Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
2056a782 | 29 | #include <linux/blktrace_api.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
1da177e4 | 31 | |
8324aa91 JA |
32 | #include "blk.h" |
33 | ||
165125e1 | 34 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
35 | |
36 | /* | |
37 | * For the allocated request tables | |
38 | */ | |
5ece6c52 | 39 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
40 | |
41 | /* | |
42 | * For queue allocation | |
43 | */ | |
6728cb0e | 44 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 45 | |
1da177e4 LT |
46 | /* |
47 | * Controlling structure to kblockd | |
48 | */ | |
ff856bad | 49 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 50 | |
26b8256e JA |
51 | static void drive_stat_acct(struct request *rq, int new_io) |
52 | { | |
28f13702 | 53 | struct hd_struct *part; |
26b8256e | 54 | int rw = rq_data_dir(rq); |
c9959059 | 55 | int cpu; |
26b8256e JA |
56 | |
57 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
58 | return; | |
59 | ||
074a7aca | 60 | cpu = part_stat_lock(); |
e71bf0d0 | 61 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
c9959059 | 62 | |
28f13702 | 63 | if (!new_io) |
074a7aca | 64 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 65 | else { |
074a7aca TH |
66 | part_round_stats(cpu, part); |
67 | part_inc_in_flight(part); | |
26b8256e | 68 | } |
e71bf0d0 | 69 | |
074a7aca | 70 | part_stat_unlock(); |
26b8256e JA |
71 | } |
72 | ||
8324aa91 | 73 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
74 | { |
75 | int nr; | |
76 | ||
77 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
78 | if (nr > q->nr_requests) | |
79 | nr = q->nr_requests; | |
80 | q->nr_congestion_on = nr; | |
81 | ||
82 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
83 | if (nr < 1) | |
84 | nr = 1; | |
85 | q->nr_congestion_off = nr; | |
86 | } | |
87 | ||
1da177e4 LT |
88 | /** |
89 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
90 | * @bdev: device | |
91 | * | |
92 | * Locates the passed device's request queue and returns the address of its | |
93 | * backing_dev_info | |
94 | * | |
95 | * Will return NULL if the request queue cannot be located. | |
96 | */ | |
97 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
98 | { | |
99 | struct backing_dev_info *ret = NULL; | |
165125e1 | 100 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
101 | |
102 | if (q) | |
103 | ret = &q->backing_dev_info; | |
104 | return ret; | |
105 | } | |
1da177e4 LT |
106 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
107 | ||
2a4aa30c | 108 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 109 | { |
1afb20f3 FT |
110 | memset(rq, 0, sizeof(*rq)); |
111 | ||
1da177e4 | 112 | INIT_LIST_HEAD(&rq->queuelist); |
c7c22e4d | 113 | rq->cpu = -1; |
63a71386 JA |
114 | rq->q = q; |
115 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
116 | INIT_HLIST_NODE(&rq->hash); |
117 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 118 | rq->cmd = rq->__cmd; |
63a71386 | 119 | rq->tag = -1; |
1da177e4 | 120 | rq->ref_count = 1; |
1da177e4 | 121 | } |
2a4aa30c | 122 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 123 | |
5bb23a68 N |
124 | static void req_bio_endio(struct request *rq, struct bio *bio, |
125 | unsigned int nbytes, int error) | |
1da177e4 | 126 | { |
165125e1 | 127 | struct request_queue *q = rq->q; |
797e7dbb | 128 | |
5bb23a68 N |
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; | |
797e7dbb | 134 | |
5bb23a68 | 135 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 136 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 137 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
138 | nbytes = bio->bi_size; |
139 | } | |
797e7dbb | 140 | |
5bb23a68 N |
141 | bio->bi_size -= nbytes; |
142 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
143 | |
144 | if (bio_integrity(bio)) | |
145 | bio_integrity_advance(bio, nbytes); | |
146 | ||
5bb23a68 | 147 | if (bio->bi_size == 0) |
6712ecf8 | 148 | bio_endio(bio, error); |
5bb23a68 N |
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 | } | |
1da177e4 | 158 | } |
1da177e4 | 159 | |
1da177e4 LT |
160 | void blk_dump_rq_flags(struct request *rq, char *msg) |
161 | { | |
162 | int bit; | |
163 | ||
6728cb0e | 164 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
165 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
166 | rq->cmd_flags); | |
1da177e4 | 167 | |
6728cb0e JA |
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); | |
1da177e4 | 176 | |
4aff5e23 | 177 | if (blk_pc_request(rq)) { |
6728cb0e | 178 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 179 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
180 | printk("%02x ", rq->cmd[bit]); |
181 | printk("\n"); | |
182 | } | |
183 | } | |
1da177e4 LT |
184 | EXPORT_SYMBOL(blk_dump_rq_flags); |
185 | ||
1da177e4 LT |
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 | */ | |
165125e1 | 194 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
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 | */ | |
7daac490 | 202 | if (blk_queue_stopped(q)) |
1da177e4 LT |
203 | return; |
204 | ||
e48ec690 | 205 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 206 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
207 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
208 | } | |
1da177e4 | 209 | } |
1da177e4 LT |
210 | EXPORT_SYMBOL(blk_plug_device); |
211 | ||
6c5e0c4d JA |
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 | ||
1da177e4 LT |
230 | /* |
231 | * remove the queue from the plugged list, if present. called with | |
232 | * queue lock held and interrupts disabled. | |
233 | */ | |
165125e1 | 234 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
235 | { |
236 | WARN_ON(!irqs_disabled()); | |
237 | ||
e48ec690 | 238 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
239 | return 0; |
240 | ||
241 | del_timer(&q->unplug_timer); | |
242 | return 1; | |
243 | } | |
1da177e4 LT |
244 | EXPORT_SYMBOL(blk_remove_plug); |
245 | ||
246 | /* | |
247 | * remove the plug and let it rip.. | |
248 | */ | |
165125e1 | 249 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 250 | { |
7daac490 | 251 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
252 | return; |
253 | ||
254 | if (!blk_remove_plug(q)) | |
255 | return; | |
256 | ||
22e2c507 | 257 | q->request_fn(q); |
1da177e4 LT |
258 | } |
259 | EXPORT_SYMBOL(__generic_unplug_device); | |
260 | ||
261 | /** | |
262 | * generic_unplug_device - fire a request queue | |
165125e1 | 263 | * @q: The &struct request_queue in question |
1da177e4 LT |
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 | **/ | |
165125e1 | 272 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 273 | { |
dbaf2c00 JA |
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 | } | |
1da177e4 LT |
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 | { | |
165125e1 | 285 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 286 | |
2ad8b1ef | 287 | blk_unplug(q); |
1da177e4 LT |
288 | } |
289 | ||
86db1e29 | 290 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 291 | { |
165125e1 JA |
292 | struct request_queue *q = |
293 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 294 | |
2056a782 JA |
295 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
296 | q->rq.count[READ] + q->rq.count[WRITE]); | |
297 | ||
1da177e4 LT |
