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> |
ff856bad JA |
29 | #include <linux/interrupt.h> |
30 | #include <linux/cpu.h> | |
2056a782 | 31 | #include <linux/blktrace_api.h> |
c17bb495 | 32 | #include <linux/fault-inject.h> |
1da177e4 | 33 | |
8324aa91 JA |
34 | #include "blk.h" |
35 | ||
165125e1 | 36 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
37 | |
38 | /* | |
39 | * For the allocated request tables | |
40 | */ | |
8324aa91 | 41 | struct kmem_cache *request_cachep; |
1da177e4 LT |
42 | |
43 | /* | |
44 | * For queue allocation | |
45 | */ | |
6728cb0e | 46 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 47 | |
1da177e4 LT |
48 | /* |
49 | * Controlling structure to kblockd | |
50 | */ | |
ff856bad | 51 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 52 | |
ff856bad JA |
53 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); |
54 | ||
26b8256e JA |
55 | static void drive_stat_acct(struct request *rq, int new_io) |
56 | { | |
57 | int rw = rq_data_dir(rq); | |
58 | ||
59 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
60 | return; | |
61 | ||
62 | if (!new_io) { | |
63 | __disk_stat_inc(rq->rq_disk, merges[rw]); | |
64 | } else { | |
65 | disk_round_stats(rq->rq_disk); | |
66 | rq->rq_disk->in_flight++; | |
67 | } | |
68 | } | |
69 | ||
8324aa91 | 70 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
71 | { |
72 | int nr; | |
73 | ||
74 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
75 | if (nr > q->nr_requests) | |
76 | nr = q->nr_requests; | |
77 | q->nr_congestion_on = nr; | |
78 | ||
79 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
80 | if (nr < 1) | |
81 | nr = 1; | |
82 | q->nr_congestion_off = nr; | |
83 | } | |
84 | ||
1da177e4 LT |
85 | /** |
86 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
87 | * @bdev: device | |
88 | * | |
89 | * Locates the passed device's request queue and returns the address of its | |
90 | * backing_dev_info | |
91 | * | |
92 | * Will return NULL if the request queue cannot be located. | |
93 | */ | |
94 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
95 | { | |
96 | struct backing_dev_info *ret = NULL; | |
165125e1 | 97 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
98 | |
99 | if (q) | |
100 | ret = &q->backing_dev_info; | |
101 | return ret; | |
102 | } | |
1da177e4 LT |
103 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
104 | ||
86db1e29 | 105 | void rq_init(struct request_queue *q, struct request *rq) |
1da177e4 LT |
106 | { |
107 | INIT_LIST_HEAD(&rq->queuelist); | |
ff856bad | 108 | INIT_LIST_HEAD(&rq->donelist); |
1da177e4 LT |
109 | |
110 | rq->errors = 0; | |
1da177e4 | 111 | rq->bio = rq->biotail = NULL; |
2e662b65 JA |
112 | INIT_HLIST_NODE(&rq->hash); |
113 | RB_CLEAR_NODE(&rq->rb_node); | |
22e2c507 | 114 | rq->ioprio = 0; |
1da177e4 LT |
115 | rq->buffer = NULL; |
116 | rq->ref_count = 1; | |
117 | rq->q = q; | |
1da177e4 LT |
118 | rq->special = NULL; |
119 | rq->data_len = 0; | |
120 | rq->data = NULL; | |
df46b9a4 | 121 | rq->nr_phys_segments = 0; |
1da177e4 LT |
122 | rq->sense = NULL; |
123 | rq->end_io = NULL; | |
124 | rq->end_io_data = NULL; | |
ff856bad | 125 | rq->completion_data = NULL; |
abae1fde | 126 | rq->next_rq = NULL; |
1da177e4 LT |
127 | } |
128 | ||
5bb23a68 N |
129 | static void req_bio_endio(struct request *rq, struct bio *bio, |
130 | unsigned int nbytes, int error) | |
1da177e4 | 131 | { |
165125e1 | 132 | struct request_queue *q = rq->q; |
797e7dbb | 133 | |
5bb23a68 N |
134 | if (&q->bar_rq != rq) { |
135 | if (error) | |
136 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
137 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
138 | error = -EIO; | |
797e7dbb | 139 | |
5bb23a68 | 140 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 141 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
5bb23a68 N |
142 | __FUNCTION__, nbytes, bio->bi_size); |
143 | nbytes = bio->bi_size; | |
144 | } | |
797e7dbb | 145 | |
5bb23a68 N |
146 | bio->bi_size -= nbytes; |
147 | bio->bi_sector += (nbytes >> 9); | |
148 | if (bio->bi_size == 0) | |
6712ecf8 | 149 | bio_endio(bio, error); |
5bb23a68 N |
150 | } else { |
151 | ||
152 | /* | |
153 | * Okay, this is the barrier request in progress, just | |
154 | * record the error; | |
155 | */ | |
156 | if (error && !q->orderr) | |
157 | q->orderr = error; | |
158 | } | |
1da177e4 | 159 | } |
1da177e4 | 160 | |
1da177e4 LT |
161 | void blk_dump_rq_flags(struct request *rq, char *msg) |
162 | { | |
163 | int bit; | |
164 | ||
6728cb0e | 165 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
166 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
167 | rq->cmd_flags); | |
1da177e4 | 168 | |
6728cb0e JA |
169 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
170 | (unsigned long long)rq->sector, | |
171 | rq->nr_sectors, | |
172 | rq->current_nr_sectors); | |
173 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
174 | rq->bio, rq->biotail, | |
175 | rq->buffer, rq->data, | |
176 | rq->data_len); | |
1da177e4 | 177 | |
4aff5e23 | 178 | if (blk_pc_request(rq)) { |
6728cb0e | 179 | printk(KERN_INFO " cdb: "); |
1da177e4 LT |
180 | for (bit = 0; bit < sizeof(rq->cmd); bit++) |
181 | printk("%02x ", rq->cmd[bit]); | |
182 | printk("\n"); | |
183 | } | |
184 | } | |
1da177e4 LT |
185 | EXPORT_SYMBOL(blk_dump_rq_flags); |
186 | ||
1da177e4 LT |
187 | /* |
188 | * "plug" the device if there are no outstanding requests: this will | |
189 | * force the transfer to start only after we have put all the requests | |
190 | * on the list. | |
191 | * | |
192 | * This is called with interrupts off and no requests on the queue and | |
193 | * with the queue lock held. | |
194 | */ | |
165125e1 | 195 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
196 | { |
197 | WARN_ON(!irqs_disabled()); | |
198 | ||
199 | /* | |
200 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
201 | * which will restart the queueing | |
202 | */ | |
7daac490 | 203 | if (blk_queue_stopped(q)) |
1da177e4 LT |
204 | return; |
205 | ||
2056a782 | 206 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 207 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
208 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
209 | } | |
1da177e4 | 210 | } |
1da177e4 LT |
211 | EXPORT_SYMBOL(blk_plug_device); |
212 | ||
213 | /* | |
214 | * remove the queue from the plugged list, if present. called with | |
215 | * queue lock held and interrupts disabled. | |
216 | */ | |
165125e1 | 217 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
218 | { |
219 | WARN_ON(!irqs_disabled()); | |
220 | ||
221 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
222 | return 0; | |
223 | ||
224 | del_timer(&q->unplug_timer); | |
225 | return 1; | |
226 | } | |
1da177e4 LT |
227 | EXPORT_SYMBOL(blk_remove_plug); |
228 | ||
229 | /* | |
230 | * remove the plug and let it rip.. | |
231 | */ | |
165125e1 | 232 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 233 | { |
7daac490 | 234 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
235 | return; |
236 | ||
237 | if (!blk_remove_plug(q)) | |
238 | return; | |
239 | ||
22e2c507 | 240 | q->request_fn(q); |
1da177e4 LT |
241 | } |
242 | EXPORT_SYMBOL(__generic_unplug_device); | |
243 | ||
244 | /** | |
245 | * generic_unplug_device - fire a request queue | |
165125e1 | 246 | * @q: The &struct request_queue in question |
1da177e4 LT |
247 | * |
248 | * Description: | |
249 | * Linux uses plugging to build bigger requests queues before letting | |
250 | * the device have at them. If a queue is plugged, the I/O scheduler | |
251 | * is still adding and merging requests on the queue. Once the queue | |
252 | * gets unplugged, the request_fn defined for the queue is invoked and | |
253 | * transfers started. | |
254 | **/ | |
165125e1 | 255 | void generic_unplug_device(struct request_queue *q) |
1da177e4 LT |
256 | { |
257 | spin_lock_irq(q->queue_lock); | |
258 | __generic_unplug_device(q); | |
259 | spin_unlock_irq(q->queue_lock); | |
260 | } | |
261 | EXPORT_SYMBOL(generic_unplug_device); | |
262 | ||
263 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
264 | struct page *page) | |
265 | { | |
165125e1 | 266 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 267 | |
2ad8b1ef | 268 | blk_unplug(q); |
1da177e4 LT |
269 | } |
270 | ||
86db1e29 | 271 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 272 | { |
165125e1 JA |
273 | struct request_queue *q = |
274 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 275 | |
2056a782 JA |
276 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
277 | q->rq.count[READ] + q->rq.count[WRITE]); | |
278 | ||
1da177e4 LT |
279 | q->unplug_fn(q); |
280 | } | |
281 | ||
86db1e29 | 282 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 283 | { |
165125e1 | 284 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 285 | |
2056a782 JA |
286 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
287 | q->rq.count[READ] + q->rq.count[WRITE]); | |
288 | ||
1da177e4 LT |
289 | kblockd_schedule_work(&q->unplug_work); |
290 | } | |
291 | ||
2ad8b1ef AB |
292 | void blk_unplug(struct request_queue *q) |
293 | { | |
294 | /* | |
295 | * devices don't necessarily have an ->unplug_fn defined | |
296 | */ | |
297 | if (q->unplug_fn) { | |
298 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
