | } else if (ret == BLKPREP_DEFER) { | |
1811 | /* | |
1812 | * the request may have been (partially) prepped. | |
1813 | * we need to keep this request in the front to | |
1814 | * avoid resource deadlock. REQ_STARTED will | |
1815 | * prevent other fs requests from passing this one. | |
1816 | */ | |
1817 | if (q->dma_drain_size && blk_rq_bytes(rq) && | |
1818 | !(rq->cmd_flags & REQ_DONTPREP)) { | |
1819 | /* | |
1820 | * remove the space for the drain we added | |
1821 | * so that we don't add it again | |
1822 | */ | |
1823 | --rq->nr_phys_segments; | |
1824 | } | |
1825 | ||
1826 | rq = NULL; | |
1827 | break; | |
1828 | } else if (ret == BLKPREP_KILL) { | |
1829 | rq->cmd_flags |= REQ_QUIET; | |
1830 | /* | |
1831 | * Mark this request as started so we don't trigger | |
1832 | * any debug logic in the end I/O path. | |
1833 | */ | |
1834 | blk_start_request(rq); | |
1835 | __blk_end_request_all(rq, -EIO); | |
1836 | } else { | |
1837 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1838 | break; | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | return rq; | |
1843 | } | |
1844 | EXPORT_SYMBOL(blk_peek_request); | |
1845 | ||
1846 | void blk_dequeue_request(struct request *rq) | |
1847 | { | |
1848 | struct request_queue *q = rq->q; | |
1849 | ||
1850 | BUG_ON(list_empty(&rq->queuelist)); | |
1851 | BUG_ON(ELV_ON_HASH(rq)); | |
1852 | ||
1853 | list_del_init(&rq->queuelist); | |
1854 | ||
1855 | /* | |
1856 | * the time frame between a request being removed from the lists | |
1857 | * and to it is freed is accounted as io that is in progress at | |
1858 | * the driver side. | |
1859 | */ | |
1860 | if (blk_account_rq(rq)) { | |
1861 | q->in_flight[rq_is_sync(rq)]++; | |
1862 | /* | |
1863 | * Mark this device as supporting hardware queuing, if | |
1864 | * we have more IOs in flight than 4. | |
1865 | */ | |
1866 | if (!blk_queue_queuing(q) && queue_in_flight(q) > 4) | |
1867 | set_bit(QUEUE_FLAG_CQ, &q->queue_flags); | |
1868 | } | |
1869 | } | |
1870 | ||
1871 | /** | |
1872 | * blk_start_request - start request processing on the driver | |
1873 | * @req: request to dequeue | |
1874 | * | |
1875 | * Description: | |
1876 | * Dequeue @req and start timeout timer on it. This hands off the | |
1877 | * request to the driver. | |
1878 | * | |
1879 | * Block internal functions which don't want to start timer should | |
1880 | * call blk_dequeue_request(). | |
1881 | * | |
1882 | * Context: | |
1883 | * queue_lock must be held. | |
1884 | */ | |
1885 | void blk_start_request(struct request *req) | |
1886 | { | |
1887 | blk_dequeue_request(req); | |
1888 | ||
1889 | /* | |
1890 | * We are now handing the request to the hardware, initialize | |
1891 | * resid_len to full count and add the timeout handler. | |
1892 | */ | |
1893 | req->resid_len = blk_rq_bytes(req); | |
1894 | if (unlikely(blk_bidi_rq(req))) | |
1895 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
1896 | ||
1897 | blk_add_timer(req); | |
1898 | } | |
1899 | EXPORT_SYMBOL(blk_start_request); | |
1900 | ||
1901 | /** | |
1902 | * blk_fetch_request - fetch a request from a request queue | |
1903 | * @q: request queue to fetch a request from | |
1904 | * | |
1905 | * Description: | |
1906 | * Return the request at the top of @q. The request is started on | |
1907 | * return and LLD can start processing it immediately. | |
1908 | * | |
1909 | * Return: | |
1910 | * Pointer to the request at the top of @q if available. Null | |
1911 | * otherwise. | |
1912 | * | |
1913 | * Context: | |
1914 | * queue_lock must be held. | |
1915 | */ | |
1916 | struct request *blk_fetch_request(struct request_queue *q) | |
1917 | { | |
1918 | struct request *rq; | |
1919 | ||
1920 | rq = blk_peek_request(q); | |
1921 | if (rq) | |
1922 | blk_start_request(rq); | |
1923 | return rq; | |
1924 | } | |
1925 | EXPORT_SYMBOL(blk_fetch_request); | |
1926 | ||
1927 | /** | |
1928 | * blk_update_request - Special helper function for request stacking drivers | |
1929 | * @req: the request being processed | |
1930 | * @error: %0 for success, < %0 for error | |
1931 | * @nr_bytes: number of bytes to complete @req | |