298 | q->unplug_fn(q); |
299 | } | |
300 | ||
86db1e29 | 301 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 302 | { |
165125e1 | 303 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 304 | |
2056a782 JA |
305 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
306 | q->rq.count[READ] + q->rq.count[WRITE]); | |
307 | ||
18887ad9 | 308 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
309 | } |
310 | ||
2ad8b1ef AB |
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 | ||
c7c22e4d JA |
325 | static void blk_invoke_request_fn(struct request_queue *q) |
326 | { | |
327 | /* | |
328 | * one level of recursion is ok and is much faster than kicking | |
329 | * the unplug handling | |
330 | */ | |
331 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
332 | q->request_fn(q); | |
333 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
334 | } else { | |
335 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
336 | kblockd_schedule_work(q, &q->unplug_work); | |
337 | } | |
338 | } | |
339 | ||
1da177e4 LT |
340 | /** |
341 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 342 | * @q: The &struct request_queue in question |
1da177e4 LT |
343 | * |
344 | * Description: | |
345 | * blk_start_queue() will clear the stop flag on the queue, and call | |
346 | * the request_fn for the queue if it was in a stopped state when | |
347 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
348 | **/ | |
165125e1 | 349 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 350 | { |
a038e253 PBG |
351 | WARN_ON(!irqs_disabled()); |
352 | ||
75ad23bc | 353 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
c7c22e4d | 354 | blk_invoke_request_fn(q); |
1da177e4 | 355 | } |
1da177e4 LT |
356 | EXPORT_SYMBOL(blk_start_queue); |
357 | ||
358 | /** | |
359 | * blk_stop_queue - stop a queue | |
165125e1 | 360 | * @q: The &struct request_queue in question |
1da177e4 LT |
361 | * |
362 | * Description: | |
363 | * The Linux block layer assumes that a block driver will consume all | |
364 | * entries on the request queue when the request_fn strategy is called. | |
365 | * Often this will not happen, because of hardware limitations (queue | |
366 | * depth settings). If a device driver gets a 'queue full' response, | |
367 | * or if it simply chooses not to queue more I/O at one point, it can | |
368 | * call this function to prevent the request_fn from being called until | |
369 | * the driver has signalled it's ready to go again. This happens by calling | |
370 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
371 | **/ | |
165125e1 | 372 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
373 | { |
374 | blk_remove_plug(q); | |
75ad23bc | 375 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
376 | } |
377 | EXPORT_SYMBOL(blk_stop_queue); | |
378 | ||
379 | /** | |
380 | * blk_sync_queue - cancel any pending callbacks on a queue | |
381 | * @q: the queue | |
382 | * | |
383 | * Description: | |
384 | * The block layer may perform asynchronous callback activity | |
385 | * on a queue, such as calling the unplug function after a timeout. | |
386 | * A block device may call blk_sync_queue to ensure that any | |
387 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 388 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
389 | * that its ->make_request_fn will not re-add plugging prior to calling |
390 | * this function. | |
391 | * | |
392 | */ | |
393 | void blk_sync_queue(struct request_queue *q) | |
394 | { | |
395 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 396 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
397 | } |
398 | EXPORT_SYMBOL(blk_sync_queue); | |
399 | ||
400 | /** | |
401 | * blk_run_queue - run a single device queue | |
402 | * @q: The queue to run | |
403 | */ | |
75ad23bc | 404 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 405 | { |
1da177e4 | 406 | blk_remove_plug(q); |
dac07ec1 JA |
407 | |
408 | /* | |
409 | * Only recurse once to avoid overrunning the stack, let the unplug | |
410 | * handling reinvoke the handler shortly if we already got there. | |
411 | */ | |
c7c22e4d JA |
412 | if (!elv_queue_empty(q)) |
413 | blk_invoke_request_fn(q); | |
75ad23bc NP |
414 | } |
415 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 416 | |
75ad23bc NP |
417 | /** |
418 | * blk_run_queue - run a single device queue | |
419 | * @q: The queue to run | |
420 | */ | |
421 | void blk_run_queue(struct request_queue *q) | |
422 | { | |
423 | unsigned long flags; | |
424 | ||
425 | spin_lock_irqsave(q->queue_lock, flags); | |
426 | __blk_run_queue(q); | |
1da177e4 LT |
427 | spin_unlock_irqrestore(q->queue_lock, flags); |
428 | } | |
429 | EXPORT_SYMBOL(blk_run_queue); | |
430 | ||
165125e1 | 431 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
432 | { |
433 | kobject_put(&q->kobj); | |
434 | } | |
483f4afc | 435 | |
6728cb0e | 436 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc AV |
437 | { |
438 | mutex_lock(&q->sysfs_lock); | |
75ad23bc | 439 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
440 | mutex_unlock(&q->sysfs_lock); |
441 | ||
442 | if (q->elevator) | |
443 | elevator_exit(q->elevator); | |
444 | ||
445 | blk_put_queue(q); | |
446 | } | |
1da177e4 LT |
447 | EXPORT_SYMBOL(blk_cleanup_queue); |
448 | ||
165125e1 | 449 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
450 | { |
451 | struct request_list *rl = &q->rq; | |
452 | ||
453 | rl->count[READ] = rl->count[WRITE] = 0; | |
454 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 455 | rl->elvpriv = 0; |
1da177e4 LT |
456 | init_waitqueue_head(&rl->wait[READ]); |
457 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 458 | |
1946089a CL |
459 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
460 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
461 | |
462 | if (!rl->rq_pool) | |
463 | return -ENOMEM; | |
464 | ||
465 | return 0; | |
466 | } | |
467 | ||
165125e1 | 468 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 469 | { |
1946089a CL |
470 | return blk_alloc_queue_node(gfp_mask, -1); |
471 | } | |
472 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 473 | |
165125e1 | 474 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 475 | { |
165125e1 | 476 | struct request_queue *q; |
e0bf68dd | 477 | int err; |
1946089a | 478 | |
8324aa91 | 479 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 480 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
481 | if (!q) |
482 | return NULL; | |
483 | ||
e0bf68dd PZ |
484 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
485 | q->backing_dev_info.unplug_io_data = q; | |
486 | err = bdi_init(&q->backing_dev_info); | |
487 | if (err) { | |
8324aa91 | 488 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
489 | return NULL; |
490 | } | |
491 | ||
1da177e4 | 492 | init_timer(&q->unplug_timer); |
483f4afc | 493 | |
8324aa91 | 494 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 495 | |
483f4afc | 496 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 497 | spin_lock_init(&q->__queue_lock); |
483f4afc | 498 | |
1da177e4 LT |
499 | return q; |
500 | } | |
1946089a | 501 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
502 | |
503 | /** | |
504 | * blk_init_queue - prepare a request queue for use with a block device | |
505 | * @rfn: The function to be called to process requests that have been | |
