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> |
55782138 LZ |
31 | |
32 | #define CREATE_TRACE_POINTS | |
33 | #include <trace/events/block.h> | |
1da177e4 | 34 | |
8324aa91 JA |
35 | #include "blk.h" |
36 | ||
0bfc2455 | 37 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); |
55782138 | 38 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 39 | |
165125e1 | 40 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
41 | |
42 | /* | |
43 | * For the allocated request tables | |
44 | */ | |
5ece6c52 | 45 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
46 | |
47 | /* | |
48 | * For queue allocation | |
49 | */ | |
6728cb0e | 50 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 51 | |
1da177e4 LT |
52 | /* |
53 | * Controlling structure to kblockd | |
54 | */ | |
ff856bad | 55 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 56 | |
26b8256e JA |
57 | static void drive_stat_acct(struct request *rq, int new_io) |
58 | { | |
28f13702 | 59 | struct hd_struct *part; |
26b8256e | 60 | int rw = rq_data_dir(rq); |
c9959059 | 61 | int cpu; |
26b8256e | 62 | |
26308eab | 63 | if (!blk_fs_request(rq) || !blk_do_io_stat(rq)) |
26b8256e JA |
64 | return; |
65 | ||
074a7aca | 66 | cpu = part_stat_lock(); |
e71bf0d0 | 67 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
c9959059 | 68 | |
28f13702 | 69 | if (!new_io) |
074a7aca | 70 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 71 | else { |
074a7aca TH |
72 | part_round_stats(cpu, part); |
73 | part_inc_in_flight(part); | |
26b8256e | 74 | } |
e71bf0d0 | 75 | |
074a7aca | 76 | part_stat_unlock(); |
26b8256e JA |
77 | } |
78 | ||
8324aa91 | 79 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
80 | { |
81 | int nr; | |
82 | ||
83 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
84 | if (nr > q->nr_requests) | |
85 | nr = q->nr_requests; | |
86 | q->nr_congestion_on = nr; | |
87 | ||
88 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
89 | if (nr < 1) | |
90 | nr = 1; | |
91 | q->nr_congestion_off = nr; | |
92 | } | |
93 | ||
1da177e4 LT |
94 | /** |
95 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
96 | * @bdev: device | |
97 | * | |
98 | * Locates the passed device's request queue and returns the address of its | |
99 | * backing_dev_info | |
100 | * | |
101 | * Will return NULL if the request queue cannot be located. | |
102 | */ | |
103 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
104 | { | |
105 | struct backing_dev_info *ret = NULL; | |
165125e1 | 106 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
107 | |
108 | if (q) | |
109 | ret = &q->backing_dev_info; | |
110 | return ret; | |
111 | } | |
1da177e4 LT |
112 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
113 | ||
2a4aa30c | 114 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 115 | { |
1afb20f3 FT |
116 | memset(rq, 0, sizeof(*rq)); |
117 | ||
1da177e4 | 118 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 119 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 120 | rq->cpu = -1; |
63a71386 JA |
121 | rq->q = q; |
122 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
123 | INIT_HLIST_NODE(&rq->hash); |
124 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 125 | rq->cmd = rq->__cmd; |
e2494e1b | 126 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 127 | rq->tag = -1; |
1da177e4 | 128 | rq->ref_count = 1; |
1da177e4 | 129 | } |
2a4aa30c | 130 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 131 | |
5bb23a68 N |
132 | static void req_bio_endio(struct request *rq, struct bio *bio, |
133 | unsigned int nbytes, int error) | |
1da177e4 | 134 | { |
165125e1 | 135 | struct request_queue *q = rq->q; |
797e7dbb | 136 | |
5bb23a68 N |
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; | |
797e7dbb | 142 | |
5bb23a68 | 143 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 144 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 145 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
146 | nbytes = bio->bi_size; |
147 | } | |
797e7dbb | 148 | |
08bafc03 KM |
149 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
150 | set_bit(BIO_QUIET, &bio->bi_flags); | |
151 | ||
5bb23a68 N |
152 | bio->bi_size -= nbytes; |
153 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
154 | |
155 | if (bio_integrity(bio)) | |
156 | bio_integrity_advance(bio, nbytes); | |
157 | ||
5bb23a68 | 158 | if (bio->bi_size == 0) |
6712ecf8 | 159 | bio_endio(bio, error); |
5bb23a68 N |
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 | } | |
1da177e4 | 169 | } |
1da177e4 | 170 | |
1da177e4 LT |
171 | void blk_dump_rq_flags(struct request *rq, char *msg) |
172 | { | |
173 | int bit; | |
174 | ||
6728cb0e | 175 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
176 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
177 | rq->cmd_flags); | |
1da177e4 | 178 | |
6728cb0e JA |
179 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
180 | (unsigned long long)rq->sector, | |
181 | rq->nr_sectors, | |
182 | rq->current_nr_sectors); | |
183 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
184 | rq->bio, rq->biotail, | |
185 | rq->buffer, rq->data, | |
186 | rq->data_len); | |
1da177e4 | 187 | |
4aff5e23 | 188 | if (blk_pc_request(rq)) { |
6728cb0e | 189 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 190 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
191 | printk("%02x ", rq->cmd[bit]); |
192 | printk("\n"); | |
193 | } | |
194 | } | |
1da177e4 LT |
195 | EXPORT_SYMBOL(blk_dump_rq_flags); |
196 | ||
1da177e4 LT |
197 | /* |
198 | * "plug" the device if there are no outstanding requests: this will | |
199 | * force the transfer to start only after we have put all the requests | |
200 | * on the list. | |
201 | * | |
202 | * This is called with interrupts off and no requests on the queue and | |
203 | * with the queue lock held. | |
204 | */ | |
165125e1 | 205 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
206 | { |
207 | WARN_ON(!irqs_disabled()); | |
208 | ||
209 | /* | |
210 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
211 | * which will restart the queueing | |
212 | */ | |
7daac490 | 213 | if (blk_queue_stopped(q)) |
1da177e4 LT |
214 | return; |
215 | ||
e48ec690 | 216 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 217 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 218 | trace_block_plug(q); |
2056a782 | 219 | } |
1da177e4 | 220 | } |
1da177e4 LT |
221 | EXPORT_SYMBOL(blk_plug_device); |
222 | ||
6c5e0c4d JA |
223 | /** |
224 | * blk_plug_device_unlocked - plug a device without queue lock held | |
225 | * @q: The &struct request_queue to plug | |
226 | * | |
227 | * Description: | |
228 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
229 | * interrupts. | |
230 | **/ | |
231 | void blk_plug_device_unlocked(struct request_queue *q) | |
232 | { | |
233 | unsigned long flags; | |
234 | ||
235 | spin_lock_irqsave(q->queue_lock, flags); | |
236 | blk_plug_device(q); | |
237 | spin_unlock_irqrestore(q->queue_lock, flags); | |
238 | } | |
239 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
240 | ||
1da177e4 LT |
241 | /* |
242 | * remove the queue from the plugged list, if present. called with | |
243 | * queue lock held and interrupts disabled. | |
244 | */ | |
165125e1 | 245 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
246 | { |
247 | WARN_ON(!irqs_disabled()); | |
248 | ||
e48ec690 | 249 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
250 | return 0; |
251 | ||
252 | del_timer(&q->unplug_timer); | |
253 | return 1; | |
254 | } | |
1da177e4 LT |
255 | EXPORT_SYMBOL(blk_remove_plug); |
256 | ||
257 | /* | |
258 | * remove the plug and let it rip.. | |
259 | */ | |
165125e1 | 260 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 261 | { |
7daac490 | 262 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 | 263 | return; |
a31a9738 | 264 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) |
1da177e4 LT |
265 | return; |
266 | ||
22e2c507 | 267 | q->request_fn(q); |
1da177e4 | 268 | } |
1da177e4 LT |
269 | |
270 | /** | |
271 | * generic_unplug_device - fire a request queue | |
165125e1 | 272 | * @q: The &struct request_queue in question |
1da177e4 LT |
273 | * |
274 | * Description: | |
275 | * Linux uses plugging to build bigger requests queues before letting | |
276 | * the device have at them. If a queue is plugged, the I/O scheduler | |
277 | * is still adding and merging requests on the queue. Once the queue | |
278 | * gets unplugged, the request_fn defined for the queue is invoked and | |
279 | * transfers started. | |
280 | **/ | |
165125e1 | 281 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 282 | { |
dbaf2c00 JA |
283 | if (blk_queue_plugged(q)) { |
284 | spin_lock_irq(q->queue_lock); | |
285 | __generic_unplug_device(q); | |
286 | spin_unlock_irq(q->queue_lock); | |
287 | } | |
1da177e4 LT |
288 | } |
289 | EXPORT_SYMBOL(generic_unplug_device); | |
290 | ||
291 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
292 | struct page *page) | |
293 | { | |
165125e1 | 294 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 295 | |
2ad8b1ef | 296 | blk_unplug(q); |
1da177e4 LT |
297 | } |
298 | ||
86db1e29 | 299 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 300 | { |
165125e1 JA |
301 | struct request_queue *q = |
302 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 303 | |
5f3ea37c | 304 | trace_block_unplug_io(q); |
1da177e4 LT |
305 | q->unplug_fn(q); |
306 | } | |
307 | ||
86db1e29 | 308 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 309 | { |
165125e1 | 310 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 311 | |
5f3ea37c | 312 | trace_block_unplug_timer(q); |
18887ad9 | 313 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
314 | } |
315 | ||
2ad8b1ef AB |
316 | void blk_unplug(struct request_queue *q) |
317 | { | |
318 | /* | |
319 | * devices don't necessarily have an ->unplug_fn defined | |
320 | */ | |
321 | if (q->unplug_fn) { | |
5f3ea37c | 322 | trace_block_unplug_io(q); |
2ad8b1ef AB |
323 | q->unplug_fn(q); |
324 | } | |
325 | } | |
326 | EXPORT_SYMBOL(blk_unplug); | |
327 | ||
c7c22e4d JA |
328 | static void blk_invoke_request_fn(struct request_queue *q) |
329 | { | |