1932 | * | |
1933 | * Description: | |
1934 | * Ends I/O on a number of bytes attached to @req, but doesn't complete | |
1935 | * the request structure even if @req doesn't have leftover. | |
1936 | * If @req has leftover, sets it up for the next range of segments. | |
1937 | * | |
1938 | * This special helper function is only for request stacking drivers | |
1939 | * (e.g. request-based dm) so that they can handle partial completion. | |
1940 | * Actual device drivers should use blk_end_request instead. | |
1941 | * | |
1942 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
1943 | * %false return from this function. | |
1944 | * | |
1945 | * Return: | |
1946 | * %false - this request doesn't have any more data | |
1947 | * %true - this request has more data | |
1948 | **/ | |
1949 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) | |
1950 | { | |
1951 | int total_bytes, bio_nbytes, next_idx = 0; | |
1952 | struct bio *bio; | |
1953 | ||
1954 | if (!req->bio) | |
1955 | return false; | |
1956 | ||
1957 | trace_block_rq_complete(req->q, req); | |
1958 | ||
1959 | /* | |
1960 | * For fs requests, rq is just carrier of independent bio's | |
1961 | * and each partial completion should be handled separately. | |
1962 | * Reset per-request error on each partial completion. | |
1963 | * | |
1964 | * TODO: tj: This is too subtle. It would be better to let | |
1965 | * low level drivers do what they see fit. | |
1966 | */ | |
1967 | if (blk_fs_request(req)) | |
1968 | req->errors = 0; | |
1969 | ||
1970 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { | |
1971 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1972 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
1973 | (unsigned long long)blk_rq_pos(req)); | |
1974 | } | |
1975 | ||
1976 | blk_account_io_completion(req, nr_bytes); | |
1977 | ||
1978 | total_bytes = bio_nbytes = 0; | |
1979 | while ((bio = req->bio) != NULL) { | |
1980 | int nbytes; | |
1981 | ||
1982 | if (nr_bytes >= bio->bi_size) { | |
1983 | req->bio = bio->bi_next; | |
1984 | nbytes = bio->bi_size; | |
1985 | req_bio_endio(req, bio, nbytes, error); | |
1986 | next_idx = 0; | |
1987 | bio_nbytes = 0; | |
1988 | } else { | |
1989 | int idx = bio->bi_idx + next_idx; | |
1990 | ||
1991 | if (unlikely(idx >= bio->bi_vcnt)) { | |
1992 | blk_dump_rq_flags(req, "__end_that"); | |
1993 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", | |
1994 | __func__, idx, bio->bi_vcnt); | |
1995 | break; | |
1996 | } | |
1997 | ||
1998 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1999 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2000 | ||
2001 | /* | |
2002 | * not a complete bvec done | |
2003 | */ | |
2004 | if (unlikely(nbytes > nr_bytes)) { | |
2005 | bio_nbytes += nr_bytes; | |
2006 | total_bytes += nr_bytes; | |
2007 | break; | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * advance to the next vector | |
2012 | */ | |
2013 | next_idx++; | |
2014 | bio_nbytes += nbytes; | |
2015 | } | |
2016 | ||
2017 | total_bytes += nbytes; | |
2018 | nr_bytes -= nbytes; | |
2019 | ||
2020 | bio = req->bio; | |
2021 | if (bio) { | |
2022 | /* | |
2023 | * end more in this run, or just return 'not-done' | |
2024 | */ | |
2025 | if (unlikely(nr_bytes <= 0)) | |
2026 | break; | |
2027 | } | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * completely done | |
2032 | */ | |
2033 | if (!req->bio) { | |
2034 | /* | |
2035 | * Reset counters so that the request stacking driver | |
2036 | * can find how many bytes remain in the request | |
2037 | * later. | |
2038 | */ | |
2039 | req->__data_len = 0; | |
2040 | return false; | |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * if the request wasn't completed, update state | |
2045 | */ | |
2046 | if (bio_nbytes) { | |
2047 | req_bio_endio(req, bio, bio_nbytes, error); | |
2048 | bio->bi_idx += next_idx; | |
2049 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2050 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2051 | } | |
2052 | ||
2053 | req->__data_len -= total_bytes; | |
2054 | req->buffer = bio_data(req->bio); | |
