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> |
c17bb495 | 29 | #include <linux/fault-inject.h> |
73c10101 | 30 | #include <linux/list_sort.h> |
55782138 LZ |
31 | |
32 | #define CREATE_TRACE_POINTS | |
33 | #include <trace/events/block.h> | |
1da177e4 | 34 | |
8324aa91 JA |
35 | #include "blk.h" |
36 | ||
d07335e5 | 37 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
b0da3f0d | 38 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
55782138 | 39 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
0bfc2455 | 40 | |
165125e1 | 41 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
42 | |
43 | /* | |
44 | * For the allocated request tables | |
45 | */ | |
5ece6c52 | 46 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
47 | |
48 | /* | |
49 | * For queue allocation | |
50 | */ | |
6728cb0e | 51 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 52 | |
1da177e4 LT |
53 | /* |
54 | * Controlling structure to kblockd | |
55 | */ | |
ff856bad | 56 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 57 | |
26b8256e JA |
58 | static void drive_stat_acct(struct request *rq, int new_io) |
59 | { | |
28f13702 | 60 | struct hd_struct *part; |
26b8256e | 61 | int rw = rq_data_dir(rq); |
c9959059 | 62 | int cpu; |
26b8256e | 63 | |
c2553b58 | 64 | if (!blk_do_io_stat(rq)) |
26b8256e JA |
65 | return; |
66 | ||
074a7aca | 67 | cpu = part_stat_lock(); |
c9959059 | 68 | |
09e099d4 JM |
69 | if (!new_io) { |
70 | part = rq->part; | |
074a7aca | 71 | part_stat_inc(cpu, part, merges[rw]); |
09e099d4 JM |
72 | } else { |
73 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); | |
6c23a968 | 74 | if (!hd_struct_try_get(part)) { |
09e099d4 JM |
75 | /* |
76 | * The partition is already being removed, | |
77 | * the request will be accounted on the disk only | |
78 | * | |
79 | * We take a reference on disk->part0 although that | |
80 | * partition will never be deleted, so we can treat | |
81 | * it as any other partition. | |
82 | */ | |
83 | part = &rq->rq_disk->part0; | |
6c23a968 | 84 | hd_struct_get(part); |
09e099d4 | 85 | } |
074a7aca | 86 | part_round_stats(cpu, part); |
316d315b | 87 | part_inc_in_flight(part, rw); |
09e099d4 | 88 | rq->part = part; |
26b8256e | 89 | } |
e71bf0d0 | 90 | |
074a7aca | 91 | part_stat_unlock(); |
26b8256e JA |
92 | } |
93 | ||
8324aa91 | 94 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
95 | { |
96 | int nr; | |
97 | ||
98 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
99 | if (nr > q->nr_requests) | |
100 | nr = q->nr_requests; | |
101 | q->nr_congestion_on = nr; | |
102 | ||
103 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
104 | if (nr < 1) | |
105 | nr = 1; | |
106 | q->nr_congestion_off = nr; | |
107 | } | |
108 | ||
1da177e4 LT |
109 | /** |
110 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
111 | * @bdev: device | |
112 | * | |
113 | * Locates the passed device's request queue and returns the address of its | |
114 | * backing_dev_info | |
115 | * | |
116 | * Will return NULL if the request queue cannot be located. | |
117 | */ | |
118 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
119 | { | |
120 | struct backing_dev_info *ret = NULL; | |
165125e1 | 121 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
122 | |
123 | if (q) | |
124 | ret = &q->backing_dev_info; | |
125 | return ret; | |
126 | } | |
1da177e4 LT |
127 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
128 | ||
2a4aa30c | 129 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 130 | { |
1afb20f3 FT |
131 | memset(rq, 0, sizeof(*rq)); |
132 | ||
1da177e4 | 133 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 134 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 135 | rq->cpu = -1; |
63a71386 | 136 | rq->q = q; |
a2dec7b3 | 137 | rq->__sector = (sector_t) -1; |
2e662b65 JA |
138 | INIT_HLIST_NODE(&rq->hash); |
139 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 140 | rq->cmd = rq->__cmd; |
e2494e1b | 141 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 142 | rq->tag = -1; |
1da177e4 | 143 | rq->ref_count = 1; |
b243ddcb | 144 | rq->start_time = jiffies; |
9195291e | 145 | set_start_time_ns(rq); |
09e099d4 | 146 | rq->part = NULL; |
1da177e4 | 147 | } |
2a4aa30c | 148 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 149 | |
5bb23a68 N |
150 | static void req_bio_endio(struct request *rq, struct bio *bio, |
151 | unsigned int nbytes, int error) | |
1da177e4 | 152 | { |
143a87f4 TH |
153 | if (error) |
154 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
155 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
156 | error = -EIO; | |
797e7dbb | 157 | |
143a87f4 TH |
158 | if (unlikely(nbytes > bio->bi_size)) { |
159 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", | |
160 | __func__, nbytes, bio->bi_size); | |
161 | nbytes = bio->bi_size; | |
5bb23a68 | 162 | } |
797e7dbb | 163 | |
143a87f4 TH |
164 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
165 | set_bit(BIO_QUIET, &bio->bi_flags); | |
08bafc03 | 166 | |
143a87f4 TH |
167 | bio->bi_size -= nbytes; |
168 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 | 169 | |
143a87f4 TH |
170 | if (bio_integrity(bio)) |
171 | bio_integrity_advance(bio, nbytes); | |
7ba1ba12 | 172 | |
143a87f4 TH |
173 | /* don't actually finish bio if it's part of flush sequence */ |
174 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) | |
175 | bio_endio(bio, error); | |
1da177e4 | 176 | } |
1da177e4 | 177 | |
1da177e4 LT |
178 | void blk_dump_rq_flags(struct request *rq, char *msg) |
179 | { | |
180 | int bit; | |
181 | ||
6728cb0e | 182 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
183 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
184 | rq->cmd_flags); | |
1da177e4 | 185 | |
83096ebf TH |
186 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
187 | (unsigned long long)blk_rq_pos(rq), | |
188 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); | |
731ec497 | 189 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
2e46e8b2 | 190 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
1da177e4 | 191 | |
33659ebb | 192 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
6728cb0e | 193 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 194 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
195 | printk("%02x ", rq->cmd[bit]); |
196 | printk("\n"); | |
197 | } | |
198 | } | |
1da177e4 LT |
199 | EXPORT_SYMBOL(blk_dump_rq_flags); |
200 | ||
1da177e4 | 201 | /* |
7eaceacc JA |
202 | * Make sure that plugs that were pending when this function was entered, |
203 | * are now complete and requests pushed to the queue. | |
204 | */ | |
205 | static inline void queue_sync_plugs(struct request_queue *q) | |
1da177e4 | 206 | { |
1da177e4 | 207 | /* |
7eaceacc JA |
208 | * If the current process is plugged and has barriers submitted, |
209 | * we will livelock if we don't unplug first. | |
1da177e4 | 210 | */ |
7eaceacc | 211 | blk_flush_plug(current); |
1da177e4 | 212 | } |
1da177e4 | 213 | |
3cca6dc1 | 214 | static void blk_delay_work(struct work_struct *work) |
1da177e4 | 215 | { |
3cca6dc1 | 216 | struct request_queue *q; |
1da177e4 | 217 | |
3cca6dc1 JA |
218 | q = container_of(work, struct request_queue, delay_work.work); |
219 | spin_lock_irq(q->queue_lock); | |
4c63f564 | 220 | __blk_run_queue(q, false); |
3cca6dc1 | 221 | spin_unlock_irq(q->queue_lock); |
1da177e4 | 222 | } |
1da177e4 LT |
223 | |
224 | /** | |
3cca6dc1 JA |
225 | * blk_delay_queue - restart queueing after defined interval |
226 | * @q: The &struct request_queue in question | |
227 | * @msecs: Delay in msecs | |
1da177e4 LT |
228 | * |
229 | * Description: | |
3cca6dc1 JA |
230 | * Sometimes queueing needs to be postponed for a little while, to allow |
231 | * resources to come back. This function will make sure that queueing is | |
232 | * restarted around the specified time. | |
233 | */ | |
234 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) | |
2ad8b1ef | 235 | { |
3cca6dc1 | 236 | schedule_delayed_work(&q->delay_work, msecs_to_jiffies(msecs)); |
2ad8b1ef | 237 | } |
3cca6dc1 | 238 | EXPORT_SYMBOL(blk_delay_queue); |
2ad8b1ef | 239 | |
1da177e4 LT |
240 | /** |
241 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 242 | * @q: The &struct request_queue in question |
1da177e4 LT |
243 | * |
244 | * Description: | |
245 | * blk_start_queue() will clear the stop flag on the queue, and call | |
246 | * the request_fn for the queue if it was in a stopped state when | |
247 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
248 | **/ | |
165125e1 | 249 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 250 | { |
a038e253 PBG |
251 | WARN_ON(!irqs_disabled()); |
252 | ||
75ad23bc | 253 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
1654e741 | 254 | __blk_run_queue(q, false); |
1da177e4 | 255 | } |
1da177e4 LT |
256 | EXPORT_SYMBOL(blk_start_queue); |
257 | ||
258 | /** | |
259 | * blk_stop_queue - stop a queue | |
165125e1 | 260 | * @q: The &struct request_queue in question |
1da177e4 LT |
261 | * |
262 | * Description: | |
263 | * The Linux block layer assumes that a block driver will consume all | |
264 | * entries on the request queue when the request_fn strategy is called. | |
265 | * Often this will not happen, because of hardware limitations (queue | |
266 | * depth settings). If a device driver gets a 'queue full' response, | |
267 | * or if it simply chooses not to queue more I/O at one point, it can | |
268 | * call this function to prevent the request_fn from being called until | |
269 | * the driver has signalled it's ready to go again. This happens by calling | |
270 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
271 | **/ | |
165125e1 | 272 | void blk_stop_queue(struct request_queue *q) |
1da177e4 | 273 | { |
ad3d9d7e | 274 | __cancel_delayed_work(&q->delay_work); |
75ad23bc | 275 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
276 | } |
277 | EXPORT_SYMBOL(blk_stop_queue); | |
278 | ||
279 | /** | |
280 | * blk_sync_queue - cancel any pending callbacks on a queue | |
281 | * @q: the queue | |
282 | * | |
283 | * Description: | |
284 | * The block layer may perform asynchronous callback activity | |
285 | * on a queue, such as calling the unplug function after a timeout. | |
286 | * A block device may call blk_sync_queue to ensure that any | |
287 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 288 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
289 | * that its ->make_request_fn will not re-add plugging prior to calling |
290 | * this function. | |
291 | * | |
da527770 VG |
292 | * This function does not cancel any asynchronous activity arising |
293 | * out of elevator or throttling code. That would require elevaotor_exit() | |
294 | * and blk_throtl_exit() to be called with queue lock initialized. | |
295 | * | |
1da177e4 LT |
296 | */ |
297 | void blk_sync_queue(struct request_queue *q) | |
298 | { | |
70ed28b9 | 299 | del_timer_sync(&q->timeout); |
3cca6dc1 | 300 | cancel_delayed_work_sync(&q->delay_work); |
7eaceacc | 301 | queue_sync_plugs(q); |
1da177e4 LT |
302 | } |
303 | EXPORT_SYMBOL(blk_sync_queue); | |
304 | ||
305 | /** | |
80a4b58e | 306 | * __blk_run_queue - run a single device queue |
1da177e4 | 307 | * @q: The queue to run |
1654e741 | 308 | * @force_kblockd: Don't run @q->request_fn directly. Use kblockd. |
80a4b58e JA |
309 | * |
310 | * Description: | |
311 | * See @blk_run_queue. This variant must be called with the queue lock | |
312 | * held and interrupts disabled. | |
313 | * | |
1da177e4 | 314 | */ |
1654e741 | 315 | void __blk_run_queue(struct request_queue *q, bool force_kblockd) |
1da177e4 | 316 | { |
a538cd03 TH |
317 | if (unlikely(blk_queue_stopped(q))) |
318 | return; | |
319 | ||
dac07ec1 JA |
320 | /* |
321 | * Only recurse once to avoid overrunning the stack, let the unplug | |
322 | * handling reinvoke the handler shortly if we already got there. | |
323 | */ | |
1654e741 | 324 | if (!force_kblockd && !queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { |
a538cd03 TH |
325 | q->request_fn(q); |
326 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
7eaceacc JA |
327 | } else |
328 | queue_delayed_work(kblockd_workqueue, &q->delay_work, 0); | |
75ad23bc NP |
329 | } |
330 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 331 | |
75ad23bc NP |
332 | /** |
333 | * blk_run_queue - run a single device queue | |
334 | * @q: The queue to run | |
80a4b58e JA |
335 | * |
336 | * Description: | |
337 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 338 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
339 | */ |
340 | void blk_run_queue(struct request_queue *q) | |
341 | { | |
342 | unsigned long flags; | |
343 | ||
344 | spin_lock_irqsave(q->queue_lock, flags); | |
1654e741 | 345 | __blk_run_queue(q, false); |
1da177e4 LT |
346 | spin_unlock_irqrestore(q->queue_lock, flags); |
347 | } | |
348 | EXPORT_SYMBOL(blk_run_queue); | |
349 | ||
165125e1 | 350 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
351 | { |
352 | kobject_put(&q->kobj); | |