80a4b58e JA |
330 | if (unlikely(blk_queue_stopped(q))) |
331 | return; | |
332 | ||
c7c22e4d JA |
333 | /* |
334 | * one level of recursion is ok and is much faster than kicking | |
335 | * the unplug handling | |
336 | */ | |
337 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { | |
338 | q->request_fn(q); | |
339 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
340 | } else { | |
341 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
342 | kblockd_schedule_work(q, &q->unplug_work); | |
343 | } | |
344 | } | |
345 | ||
1da177e4 LT |
346 | /** |
347 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 348 | * @q: The &struct request_queue in question |
1da177e4 LT |
349 | * |
350 | * Description: | |
351 | * blk_start_queue() will clear the stop flag on the queue, and call | |
352 | * the request_fn for the queue if it was in a stopped state when | |
353 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
354 | **/ | |
165125e1 | 355 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 356 | { |
a038e253 PBG |
357 | WARN_ON(!irqs_disabled()); |
358 | ||
75ad23bc | 359 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
c7c22e4d | 360 | blk_invoke_request_fn(q); |
1da177e4 | 361 | } |
1da177e4 LT |
362 | EXPORT_SYMBOL(blk_start_queue); |
363 | ||
364 | /** | |
365 | * blk_stop_queue - stop a queue | |
165125e1 | 366 | * @q: The &struct request_queue in question |
1da177e4 LT |
367 | * |
368 | * Description: | |
369 | * The Linux block layer assumes that a block driver will consume all | |
370 | * entries on the request queue when the request_fn strategy is called. | |
371 | * Often this will not happen, because of hardware limitations (queue | |
372 | * depth settings). If a device driver gets a 'queue full' response, | |
373 | * or if it simply chooses not to queue more I/O at one point, it can | |
374 | * call this function to prevent the request_fn from being called until | |
375 | * the driver has signalled it's ready to go again. This happens by calling | |
376 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
377 | **/ | |
165125e1 | 378 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
379 | { |
380 | blk_remove_plug(q); | |
75ad23bc | 381 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
382 | } |
383 | EXPORT_SYMBOL(blk_stop_queue); | |
384 | ||
385 | /** | |
386 | * blk_sync_queue - cancel any pending callbacks on a queue | |
387 | * @q: the queue | |
388 | * | |
389 | * Description: | |
390 | * The block layer may perform asynchronous callback activity | |
391 | * on a queue, such as calling the unplug function after a timeout. | |
392 | * A block device may call blk_sync_queue to ensure that any | |
393 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 394 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
395 | * that its ->make_request_fn will not re-add plugging prior to calling |
396 | * this function. | |
397 | * | |
398 | */ | |
399 | void blk_sync_queue(struct request_queue *q) | |
400 | { | |
401 | del_timer_sync(&q->unplug_timer); | |
70ed28b9 | 402 | del_timer_sync(&q->timeout); |
64d01dc9 | 403 | cancel_work_sync(&q->unplug_work); |
1da177e4 LT |
404 | } |
405 | EXPORT_SYMBOL(blk_sync_queue); | |
406 | ||
407 | /** | |
80a4b58e | 408 | * __blk_run_queue - run a single device queue |
1da177e4 | 409 | * @q: The queue to run |
80a4b58e JA |
410 | * |
411 | * Description: | |
412 | * See @blk_run_queue. This variant must be called with the queue lock | |
413 | * held and interrupts disabled. | |
414 | * | |
1da177e4 | 415 | */ |
75ad23bc | 416 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 417 | { |
1da177e4 | 418 | blk_remove_plug(q); |
dac07ec1 JA |
419 | |
420 | /* | |
421 | * Only recurse once to avoid overrunning the stack, let the unplug | |
422 | * handling reinvoke the handler shortly if we already got there. | |
423 | */ | |
c7c22e4d JA |
424 | if (!elv_queue_empty(q)) |
425 | blk_invoke_request_fn(q); | |
75ad23bc NP |
426 | } |
427 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 428 | |
75ad23bc NP |
429 | /** |
430 | * blk_run_queue - run a single device queue | |
431 | * @q: The queue to run | |
80a4b58e JA |
432 | * |
433 | * Description: | |
434 | * Invoke request handling on this queue, if it has pending work to do. | |
435 | * May be used to restart queueing when a request has completed. Also | |
436 | * See @blk_start_queueing. | |
437 | * | |
75ad23bc NP |
438 | */ |
439 | void blk_run_queue(struct request_queue *q) | |
440 | { | |
441 | unsigned long flags; | |
442 | ||
443 | spin_lock_irqsave(q->queue_lock, flags); | |
444 | __blk_run_queue(q); | |
1da177e4 LT |
445 | spin_unlock_irqrestore(q->queue_lock, flags); |
446 | } | |
447 | EXPORT_SYMBOL(blk_run_queue); | |
448 | ||
165125e1 | 449 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
450 | { |
451 | kobject_put(&q->kobj); | |
452 | } | |
483f4afc | 453 | |
6728cb0e | 454 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 455 | { |
e3335de9 JA |
456 | /* |
457 | * We know we have process context here, so we can be a little | |
458 | * cautious and ensure that pending block actions on this device | |
459 | * are done before moving on. Going into this function, we should | |
460 | * not have processes doing IO to this device. | |
461 | */ | |
462 | blk_sync_queue(q); | |
463 | ||
483f4afc | 464 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 465 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
466 | mutex_unlock(&q->sysfs_lock); |
467 | ||
468 | if (q->elevator) | |
469 | elevator_exit(q->elevator); | |
470 | ||
471 | blk_put_queue(q); | |
472 | } | |
1da177e4 LT |
473 | EXPORT_SYMBOL(blk_cleanup_queue); |
474 | ||
165125e1 | 475 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
476 | { |
477 | struct request_list *rl = &q->rq; | |
478 | ||
1faa16d2 JA |
479 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
480 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 481 | rl->elvpriv = 0; |
1faa16d2 JA |
482 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
483 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 484 | |
1946089a CL |
485 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
486 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
487 | |
488 | if (!rl->rq_pool) | |
489 | return -ENOMEM; | |
490 | ||
491 | return 0; | |
492 | } | |
493 | ||
165125e1 | 494 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 495 | { |
1946089a CL |
496 | return blk_alloc_queue_node(gfp_mask, -1); |
497 | } | |
498 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 499 | |
165125e1 | 500 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 501 | { |
165125e1 | 502 | struct request_queue *q; |
e0bf68dd | 503 | int err; |
1946089a | 504 | |
8324aa91 | 505 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 506 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
507 | if (!q) |
508 | return NULL; | |
509 | ||
e0bf68dd PZ |
510 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
511 | q->backing_dev_info.unplug_io_data = q; | |
512 | err = bdi_init(&q->backing_dev_info); | |
513 | if (err) { | |
8324aa91 | 514 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
515 | return NULL; |
516 | } | |
517 | ||
1da177e4 | 518 | init_timer(&q->unplug_timer); |
242f9dcb JA |
519 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
520 | INIT_LIST_HEAD(&q->timeout_list); | |
713ada9b | 521 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 522 | |
8324aa91 | 523 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 524 | |
483f4afc | 525 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 526 | spin_lock_init(&q->__queue_lock); |
483f4afc | 527 | |
1da177e4 LT |
528 | return q; |
529 | } | |
1946089a | 530 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
531 | |
532 | /** | |
533 | * blk_init_queue - prepare a request queue for use with a block device | |
534 | * @rfn: The function to be called to process requests that have been | |
535 | * placed on the queue. | |
536 | * @lock: Request queue spin lock | |
537 | * | |
538 | * Description: | |
539 | * If a block device wishes to use the standard request handling procedures, | |
540 | * which sorts requests and coalesces adjacent requests, then it must | |
541 | * call blk_init_queue(). The function @rfn will be called when there | |
542 | * are requests on the queue that need to be processed. If the device | |
543 | * supports plugging, then @rfn may not be called immediately when requests | |
544 | * are available on the queue, but may be called at some time later instead. | |
545 | * Plugged queues are generally unplugged when a buffer belonging to one | |
546 | * of the requests on the queue is needed, or due to memory pressure. | |
547 | * | |
548 | * @rfn is not required, or even expected, to remove all requests off the | |
549 | * queue, but only as many as it can handle at a time. If it does leave | |
550 | * requests on the queue, it is responsible for arranging that the requests | |
551 | * get dealt with eventually. | |
552 | * | |
553 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
554 | * request queue; this lock will be taken also from interrupt context, so irq |
555 | * disabling is needed for it. | |
1da177e4 | 556 | * |
710027a4 | 557 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
558 | * it didn't succeed. |
559 | * | |
560 | * Note: | |
561 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
562 | * when the block device is deactivated (such as at module unload). | |
563 | **/ | |
1946089a | 564 | |
165125e1 | 565 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 566 | { |
1946089a CL |
567 | return blk_init_queue_node(rfn, lock, -1); |
568 | } | |
569 | EXPORT_SYMBOL(blk_init_queue); | |
570 | ||
165125e1 | 571 | struct request_queue * |
1946089a CL |
572 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
573 | { | |
165125e1 | 574 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
575 | |
576 | if (!q) | |
577 | return NULL; | |
578 | ||
1946089a | 579 | q->node = node_id; |
8669aafd | 580 | if (blk_init_free_list(q)) { |
8324aa91 | 581 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
582 | return NULL; |