506 | * placed on the queue. | |
507 | * @lock: Request queue spin lock | |
508 | * | |
509 | * Description: | |
510 | * If a block device wishes to use the standard request handling procedures, | |
511 | * which sorts requests and coalesces adjacent requests, then it must | |
512 | * call blk_init_queue(). The function @rfn will be called when there | |
513 | * are requests on the queue that need to be processed. If the device | |
514 | * supports plugging, then @rfn may not be called immediately when requests | |
515 | * are available on the queue, but may be called at some time later instead. | |
516 | * Plugged queues are generally unplugged when a buffer belonging to one | |
517 | * of the requests on the queue is needed, or due to memory pressure. | |
518 | * | |
519 | * @rfn is not required, or even expected, to remove all requests off the | |
520 | * queue, but only as many as it can handle at a time. If it does leave | |
521 | * requests on the queue, it is responsible for arranging that the requests | |
522 | * get dealt with eventually. | |
523 | * | |
524 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
525 | * request queue; this lock will be taken also from interrupt context, so irq |
526 | * disabling is needed for it. | |
1da177e4 | 527 | * |
710027a4 | 528 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
529 | * it didn't succeed. |
530 | * | |
531 | * Note: | |
532 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
533 | * when the block device is deactivated (such as at module unload). | |
534 | **/ | |
1946089a | 535 | |
165125e1 | 536 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 537 | { |
1946089a CL |
538 | return blk_init_queue_node(rfn, lock, -1); |
539 | } | |
540 | EXPORT_SYMBOL(blk_init_queue); | |
541 | ||
165125e1 | 542 | struct request_queue * |
1946089a CL |
543 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
544 | { | |
165125e1 | 545 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
546 | |
547 | if (!q) | |
548 | return NULL; | |
549 | ||
1946089a | 550 | q->node = node_id; |
8669aafd | 551 | if (blk_init_free_list(q)) { |
8324aa91 | 552 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
553 | return NULL; |
554 | } | |
1da177e4 | 555 | |
152587de | 556 | /* |
557 | * if caller didn't supply a lock, they get per-queue locking with | |
558 | * our embedded lock | |
559 | */ | |
e7e72bf6 | 560 | if (!lock) |
152587de | 561 | lock = &q->__queue_lock; |
152587de | 562 | |
1da177e4 | 563 | q->request_fn = rfn; |
1da177e4 LT |
564 | q->prep_rq_fn = NULL; |
565 | q->unplug_fn = generic_unplug_device; | |
566 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
567 | q->queue_lock = lock; | |
568 | ||
569 | blk_queue_segment_boundary(q, 0xffffffff); | |
570 | ||
571 | blk_queue_make_request(q, __make_request); | |
572 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
573 | ||
574 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
575 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
576 | ||
44ec9542 AS |
577 | q->sg_reserved_size = INT_MAX; |
578 | ||
abf54393 FT |
579 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
580 | ||
1da177e4 LT |
581 | /* |
582 | * all done | |
583 | */ | |
584 | if (!elevator_init(q, NULL)) { | |
585 | blk_queue_congestion_threshold(q); | |
586 | return q; | |
587 | } | |
588 | ||
8669aafd | 589 | blk_put_queue(q); |
1da177e4 LT |
590 | return NULL; |
591 | } | |
1946089a | 592 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 593 | |
165125e1 | 594 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 595 | { |
fde6ad22 | 596 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 597 | kobject_get(&q->kobj); |
1da177e4 LT |
598 | return 0; |
599 | } | |
600 | ||
601 | return 1; | |
602 | } | |
1da177e4 | 603 | |
165125e1 | 604 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 605 | { |
4aff5e23 | 606 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 607 | elv_put_request(q, rq); |
1da177e4 LT |
608 | mempool_free(rq, q->rq.rq_pool); |
609 | } | |
610 | ||
1ea25ecb | 611 | static struct request * |
165125e1 | 612 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
613 | { |
614 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
615 | ||
616 | if (!rq) | |
617 | return NULL; | |
618 | ||
2a4aa30c | 619 | blk_rq_init(q, rq); |
1afb20f3 | 620 | |
49171e5c | 621 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 622 | |
cb98fc8b | 623 | if (priv) { |
cb78b285 | 624 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
625 | mempool_free(rq, q->rq.rq_pool); |
626 | return NULL; | |
627 | } | |
4aff5e23 | 628 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 629 | } |
1da177e4 | 630 | |
cb98fc8b | 631 | return rq; |
1da177e4 LT |
632 | } |
633 | ||
634 | /* | |
635 | * ioc_batching returns true if the ioc is a valid batching request and | |
636 | * should be given priority access to a request. | |
637 | */ | |
165125e1 | 638 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
639 | { |
640 | if (!ioc) | |
641 | return 0; | |
642 | ||
643 | /* | |
644 | * Make sure the process is able to allocate at least 1 request | |
645 | * even if the batch times out, otherwise we could theoretically | |
646 | * lose wakeups. | |
647 | */ | |
648 | return ioc->nr_batch_requests == q->nr_batching || | |
649 | (ioc->nr_batch_requests > 0 | |
650 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
651 | } | |
652 | ||
653 | /* | |
654 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
655 | * will cause the process to be a "batcher" on all queues in the system. This | |
656 | * is the behaviour we want though - once it gets a wakeup it should be given | |
657 | * a nice run. | |
658 | */ | |
165125e1 | 659 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
660 | { |
661 | if (!ioc || ioc_batching(q, ioc)) | |
662 | return; | |
663 | ||
664 | ioc->nr_batch_requests = q->nr_batching; | |
665 | ioc->last_waited = jiffies; | |
666 | } | |
667 | ||
165125e1 | 668 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
669 | { |
670 | struct request_list *rl = &q->rq; | |
671 | ||
672 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 673 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
674 | |
675 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
676 | if (waitqueue_active(&rl->wait[rw])) |
677 | wake_up(&rl->wait[rw]); | |
678 | ||
679 | blk_clear_queue_full(q, rw); | |
680 | } | |
681 | } | |
682 | ||
683 | /* | |
684 | * A request has just been released. Account for it, update the full and | |
685 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
686 | */ | |
165125e1 | 687 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
688 | { |
689 | struct request_list *rl = &q->rq; | |
690 | ||
691 | rl->count[rw]--; | |
cb98fc8b TH |
692 | if (priv) |
693 | rl->elvpriv--; | |
1da177e4 LT |
694 | |
695 | __freed_request(q, rw); | |
696 | ||
697 | if (unlikely(rl->starved[rw ^ 1])) | |
698 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
699 | } |
700 | ||
701 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
702 | /* | |
d6344532 NP |
703 | * Get a free request, queue_lock must be held. |
704 | * Returns NULL on failure, with queue_lock held. | |
705 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 706 | */ |
165125e1 | 707 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 708 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