299 | q->rq.count[READ] + q->rq.count[WRITE]); | |
300 | ||
301 | q->unplug_fn(q); | |
302 | } | |
303 | } | |
304 | EXPORT_SYMBOL(blk_unplug); | |
305 | ||
1da177e4 LT |
306 | /** |
307 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 308 | * @q: The &struct request_queue in question |
1da177e4 LT |
309 | * |
310 | * Description: | |
311 | * blk_start_queue() will clear the stop flag on the queue, and call | |
312 | * the request_fn for the queue if it was in a stopped state when | |
313 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
314 | **/ | |
165125e1 | 315 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 316 | { |
a038e253 PBG |
317 | WARN_ON(!irqs_disabled()); |
318 | ||
1da177e4 LT |
319 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
320 | ||
321 | /* | |
322 | * one level of recursion is ok and is much faster than kicking | |
323 | * the unplug handling | |
324 | */ | |
325 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
326 | q->request_fn(q); | |
327 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
328 | } else { | |
329 | blk_plug_device(q); | |
330 | kblockd_schedule_work(&q->unplug_work); | |
331 | } | |
332 | } | |
1da177e4 LT |
333 | EXPORT_SYMBOL(blk_start_queue); |
334 | ||
335 | /** | |
336 | * blk_stop_queue - stop a queue | |
165125e1 | 337 | * @q: The &struct request_queue in question |
1da177e4 LT |
338 | * |
339 | * Description: | |
340 | * The Linux block layer assumes that a block driver will consume all | |
341 | * entries on the request queue when the request_fn strategy is called. | |
342 | * Often this will not happen, because of hardware limitations (queue | |
343 | * depth settings). If a device driver gets a 'queue full' response, | |
344 | * or if it simply chooses not to queue more I/O at one point, it can | |
345 | * call this function to prevent the request_fn from being called until | |
346 | * the driver has signalled it's ready to go again. This happens by calling | |
347 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
348 | **/ | |
165125e1 | 349 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
350 | { |
351 | blk_remove_plug(q); | |
352 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
353 | } | |
354 | EXPORT_SYMBOL(blk_stop_queue); | |
355 | ||
356 | /** | |
357 | * blk_sync_queue - cancel any pending callbacks on a queue | |
358 | * @q: the queue | |
359 | * | |
360 | * Description: | |
361 | * The block layer may perform asynchronous callback activity | |
362 | * on a queue, such as calling the unplug function after a timeout. | |
363 | * A block device may call blk_sync_queue to ensure that any | |
364 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 365 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
366 | * that its ->make_request_fn will not re-add plugging prior to calling |
367 | * this function. | |
368 | * | |
369 | */ | |
370 | void blk_sync_queue(struct request_queue *q) | |
371 | { | |
372 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 373 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
374 | } |
375 | EXPORT_SYMBOL(blk_sync_queue); | |
376 | ||
377 | /** | |
378 | * blk_run_queue - run a single device queue | |
379 | * @q: The queue to run | |
380 | */ | |
381 | void blk_run_queue(struct request_queue *q) | |
382 | { | |
383 | unsigned long flags; | |
384 | ||
385 | spin_lock_irqsave(q->queue_lock, flags); | |
386 | blk_remove_plug(q); | |
dac07ec1 JA |
387 | |
388 | /* | |
389 | * Only recurse once to avoid overrunning the stack, let the unplug | |
390 | * handling reinvoke the handler shortly if we already got there. | |
391 | */ | |
392 | if (!elv_queue_empty(q)) { | |
393 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
394 | q->request_fn(q); | |
395 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
396 | } else { | |
397 | blk_plug_device(q); | |
398 | kblockd_schedule_work(&q->unplug_work); | |
399 | } | |
400 | } | |
401 | ||
1da177e4 LT |
402 | spin_unlock_irqrestore(q->queue_lock, flags); |
403 | } | |
404 | EXPORT_SYMBOL(blk_run_queue); | |
405 | ||
165125e1 | 406 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
407 | { |
408 | kobject_put(&q->kobj); | |
409 | } | |
410 | EXPORT_SYMBOL(blk_put_queue); | |
411 | ||
6728cb0e | 412 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc AV |
413 | { |
414 | mutex_lock(&q->sysfs_lock); | |
415 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
416 | mutex_unlock(&q->sysfs_lock); | |
417 | ||
418 | if (q->elevator) | |
419 | elevator_exit(q->elevator); | |
420 | ||
421 | blk_put_queue(q); | |
422 | } | |
1da177e4 LT |
423 | EXPORT_SYMBOL(blk_cleanup_queue); |
424 | ||
165125e1 | 425 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
426 | { |
427 | struct request_list *rl = &q->rq; | |
428 | ||
429 | rl->count[READ] = rl->count[WRITE] = 0; | |
430 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 431 | rl->elvpriv = 0; |
1da177e4 LT |
432 | init_waitqueue_head(&rl->wait[READ]); |
433 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 434 | |
1946089a CL |
435 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
436 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
437 | |
438 | if (!rl->rq_pool) | |
439 | return -ENOMEM; | |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
165125e1 | 444 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 445 | { |
1946089a CL |
446 | return blk_alloc_queue_node(gfp_mask, -1); |
447 | } | |
448 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 449 | |
165125e1 | 450 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 451 | { |
165125e1 | 452 | struct request_queue *q; |
e0bf68dd | 453 | int err; |
1946089a | 454 | |
8324aa91 | 455 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 456 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
457 | if (!q) |
458 | return NULL; | |
459 | ||
e0bf68dd PZ |
460 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
461 | q->backing_dev_info.unplug_io_data = q; | |
462 | err = bdi_init(&q->backing_dev_info); | |
463 | if (err) { | |
8324aa91 | 464 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
465 | return NULL; |
466 | } | |
467 | ||
1da177e4 | 468 | init_timer(&q->unplug_timer); |
483f4afc | 469 | |
8324aa91 | 470 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 471 | |
483f4afc AV |
472 | mutex_init(&q->sysfs_lock); |
473 | ||
1da177e4 LT |
474 | return q; |
475 | } | |
1946089a | 476 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
477 | |
478 | /** | |
479 | * blk_init_queue - prepare a request queue for use with a block device | |
480 | * @rfn: The function to be called to process requests that have been | |
481 | * placed on the queue. | |
482 | * @lock: Request queue spin lock | |
483 | * | |
484 | * Description: | |
485 | * If a block device wishes to use the standard request handling procedures, | |
486 | * which sorts requests and coalesces adjacent requests, then it must | |
487 | * call blk_init_queue(). The function @rfn will be called when there | |
488 | * are requests on the queue that need to be processed. If the device | |
489 | * supports plugging, then @rfn may not be called immediately when requests | |
490 | * are available on the queue, but may be called at some time later instead. | |
491 | * Plugged queues are generally unplugged when a buffer belonging to one | |
492 | * of the requests on the queue is needed, or due to memory pressure. | |
493 | * | |
494 | * @rfn is not required, or even expected, to remove all requests off the | |
495 | * queue, but only as many as it can handle at a time. If it does leave | |
496 | * requests on the queue, it is responsible for arranging that the requests | |
497 | * get dealt with eventually. | |
498 | * | |
499 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
500 | * request queue; this lock will be taken also from interrupt context, so irq |
501 | * disabling is needed for it. | |
1da177e4 LT |
502 | * |
503 | * Function returns a pointer to the initialized request queue, or NULL if | |
504 | * it didn't succeed. | |
505 | * | |
506 | * Note: | |
507 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
508 | * when the block device is deactivated (such as at module unload). | |
509 | **/ | |
1946089a | 510 | |
165125e1 | 511 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 512 | { |
1946089a CL |
513 | return blk_init_queue_node(rfn, lock, -1); |
514 | } | |
515 | EXPORT_SYMBOL(blk_init_queue); | |
516 | ||
165125e1 | 517 | struct request_queue * |
1946089a CL |
518 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
519 | { | |
165125e1 | 520 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
521 | |
522 | if (!q) | |
523 | return NULL; | |
524 | ||
1946089a | 525 | q->node = node_id; |
8669aafd | 526 | if (blk_init_free_list(q)) { |
8324aa91 | 527 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
528 | return NULL; |
529 | } | |
1da177e4 | 530 | |
152587de | 531 | /* |
532 | * if caller didn't supply a lock, they get per-queue locking with | |
533 | * our embedded lock | |
534 | */ | |
535 | if (!lock) { | |
536 | spin_lock_init(&q->__queue_lock); | |
537 | lock = &q->__queue_lock; | |
538 | } | |
539 | ||
1da177e4 | 540 | q->request_fn = rfn; |
1da177e4 LT |
541 | q->prep_rq_fn = NULL; |