583 | } | |
1da177e4 | 584 | |
152587de | 585 | /* |
586 | * if caller didn't supply a lock, they get per-queue locking with | |
587 | * our embedded lock | |
588 | */ | |
e7e72bf6 | 589 | if (!lock) |
152587de | 590 | lock = &q->__queue_lock; |
152587de | 591 | |
1da177e4 | 592 | q->request_fn = rfn; |
1da177e4 LT |
593 | q->prep_rq_fn = NULL; |
594 | q->unplug_fn = generic_unplug_device; | |
bc58ba94 | 595 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
1da177e4 LT |
596 | q->queue_lock = lock; |
597 | ||
f3b144aa JA |
598 | /* |
599 | * This also sets hw/phys segments, boundary and size | |
600 | */ | |
1da177e4 | 601 | blk_queue_make_request(q, __make_request); |
1da177e4 | 602 | |
44ec9542 AS |
603 | q->sg_reserved_size = INT_MAX; |
604 | ||
abf54393 FT |
605 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
606 | ||
1da177e4 LT |
607 | /* |
608 | * all done | |
609 | */ | |
610 | if (!elevator_init(q, NULL)) { | |
611 | blk_queue_congestion_threshold(q); | |
612 | return q; | |
613 | } | |
614 | ||
8669aafd | 615 | blk_put_queue(q); |
1da177e4 LT |
616 | return NULL; |
617 | } | |
1946089a | 618 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 619 | |
165125e1 | 620 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 621 | { |
fde6ad22 | 622 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 623 | kobject_get(&q->kobj); |
1da177e4 LT |
624 | return 0; |
625 | } | |
626 | ||
627 | return 1; | |
628 | } | |
1da177e4 | 629 | |
165125e1 | 630 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 631 | { |
4aff5e23 | 632 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 633 | elv_put_request(q, rq); |
1da177e4 LT |
634 | mempool_free(rq, q->rq.rq_pool); |
635 | } | |
636 | ||
1ea25ecb | 637 | static struct request * |
42dad764 | 638 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) |
1da177e4 LT |
639 | { |
640 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
641 | ||
642 | if (!rq) | |
643 | return NULL; | |
644 | ||
2a4aa30c | 645 | blk_rq_init(q, rq); |
1afb20f3 | 646 | |
42dad764 | 647 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 648 | |
cb98fc8b | 649 | if (priv) { |
cb78b285 | 650 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
651 | mempool_free(rq, q->rq.rq_pool); |
652 | return NULL; | |
653 | } | |
4aff5e23 | 654 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 655 | } |
1da177e4 | 656 | |
cb98fc8b | 657 | return rq; |
1da177e4 LT |
658 | } |
659 | ||
660 | /* | |
661 | * ioc_batching returns true if the ioc is a valid batching request and | |
662 | * should be given priority access to a request. | |
663 | */ | |
165125e1 | 664 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
665 | { |
666 | if (!ioc) | |
667 | return 0; | |
668 | ||
669 | /* | |
670 | * Make sure the process is able to allocate at least 1 request | |
671 | * even if the batch times out, otherwise we could theoretically | |
672 | * lose wakeups. | |
673 | */ | |
674 | return ioc->nr_batch_requests == q->nr_batching || | |
675 | (ioc->nr_batch_requests > 0 | |
676 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
677 | } | |
678 | ||
679 | /* | |
680 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
681 | * will cause the process to be a "batcher" on all queues in the system. This | |
682 | * is the behaviour we want though - once it gets a wakeup it should be given | |
683 | * a nice run. | |
684 | */ | |
165125e1 | 685 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
686 | { |
687 | if (!ioc || ioc_batching(q, ioc)) | |
688 | return; | |
689 | ||
690 | ioc->nr_batch_requests = q->nr_batching; | |
691 | ioc->last_waited = jiffies; | |
692 | } | |
693 | ||
1faa16d2 | 694 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
695 | { |
696 | struct request_list *rl = &q->rq; | |
697 | ||
1faa16d2 JA |
698 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
699 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 700 | |
1faa16d2 JA |
701 | if (rl->count[sync] + 1 <= q->nr_requests) { |
702 | if (waitqueue_active(&rl->wait[sync])) | |
703 | wake_up(&rl->wait[sync]); | |
1da177e4 | 704 | |
1faa16d2 | 705 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
706 | } |
707 | } | |
708 | ||
709 | /* | |
710 | * A request has just been released. Account for it, update the full and | |
711 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
712 | */ | |
1faa16d2 | 713 | static void freed_request(struct request_queue *q, int sync, int priv) |
1da177e4 LT |
714 | { |
715 | struct request_list *rl = &q->rq; | |
716 | ||
1faa16d2 | 717 | rl->count[sync]--; |
cb98fc8b TH |
718 | if (priv) |
719 | rl->elvpriv--; | |
1da177e4 | 720 | |
1faa16d2 | 721 | __freed_request(q, sync); |
1da177e4 | 722 | |
1faa16d2 JA |
723 | if (unlikely(rl->starved[sync ^ 1])) |
724 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
725 | } |
726 | ||
1da177e4 | 727 | /* |
d6344532 NP |
728 | * Get a free request, queue_lock must be held. |
729 | * Returns NULL on failure, with queue_lock held. | |
730 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 731 | */ |
165125e1 | 732 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 733 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
734 | { |
735 | struct request *rq = NULL; | |
736 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 737 | struct io_context *ioc = NULL; |
1faa16d2 | 738 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
88ee5ef1 JA |
739 | int may_queue, priv; |
740 | ||
7749a8d4 | 741 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
742 | if (may_queue == ELV_MQUEUE_NO) |
743 | goto rq_starved; | |
744 | ||
1faa16d2 JA |
745 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
746 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 747 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
748 | /* |
749 | * The queue will fill after this allocation, so set | |
750 | * it as full, and mark this process as "batching". | |
751 | * This process will be allowed to complete a batch of | |
752 | * requests, others will be blocked. | |
753 | */ | |
1faa16d2 | 754 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 755 | ioc_set_batching(q, ioc); |
1faa16d2 | 756 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
757 | } else { |
758 | if (may_queue != ELV_MQUEUE_MUST | |
759 | && !ioc_batching(q, ioc)) { | |
760 | /* | |
761 | * The queue is full and the allocating | |
762 | * process is not a "batcher", and not | |
763 | * exempted by the IO scheduler | |
764 | */ | |
765 | goto out; | |
766 | } | |
767 | } | |
1da177e4 | 768 | } |
1faa16d2 | 769 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
770 | } |
771 | ||
082cf69e JA |
772 | /* |
773 | * Only allow batching queuers to allocate up to 50% over the defined | |
774 | * limit of requests, otherwise we could have thousands of requests | |
775 | * allocated with any setting of ->nr_requests | |
776 | */ | |
1faa16d2 | 777 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 778 | goto out; |
fd782a4a | 779 | |
1faa16d2 JA |
780 | rl->count[is_sync]++; |
781 | rl->starved[is_sync] = 0; | |
cb98fc8b | 782 | |
64521d1a | 783 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
784 | if (priv) |
785 | rl->elvpriv++; | |
786 | ||
42dad764 JM |
787 | if (blk_queue_io_stat(q)) |
788 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
789 | spin_unlock_irq(q->queue_lock); |
790 | ||
7749a8d4 | 791 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 792 | if (unlikely(!rq)) { |
1da177e4 LT |
793 | /* |
794 | * Allocation failed presumably due to memory. Undo anything | |
795 | * we might have messed up. | |
796 | * | |
797 | * Allocating task should really be put onto the front of the | |
798 | * wait queue, but this is pretty rare. | |
799 | */ | |
800 | spin_lock_irq(q->queue_lock); | |
1faa16d2 | 801 | freed_request(q, is_sync, priv); |
1da177e4 LT |
802 | |
803 | /* | |
804 | * in the very unlikely event that allocation failed and no | |
805 | * requests for this direction was pending, mark us starved | |
806 | * so that freeing of a request in the other direction will | |
807 | * notice us. another possible fix would be to split the | |
808 | * rq mempool into READ and WRITE | |
809 | */ | |
810 | rq_starved: | |
1faa16d2 JA |
811 | if (unlikely(rl->count[is_sync] == 0)) |
812 | rl->starved[is_sync] = 1; | |
1da177e4 | 813 | |
1da177e4 LT |
814 | goto out; |
815 | } | |
816 | ||
88ee5ef1 JA |
817 | /* |
818 | * ioc may be NULL here, and ioc_batching will be false. That's | |
819 | * OK, if the queue is under the request limit then requests need | |
820 | * not count toward the nr_batch_requests limit. There will always | |
821 | * be some limit enforced by BLK_BATCH_TIME. | |
822 | */ | |
1da177e4 LT |
823 | if (ioc_batching(q, ioc)) |
824 | ioc->nr_batch_requests--; | |
6728cb0e | 825 | |
1faa16d2 | 826 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 827 | out: |
1da177e4 LT |
828 | return rq; |
829 | } | |
830 | ||
831 | /* | |
832 | * No available requests for this queue, unplug the device and wait for some | |
833 | * requests to become available. | |
d6344532 NP |
834 | * |
835 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 836 | */ |
165125e1 | 837 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 838 | struct bio *bio) |
1da177e4 | 839 | { |
1faa16d2 | 840 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
841 | struct request *rq; |
842 | ||
7749a8d4 | 843 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
844 | while (!rq) { |
845 | DEFINE_WAIT(wait); | |
05caf8db | 846 | struct io_context *ioc; |
1da177e4 LT |
847 | struct request_list *rl = &q->rq; |
848 | ||
1faa16d2 | 849 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
850 | TASK_UNINTERRUPTIBLE); |
851 | ||
1faa16d2 | 852 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 853 | |
05caf8db ZY |
854 | __generic_unplug_device(q); |
855 | spin_unlock_irq(q->queue_lock); | |
856 | io_schedule(); | |
1da177e4 | 857 | |
05caf8db ZY |
858 | /* |
859 | * After sleeping, we become a "batching" process and | |
860 | * will be able to allocate at least one request, and | |
861 | * up to a big batch of them for a small period time. | |