2055 | ||
2056 | /* update sector only for requests with clear definition of sector */ | |
2057 | if (blk_fs_request(req) || blk_discard_rq(req)) | |
2058 | req->__sector += total_bytes >> 9; | |
2059 | ||
2060 | /* mixed attributes always follow the first bio */ | |
2061 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2062 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2063 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2064 | } | |
2065 | ||
2066 | /* | |
2067 | * If total number of sectors is less than the first segment | |
2068 | * size, something has gone terribly wrong. | |
2069 | */ | |
2070 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
2071 | printk(KERN_ERR "blk: request botched\n"); | |
2072 | req->__data_len = blk_rq_cur_bytes(req); | |
2073 | } | |
2074 | ||
2075 | /* recalculate the number of segments */ | |
2076 | blk_recalc_rq_segments(req); | |
2077 | ||
2078 | return true; | |
2079 | } | |
2080 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2081 | ||
2082 | static bool blk_update_bidi_request(struct request *rq, int error, | |
2083 | unsigned int nr_bytes, | |
2084 | unsigned int bidi_bytes) | |
2085 | { | |
2086 | if (blk_update_request(rq, error, nr_bytes)) | |
2087 | return true; | |
2088 | ||
2089 | /* Bidi request must be completed as a whole */ | |
2090 | if (unlikely(blk_bidi_rq(rq)) && | |
2091 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2092 | return true; | |
2093 | ||
2094 | add_disk_randomness(rq->rq_disk); | |
2095 | ||
2096 | return false; | |
2097 | } | |
2098 | ||
2099 | /* | |
2100 | * queue lock must be held | |
2101 | */ | |
2102 | static void blk_finish_request(struct request *req, int error) | |
2103 | { | |
2104 | if (blk_rq_tagged(req)) | |
2105 | blk_queue_end_tag(req->q, req); | |
2106 | ||
2107 | BUG_ON(blk_queued_rq(req)); | |
2108 | ||
2109 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
2110 | laptop_io_completion(); | |
2111 | ||
2112 | blk_delete_timer(req); | |
2113 | ||
2114 | blk_account_io_done(req); | |
2115 | ||
2116 | if (req->end_io) | |
2117 | req->end_io(req, error); | |
2118 | else { | |
2119 | if (blk_bidi_rq(req)) | |
2120 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2121 | ||
2122 | __blk_put_request(req->q, req); | |
2123 | } | |
2124 | } | |
2125 | ||
2126 | /** | |
2127 | * blk_end_bidi_request - Complete a bidi request | |
2128 | * @rq: the request to complete | |
2129 | * @error: %0 for success, < %0 for error | |
2130 | * @nr_bytes: number of bytes to complete @rq | |
2131 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2132 | * | |
2133 | * Description: | |
2134 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2135 | * Drivers that supports bidi can safely call this member for any | |
2136 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2137 | * just ignored. | |
2138 | * | |
2139 | * Return: | |
2140 | * %false - we are done with this request | |
2141 | * %true - still buffers pending for this request | |
2142 | **/ | |
2143 | static bool blk_end_bidi_request(struct request *rq, int error, | |
2144 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
2145 | { | |
2146 | struct request_queue *q = rq->q; | |
2147 | unsigned long flags; | |
2148 | ||
2149 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
2150 | return true; | |
2151 | ||
2152 | spin_lock_irqsave(q->queue_lock, flags); | |
2153 | blk_finish_request(rq, error); | |
2154 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2155 | ||
2156 | return false; | |
2157 | } | |
2158 | ||
2159 | /** | |
2160 | * __blk_end_bidi_request - Complete a bidi request with queue lock held | |
2161 | * @rq: the request to complete | |
2162 | * @error: %0 for success, < %0 for error | |
2163 | * @nr_bytes: number of bytes to complete @rq | |
2164 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2165 | * | |
2166 | * Description: | |
2167 | * Identical to blk_end_bidi_request() except that queue lock is | |
2168 | * assumed to be locked on entry and remains so on return. | |
2169 | * | |
2170 | * Return: | |
2171 | * %false - we are done with this request | |