353 | } | |
483f4afc | 354 | |
c94a96ac VG |
355 | /* |
356 | * Note: If a driver supplied the queue lock, it should not zap that lock | |
357 | * unexpectedly as some queue cleanup components like elevator_exit() and | |
358 | * blk_throtl_exit() need queue lock. | |
359 | */ | |
6728cb0e | 360 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 361 | { |
e3335de9 JA |
362 | /* |
363 | * We know we have process context here, so we can be a little | |
364 | * cautious and ensure that pending block actions on this device | |
365 | * are done before moving on. Going into this function, we should | |
366 | * not have processes doing IO to this device. | |
367 | */ | |
368 | blk_sync_queue(q); | |
369 | ||
31373d09 | 370 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); |
483f4afc | 371 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 372 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
373 | mutex_unlock(&q->sysfs_lock); |
374 | ||
375 | if (q->elevator) | |
376 | elevator_exit(q->elevator); | |
377 | ||
da527770 VG |
378 | blk_throtl_exit(q); |
379 | ||
483f4afc AV |
380 | blk_put_queue(q); |
381 | } | |
1da177e4 LT |
382 | EXPORT_SYMBOL(blk_cleanup_queue); |
383 | ||
165125e1 | 384 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
385 | { |
386 | struct request_list *rl = &q->rq; | |
387 | ||
1abec4fd MS |
388 | if (unlikely(rl->rq_pool)) |
389 | return 0; | |
390 | ||
1faa16d2 JA |
391 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
392 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 393 | rl->elvpriv = 0; |
1faa16d2 JA |
394 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
395 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 396 | |
1946089a CL |
397 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
398 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
399 | |
400 | if (!rl->rq_pool) | |
401 | return -ENOMEM; | |
402 | ||
403 | return 0; | |
404 | } | |
405 | ||
165125e1 | 406 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 407 | { |
1946089a CL |
408 | return blk_alloc_queue_node(gfp_mask, -1); |
409 | } | |
410 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 411 | |
165125e1 | 412 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 413 | { |
165125e1 | 414 | struct request_queue *q; |
e0bf68dd | 415 | int err; |
1946089a | 416 | |
8324aa91 | 417 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 418 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
419 | if (!q) |
420 | return NULL; | |
421 | ||
0989a025 JA |
422 | q->backing_dev_info.ra_pages = |
423 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
424 | q->backing_dev_info.state = 0; | |
425 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
d993831f | 426 | q->backing_dev_info.name = "block"; |
0989a025 | 427 | |
e0bf68dd PZ |
428 | err = bdi_init(&q->backing_dev_info); |
429 | if (err) { | |
8324aa91 | 430 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
431 | return NULL; |
432 | } | |
433 | ||
e43473b7 VG |
434 | if (blk_throtl_init(q)) { |
435 | kmem_cache_free(blk_requestq_cachep, q); | |
436 | return NULL; | |
437 | } | |
438 | ||
31373d09 MG |
439 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
440 | laptop_mode_timer_fn, (unsigned long) q); | |
242f9dcb JA |
441 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
442 | INIT_LIST_HEAD(&q->timeout_list); | |
ae1b1539 TH |
443 | INIT_LIST_HEAD(&q->flush_queue[0]); |
444 | INIT_LIST_HEAD(&q->flush_queue[1]); | |
445 | INIT_LIST_HEAD(&q->flush_data_in_flight); | |
3cca6dc1 | 446 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
483f4afc | 447 | |
8324aa91 | 448 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 449 | |
483f4afc | 450 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 451 | spin_lock_init(&q->__queue_lock); |
483f4afc | 452 | |
c94a96ac VG |
453 | /* |
454 | * By default initialize queue_lock to internal lock and driver can | |
455 | * override it later if need be. | |
456 | */ | |
457 | q->queue_lock = &q->__queue_lock; | |
458 | ||
1da177e4 LT |
459 | return q; |
460 | } | |
1946089a | 461 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
462 | |
463 | /** | |
464 | * blk_init_queue - prepare a request queue for use with a block device | |
465 | * @rfn: The function to be called to process requests that have been | |
466 | * placed on the queue. | |
467 | * @lock: Request queue spin lock | |
468 | * | |
469 | * Description: | |
470 | * If a block device wishes to use the standard request handling procedures, | |
471 | * which sorts requests and coalesces adjacent requests, then it must | |
472 | * call blk_init_queue(). The function @rfn will be called when there | |
473 | * are requests on the queue that need to be processed. If the device | |
474 | * supports plugging, then @rfn may not be called immediately when requests | |
475 | * are available on the queue, but may be called at some time later instead. | |
476 | * Plugged queues are generally unplugged when a buffer belonging to one | |
477 | * of the requests on the queue is needed, or due to memory pressure. | |
478 | * | |
479 | * @rfn is not required, or even expected, to remove all requests off the | |
480 | * queue, but only as many as it can handle at a time. If it does leave | |
481 | * requests on the queue, it is responsible for arranging that the requests | |
482 | * get dealt with eventually. | |
483 | * | |
484 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
485 | * request queue; this lock will be taken also from interrupt context, so irq |
486 | * disabling is needed for it. | |
1da177e4 | 487 | * |
710027a4 | 488 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
489 | * it didn't succeed. |
490 | * | |
491 | * Note: | |
492 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
493 | * when the block device is deactivated (such as at module unload). | |
494 | **/ | |
1946089a | 495 | |
165125e1 | 496 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 497 | { |
1946089a CL |
498 | return blk_init_queue_node(rfn, lock, -1); |
499 | } | |
500 | EXPORT_SYMBOL(blk_init_queue); | |
501 | ||
165125e1 | 502 | struct request_queue * |
1946089a CL |
503 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
504 | { | |
c86d1b8a | 505 | struct request_queue *uninit_q, *q; |
1da177e4 | 506 | |
c86d1b8a MS |
507 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
508 | if (!uninit_q) | |
509 | return NULL; | |
510 | ||
511 | q = blk_init_allocated_queue_node(uninit_q, rfn, lock, node_id); | |
512 | if (!q) | |
513 | blk_cleanup_queue(uninit_q); | |
514 | ||
515 | return q; | |
01effb0d MS |
516 | } |
517 | EXPORT_SYMBOL(blk_init_queue_node); | |
518 | ||
519 | struct request_queue * | |
520 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, | |
521 | spinlock_t *lock) | |
522 | { | |
523 | return blk_init_allocated_queue_node(q, rfn, lock, -1); | |
524 | } | |
525 | EXPORT_SYMBOL(blk_init_allocated_queue); | |
526 | ||
527 | struct request_queue * | |
528 | blk_init_allocated_queue_node(struct request_queue *q, request_fn_proc *rfn, | |
529 | spinlock_t *lock, int node_id) | |
530 | { | |
1da177e4 LT |
531 | if (!q) |
532 | return NULL; | |
533 | ||
1946089a | 534 | q->node = node_id; |
c86d1b8a | 535 | if (blk_init_free_list(q)) |
8669aafd | 536 | return NULL; |
1da177e4 LT |
537 | |
538 | q->request_fn = rfn; | |
1da177e4 | 539 | q->prep_rq_fn = NULL; |
28018c24 | 540 | q->unprep_rq_fn = NULL; |
bc58ba94 | 541 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
c94a96ac VG |
542 | |
543 | /* Override internal queue lock with supplied lock pointer */ | |
544 | if (lock) | |
545 | q->queue_lock = lock; | |
1da177e4 | 546 | |
f3b144aa JA |
547 | /* |
548 | * This also sets hw/phys segments, boundary and size | |
549 | */ | |
1da177e4 | 550 | blk_queue_make_request(q, __make_request); |
1da177e4 | 551 | |
44ec9542 AS |
552 | q->sg_reserved_size = INT_MAX; |
553 | ||
1da177e4 LT |
554 | /* |
555 | * all done | |
556 | */ | |
557 | if (!elevator_init(q, NULL)) { | |
558 | blk_queue_congestion_threshold(q); | |
559 | return q; | |
560 | } | |
561 | ||
1da177e4 LT |
562 | return NULL; |
563 | } | |
01effb0d | 564 | EXPORT_SYMBOL(blk_init_allocated_queue_node); |
1da177e4 | 565 | |
165125e1 | 566 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 567 | { |
fde6ad22 | 568 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 569 | kobject_get(&q->kobj); |
1da177e4 LT |
570 | return 0; |
571 | } | |
572 | ||
573 | return 1; | |
574 | } | |
1da177e4 | 575 | |
165125e1 | 576 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 577 | { |
73c10101 JA |
578 | BUG_ON(rq->cmd_flags & REQ_ON_PLUG); |
579 | ||
4aff5e23 | 580 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 581 | elv_put_request(q, rq); |
1da177e4 LT |
582 | mempool_free(rq, q->rq.rq_pool); |
583 | } | |
584 | ||
1ea25ecb | 585 | static struct request * |
42dad764 | 586 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) |
1da177e4 LT |
587 | { |
588 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
589 | ||
590 | if (!rq) | |
591 | return NULL; | |
592 | ||
2a4aa30c | 593 | blk_rq_init(q, rq); |
1afb20f3 | 594 | |
42dad764 | 595 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 596 | |
cb98fc8b | 597 | if (priv) { |
cb78b285 | 598 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
599 | mempool_free(rq, q->rq.rq_pool); |
600 | return NULL; | |
601 | } | |
4aff5e23 | 602 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 603 | } |
1da177e4 | 604 | |
cb98fc8b | 605 | return rq; |
1da177e4 LT |
606 | } |
607 | ||
608 | /* | |
609 | * ioc_batching returns true if the ioc is a valid batching request and | |
610 | * should be given priority access to a request. | |
611 | */ | |
165125e1 | 612 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
613 | { |
614 | if (!ioc) | |
615 | return 0; | |
616 | ||
617 | /* | |
618 | * Make sure the process is able to allocate at least 1 request | |
619 | * even if the batch times out, otherwise we could theoretically | |
620 | * lose wakeups. | |
621 | */ | |
622 | return ioc->nr_batch_requests == q->nr_batching || | |
623 | (ioc->nr_batch_requests > 0 | |
624 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
625 | } | |
626 | ||
627 | /* | |
628 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
629 | * will cause the process to be a "batcher" on all queues in the system. This | |
630 | * is the behaviour we want though - once it gets a wakeup it should be given | |
631 | * a nice run. | |
632 | */ | |
165125e1 | 633 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
634 | { |
635 | if (!ioc || ioc_batching(q, ioc)) | |
636 | return; | |
637 | ||
638 | ioc->nr_batch_requests = q->nr_batching; | |
639 | ioc->last_waited = jiffies; | |
640 | } | |
641 | ||
1faa16d2 | 642 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
643 | { |
644 | struct request_list *rl = &q->rq; | |
645 | ||
1faa16d2 JA |
646 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
647 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 648 | |
1faa16d2 JA |
649 | if (rl->count[sync] + 1 <= q->nr_requests) { |
650 | if (waitqueue_active(&rl->wait[sync])) | |
651 | wake_up(&rl->wait[sync]); | |
1da177e4 | 652 | |
1faa16d2 | 653 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
654 | } |
655 | } | |
656 | ||
657 | /* | |
658 | * A request has just been released. Account for it, update the full and | |
659 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
660 | */ | |
1faa16d2 | 661 | static void freed_request(struct request_queue *q, int sync, int priv) |
1da177e4 LT |
662 | { |
663 | struct request_list *rl = &q->rq; | |
664 | ||
1faa16d2 | 665 | rl->count[sync]--; |
cb98fc8b TH |
666 | if (priv) |
667 | rl->elvpriv--; | |
1da177e4 | 668 | |
1faa16d2 | 669 | __freed_request(q, sync); |
1da177e4 | 670 | |
1faa16d2 JA |
671 | if (unlikely(rl->starved[sync ^ 1])) |
672 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
673 | } |
674 | ||
9d5a4e94 MS |
675 | /* |
676 | * Determine if elevator data should be initialized when allocating the | |
677 | * request associated with @bio. | |
678 | */ | |
679 | static bool blk_rq_should_init_elevator(struct bio *bio) | |
680 | { | |
681 | if (!bio) | |
682 | return true; | |
683 | ||
684 | /* | |
685 | * Flush requests do not use the elevator so skip initialization. | |
686 | * This allows a request to share the flush and elevator data. | |
687 | */ | |
688 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) | |
689 | return false; | |
690 | ||
691 | return true; | |
692 | } | |
693 | ||
1da177e4 | 694 | /* |
d6344532 NP |
695 | * Get a free request, queue_lock must be held. |
696 | * Returns NULL on failure, with queue_lock held. | |
697 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 698 | */ |
165125e1 | 699 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 700 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
701 | { |
702 | struct request *rq = NULL; | |