862 | * See ioc_batching, ioc_set_batching | |
863 | */ | |
864 | ioc = current_io_context(GFP_NOIO, q->node); | |
865 | ioc_set_batching(q, ioc); | |
d6344532 | 866 | |
05caf8db | 867 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 868 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
869 | |
870 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
871 | }; | |
1da177e4 LT |
872 | |
873 | return rq; | |
874 | } | |
875 | ||
165125e1 | 876 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
877 | { |
878 | struct request *rq; | |
879 | ||
880 | BUG_ON(rw != READ && rw != WRITE); | |
881 | ||
d6344532 NP |
882 | spin_lock_irq(q->queue_lock); |
883 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 884 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 885 | } else { |
22e2c507 | 886 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
887 | if (!rq) |
888 | spin_unlock_irq(q->queue_lock); | |
889 | } | |
890 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
891 | |
892 | return rq; | |
893 | } | |
1da177e4 LT |
894 | EXPORT_SYMBOL(blk_get_request); |
895 | ||
dc72ef4a JA |
896 | /** |
897 | * blk_start_queueing - initiate dispatch of requests to device | |
898 | * @q: request queue to kick into gear | |
899 | * | |
900 | * This is basically a helper to remove the need to know whether a queue | |
901 | * is plugged or not if someone just wants to initiate dispatch of requests | |
80a4b58e JA |
902 | * for this queue. Should be used to start queueing on a device outside |
903 | * of ->request_fn() context. Also see @blk_run_queue. | |
dc72ef4a JA |
904 | * |
905 | * The queue lock must be held with interrupts disabled. | |
906 | */ | |
165125e1 | 907 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a | 908 | { |
336c3d8c EO |
909 | if (!blk_queue_plugged(q)) { |
910 | if (unlikely(blk_queue_stopped(q))) | |
911 | return; | |
dc72ef4a | 912 | q->request_fn(q); |
336c3d8c | 913 | } else |
dc72ef4a JA |
914 | __generic_unplug_device(q); |
915 | } | |
916 | EXPORT_SYMBOL(blk_start_queueing); | |
917 | ||
1da177e4 LT |
918 | /** |
919 | * blk_requeue_request - put a request back on queue | |
920 | * @q: request queue where request should be inserted | |
921 | * @rq: request to be inserted | |
922 | * | |
923 | * Description: | |
924 | * Drivers often keep queueing requests until the hardware cannot accept | |
925 | * more, when that condition happens we need to put the request back | |
926 | * on the queue. Must be called with queue lock held. | |
927 | */ | |
165125e1 | 928 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 929 | { |
242f9dcb JA |
930 | blk_delete_timer(rq); |
931 | blk_clear_rq_complete(rq); | |
5f3ea37c | 932 | trace_block_rq_requeue(q, rq); |
2056a782 | 933 | |
1da177e4 LT |
934 | if (blk_rq_tagged(rq)) |
935 | blk_queue_end_tag(q, rq); | |
936 | ||
937 | elv_requeue_request(q, rq); | |
938 | } | |
1da177e4 LT |
939 | EXPORT_SYMBOL(blk_requeue_request); |
940 | ||
941 | /** | |
710027a4 | 942 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
943 | * @q: request queue where request should be inserted |
944 | * @rq: request to be inserted | |
945 | * @at_head: insert request at head or tail of queue | |
946 | * @data: private data | |
1da177e4 LT |
947 | * |
948 | * Description: | |
949 | * Many block devices need to execute commands asynchronously, so they don't | |
950 | * block the whole kernel from preemption during request execution. This is | |
951 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
952 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
953 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
954 | * |
955 | * We have the option of inserting the head or the tail of the queue. | |
956 | * Typically we use the tail for new ioctls and so forth. We use the head | |
957 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
958 | * host that is unable to accept a particular command. | |
959 | */ | |
165125e1 | 960 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 961 | int at_head, void *data) |
1da177e4 | 962 | { |
867d1191 | 963 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
964 | unsigned long flags; |
965 | ||
966 | /* | |
967 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
968 | * must not attempt merges on this) and that it acts as a soft | |
969 | * barrier | |
970 | */ | |
4aff5e23 JA |
971 | rq->cmd_type = REQ_TYPE_SPECIAL; |
972 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
973 | |
974 | rq->special = data; | |
975 | ||
976 | spin_lock_irqsave(q->queue_lock, flags); | |
977 | ||
978 | /* | |
979 | * If command is tagged, release the tag | |
980 | */ | |
867d1191 TH |
981 | if (blk_rq_tagged(rq)) |
982 | blk_queue_end_tag(q, rq); | |
1da177e4 | 983 | |
b238b3d4 | 984 | drive_stat_acct(rq, 1); |
867d1191 | 985 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 986 | blk_start_queueing(q); |
1da177e4 LT |
987 | spin_unlock_irqrestore(q->queue_lock, flags); |
988 | } | |
1da177e4 LT |
989 | EXPORT_SYMBOL(blk_insert_request); |
990 | ||
1da177e4 LT |
991 | /* |
992 | * add-request adds a request to the linked list. | |
993 | * queue lock is held and interrupts disabled, as we muck with the | |
994 | * request queue list. | |
995 | */ | |
6728cb0e | 996 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 997 | { |
b238b3d4 | 998 | drive_stat_acct(req, 1); |
1da177e4 | 999 | |
1da177e4 LT |
1000 | /* |
1001 | * elevator indicated where it wants this request to be | |
1002 | * inserted at elevator_merge time | |
1003 | */ | |
1004 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
1005 | } | |
6728cb0e | 1006 | |
074a7aca TH |
1007 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1008 | unsigned long now) | |
1009 | { | |
1010 | if (now == part->stamp) | |
1011 | return; | |
1012 | ||
1013 | if (part->in_flight) { | |
1014 | __part_stat_add(cpu, part, time_in_queue, | |
1015 | part->in_flight * (now - part->stamp)); | |
1016 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
1017 | } | |
1018 | part->stamp = now; | |
1019 | } | |
1020 | ||
1021 | /** | |
496aa8a9 RD |
1022 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1023 | * @cpu: cpu number for stats access | |
1024 | * @part: target partition | |
1da177e4 LT |
1025 | * |
1026 | * The average IO queue length and utilisation statistics are maintained | |
1027 | * by observing the current state of the queue length and the amount of | |
1028 | * time it has been in this state for. | |
1029 | * | |
1030 | * Normally, that accounting is done on IO completion, but that can result | |
1031 | * in more than a second's worth of IO being accounted for within any one | |
1032 | * second, leading to >100% utilisation. To deal with that, we call this | |
1033 | * function to do a round-off before returning the results when reading | |
1034 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1035 | * the current jiffies and restarts the counters again. | |
1036 | */ | |
c9959059 | 1037 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1038 | { |
1039 | unsigned long now = jiffies; | |
1040 | ||
074a7aca TH |
1041 | if (part->partno) |
1042 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1043 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1044 | } |
074a7aca | 1045 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1046 | |
1da177e4 LT |
1047 | /* |
1048 | * queue lock must be held | |
1049 | */ | |
165125e1 | 1050 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1051 | { |
1da177e4 LT |
1052 | if (unlikely(!q)) |
1053 | return; | |
1054 | if (unlikely(--req->ref_count)) | |
1055 | return; | |
1056 | ||
8922e16c TH |
1057 | elv_completed_request(q, req); |
1058 | ||
1cd96c24 BH |
1059 | /* this is a bio leak */ |
1060 | WARN_ON(req->bio != NULL); | |
1061 | ||
1da177e4 LT |
1062 | /* |
1063 | * Request may not have originated from ll_rw_blk. if not, | |
1064 | * it didn't come out of our reserved rq pools | |
1065 | */ | |
49171e5c | 1066 | if (req->cmd_flags & REQ_ALLOCED) { |
1faa16d2 | 1067 | int is_sync = rq_is_sync(req) != 0; |
4aff5e23 | 1068 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1069 | |
1da177e4 | 1070 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1071 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1072 | |
1073 | blk_free_request(q, req); | |
1faa16d2 | 1074 | freed_request(q, is_sync, priv); |
1da177e4 LT |
1075 | } |
1076 | } | |
6e39b69e MC |
1077 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1078 | ||
1da177e4 LT |
1079 | void blk_put_request(struct request *req) |
1080 | { | |
8922e16c | 1081 | unsigned long flags; |
165125e1 | 1082 | struct request_queue *q = req->q; |
8922e16c | 1083 | |
52a93ba8 FT |
1084 | spin_lock_irqsave(q->queue_lock, flags); |
1085 | __blk_put_request(q, req); | |
1086 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1087 | } |
1da177e4 LT |
1088 | EXPORT_SYMBOL(blk_put_request); |
1089 | ||
86db1e29 | 1090 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1091 | { |
c7c22e4d | 1092 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1093 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1094 | |
1095 | /* | |
1096 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1097 | */ | |
6000a368 MC |
1098 | if (bio_rw_ahead(bio)) |
1099 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1100 | REQ_FAILFAST_DRIVER); | |
1101 | if (bio_failfast_dev(bio)) | |
1102 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1103 | if (bio_failfast_transport(bio)) | |
1104 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1105 | if (bio_failfast_driver(bio)) | |
1106 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
52d9e675 TH |
1107 | |
1108 | /* | |
1109 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1110 | */ | |
fb2dce86 | 1111 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1112 | req->cmd_flags |= REQ_DISCARD; |
1113 | if (bio_barrier(bio)) | |
1114 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1115 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 DW |
1116 | } else if (unlikely(bio_barrier(bio))) |
1117 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | |
52d9e675 | 1118 | |
