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> | |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 | |
7 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 | |
8 | */ | |
9 | ||
10 | /* | |
11 | * This handles all read/write requests to block devices | |
12 | */ | |
1da177e4 LT |
13 | #include <linux/kernel.h> |
14 | #include <linux/module.h> | |
15 | #include <linux/backing-dev.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/blkdev.h> | |
18 | #include <linux/highmem.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/kernel_stat.h> | |
21 | #include <linux/string.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | |
24 | #include <linux/completion.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
ff856bad JA |
29 | #include <linux/interrupt.h> |
30 | #include <linux/cpu.h> | |
2056a782 | 31 | #include <linux/blktrace_api.h> |
c17bb495 | 32 | #include <linux/fault-inject.h> |
1da177e4 LT |
33 | |
34 | /* | |
35 | * for max sense size | |
36 | */ | |
37 | #include <scsi/scsi_cmnd.h> | |
38 | ||
65f27f38 | 39 | static void blk_unplug_work(struct work_struct *work); |
1da177e4 | 40 | static void blk_unplug_timeout(unsigned long data); |
93d17d3d | 41 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); |
52d9e675 TH |
42 | static void init_request_from_bio(struct request *req, struct bio *bio); |
43 | static int __make_request(request_queue_t *q, struct bio *bio); | |
b5deef90 | 44 | static struct io_context *current_io_context(gfp_t gfp_flags, int node); |
1da177e4 LT |
45 | |
46 | /* | |
47 | * For the allocated request tables | |
48 | */ | |
e18b890b | 49 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
50 | |
51 | /* | |
52 | * For queue allocation | |
53 | */ | |
e18b890b | 54 | static struct kmem_cache *requestq_cachep; |
1da177e4 LT |
55 | |
56 | /* | |
57 | * For io context allocations | |
58 | */ | |
e18b890b | 59 | static struct kmem_cache *iocontext_cachep; |
1da177e4 | 60 | |
1da177e4 LT |
61 | /* |
62 | * Controlling structure to kblockd | |
63 | */ | |
ff856bad | 64 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 LT |
65 | |
66 | unsigned long blk_max_low_pfn, blk_max_pfn; | |
67 | ||
68 | EXPORT_SYMBOL(blk_max_low_pfn); | |
69 | EXPORT_SYMBOL(blk_max_pfn); | |
70 | ||
ff856bad JA |
71 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); |
72 | ||
1da177e4 LT |
73 | /* Amount of time in which a process may batch requests */ |
74 | #define BLK_BATCH_TIME (HZ/50UL) | |
75 | ||
76 | /* Number of requests a "batching" process may submit */ | |
77 | #define BLK_BATCH_REQ 32 | |
78 | ||
79 | /* | |
80 | * Return the threshold (number of used requests) at which the queue is | |
81 | * considered to be congested. It include a little hysteresis to keep the | |
82 | * context switch rate down. | |
83 | */ | |
84 | static inline int queue_congestion_on_threshold(struct request_queue *q) | |
85 | { | |
86 | return q->nr_congestion_on; | |
87 | } | |
88 | ||
89 | /* | |
90 | * The threshold at which a queue is considered to be uncongested | |
91 | */ | |
92 | static inline int queue_congestion_off_threshold(struct request_queue *q) | |
93 | { | |
94 | return q->nr_congestion_off; | |
95 | } | |
96 | ||
97 | static void blk_queue_congestion_threshold(struct request_queue *q) | |
98 | { | |
99 | int nr; | |
100 | ||
101 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
102 | if (nr > q->nr_requests) | |
103 | nr = q->nr_requests; | |
104 | q->nr_congestion_on = nr; | |
105 | ||
106 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
107 | if (nr < 1) | |
108 | nr = 1; | |
109 | q->nr_congestion_off = nr; | |
110 | } | |
111 | ||
1da177e4 LT |
112 | /** |
113 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
114 | * @bdev: device | |
115 | * | |
116 | * Locates the passed device's request queue and returns the address of its | |
117 | * backing_dev_info | |
118 | * | |
119 | * Will return NULL if the request queue cannot be located. | |
120 | */ | |
121 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
122 | { | |
123 | struct backing_dev_info *ret = NULL; | |
124 | request_queue_t *q = bdev_get_queue(bdev); | |
125 | ||
126 | if (q) | |
127 | ret = &q->backing_dev_info; | |
128 | return ret; | |
129 | } | |