703 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 704 | struct io_context *ioc = NULL; |
1faa16d2 | 705 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
9d5a4e94 | 706 | int may_queue, priv = 0; |
88ee5ef1 | 707 | |
7749a8d4 | 708 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
709 | if (may_queue == ELV_MQUEUE_NO) |
710 | goto rq_starved; | |
711 | ||
1faa16d2 JA |
712 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
713 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 714 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
715 | /* |
716 | * The queue will fill after this allocation, so set | |
717 | * it as full, and mark this process as "batching". | |
718 | * This process will be allowed to complete a batch of | |
719 | * requests, others will be blocked. | |
720 | */ | |
1faa16d2 | 721 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 722 | ioc_set_batching(q, ioc); |
1faa16d2 | 723 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
724 | } else { |
725 | if (may_queue != ELV_MQUEUE_MUST | |
726 | && !ioc_batching(q, ioc)) { | |
727 | /* | |
728 | * The queue is full and the allocating | |
729 | * process is not a "batcher", and not | |
730 | * exempted by the IO scheduler | |
731 | */ | |
732 | goto out; | |
733 | } | |
734 | } | |
1da177e4 | 735 | } |
1faa16d2 | 736 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
737 | } |
738 | ||
082cf69e JA |
739 | /* |
740 | * Only allow batching queuers to allocate up to 50% over the defined | |
741 | * limit of requests, otherwise we could have thousands of requests | |
742 | * allocated with any setting of ->nr_requests | |
743 | */ | |
1faa16d2 | 744 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 745 | goto out; |
fd782a4a | 746 | |
1faa16d2 JA |
747 | rl->count[is_sync]++; |
748 | rl->starved[is_sync] = 0; | |
cb98fc8b | 749 | |
9d5a4e94 MS |
750 | if (blk_rq_should_init_elevator(bio)) { |
751 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | |
752 | if (priv) | |
753 | rl->elvpriv++; | |
754 | } | |
cb98fc8b | 755 | |
f253b86b JA |
756 | if (blk_queue_io_stat(q)) |
757 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
758 | spin_unlock_irq(q->queue_lock); |
759 | ||
7749a8d4 | 760 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 761 | if (unlikely(!rq)) { |
1da177e4 LT |
762 | /* |
763 | * Allocation failed presumably due to memory. Undo anything | |
764 | * we might have messed up. | |
765 | * | |
766 | * Allocating task should really be put onto the front of the | |
767 | * wait queue, but this is pretty rare. | |
768 | */ | |
769 | spin_lock_irq(q->queue_lock); | |
1faa16d2 | 770 | freed_request(q, is_sync, priv); |
1da177e4 LT |
771 | |
772 | /* | |
773 | * in the very unlikely event that allocation failed and no | |
774 | * requests for this direction was pending, mark us starved | |
775 | * so that freeing of a request in the other direction will | |
776 | * notice us. another possible fix would be to split the | |
777 | * rq mempool into READ and WRITE | |
778 | */ | |
779 | rq_starved: | |
1faa16d2 JA |
780 | if (unlikely(rl->count[is_sync] == 0)) |
781 | rl->starved[is_sync] = 1; | |
1da177e4 | 782 | |
1da177e4 LT |
783 | goto out; |
784 | } | |
785 | ||
88ee5ef1 JA |
786 | /* |
787 | * ioc may be NULL here, and ioc_batching will be false. That's | |
788 | * OK, if the queue is under the request limit then requests need | |
789 | * not count toward the nr_batch_requests limit. There will always | |
790 | * be some limit enforced by BLK_BATCH_TIME. | |
791 | */ | |
1da177e4 LT |
792 | if (ioc_batching(q, ioc)) |
793 | ioc->nr_batch_requests--; | |
6728cb0e | 794 | |
1faa16d2 | 795 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 796 | out: |
1da177e4 LT |
797 | return rq; |
798 | } | |
799 | ||
800 | /* | |
7eaceacc JA |
801 | * No available requests for this queue, wait for some requests to become |
802 | * available. | |
d6344532 NP |
803 | * |
804 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 805 | */ |
165125e1 | 806 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 807 | struct bio *bio) |
1da177e4 | 808 | { |
1faa16d2 | 809 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
810 | struct request *rq; |
811 | ||
7749a8d4 | 812 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
813 | while (!rq) { |
814 | DEFINE_WAIT(wait); | |
05caf8db | 815 | struct io_context *ioc; |
1da177e4 LT |
816 | struct request_list *rl = &q->rq; |
817 | ||
1faa16d2 | 818 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
819 | TASK_UNINTERRUPTIBLE); |
820 | ||
1faa16d2 | 821 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 822 | |
05caf8db ZY |
823 | spin_unlock_irq(q->queue_lock); |
824 | io_schedule(); | |
1da177e4 | 825 | |
05caf8db ZY |
826 | /* |
827 | * After sleeping, we become a "batching" process and | |
828 | * will be able to allocate at least one request, and | |
829 | * up to a big batch of them for a small period time. | |
830 | * See ioc_batching, ioc_set_batching | |
831 | */ | |
832 | ioc = current_io_context(GFP_NOIO, q->node); | |
833 | ioc_set_batching(q, ioc); | |
d6344532 | 834 | |
05caf8db | 835 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 836 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
837 | |
838 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
839 | }; | |
1da177e4 LT |
840 | |
841 | return rq; | |
842 | } | |
843 | ||
165125e1 | 844 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
845 | { |
846 | struct request *rq; | |
847 | ||
848 | BUG_ON(rw != READ && rw != WRITE); | |
849 | ||
d6344532 NP |
850 | spin_lock_irq(q->queue_lock); |
851 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 852 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 853 | } else { |
22e2c507 | 854 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
855 | if (!rq) |
856 | spin_unlock_irq(q->queue_lock); | |
857 | } | |
858 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
859 | |
860 | return rq; | |
861 | } | |
1da177e4 LT |
862 | EXPORT_SYMBOL(blk_get_request); |
863 | ||
dc72ef4a | 864 | /** |
79eb63e9 | 865 | * blk_make_request - given a bio, allocate a corresponding struct request. |
8ebf9756 | 866 | * @q: target request queue |
79eb63e9 BH |
867 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
868 | * It may be a chained-bio properly constructed by block/bio layer. | |
8ebf9756 | 869 | * @gfp_mask: gfp flags to be used for memory allocation |
dc72ef4a | 870 | * |
79eb63e9 BH |
871 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
872 | * type commands. Where the struct request needs to be farther initialized by | |
873 | * the caller. It is passed a &struct bio, which describes the memory info of | |
874 | * the I/O transfer. | |
dc72ef4a | 875 | * |
79eb63e9 BH |
876 | * The caller of blk_make_request must make sure that bi_io_vec |
877 | * are set to describe the memory buffers. That bio_data_dir() will return | |
878 | * the needed direction of the request. (And all bio's in the passed bio-chain | |
879 | * are properly set accordingly) | |
880 | * | |
881 | * If called under none-sleepable conditions, mapped bio buffers must not | |
882 | * need bouncing, by calling the appropriate masked or flagged allocator, | |
883 | * suitable for the target device. Otherwise the call to blk_queue_bounce will | |
884 | * BUG. | |
53674ac5 JA |
885 | * |
886 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be | |
887 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for | |
888 | * anything but the first bio in the chain. Otherwise you risk waiting for IO | |
889 | * completion of a bio that hasn't been submitted yet, thus resulting in a | |
890 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead | |
891 | * of bio_alloc(), as that avoids the mempool deadlock. | |
892 | * If possible a big IO should be split into smaller parts when allocation | |
893 | * fails. Partial allocation should not be an error, or you risk a live-lock. | |
dc72ef4a | 894 | */ |
79eb63e9 BH |
895 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
896 | gfp_t gfp_mask) | |
dc72ef4a | 897 | { |
79eb63e9 BH |
898 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
899 | ||
900 | if (unlikely(!rq)) | |
901 | return ERR_PTR(-ENOMEM); | |
902 | ||
903 | for_each_bio(bio) { | |
904 | struct bio *bounce_bio = bio; | |
905 | int ret; | |
906 | ||
907 | blk_queue_bounce(q, &bounce_bio); | |
908 | ret = blk_rq_append_bio(q, rq, bounce_bio); | |
909 | if (unlikely(ret)) { | |
910 | blk_put_request(rq); | |
911 | return ERR_PTR(ret); | |
912 | } | |
913 | } | |
914 | ||
915 | return rq; | |
dc72ef4a | 916 | } |
79eb63e9 | 917 | EXPORT_SYMBOL(blk_make_request); |
dc72ef4a | 918 | |
1da177e4 LT |
919 | /** |
920 | * blk_requeue_request - put a request back on queue | |
921 | * @q: request queue where request should be inserted | |
922 | * @rq: request to be inserted | |
923 | * | |
924 | * Description: | |
925 | * Drivers often keep queueing requests until the hardware cannot accept | |
926 | * more, when that condition happens we need to put the request back | |
927 | * on the queue. Must be called with queue lock held. | |
928 | */ | |
165125e1 | 929 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 930 | { |
242f9dcb JA |
931 | blk_delete_timer(rq); |
932 | blk_clear_rq_complete(rq); | |
5f3ea37c | 933 | trace_block_rq_requeue(q, rq); |
2056a782 | 934 | |
1da177e4 LT |
935 | if (blk_rq_tagged(rq)) |
936 | blk_queue_end_tag(q, rq); | |
937 | ||
ba396a6c JB |
938 | BUG_ON(blk_queued_rq(rq)); |
939 | ||
1da177e4 LT |
940 | elv_requeue_request(q, rq); |
941 | } | |
1da177e4 LT |
942 | EXPORT_SYMBOL(blk_requeue_request); |
943 | ||
73c10101 JA |
944 | static void add_acct_request(struct request_queue *q, struct request *rq, |
945 | int where) | |
946 | { | |
947 | drive_stat_acct(rq, 1); | |
7eaceacc | 948 | __elv_add_request(q, rq, where); |
73c10101 JA |
949 | } |
950 | ||
1da177e4 | 951 | /** |
710027a4 | 952 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
953 | * @q: request queue where request should be inserted |
954 | * @rq: request to be inserted | |
955 | * @at_head: insert request at head or tail of queue | |
956 | * @data: private data | |
1da177e4 LT |
957 | * |
958 | * Description: | |
959 | * Many block devices need to execute commands asynchronously, so they don't | |
960 | * block the whole kernel from preemption during request execution. This is | |
961 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
962 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
963 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
964 | * |
965 | * We have the option of inserting the head or the tail of the queue. | |
966 | * Typically we use the tail for new ioctls and so forth. We use the head | |
967 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
968 | * host that is unable to accept a particular command. | |
969 | */ | |
165125e1 | 970 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 971 | int at_head, void *data) |
1da177e4 | 972 | { |
867d1191 | 973 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
974 | unsigned long flags; |
975 | ||
976 | /* | |
977 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
978 | * must not attempt merges on this) and that it acts as a soft | |
979 | * barrier | |
980 | */ | |
4aff5e23 | 981 | rq->cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
982 | |
983 | rq->special = data; | |
984 | ||
985 | spin_lock_irqsave(q->queue_lock, flags); | |
986 | ||
987 | /* | |
988 | * If command is tagged, release the tag | |
989 | */ | |
867d1191 TH |
990 | if (blk_rq_tagged(rq)) |
991 | blk_queue_end_tag(q, rq); | |
1da177e4 | 992 | |
73c10101 | 993 | add_acct_request(q, rq, where); |
1654e741 | 994 | __blk_run_queue(q, false); |
1da177e4 LT |
995 | spin_unlock_irqrestore(q->queue_lock, flags); |
996 | } | |
1da177e4 LT |
997 | EXPORT_SYMBOL(blk_insert_request); |
998 | ||
074a7aca TH |
999 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
1000 | unsigned long now) | |
1001 | { | |
1002 | if (now == part->stamp) | |
1003 | return; | |
1004 | ||
316d315b | 1005 | if (part_in_flight(part)) { |
074a7aca | 1006 | __part_stat_add(cpu, part, time_in_queue, |
316d315b | 1007 | part_in_flight(part) * (now - part->stamp)); |
074a7aca TH |
1008 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
1009 | } | |
1010 | part->stamp = now; | |
1011 | } | |
1012 | ||
1013 | /** | |
496aa8a9 RD |
1014 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
1015 | * @cpu: cpu number for stats access | |
1016 | * @part: target partition | |
1da177e4 LT |
1017 | * |
1018 | * The average IO queue length and utilisation statistics are maintained | |
1019 | * by observing the current state of the queue length and the amount of | |
1020 | * time it has been in this state for. | |
1021 | * | |
1022 | * Normally, that accounting is done on IO completion, but that can result | |
1023 | * in more than a second's worth of IO being accounted for within any one | |