b31dc66a | 1119 | if (bio_sync(bio)) |
4aff5e23 | 1120 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1121 | if (bio_rw_meta(bio)) |
1122 | req->cmd_flags |= REQ_RW_META; | |
aeb6fafb JA |
1123 | if (bio_noidle(bio)) |
1124 | req->cmd_flags |= REQ_NOIDLE; | |
b31dc66a | 1125 | |
52d9e675 TH |
1126 | req->errors = 0; |
1127 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1128 | req->ioprio = bio_prio(bio); |
52d9e675 | 1129 | req->start_time = jiffies; |
bc1c56fd | 1130 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1131 | } |
1132 | ||
644b2d99 JA |
1133 | /* |
1134 | * Only disabling plugging for non-rotational devices if it does tagging | |
1135 | * as well, otherwise we do need the proper merging | |
1136 | */ | |
1137 | static inline bool queue_should_plug(struct request_queue *q) | |
1138 | { | |
1139 | return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); | |
1140 | } | |
1141 | ||
165125e1 | 1142 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1143 | { |
450991bc | 1144 | struct request *req; |
a7384677 | 1145 | int el_ret, nr_sectors; |
51da90fc JA |
1146 | const unsigned short prio = bio_prio(bio); |
1147 | const int sync = bio_sync(bio); | |
213d9417 | 1148 | const int unplug = bio_unplug(bio); |
7749a8d4 | 1149 | int rw_flags; |
1da177e4 | 1150 | |
1da177e4 | 1151 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1152 | |
1153 | /* | |
1154 | * low level driver can indicate that it wants pages above a | |
1155 | * certain limit bounced to low memory (ie for highmem, or even | |
1156 | * ISA dma in theory) | |
1157 | */ | |
1158 | blk_queue_bounce(q, &bio); | |
1159 | ||
1da177e4 LT |
1160 | spin_lock_irq(q->queue_lock); |
1161 | ||
a7384677 | 1162 | if (unlikely(bio_barrier(bio)) || elv_queue_empty(q)) |
1da177e4 LT |
1163 | goto get_rq; |
1164 | ||
1165 | el_ret = elv_merge(q, &req, bio); | |
1166 | switch (el_ret) { | |
6728cb0e JA |
1167 | case ELEVATOR_BACK_MERGE: |
1168 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1169 | |
6728cb0e JA |
1170 | if (!ll_back_merge_fn(q, req, bio)) |
1171 | break; | |
1da177e4 | 1172 | |
5f3ea37c | 1173 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1174 | |
6728cb0e JA |
1175 | req->biotail->bi_next = bio; |
1176 | req->biotail = bio; | |
1177 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1178 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1179 | if (!blk_rq_cpu_valid(req)) |
1180 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1181 | drive_stat_acct(req, 0); |
1182 | if (!attempt_back_merge(q, req)) | |
1183 | elv_merged_request(q, req, el_ret); | |
1184 | goto out; | |
1da177e4 | 1185 | |
6728cb0e JA |
1186 | case ELEVATOR_FRONT_MERGE: |
1187 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1188 | |
6728cb0e JA |
1189 | if (!ll_front_merge_fn(q, req, bio)) |
1190 | break; | |
1da177e4 | 1191 | |
5f3ea37c | 1192 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1193 | |
6728cb0e JA |
1194 | bio->bi_next = req->bio; |
1195 | req->bio = bio; | |
1da177e4 | 1196 | |
6728cb0e JA |
1197 | /* |
1198 | * may not be valid. if the low level driver said | |
1199 | * it didn't need a bounce buffer then it better | |
1200 | * not touch req->buffer either... | |
1201 | */ | |
1202 | req->buffer = bio_data(bio); | |
1203 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1204 | req->hard_cur_sectors = req->current_nr_sectors; | |
1205 | req->sector = req->hard_sector = bio->bi_sector; | |
1206 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1207 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1208 | if (!blk_rq_cpu_valid(req)) |
1209 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1210 | drive_stat_acct(req, 0); |
1211 | if (!attempt_front_merge(q, req)) | |
1212 | elv_merged_request(q, req, el_ret); | |
1213 | goto out; | |
1214 | ||
1215 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1216 | default: | |
1217 | ; | |
1da177e4 LT |
1218 | } |
1219 | ||
450991bc | 1220 | get_rq: |
7749a8d4 JA |
1221 | /* |
1222 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1223 | * but we need to set it earlier to expose the sync flag to the | |
1224 | * rq allocator and io schedulers. | |
1225 | */ | |
1226 | rw_flags = bio_data_dir(bio); | |
1227 | if (sync) | |
1228 | rw_flags |= REQ_RW_SYNC; | |
1229 | ||
1da177e4 | 1230 | /* |
450991bc | 1231 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1232 | * Returns with the queue unlocked. |
450991bc | 1233 | */ |
7749a8d4 | 1234 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1235 | |
450991bc NP |
1236 | /* |
1237 | * After dropping the lock and possibly sleeping here, our request | |
1238 | * may now be mergeable after it had proven unmergeable (above). | |
1239 | * We don't worry about that case for efficiency. It won't happen | |
1240 | * often, and the elevators are able to handle it. | |
1da177e4 | 1241 | */ |
52d9e675 | 1242 | init_request_from_bio(req, bio); |
1da177e4 | 1243 | |
450991bc | 1244 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1245 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1246 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1247 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
644b2d99 | 1248 | if (queue_should_plug(q) && elv_queue_empty(q)) |
450991bc | 1249 | blk_plug_device(q); |
1da177e4 LT |
1250 | add_request(q, req); |
1251 | out: | |
644b2d99 | 1252 | if (unplug || !queue_should_plug(q)) |
1da177e4 | 1253 | __generic_unplug_device(q); |
1da177e4 LT |
1254 | spin_unlock_irq(q->queue_lock); |
1255 | return 0; | |
1da177e4 LT |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * If bio->bi_dev is a partition, remap the location | |
1260 | */ | |
1261 | static inline void blk_partition_remap(struct bio *bio) | |
1262 | { | |
1263 | struct block_device *bdev = bio->bi_bdev; | |
1264 | ||
bf2de6f5 | 1265 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1266 | struct hd_struct *p = bdev->bd_part; |
1267 | ||
1da177e4 LT |
1268 | bio->bi_sector += p->start_sect; |
1269 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1270 | |
5f3ea37c | 1271 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, |
22a7c31a | 1272 | bdev->bd_dev, |
c7149d6b | 1273 | bio->bi_sector - p->start_sect); |
1da177e4 LT |
1274 | } |
1275 | } | |
1276 | ||
1da177e4 LT |
1277 | static void handle_bad_sector(struct bio *bio) |
1278 | { | |
1279 | char b[BDEVNAME_SIZE]; | |
1280 | ||
1281 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1282 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1283 | bdevname(bio->bi_bdev, b), | |
1284 | bio->bi_rw, | |
1285 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1286 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1287 | ||
1288 | set_bit(BIO_EOF, &bio->bi_flags); | |
1289 | } | |
1290 | ||
c17bb495 AM |
1291 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1292 | ||
1293 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1294 | ||
1295 | static int __init setup_fail_make_request(char *str) | |
1296 | { | |
1297 | return setup_fault_attr(&fail_make_request, str); | |
1298 | } | |
1299 | __setup("fail_make_request=", setup_fail_make_request); | |
1300 | ||
1301 | static int should_fail_request(struct bio *bio) | |
1302 | { | |
eddb2e26 TH |
1303 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1304 | ||
1305 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1306 | return should_fail(&fail_make_request, bio->bi_size); |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | static int __init fail_make_request_debugfs(void) | |
1312 | { | |
1313 | return init_fault_attr_dentries(&fail_make_request, | |
1314 | "fail_make_request"); | |
1315 | } | |
1316 | ||
1317 | late_initcall(fail_make_request_debugfs); | |
1318 | ||
1319 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1320 | ||
1321 | static inline int should_fail_request(struct bio *bio) | |
1322 | { | |
1323 | return 0; | |
1324 | } | |
1325 | ||
1326 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1327 | ||
c07e2b41 JA |
1328 | /* |
1329 | * Check whether this bio extends beyond the end of the device. | |
1330 | */ | |
1331 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1332 | { | |
1333 | sector_t maxsector; | |
1334 | ||
1335 | if (!nr_sectors) | |
1336 | return 0; | |
1337 | ||
1338 | /* Test device or partition size, when known. */ | |
1339 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1340 | if (maxsector) { | |
1341 | sector_t sector = bio->bi_sector; | |
1342 | ||
1343 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1344 | /* | |
1345 | * This may well happen - the kernel calls bread() | |
1346 | * without checking the size of the device, e.g., when | |
1347 | * mounting a device. | |
1348 | */ | |
1349 | handle_bad_sector(bio); | |
1350 | return 1; | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | return 0; | |
1355 | } | |
1356 | ||
1da177e4 | 1357 | /** |
710027a4 | 1358 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1359 | * @bio: The bio describing the location in memory and on the device. |
1360 | * | |
1361 | * generic_make_request() is used to make I/O requests of block | |
1362 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1363 | * to be done. | |
1364 | * | |
1365 | * generic_make_request() does not return any status. The | |
1366 | * success/failure status of the request, along with notification of | |
1367 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1368 | * function described (one day) else where. | |
1369 | * | |
1370 | * The caller of generic_make_request must make sure that bi_io_vec | |
1371 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1372 | * set to describe the device address, and the | |
1373 | * bi_end_io and optionally bi_private are set to describe how | |
1374 | * completion notification should be signaled. | |
1375 | * | |
1376 | * generic_make_request and the drivers it calls may use bi_next if this | |
1377 | * bio happens to be merged with someone else, and may change bi_dev and | |
1378 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1379 | * should NOT be depended on after the call to generic_make_request. | |
1380 | */ | |