1da177e4 LT |
130 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
131 | ||
1da177e4 LT |
132 | /** |
133 | * blk_queue_prep_rq - set a prepare_request function for queue | |
134 | * @q: queue | |
135 | * @pfn: prepare_request function | |
136 | * | |
137 | * It's possible for a queue to register a prepare_request callback which | |
138 | * is invoked before the request is handed to the request_fn. The goal of | |
139 | * the function is to prepare a request for I/O, it can be used to build a | |
140 | * cdb from the request data for instance. | |
141 | * | |
142 | */ | |
143 | void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) | |
144 | { | |
145 | q->prep_rq_fn = pfn; | |
146 | } | |
147 | ||
148 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
149 | ||
150 | /** | |
151 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
152 | * @q: queue | |
153 | * @mbfn: merge_bvec_fn | |
154 | * | |
155 | * Usually queues have static limitations on the max sectors or segments that | |
156 | * we can put in a request. Stacking drivers may have some settings that | |
157 | * are dynamic, and thus we have to query the queue whether it is ok to | |
158 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
159 | * has such limitations, it needs to register a merge_bvec_fn to control | |
160 | * the size of bio's sent to it. Note that a block device *must* allow a | |
161 | * single page to be added to an empty bio. The block device driver may want | |
162 | * to use the bio_split() function to deal with these bio's. By default | |
163 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
164 | * honored. | |
165 | */ | |
166 | void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) | |
167 | { | |
168 | q->merge_bvec_fn = mbfn; | |
169 | } | |
170 | ||
171 | EXPORT_SYMBOL(blk_queue_merge_bvec); | |
172 | ||
ff856bad JA |
173 | void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn) |
174 | { | |
175 | q->softirq_done_fn = fn; | |
176 | } | |
177 | ||
178 | EXPORT_SYMBOL(blk_queue_softirq_done); | |
179 | ||
1da177e4 LT |
180 | /** |
181 | * blk_queue_make_request - define an alternate make_request function for a device | |
182 | * @q: the request queue for the device to be affected | |
183 | * @mfn: the alternate make_request function | |
184 | * | |
185 | * Description: | |
186 | * The normal way for &struct bios to be passed to a device | |
187 | * driver is for them to be collected into requests on a request | |
188 | * queue, and then to allow the device driver to select requests | |
189 | * off that queue when it is ready. This works well for many block | |
190 | * devices. However some block devices (typically virtual devices | |
191 | * such as md or lvm) do not benefit from the processing on the | |
192 | * request queue, and are served best by having the requests passed | |
193 | * directly to them. This can be achieved by providing a function | |
194 | * to blk_queue_make_request(). | |
195 | * | |
196 | * Caveat: | |
197 | * The driver that does this *must* be able to deal appropriately | |
198 | * with buffers in "highmemory". This can be accomplished by either calling | |
199 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
200 | * blk_queue_bounce() to create a buffer in normal memory. | |
201 | **/ | |
202 | void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) | |
203 | { | |
204 | /* | |
205 | * set defaults | |
206 | */ | |
207 | q->nr_requests = BLKDEV_MAX_RQ; | |
309c0a1d SM |
208 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
209 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1da177e4 LT |
210 | q->make_request_fn = mfn; |
211 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
212 | q->backing_dev_info.state = 0; | |
213 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
defd94b7 | 214 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
1da177e4 LT |
215 | blk_queue_hardsect_size(q, 512); |
216 | blk_queue_dma_alignment(q, 511); | |
217 | blk_queue_congestion_threshold(q); | |
218 | q->nr_batching = BLK_BATCH_REQ; | |
219 | ||
220 | q->unplug_thresh = 4; /* hmm */ | |
221 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
222 | if (q->unplug_delay == 0) | |
223 | q->unplug_delay = 1; | |
224 | ||