2172 | * %true - still buffers pending for this request | |
2173 | **/ | |
2174 | static bool __blk_end_bidi_request(struct request *rq, int error, | |
2175 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
2176 | { | |
2177 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) | |
2178 | return true; | |
2179 | ||
2180 | blk_finish_request(rq, error); | |
2181 | ||
2182 | return false; | |
2183 | } | |
2184 | ||
2185 | /** | |
2186 | * blk_end_request - Helper function for drivers to complete the request. | |
2187 | * @rq: the request being processed | |
2188 | * @error: %0 for success, < %0 for error | |
2189 | * @nr_bytes: number of bytes to complete | |
2190 | * | |
2191 | * Description: | |
2192 | * Ends I/O on a number of bytes attached to @rq. | |
2193 | * If @rq has leftover, sets it up for the next range of segments. | |
2194 | * | |
2195 | * Return: | |
2196 | * %false - we are done with this request | |
2197 | * %true - still buffers pending for this request | |
2198 | **/ | |
2199 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) | |
2200 | { | |
2201 | return blk_end_bidi_request(rq, error, nr_bytes, 0); | |
2202 | } | |
2203 | EXPORT_SYMBOL(blk_end_request); | |
2204 | ||
2205 | /** | |
2206 | * blk_end_request_all - Helper function for drives to finish the request. | |
2207 | * @rq: the request to finish | |
2208 | * @error: %0 for success, < %0 for error | |
2209 | * | |
2210 | * Description: | |
2211 | * Completely finish @rq. | |
2212 | */ | |
2213 | void blk_end_request_all(struct request *rq, int error) | |
2214 | { | |
2215 | bool pending; | |
2216 | unsigned int bidi_bytes = 0; | |
2217 | ||
2218 | if (unlikely(blk_bidi_rq(rq))) | |
2219 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2220 | ||
2221 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2222 | BUG_ON(pending); | |
2223 | } | |
2224 | EXPORT_SYMBOL(blk_end_request_all); | |
2225 | ||
2226 | /** | |
2227 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2228 | * @rq: the request to finish the current chunk for | |
2229 | * @error: %0 for success, < %0 for error | |
2230 | * | |
2231 | * Description: | |
2232 | * Complete the current consecutively mapped chunk from @rq. | |
2233 | * | |
2234 | * Return: | |
2235 | * %false - we are done with this request | |
2236 | * %true - still buffers pending for this request | |
2237 | */ | |
2238 | bool blk_end_request_cur(struct request *rq, int error) | |
2239 | { | |
2240 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
2241 | } | |
2242 | EXPORT_SYMBOL(blk_end_request_cur); | |
2243 | ||
2244 | /** | |
2245 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2246 | * @rq: the request to finish till the next failure boundary for | |
2247 | * @error: must be negative errno | |
2248 | * | |
2249 | * Description: | |
2250 | * Complete @rq till the next failure boundary. | |
2251 | * | |
2252 | * Return: | |
2253 | * %false - we are done with this request | |
2254 | * %true - still buffers pending for this request | |
2255 | */ | |
2256 | bool blk_end_request_err(struct request *rq, int error) | |
2257 | { | |
2258 | WARN_ON(error >= 0); | |
2259 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2260 | } | |
2261 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2262 | ||
2263 | /** | |
2264 | * __blk_end_request - Helper function for drivers to complete the request. | |
2265 | * @rq: the request being processed | |
2266 | * @error: %0 for success, < %0 for error | |
2267 | * @nr_bytes: number of bytes to complete | |
2268 | * | |
2269 | * Description: | |
2270 | * Must be called with queue lock held unlike blk_end_request(). | |
2271 | * | |
2272 | * Return: | |
2273 | * %false - we are done with this request | |
2274 | * %true - still buffers pending for this request | |
2275 | **/ | |
2276 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) | |
2277 | { | |
2278 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); | |
2279 | } | |
2280 | EXPORT_SYMBOL(__blk_end_request); | |
2281 | ||
2282 | /** | |
2283 | * __blk_end_request_all - Helper function for drives to finish the request. | |