1024 | * second, leading to >100% utilisation. To deal with that, we call this | |
1025 | * function to do a round-off before returning the results when reading | |
1026 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1027 | * the current jiffies and restarts the counters again. | |
1028 | */ | |
c9959059 | 1029 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1030 | { |
1031 | unsigned long now = jiffies; | |
1032 | ||
074a7aca TH |
1033 | if (part->partno) |
1034 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1035 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1036 | } |
074a7aca | 1037 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1038 | |
1da177e4 LT |
1039 | /* |
1040 | * queue lock must be held | |
1041 | */ | |
165125e1 | 1042 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1043 | { |
1da177e4 LT |
1044 | if (unlikely(!q)) |
1045 | return; | |
1046 | if (unlikely(--req->ref_count)) | |
1047 | return; | |
1048 | ||
8922e16c TH |
1049 | elv_completed_request(q, req); |
1050 | ||
1cd96c24 BH |
1051 | /* this is a bio leak */ |
1052 | WARN_ON(req->bio != NULL); | |
1053 | ||
1da177e4 LT |
1054 | /* |
1055 | * Request may not have originated from ll_rw_blk. if not, | |
1056 | * it didn't come out of our reserved rq pools | |
1057 | */ | |
49171e5c | 1058 | if (req->cmd_flags & REQ_ALLOCED) { |
1faa16d2 | 1059 | int is_sync = rq_is_sync(req) != 0; |
4aff5e23 | 1060 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1061 | |
1da177e4 | 1062 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1063 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1064 | |
1065 | blk_free_request(q, req); | |
1faa16d2 | 1066 | freed_request(q, is_sync, priv); |
1da177e4 LT |
1067 | } |
1068 | } | |
6e39b69e MC |
1069 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1070 | ||
1da177e4 LT |
1071 | void blk_put_request(struct request *req) |
1072 | { | |
8922e16c | 1073 | unsigned long flags; |
165125e1 | 1074 | struct request_queue *q = req->q; |
8922e16c | 1075 | |
52a93ba8 FT |
1076 | spin_lock_irqsave(q->queue_lock, flags); |
1077 | __blk_put_request(q, req); | |
1078 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1079 | } |
1da177e4 LT |
1080 | EXPORT_SYMBOL(blk_put_request); |
1081 | ||
66ac0280 CH |
1082 | /** |
1083 | * blk_add_request_payload - add a payload to a request | |
1084 | * @rq: request to update | |
1085 | * @page: page backing the payload | |
1086 | * @len: length of the payload. | |
1087 | * | |
1088 | * This allows to later add a payload to an already submitted request by | |
1089 | * a block driver. The driver needs to take care of freeing the payload | |
1090 | * itself. | |
1091 | * | |
1092 | * Note that this is a quite horrible hack and nothing but handling of | |
1093 | * discard requests should ever use it. | |
1094 | */ | |
1095 | void blk_add_request_payload(struct request *rq, struct page *page, | |
1096 | unsigned int len) | |
1097 | { | |
1098 | struct bio *bio = rq->bio; | |
1099 | ||
1100 | bio->bi_io_vec->bv_page = page; | |
1101 | bio->bi_io_vec->bv_offset = 0; | |
1102 | bio->bi_io_vec->bv_len = len; | |
1103 | ||
1104 | bio->bi_size = len; | |
1105 | bio->bi_vcnt = 1; | |
1106 | bio->bi_phys_segments = 1; | |
1107 | ||
1108 | rq->__data_len = rq->resid_len = len; | |
1109 | rq->nr_phys_segments = 1; | |
1110 | rq->buffer = bio_data(bio); | |
1111 | } | |
1112 | EXPORT_SYMBOL_GPL(blk_add_request_payload); | |
1113 | ||
73c10101 JA |
1114 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
1115 | struct bio *bio) | |
1116 | { | |
1117 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1118 | ||
1119 | /* | |
1120 | * Debug stuff, kill later | |
1121 | */ | |
1122 | if (!rq_mergeable(req)) { | |
1123 | blk_dump_rq_flags(req, "back"); | |
1124 | return false; | |
1125 | } | |
1126 | ||
1127 | if (!ll_back_merge_fn(q, req, bio)) | |
1128 | return false; | |
1129 | ||
1130 | trace_block_bio_backmerge(q, bio); | |
1131 | ||
1132 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1133 | blk_rq_set_mixed_merge(req); | |
1134 | ||
1135 | req->biotail->bi_next = bio; | |
1136 | req->biotail = bio; | |
1137 | req->__data_len += bio->bi_size; | |
1138 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1139 | ||
1140 | drive_stat_acct(req, 0); | |
1141 | return true; | |
1142 | } | |
1143 | ||
1144 | static bool bio_attempt_front_merge(struct request_queue *q, | |
1145 | struct request *req, struct bio *bio) | |
1146 | { | |
1147 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; | |
1148 | sector_t sector; | |
1149 | ||
1150 | /* | |
1151 | * Debug stuff, kill later | |
1152 | */ | |
1153 | if (!rq_mergeable(req)) { | |
1154 | blk_dump_rq_flags(req, "front"); | |
1155 | return false; | |
1156 | } | |
1157 | ||
1158 | if (!ll_front_merge_fn(q, req, bio)) | |
1159 | return false; | |
1160 | ||
1161 | trace_block_bio_frontmerge(q, bio); | |
1162 | ||
1163 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1164 | blk_rq_set_mixed_merge(req); | |
1165 | ||
1166 | sector = bio->bi_sector; | |
1167 | ||
1168 | bio->bi_next = req->bio; | |
1169 | req->bio = bio; | |
1170 | ||
1171 | /* | |
1172 | * may not be valid. if the low level driver said | |
1173 | * it didn't need a bounce buffer then it better | |
1174 | * not touch req->buffer either... | |
1175 | */ | |
1176 | req->buffer = bio_data(bio); | |
1177 | req->__sector = bio->bi_sector; | |
1178 | req->__data_len += bio->bi_size; | |
1179 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); | |
1180 | ||
1181 | drive_stat_acct(req, 0); | |
1182 | return true; | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * Attempts to merge with the plugged list in the current process. Returns | |
25985edc | 1187 | * true if merge was successful, otherwise false. |
73c10101 JA |
1188 | */ |
1189 | static bool attempt_plug_merge(struct task_struct *tsk, struct request_queue *q, | |
1190 | struct bio *bio) | |
1191 | { | |
1192 | struct blk_plug *plug; | |
1193 | struct request *rq; | |
1194 | bool ret = false; | |
1195 | ||
1196 | plug = tsk->plug; | |
1197 | if (!plug) | |
1198 | goto out; | |
1199 | ||
1200 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { | |
1201 | int el_ret; | |
1202 | ||
1203 | if (rq->q != q) | |
1204 | continue; | |
1205 | ||
1206 | el_ret = elv_try_merge(rq, bio); | |
1207 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
1208 | ret = bio_attempt_back_merge(q, rq, bio); | |
1209 | if (ret) | |
1210 | break; | |
1211 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1212 | ret = bio_attempt_front_merge(q, rq, bio); | |
1213 | if (ret) | |
1214 | break; | |
1215 | } | |
1216 | } | |
1217 | out: | |
1218 | return ret; | |
1219 | } | |
1220 | ||
86db1e29 | 1221 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1222 | { |
c7c22e4d | 1223 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1224 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 | 1225 | |
7b6d91da CH |
1226 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
1227 | if (bio->bi_rw & REQ_RAHEAD) | |
a82afdfc | 1228 | req->cmd_flags |= REQ_FAILFAST_MASK; |
b31dc66a | 1229 | |
52d9e675 | 1230 | req->errors = 0; |
a2dec7b3 | 1231 | req->__sector = bio->bi_sector; |
52d9e675 | 1232 | req->ioprio = bio_prio(bio); |
bc1c56fd | 1233 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1234 | } |
1235 | ||
165125e1 | 1236 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1237 | { |
5e00d1b5 | 1238 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
73c10101 JA |
1239 | struct blk_plug *plug; |
1240 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; | |
1241 | struct request *req; | |
1da177e4 | 1242 | |
1da177e4 LT |
1243 | /* |
1244 | * low level driver can indicate that it wants pages above a | |
1245 | * certain limit bounced to low memory (ie for highmem, or even | |
1246 | * ISA dma in theory) | |
1247 | */ | |
1248 | blk_queue_bounce(q, &bio); | |
1249 | ||
4fed947c | 1250 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
73c10101 | 1251 | spin_lock_irq(q->queue_lock); |
ae1b1539 | 1252 | where = ELEVATOR_INSERT_FLUSH; |
28e7d184 TH |
1253 | goto get_rq; |
1254 | } | |
1255 | ||
73c10101 JA |
1256 | /* |
1257 | * Check if we can merge with the plugged list before grabbing | |
1258 | * any locks. | |
1259 | */ | |
1260 | if (attempt_plug_merge(current, q, bio)) | |
6728cb0e | 1261 | goto out; |
1da177e4 | 1262 | |
73c10101 | 1263 | spin_lock_irq(q->queue_lock); |
2056a782 | 1264 | |
73c10101 JA |
1265 | el_ret = elv_merge(q, &req, bio); |
1266 | if (el_ret == ELEVATOR_BACK_MERGE) { | |
1267 | BUG_ON(req->cmd_flags & REQ_ON_PLUG); | |
1268 | if (bio_attempt_back_merge(q, req, bio)) { | |
1269 | if (!attempt_back_merge(q, req)) | |
1270 | elv_merged_request(q, req, el_ret); | |
1271 | goto out_unlock; | |
1272 | } | |
1273 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { | |
1274 | BUG_ON(req->cmd_flags & REQ_ON_PLUG); | |
1275 | if (bio_attempt_front_merge(q, req, bio)) { | |
1276 | if (!attempt_front_merge(q, req)) | |
1277 | elv_merged_request(q, req, el_ret); | |
1278 | goto out_unlock; | |
80a761fd | 1279 | } |
1da177e4 LT |
1280 | } |
1281 | ||
450991bc | 1282 | get_rq: |
7749a8d4 JA |
1283 | /* |
1284 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1285 | * but we need to set it earlier to expose the sync flag to the | |
1286 | * rq allocator and io schedulers. | |
1287 | */ | |
1288 | rw_flags = bio_data_dir(bio); | |
1289 | if (sync) | |
7b6d91da | 1290 | rw_flags |= REQ_SYNC; |
7749a8d4 | 1291 | |
1da177e4 | 1292 | /* |
450991bc | 1293 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1294 | * Returns with the queue unlocked. |
450991bc | 1295 | */ |
7749a8d4 | 1296 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1297 | |
450991bc NP |
1298 | /* |
1299 | * After dropping the lock and possibly sleeping here, our request | |
1300 | * may now be mergeable after it had proven unmergeable (above). | |
1301 | * We don't worry about that case for efficiency. It won't happen | |
1302 | * often, and the elevators are able to handle it. | |
1da177e4 | 1303 | */ |
52d9e675 | 1304 | init_request_from_bio(req, bio); |
1da177e4 | 1305 | |
c7c22e4d | 1306 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
73c10101 JA |
1307 | bio_flagged(bio, BIO_CPU_AFFINE)) { |
1308 | req->cpu = blk_cpu_to_group(get_cpu()); | |
1309 | put_cpu(); | |
1310 | } | |
1311 | ||
1312 | plug = current->plug; | |
721a9602 | 1313 | if (plug) { |
73c10101 JA |
1314 | if (!plug->should_sort && !list_empty(&plug->list)) { |
1315 | struct request *__rq; | |
1316 | ||
1317 | __rq = list_entry_rq(plug->list.prev); | |
1318 | if (__rq->q != q) | |
1319 | plug->should_sort = 1; | |
1320 | } | |
1321 | /* | |
1322 | * Debug flag, kill later | |
1323 | */ | |
1324 | req->cmd_flags |= REQ_ON_PLUG; | |
1325 | list_add_tail(&req->queuelist, &plug->list); | |
1326 | drive_stat_acct(req, 1); | |
1327 | } else { | |
1328 | spin_lock_irq(q->queue_lock); | |
1329 | add_acct_request(q, req, where); | |
4c63f564 | 1330 | __blk_run_queue(q, false); |
73c10101 JA |
1331 | out_unlock: |
1332 | spin_unlock_irq(q->queue_lock); | |
1333 | } | |
1da177e4 | 1334 | out: |
1da177e4 | 1335 | return 0; |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | /* | |
1339 | * If bio->bi_dev is a partition, remap the location | |
1340 | */ | |
1341 | static inline void blk_partition_remap(struct bio *bio) | |
1342 | { | |
1343 | struct block_device *bdev = bio->bi_bdev; | |
1344 | ||
bf2de6f5 | 1345 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1346 | struct hd_struct *p = bdev->bd_part; |
1347 | ||
1da177e4 LT |
1348 | bio->bi_sector += p->start_sect; |
1349 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1350 | |
d07335e5 MS |
1351 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
1352 | bdev->bd_dev, | |
1353 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1354 | } |
1355 | } | |
1356 | ||
1da177e4 LT |
1357 | static void handle_bad_sector(struct bio *bio) |
1358 | { | |
1359 | char b[BDEVNAME_SIZE]; | |
1360 | ||
1361 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1362 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1363 | bdevname(bio->bi_bdev, b), | |
1364 | bio->bi_rw, | |
1365 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
77304d2a | 1366 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
1da177e4 LT |
1367 | |
1368 | set_bit(BIO_EOF, &bio->bi_flags); | |
1369 | } | |
1370 | ||
c17bb495 AM |
1371 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1372 | ||
1373 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1374 | ||
1375 | static int __init setup_fail_make_request(char *str) | |
1376 | { | |
1377 | return setup_fault_attr(&fail_make_request, str); | |
1378 | } | |
1379 | __setup("fail_make_request=", setup_fail_make_request); | |
1380 | ||
1381 | static int should_fail_request(struct bio *bio) | |
1382 | { | |
eddb2e26 TH |
1383 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1384 | ||
1385 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1386 | return should_fail(&fail_make_request, bio->bi_size); |