542 | q->unplug_fn = generic_unplug_device; | |
543 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
544 | q->queue_lock = lock; | |
545 | ||
546 | blk_queue_segment_boundary(q, 0xffffffff); | |
547 | ||
548 | blk_queue_make_request(q, __make_request); | |
549 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
550 | ||
551 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
552 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
553 | ||
44ec9542 AS |
554 | q->sg_reserved_size = INT_MAX; |
555 | ||
1da177e4 LT |
556 | /* |
557 | * all done | |
558 | */ | |
559 | if (!elevator_init(q, NULL)) { | |
560 | blk_queue_congestion_threshold(q); | |
561 | return q; | |
562 | } | |
563 | ||
8669aafd | 564 | blk_put_queue(q); |
1da177e4 LT |
565 | return NULL; |
566 | } | |
1946089a | 567 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 568 | |
165125e1 | 569 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 570 | { |
fde6ad22 | 571 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 572 | kobject_get(&q->kobj); |
1da177e4 LT |
573 | return 0; |
574 | } | |
575 | ||
576 | return 1; | |
577 | } | |
1da177e4 LT |
578 | EXPORT_SYMBOL(blk_get_queue); |
579 | ||
165125e1 | 580 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 581 | { |
4aff5e23 | 582 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 583 | elv_put_request(q, rq); |
1da177e4 LT |
584 | mempool_free(rq, q->rq.rq_pool); |
585 | } | |
586 | ||
1ea25ecb | 587 | static struct request * |
165125e1 | 588 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
589 | { |
590 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
591 | ||
592 | if (!rq) | |
593 | return NULL; | |
594 | ||
595 | /* | |
4aff5e23 | 596 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
597 | * see bio.h and blkdev.h |
598 | */ | |
49171e5c | 599 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 600 | |
cb98fc8b | 601 | if (priv) { |
cb78b285 | 602 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
603 | mempool_free(rq, q->rq.rq_pool); |
604 | return NULL; | |
605 | } | |
4aff5e23 | 606 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 607 | } |
1da177e4 | 608 | |
cb98fc8b | 609 | return rq; |
1da177e4 LT |
610 | } |
611 | ||
612 | /* | |
613 | * ioc_batching returns true if the ioc is a valid batching request and | |
614 | * should be given priority access to a request. | |
615 | */ | |
165125e1 | 616 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
617 | { |
618 | if (!ioc) | |
619 | return 0; | |
620 | ||
621 | /* | |
622 | * Make sure the process is able to allocate at least 1 request | |
623 | * even if the batch times out, otherwise we could theoretically | |
624 | * lose wakeups. | |
625 | */ | |
626 | return ioc->nr_batch_requests == q->nr_batching || | |
627 | (ioc->nr_batch_requests > 0 | |
628 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
629 | } | |
630 | ||
631 | /* | |
632 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
633 | * will cause the process to be a "batcher" on all queues in the system. This | |
634 | * is the behaviour we want though - once it gets a wakeup it should be given | |
635 | * a nice run. | |
636 | */ | |
165125e1 | 637 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
638 | { |
639 | if (!ioc || ioc_batching(q, ioc)) | |
640 | return; | |
641 | ||
642 | ioc->nr_batch_requests = q->nr_batching; | |
643 | ioc->last_waited = jiffies; | |
644 | } | |
645 | ||
165125e1 | 646 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
647 | { |
648 | struct request_list *rl = &q->rq; | |
649 | ||
650 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 651 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
652 | |
653 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
654 | if (waitqueue_active(&rl->wait[rw])) |
655 | wake_up(&rl->wait[rw]); | |
656 | ||
657 | blk_clear_queue_full(q, rw); | |
658 | } | |
659 | } | |
660 | ||
661 | /* | |
662 | * A request has just been released. Account for it, update the full and | |
663 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
664 | */ | |
165125e1 | 665 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
666 | { |
667 | struct request_list *rl = &q->rq; | |
668 | ||
669 | rl->count[rw]--; | |
cb98fc8b TH |
670 | if (priv) |
671 | rl->elvpriv--; | |
1da177e4 LT |
672 | |
673 | __freed_request(q, rw); | |
674 | ||
675 | if (unlikely(rl->starved[rw ^ 1])) | |
676 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
677 | } |
678 | ||
679 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
680 | /* | |
d6344532 NP |
681 | * Get a free request, queue_lock must be held. |
682 | * Returns NULL on failure, with queue_lock held. | |
683 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 684 | */ |
165125e1 | 685 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 686 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
687 | { |
688 | struct request *rq = NULL; | |
689 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 690 | struct io_context *ioc = NULL; |
7749a8d4 | 691 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
692 | int may_queue, priv; |
693 | ||
7749a8d4 | 694 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
695 | if (may_queue == ELV_MQUEUE_NO) |
696 | goto rq_starved; | |
697 | ||
698 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
699 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 700 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
701 | /* |
702 | * The queue will fill after this allocation, so set | |
703 | * it as full, and mark this process as "batching". | |
704 | * This process will be allowed to complete a batch of | |
705 | * requests, others will be blocked. | |
706 | */ | |
707 | if (!blk_queue_full(q, rw)) { | |
708 | ioc_set_batching(q, ioc); | |
709 | blk_set_queue_full(q, rw); | |
710 | } else { | |
711 | if (may_queue != ELV_MQUEUE_MUST | |
712 | && !ioc_batching(q, ioc)) { | |
713 | /* | |
714 | * The queue is full and the allocating | |
715 | * process is not a "batcher", and not | |
716 | * exempted by the IO scheduler | |
717 | */ | |
718 | goto out; | |
719 | } | |
720 | } | |
1da177e4 | 721 | } |
79e2de4b | 722 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
723 | } |
724 | ||
082cf69e JA |
725 | /* |
726 | * Only allow batching queuers to allocate up to 50% over the defined | |
727 | * limit of requests, otherwise we could have thousands of requests | |
728 | * allocated with any setting of ->nr_requests | |
729 | */ | |
fd782a4a | 730 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 731 | goto out; |
fd782a4a | 732 | |
1da177e4 LT |
733 | rl->count[rw]++; |
734 | rl->starved[rw] = 0; | |
cb98fc8b | 735 | |
64521d1a | 736 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
737 | if (priv) |
738 | rl->elvpriv++; | |
739 | ||
1da177e4 LT |
740 | spin_unlock_irq(q->queue_lock); |
741 | ||
7749a8d4 | 742 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 743 | if (unlikely(!rq)) { |
1da177e4 LT |
744 | /* |
745 | * Allocation failed presumably due to memory. Undo anything | |
746 | * we might have messed up. | |
747 | * | |
748 | * Allocating task should really be put onto the front of the | |
749 | * wait queue, but this is pretty rare. | |
750 | */ | |
751 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 752 | freed_request(q, rw, priv); |
1da177e4 LT |
753 | |
754 | /* | |
755 | * in the very unlikely event that allocation failed and no | |
756 | * requests for this direction was pending, mark us starved | |
757 | * so that freeing of a request in the other direction will | |
758 | * notice us. another possible fix would be to split the | |
759 | * rq mempool into READ and WRITE | |
760 | */ | |
761 | rq_starved: | |
762 | if (unlikely(rl->count[rw] == 0)) | |
763 | rl->starved[rw] = 1; | |
764 | ||
1da177e4 LT |
765 | goto out; |
766 | } | |
767 | ||
88ee5ef1 JA |
768 | /* |
769 | * ioc may be NULL here, and ioc_batching will be false. That's | |
770 | * OK, if the queue is under the request limit then requests need | |
771 | * not count toward the nr_batch_requests limit. There will always | |
772 | * be some limit enforced by BLK_BATCH_TIME. | |
773 | */ | |
1da177e4 LT |
774 | if (ioc_batching(q, ioc)) |
775 | ioc->nr_batch_requests--; | |
6728cb0e | 776 | |
1da177e4 | 777 | rq_init(q, rq); |
2056a782 JA |
778 | |
779 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 780 | out: |
1da177e4 LT |
781 | return rq; |
782 | } | |
783 | ||
784 | /* | |
785 | * No available requests for this queue, unplug the device and wait for some | |
786 | * requests to become available. | |
d6344532 NP |
787 | * |
788 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 789 | */ |
165125e1 | 790 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 791 | struct bio *bio) |
1da177e4 | 792 | { |
7749a8d4 | 793 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
794 | struct request *rq; |
795 | ||
7749a8d4 | 796 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
797 | while (!rq) { |
798 | DEFINE_WAIT(wait); | |
1da177e4 LT |
799 | struct request_list *rl = &q->rq; |
800 | ||
801 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
802 | TASK_UNINTERRUPTIBLE); | |
803 | ||
7749a8d4 | 804 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
805 | |
806 | if (!rq) { | |
807 | struct io_context *ioc; | |