65f27f38 | 225 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
1da177e4 LT |
226 | |
227 | q->unplug_timer.function = blk_unplug_timeout; | |
228 | q->unplug_timer.data = (unsigned long)q; | |
229 | ||
230 | /* | |
231 | * by default assume old behaviour and bounce for any highmem page | |
232 | */ | |
233 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
1da177e4 LT |
234 | } |
235 | ||
236 | EXPORT_SYMBOL(blk_queue_make_request); | |
237 | ||
1ea25ecb | 238 | static void rq_init(request_queue_t *q, struct request *rq) |
1da177e4 LT |
239 | { |
240 | INIT_LIST_HEAD(&rq->queuelist); | |
ff856bad | 241 | INIT_LIST_HEAD(&rq->donelist); |
1da177e4 LT |
242 | |
243 | rq->errors = 0; | |
1da177e4 | 244 | rq->bio = rq->biotail = NULL; |
2e662b65 JA |
245 | INIT_HLIST_NODE(&rq->hash); |
246 | RB_CLEAR_NODE(&rq->rb_node); | |
22e2c507 | 247 | rq->ioprio = 0; |
1da177e4 LT |
248 | rq->buffer = NULL; |
249 | rq->ref_count = 1; | |
250 | rq->q = q; | |
1da177e4 LT |
251 | rq->special = NULL; |
252 | rq->data_len = 0; | |
253 | rq->data = NULL; | |
df46b9a4 | 254 | rq->nr_phys_segments = 0; |
1da177e4 LT |
255 | rq->sense = NULL; |
256 | rq->end_io = NULL; | |
257 | rq->end_io_data = NULL; | |
ff856bad | 258 | rq->completion_data = NULL; |
1da177e4 LT |
259 | } |
260 | ||
261 | /** | |
262 | * blk_queue_ordered - does this queue support ordered writes | |
797e7dbb TH |
263 | * @q: the request queue |
264 | * @ordered: one of QUEUE_ORDERED_* | |
fddfdeaf | 265 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes |
1da177e4 LT |
266 | * |
267 | * Description: | |
268 | * For journalled file systems, doing ordered writes on a commit | |
269 | * block instead of explicitly doing wait_on_buffer (which is bad | |
270 | * for performance) can be a big win. Block drivers supporting this | |
271 | * feature should call this function and indicate so. | |
272 | * | |
273 | **/ | |
797e7dbb TH |
274 | int blk_queue_ordered(request_queue_t *q, unsigned ordered, |
275 | prepare_flush_fn *prepare_flush_fn) | |
276 | { | |
277 | if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && | |
278 | prepare_flush_fn == NULL) { | |
279 | printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); | |
280 | return -EINVAL; | |
281 | } | |
282 | ||
283 | if (ordered != QUEUE_ORDERED_NONE && | |
284 | ordered != QUEUE_ORDERED_DRAIN && | |
285 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
286 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
287 | ordered != QUEUE_ORDERED_TAG && | |
288 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
289 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
290 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
291 | return -EINVAL; | |
1da177e4 | 292 | } |
797e7dbb | 293 | |
60481b12 | 294 | q->ordered = ordered; |
797e7dbb TH |
295 | q->next_ordered = ordered; |
296 | q->prepare_flush_fn = prepare_flush_fn; | |
297 | ||
298 | return 0; | |
1da177e4 LT |
299 | } |
300 | ||
301 | EXPORT_SYMBOL(blk_queue_ordered); | |
302 | ||
303 | /** | |
304 | * blk_queue_issue_flush_fn - set function for issuing a flush | |
305 | * @q: the request queue | |
306 | * @iff: the function to be called issuing the flush | |
307 | * | |
308 | * Description: | |
309 | * If a driver supports issuing a flush command, the support is notified | |
310 | * to the block layer by defining it through this call. | |
311 | * | |
312 | **/ | |
313 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) | |
314 | { | |
315 | q->issue_flush_fn = iff; | |
316 | } | |
317 | ||
318 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); | |
319 | ||
320 | /* | |
321 | * Cache flushing for ordered writes handling | |
322 | */ | |
797e7dbb | 323 | inline unsigned blk_ordered_cur_seq(request_queue_t *q) |
1da177e4 | 324 | { |
797e7dbb TH |
325 | if (!q->ordseq) |
326 | return 0; | |
327 | return 1 << ffz(q->ordseq); | |
1da177e4 LT |
328 | } |
329 | ||
797e7dbb | 330 | unsigned blk_ordered_req_seq(struct request *rq) |
1da177e4 | 331 | { |
1da177e4 LT |
332 | request_queue_t *q = rq->q; |
333 | ||
797e7dbb | 334 | BUG_ON(q->ordseq == 0); |
8922e16c | 335 | |
797e7dbb TH |
336 | if (rq == &q->pre_flush_rq) |
337 | return QUEUE_ORDSEQ_PREFLUSH; | |
338 | if (rq == &q->bar_rq) | |