1387 | ||
1388 | return 0; | |
1389 | } | |
1390 | ||
1391 | static int __init fail_make_request_debugfs(void) | |
1392 | { | |
1393 | return init_fault_attr_dentries(&fail_make_request, | |
1394 | "fail_make_request"); | |
1395 | } | |
1396 | ||
1397 | late_initcall(fail_make_request_debugfs); | |
1398 | ||
1399 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1400 | ||
1401 | static inline int should_fail_request(struct bio *bio) | |
1402 | { | |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1407 | ||
c07e2b41 JA |
1408 | /* |
1409 | * Check whether this bio extends beyond the end of the device. | |
1410 | */ | |
1411 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1412 | { | |
1413 | sector_t maxsector; | |
1414 | ||
1415 | if (!nr_sectors) | |
1416 | return 0; | |
1417 | ||
1418 | /* Test device or partition size, when known. */ | |
77304d2a | 1419 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
c07e2b41 JA |
1420 | if (maxsector) { |
1421 | sector_t sector = bio->bi_sector; | |
1422 | ||
1423 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1424 | /* | |
1425 | * This may well happen - the kernel calls bread() | |
1426 | * without checking the size of the device, e.g., when | |
1427 | * mounting a device. | |
1428 | */ | |
1429 | handle_bad_sector(bio); | |
1430 | return 1; | |
1431 | } | |
1432 | } | |
1433 | ||
1434 | return 0; | |
1435 | } | |
1436 | ||
1da177e4 | 1437 | /** |
710027a4 | 1438 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1439 | * @bio: The bio describing the location in memory and on the device. |
1440 | * | |
1441 | * generic_make_request() is used to make I/O requests of block | |
1442 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1443 | * to be done. | |
1444 | * | |
1445 | * generic_make_request() does not return any status. The | |
1446 | * success/failure status of the request, along with notification of | |
1447 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1448 | * function described (one day) else where. | |
1449 | * | |
1450 | * The caller of generic_make_request must make sure that bi_io_vec | |
1451 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1452 | * set to describe the device address, and the | |
1453 | * bi_end_io and optionally bi_private are set to describe how | |
1454 | * completion notification should be signaled. | |
1455 | * | |
1456 | * generic_make_request and the drivers it calls may use bi_next if this | |
1457 | * bio happens to be merged with someone else, and may change bi_dev and | |
1458 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1459 | * should NOT be depended on after the call to generic_make_request. | |
1460 | */ | |
d89d8796 | 1461 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1462 | { |
165125e1 | 1463 | struct request_queue *q; |
5ddfe969 | 1464 | sector_t old_sector; |
1da177e4 | 1465 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1466 | dev_t old_dev; |
51fd77bd | 1467 | int err = -EIO; |
1da177e4 LT |
1468 | |
1469 | might_sleep(); | |
1da177e4 | 1470 | |
c07e2b41 JA |
1471 | if (bio_check_eod(bio, nr_sectors)) |
1472 | goto end_io; | |
1da177e4 LT |
1473 | |
1474 | /* | |
1475 | * Resolve the mapping until finished. (drivers are | |
1476 | * still free to implement/resolve their own stacking | |
1477 | * by explicitly returning 0) | |
1478 | * | |
1479 | * NOTE: we don't repeat the blk_size check for each new device. | |
1480 | * Stacking drivers are expected to know what they are doing. | |
1481 | */ | |
5ddfe969 | 1482 | old_sector = -1; |
2056a782 | 1483 | old_dev = 0; |
1da177e4 LT |
1484 | do { |
1485 | char b[BDEVNAME_SIZE]; | |
1486 | ||
1487 | q = bdev_get_queue(bio->bi_bdev); | |
a7384677 | 1488 | if (unlikely(!q)) { |
1da177e4 LT |
1489 | printk(KERN_ERR |
1490 | "generic_make_request: Trying to access " | |
1491 | "nonexistent block-device %s (%Lu)\n", | |
1492 | bdevname(bio->bi_bdev, b), | |
1493 | (long long) bio->bi_sector); | |
a7384677 | 1494 | goto end_io; |
1da177e4 LT |
1495 | } |
1496 | ||
7b6d91da | 1497 | if (unlikely(!(bio->bi_rw & REQ_DISCARD) && |
67efc925 | 1498 | nr_sectors > queue_max_hw_sectors(q))) { |
6728cb0e | 1499 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
ae03bf63 MP |
1500 | bdevname(bio->bi_bdev, b), |
1501 | bio_sectors(bio), | |
1502 | queue_max_hw_sectors(q)); | |
1da177e4 LT |
1503 | goto end_io; |
1504 | } | |
1505 | ||
fde6ad22 | 1506 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1507 | goto end_io; |
1508 | ||
c17bb495 AM |
1509 | if (should_fail_request(bio)) |
1510 | goto end_io; | |
1511 | ||
1da177e4 LT |
1512 | /* |
1513 | * If this device has partitions, remap block n | |
1514 | * of partition p to block n+start(p) of the disk. | |
1515 | */ | |
1516 | blk_partition_remap(bio); | |
1517 | ||
7ba1ba12 MP |
1518 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1519 | goto end_io; | |
1520 | ||
5ddfe969 | 1521 | if (old_sector != -1) |
d07335e5 | 1522 | trace_block_bio_remap(q, bio, old_dev, old_sector); |
2056a782 | 1523 | |
5ddfe969 | 1524 | old_sector = bio->bi_sector; |
2056a782 JA |
1525 | old_dev = bio->bi_bdev->bd_dev; |
1526 | ||
c07e2b41 JA |
1527 | if (bio_check_eod(bio, nr_sectors)) |
1528 | goto end_io; | |
a7384677 | 1529 | |
1e87901e TH |
1530 | /* |
1531 | * Filter flush bio's early so that make_request based | |
1532 | * drivers without flush support don't have to worry | |
1533 | * about them. | |
1534 | */ | |
1535 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { | |
1536 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); | |
1537 | if (!nr_sectors) { | |
1538 | err = 0; | |
1539 | goto end_io; | |
1540 | } | |
1541 | } | |
1542 | ||
8d57a98c AH |
1543 | if ((bio->bi_rw & REQ_DISCARD) && |
1544 | (!blk_queue_discard(q) || | |
1545 | ((bio->bi_rw & REQ_SECURE) && | |
1546 | !blk_queue_secdiscard(q)))) { | |
51fd77bd JA |
1547 | err = -EOPNOTSUPP; |
1548 | goto end_io; | |
1549 | } | |
5ddfe969 | 1550 | |
e43473b7 VG |
1551 | blk_throtl_bio(q, &bio); |
1552 | ||
1553 | /* | |
1554 | * If bio = NULL, bio has been throttled and will be submitted | |
1555 | * later. | |
1556 | */ | |
1557 | if (!bio) | |
1558 | break; | |
1559 | ||
01edede4 MK |
1560 | trace_block_bio_queue(q, bio); |
1561 | ||
1da177e4 LT |
1562 | ret = q->make_request_fn(q, bio); |
1563 | } while (ret); | |
a7384677 TH |
1564 | |
1565 | return; | |
1566 | ||
1567 | end_io: | |
1568 | bio_endio(bio, err); | |
1da177e4 LT |
1569 | } |
1570 | ||
d89d8796 NB |
1571 | /* |
1572 | * We only want one ->make_request_fn to be active at a time, | |
1573 | * else stack usage with stacked devices could be a problem. | |
bddd87c7 | 1574 | * So use current->bio_list to keep a list of requests |
d89d8796 | 1575 | * submited by a make_request_fn function. |
bddd87c7 | 1576 | * current->bio_list is also used as a flag to say if |
d89d8796 NB |
1577 | * generic_make_request is currently active in this task or not. |
1578 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1579 | * then a make_request is active, and new requests should be added | |
1580 | * at the tail | |
1581 | */ | |
1582 | void generic_make_request(struct bio *bio) | |
1583 | { | |
bddd87c7 AM |
1584 | struct bio_list bio_list_on_stack; |
1585 | ||
1586 | if (current->bio_list) { | |
d89d8796 | 1587 | /* make_request is active */ |
bddd87c7 | 1588 | bio_list_add(current->bio_list, bio); |
d89d8796 NB |
1589 | return; |
1590 | } | |
1591 | /* following loop may be a bit non-obvious, and so deserves some | |
1592 | * explanation. | |
1593 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1594 | * ensure that) so we have a list with a single bio. | |
1595 | * We pretend that we have just taken it off a longer list, so | |
bddd87c7 AM |
1596 | * we assign bio_list to a pointer to the bio_list_on_stack, |
1597 | * thus initialising the bio_list of new bios to be | |
d89d8796 NB |
1598 | * added. __generic_make_request may indeed add some more bios |
1599 | * through a recursive call to generic_make_request. If it | |
1600 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1601 | * from the top. In this case we really did just take the bio | |
bddd87c7 AM |
1602 | * of the top of the list (no pretending) and so remove it from |
1603 | * bio_list, and call into __generic_make_request again. | |
d89d8796 NB |
1604 | * |
1605 | * The loop was structured like this to make only one call to | |
1606 | * __generic_make_request (which is important as it is large and | |
1607 | * inlined) and to keep the structure simple. | |
1608 | */ | |
1609 | BUG_ON(bio->bi_next); | |
bddd87c7 AM |
1610 | bio_list_init(&bio_list_on_stack); |
1611 | current->bio_list = &bio_list_on_stack; | |
d89d8796 | 1612 | do { |
d89d8796 | 1613 | __generic_make_request(bio); |
bddd87c7 | 1614 | bio = bio_list_pop(current->bio_list); |
d89d8796 | 1615 | } while (bio); |
bddd87c7 | 1616 | current->bio_list = NULL; /* deactivate */ |
d89d8796 | 1617 | } |
1da177e4 LT |
1618 | EXPORT_SYMBOL(generic_make_request); |
1619 | ||
1620 | /** | |
710027a4 | 1621 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1622 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1623 | * @bio: The &struct bio which describes the I/O | |
1624 | * | |
1625 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1626 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1627 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1628 | * |
1629 | */ | |
1630 | void submit_bio(int rw, struct bio *bio) | |
1631 | { | |
1632 | int count = bio_sectors(bio); | |
1633 | ||
22e2c507 | 1634 | bio->bi_rw |= rw; |
1da177e4 | 1635 | |
bf2de6f5 JA |
1636 | /* |
1637 | * If it's a regular read/write or a barrier with data attached, | |
1638 | * go through the normal accounting stuff before submission. | |
1639 | */ | |
3ffb52e7 | 1640 | if (bio_has_data(bio) && !(rw & REQ_DISCARD)) { |
bf2de6f5 JA |
1641 | if (rw & WRITE) { |
1642 | count_vm_events(PGPGOUT, count); | |
1643 | } else { | |
1644 | task_io_account_read(bio->bi_size); | |
1645 | count_vm_events(PGPGIN, count); | |
1646 | } | |
1647 | ||
1648 | if (unlikely(block_dump)) { | |
1649 | char b[BDEVNAME_SIZE]; | |
8dcbdc74 | 1650 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
ba25f9dc | 1651 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1652 | (rw & WRITE) ? "WRITE" : "READ", |
1653 | (unsigned long long)bio->bi_sector, | |
8dcbdc74 SM |
1654 | bdevname(bio->bi_bdev, b), |
1655 | count); | |
bf2de6f5 | 1656 | } |
1da177e4 LT |
1657 | } |
1658 | ||
1659 | generic_make_request(bio); | |
1660 | } | |
1da177e4 LT |
1661 | EXPORT_SYMBOL(submit_bio); |
1662 | ||
82124d60 KU |
1663 | /** |
1664 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1665 | * @q: the queue | |
1666 | * @rq: the request being checked | |
1667 | * | |
1668 | * Description: | |
1669 | * @rq may have been made based on weaker limitations of upper-level queues | |
1670 | * in request stacking drivers, and it may violate the limitation of @q. | |
1671 | * Since the block layer and the underlying device driver trust @rq | |
1672 | * after it is inserted to @q, it should be checked against @q before | |
1673 | * the insertion using this generic function. | |
1674 | * | |
1675 | * This function should also be useful for request stacking drivers | |
eef35c2d | 1676 | * in some cases below, so export this function. |
82124d60 KU |
1677 | * Request stacking drivers like request-based dm may change the queue |
1678 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1679 | * Such request stacking drivers should check those requests agaist | |
1680 | * the new queue limits again when they dispatch those requests, | |
1681 | * although such checkings are also done against the old queue limits | |
1682 | * when submitting requests. | |
1683 | */ | |
1684 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1685 | { | |
3383977f S |
1686 | if (rq->cmd_flags & REQ_DISCARD) |
1687 | return 0; | |
1688 | ||
ae03bf63 MP |
1689 | if (blk_rq_sectors(rq) > queue_max_sectors(q) || |
1690 | blk_rq_bytes(rq) > queue_max_hw_sectors(q) << 9) { | |
82124d60 KU |
1691 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
1692 | return -EIO; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * queue's settings related to segment counting like q->bounce_pfn | |
1697 | * may differ from that of other stacking queues. | |
1698 | * Recalculate it to check the request correctly on this queue's | |
1699 | * limitation. | |
1700 | */ | |
1701 | blk_recalc_rq_segments(rq); | |
8a78362c | 1702 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
82124d60 KU |
1703 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
1704 | return -EIO; | |
1705 | } | |
1706 | ||
1707 | return 0; | |
1708 | } | |
1709 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1710 | ||
1711 | /** | |
1712 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1713 | * @q: the queue to submit the request | |
1714 | * @rq: the request being queued | |
1715 | */ | |