2284 | * @rq: the request to finish | |
2285 | * @error: %0 for success, < %0 for error | |
2286 | * | |
2287 | * Description: | |
2288 | * Completely finish @rq. Must be called with queue lock held. | |
2289 | */ | |
2290 | void __blk_end_request_all(struct request *rq, int error) | |
2291 | { | |
2292 | bool pending; | |
2293 | unsigned int bidi_bytes = 0; | |
2294 | ||
2295 | if (unlikely(blk_bidi_rq(rq))) | |
2296 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2297 | ||
2298 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2299 | BUG_ON(pending); | |
2300 | } | |
2301 | EXPORT_SYMBOL(__blk_end_request_all); | |
2302 | ||
2303 | /** | |
2304 | * __blk_end_request_cur - Helper function to finish the current request chunk. | |
2305 | * @rq: the request to finish the current chunk for | |
2306 | * @error: %0 for success, < %0 for error | |
2307 | * | |
2308 | * Description: | |
2309 | * Complete the current consecutively mapped chunk from @rq. Must | |
2310 | * be called with queue lock held. | |
2311 | * | |
2312 | * Return: | |
2313 | * %false - we are done with this request | |
2314 | * %true - still buffers pending for this request | |
2315 | */ | |
2316 | bool __blk_end_request_cur(struct request *rq, int error) | |
2317 | { | |
2318 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
2319 | } | |
2320 | EXPORT_SYMBOL(__blk_end_request_cur); | |
2321 | ||
2322 | /** | |
2323 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2324 | * @rq: the request to finish till the next failure boundary for | |
2325 | * @error: must be negative errno | |
2326 | * | |
2327 | * Description: | |
2328 | * Complete @rq till the next failure boundary. Must be called | |
2329 | * with queue lock held. | |
2330 | * | |
2331 | * Return: | |
2332 | * %false - we are done with this request | |
2333 | * %true - still buffers pending for this request | |
2334 | */ | |
2335 | bool __blk_end_request_err(struct request *rq, int error) | |
2336 | { | |
2337 | WARN_ON(error >= 0); | |
2338 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2339 | } | |
2340 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2341 | ||
2342 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, | |
2343 | struct bio *bio) | |
2344 | { | |
2345 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ | |
2346 | rq->cmd_flags |= bio->bi_rw & REQ_RW; | |
2347 | ||
2348 | if (bio_has_data(bio)) { | |
2349 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
2350 | rq->buffer = bio_data(bio); | |
2351 | } | |
2352 | rq->__data_len = bio->bi_size; | |
2353 | rq->bio = rq->biotail = bio; | |
2354 | ||
2355 | if (bio->bi_bdev) | |
2356 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2357 | } | |
2358 | ||
2359 | /** | |
2360 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2361 | * @q : the queue of the device being checked | |
2362 | * | |
2363 | * Description: | |
2364 | * Check if underlying low-level drivers of a device are busy. | |
2365 | * If the drivers want to export their busy state, they must set own | |
2366 | * exporting function using blk_queue_lld_busy() first. | |
2367 | * | |
2368 | * Basically, this function is used only by request stacking drivers | |
2369 | * to stop dispatching requests to underlying devices when underlying | |
2370 | * devices are busy. This behavior helps more I/O merging on the queue | |
2371 | * of the request stacking driver and prevents I/O throughput regression | |
2372 | * on burst I/O load. | |
2373 | * | |
2374 | * Return: | |
2375 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2376 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2377 | */ | |
2378 | int blk_lld_busy(struct request_queue *q) | |
2379 | { | |
2380 | if (q->lld_busy_fn) | |
2381 | return q->lld_busy_fn(q); | |
2382 | ||
2383 | return 0; | |
2384 | } | |
2385 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2386 | ||
2387 | /** | |
2388 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2389 | * @rq: the clone request to be cleaned up | |
2390 | * | |
2391 | * Description: | |
2392 | * Free all bios in @rq for a cloned request. | |