709 | { |
710 | struct request *rq = NULL; | |
711 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 712 | struct io_context *ioc = NULL; |
7749a8d4 | 713 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
714 | int may_queue, priv; |
715 | ||
7749a8d4 | 716 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
717 | if (may_queue == ELV_MQUEUE_NO) |
718 | goto rq_starved; | |
719 | ||
720 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
721 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 722 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
723 | /* |
724 | * The queue will fill after this allocation, so set | |
725 | * it as full, and mark this process as "batching". | |
726 | * This process will be allowed to complete a batch of | |
727 | * requests, others will be blocked. | |
728 | */ | |
729 | if (!blk_queue_full(q, rw)) { | |
730 | ioc_set_batching(q, ioc); | |
731 | blk_set_queue_full(q, rw); | |
732 | } else { | |
733 | if (may_queue != ELV_MQUEUE_MUST | |
734 | && !ioc_batching(q, ioc)) { | |
735 | /* | |
736 | * The queue is full and the allocating | |
737 | * process is not a "batcher", and not | |
738 | * exempted by the IO scheduler | |
739 | */ | |
740 | goto out; | |
741 | } | |
742 | } | |
1da177e4 | 743 | } |
79e2de4b | 744 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
745 | } |
746 | ||
082cf69e JA |
747 | /* |
748 | * Only allow batching queuers to allocate up to 50% over the defined | |
749 | * limit of requests, otherwise we could have thousands of requests | |
750 | * allocated with any setting of ->nr_requests | |
751 | */ | |
fd782a4a | 752 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 753 | goto out; |
fd782a4a | 754 | |
1da177e4 LT |
755 | rl->count[rw]++; |
756 | rl->starved[rw] = 0; | |
cb98fc8b | 757 | |
64521d1a | 758 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
759 | if (priv) |
760 | rl->elvpriv++; | |
761 | ||
1da177e4 LT |
762 | spin_unlock_irq(q->queue_lock); |
763 | ||
7749a8d4 | 764 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 765 | if (unlikely(!rq)) { |
1da177e4 LT |
766 | /* |
767 | * Allocation failed presumably due to memory. Undo anything | |
768 | * we might have messed up. | |
769 | * | |
770 | * Allocating task should really be put onto the front of the | |
771 | * wait queue, but this is pretty rare. | |
772 | */ | |
773 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 774 | freed_request(q, rw, priv); |
1da177e4 LT |
775 | |
776 | /* | |
777 | * in the very unlikely event that allocation failed and no | |
778 | * requests for this direction was pending, mark us starved | |
779 | * so that freeing of a request in the other direction will | |
780 | * notice us. another possible fix would be to split the | |
781 | * rq mempool into READ and WRITE | |
782 | */ | |
783 | rq_starved: | |
784 | if (unlikely(rl->count[rw] == 0)) | |
785 | rl->starved[rw] = 1; | |
786 | ||
1da177e4 LT |
787 | goto out; |
788 | } | |
789 | ||
88ee5ef1 JA |
790 | /* |
791 | * ioc may be NULL here, and ioc_batching will be false. That's | |
792 | * OK, if the queue is under the request limit then requests need | |
793 | * not count toward the nr_batch_requests limit. There will always | |
794 | * be some limit enforced by BLK_BATCH_TIME. | |
795 | */ | |
1da177e4 LT |
796 | if (ioc_batching(q, ioc)) |
797 | ioc->nr_batch_requests--; | |
6728cb0e | 798 | |
2056a782 | 799 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); |
1da177e4 | 800 | out: |
1da177e4 LT |
801 | return rq; |
802 | } | |
803 | ||
804 | /* | |
805 | * No available requests for this queue, unplug the device and wait for some | |
806 | * requests to become available. | |
d6344532 NP |
807 | * |
808 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 809 | */ |
165125e1 | 810 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 811 | struct bio *bio) |
1da177e4 | 812 | { |
7749a8d4 | 813 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
814 | struct request *rq; |
815 | ||
7749a8d4 | 816 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
817 | while (!rq) { |
818 | DEFINE_WAIT(wait); | |
05caf8db | 819 | struct io_context *ioc; |
1da177e4 LT |
820 | struct request_list *rl = &q->rq; |
821 | ||
822 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
823 | TASK_UNINTERRUPTIBLE); | |
824 | ||
05caf8db | 825 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
1da177e4 | 826 | |
05caf8db ZY |
827 | __generic_unplug_device(q); |
828 | spin_unlock_irq(q->queue_lock); | |
829 | io_schedule(); | |
1da177e4 | 830 | |
05caf8db ZY |
831 | /* |
832 | * After sleeping, we become a "batching" process and | |
833 | * will be able to allocate at least one request, and | |
834 | * up to a big batch of them for a small period time. | |
835 | * See ioc_batching, ioc_set_batching | |
836 | */ | |
837 | ioc = current_io_context(GFP_NOIO, q->node); | |
838 | ioc_set_batching(q, ioc); | |
d6344532 | 839 | |
05caf8db | 840 | spin_lock_irq(q->queue_lock); |
1da177e4 | 841 | finish_wait(&rl->wait[rw], &wait); |
05caf8db ZY |
842 | |
843 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
844 | }; | |
1da177e4 LT |
845 | |
846 | return rq; | |
847 | } | |
848 | ||
165125e1 | 849 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
850 | { |
851 | struct request *rq; | |
852 | ||
853 | BUG_ON(rw != READ && rw != WRITE); | |
854 | ||
d6344532 NP |
855 | spin_lock_irq(q->queue_lock); |
856 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 857 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 858 | } else { |
22e2c507 | 859 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
860 | if (!rq) |
861 | spin_unlock_irq(q->queue_lock); | |
862 | } | |
863 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
864 | |
865 | return rq; | |
866 | } | |
1da177e4 LT |
867 | EXPORT_SYMBOL(blk_get_request); |
868 | ||
dc72ef4a JA |
869 | /** |
870 | * blk_start_queueing - initiate dispatch of requests to device | |
871 | * @q: request queue to kick into gear | |
872 | * | |
873 | * This is basically a helper to remove the need to know whether a queue | |
874 | * is plugged or not if someone just wants to initiate dispatch of requests | |
875 | * for this queue. | |
876 | * | |
877 | * The queue lock must be held with interrupts disabled. | |
878 | */ | |
165125e1 | 879 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a JA |
880 | { |
881 | if (!blk_queue_plugged(q)) | |
882 | q->request_fn(q); | |
883 | else | |
884 | __generic_unplug_device(q); | |
885 | } | |
886 | EXPORT_SYMBOL(blk_start_queueing); | |
887 | ||
1da177e4 LT |
888 | /** |
889 | * blk_requeue_request - put a request back on queue | |
890 | * @q: request queue where request should be inserted | |
891 | * @rq: request to be inserted | |
892 | * | |
893 | * Description: | |
894 | * Drivers often keep queueing requests until the hardware cannot accept | |
895 | * more, when that condition happens we need to put the request back | |
896 | * on the queue. Must be called with queue lock held. | |
897 | */ | |
165125e1 | 898 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 899 | { |
2056a782 JA |
900 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
901 | ||
1da177e4 LT |
902 | if (blk_rq_tagged(rq)) |
903 | blk_queue_end_tag(q, rq); | |
904 | ||