1716 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1717 | { | |
1718 | unsigned long flags; | |
1719 | ||
1720 | if (blk_rq_check_limits(q, rq)) | |
1721 | return -EIO; | |
1722 | ||
1723 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1724 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1725 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1726 | return -EIO; | |
1727 | #endif | |
1728 | ||
1729 | spin_lock_irqsave(q->queue_lock, flags); | |
1730 | ||
1731 | /* | |
1732 | * Submitting request must be dequeued before calling this function | |
1733 | * because it will be linked to another request_queue | |
1734 | */ | |
1735 | BUG_ON(blk_queued_rq(rq)); | |
1736 | ||
73c10101 | 1737 | add_acct_request(q, rq, ELEVATOR_INSERT_BACK); |
82124d60 KU |
1738 | spin_unlock_irqrestore(q->queue_lock, flags); |
1739 | ||
1740 | return 0; | |
1741 | } | |
1742 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1743 | ||
80a761fd TH |
1744 | /** |
1745 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary | |
1746 | * @rq: request to examine | |
1747 | * | |
1748 | * Description: | |
1749 | * A request could be merge of IOs which require different failure | |
1750 | * handling. This function determines the number of bytes which | |
1751 | * can be failed from the beginning of the request without | |
1752 | * crossing into area which need to be retried further. | |
1753 | * | |
1754 | * Return: | |
1755 | * The number of bytes to fail. | |
1756 | * | |
1757 | * Context: | |
1758 | * queue_lock must be held. | |
1759 | */ | |
1760 | unsigned int blk_rq_err_bytes(const struct request *rq) | |
1761 | { | |
1762 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
1763 | unsigned int bytes = 0; | |
1764 | struct bio *bio; | |
1765 | ||
1766 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) | |
1767 | return blk_rq_bytes(rq); | |
1768 | ||
1769 | /* | |
1770 | * Currently the only 'mixing' which can happen is between | |
1771 | * different fastfail types. We can safely fail portions | |
1772 | * which have all the failfast bits that the first one has - | |
1773 | * the ones which are at least as eager to fail as the first | |
1774 | * one. | |
1775 | */ | |
1776 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1777 | if ((bio->bi_rw & ff) != ff) | |
1778 | break; | |
1779 | bytes += bio->bi_size; | |
1780 | } | |
1781 | ||
1782 | /* this could lead to infinite loop */ | |
1783 | BUG_ON(blk_rq_bytes(rq) && !bytes); | |
1784 | return bytes; | |
1785 | } | |
1786 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); | |
1787 | ||
bc58ba94 JA |
1788 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1789 | { | |
c2553b58 | 1790 | if (blk_do_io_stat(req)) { |
bc58ba94 JA |
1791 | const int rw = rq_data_dir(req); |
1792 | struct hd_struct *part; | |
1793 | int cpu; | |
1794 | ||
1795 | cpu = part_stat_lock(); | |
09e099d4 | 1796 | part = req->part; |
bc58ba94 JA |
1797 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
1798 | part_stat_unlock(); | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | static void blk_account_io_done(struct request *req) | |
1803 | { | |
bc58ba94 | 1804 | /* |
dd4c133f TH |
1805 | * Account IO completion. flush_rq isn't accounted as a |
1806 | * normal IO on queueing nor completion. Accounting the | |
1807 | * containing request is enough. | |
bc58ba94 | 1808 | */ |
414b4ff5 | 1809 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
bc58ba94 JA |
1810 | unsigned long duration = jiffies - req->start_time; |
1811 | const int rw = rq_data_dir(req); | |
1812 | struct hd_struct *part; | |
1813 | int cpu; | |
1814 | ||
1815 | cpu = part_stat_lock(); | |
09e099d4 | 1816 | part = req->part; |
bc58ba94 JA |
1817 | |
1818 | part_stat_inc(cpu, part, ios[rw]); | |
1819 | part_stat_add(cpu, part, ticks[rw], duration); | |
1820 | part_round_stats(cpu, part); | |
316d315b | 1821 | part_dec_in_flight(part, rw); |
bc58ba94 | 1822 | |
6c23a968 | 1823 | hd_struct_put(part); |
bc58ba94 JA |
1824 | part_stat_unlock(); |
1825 | } | |
1826 | } | |
1827 | ||
3bcddeac | 1828 | /** |
9934c8c0 TH |
1829 | * blk_peek_request - peek at the top of a request queue |
1830 | * @q: request queue to peek at | |
1831 | * | |
1832 | * Description: | |
1833 | * Return the request at the top of @q. The returned request | |
1834 | * should be started using blk_start_request() before LLD starts | |
1835 | * processing it. | |
1836 | * | |
1837 | * Return: | |
1838 | * Pointer to the request at the top of @q if available. Null | |
1839 | * otherwise. | |
1840 | * | |
1841 | * Context: | |
1842 | * queue_lock must be held. | |
1843 | */ | |
1844 | struct request *blk_peek_request(struct request_queue *q) | |
158dbda0 TH |
1845 | { |
1846 | struct request *rq; | |
1847 | int ret; | |
1848 | ||
1849 | while ((rq = __elv_next_request(q)) != NULL) { | |
1850 | if (!(rq->cmd_flags & REQ_STARTED)) { | |
1851 | /* | |
1852 | * This is the first time the device driver | |
1853 | * sees this request (possibly after | |
1854 | * requeueing). Notify IO scheduler. | |
1855 | */ | |
33659ebb | 1856 | if (rq->cmd_flags & REQ_SORTED) |
158dbda0 TH |
1857 | elv_activate_rq(q, rq); |
1858 | ||
1859 | /* | |
1860 | * just mark as started even if we don't start | |
1861 | * it, a request that has been delayed should | |
1862 | * not be passed by new incoming requests | |
1863 | */ | |
1864 | rq->cmd_flags |= REQ_STARTED; | |
1865 | trace_block_rq_issue(q, rq); | |
1866 | } | |
1867 | ||
1868 | if (!q->boundary_rq || q->boundary_rq == rq) { | |
1869 | q->end_sector = rq_end_sector(rq); | |
1870 | q->boundary_rq = NULL; | |
1871 | } | |
1872 | ||
1873 | if (rq->cmd_flags & REQ_DONTPREP) | |
1874 | break; | |
1875 | ||
2e46e8b2 | 1876 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
158dbda0 TH |
1877 | /* |
1878 | * make sure space for the drain appears we | |
1879 | * know we can do this because max_hw_segments | |
1880 | * has been adjusted to be one fewer than the | |
1881 | * device can handle | |
1882 | */ | |
1883 | rq->nr_phys_segments++; | |
1884 | } | |
1885 | ||
1886 | if (!q->prep_rq_fn) | |
1887 | break; | |
1888 | ||
1889 | ret = q->prep_rq_fn(q, rq); | |
1890 | if (ret == BLKPREP_OK) { | |
1891 | break; | |
1892 | } else if (ret == BLKPREP_DEFER) { | |
1893 | /* | |
1894 | * the request may have been (partially) prepped. | |
1895 | * we need to keep this request in the front to | |
1896 | * avoid resource deadlock. REQ_STARTED will | |
1897 | * prevent other fs requests from passing this one. | |
1898 | */ | |
2e46e8b2 | 1899 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
158dbda0 TH |
1900 | !(rq->cmd_flags & REQ_DONTPREP)) { |
1901 | /* | |
1902 | * remove the space for the drain we added | |
1903 | * so that we don't add it again | |
1904 | */ | |
1905 | --rq->nr_phys_segments; | |
1906 | } | |
1907 | ||
1908 | rq = NULL; | |
1909 | break; | |
1910 | } else if (ret == BLKPREP_KILL) { | |
1911 | rq->cmd_flags |= REQ_QUIET; | |
c143dc90 JB |
1912 | /* |
1913 | * Mark this request as started so we don't trigger | |
1914 | * any debug logic in the end I/O path. | |
1915 | */ | |
1916 | blk_start_request(rq); | |
40cbbb78 | 1917 | __blk_end_request_all(rq, -EIO); |
158dbda0 TH |
1918 | } else { |
1919 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); | |
1920 | break; | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | return rq; | |
1925 | } | |
9934c8c0 | 1926 | EXPORT_SYMBOL(blk_peek_request); |
158dbda0 | 1927 | |
9934c8c0 | 1928 | void blk_dequeue_request(struct request *rq) |
158dbda0 | 1929 | { |
9934c8c0 TH |
1930 | struct request_queue *q = rq->q; |
1931 | ||
158dbda0 TH |
1932 | BUG_ON(list_empty(&rq->queuelist)); |
1933 | BUG_ON(ELV_ON_HASH(rq)); | |
1934 | ||
1935 | list_del_init(&rq->queuelist); | |
1936 | ||
1937 | /* | |
1938 | * the time frame between a request being removed from the lists | |
1939 | * and to it is freed is accounted as io that is in progress at | |
1940 | * the driver side. | |
1941 | */ | |
9195291e | 1942 | if (blk_account_rq(rq)) { |
0a7ae2ff | 1943 | q->in_flight[rq_is_sync(rq)]++; |
9195291e DS |
1944 | set_io_start_time_ns(rq); |
1945 | } | |
158dbda0 TH |
1946 | } |
1947 | ||
9934c8c0 TH |
1948 | /** |
1949 | * blk_start_request - start request processing on the driver | |
1950 | * @req: request to dequeue | |
1951 | * | |
1952 | * Description: | |
1953 | * Dequeue @req and start timeout timer on it. This hands off the | |
1954 | * request to the driver. | |
1955 | * | |
1956 | * Block internal functions which don't want to start timer should | |
1957 | * call blk_dequeue_request(). | |
1958 | * | |
1959 | * Context: | |
1960 | * queue_lock must be held. | |
1961 | */ | |
1962 | void blk_start_request(struct request *req) | |
1963 | { | |
1964 | blk_dequeue_request(req); | |
1965 | ||
1966 | /* | |
5f49f631 TH |
1967 | * We are now handing the request to the hardware, initialize |
1968 | * resid_len to full count and add the timeout handler. | |
9934c8c0 | 1969 | */ |
5f49f631 | 1970 | req->resid_len = blk_rq_bytes(req); |
dbb66c4b FT |
1971 | if (unlikely(blk_bidi_rq(req))) |
1972 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); | |
1973 | ||
9934c8c0 TH |
1974 | blk_add_timer(req); |
1975 | } | |
1976 | EXPORT_SYMBOL(blk_start_request); | |
1977 | ||
1978 | /** | |
1979 | * blk_fetch_request - fetch a request from a request queue | |
1980 | * @q: request queue to fetch a request from | |
1981 | * | |
1982 | * Description: | |
1983 | * Return the request at the top of @q. The request is started on | |
1984 | * return and LLD can start processing it immediately. | |
1985 | * | |
1986 | * Return: | |
1987 | * Pointer to the request at the top of @q if available. Null | |
1988 | * otherwise. | |
1989 | * | |
1990 | * Context: | |
1991 | * queue_lock must be held. | |
1992 | */ | |
1993 | struct request *blk_fetch_request(struct request_queue *q) | |
1994 | { | |
1995 | struct request *rq; | |
1996 | ||
1997 | rq = blk_peek_request(q); | |
1998 | if (rq) | |
1999 | blk_start_request(rq); | |
2000 | return rq; | |
2001 | } | |
2002 | EXPORT_SYMBOL(blk_fetch_request); | |
2003 | ||
3bcddeac | 2004 | /** |
2e60e022 | 2005 | * blk_update_request - Special helper function for request stacking drivers |
8ebf9756 | 2006 | * @req: the request being processed |
710027a4 | 2007 | * @error: %0 for success, < %0 for error |
8ebf9756 | 2008 | * @nr_bytes: number of bytes to complete @req |
3bcddeac KU |
2009 | * |
2010 | * Description: | |
8ebf9756 RD |
2011 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
2012 | * the request structure even if @req doesn't have leftover. | |
2013 | * If @req has leftover, sets it up for the next range of segments. | |
2e60e022 TH |
2014 | * |
2015 | * This special helper function is only for request stacking drivers | |
2016 | * (e.g. request-based dm) so that they can handle partial completion. | |
2017 | * Actual device drivers should use blk_end_request instead. | |
2018 | * | |
2019 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees | |
2020 | * %false return from this function. | |
3bcddeac KU |
2021 | * |
2022 | * Return: | |
2e60e022 TH |
2023 | * %false - this request doesn't have any more data |
2024 | * %true - this request has more data | |
3bcddeac | 2025 | **/ |
2e60e022 | 2026 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
1da177e4 | 2027 | { |
5450d3e1 | 2028 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
2029 | struct bio *bio; |
2030 | ||
2e60e022 TH |
2031 | if (!req->bio) |
2032 | return false; | |
2033 | ||
5f3ea37c | 2034 | trace_block_rq_complete(req->q, req); |
2056a782 | 2035 | |
1da177e4 | 2036 | /* |
6f41469c TH |
2037 | * For fs requests, rq is just carrier of independent bio's |
2038 | * and each partial completion should be handled separately. | |
2039 | * Reset per-request error on each partial completion. | |
2040 | * | |
2041 | * TODO: tj: This is too subtle. It would be better to let | |
2042 | * low level drivers do what they see fit. | |
1da177e4 | 2043 | */ |
33659ebb | 2044 | if (req->cmd_type == REQ_TYPE_FS) |
1da177e4 LT |
2045 | req->errors = 0; |
2046 | ||
33659ebb CH |
2047 | if (error && req->cmd_type == REQ_TYPE_FS && |
2048 | !(req->cmd_flags & REQ_QUIET)) { | |
79775567 HR |
2049 | char *error_type; |
2050 | ||
2051 | switch (error) { | |
2052 | case -ENOLINK: | |
2053 | error_type = "recoverable transport"; | |
2054 | break; | |
2055 | case -EREMOTEIO: | |
2056 | error_type = "critical target"; | |
2057 | break; | |
2058 | case -EBADE: | |
2059 | error_type = "critical nexus"; | |
2060 | break; | |
2061 | case -EIO: | |
2062 | default: | |
2063 | error_type = "I/O"; | |
2064 | break; | |
2065 | } | |
2066 | printk(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", | |
2067 | error_type, req->rq_disk ? req->rq_disk->disk_name : "?", | |
2068 | (unsigned long long)blk_rq_pos(req)); | |
1da177e4 LT |
2069 | } |
2070 | ||
bc58ba94 | 2071 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 2072 | |
1da177e4 LT |