d89d8796 | 1381 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1382 | { |
165125e1 | 1383 | struct request_queue *q; |
5ddfe969 | 1384 | sector_t old_sector; |
1da177e4 | 1385 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1386 | dev_t old_dev; |
51fd77bd | 1387 | int err = -EIO; |
1da177e4 LT |
1388 | |
1389 | might_sleep(); | |
1da177e4 | 1390 | |
c07e2b41 JA |
1391 | if (bio_check_eod(bio, nr_sectors)) |
1392 | goto end_io; | |
1da177e4 LT |
1393 | |
1394 | /* | |
1395 | * Resolve the mapping until finished. (drivers are | |
1396 | * still free to implement/resolve their own stacking | |
1397 | * by explicitly returning 0) | |
1398 | * | |
1399 | * NOTE: we don't repeat the blk_size check for each new device. | |
1400 | * Stacking drivers are expected to know what they are doing. | |
1401 | */ | |
5ddfe969 | 1402 | old_sector = -1; |
2056a782 | 1403 | old_dev = 0; |
1da177e4 LT |
1404 | do { |
1405 | char b[BDEVNAME_SIZE]; | |
1406 | ||
1407 | q = bdev_get_queue(bio->bi_bdev); | |
a7384677 | 1408 | if (unlikely(!q)) { |
1da177e4 LT |
1409 | printk(KERN_ERR |
1410 | "generic_make_request: Trying to access " | |
1411 | "nonexistent block-device %s (%Lu)\n", | |
1412 | bdevname(bio->bi_bdev, b), | |
1413 | (long long) bio->bi_sector); | |
a7384677 | 1414 | goto end_io; |
1da177e4 LT |
1415 | } |
1416 | ||
4fa253f3 | 1417 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1418 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1419 | bdevname(bio->bi_bdev, b), |
1420 | bio_sectors(bio), | |
1421 | q->max_hw_sectors); | |
1422 | goto end_io; | |
1423 | } | |
1424 | ||
fde6ad22 | 1425 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1426 | goto end_io; |
1427 | ||
c17bb495 AM |
1428 | if (should_fail_request(bio)) |
1429 | goto end_io; | |
1430 | ||
1da177e4 LT |
1431 | /* |
1432 | * If this device has partitions, remap block n | |
1433 | * of partition p to block n+start(p) of the disk. | |
1434 | */ | |
1435 | blk_partition_remap(bio); | |
1436 | ||
7ba1ba12 MP |
1437 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1438 | goto end_io; | |
1439 | ||
5ddfe969 | 1440 | if (old_sector != -1) |
22a7c31a | 1441 | trace_block_remap(q, bio, old_dev, old_sector); |
2056a782 | 1442 | |
5f3ea37c | 1443 | trace_block_bio_queue(q, bio); |
2056a782 | 1444 | |
5ddfe969 | 1445 | old_sector = bio->bi_sector; |
2056a782 JA |
1446 | old_dev = bio->bi_bdev->bd_dev; |
1447 | ||
c07e2b41 JA |
1448 | if (bio_check_eod(bio, nr_sectors)) |
1449 | goto end_io; | |
a7384677 TH |
1450 | |
1451 | if (bio_discard(bio) && !q->prepare_discard_fn) { | |
51fd77bd JA |
1452 | err = -EOPNOTSUPP; |
1453 | goto end_io; | |
1454 | } | |
cec0707e JA |
1455 | if (bio_barrier(bio) && bio_has_data(bio) && |
1456 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1457 | err = -EOPNOTSUPP; | |
1458 | goto end_io; | |
1459 | } | |
5ddfe969 | 1460 | |
1da177e4 LT |
1461 | ret = q->make_request_fn(q, bio); |
1462 | } while (ret); | |
a7384677 TH |
1463 | |
1464 | return; | |
1465 | ||
1466 | end_io: | |
1467 | bio_endio(bio, err); | |
1da177e4 LT |
1468 | } |
1469 | ||
d89d8796 NB |
1470 | /* |
1471 | * We only want one ->make_request_fn to be active at a time, | |
1472 | * else stack usage with stacked devices could be a problem. | |
1473 | * So use current->bio_{list,tail} to keep a list of requests | |
1474 | * submited by a make_request_fn function. | |
1475 | * current->bio_tail is also used as a flag to say if | |
1476 | * generic_make_request is currently active in this task or not. | |
1477 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1478 | * then a make_request is active, and new requests should be added | |
1479 | * at the tail | |
1480 | */ | |
1481 | void generic_make_request(struct bio *bio) | |
1482 | { | |
1483 | if (current->bio_tail) { | |
1484 | /* make_request is active */ | |
1485 | *(current->bio_tail) = bio; | |
1486 | bio->bi_next = NULL; | |
1487 | current->bio_tail = &bio->bi_next; | |
1488 | return; | |
1489 | } | |
1490 | /* following loop may be a bit non-obvious, and so deserves some | |
1491 | * explanation. | |
1492 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1493 | * ensure that) so we have a list with a single bio. | |
1494 | * We pretend that we have just taken it off a longer list, so | |
1495 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1496 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1497 | * added. __generic_make_request may indeed add some more bios | |
1498 | * through a recursive call to generic_make_request. If it | |
1499 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1500 | * from the top. In this case we really did just take the bio | |
1501 | * of the top of the list (no pretending) and so fixup bio_list and | |
1502 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1503 | * | |
1504 | * The loop was structured like this to make only one call to | |
1505 | * __generic_make_request (which is important as it is large and | |
1506 | * inlined) and to keep the structure simple. | |
1507 | */ | |
1508 | BUG_ON(bio->bi_next); | |
1509 | do { | |
1510 | current->bio_list = bio->bi_next; | |
1511 | if (bio->bi_next == NULL) | |
1512 | current->bio_tail = ¤t->bio_list; | |
1513 | else | |
1514 | bio->bi_next = NULL; | |
1515 | __generic_make_request(bio); | |
1516 | bio = current->bio_list; | |
1517 | } while (bio); | |
1518 | current->bio_tail = NULL; /* deactivate */ | |
1519 | } | |
1da177e4 LT |
1520 | EXPORT_SYMBOL(generic_make_request); |
1521 | ||
1522 | /** | |
710027a4 | 1523 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1524 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1525 | * @bio: The &struct bio which describes the I/O | |
1526 | * | |
1527 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1528 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1529 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1530 | * |
1531 | */ | |
1532 | void submit_bio(int rw, struct bio *bio) | |
1533 | { | |
1534 | int count = bio_sectors(bio); | |
1535 | ||
22e2c507 | 1536 | bio->bi_rw |= rw; |
1da177e4 | 1537 | |
bf2de6f5 JA |
1538 | /* |
1539 | * If it's a regular read/write or a barrier with data attached, | |
1540 | * go through the normal accounting stuff before submission. | |
1541 | */ | |
a9c701e5 | 1542 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1543 | if (rw & WRITE) { |
1544 | count_vm_events(PGPGOUT, count); | |
1545 | } else { | |
1546 | task_io_account_read(bio->bi_size); | |
1547 | count_vm_events(PGPGIN, count); | |
1548 | } | |
1549 | ||
1550 | if (unlikely(block_dump)) { | |
1551 | char b[BDEVNAME_SIZE]; | |
1552 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1553 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1554 | (rw & WRITE) ? "WRITE" : "READ", |
1555 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1556 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1557 | } |
1da177e4 LT |
1558 | } |
1559 | ||
1560 | generic_make_request(bio); | |
1561 | } | |
1da177e4 LT |
1562 | EXPORT_SYMBOL(submit_bio); |
1563 | ||
82124d60 KU |
1564 | /** |
1565 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1566 | * @q: the queue | |
1567 | * @rq: the request being checked | |
1568 | * | |
1569 | * Description: | |
1570 | * @rq may have been made based on weaker limitations of upper-level queues | |
1571 | * in request stacking drivers, and it may violate the limitation of @q. | |
1572 | * Since the block layer and the underlying device driver trust @rq | |
1573 | * after it is inserted to @q, it should be checked against @q before | |
1574 | * the insertion using this generic function. | |
1575 | * | |
1576 | * This function should also be useful for request stacking drivers | |
1577 | * in some cases below, so export this fuction. | |
1578 | * Request stacking drivers like request-based dm may change the queue | |
1579 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1580 | * Such request stacking drivers should check those requests agaist | |
1581 | * the new queue limits again when they dispatch those requests, | |
1582 | * although such checkings are also done against the old queue limits | |
1583 | * when submitting requests. | |
1584 | */ | |
1585 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1586 | { | |
1587 | if (rq->nr_sectors > q->max_sectors || | |
1588 | rq->data_len > q->max_hw_sectors << 9) { | |
1589 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1590 | return -EIO; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * queue's settings related to segment counting like q->bounce_pfn | |
1595 | * may differ from that of other stacking queues. | |
1596 | * Recalculate it to check the request correctly on this queue's | |
1597 | * limitation. | |
1598 | */ | |
1599 | blk_recalc_rq_segments(rq); | |
1600 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1601 | rq->nr_phys_segments > q->max_hw_segments) { | |
1602 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1603 | return -EIO; | |
1604 | } | |
1605 | ||
1606 | return 0; | |
1607 | } | |
1608 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1609 | ||
1610 | /** | |
1611 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1612 | * @q: the queue to submit the request | |
1613 | * @rq: the request being queued | |
1614 | */ | |
1615 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1616 | { | |
1617 | unsigned long flags; | |
1618 | ||
1619 | if (blk_rq_check_limits(q, rq)) | |
1620 | return -EIO; | |
1621 | ||
1622 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1623 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1624 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1625 | return -EIO; | |
1626 | #endif | |
1627 | ||
1628 | spin_lock_irqsave(q->queue_lock, flags); | |
1629 | ||
1630 | /* | |
1631 | * Submitting request must be dequeued before calling this function | |
1632 | * because it will be linked to another request_queue | |
1633 | */ | |
1634 | BUG_ON(blk_queued_rq(rq)); | |
1635 | ||
1636 | drive_stat_acct(rq, 1); | |
1637 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1638 | ||
1639 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1640 | ||
1641 | return 0; | |
1642 | } | |