808 | ||
2056a782 JA |
809 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
810 | ||
d6344532 NP |
811 | __generic_unplug_device(q); |
812 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
813 | io_schedule(); |
814 | ||
815 | /* | |
816 | * After sleeping, we become a "batching" process and | |
817 | * will be able to allocate at least one request, and | |
818 | * up to a big batch of them for a small period time. | |
819 | * See ioc_batching, ioc_set_batching | |
820 | */ | |
b5deef90 | 821 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 822 | ioc_set_batching(q, ioc); |
d6344532 NP |
823 | |
824 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
825 | } |
826 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 827 | } |
1da177e4 LT |
828 | |
829 | return rq; | |
830 | } | |
831 | ||
165125e1 | 832 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
833 | { |
834 | struct request *rq; | |
835 | ||
836 | BUG_ON(rw != READ && rw != WRITE); | |
837 | ||
d6344532 NP |
838 | spin_lock_irq(q->queue_lock); |
839 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 840 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 841 | } else { |
22e2c507 | 842 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
843 | if (!rq) |
844 | spin_unlock_irq(q->queue_lock); | |
845 | } | |
846 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
847 | |
848 | return rq; | |
849 | } | |
1da177e4 LT |
850 | EXPORT_SYMBOL(blk_get_request); |
851 | ||
dc72ef4a JA |
852 | /** |
853 | * blk_start_queueing - initiate dispatch of requests to device | |
854 | * @q: request queue to kick into gear | |
855 | * | |
856 | * This is basically a helper to remove the need to know whether a queue | |
857 | * is plugged or not if someone just wants to initiate dispatch of requests | |
858 | * for this queue. | |
859 | * | |
860 | * The queue lock must be held with interrupts disabled. | |
861 | */ | |
165125e1 | 862 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a JA |
863 | { |
864 | if (!blk_queue_plugged(q)) | |
865 | q->request_fn(q); | |
866 | else | |
867 | __generic_unplug_device(q); | |
868 | } | |
869 | EXPORT_SYMBOL(blk_start_queueing); | |
870 | ||
1da177e4 LT |
871 | /** |
872 | * blk_requeue_request - put a request back on queue | |
873 | * @q: request queue where request should be inserted | |
874 | * @rq: request to be inserted | |
875 | * | |
876 | * Description: | |
877 | * Drivers often keep queueing requests until the hardware cannot accept | |
878 | * more, when that condition happens we need to put the request back | |
879 | * on the queue. Must be called with queue lock held. | |
880 | */ | |
165125e1 | 881 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 882 | { |
2056a782 JA |
883 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
884 | ||
1da177e4 LT |
885 | if (blk_rq_tagged(rq)) |
886 | blk_queue_end_tag(q, rq); | |
887 | ||
888 | elv_requeue_request(q, rq); | |
889 | } | |
1da177e4 LT |
890 | EXPORT_SYMBOL(blk_requeue_request); |
891 | ||
892 | /** | |
893 | * blk_insert_request - insert a special request in to a request queue | |
894 | * @q: request queue where request should be inserted | |
895 | * @rq: request to be inserted | |
896 | * @at_head: insert request at head or tail of queue | |
897 | * @data: private data | |
1da177e4 LT |
898 | * |
899 | * Description: | |
900 | * Many block devices need to execute commands asynchronously, so they don't | |
901 | * block the whole kernel from preemption during request execution. This is | |
902 | * accomplished normally by inserting aritficial requests tagged as | |
903 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
904 | * scheduled for actual execution by the request queue. | |
905 | * | |
906 | * We have the option of inserting the head or the tail of the queue. | |
907 | * Typically we use the tail for new ioctls and so forth. We use the head | |
908 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
909 | * host that is unable to accept a particular command. | |
910 | */ | |
165125e1 | 911 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 912 | int at_head, void *data) |
1da177e4 | 913 | { |
867d1191 | 914 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
915 | unsigned long flags; |
916 | ||
917 | /* | |
918 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
919 | * must not attempt merges on this) and that it acts as a soft | |
920 | * barrier | |
921 | */ | |
4aff5e23 JA |
922 | rq->cmd_type = REQ_TYPE_SPECIAL; |
923 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
924 | |
925 | rq->special = data; | |
926 | ||
927 | spin_lock_irqsave(q->queue_lock, flags); | |
928 | ||
929 | /* | |
930 | * If command is tagged, release the tag | |
931 | */ | |
867d1191 TH |
932 | if (blk_rq_tagged(rq)) |
933 | blk_queue_end_tag(q, rq); | |
1da177e4 | 934 | |
b238b3d4 | 935 | drive_stat_acct(rq, 1); |
867d1191 | 936 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 937 | blk_start_queueing(q); |
1da177e4 LT |
938 | spin_unlock_irqrestore(q->queue_lock, flags); |
939 | } | |
1da177e4 LT |
940 | EXPORT_SYMBOL(blk_insert_request); |
941 | ||
1da177e4 LT |
942 | /* |
943 | * add-request adds a request to the linked list. | |
944 | * queue lock is held and interrupts disabled, as we muck with the | |
945 | * request queue list. | |
946 | */ | |
6728cb0e | 947 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 948 | { |
b238b3d4 | 949 | drive_stat_acct(req, 1); |
1da177e4 | 950 | |
1da177e4 LT |
951 | /* |
952 | * elevator indicated where it wants this request to be | |
953 | * inserted at elevator_merge time | |
954 | */ | |
955 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
956 | } | |
6728cb0e | 957 | |
1da177e4 LT |
958 | /* |
959 | * disk_round_stats() - Round off the performance stats on a struct | |
960 | * disk_stats. | |
961 | * | |
962 | * The average IO queue length and utilisation statistics are maintained | |
963 | * by observing the current state of the queue length and the amount of | |
964 | * time it has been in this state for. | |
965 | * | |
966 | * Normally, that accounting is done on IO completion, but that can result | |
967 | * in more than a second's worth of IO being accounted for within any one | |
968 | * second, leading to >100% utilisation. To deal with that, we call this | |
969 | * function to do a round-off before returning the results when reading | |
970 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
971 | * the current jiffies and restarts the counters again. | |
972 | */ | |
973 | void disk_round_stats(struct gendisk *disk) | |
974 | { | |
975 | unsigned long now = jiffies; | |
976 | ||
b2982649 CK |
977 | if (now == disk->stamp) |
978 | return; | |
1da177e4 | 979 | |
20e5c81f CK |
980 | if (disk->in_flight) { |
981 | __disk_stat_add(disk, time_in_queue, | |
982 | disk->in_flight * (now - disk->stamp)); | |
983 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
984 | } | |
1da177e4 | 985 | disk->stamp = now; |
1da177e4 | 986 | } |
3eaf840e JNN |
987 | EXPORT_SYMBOL_GPL(disk_round_stats); |
988 | ||
1da177e4 LT |
989 | /* |
990 | * queue lock must be held | |
991 | */ | |
165125e1 | 992 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 993 | { |
1da177e4 LT |
994 | if (unlikely(!q)) |
995 | return; | |
996 | if (unlikely(--req->ref_count)) | |
997 | return; | |
998 | ||
8922e16c TH |
999 | elv_completed_request(q, req); |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * Request may not have originated from ll_rw_blk. if not, | |
1003 | * it didn't come out of our reserved rq pools | |
1004 | */ | |
49171e5c | 1005 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 1006 | int rw = rq_data_dir(req); |
4aff5e23 | 1007 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1008 | |
1da177e4 | 1009 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1010 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1011 | |
1012 | blk_free_request(q, req); | |
cb98fc8b | 1013 | freed_request(q, rw, priv); |
1da177e4 LT |
1014 | } |
1015 | } | |
6e39b69e MC |
1016 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1017 | ||
1da177e4 LT |
1018 | void blk_put_request(struct request *req) |
1019 | { | |
8922e16c | 1020 | unsigned long flags; |
165125e1 | 1021 | struct request_queue *q = req->q; |
8922e16c | 1022 | |
1da177e4 | 1023 | /* |
8922e16c TH |
1024 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
1025 | * following if (q) test. | |
1da177e4 | 1026 | */ |
8922e16c | 1027 | if (q) { |
1da177e4 LT |
1028 | spin_lock_irqsave(q->queue_lock, flags); |
1029 | __blk_put_request(q, req); | |
1030 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1031 | } | |
1032 | } | |
1da177e4 LT |
1033 | EXPORT_SYMBOL(blk_put_request); |
1034 | ||
86db1e29 | 1035 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1036 | { |
4aff5e23 | 1037 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1038 | |
1039 | /* | |
1040 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1041 | */ | |
1042 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 1043 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
1044 | |
1045 | /* | |
1046 | * REQ_BARRIER implies no merging, but lets make it explicit | |
1047 | */ | |
1048 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 1049 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 1050 | |