905 | elv_requeue_request(q, rq); | |
906 | } | |
1da177e4 LT |
907 | EXPORT_SYMBOL(blk_requeue_request); |
908 | ||
909 | /** | |
710027a4 | 910 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
911 | * @q: request queue where request should be inserted |
912 | * @rq: request to be inserted | |
913 | * @at_head: insert request at head or tail of queue | |
914 | * @data: private data | |
1da177e4 LT |
915 | * |
916 | * Description: | |
917 | * Many block devices need to execute commands asynchronously, so they don't | |
918 | * block the whole kernel from preemption during request execution. This is | |
919 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
920 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
921 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
922 | * |
923 | * We have the option of inserting the head or the tail of the queue. | |
924 | * Typically we use the tail for new ioctls and so forth. We use the head | |
925 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
926 | * host that is unable to accept a particular command. | |
927 | */ | |
165125e1 | 928 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 929 | int at_head, void *data) |
1da177e4 | 930 | { |
867d1191 | 931 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
932 | unsigned long flags; |
933 | ||
934 | /* | |
935 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
936 | * must not attempt merges on this) and that it acts as a soft | |
937 | * barrier | |
938 | */ | |
4aff5e23 JA |
939 | rq->cmd_type = REQ_TYPE_SPECIAL; |
940 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
941 | |
942 | rq->special = data; | |
943 | ||
944 | spin_lock_irqsave(q->queue_lock, flags); | |
945 | ||
946 | /* | |
947 | * If command is tagged, release the tag | |
948 | */ | |
867d1191 TH |
949 | if (blk_rq_tagged(rq)) |
950 | blk_queue_end_tag(q, rq); | |
1da177e4 | 951 | |
b238b3d4 | 952 | drive_stat_acct(rq, 1); |
867d1191 | 953 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 954 | blk_start_queueing(q); |
1da177e4 LT |
955 | spin_unlock_irqrestore(q->queue_lock, flags); |
956 | } | |
1da177e4 LT |
957 | EXPORT_SYMBOL(blk_insert_request); |
958 | ||
1da177e4 LT |
959 | /* |
960 | * add-request adds a request to the linked list. | |
961 | * queue lock is held and interrupts disabled, as we muck with the | |
962 | * request queue list. | |
963 | */ | |
6728cb0e | 964 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 965 | { |
b238b3d4 | 966 | drive_stat_acct(req, 1); |
1da177e4 | 967 | |
1da177e4 LT |
968 | /* |
969 | * elevator indicated where it wants this request to be | |
970 | * inserted at elevator_merge time | |
971 | */ | |
972 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
973 | } | |
6728cb0e | 974 | |
074a7aca TH |
975 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
976 | unsigned long now) | |
977 | { | |
978 | if (now == part->stamp) | |
979 | return; | |
980 | ||
981 | if (part->in_flight) { | |
982 | __part_stat_add(cpu, part, time_in_queue, | |
983 | part->in_flight * (now - part->stamp)); | |
984 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
985 | } | |
986 | part->stamp = now; | |
987 | } | |
988 | ||
989 | /** | |
990 | * part_round_stats() - Round off the performance stats on a struct | |
1da177e4 LT |
991 | * disk_stats. |
992 | * | |
993 | * The average IO queue length and utilisation statistics are maintained | |
994 | * by observing the current state of the queue length and the amount of | |
995 | * time it has been in this state for. | |
996 | * | |
997 | * Normally, that accounting is done on IO completion, but that can result | |
998 | * in more than a second's worth of IO being accounted for within any one | |
999 | * second, leading to >100% utilisation. To deal with that, we call this | |
1000 | * function to do a round-off before returning the results when reading | |
1001 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1002 | * the current jiffies and restarts the counters again. | |
1003 | */ | |
c9959059 | 1004 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1005 | { |
1006 | unsigned long now = jiffies; | |
1007 | ||
074a7aca TH |
1008 | if (part->partno) |
1009 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1010 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1011 | } |
074a7aca | 1012 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1013 | |
1da177e4 LT |
1014 | /* |
1015 | * queue lock must be held | |
1016 | */ | |
165125e1 | 1017 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1018 | { |
1da177e4 LT |
1019 | if (unlikely(!q)) |
1020 | return; | |
1021 | if (unlikely(--req->ref_count)) | |
1022 | return; | |
1023 | ||
8922e16c TH |
1024 | elv_completed_request(q, req); |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * Request may not have originated from ll_rw_blk. if not, | |
1028 | * it didn't come out of our reserved rq pools | |
1029 | */ | |
49171e5c | 1030 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1031 | int rw = rq_data_dir(req); |
4aff5e23 | 1032 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1033 | |
1da177e4 | 1034 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1035 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1036 | |
1037 | blk_free_request(q, req); | |
cb98fc8b | 1038 | freed_request(q, rw, priv); |
1da177e4 LT |
1039 | } |
1040 | } | |
6e39b69e MC |
1041 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1042 | ||
1da177e4 LT |
1043 | void blk_put_request(struct request *req) |
1044 | { | |
8922e16c | 1045 | unsigned long flags; |
165125e1 | 1046 | struct request_queue *q = req->q; |
8922e16c | 1047 | |
52a93ba8 FT |
1048 | spin_lock_irqsave(q->queue_lock, flags); |
1049 | __blk_put_request(q, req); | |
1050 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1051 | } |
1da177e4 LT |
1052 | EXPORT_SYMBOL(blk_put_request); |
1053 | ||
86db1e29 | 1054 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1055 | { |
c7c22e4d | 1056 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1057 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1058 | |
1059 | /* | |
1060 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1061 | */ | |
1062 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 1063 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
1064 | |
1065 | /* | |
1066 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1067 | */ | |
fb2dce86 | 1068 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1069 | req->cmd_flags |= REQ_DISCARD; |
1070 | if (bio_barrier(bio)) | |
1071 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1072 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 DW |
1073 | } else if (unlikely(bio_barrier(bio))) |
1074 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
52d9e675 | 1075 | |
b31dc66a | 1076 | if (bio_sync(bio)) |
4aff5e23 | 1077 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1078 | if (bio_rw_meta(bio)) |
1079 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1080 | |
52d9e675 TH |
1081 | req->errors = 0; |
1082 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1083 | req->ioprio = bio_prio(bio); |
52d9e675 | 1084 | req->start_time = jiffies; |
bc1c56fd | 1085 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1086 | } |
1087 | ||
165125e1 | 1088 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1089 | { |