1643 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1644 | ||
53a08807 TH |
1645 | /** |
1646 | * blkdev_dequeue_request - dequeue request and start timeout timer | |
1647 | * @req: request to dequeue | |
1648 | * | |
1649 | * Dequeue @req and start timeout timer on it. This hands off the | |
1650 | * request to the driver. | |
1651 | * | |
1652 | * Block internal functions which don't want to start timer should | |
1653 | * call elv_dequeue_request(). | |
1654 | */ | |
1655 | void blkdev_dequeue_request(struct request *req) | |
1656 | { | |
1657 | elv_dequeue_request(req->q, req); | |
1658 | ||
1659 | /* | |
1660 | * We are now handing the request to the hardware, add the | |
1661 | * timeout handler. | |
1662 | */ | |
1663 | blk_add_timer(req); | |
1664 | } | |
1665 | EXPORT_SYMBOL(blkdev_dequeue_request); | |
1666 | ||
bc58ba94 JA |
1667 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1668 | { | |
26308eab | 1669 | if (!blk_do_io_stat(req)) |
bc58ba94 JA |
1670 | return; |
1671 | ||
1672 | if (blk_fs_request(req)) { | |
1673 | const int rw = rq_data_dir(req); | |
1674 | struct hd_struct *part; | |
1675 | int cpu; | |
1676 | ||
1677 | cpu = part_stat_lock(); | |
1678 | part = disk_map_sector_rcu(req->rq_disk, req->sector); | |
1679 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
1680 | part_stat_unlock(); | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | static void blk_account_io_done(struct request *req) | |
1685 | { | |
26308eab | 1686 | if (!blk_do_io_stat(req)) |
bc58ba94 JA |
1687 | return; |
1688 | ||
1689 | /* | |
1690 | * Account IO completion. bar_rq isn't accounted as a normal | |
1691 | * IO on queueing nor completion. Accounting the containing | |
1692 | * request is enough. | |
1693 | */ | |
1694 | if (blk_fs_request(req) && req != &req->q->bar_rq) { | |
1695 | unsigned long duration = jiffies - req->start_time; | |
1696 | const int rw = rq_data_dir(req); | |
1697 | struct hd_struct *part; | |
1698 | int cpu; | |
1699 | ||
1700 | cpu = part_stat_lock(); | |
26308eab | 1701 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
bc58ba94 JA |
1702 | |
1703 | part_stat_inc(cpu, part, ios[rw]); | |
1704 | part_stat_add(cpu, part, ticks[rw], duration); | |
1705 | part_round_stats(cpu, part); | |
1706 | part_dec_in_flight(part); | |
1707 | ||
1708 | part_stat_unlock(); | |
1709 | } | |
1710 | } | |
1711 | ||
3bcddeac KU |
1712 | /** |
1713 | * __end_that_request_first - end I/O on a request | |
1714 | * @req: the request being processed | |
710027a4 | 1715 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
1716 | * @nr_bytes: number of bytes to complete |
1717 | * | |
1718 | * Description: | |
1719 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1720 | * for the next range of segments (if any) in the cluster. | |
1721 | * | |
1722 | * Return: | |
710027a4 RD |
1723 | * %0 - we are done with this request, call end_that_request_last() |
1724 | * %1 - still buffers pending for this request | |
3bcddeac | 1725 | **/ |
5450d3e1 | 1726 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1727 | int nr_bytes) |
1728 | { | |
5450d3e1 | 1729 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1730 | struct bio *bio; |
1731 | ||
5f3ea37c | 1732 | trace_block_rq_complete(req->q, req); |
2056a782 | 1733 | |
1da177e4 | 1734 | /* |
6f41469c TH |
1735 | * For fs requests, rq is just carrier of independent bio's |
1736 | * and each partial completion should be handled separately. | |
1737 | * Reset per-request error on each partial completion. | |
1738 | * | |
1739 | * TODO: tj: This is too subtle. It would be better to let | |
1740 | * low level drivers do what they see fit. | |
1da177e4 | 1741 | */ |
6f41469c | 1742 | if (blk_fs_request(req)) |
1da177e4 LT |
1743 | req->errors = 0; |
1744 | ||
6728cb0e JA |
1745 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1746 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1747 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1748 | (unsigned long long)req->sector); | |
1749 | } | |
1750 | ||
bc58ba94 | 1751 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1752 | |
1da177e4 LT |
1753 | total_bytes = bio_nbytes = 0; |
1754 | while ((bio = req->bio) != NULL) { | |
1755 | int nbytes; | |
1756 | ||
1757 | if (nr_bytes >= bio->bi_size) { | |
1758 | req->bio = bio->bi_next; | |
1759 | nbytes = bio->bi_size; | |
5bb23a68 | 1760 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1761 | next_idx = 0; |
1762 | bio_nbytes = 0; | |
1763 | } else { | |
1764 | int idx = bio->bi_idx + next_idx; | |
1765 | ||
af498d7f | 1766 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 1767 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 1768 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 1769 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
1770 | break; |
1771 | } | |
1772 | ||
1773 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1774 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1775 | ||
1776 | /* | |
1777 | * not a complete bvec done | |
1778 | */ | |
1779 | if (unlikely(nbytes > nr_bytes)) { | |
1780 | bio_nbytes += nr_bytes; | |
1781 | total_bytes += nr_bytes; | |
1782 | break; | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * advance to the next vector | |
1787 | */ | |
1788 | next_idx++; | |
1789 | bio_nbytes += nbytes; | |
1790 | } | |
1791 | ||
1792 | total_bytes += nbytes; | |
1793 | nr_bytes -= nbytes; | |
1794 | ||
6728cb0e JA |
1795 | bio = req->bio; |
1796 | if (bio) { | |
1da177e4 LT |
1797 | /* |
1798 | * end more in this run, or just return 'not-done' | |
1799 | */ | |
1800 | if (unlikely(nr_bytes <= 0)) | |
1801 | break; | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | /* | |
1806 | * completely done | |
1807 | */ | |
1808 | if (!req->bio) | |
1809 | return 0; | |
1810 | ||
1811 | /* | |
1812 | * if the request wasn't completed, update state | |
1813 | */ | |
1814 | if (bio_nbytes) { | |
5bb23a68 | 1815 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1816 | bio->bi_idx += next_idx; |
1817 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1818 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1819 | } | |
1820 | ||
1821 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1822 | blk_recalc_rq_segments(req); | |
1823 | return 1; | |
1824 | } | |
1825 | ||
1da177e4 LT |
1826 | /* |
1827 | * queue lock must be held | |
1828 | */ | |
5450d3e1 | 1829 | static void end_that_request_last(struct request *req, int error) |
1da177e4 | 1830 | { |
b8286239 KU |
1831 | if (blk_rq_tagged(req)) |
1832 | blk_queue_end_tag(req->q, req); | |
1833 | ||
1834 | if (blk_queued_rq(req)) | |
53a08807 | 1835 | elv_dequeue_request(req->q, req); |
1da177e4 LT |
1836 | |
1837 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1838 | laptop_io_completion(); | |
1839 | ||
e78042e5 MA |
1840 | blk_delete_timer(req); |
1841 | ||
bc58ba94 | 1842 | blk_account_io_done(req); |
b8286239 | 1843 | |
1da177e4 | 1844 | if (req->end_io) |
8ffdc655 | 1845 | req->end_io(req, error); |
b8286239 KU |
1846 | else { |
1847 | if (blk_bidi_rq(req)) | |
1848 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1849 | ||
1da177e4 | 1850 | __blk_put_request(req->q, req); |
b8286239 | 1851 | } |
1da177e4 LT |
1852 | } |
1853 | ||
3b11313a KU |
1854 | /** |
1855 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1856 | * @rq: the request being processed |
3b11313a KU |
1857 | **/ |
1858 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1859 | { |
1860 | if (blk_fs_request(rq)) | |
1861 | return rq->hard_nr_sectors << 9; | |
1862 | ||
1863 | return rq->data_len; | |
1864 | } | |
3b11313a KU |
1865 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1866 | ||
1867 | /** | |
1868 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1869 | * @rq: the request being processed |
3b11313a KU |
1870 | **/ |
1871 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1872 | { | |
1873 | if (blk_fs_request(rq)) | |
1874 | return rq->current_nr_sectors << 9; | |
1875 | ||
1876 | if (rq->bio) | |
1877 | return rq->bio->bi_size; | |
1878 | ||
1879 | return rq->data_len; | |
1880 | } | |
1881 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 | 1882 | |
a0cd1285 JA |
1883 | /** |
1884 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1885 | * @req: the request being processed |
710027a4 | 1886 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1887 | * |
1888 | * Description: | |
1889 | * Ends I/O on the current segment of a request. If that is the only | |
1890 | * remaining segment, the request is also completed and freed. | |
1891 | * | |
710027a4 RD |
1892 | * This is a remnant of how older block drivers handled I/O completions. |
1893 | * Modern drivers typically end I/O on the full request in one go, unless | |
a0cd1285 JA |
1894 | * they have a residual value to account for. For that case this function |
1895 | * isn't really useful, unless the residual just happens to be the | |
1896 | * full current segment. In other words, don't use this function in new | |
d00e29fd | 1897 | * code. Use blk_end_request() or __blk_end_request() to end a request. |
a0cd1285 JA |
1898 | **/ |
1899 | void end_request(struct request *req, int uptodate) | |
1900 | { | |
d00e29fd KU |
1901 | int error = 0; |
1902 | ||
1903 | if (uptodate <= 0) | |
1904 | error = uptodate ? uptodate : -EIO; | |
1905 | ||
1906 | __blk_end_request(req, error, req->hard_cur_sectors << 9); | |
a0cd1285 | 1907 | } |
1da177e4 LT |
1908 | EXPORT_SYMBOL(end_request); |
1909 | ||
32fab448 KU |
1910 | static int end_that_request_data(struct request *rq, int error, |
1911 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1912 | { | |
1913 | if (rq->bio) { | |
1914 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1915 | return 1; | |
1916 | ||
1917 | /* Bidi request must be completed as a whole */ | |
1918 | if (blk_bidi_rq(rq) && | |
1919 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1920 | return 1; | |
1921 | } | |
1922 | ||
1923 | return 0; | |
1924 | } | |
1925 | ||
336cdb40 | 1926 | /** |
e19a3ab0 KU |
1927 | * blk_end_io - Generic end_io function to complete a request. |
1928 | * @rq: the request being processed | |