b31dc66a | 1051 | if (bio_sync(bio)) |
4aff5e23 | 1052 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1053 | if (bio_rw_meta(bio)) |
1054 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 1055 | |
52d9e675 TH |
1056 | req->errors = 0; |
1057 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1058 | req->ioprio = bio_prio(bio); |
52d9e675 | 1059 | req->start_time = jiffies; |
bc1c56fd | 1060 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1061 | } |
1062 | ||
165125e1 | 1063 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1064 | { |
450991bc | 1065 | struct request *req; |
51da90fc JA |
1066 | int el_ret, nr_sectors, barrier, err; |
1067 | const unsigned short prio = bio_prio(bio); | |
1068 | const int sync = bio_sync(bio); | |
7749a8d4 | 1069 | int rw_flags; |
1da177e4 | 1070 | |
1da177e4 | 1071 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1072 | |
1073 | /* | |
1074 | * low level driver can indicate that it wants pages above a | |
1075 | * certain limit bounced to low memory (ie for highmem, or even | |
1076 | * ISA dma in theory) | |
1077 | */ | |
1078 | blk_queue_bounce(q, &bio); | |
1079 | ||
1da177e4 | 1080 | barrier = bio_barrier(bio); |
797e7dbb | 1081 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
1082 | err = -EOPNOTSUPP; |
1083 | goto end_io; | |
1084 | } | |
1085 | ||
1da177e4 LT |
1086 | spin_lock_irq(q->queue_lock); |
1087 | ||
450991bc | 1088 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
1089 | goto get_rq; |
1090 | ||
1091 | el_ret = elv_merge(q, &req, bio); | |
1092 | switch (el_ret) { | |
6728cb0e JA |
1093 | case ELEVATOR_BACK_MERGE: |
1094 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1095 | |
6728cb0e JA |
1096 | if (!ll_back_merge_fn(q, req, bio)) |
1097 | break; | |
1da177e4 | 1098 | |
6728cb0e | 1099 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2056a782 | 1100 | |
6728cb0e JA |
1101 | req->biotail->bi_next = bio; |
1102 | req->biotail = bio; | |
1103 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1104 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1105 | drive_stat_acct(req, 0); | |
1106 | if (!attempt_back_merge(q, req)) | |
1107 | elv_merged_request(q, req, el_ret); | |
1108 | goto out; | |
1da177e4 | 1109 | |
6728cb0e JA |
1110 | case ELEVATOR_FRONT_MERGE: |
1111 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1112 | |
6728cb0e JA |
1113 | if (!ll_front_merge_fn(q, req, bio)) |
1114 | break; | |
1da177e4 | 1115 | |
6728cb0e | 1116 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2056a782 | 1117 | |
6728cb0e JA |
1118 | bio->bi_next = req->bio; |
1119 | req->bio = bio; | |
1da177e4 | 1120 | |
6728cb0e JA |
1121 | /* |
1122 | * may not be valid. if the low level driver said | |
1123 | * it didn't need a bounce buffer then it better | |
1124 | * not touch req->buffer either... | |
1125 | */ | |
1126 | req->buffer = bio_data(bio); | |
1127 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1128 | req->hard_cur_sectors = req->current_nr_sectors; | |
1129 | req->sector = req->hard_sector = bio->bi_sector; | |
1130 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1131 | req->ioprio = ioprio_best(req->ioprio, prio); | |
1132 | drive_stat_acct(req, 0); | |
1133 | if (!attempt_front_merge(q, req)) | |
1134 | elv_merged_request(q, req, el_ret); | |
1135 | goto out; | |
1136 | ||
1137 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1138 | default: | |
1139 | ; | |
1da177e4 LT |
1140 | } |
1141 | ||
450991bc | 1142 | get_rq: |
7749a8d4 JA |
1143 | /* |
1144 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1145 | * but we need to set it earlier to expose the sync flag to the | |
1146 | * rq allocator and io schedulers. | |
1147 | */ | |
1148 | rw_flags = bio_data_dir(bio); | |
1149 | if (sync) | |
1150 | rw_flags |= REQ_RW_SYNC; | |
1151 | ||
1da177e4 | 1152 | /* |
450991bc | 1153 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1154 | * Returns with the queue unlocked. |
450991bc | 1155 | */ |
7749a8d4 | 1156 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1157 | |
450991bc NP |
1158 | /* |
1159 | * After dropping the lock and possibly sleeping here, our request | |
1160 | * may now be mergeable after it had proven unmergeable (above). | |
1161 | * We don't worry about that case for efficiency. It won't happen | |
1162 | * often, and the elevators are able to handle it. | |
1da177e4 | 1163 | */ |
52d9e675 | 1164 | init_request_from_bio(req, bio); |
1da177e4 | 1165 | |
450991bc NP |
1166 | spin_lock_irq(q->queue_lock); |
1167 | if (elv_queue_empty(q)) | |
1168 | blk_plug_device(q); | |
1da177e4 LT |
1169 | add_request(q, req); |
1170 | out: | |
4a534f93 | 1171 | if (sync) |
1da177e4 LT |
1172 | __generic_unplug_device(q); |
1173 | ||
1174 | spin_unlock_irq(q->queue_lock); | |
1175 | return 0; | |
1176 | ||
1177 | end_io: | |
6712ecf8 | 1178 | bio_endio(bio, err); |
1da177e4 LT |
1179 | return 0; |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * If bio->bi_dev is a partition, remap the location | |
1184 | */ | |
1185 | static inline void blk_partition_remap(struct bio *bio) | |
1186 | { | |
1187 | struct block_device *bdev = bio->bi_bdev; | |
1188 | ||
bf2de6f5 | 1189 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 | 1190 | struct hd_struct *p = bdev->bd_part; |
a362357b JA |
1191 | const int rw = bio_data_dir(bio); |
1192 | ||
1193 | p->sectors[rw] += bio_sectors(bio); | |
1194 | p->ios[rw]++; | |
1da177e4 | 1195 | |
1da177e4 LT |
1196 | bio->bi_sector += p->start_sect; |
1197 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b AB |
1198 | |
1199 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
1200 | bdev->bd_dev, bio->bi_sector, | |
1201 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1202 | } |
1203 | } | |
1204 | ||
1da177e4 LT |
1205 | static void handle_bad_sector(struct bio *bio) |
1206 | { | |
1207 | char b[BDEVNAME_SIZE]; | |
1208 | ||
1209 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1210 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1211 | bdevname(bio->bi_bdev, b), | |
1212 | bio->bi_rw, | |
1213 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1214 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1215 | ||
1216 | set_bit(BIO_EOF, &bio->bi_flags); | |
1217 | } | |
1218 | ||
c17bb495 AM |
1219 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1220 | ||
1221 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1222 | ||
1223 | static int __init setup_fail_make_request(char *str) | |
1224 | { | |
1225 | return setup_fault_attr(&fail_make_request, str); | |
1226 | } | |
1227 | __setup("fail_make_request=", setup_fail_make_request); | |
1228 | ||
1229 | static int should_fail_request(struct bio *bio) | |
1230 | { | |
1231 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
1232 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
1233 | return should_fail(&fail_make_request, bio->bi_size); | |
1234 | ||
1235 | return 0; | |
1236 | } | |
1237 | ||
1238 | static int __init fail_make_request_debugfs(void) | |
1239 | { | |
1240 | return init_fault_attr_dentries(&fail_make_request, | |
1241 | "fail_make_request"); | |
1242 | } | |
1243 | ||
1244 | late_initcall(fail_make_request_debugfs); | |
1245 | ||
1246 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1247 | ||
1248 | static inline int should_fail_request(struct bio *bio) | |
1249 | { | |
1250 | return 0; | |
1251 | } | |
1252 | ||
1253 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1254 | ||
c07e2b41 JA |
1255 | /* |
1256 | * Check whether this bio extends beyond the end of the device. | |
1257 | */ | |
1258 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1259 | { | |
1260 | sector_t maxsector; | |
1261 | ||
1262 | if (!nr_sectors) | |
1263 | return 0; | |
1264 | ||
1265 | /* Test device or partition size, when known. */ | |
1266 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1267 | if (maxsector) { | |
1268 | sector_t sector = bio->bi_sector; | |
1269 | ||
1270 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1271 | /* | |
1272 | * This may well happen - the kernel calls bread() | |
1273 | * without checking the size of the device, e.g., when | |
1274 | * mounting a device. | |
1275 | */ | |
1276 | handle_bad_sector(bio); | |
1277 | return 1; | |
1278 | } | |
1279 | } | |
1280 | ||
1281 | return 0; | |
1282 | } | |
1283 | ||
1da177e4 LT |
1284 | /** |
1285 | * generic_make_request: hand a buffer to its device driver for I/O | |
1286 | * @bio: The bio describing the location in memory and on the device. | |
1287 | * | |
1288 | * generic_make_request() is used to make I/O requests of block | |
1289 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1290 | * to be done. | |
1291 | * | |
1292 | * generic_make_request() does not return any status. The | |
1293 | * success/failure status of the request, along with notification of | |
1294 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1295 | * function described (one day) else where. | |
1296 | * | |
1297 | * The caller of generic_make_request must make sure that bi_io_vec | |
1298 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1299 | * set to describe the device address, and the | |
1300 | * bi_end_io and optionally bi_private are set to describe how | |