2073 | total_bytes = bio_nbytes = 0; |
2074 | while ((bio = req->bio) != NULL) { | |
2075 | int nbytes; | |
2076 | ||
2077 | if (nr_bytes >= bio->bi_size) { | |
2078 | req->bio = bio->bi_next; | |
2079 | nbytes = bio->bi_size; | |
5bb23a68 | 2080 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
2081 | next_idx = 0; |
2082 | bio_nbytes = 0; | |
2083 | } else { | |
2084 | int idx = bio->bi_idx + next_idx; | |
2085 | ||
af498d7f | 2086 | if (unlikely(idx >= bio->bi_vcnt)) { |
1da177e4 | 2087 | blk_dump_rq_flags(req, "__end_that"); |
6728cb0e | 2088 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
af498d7f | 2089 | __func__, idx, bio->bi_vcnt); |
1da177e4 LT |
2090 | break; |
2091 | } | |
2092 | ||
2093 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
2094 | BIO_BUG_ON(nbytes > bio->bi_size); | |
2095 | ||
2096 | /* | |
2097 | * not a complete bvec done | |
2098 | */ | |
2099 | if (unlikely(nbytes > nr_bytes)) { | |
2100 | bio_nbytes += nr_bytes; | |
2101 | total_bytes += nr_bytes; | |
2102 | break; | |
2103 | } | |
2104 | ||
2105 | /* | |
2106 | * advance to the next vector | |
2107 | */ | |
2108 | next_idx++; | |
2109 | bio_nbytes += nbytes; | |
2110 | } | |
2111 | ||
2112 | total_bytes += nbytes; | |
2113 | nr_bytes -= nbytes; | |
2114 | ||
6728cb0e JA |
2115 | bio = req->bio; |
2116 | if (bio) { | |
1da177e4 LT |
2117 | /* |
2118 | * end more in this run, or just return 'not-done' | |
2119 | */ | |
2120 | if (unlikely(nr_bytes <= 0)) | |
2121 | break; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | /* | |
2126 | * completely done | |
2127 | */ | |
2e60e022 TH |
2128 | if (!req->bio) { |
2129 | /* | |
2130 | * Reset counters so that the request stacking driver | |
2131 | * can find how many bytes remain in the request | |
2132 | * later. | |
2133 | */ | |
a2dec7b3 | 2134 | req->__data_len = 0; |
2e60e022 TH |
2135 | return false; |
2136 | } | |
1da177e4 LT |
2137 | |
2138 | /* | |
2139 | * if the request wasn't completed, update state | |
2140 | */ | |
2141 | if (bio_nbytes) { | |
5bb23a68 | 2142 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
2143 | bio->bi_idx += next_idx; |
2144 | bio_iovec(bio)->bv_offset += nr_bytes; | |
2145 | bio_iovec(bio)->bv_len -= nr_bytes; | |
2146 | } | |
2147 | ||
a2dec7b3 | 2148 | req->__data_len -= total_bytes; |
2e46e8b2 TH |
2149 | req->buffer = bio_data(req->bio); |
2150 | ||
2151 | /* update sector only for requests with clear definition of sector */ | |
33659ebb | 2152 | if (req->cmd_type == REQ_TYPE_FS || (req->cmd_flags & REQ_DISCARD)) |
a2dec7b3 | 2153 | req->__sector += total_bytes >> 9; |
2e46e8b2 | 2154 | |
80a761fd TH |
2155 | /* mixed attributes always follow the first bio */ |
2156 | if (req->cmd_flags & REQ_MIXED_MERGE) { | |
2157 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
2158 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; | |
2159 | } | |
2160 | ||
2e46e8b2 TH |
2161 | /* |
2162 | * If total number of sectors is less than the first segment | |
2163 | * size, something has gone terribly wrong. | |
2164 | */ | |
2165 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { | |
8182924b | 2166 | blk_dump_rq_flags(req, "request botched"); |
a2dec7b3 | 2167 | req->__data_len = blk_rq_cur_bytes(req); |
2e46e8b2 TH |
2168 | } |
2169 | ||
2170 | /* recalculate the number of segments */ | |
1da177e4 | 2171 | blk_recalc_rq_segments(req); |
2e46e8b2 | 2172 | |
2e60e022 | 2173 | return true; |
1da177e4 | 2174 | } |
2e60e022 | 2175 | EXPORT_SYMBOL_GPL(blk_update_request); |
1da177e4 | 2176 | |
2e60e022 TH |
2177 | static bool blk_update_bidi_request(struct request *rq, int error, |
2178 | unsigned int nr_bytes, | |
2179 | unsigned int bidi_bytes) | |
5efccd17 | 2180 | { |
2e60e022 TH |
2181 | if (blk_update_request(rq, error, nr_bytes)) |
2182 | return true; | |
5efccd17 | 2183 | |
2e60e022 TH |
2184 | /* Bidi request must be completed as a whole */ |
2185 | if (unlikely(blk_bidi_rq(rq)) && | |
2186 | blk_update_request(rq->next_rq, error, bidi_bytes)) | |
2187 | return true; | |
5efccd17 | 2188 | |
e2e1a148 JA |
2189 | if (blk_queue_add_random(rq->q)) |
2190 | add_disk_randomness(rq->rq_disk); | |
2e60e022 TH |
2191 | |
2192 | return false; | |
1da177e4 LT |
2193 | } |
2194 | ||
28018c24 JB |
2195 | /** |
2196 | * blk_unprep_request - unprepare a request | |
2197 | * @req: the request | |
2198 | * | |
2199 | * This function makes a request ready for complete resubmission (or | |
2200 | * completion). It happens only after all error handling is complete, | |
2201 | * so represents the appropriate moment to deallocate any resources | |
2202 | * that were allocated to the request in the prep_rq_fn. The queue | |
2203 | * lock is held when calling this. | |
2204 | */ | |
2205 | void blk_unprep_request(struct request *req) | |
2206 | { | |
2207 | struct request_queue *q = req->q; | |
2208 | ||
2209 | req->cmd_flags &= ~REQ_DONTPREP; | |
2210 | if (q->unprep_rq_fn) | |
2211 | q->unprep_rq_fn(q, req); | |
2212 | } | |
2213 | EXPORT_SYMBOL_GPL(blk_unprep_request); | |
2214 | ||
1da177e4 LT |
2215 | /* |
2216 | * queue lock must be held | |
2217 | */ | |
2e60e022 | 2218 | static void blk_finish_request(struct request *req, int error) |
1da177e4 | 2219 | { |
b8286239 KU |
2220 | if (blk_rq_tagged(req)) |
2221 | blk_queue_end_tag(req->q, req); | |
2222 | ||
ba396a6c | 2223 | BUG_ON(blk_queued_rq(req)); |
1da177e4 | 2224 | |
33659ebb | 2225 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
31373d09 | 2226 | laptop_io_completion(&req->q->backing_dev_info); |
1da177e4 | 2227 | |
e78042e5 MA |
2228 | blk_delete_timer(req); |
2229 | ||
28018c24 JB |
2230 | if (req->cmd_flags & REQ_DONTPREP) |
2231 | blk_unprep_request(req); | |
2232 | ||
2233 | ||
bc58ba94 | 2234 | blk_account_io_done(req); |
b8286239 | 2235 | |
1da177e4 | 2236 | if (req->end_io) |
8ffdc655 | 2237 | req->end_io(req, error); |
b8286239 KU |
2238 | else { |
2239 | if (blk_bidi_rq(req)) | |
2240 | __blk_put_request(req->next_rq->q, req->next_rq); | |
2241 | ||
1da177e4 | 2242 | __blk_put_request(req->q, req); |
b8286239 | 2243 | } |
1da177e4 LT |
2244 | } |
2245 | ||
3b11313a | 2246 | /** |
2e60e022 TH |
2247 | * blk_end_bidi_request - Complete a bidi request |
2248 | * @rq: the request to complete | |
2249 | * @error: %0 for success, < %0 for error | |
2250 | * @nr_bytes: number of bytes to complete @rq | |
2251 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
a0cd1285 JA |
2252 | * |
2253 | * Description: | |
e3a04fe3 | 2254 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
2e60e022 TH |
2255 | * Drivers that supports bidi can safely call this member for any |
2256 | * type of request, bidi or uni. In the later case @bidi_bytes is | |
2257 | * just ignored. | |
336cdb40 KU |
2258 | * |
2259 | * Return: | |
2e60e022 TH |
2260 | * %false - we are done with this request |
2261 | * %true - still buffers pending for this request | |
a0cd1285 | 2262 | **/ |
b1f74493 | 2263 | static bool blk_end_bidi_request(struct request *rq, int error, |
32fab448 KU |
2264 | unsigned int nr_bytes, unsigned int bidi_bytes) |
2265 | { | |
336cdb40 | 2266 | struct request_queue *q = rq->q; |
2e60e022 | 2267 | unsigned long flags; |
32fab448 | 2268 | |
2e60e022 TH |
2269 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2270 | return true; | |
32fab448 | 2271 | |
336cdb40 | 2272 | spin_lock_irqsave(q->queue_lock, flags); |
2e60e022 | 2273 | blk_finish_request(rq, error); |
336cdb40 KU |
2274 | spin_unlock_irqrestore(q->queue_lock, flags); |
2275 | ||
2e60e022 | 2276 | return false; |
32fab448 KU |
2277 | } |
2278 | ||
336cdb40 | 2279 | /** |
2e60e022 TH |
2280 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
2281 | * @rq: the request to complete | |
710027a4 | 2282 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
2283 | * @nr_bytes: number of bytes to complete @rq |
2284 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
336cdb40 KU |
2285 | * |
2286 | * Description: | |
2e60e022 TH |
2287 | * Identical to blk_end_bidi_request() except that queue lock is |
2288 | * assumed to be locked on entry and remains so on return. | |
336cdb40 KU |
2289 | * |
2290 | * Return: | |
2e60e022 TH |
2291 | * %false - we are done with this request |
2292 | * %true - still buffers pending for this request | |
336cdb40 | 2293 | **/ |
b1f74493 FT |
2294 | static bool __blk_end_bidi_request(struct request *rq, int error, |
2295 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
336cdb40 | 2296 | { |
2e60e022 TH |
2297 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
2298 | return true; | |
336cdb40 | 2299 | |
2e60e022 | 2300 | blk_finish_request(rq, error); |
336cdb40 | 2301 | |
2e60e022 | 2302 | return false; |
336cdb40 | 2303 | } |
e19a3ab0 KU |
2304 | |
2305 | /** | |
2306 | * blk_end_request - Helper function for drivers to complete the request. | |
2307 | * @rq: the request being processed | |
710027a4 | 2308 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2309 | * @nr_bytes: number of bytes to complete |
2310 | * | |
2311 | * Description: | |
2312 | * Ends I/O on a number of bytes attached to @rq. | |
2313 | * If @rq has leftover, sets it up for the next range of segments. | |
2314 | * | |
2315 | * Return: | |
b1f74493 FT |
2316 | * %false - we are done with this request |
2317 | * %true - still buffers pending for this request | |
e19a3ab0 | 2318 | **/ |
b1f74493 | 2319 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 2320 | { |
b1f74493 | 2321 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
e19a3ab0 | 2322 | } |
56ad1740 | 2323 | EXPORT_SYMBOL(blk_end_request); |
336cdb40 KU |
2324 | |
2325 | /** | |
b1f74493 FT |
2326 | * blk_end_request_all - Helper function for drives to finish the request. |
2327 | * @rq: the request to finish | |
8ebf9756 | 2328 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
2329 | * |
2330 | * Description: | |
b1f74493 FT |
2331 | * Completely finish @rq. |
2332 | */ | |
2333 | void blk_end_request_all(struct request *rq, int error) | |
336cdb40 | 2334 | { |
b1f74493 FT |
2335 | bool pending; |
2336 | unsigned int bidi_bytes = 0; | |
336cdb40 | 2337 | |
b1f74493 FT |
2338 | if (unlikely(blk_bidi_rq(rq))) |
2339 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
336cdb40 | 2340 | |
b1f74493 FT |
2341 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
2342 | BUG_ON(pending); | |
2343 | } | |
56ad1740 | 2344 | EXPORT_SYMBOL(blk_end_request_all); |
336cdb40 | 2345 | |
b1f74493 FT |
2346 | /** |
2347 | * blk_end_request_cur - Helper function to finish the current request chunk. | |
2348 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2349 | * @error: %0 for success, < %0 for error |
b1f74493 FT |
2350 | * |
2351 | * Description: | |
2352 | * Complete the current consecutively mapped chunk from @rq. | |
2353 | * | |
2354 | * Return: | |
2355 | * %false - we are done with this request | |
2356 | * %true - still buffers pending for this request | |
2357 | */ | |
2358 | bool blk_end_request_cur(struct request *rq, int error) | |
2359 | { | |
2360 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); | |
336cdb40 | 2361 | } |
56ad1740 | 2362 | EXPORT_SYMBOL(blk_end_request_cur); |
336cdb40 | 2363 | |
80a761fd TH |
2364 | /** |
2365 | * blk_end_request_err - Finish a request till the next failure boundary. | |
2366 | * @rq: the request to finish till the next failure boundary for | |
2367 | * @error: must be negative errno | |
2368 | * | |
2369 | * Description: | |
2370 | * Complete @rq till the next failure boundary. | |
2371 | * | |
2372 | * Return: | |
2373 | * %false - we are done with this request | |
2374 | * %true - still buffers pending for this request | |
2375 | */ | |
2376 | bool blk_end_request_err(struct request *rq, int error) | |
2377 | { | |
2378 | WARN_ON(error >= 0); | |
2379 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2380 | } | |
2381 | EXPORT_SYMBOL_GPL(blk_end_request_err); | |
2382 | ||
e3a04fe3 | 2383 | /** |
b1f74493 FT |
2384 | * __blk_end_request - Helper function for drivers to complete the request. |
2385 | * @rq: the request being processed | |
2386 | * @error: %0 for success, < %0 for error | |
2387 | * @nr_bytes: number of bytes to complete | |
e3a04fe3 KU |
2388 | * |
2389 | * Description: | |
b1f74493 | 2390 | * Must be called with queue lock held unlike blk_end_request(). |
e3a04fe3 KU |
2391 | * |
2392 | * Return: | |
b1f74493 FT |
2393 | * %false - we are done with this request |
2394 | * %true - still buffers pending for this request | |
e3a04fe3 | 2395 | **/ |
b1f74493 | 2396 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e3a04fe3 | 2397 | { |
b1f74493 | 2398 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
e3a04fe3 | 2399 | } |
56ad1740 | 2400 | EXPORT_SYMBOL(__blk_end_request); |
e3a04fe3 | 2401 | |
32fab448 | 2402 | /** |
b1f74493 FT |
2403 | * __blk_end_request_all - Helper function for drives to finish the request. |
2404 | * @rq: the request to finish | |
8ebf9756 | 2405 | * @error: %0 for success, < %0 for error |
32fab448 KU |