339 | return QUEUE_ORDSEQ_BAR; | |
340 | if (rq == &q->post_flush_rq) | |
341 | return QUEUE_ORDSEQ_POSTFLUSH; | |
1da177e4 | 342 | |
bc90ba09 TH |
343 | /* |
344 | * !fs requests don't need to follow barrier ordering. Always | |
345 | * put them at the front. This fixes the following deadlock. | |
346 | * | |
347 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
348 | */ | |
349 | if (!blk_fs_request(rq)) | |
350 | return QUEUE_ORDSEQ_DRAIN; | |
351 | ||
4aff5e23 JA |
352 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == |
353 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
797e7dbb TH |
354 | return QUEUE_ORDSEQ_DRAIN; |
355 | else | |
356 | return QUEUE_ORDSEQ_DONE; | |
1da177e4 LT |
357 | } |
358 | ||
797e7dbb | 359 | void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error) |
1da177e4 | 360 | { |
797e7dbb TH |
361 | struct request *rq; |
362 | int uptodate; | |
1da177e4 | 363 | |
797e7dbb TH |
364 | if (error && !q->orderr) |
365 | q->orderr = error; | |
1da177e4 | 366 | |
797e7dbb TH |
367 | BUG_ON(q->ordseq & seq); |
368 | q->ordseq |= seq; | |
1da177e4 | 369 | |
797e7dbb TH |
370 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) |
371 | return; | |
1da177e4 LT |
372 | |
373 | /* | |
797e7dbb | 374 | * Okay, sequence complete. |
1da177e4 | 375 | */ |
797e7dbb TH |
376 | rq = q->orig_bar_rq; |
377 | uptodate = q->orderr ? q->orderr : 1; | |
1da177e4 | 378 | |
797e7dbb | 379 | q->ordseq = 0; |
1da177e4 | 380 | |
797e7dbb TH |
381 | end_that_request_first(rq, uptodate, rq->hard_nr_sectors); |
382 | end_that_request_last(rq, uptodate); | |
1da177e4 LT |
383 | } |
384 | ||
797e7dbb | 385 | static void pre_flush_end_io(struct request *rq, int error) |
1da177e4 | 386 | { |
797e7dbb TH |
387 | elv_completed_request(rq->q, rq); |
388 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
389 | } | |
1da177e4 | 390 | |
797e7dbb TH |
391 | static void bar_end_io(struct request *rq, int error) |
392 | { | |
393 | elv_completed_request(rq->q, rq); | |
394 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
395 | } | |
1da177e4 | 396 | |
797e7dbb TH |
397 | static void post_flush_end_io(struct request *rq, int error) |
398 | { | |
399 | elv_completed_request(rq->q, rq); | |
400 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
401 | } | |
1da177e4 | 402 | |
797e7dbb TH |
403 | static void queue_flush(request_queue_t *q, unsigned which) |
404 | { | |
405 | struct request *rq; | |
406 | rq_end_io_fn *end_io; | |
1da177e4 | 407 | |
797e7dbb TH |
408 | if (which == QUEUE_ORDERED_PREFLUSH) { |
409 | rq = &q->pre_flush_rq; | |
410 | end_io = pre_flush_end_io; | |
411 | } else { | |
412 | rq = &q->post_flush_rq; | |
413 | end_io = post_flush_end_io; | |
1da177e4 | 414 | } |
797e7dbb | 415 | |
4aff5e23 | 416 | rq->cmd_flags = REQ_HARDBARRIER; |
797e7dbb | 417 | rq_init(q, rq); |
797e7dbb | 418 | rq->elevator_private = NULL; |
c00895ab | 419 | rq->elevator_private2 = NULL; |
797e7dbb | 420 | rq->rq_disk = q->bar_rq.rq_disk; |
797e7dbb TH |
421 | rq->end_io = end_io; |
422 | q->prepare_flush_fn(q, rq); | |
423 | ||
30e9656c | 424 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
1da177e4 LT |
425 | } |
426 | ||
797e7dbb TH |
427 | static inline struct request *start_ordered(request_queue_t *q, |
428 | struct request *rq) | |
1da177e4 | 429 | { |
797e7dbb TH |
430 | q->bi_size = 0; |
431 | q->orderr = 0; | |
432 | q->ordered = q->next_ordered; | |
433 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
434 | ||
435 | /* | |
436 | * Prep proxy barrier request. | |
437 | */ | |
438 | blkdev_dequeue_request(rq); | |
439 | q->orig_bar_rq = rq; | |
440 | rq = &q->bar_rq; | |
4aff5e23 | 441 | rq->cmd_flags = 0; |
797e7dbb | 442 | rq_init(q, rq); |
4aff5e23 JA |
443 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) |
444 | rq->cmd_flags |= REQ_RW; | |
445 | rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0; | |
797e7dbb | 446 | rq->elevator_private = NULL; |
c00895ab | 447 | rq->elevator_private2 = NULL; |
797e7dbb TH |
448 | init_request_from_bio(rq, q->orig_bar_rq->bio); |
449 | rq->end_io = bar_end_io; | |
450 | ||
451 | /* | |
452 | * Queue ordered sequence. As we stack them at the head, we | |
453 | * need to queue in reverse order. Note that we rely on that | |