2393 | */ | |
2394 | void blk_rq_unprep_clone(struct request *rq) | |
2395 | { | |
2396 | struct bio *bio; | |
2397 | ||
2398 | while ((bio = rq->bio) != NULL) { | |
2399 | rq->bio = bio->bi_next; | |
2400 | ||
2401 | bio_put(bio); | |
2402 | } | |
2403 | } | |
2404 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2405 | ||
2406 | /* | |
2407 | * Copy attributes of the original request to the clone request. | |
2408 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2409 | */ | |
2410 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2411 | { | |
2412 | dst->cpu = src->cpu; | |
2413 | dst->cmd_flags = (rq_data_dir(src) | REQ_NOMERGE); | |
2414 | dst->cmd_type = src->cmd_type; | |
2415 | dst->__sector = blk_rq_pos(src); | |
2416 | dst->__data_len = blk_rq_bytes(src); | |
2417 | dst->nr_phys_segments = src->nr_phys_segments; | |
2418 | dst->ioprio = src->ioprio; | |
2419 | dst->extra_len = src->extra_len; | |
2420 | } | |
2421 | ||
2422 | /** | |
2423 | * blk_rq_prep_clone - Helper function to setup clone request | |
2424 | * @rq: the request to be setup | |
2425 | * @rq_src: original request to be cloned | |
2426 | * @bs: bio_set that bios for clone are allocated from | |
2427 | * @gfp_mask: memory allocation mask for bio | |
2428 | * @bio_ctr: setup function to be called for each clone bio. | |
2429 | * Returns %0 for success, non %0 for failure. | |
2430 | * @data: private data to be passed to @bio_ctr | |
2431 | * | |
2432 | * Description: | |
2433 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2434 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2435 | * are not copied, and copying such parts is the caller's responsibility. | |
2436 | * Also, pages which the original bios are pointing to are not copied | |
2437 | * and the cloned bios just point same pages. | |
2438 | * So cloned bios must be completed before original bios, which means | |
2439 | * the caller must complete @rq before @rq_src. | |
2440 | */ | |
2441 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2442 | struct bio_set *bs, gfp_t gfp_mask, | |
2443 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2444 | void *data) | |
2445 | { | |
2446 | struct bio *bio, *bio_src; | |
2447 | ||
2448 | if (!bs) | |
2449 | bs = fs_bio_set; | |
2450 | ||
2451 | blk_rq_init(NULL, rq); | |
2452 | ||
2453 | __rq_for_each_bio(bio_src, rq_src) { | |
2454 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2455 | if (!bio) | |
2456 | goto free_and_out; | |
2457 | ||
2458 | __bio_clone(bio, bio_src); | |
2459 | ||
2460 | if (bio_integrity(bio_src) && | |
2461 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) | |
2462 | goto free_and_out; | |
2463 | ||
2464 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2465 | goto free_and_out; | |
2466 | ||
2467 | if (rq->bio) { | |
2468 | rq->biotail->bi_next = bio; | |
2469 | rq->biotail = bio; | |
2470 | } else | |
2471 | rq->bio = rq->biotail = bio; | |
2472 | } | |
2473 | ||
2474 | __blk_rq_prep_clone(rq, rq_src); | |
2475 | ||
2476 | return 0; | |
2477 | ||
2478 | free_and_out: | |
2479 | if (bio) | |
2480 | bio_free(bio, bs); | |
2481 | blk_rq_unprep_clone(rq); | |
2482 | ||
2483 | return -ENOMEM; | |
2484 | } | |
2485 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2486 | ||
2487 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) | |
2488 | { | |
2489 | return queue_work(kblockd_workqueue, work); | |
2490 | } | |
2491 | EXPORT_SYMBOL(kblockd_schedule_work); | |
2492 | ||
2493 | int __init blk_dev_init(void) | |
2494 | { | |
2495 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * | |
2496 | sizeof(((struct request *)0)->cmd_flags)); | |
2497 | ||
2498 | kblockd_workqueue = create_workqueue("kblockd"); | |
2499 | if (!kblockd_workqueue) | |
2500 | panic("Failed to create kblockd\n"); | |
2501 | ||
2502 | request_cachep = kmem_cache_create("blkdev_requests", | |
2503 | sizeof(struct request), 0, SLAB_PANIC, NULL); | |
2504 | ||
2505 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", | |
2506 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); | |
2507 | ||
2508 | return 0; | |
2509 | } | |
2510 |