450991bc | 1090 | struct request *req; |
fb2dce86 | 1091 | int el_ret, nr_sectors, barrier, discard, err; |
51da90fc JA |
1092 | const unsigned short prio = bio_prio(bio); |
1093 | const int sync = bio_sync(bio); | |
7749a8d4 | 1094 | int rw_flags; |
1da177e4 | 1095 | |
1da177e4 | 1096 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1097 | |
1098 | /* | |
1099 | * low level driver can indicate that it wants pages above a | |
1100 | * certain limit bounced to low memory (ie for highmem, or even | |
1101 | * ISA dma in theory) | |
1102 | */ | |
1103 | blk_queue_bounce(q, &bio); | |
1104 | ||
1da177e4 | 1105 | barrier = bio_barrier(bio); |
e17fc0a1 DW |
1106 | if (unlikely(barrier) && bio_has_data(bio) && |
1107 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1da177e4 LT |
1108 | err = -EOPNOTSUPP; |
1109 | goto end_io; | |
1110 | } | |
1111 | ||
fb2dce86 DW |
1112 | discard = bio_discard(bio); |
1113 | if (unlikely(discard) && !q->prepare_discard_fn) { | |
1114 | err = -EOPNOTSUPP; | |
1115 | goto end_io; | |
1116 | } | |
1117 | ||
1da177e4 LT |
1118 | spin_lock_irq(q->queue_lock); |
1119 | ||
450991bc | 1120 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1121 | goto get_rq; |
1122 | ||
1123 | el_ret = elv_merge(q, &req, bio); | |
1124 | switch (el_ret) { | |
6728cb0e JA |
1125 | case ELEVATOR_BACK_MERGE: |
1126 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1127 | |
6728cb0e JA |
1128 | if (!ll_back_merge_fn(q, req, bio)) |
1129 | break; | |
1da177e4 | 1130 | |
6728cb0e | 1131 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2056a782 | 1132 | |
6728cb0e JA |
1133 | req->biotail->bi_next = bio; |
1134 | req->biotail = bio; | |
1135 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1136 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1137 | drive_stat_acct(req, 0); | |
1138 | if (!attempt_back_merge(q, req)) | |
1139 | elv_merged_request(q, req, el_ret); | |
1140 | goto out; | |
1da177e4 | 1141 | |
6728cb0e JA |
1142 | case ELEVATOR_FRONT_MERGE: |
1143 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1144 | |
6728cb0e JA |
1145 | if (!ll_front_merge_fn(q, req, bio)) |
1146 | break; | |
1da177e4 | 1147 | |
6728cb0e | 1148 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2056a782 | 1149 | |
6728cb0e JA |
1150 | bio->bi_next = req->bio; |
1151 | req->bio = bio; | |
1da177e4 | 1152 | |
6728cb0e JA |
1153 | /* |
1154 | * may not be valid. if the low level driver said | |
1155 | * it didn't need a bounce buffer then it better | |
1156 | * not touch req->buffer either... | |
1157 | */ | |
1158 | req->buffer = bio_data(bio); | |
1159 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1160 | req->hard_cur_sectors = req->current_nr_sectors; | |
1161 | req->sector = req->hard_sector = bio->bi_sector; | |
1162 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1163 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1164 | drive_stat_acct(req, 0); | |
1165 | if (!attempt_front_merge(q, req)) | |
1166 | elv_merged_request(q, req, el_ret); | |
1167 | goto out; | |
1168 | ||
1169 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1170 | default: | |
1171 | ; | |
1da177e4 LT |
1172 | } |
1173 | ||
450991bc | 1174 | get_rq: |
7749a8d4 JA |
1175 | /* |
1176 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1177 | * but we need to set it earlier to expose the sync flag to the | |
1178 | * rq allocator and io schedulers. | |
1179 | */ | |
1180 | rw_flags = bio_data_dir(bio); | |
1181 | if (sync) | |
1182 | rw_flags |= REQ_RW_SYNC; | |
1183 | ||
1da177e4 | 1184 | /* |
450991bc | 1185 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1186 | * Returns with the queue unlocked. |
450991bc | 1187 | */ |
7749a8d4 | 1188 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1189 | |
450991bc NP |
1190 | /* |
1191 | * After dropping the lock and possibly sleeping here, our request | |
1192 | * may now be mergeable after it had proven unmergeable (above). | |
1193 | * We don't worry about that case for efficiency. It won't happen | |
1194 | * often, and the elevators are able to handle it. | |
1da177e4 | 1195 | */ |
52d9e675 | 1196 | init_request_from_bio(req, bio); |
1da177e4 | 1197 | |
450991bc | 1198 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1199 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1200 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1201 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
450991bc NP |
1202 | if (elv_queue_empty(q)) |
1203 | blk_plug_device(q); | |
1da177e4 LT |
1204 | add_request(q, req); |
1205 | out: | |
4a534f93 | 1206 | if (sync) |
1da177e4 | 1207 | __generic_unplug_device(q); |
1da177e4 LT |
1208 | spin_unlock_irq(q->queue_lock); |
1209 | return 0; | |
1210 | ||
1211 | end_io: | |
6712ecf8 | 1212 | bio_endio(bio, err); |
1da177e4 LT |
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 | ||
bf2de6f5 | 1223 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1224 | struct hd_struct *p = bdev->bd_part; |
1225 | ||
1da177e4 LT |
1226 | bio->bi_sector += p->start_sect; |
1227 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b AB |
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); | |
1da177e4 LT |
1232 | } |
1233 | } | |
1234 | ||
1da177e4 LT |
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 | ||
c17bb495 AM |
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 | { | |
eddb2e26 TH |
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) | |
c17bb495 AM |
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 | ||
c07e2b41 JA |
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 | ||
1da177e4 | 1315 | /** |
710027a4 | 1316 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
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 | */ | |
d89d8796 | 1339 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1340 | { |
165125e1 | 1341 | struct request_queue *q; |
5ddfe969 | 1342 | sector_t old_sector; |
1da177e4 | 1343 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1344 | dev_t old_dev; |
51fd77bd | 1345 | int err = -EIO; |
1da177e4 LT |
1346 | |
1347 | might_sleep(); | |
1da177e4 | 1348 | |
c07e2b41 JA |
1349 | if (bio_check_eod(bio, nr_sectors)) |
1350 | goto end_io; | |
1da177e4 LT |
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 | */ | |
5ddfe969 | 1360 | old_sector = -1; |
2056a782 | 1361 | old_dev = 0; |
1da177e4 LT |
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: | |
51fd77bd | 1373 | bio_endio(bio, err); |
1da177e4 LT |
1374 | break; |
1375 | } | |
1376 | ||
4fa253f3 | 1377 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1378 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1379 | bdevname(bio->bi_bdev, b), |
1380 | bio_sectors(bio), | |
1381 | q->max_hw_sectors); | |
1382 | goto end_io; | |
1383 | } | |
1384 | ||
fde6ad22 | 1385 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1386 | goto end_io; |
1387 | ||
c17bb495 AM |
1388 | if (should_fail_request(bio)) |
1389 | goto end_io; | |
1390 | ||
1da177e4 LT |
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 | ||
7ba1ba12 MP |
1397 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1398 | goto end_io; | |
1399 | ||
5ddfe969 | 1400 | if (old_sector != -1) |
4fa253f3 | 1401 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1402 | old_sector); |
2056a782 JA |
1403 | |
1404 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
1405 | ||