454 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
455 | * request gets inbetween ordered sequence. | |
456 | */ | |
457 | if (q->ordered & QUEUE_ORDERED_POSTFLUSH) | |
458 | queue_flush(q, QUEUE_ORDERED_POSTFLUSH); | |
459 | else | |
460 | q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; | |
461 | ||
30e9656c | 462 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
797e7dbb TH |
463 | |
464 | if (q->ordered & QUEUE_ORDERED_PREFLUSH) { | |
465 | queue_flush(q, QUEUE_ORDERED_PREFLUSH); | |
466 | rq = &q->pre_flush_rq; | |
467 | } else | |
468 | q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; | |
1da177e4 | 469 | |
797e7dbb TH |
470 | if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) |
471 | q->ordseq |= QUEUE_ORDSEQ_DRAIN; | |
472 | else | |
473 | rq = NULL; | |
474 | ||
475 | return rq; | |
1da177e4 LT |
476 | } |
477 | ||
797e7dbb | 478 | int blk_do_ordered(request_queue_t *q, struct request **rqp) |
1da177e4 | 479 | { |
9a7a67af | 480 | struct request *rq = *rqp; |
797e7dbb | 481 | int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); |
1da177e4 | 482 | |
797e7dbb TH |
483 | if (!q->ordseq) { |
484 | if (!is_barrier) | |
485 | return 1; | |
1da177e4 | 486 | |
797e7dbb TH |
487 | if (q->next_ordered != QUEUE_ORDERED_NONE) { |
488 | *rqp = start_ordered(q, rq); | |
489 | return 1; | |
490 | } else { | |
491 | /* | |
492 | * This can happen when the queue switches to | |
493 | * ORDERED_NONE while this request is on it. | |
494 | */ | |
495 | blkdev_dequeue_request(rq); | |
496 | end_that_request_first(rq, -EOPNOTSUPP, | |
497 | rq->hard_nr_sectors); | |
498 | end_that_request_last(rq, -EOPNOTSUPP); | |
499 | *rqp = NULL; | |
500 | return 0; | |
501 | } | |
502 | } | |
1da177e4 | 503 | |
9a7a67af JA |
504 | /* |
505 | * Ordered sequence in progress | |
506 | */ | |
507 | ||
508 | /* Special requests are not subject to ordering rules. */ | |
509 | if (!blk_fs_request(rq) && | |
510 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
511 | return 1; | |
512 | ||
797e7dbb | 513 | if (q->ordered & QUEUE_ORDERED_TAG) { |
9a7a67af | 514 | /* Ordered by tag. Blocking the next barrier is enough. */ |
797e7dbb TH |
515 | if (is_barrier && rq != &q->bar_rq) |
516 | *rqp = NULL; | |
9a7a67af JA |
517 | } else { |
518 | /* Ordered by draining. Wait for turn. */ | |
519 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
520 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
521 | *rqp = NULL; | |
1da177e4 LT |
522 | } |
523 | ||
524 | return 1; | |
525 | } | |
526 | ||
797e7dbb | 527 | static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error) |
1da177e4 | 528 | { |
797e7dbb TH |
529 | request_queue_t *q = bio->bi_private; |
530 | struct bio_vec *bvec; | |
531 | int i; | |
532 | ||
533 | /* | |
534 | * This is dry run, restore bio_sector and size. We'll finish | |
535 | * this request again with the original bi_end_io after an | |
536 | * error occurs or post flush is complete. | |
537 | */ | |
538 | q->bi_size += bytes; | |
539 | ||
540 | if (bio->bi_size) | |
541 | return 1; | |
542 | ||
543 | /* Rewind bvec's */ | |
544 | bio->bi_idx = 0; | |
545 | bio_for_each_segment(bvec, bio, i) { | |
546 | bvec->bv_len += bvec->bv_offset; | |
547 | bvec->bv_offset = 0; | |
548 | } | |
549 | ||
550 | /* Reset bio */ | |
551 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
552 | bio->bi_size = q->bi_size; | |
553 | bio->bi_sector -= (q->bi_size >> 9); | |
554 | q->bi_size = 0; | |
555 | ||
556 | return 0; | |
1da177e4 | 557 | } |
1da177e4 | 558 | |
1ea25ecb JA |
559 | static int ordered_bio_endio(struct request *rq, struct bio *bio, |
560 | unsigned int nbytes, int error) | |
1da177e4 | 561 | { |
797e7dbb TH |
562 | request_queue_t *q = rq->q; |
563 | bio_end_io_t *endio; | |
564 | void *private; | |
565 | ||
566 | if (&q->bar_rq != rq) | |
567 | return 0; | |
568 | ||
569 | /* | |
570 | * Okay, this is the barrier request in progress, dry finish it. | |
571 | */ | |
572 | if (error && !q->orderr) | |
573 | q->orderr = error; | |
574 | ||
575 | endio = bio->bi_end_io; | |
576 | private = bio->bi_private; | |
577 | bio->bi_end_io = flush_dry_bio_endio; | |
578 | bio->bi_private = q; | |
579 | ||
580 | bio_endio(bio, nbytes, error); | |