710027a4 | 1929 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1930 | * @nr_bytes: number of bytes to complete @rq |
1931 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1932 | * @drv_callback: function called between completion of bios in the request |
1933 | * and completion of the request. | |
710027a4 | 1934 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1935 | * completion of the request. |
336cdb40 KU |
1936 | * |
1937 | * Description: | |
e3a04fe3 | 1938 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1939 | * If @rq has leftover, sets it up for the next range of segments. |
1940 | * | |
1941 | * Return: | |
710027a4 RD |
1942 | * %0 - we are done with this request |
1943 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1944 | **/ |
22b13210 JA |
1945 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1946 | unsigned int bidi_bytes, | |
1947 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1948 | { |
1949 | struct request_queue *q = rq->q; | |
1950 | unsigned long flags = 0UL; | |
336cdb40 | 1951 | |
32fab448 KU |
1952 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
1953 | return 1; | |
336cdb40 | 1954 | |
e19a3ab0 KU |
1955 | /* Special feature for tricky drivers */ |
1956 | if (drv_callback && drv_callback(rq)) | |
1957 | return 1; | |
1958 | ||
336cdb40 KU |
1959 | add_disk_randomness(rq->rq_disk); |
1960 | ||
1961 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1962 | end_that_request_last(rq, error); |
336cdb40 KU |
1963 | spin_unlock_irqrestore(q->queue_lock, flags); |
1964 | ||
1965 | return 0; | |
1966 | } | |
e19a3ab0 KU |
1967 | |
1968 | /** | |
1969 | * blk_end_request - Helper function for drivers to complete the request. | |
1970 | * @rq: the request being processed | |
710027a4 | 1971 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
1972 | * @nr_bytes: number of bytes to complete |
1973 | * | |
1974 | * Description: | |
1975 | * Ends I/O on a number of bytes attached to @rq. | |
1976 | * If @rq has leftover, sets it up for the next range of segments. | |
1977 | * | |
1978 | * Return: | |
710027a4 RD |
1979 | * %0 - we are done with this request |
1980 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1981 | **/ |
22b13210 | 1982 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1983 | { |
e3a04fe3 | 1984 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1985 | } |
336cdb40 KU |
1986 | EXPORT_SYMBOL_GPL(blk_end_request); |
1987 | ||
1988 | /** | |
1989 | * __blk_end_request - Helper function for drivers to complete the request. | |
1990 | * @rq: the request being processed | |
710027a4 | 1991 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
1992 | * @nr_bytes: number of bytes to complete |
1993 | * | |
1994 | * Description: | |
1995 | * Must be called with queue lock held unlike blk_end_request(). | |
1996 | * | |
1997 | * Return: | |
710027a4 RD |
1998 | * %0 - we are done with this request |
1999 | * %1 - still buffers pending for this request | |
336cdb40 | 2000 | **/ |
22b13210 | 2001 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 2002 | { |
60540161 | 2003 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
051cc395 | 2004 | return 1; |
336cdb40 KU |
2005 | |
2006 | add_disk_randomness(rq->rq_disk); | |
2007 | ||
b8286239 | 2008 | end_that_request_last(rq, error); |
336cdb40 KU |
2009 | |
2010 | return 0; | |
2011 | } | |
2012 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
2013 | ||
e3a04fe3 KU |
2014 | /** |
2015 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
2016 | * @rq: the bidi request being processed | |
710027a4 | 2017 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2018 | * @nr_bytes: number of bytes to complete @rq |
2019 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
2020 | * | |
2021 | * Description: | |
2022 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
2023 | * | |
2024 | * Return: | |
710027a4 RD |
2025 | * %0 - we are done with this request |
2026 | * %1 - still buffers pending for this request | |
e3a04fe3 | 2027 | **/ |
22b13210 JA |
2028 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
2029 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
2030 | { |
2031 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
2032 | } | |
2033 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2034 | ||
32fab448 KU |
2035 | /** |
2036 | * blk_update_request - Special helper function for request stacking drivers | |
2037 | * @rq: the request being processed | |
2038 | * @error: %0 for success, < %0 for error | |
2039 | * @nr_bytes: number of bytes to complete @rq | |
2040 | * | |
2041 | * Description: | |
2042 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2043 | * the request structure even if @rq doesn't have leftover. | |
2044 | * If @rq has leftover, sets it up for the next range of segments. | |
2045 | * | |
2046 | * This special helper function is only for request stacking drivers | |
2047 | * (e.g. request-based dm) so that they can handle partial completion. | |
2048 | * Actual device drivers should use blk_end_request instead. | |
2049 | */ | |
2050 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2051 | { | |
2052 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2053 | /* | |
2054 | * These members are not updated in end_that_request_data() | |
2055 | * when all bios are completed. | |
2056 | * Update them so that the request stacking driver can find | |
2057 | * how many bytes remain in the request later. | |
2058 | */ | |
2059 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2060 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2061 | } | |
2062 | } | |
2063 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2064 | ||
e19a3ab0 KU |
2065 | /** |
2066 | * blk_end_request_callback - Special helper function for tricky drivers | |
2067 | * @rq: the request being processed | |
710027a4 | 2068 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2069 | * @nr_bytes: number of bytes to complete |
2070 | * @drv_callback: function called between completion of bios in the request | |
2071 | * and completion of the request. | |
710027a4 | 2072 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
2073 | * completion of the request. |
2074 | * | |
2075 | * Description: | |
2076 | * Ends I/O on a number of bytes attached to @rq. | |
2077 | * If @rq has leftover, sets it up for the next range of segments. | |
2078 | * | |
2079 | * This special helper function is used only for existing tricky drivers. | |
2080 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2081 | * This interface will be removed when such drivers are rewritten. | |
2082 | * Don't use this interface in other places anymore. | |
2083 | * | |
2084 | * Return: | |
710027a4 RD |
2085 | * %0 - we are done with this request |
2086 | * %1 - this request is not freed yet. | |
2087 | * this request still has pending buffers or | |
2088 | * the driver doesn't want to finish this request yet. | |
e19a3ab0 | 2089 | **/ |
22b13210 JA |
2090 | int blk_end_request_callback(struct request *rq, int error, |
2091 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2092 | int (drv_callback)(struct request *)) |
2093 | { | |
e3a04fe3 | 2094 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2095 | } |
2096 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2097 | ||
86db1e29 JA |
2098 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2099 | struct bio *bio) | |
1da177e4 | 2100 | { |
d628eaef DW |
2101 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2102 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2103 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2104 | |
fb2dce86 DW |
2105 | if (bio_has_data(bio)) { |
2106 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2107 | rq->buffer = bio_data(bio); |
2108 | } | |
1da177e4 LT |
2109 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2110 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2111 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2112 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2113 | |
2114 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2115 | |
66846572 N |
2116 | if (bio->bi_bdev) |
2117 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2118 | } | |
1da177e4 | 2119 | |
ef9e3fac KU |
2120 | /** |
2121 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2122 | * @q : the queue of the device being checked | |
2123 | * | |
2124 | * Description: | |
2125 | * Check if underlying low-level drivers of a device are busy. | |
2126 | * If the drivers want to export their busy state, they must set own | |
2127 | * exporting function using blk_queue_lld_busy() first. | |
2128 | * | |
2129 | * Basically, this function is used only by request stacking drivers | |
2130 | * to stop dispatching requests to underlying devices when underlying | |
2131 | * devices are busy. This behavior helps more I/O merging on the queue | |
2132 | * of the request stacking driver and prevents I/O throughput regression | |
2133 | * on burst I/O load. | |
2134 | * | |
2135 | * Return: | |
2136 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2137 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2138 | */ | |
2139 | int blk_lld_busy(struct request_queue *q) | |
2140 | { | |
2141 | if (q->lld_busy_fn) | |
2142 | return q->lld_busy_fn(q); | |
2143 | ||
2144 | return 0; | |
2145 | } | |
2146 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2147 | ||
18887ad9 | 2148 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2149 | { |
2150 | return queue_work(kblockd_workqueue, work); | |
2151 | } | |
1da177e4 LT |
2152 | EXPORT_SYMBOL(kblockd_schedule_work); |
2153 | ||
1da177e4 LT |
2154 | int __init blk_dev_init(void) |
2155 | { | |
2156 | kblockd_workqueue = create_workqueue("kblockd"); | |
2157 | if (!kblockd_workqueue) | |
2158 | panic("Failed to create kblockd\n"); | |
2159 | ||
2160 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2161 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2162 | |
8324aa91 | 2163 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2164 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2165 | |
d38ecf93 | 2166 | return 0; |
1da177e4 | 2167 | } |
1da177e4 | 2168 |