5ddfe969 | 1406 | old_sector = bio->bi_sector; |
2056a782 JA |
1407 | old_dev = bio->bi_bdev->bd_dev; |
1408 | ||
c07e2b41 JA |
1409 | if (bio_check_eod(bio, nr_sectors)) |
1410 | goto end_io; | |
fb2dce86 DW |
1411 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || |
1412 | (bio_discard(bio) && !q->prepare_discard_fn)) { | |
51fd77bd JA |
1413 | err = -EOPNOTSUPP; |
1414 | goto end_io; | |
1415 | } | |
5ddfe969 | 1416 | |
1da177e4 LT |
1417 | ret = q->make_request_fn(q, bio); |
1418 | } while (ret); | |
1419 | } | |
1420 | ||
d89d8796 NB |
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 | } | |
1da177e4 LT |
1471 | EXPORT_SYMBOL(generic_make_request); |
1472 | ||
1473 | /** | |
710027a4 | 1474 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
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 | |
710027a4 | 1480 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1481 | * |
1482 | */ | |
1483 | void submit_bio(int rw, struct bio *bio) | |
1484 | { | |
1485 | int count = bio_sectors(bio); | |
1486 | ||
22e2c507 | 1487 | bio->bi_rw |= rw; |
1da177e4 | 1488 | |
bf2de6f5 JA |
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 | */ | |
a9c701e5 | 1493 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
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", | |
ba25f9dc | 1504 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1505 | (rw & WRITE) ? "WRITE" : "READ", |
1506 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1507 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1508 | } |
1da177e4 LT |
1509 | } |
1510 | ||
1511 | generic_make_request(bio); | |
1512 | } | |
1da177e4 LT |
1513 | EXPORT_SYMBOL(submit_bio); |
1514 | ||
3bcddeac KU |
1515 | /** |
1516 | * __end_that_request_first - end I/O on a request | |
1517 | * @req: the request being processed | |
710027a4 | 1518 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
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: | |
710027a4 RD |
1526 | * %0 - we are done with this request, call end_that_request_last() |
1527 | * %1 - still buffers pending for this request | |
3bcddeac | 1528 | **/ |
5450d3e1 | 1529 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1530 | int nr_bytes) |
1531 | { | |
5450d3e1 | 1532 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1533 | struct bio *bio; |
1534 | ||
2056a782 JA |
1535 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
1536 | ||
1da177e4 | 1537 | /* |
710027a4 | 1538 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual |
1da177e4 LT |
1539 | * sense key with us all the way through |
1540 | */ | |
1541 | if (!blk_pc_request(req)) | |
1542 | req->errors = 0; | |
1543 | ||
6728cb0e JA |
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", | |
1da177e4 LT |
1546 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1547 | (unsigned long long)req->sector); | |
1548 | } | |
1549 | ||
d72d904a | 1550 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b | 1551 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1552 | struct hd_struct *part; |
c9959059 | 1553 | int cpu; |
a362357b | 1554 | |
074a7aca | 1555 | cpu = part_stat_lock(); |
e71bf0d0 | 1556 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
074a7aca TH |
1557 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); |
1558 | part_stat_unlock(); | |
d72d904a JA |
1559 | } |
1560 | ||
1da177e4 LT |
1561 | total_bytes = bio_nbytes = 0; |
1562 | while ((bio = req->bio) != NULL) { | |
1563 | int nbytes; | |
1564 | ||
bf2de6f5 JA |
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 | ||
1da177e4 LT |
1573 | if (nr_bytes >= bio->bi_size) { |
1574 | req->bio = bio->bi_next; | |
1575 | nbytes = bio->bi_size; | |
5bb23a68 | 1576 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
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"); | |
6728cb0e | 1584 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1585 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
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 | ||
6728cb0e JA |
1611 | bio = req->bio; |
1612 | if (bio) { | |
1da177e4 LT |
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) { | |
5bb23a68 | 1631 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
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 | ||
1da177e4 LT |
1642 | /* |
1643 | * queue lock must be held | |
1644 | */ | |
5450d3e1 | 1645 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1646 | { |
1647 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1648 | |
b8286239 KU |
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); | |
1da177e4 LT |
1654 | |
1655 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1656 | laptop_io_completion(); | |
1657 | ||
fd0ff8aa JA |
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) { | |
1da177e4 | 1664 | unsigned long duration = jiffies - req->start_time; |
a362357b | 1665 | const int rw = rq_data_dir(req); |
e71bf0d0 | 1666 | struct hd_struct *part; |
c9959059 | 1667 | int cpu; |
e71bf0d0 | 1668 | |
074a7aca | 1669 | cpu = part_stat_lock(); |
e71bf0d0 | 1670 | part = disk_map_sector_rcu(disk, req->sector); |
a362357b | 1671 | |
074a7aca TH |
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); | |
e71bf0d0 | 1676 | |
074a7aca | 1677 | part_stat_unlock(); |
1da177e4 | 1678 | } |
b8286239 | 1679 | |
1da177e4 | 1680 | if (req->end_io) |
8ffdc655 | 1681 | req->end_io(req, error); |
b8286239 KU |
1682 | else { |
1683 | if (blk_bidi_rq(req)) | |
1684 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1685 | ||
1da177e4 | 1686 | __blk_put_request(req->q, req); |
b8286239 | 1687 | } |
1da177e4 LT |
1688 | } |
1689 | ||
a0cd1285 | 1690 | static inline void __end_request(struct request *rq, int uptodate, |
9e6e39f2 | 1691 | unsigned int nr_bytes) |
1da177e4 | 1692 | { |
9e6e39f2 KU |
1693 | int error = 0; |
1694 | ||
1695 | if (uptodate <= 0) | |
1696 | error = uptodate ? uptodate : -EIO; | |
1697 | ||
1698 | __blk_end_request(rq, error, nr_bytes); | |
1da177e4 LT |
1699 | } |
1700 | ||
3b11313a KU |
1701 | /** |
1702 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1703 | * @rq: the request being processed |
3b11313a KU |
1704 | **/ |
1705 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1706 | { |
1707 | if (blk_fs_request(rq)) | |
1708 | return rq->hard_nr_sectors << 9; | |
1709 | ||
1710 | return rq->data_len; | |
1711 | } | |
3b11313a KU |
1712 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1713 | ||
1714 | /** | |
1715 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1716 | * @rq: the request being processed |
3b11313a KU |
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); | |
a0cd1285 JA |
1729 | |
1730 | /** | |
1731 | * end_queued_request - end all I/O on a queued request | |
1732 | * @rq: the request being processed | |
710027a4 | 1733 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1734 | * |
1735 | * Description: | |
1736 | * Ends all I/O on a request, and removes it from the block layer queues. | |
710027a4 | 1737 | * Not suitable for normal I/O completion, unless the driver still has |
a0cd1285 JA |
1738 | * the request attached to the block layer. |
1739 | * | |
1740 | **/ | |
1741 | void end_queued_request(struct request *rq, int uptodate) | |
1742 | { | |
9e6e39f2 | 1743 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
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 | |