581 | ||
582 | bio->bi_end_io = endio; | |
583 | bio->bi_private = private; | |
584 | ||
585 | return 1; | |
1da177e4 | 586 | } |
1da177e4 LT |
587 | |
588 | /** | |
589 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
590 | * @q: the request queue for the device | |
591 | * @dma_addr: bus address limit | |
592 | * | |
593 | * Description: | |
594 | * Different hardware can have different requirements as to what pages | |
595 | * it can do I/O directly to. A low level driver can call | |
596 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
5ee1af9f | 597 | * buffers for doing I/O to pages residing above @page. |
1da177e4 LT |
598 | **/ |
599 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) | |
600 | { | |
601 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
5ee1af9f AK |
602 | int dma = 0; |
603 | ||
604 | q->bounce_gfp = GFP_NOIO; | |
605 | #if BITS_PER_LONG == 64 | |
606 | /* Assume anything <= 4GB can be handled by IOMMU. | |
607 | Actually some IOMMUs can handle everything, but I don't | |
608 | know of a way to test this here. */ | |
8269730b | 609 | if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
5ee1af9f AK |
610 | dma = 1; |
611 | q->bounce_pfn = max_low_pfn; | |
612 | #else | |
613 | if (bounce_pfn < blk_max_low_pfn) | |
614 | dma = 1; | |
615 | q->bounce_pfn = bounce_pfn; | |
616 | #endif | |
617 | if (dma) { | |
1da177e4 LT |
618 | init_emergency_isa_pool(); |
619 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
5ee1af9f AK |
620 | q->bounce_pfn = bounce_pfn; |
621 | } | |
1da177e4 LT |
622 | } |
623 | ||
624 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
625 | ||
626 | /** | |
627 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
628 | * @q: the request queue for the device | |
629 | * @max_sectors: max sectors in the usual 512b unit | |
630 | * | |
631 | * Description: | |
632 | * Enables a low level driver to set an upper limit on the size of | |
633 | * received requests. | |
634 | **/ | |
2cb2e147 | 635 | void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors) |
1da177e4 LT |
636 | { |
637 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
638 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
639 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
640 | } | |
641 | ||
defd94b7 MC |
642 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
643 | q->max_hw_sectors = q->max_sectors = max_sectors; | |
644 | else { | |
645 | q->max_sectors = BLK_DEF_MAX_SECTORS; | |
646 | q->max_hw_sectors = max_sectors; | |
647 | } | |
1da177e4 LT |
648 | } |
649 | ||
650 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
651 | ||
652 | /** | |
653 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
654 | * @q: the request queue for the device | |
655 | * @max_segments: max number of segments | |
656 | * | |
657 | * Description: | |
658 | * Enables a low level driver to set an upper limit on the number of | |
659 | * physical data segments in a request. This would be the largest sized | |
660 | * scatter list the driver could handle. | |
661 | **/ | |
662 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) | |
663 | { | |
664 | if (!max_segments) { | |
665 | max_segments = 1; | |
666 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
667 | } | |
668 | ||
669 | q->max_phys_segments = max_segments; | |
670 | } | |
671 | ||
672 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
673 | ||
674 | /** | |
675 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
676 | * @q: the request queue for the device | |
677 | * @max_segments: max number of segments | |
678 | * | |
679 | * Description: | |
680 | * Enables a low level driver to set an upper limit on the number of | |
681 | * hw data segments in a request. This would be the largest number of | |
682 | * address/length pairs the host adapter can actually give as once | |
683 | * to the device. | |
684 | **/ | |
685 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) | |
686 | { | |
687 | if (!max_segments) { | |
688 | max_segments = 1; | |
689 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
690 | } | |
691 | ||
692 | q->max_hw_segments = max_segments; | |
693 | } | |
694 | ||
695 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