1301 | * completion notification should be signaled. | |
1302 | * | |
1303 | * generic_make_request and the drivers it calls may use bi_next if this | |
1304 | * bio happens to be merged with someone else, and may change bi_dev and | |
1305 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1306 | * should NOT be depended on after the call to generic_make_request. | |
1307 | */ | |
d89d8796 | 1308 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1309 | { |
165125e1 | 1310 | struct request_queue *q; |
5ddfe969 | 1311 | sector_t old_sector; |
1da177e4 | 1312 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1313 | dev_t old_dev; |
51fd77bd | 1314 | int err = -EIO; |
1da177e4 LT |
1315 | |
1316 | might_sleep(); | |
1da177e4 | 1317 | |
c07e2b41 JA |
1318 | if (bio_check_eod(bio, nr_sectors)) |
1319 | goto end_io; | |
1da177e4 LT |
1320 | |
1321 | /* | |
1322 | * Resolve the mapping until finished. (drivers are | |
1323 | * still free to implement/resolve their own stacking | |
1324 | * by explicitly returning 0) | |
1325 | * | |
1326 | * NOTE: we don't repeat the blk_size check for each new device. | |
1327 | * Stacking drivers are expected to know what they are doing. | |
1328 | */ | |
5ddfe969 | 1329 | old_sector = -1; |
2056a782 | 1330 | old_dev = 0; |
1da177e4 LT |
1331 | do { |
1332 | char b[BDEVNAME_SIZE]; | |
1333 | ||
1334 | q = bdev_get_queue(bio->bi_bdev); | |
1335 | if (!q) { | |
1336 | printk(KERN_ERR | |
1337 | "generic_make_request: Trying to access " | |
1338 | "nonexistent block-device %s (%Lu)\n", | |
1339 | bdevname(bio->bi_bdev, b), | |
1340 | (long long) bio->bi_sector); | |
1341 | end_io: | |
51fd77bd | 1342 | bio_endio(bio, err); |
1da177e4 LT |
1343 | break; |
1344 | } | |
1345 | ||
4fa253f3 | 1346 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1347 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1348 | bdevname(bio->bi_bdev, b), |
1349 | bio_sectors(bio), | |
1350 | q->max_hw_sectors); | |
1351 | goto end_io; | |
1352 | } | |
1353 | ||
fde6ad22 | 1354 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1355 | goto end_io; |
1356 | ||
c17bb495 AM |
1357 | if (should_fail_request(bio)) |
1358 | goto end_io; | |
1359 | ||
1da177e4 LT |
1360 | /* |
1361 | * If this device has partitions, remap block n | |
1362 | * of partition p to block n+start(p) of the disk. | |
1363 | */ | |
1364 | blk_partition_remap(bio); | |
1365 | ||
5ddfe969 | 1366 | if (old_sector != -1) |
4fa253f3 | 1367 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1368 | old_sector); |
2056a782 JA |
1369 | |
1370 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
1371 | ||
5ddfe969 | 1372 | old_sector = bio->bi_sector; |
2056a782 JA |
1373 | old_dev = bio->bi_bdev->bd_dev; |
1374 | ||
c07e2b41 JA |
1375 | if (bio_check_eod(bio, nr_sectors)) |
1376 | goto end_io; | |
51fd77bd JA |
1377 | if (bio_empty_barrier(bio) && !q->prepare_flush_fn) { |
1378 | err = -EOPNOTSUPP; | |
1379 | goto end_io; | |
1380 | } | |
5ddfe969 | 1381 | |
1da177e4 LT |
1382 | ret = q->make_request_fn(q, bio); |
1383 | } while (ret); | |
1384 | } | |
1385 | ||
d89d8796 NB |
1386 | /* |
1387 | * We only want one ->make_request_fn to be active at a time, | |
1388 | * else stack usage with stacked devices could be a problem. | |
1389 | * So use current->bio_{list,tail} to keep a list of requests | |
1390 | * submited by a make_request_fn function. | |
1391 | * current->bio_tail is also used as a flag to say if | |
1392 | * generic_make_request is currently active in this task or not. | |
1393 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1394 | * then a make_request is active, and new requests should be added | |
1395 | * at the tail | |
1396 | */ | |
1397 | void generic_make_request(struct bio *bio) | |
1398 | { | |
1399 | if (current->bio_tail) { | |
1400 | /* make_request is active */ | |
1401 | *(current->bio_tail) = bio; | |
1402 | bio->bi_next = NULL; | |
1403 | current->bio_tail = &bio->bi_next; | |
1404 | return; | |
1405 | } | |
1406 | /* following loop may be a bit non-obvious, and so deserves some | |
1407 | * explanation. | |
1408 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1409 | * ensure that) so we have a list with a single bio. | |
1410 | * We pretend that we have just taken it off a longer list, so | |
1411 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1412 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1413 | * added. __generic_make_request may indeed add some more bios | |
1414 | * through a recursive call to generic_make_request. If it | |
1415 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1416 | * from the top. In this case we really did just take the bio | |
1417 | * of the top of the list (no pretending) and so fixup bio_list and | |
1418 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1419 | * | |
1420 | * The loop was structured like this to make only one call to | |
1421 | * __generic_make_request (which is important as it is large and | |
1422 | * inlined) and to keep the structure simple. | |
1423 | */ | |
1424 | BUG_ON(bio->bi_next); | |
1425 | do { | |
1426 | current->bio_list = bio->bi_next; | |
1427 | if (bio->bi_next == NULL) | |
1428 | current->bio_tail = ¤t->bio_list; | |
1429 | else | |
1430 | bio->bi_next = NULL; | |
1431 | __generic_make_request(bio); | |
1432 | bio = current->bio_list; | |
1433 | } while (bio); | |
1434 | current->bio_tail = NULL; /* deactivate */ | |
1435 | } | |
1da177e4 LT |
1436 | EXPORT_SYMBOL(generic_make_request); |
1437 | ||
1438 | /** | |
1439 | * submit_bio: submit a bio to the block device layer for I/O | |
1440 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
1441 | * @bio: The &struct bio which describes the I/O | |
1442 | * | |
1443 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1444 | * uses that function to do most of the work. Both are fairly rough | |
1445 | * interfaces, @bio must be presetup and ready for I/O. | |
1446 | * | |
1447 | */ | |
1448 | void submit_bio(int rw, struct bio *bio) | |
1449 | { | |
1450 | int count = bio_sectors(bio); | |
1451 | ||
22e2c507 | 1452 | bio->bi_rw |= rw; |
1da177e4 | 1453 | |
bf2de6f5 JA |
1454 | /* |
1455 | * If it's a regular read/write or a barrier with data attached, | |
1456 | * go through the normal accounting stuff before submission. | |
1457 | */ | |
1458 | if (!bio_empty_barrier(bio)) { | |
1459 | ||
1460 | BIO_BUG_ON(!bio->bi_size); | |
1461 | BIO_BUG_ON(!bio->bi_io_vec); | |
1462 | ||
1463 | if (rw & WRITE) { | |
1464 | count_vm_events(PGPGOUT, count); | |
1465 | } else { | |
1466 | task_io_account_read(bio->bi_size); | |
1467 | count_vm_events(PGPGIN, count); | |
1468 | } | |
1469 | ||
1470 | if (unlikely(block_dump)) { | |
1471 | char b[BDEVNAME_SIZE]; | |
1472 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1473 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1474 | (rw & WRITE) ? "WRITE" : "READ", |
1475 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1476 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1477 | } |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | generic_make_request(bio); | |
1481 | } | |
1da177e4 LT |
1482 | EXPORT_SYMBOL(submit_bio); |
1483 | ||
3bcddeac KU |
1484 | /** |
1485 | * __end_that_request_first - end I/O on a request | |
1486 | * @req: the request being processed | |
5450d3e1 | 1487 | * @error: 0 for success, < 0 for error |
3bcddeac KU |
1488 | * @nr_bytes: number of bytes to complete |
1489 | * | |
1490 | * Description: | |
1491 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1492 | * for the next range of segments (if any) in the cluster. | |
1493 | * | |
1494 | * Return: | |
1495 | * 0 - we are done with this request, call end_that_request_last() | |
1496 | * 1 - still buffers pending for this request | |
1497 | **/ | |
5450d3e1 | 1498 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1499 | int nr_bytes) |
1500 | { | |
5450d3e1 | 1501 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1502 | struct bio *bio; |
1503 | ||
2056a782 JA |
1504 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
1505 | ||
1da177e4 LT |
1506 | /* |
1507 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
1508 | * sense key with us all the way through | |
1509 | */ | |
1510 | if (!blk_pc_request(req)) | |
1511 | req->errors = 0; | |
1512 | ||
6728cb0e JA |
1513 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1514 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1515 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1516 | (unsigned long long)req->sector); | |
1517 | } | |
1518 | ||
d72d904a | 1519 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
1520 | const int rw = rq_data_dir(req); |
1521 | ||
53e86061 | 1522 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
d72d904a JA |
1523 | } |
1524 | ||
1da177e4 LT |
1525 | total_bytes = bio_nbytes = 0; |
1526 | while ((bio = req->bio) != NULL) { | |
1527 | int nbytes; | |
1528 | ||
bf2de6f5 JA |
1529 | /* |
1530 | * For an empty barrier request, the low level driver must | |
1531 | * store a potential error location in ->sector. We pass | |
1532 | * that back up in ->bi_sector. | |
1533 | */ | |
1534 | if (blk_empty_barrier(req)) | |
1535 | bio->bi_sector = req->sector; | |
1536 | ||
1da177e4 LT |
1537 | if (nr_bytes >= bio->bi_size) { |
1538 | req->bio = bio->bi_next; | |