710027a4 | 1750 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
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 | { | |
9e6e39f2 | 1760 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1761 | } |
1762 | EXPORT_SYMBOL(end_dequeued_request); | |
1763 | ||
1764 | ||
1765 | /** | |
1766 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1767 | * @req: the request being processed |
710027a4 | 1768 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
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 | * | |
710027a4 RD |
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 | |
a0cd1285 JA |
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 | { | |
9e6e39f2 | 1786 | __end_request(req, uptodate, req->hard_cur_sectors << 9); |
a0cd1285 | 1787 | } |
1da177e4 LT |
1788 | EXPORT_SYMBOL(end_request); |
1789 | ||
336cdb40 | 1790 | /** |
e19a3ab0 KU |
1791 | * blk_end_io - Generic end_io function to complete a request. |
1792 | * @rq: the request being processed | |
710027a4 | 1793 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1794 | * @nr_bytes: number of bytes to complete @rq |
1795 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1796 | * @drv_callback: function called between completion of bios in the request |
1797 | * and completion of the request. | |
710027a4 | 1798 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1799 | * completion of the request. |
336cdb40 KU |
1800 | * |
1801 | * Description: | |
e3a04fe3 | 1802 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1803 | * If @rq has leftover, sets it up for the next range of segments. |
1804 | * | |
1805 | * Return: | |
710027a4 RD |
1806 | * %0 - we are done with this request |
1807 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1808 | **/ |
22b13210 JA |
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 *)) | |
336cdb40 KU |
1812 | { |
1813 | struct request_queue *q = rq->q; | |
1814 | unsigned long flags = 0UL; | |
336cdb40 | 1815 | |
fb2dce86 | 1816 | if (bio_has_data(rq->bio) || blk_discard_rq(rq)) { |
5450d3e1 | 1817 | if (__end_that_request_first(rq, error, nr_bytes)) |
336cdb40 | 1818 | return 1; |
e3a04fe3 KU |
1819 | |
1820 | /* Bidi request must be completed as a whole */ | |
1821 | if (blk_bidi_rq(rq) && | |
5450d3e1 | 1822 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) |
e3a04fe3 | 1823 | return 1; |
336cdb40 KU |
1824 | } |
1825 | ||
e19a3ab0 KU |
1826 | /* Special feature for tricky drivers */ |
1827 | if (drv_callback && drv_callback(rq)) | |
1828 | return 1; | |
1829 | ||
336cdb40 KU |
1830 | add_disk_randomness(rq->rq_disk); |
1831 | ||
1832 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1833 | end_that_request_last(rq, error); |
336cdb40 KU |
1834 | spin_unlock_irqrestore(q->queue_lock, flags); |
1835 | ||
1836 | return 0; | |
1837 | } | |
e19a3ab0 KU |
1838 | |
1839 | /** | |
1840 | * blk_end_request - Helper function for drivers to complete the request. | |
1841 | * @rq: the request being processed | |
710027a4 | 1842 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
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: | |
710027a4 RD |
1850 | * %0 - we are done with this request |
1851 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1852 | **/ |
22b13210 | 1853 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1854 | { |
e3a04fe3 | 1855 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1856 | } |
336cdb40 KU |
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 | |
710027a4 | 1862 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
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: | |
710027a4 RD |
1869 | * %0 - we are done with this request |
1870 | * %1 - still buffers pending for this request | |
336cdb40 | 1871 | **/ |
22b13210 | 1872 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1873 | { |
fb2dce86 | 1874 | if ((bio_has_data(rq->bio) || blk_discard_rq(rq)) && |
051cc395 JA |
1875 | __end_that_request_first(rq, error, nr_bytes)) |
1876 | return 1; | |
336cdb40 KU |
1877 | |
1878 | add_disk_randomness(rq->rq_disk); | |
1879 | ||
b8286239 | 1880 | end_that_request_last(rq, error); |
336cdb40 KU |
1881 | |
1882 | return 0; | |
1883 | } | |
1884 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1885 | ||
e3a04fe3 KU |
1886 | /** |
1887 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1888 | * @rq: the bidi request being processed | |
710027a4 | 1889 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
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: | |
710027a4 RD |
1897 | * %0 - we are done with this request |
1898 | * %1 - still buffers pending for this request | |
e3a04fe3 | 1899 | **/ |
22b13210 JA |
1900 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
1901 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
1902 | { |
1903 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
1904 | } | |
1905 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
1906 | ||
e19a3ab0 KU |
1907 | /** |
1908 | * blk_end_request_callback - Special helper function for tricky drivers | |
1909 | * @rq: the request being processed | |
710027a4 | 1910 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
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. | |
710027a4 | 1914 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
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: | |
710027a4 RD |
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. | |
e19a3ab0 | 1931 | **/ |
22b13210 JA |
1932 | int blk_end_request_callback(struct request *rq, int error, |
1933 | unsigned int nr_bytes, | |
e19a3ab0 KU |
1934 | int (drv_callback)(struct request *)) |
1935 | { | |
e3a04fe3 | 1936 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
1937 | } |
1938 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
1939 | ||
86db1e29 JA |
1940 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
1941 | struct bio *bio) | |
1da177e4 | 1942 | { |
d628eaef DW |
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). */ | |
4aff5e23 | 1945 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 1946 | |
fb2dce86 DW |
1947 | if (bio_has_data(bio)) { |
1948 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
1949 | rq->buffer = bio_data(bio); |
1950 | } | |
1da177e4 LT |
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); | |
0e75f906 | 1954 | rq->data_len = bio->bi_size; |
1da177e4 LT |
1955 | |
1956 | rq->bio = rq->biotail = bio; | |
1da177e4 | 1957 | |
66846572 N |
1958 | if (bio->bi_bdev) |
1959 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
1960 | } | |
1da177e4 | 1961 | |
18887ad9 | 1962 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
1963 | { |
1964 | return queue_work(kblockd_workqueue, work); | |
1965 | } | |
1da177e4 LT |
1966 | EXPORT_SYMBOL(kblockd_schedule_work); |
1967 | ||
19a75d83 | 1968 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 1969 | { |
28e53bdd | 1970 | cancel_work_sync(work); |
1da177e4 | 1971 | } |
19a75d83 | 1972 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
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", | |
20c2df83 | 1981 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 1982 | |
8324aa91 | 1983 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 1984 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 1985 | |
d38ecf93 | 1986 | return 0; |
1da177e4 | 1987 | } |
1da177e4 | 1988 |