2406 | * |
2407 | * Description: | |
b1f74493 | 2408 | * Completely finish @rq. Must be called with queue lock held. |
32fab448 | 2409 | */ |
b1f74493 | 2410 | void __blk_end_request_all(struct request *rq, int error) |
32fab448 | 2411 | { |
b1f74493 FT |
2412 | bool pending; |
2413 | unsigned int bidi_bytes = 0; | |
2414 | ||
2415 | if (unlikely(blk_bidi_rq(rq))) | |
2416 | bidi_bytes = blk_rq_bytes(rq->next_rq); | |
2417 | ||
2418 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); | |
2419 | BUG_ON(pending); | |
32fab448 | 2420 | } |
56ad1740 | 2421 | EXPORT_SYMBOL(__blk_end_request_all); |
32fab448 | 2422 | |
e19a3ab0 | 2423 | /** |
b1f74493 FT |
2424 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
2425 | * @rq: the request to finish the current chunk for | |
8ebf9756 | 2426 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2427 | * |
2428 | * Description: | |
b1f74493 FT |
2429 | * Complete the current consecutively mapped chunk from @rq. Must |
2430 | * be called with queue lock held. | |
e19a3ab0 KU |
2431 | * |
2432 | * Return: | |
b1f74493 FT |
2433 | * %false - we are done with this request |
2434 | * %true - still buffers pending for this request | |
2435 | */ | |
2436 | bool __blk_end_request_cur(struct request *rq, int error) | |
e19a3ab0 | 2437 | { |
b1f74493 | 2438 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
e19a3ab0 | 2439 | } |
56ad1740 | 2440 | EXPORT_SYMBOL(__blk_end_request_cur); |
e19a3ab0 | 2441 | |
80a761fd TH |
2442 | /** |
2443 | * __blk_end_request_err - Finish a request till the next failure boundary. | |
2444 | * @rq: the request to finish till the next failure boundary for | |
2445 | * @error: must be negative errno | |
2446 | * | |
2447 | * Description: | |
2448 | * Complete @rq till the next failure boundary. Must be called | |
2449 | * with queue lock held. | |
2450 | * | |
2451 | * Return: | |
2452 | * %false - we are done with this request | |
2453 | * %true - still buffers pending for this request | |
2454 | */ | |
2455 | bool __blk_end_request_err(struct request *rq, int error) | |
2456 | { | |
2457 | WARN_ON(error >= 0); | |
2458 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); | |
2459 | } | |
2460 | EXPORT_SYMBOL_GPL(__blk_end_request_err); | |
2461 | ||
86db1e29 JA |
2462 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2463 | struct bio *bio) | |
1da177e4 | 2464 | { |
a82afdfc | 2465 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
7b6d91da | 2466 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
1da177e4 | 2467 | |
fb2dce86 DW |
2468 | if (bio_has_data(bio)) { |
2469 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2470 | rq->buffer = bio_data(bio); |
2471 | } | |
a2dec7b3 | 2472 | rq->__data_len = bio->bi_size; |
1da177e4 | 2473 | rq->bio = rq->biotail = bio; |
1da177e4 | 2474 | |
66846572 N |
2475 | if (bio->bi_bdev) |
2476 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2477 | } | |
1da177e4 | 2478 | |
2d4dc890 IL |
2479 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
2480 | /** | |
2481 | * rq_flush_dcache_pages - Helper function to flush all pages in a request | |
2482 | * @rq: the request to be flushed | |
2483 | * | |
2484 | * Description: | |
2485 | * Flush all pages in @rq. | |
2486 | */ | |
2487 | void rq_flush_dcache_pages(struct request *rq) | |
2488 | { | |
2489 | struct req_iterator iter; | |
2490 | struct bio_vec *bvec; | |
2491 | ||
2492 | rq_for_each_segment(bvec, rq, iter) | |
2493 | flush_dcache_page(bvec->bv_page); | |
2494 | } | |
2495 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); | |
2496 | #endif | |
2497 | ||
ef9e3fac KU |
2498 | /** |
2499 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2500 | * @q : the queue of the device being checked | |
2501 | * | |
2502 | * Description: | |
2503 | * Check if underlying low-level drivers of a device are busy. | |
2504 | * If the drivers want to export their busy state, they must set own | |
2505 | * exporting function using blk_queue_lld_busy() first. | |
2506 | * | |
2507 | * Basically, this function is used only by request stacking drivers | |
2508 | * to stop dispatching requests to underlying devices when underlying | |
2509 | * devices are busy. This behavior helps more I/O merging on the queue | |
2510 | * of the request stacking driver and prevents I/O throughput regression | |
2511 | * on burst I/O load. | |
2512 | * | |
2513 | * Return: | |
2514 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2515 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2516 | */ | |
2517 | int blk_lld_busy(struct request_queue *q) | |
2518 | { | |
2519 | if (q->lld_busy_fn) | |
2520 | return q->lld_busy_fn(q); | |
2521 | ||
2522 | return 0; | |
2523 | } | |
2524 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2525 | ||
b0fd271d KU |
2526 | /** |
2527 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request | |
2528 | * @rq: the clone request to be cleaned up | |
2529 | * | |
2530 | * Description: | |
2531 | * Free all bios in @rq for a cloned request. | |
2532 | */ | |
2533 | void blk_rq_unprep_clone(struct request *rq) | |
2534 | { | |
2535 | struct bio *bio; | |
2536 | ||
2537 | while ((bio = rq->bio) != NULL) { | |
2538 | rq->bio = bio->bi_next; | |
2539 | ||
2540 | bio_put(bio); | |
2541 | } | |
2542 | } | |
2543 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); | |
2544 | ||
2545 | /* | |
2546 | * Copy attributes of the original request to the clone request. | |
2547 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. | |
2548 | */ | |
2549 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) | |
2550 | { | |
2551 | dst->cpu = src->cpu; | |
3a2edd0d | 2552 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
b0fd271d KU |
2553 | dst->cmd_type = src->cmd_type; |
2554 | dst->__sector = blk_rq_pos(src); | |
2555 | dst->__data_len = blk_rq_bytes(src); | |
2556 | dst->nr_phys_segments = src->nr_phys_segments; | |
2557 | dst->ioprio = src->ioprio; | |
2558 | dst->extra_len = src->extra_len; | |
2559 | } | |
2560 | ||
2561 | /** | |
2562 | * blk_rq_prep_clone - Helper function to setup clone request | |
2563 | * @rq: the request to be setup | |
2564 | * @rq_src: original request to be cloned | |
2565 | * @bs: bio_set that bios for clone are allocated from | |
2566 | * @gfp_mask: memory allocation mask for bio | |
2567 | * @bio_ctr: setup function to be called for each clone bio. | |
2568 | * Returns %0 for success, non %0 for failure. | |
2569 | * @data: private data to be passed to @bio_ctr | |
2570 | * | |
2571 | * Description: | |
2572 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. | |
2573 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) | |
2574 | * are not copied, and copying such parts is the caller's responsibility. | |
2575 | * Also, pages which the original bios are pointing to are not copied | |
2576 | * and the cloned bios just point same pages. | |
2577 | * So cloned bios must be completed before original bios, which means | |
2578 | * the caller must complete @rq before @rq_src. | |
2579 | */ | |
2580 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, | |
2581 | struct bio_set *bs, gfp_t gfp_mask, | |
2582 | int (*bio_ctr)(struct bio *, struct bio *, void *), | |
2583 | void *data) | |
2584 | { | |
2585 | struct bio *bio, *bio_src; | |
2586 | ||
2587 | if (!bs) | |
2588 | bs = fs_bio_set; | |
2589 | ||
2590 | blk_rq_init(NULL, rq); | |
2591 | ||
2592 | __rq_for_each_bio(bio_src, rq_src) { | |
2593 | bio = bio_alloc_bioset(gfp_mask, bio_src->bi_max_vecs, bs); | |
2594 | if (!bio) | |
2595 | goto free_and_out; | |
2596 | ||
2597 | __bio_clone(bio, bio_src); | |
2598 | ||
2599 | if (bio_integrity(bio_src) && | |
7878cba9 | 2600 | bio_integrity_clone(bio, bio_src, gfp_mask, bs)) |
b0fd271d KU |
2601 | goto free_and_out; |
2602 | ||
2603 | if (bio_ctr && bio_ctr(bio, bio_src, data)) | |
2604 | goto free_and_out; | |
2605 | ||
2606 | if (rq->bio) { | |
2607 | rq->biotail->bi_next = bio; | |
2608 | rq->biotail = bio; | |
2609 | } else | |
2610 | rq->bio = rq->biotail = bio; | |
2611 | } | |
2612 | ||
2613 | __blk_rq_prep_clone(rq, rq_src); | |
2614 | ||
2615 | return 0; | |
2616 | ||
2617 | free_and_out: | |
2618 | if (bio) | |
2619 | bio_free(bio, bs); | |
2620 | blk_rq_unprep_clone(rq); | |
2621 | ||
2622 | return -ENOMEM; | |
2623 | } | |
2624 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); | |
2625 | ||
18887ad9 | 2626 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2627 | { |
2628 | return queue_work(kblockd_workqueue, work); | |
2629 | } | |
1da177e4 LT |
2630 | EXPORT_SYMBOL(kblockd_schedule_work); |
2631 | ||
e43473b7 VG |
2632 | int kblockd_schedule_delayed_work(struct request_queue *q, |
2633 | struct delayed_work *dwork, unsigned long delay) | |
2634 | { | |
2635 | return queue_delayed_work(kblockd_workqueue, dwork, delay); | |
2636 | } | |
2637 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); | |
2638 | ||
73c10101 JA |
2639 | #define PLUG_MAGIC 0x91827364 |
2640 | ||
2641 | void blk_start_plug(struct blk_plug *plug) | |
2642 | { | |
2643 | struct task_struct *tsk = current; | |
2644 | ||
2645 | plug->magic = PLUG_MAGIC; | |
2646 | INIT_LIST_HEAD(&plug->list); | |
2647 | plug->should_sort = 0; | |
2648 | ||
2649 | /* | |
2650 | * If this is a nested plug, don't actually assign it. It will be | |
2651 | * flushed on its own. | |
2652 | */ | |
2653 | if (!tsk->plug) { | |
2654 | /* | |
2655 | * Store ordering should not be needed here, since a potential | |
2656 | * preempt will imply a full memory barrier | |
2657 | */ | |
2658 | tsk->plug = plug; | |
2659 | } | |
2660 | } | |
2661 | EXPORT_SYMBOL(blk_start_plug); | |
2662 | ||
2663 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) | |
2664 | { | |
2665 | struct request *rqa = container_of(a, struct request, queuelist); | |
2666 | struct request *rqb = container_of(b, struct request, queuelist); | |
2667 | ||
f83e8261 | 2668 | return !(rqa->q <= rqb->q); |
73c10101 JA |
2669 | } |
2670 | ||
2671 | static void flush_plug_list(struct blk_plug *plug) | |
2672 | { | |
2673 | struct request_queue *q; | |
2674 | unsigned long flags; | |
2675 | struct request *rq; | |
2676 | ||
2677 | BUG_ON(plug->magic != PLUG_MAGIC); | |
2678 | ||
2679 | if (list_empty(&plug->list)) | |
2680 | return; | |
2681 | ||
2682 | if (plug->should_sort) | |
2683 | list_sort(NULL, &plug->list, plug_rq_cmp); | |
2684 | ||
2685 | q = NULL; | |
2686 | local_irq_save(flags); | |
2687 | while (!list_empty(&plug->list)) { | |
2688 | rq = list_entry_rq(plug->list.next); | |
2689 | list_del_init(&rq->queuelist); | |
2690 | BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG)); | |
2691 | BUG_ON(!rq->q); | |
2692 | if (rq->q != q) { | |
2693 | if (q) { | |
4c63f564 | 2694 | __blk_run_queue(q, false); |
73c10101 JA |
2695 | spin_unlock(q->queue_lock); |
2696 | } | |
2697 | q = rq->q; | |
2698 | spin_lock(q->queue_lock); | |
2699 | } | |
2700 | rq->cmd_flags &= ~REQ_ON_PLUG; | |
2701 | ||
2702 | /* | |
2703 | * rq is already accounted, so use raw insert | |
2704 | */ | |
401a18e9 JA |
2705 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
2706 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); | |
2707 | else | |
2708 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); | |
73c10101 JA |
2709 | } |
2710 | ||
2711 | if (q) { | |
4c63f564 | 2712 | __blk_run_queue(q, false); |
73c10101 JA |
2713 | spin_unlock(q->queue_lock); |
2714 | } | |
2715 | ||
2716 | BUG_ON(!list_empty(&plug->list)); | |
2717 | local_irq_restore(flags); | |
2718 | } | |
2719 | ||
2720 | static void __blk_finish_plug(struct task_struct *tsk, struct blk_plug *plug) | |
2721 | { | |
2722 | flush_plug_list(plug); | |
2723 | ||
2724 | if (plug == tsk->plug) | |
2725 | tsk->plug = NULL; | |
2726 | } | |
2727 | ||
2728 | void blk_finish_plug(struct blk_plug *plug) | |
2729 | { | |
2730 | if (plug) | |
2731 | __blk_finish_plug(current, plug); | |
2732 | } | |
2733 | EXPORT_SYMBOL(blk_finish_plug); | |
2734 | ||
2735 | void __blk_flush_plug(struct task_struct *tsk, struct blk_plug *plug) | |
2736 | { | |
2737 | __blk_finish_plug(tsk, plug); | |
2738 | tsk->plug = plug; | |
2739 | } | |
2740 | EXPORT_SYMBOL(__blk_flush_plug); | |
2741 | ||
1da177e4 LT |
2742 | int __init blk_dev_init(void) |
2743 | { | |
9eb55b03 NK |
2744 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
2745 | sizeof(((struct request *)0)->cmd_flags)); | |
2746 | ||
89b90be2 TH |
2747 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
2748 | kblockd_workqueue = alloc_workqueue("kblockd", | |
2749 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1da177e4 LT |
2750 | if (!kblockd_workqueue) |
2751 | panic("Failed to create kblockd\n"); | |
2752 | ||
2753 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2754 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2755 | |
8324aa91 | 2756 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2757 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2758 | |
d38ecf93 | 2759 | return 0; |
1da177e4 | 2760 | } |