1539 | nbytes = bio->bi_size; | |
5bb23a68 | 1540 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1541 | next_idx = 0; |
1542 | bio_nbytes = 0; | |
1543 | } else { | |
1544 | int idx = bio->bi_idx + next_idx; | |
1545 | ||
1546 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1547 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e JA |
1548 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
1549 | __FUNCTION__, bio->bi_idx, | |
1550 | bio->bi_vcnt); | |
1da177e4 LT |
1551 | break; |
1552 | } | |
1553 | ||
1554 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1555 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1556 | ||
1557 | /* | |
1558 | * not a complete bvec done | |
1559 | */ | |
1560 | if (unlikely(nbytes > nr_bytes)) { | |
1561 | bio_nbytes += nr_bytes; | |
1562 | total_bytes += nr_bytes; | |
1563 | break; | |
1564 | } | |
1565 | ||
1566 | /* | |
1567 | * advance to the next vector | |
1568 | */ | |
1569 | next_idx++; | |
1570 | bio_nbytes += nbytes; | |
1571 | } | |
1572 | ||
1573 | total_bytes += nbytes; | |
1574 | nr_bytes -= nbytes; | |
1575 | ||
6728cb0e JA |
1576 | bio = req->bio; |
1577 | if (bio) { | |
1da177e4 LT |
1578 | /* |
1579 | * end more in this run, or just return 'not-done' | |
1580 | */ | |
1581 | if (unlikely(nr_bytes <= 0)) | |
1582 | break; | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * completely done | |
1588 | */ | |
1589 | if (!req->bio) | |
1590 | return 0; | |
1591 | ||
1592 | /* | |
1593 | * if the request wasn't completed, update state | |
1594 | */ | |
1595 | if (bio_nbytes) { | |
5bb23a68 | 1596 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1597 | bio->bi_idx += next_idx; |
1598 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1599 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1600 | } | |
1601 | ||
1602 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1603 | blk_recalc_rq_segments(req); | |
1604 | return 1; | |
1605 | } | |
1606 | ||
ff856bad JA |
1607 | /* |
1608 | * splice the completion data to a local structure and hand off to | |
1609 | * process_completion_queue() to complete the requests | |
1610 | */ | |
1611 | static void blk_done_softirq(struct softirq_action *h) | |
1612 | { | |
626ab0e6 | 1613 | struct list_head *cpu_list, local_list; |
ff856bad JA |
1614 | |
1615 | local_irq_disable(); | |
1616 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 1617 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
1618 | local_irq_enable(); |
1619 | ||
1620 | while (!list_empty(&local_list)) { | |
6728cb0e | 1621 | struct request *rq; |
ff856bad | 1622 | |
6728cb0e | 1623 | rq = list_entry(local_list.next, struct request, donelist); |
ff856bad JA |
1624 | list_del_init(&rq->donelist); |
1625 | rq->q->softirq_done_fn(rq); | |
1626 | } | |
1627 | } | |
1628 | ||
6728cb0e JA |
1629 | static int __cpuinit blk_cpu_notify(struct notifier_block *self, |
1630 | unsigned long action, void *hcpu) | |
ff856bad JA |
1631 | { |
1632 | /* | |
1633 | * If a CPU goes away, splice its entries to the current CPU | |
1634 | * and trigger a run of the softirq | |
1635 | */ | |
8bb78442 | 1636 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
1637 | int cpu = (unsigned long) hcpu; |
1638 | ||
1639 | local_irq_disable(); | |
1640 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
1641 | &__get_cpu_var(blk_cpu_done)); | |
1642 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
1643 | local_irq_enable(); | |
1644 | } | |
1645 | ||
1646 | return NOTIFY_OK; | |
1647 | } | |
1648 | ||
1649 | ||
db47d475 | 1650 | static struct notifier_block blk_cpu_notifier __cpuinitdata = { |
ff856bad JA |
1651 | .notifier_call = blk_cpu_notify, |
1652 | }; | |
1653 | ||
ff856bad JA |
1654 | /** |
1655 | * blk_complete_request - end I/O on a request | |
1656 | * @req: the request being processed | |
1657 | * | |
1658 | * Description: | |
1659 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 1660 | * unless the driver actually implements this in its completion callback |
4fa253f3 | 1661 | * through requeueing. The actual completion happens out-of-order, |
ff856bad JA |
1662 | * through a softirq handler. The user must have registered a completion |
1663 | * callback through blk_queue_softirq_done(). | |
1664 | **/ | |
1665 | ||
1666 | void blk_complete_request(struct request *req) | |
1667 | { | |
1668 | struct list_head *cpu_list; | |
1669 | unsigned long flags; | |
1670 | ||
1671 | BUG_ON(!req->q->softirq_done_fn); | |
6728cb0e | 1672 | |
ff856bad JA |
1673 | local_irq_save(flags); |
1674 | ||
1675 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
1676 | list_add_tail(&req->donelist, cpu_list); | |
1677 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
1678 | ||
1679 | local_irq_restore(flags); | |
1680 | } | |
ff856bad | 1681 | EXPORT_SYMBOL(blk_complete_request); |
6728cb0e | 1682 | |
1da177e4 LT |
1683 | /* |
1684 | * queue lock must be held | |
1685 | */ | |
5450d3e1 | 1686 | static void end_that_request_last(struct request *req, int error) |
1da177e4 LT |
1687 | { |
1688 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 | 1689 | |
b8286239 KU |
1690 | if (blk_rq_tagged(req)) |
1691 | blk_queue_end_tag(req->q, req); | |
1692 | ||
1693 | if (blk_queued_rq(req)) | |
1694 | blkdev_dequeue_request(req); | |
1da177e4 LT |
1695 | |
1696 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1697 | laptop_io_completion(); | |
1698 | ||
fd0ff8aa JA |
1699 | /* |
1700 | * Account IO completion. bar_rq isn't accounted as a normal | |
1701 | * IO on queueing nor completion. Accounting the containing | |
1702 | * request is enough. | |
1703 | */ | |
1704 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 1705 | unsigned long duration = jiffies - req->start_time; |
a362357b JA |
1706 | const int rw = rq_data_dir(req); |
1707 | ||
1708 | __disk_stat_inc(disk, ios[rw]); | |
1709 | __disk_stat_add(disk, ticks[rw], duration); | |
1da177e4 LT |
1710 | disk_round_stats(disk); |
1711 | disk->in_flight--; | |
1712 | } | |
b8286239 | 1713 | |
1da177e4 | 1714 | if (req->end_io) |
8ffdc655 | 1715 | req->end_io(req, error); |
b8286239 KU |
1716 | else { |
1717 | if (blk_bidi_rq(req)) | |
1718 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1719 | ||
1da177e4 | 1720 | __blk_put_request(req->q, req); |
b8286239 | 1721 | } |
1da177e4 LT |
1722 | } |
1723 | ||
a0cd1285 | 1724 | static inline void __end_request(struct request *rq, int uptodate, |
9e6e39f2 | 1725 | unsigned int nr_bytes) |
1da177e4 | 1726 | { |
9e6e39f2 KU |
1727 | int error = 0; |
1728 | ||
1729 | if (uptodate <= 0) | |
1730 | error = uptodate ? uptodate : -EIO; | |
1731 | ||
1732 | __blk_end_request(rq, error, nr_bytes); | |
1da177e4 LT |
1733 | } |
1734 | ||
3b11313a KU |
1735 | /** |
1736 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
1737 | **/ | |
1738 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1739 | { |
1740 | if (blk_fs_request(rq)) | |
1741 | return rq->hard_nr_sectors << 9; | |
1742 | ||
1743 | return rq->data_len; | |
1744 | } | |
3b11313a KU |
1745 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1746 | ||
1747 | /** | |
1748 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
1749 | **/ | |
1750 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1751 | { | |
1752 | if (blk_fs_request(rq)) | |
1753 | return rq->current_nr_sectors << 9; | |
1754 | ||
1755 | if (rq->bio) | |
1756 | return rq->bio->bi_size; | |
1757 | ||
1758 | return rq->data_len; | |
1759 | } | |
1760 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 JA |
1761 | |
1762 | /** | |
1763 | * end_queued_request - end all I/O on a queued request | |
1764 | * @rq: the request being processed | |
1765 | * @uptodate: error value or 0/1 uptodate flag | |
1766 | * | |
1767 | * Description: | |
1768 | * Ends all I/O on a request, and removes it from the block layer queues. | |
1769 | * Not suitable for normal IO completion, unless the driver still has | |
1770 | * the request attached to the block layer. | |
1771 | * | |
1772 | **/ | |
1773 | void end_queued_request(struct request *rq, int uptodate) | |
1774 | { | |
9e6e39f2 | 1775 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1776 | } |
1777 | EXPORT_SYMBOL(end_queued_request); | |
1778 | ||
1779 | /** | |
1780 | * end_dequeued_request - end all I/O on a dequeued request | |
1781 | * @rq: the request being processed | |
1782 | * @uptodate: error value or 0/1 uptodate flag | |
1783 | * | |
1784 | * Description: | |
1785 | * Ends all I/O on a request. The request must already have been | |
1786 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
1787 | * for most drivers. | |
1788 | * | |
1789 | **/ | |
1790 | void end_dequeued_request(struct request *rq, int uptodate) | |
1791 | { | |
9e6e39f2 | 1792 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
1793 | } |
1794 | EXPORT_SYMBOL(end_dequeued_request); | |
1795 | ||
1796 | ||
1797 | /** | |
1798 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1799 | * @req: the request being processed |
a0cd1285 JA |
1800 | * @uptodate: error value or 0/1 uptodate flag |
1801 | * | |
1802 | * Description: | |
1803 | * Ends I/O on the current segment of a request. If that is the only | |
1804 | * remaining segment, the request is also completed and freed. | |
1805 | * | |
1806 | * This is a remnant of how older block drivers handled IO completions. | |
1807 | * Modern drivers typically end IO on the full request in one go, unless | |
1808 | * they have a residual value to account for. For that case this function | |
1809 | * isn't really useful, unless the residual just happens to be the | |
1810 | * full current segment. In other words, don't use this function in new | |
1811 | * code. Either use end_request_completely(), or the | |
1812 | * end_that_request_chunk() (along with end_that_request_last()) for | |
1813 | * partial completions. | |
1814 | * | |
1815 | **/ | |
1816 | void end_request(struct request *req, int uptodate) | |
1817 | { | |
9e6e39f2 | 1818 | __end_request(req, uptodate, req->hard_cur_sectors << 9); |
a0cd1285 | 1819 | } |
1da177e4 LT |
1820 | EXPORT_SYMBOL(end_request); |
1821 | ||
336cdb40 | 1822 | /** |
e19a3ab0 KU |
1823 | * blk_end_io - Generic end_io function to complete a request. |
1824 | * @rq: the request being processed | |
1825 | * @error: 0 for success, < 0 for error | |
e3a04fe3 KU |
1826 | * @nr_bytes: number of bytes to complete @rq |
1827 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1828 | * @drv_callback: function called between completion of bios in the request |
1829 | * and completion of the request. | |
1830 | * If the callback returns non 0, this helper returns without | |
1831 | * completion of the request. | |
336cdb40 KU |
1832 | * |
1833 | * Description: | |
e3a04fe3 | 1834 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1835 | * If @rq has leftover, sets it up for the next range of segments. |
1836 | * | |
1837 | * Return: | |
1838 | * 0 - we are done with this request | |
e19a3ab0 | 1839 | * 1 - this request is not freed yet, it still has pending buffers. |
336cdb40 | 1840 | **/ |
22b13210 JA |
1841 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1842 | unsigned int bidi_bytes, | |
1843 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1844 | { |
1845 | struct request_queue *q = rq->q; | |
1846 | unsigned long flags = 0UL; | |
336cdb40 KU |
1847 | |
1848 | if (blk_fs_request(rq) || blk_pc_request(rq)) { | |
5450d3e1 | 1849 | if (__end_that_request_first(rq, error, nr_bytes)) |
336cdb40 | 1850 | return 1; |
e3a04fe3 KU |
1851 | |
1852 | /* Bidi request must be completed as a whole */ | |
1853 | if (blk_bidi_rq(rq) && | |
5450d3e1 | 1854 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) |
e3a04fe3 | 1855 | return 1; |
336cdb40 KU |
1856 | } |
1857 | ||
e19a3ab0 KU |
1858 | /* Special feature for tricky drivers */ |
1859 | if (drv_callback && drv_callback(rq)) | |
1860 | return 1; | |
1861 | ||
336cdb40 KU |
1862 | add_disk_randomness(rq->rq_disk); |
1863 | ||
1864 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1865 | end_that_request_last(rq, error); |
336cdb40 KU |
1866 | spin_unlock_irqrestore(q->queue_lock, flags); |
1867 | ||
1868 | return 0; | |
1869 | } | |
e19a3ab0 KU |
1870 | |
1871 | /** | |
1872 | * blk_end_request - Helper function for drivers to complete the request. | |
1873 | * @rq: the request being processed | |
1874 | * @error: 0 for success, < 0 for error | |
1875 | * @nr_bytes: number of bytes to complete | |
1876 | * | |
1877 | * Description: | |
1878 | * Ends I/O on a number of bytes attached to @rq. | |
1879 | * If @rq has leftover, sets it up for the next range of segments. | |
1880 | * | |
1881 | * Return: | |
1882 | * 0 - we are done with this request | |
1883 | * 1 - still buffers pending for this request | |
1884 | **/ | |
22b13210 | 1885 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1886 | { |
e3a04fe3 | 1887 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1888 | } |
336cdb40 KU |
1889 | EXPORT_SYMBOL_GPL(blk_end_request); |
1890 | ||
1891 | /** | |
1892 | * __blk_end_request - Helper function for drivers to complete the request. | |
1893 | * @rq: the request being processed | |
1894 | * @error: 0 for success, < 0 for error | |
1895 | * @nr_bytes: number of bytes to complete | |
1896 | * | |
1897 | * Description: | |
1898 | * Must be called with queue lock held unlike blk_end_request(). | |
1899 | * | |
1900 | * Return: | |
1901 | * 0 - we are done with this request | |
1902 | * 1 - still buffers pending for this request | |
1903 | **/ | |
22b13210 | 1904 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1905 | { |
336cdb40 | 1906 | if (blk_fs_request(rq) || blk_pc_request(rq)) { |
5450d3e1 | 1907 | if (__end_that_request_first(rq, error, nr_bytes)) |
336cdb40 KU |
1908 | return 1; |
1909 | } | |
1910 | ||
1911 | add_disk_randomness(rq->rq_disk); | |
1912 | ||
b8286239 | 1913 | end_that_request_last(rq, error); |
336cdb40 KU |
1914 | |
1915 | return 0; | |
1916 | } | |
1917 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1918 | ||
e3a04fe3 KU |
1919 | /** |
1920 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1921 | * @rq: the bidi request being processed | |
1922 | * @error: 0 for success, < 0 for error | |
1923 | * @nr_bytes: number of bytes to complete @rq | |
1924 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1925 | * | |
1926 | * Description: | |
1927 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1928 | * | |
1929 | * Return: | |
1930 | * 0 - we are done with this request | |
1931 | * 1 - still buffers pending for this request | |
1932 | **/ | |
22b13210 JA |
1933 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
1934 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
1935 | { |
1936 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
1937 | } | |
1938 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
1939 | ||
e19a3ab0 KU |
1940 | /** |
1941 | * blk_end_request_callback - Special helper function for tricky drivers | |
1942 | * @rq: the request being processed | |
1943 | * @error: 0 for success, < 0 for error | |
1944 | * @nr_bytes: number of bytes to complete | |
1945 | * @drv_callback: function called between completion of bios in the request | |
1946 | * and completion of the request. | |
1947 | * If the callback returns non 0, this helper returns without | |
1948 | * completion of the request. | |
1949 | * | |
1950 | * Description: | |
1951 | * Ends I/O on a number of bytes attached to @rq. | |
1952 | * If @rq has leftover, sets it up for the next range of segments. | |
1953 | * | |
1954 | * This special helper function is used only for existing tricky drivers. | |
1955 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
1956 | * This interface will be removed when such drivers are rewritten. | |
1957 | * Don't use this interface in other places anymore. | |
1958 | * | |
1959 | * Return: | |
1960 | * 0 - we are done with this request | |
1961 | * 1 - this request is not freed yet. | |
1962 | * this request still has pending buffers or | |
1963 | * the driver doesn't want to finish this request yet. | |
1964 | **/ | |
22b13210 JA |
1965 | int blk_end_request_callback(struct request *rq, int error, |
1966 | unsigned int nr_bytes, | |
e19a3ab0 KU |
1967 | int (drv_callback)(struct request *)) |
1968 | { | |
e3a04fe3 | 1969 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
1970 | } |
1971 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
1972 | ||
86db1e29 JA |
1973 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
1974 | struct bio *bio) | |
1da177e4 | 1975 | { |
4aff5e23 JA |
1976 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
1977 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
1978 | |
1979 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
1980 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
1981 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
1982 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
1983 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
1984 | rq->buffer = bio_data(bio); | |
0e75f906 | 1985 | rq->data_len = bio->bi_size; |
1da177e4 LT |
1986 | |
1987 | rq->bio = rq->biotail = bio; | |
1da177e4 | 1988 | |
66846572 N |
1989 | if (bio->bi_bdev) |
1990 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
1991 | } | |
1da177e4 LT |
1992 | |
1993 | int kblockd_schedule_work(struct work_struct *work) | |
1994 | { | |
1995 | return queue_work(kblockd_workqueue, work); | |
1996 | } | |
1da177e4 LT |
1997 | EXPORT_SYMBOL(kblockd_schedule_work); |
1998 | ||
19a75d83 | 1999 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 2000 | { |
28e53bdd | 2001 | cancel_work_sync(work); |
1da177e4 | 2002 | } |
19a75d83 | 2003 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
2004 | |
2005 | int __init blk_dev_init(void) | |
2006 | { | |
ff856bad JA |
2007 | int i; |
2008 | ||
1da177e4 LT |
2009 | kblockd_workqueue = create_workqueue("kblockd"); |
2010 | if (!kblockd_workqueue) | |
2011 | panic("Failed to create kblockd\n"); | |
2012 | ||
2013 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2014 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2015 | |
8324aa91 | 2016 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2017 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2018 | |
0a945022 | 2019 | for_each_possible_cpu(i) |
ff856bad JA |
2020 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
2021 | ||
2022 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 2023 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 2024 | |
d38ecf93 | 2025 | return 0; |
1da177e4 | 2026 | } |
1da177e4 | 2027 |