696 | ||
697 | /** | |
698 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
699 | * @q: the request queue for the device | |
700 | * @max_size: max size of segment in bytes | |
701 | * | |
702 | * Description: | |
703 | * Enables a low level driver to set an upper limit on the size of a | |
704 | * coalesced segment | |
705 | **/ | |
706 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) | |
707 | { | |
708 | if (max_size < PAGE_CACHE_SIZE) { | |
709 | max_size = PAGE_CACHE_SIZE; | |
710 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
711 | } | |
712 | ||
713 | q->max_segment_size = max_size; | |
714 | } | |
715 | ||
716 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
717 | ||
718 | /** | |
719 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
720 | * @q: the request queue for the device | |
721 | * @size: the hardware sector size, in bytes | |
722 | * | |
723 | * Description: | |
724 | * This should typically be set to the lowest possible sector size | |
725 | * that the hardware can operate on (possible without reverting to | |
726 | * even internal read-modify-write operations). Usually the default | |
727 | * of 512 covers most hardware. | |
728 | **/ | |
729 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) | |
730 | { | |
731 | q->hardsect_size = size; | |
732 | } | |
733 | ||
734 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
735 | ||
736 | /* | |
737 | * Returns the minimum that is _not_ zero, unless both are zero. | |
738 | */ | |
739 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
740 | ||
741 | /** | |
742 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
743 | * @t: the stacking driver (top) | |
744 | * @b: the underlying device (bottom) | |
745 | **/ | |
746 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) | |
747 | { | |
748 | /* zero is "infinity" */ | |
defd94b7 MC |
749 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); |
750 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | |
1da177e4 LT |
751 | |
752 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
753 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
754 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
755 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
89e5c8b5 N |
756 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
757 | clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); | |
1da177e4 LT |
758 | } |
759 | ||
760 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
761 | ||
762 | /** | |
763 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
764 | * @q: the request queue for the device | |
765 | * @mask: the memory boundary mask | |
766 | **/ | |
767 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) | |
768 | { | |
769 | if (mask < PAGE_CACHE_SIZE - 1) { | |
770 | mask = PAGE_CACHE_SIZE - 1; | |
771 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
772 | } | |
773 | ||
774 | q->seg_boundary_mask = mask; | |
775 | } | |
776 | ||
777 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
778 | ||
779 | /** | |
780 | * blk_queue_dma_alignment - set dma length and memory alignment | |
781 | * @q: the request queue for the device | |
782 | * @mask: alignment mask | |
783 | * | |
784 | * description: | |
785 | * set required memory and length aligment for direct dma transactions. | |
786 | * this is used when buiding direct io requests for the queue. | |
787 | * | |
788 | **/ | |
789 | void blk_queue_dma_alignment(request_queue_t *q, int mask) | |
790 | { | |
791 | q->dma_alignment = mask; | |
792 | } | |
793 | ||
794 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
795 | ||
796 | /** | |
797 | * blk_queue_find_tag - find a request by its tag and queue | |
1da177e4 LT |
798 | * @q: The request queue for the device |
799 | * @tag: The tag of the request | |
800 | * | |
801 | * Notes: | |
802 | * Should be used when a device returns a tag and you want to match | |
803 | * it with a request. | |
804 | * | |
805 | * no locks need be held. | |
806 | **/ | |
807 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) | |
808 | { | |
f583f492 | 809 | return blk_map_queue_find_tag(q->queue_tags, tag); |
1da177e4 LT |
810 | } |
811 | ||
812 | EXPORT_SYMBOL(blk_queue_find_tag); | |
813 | ||
814 | /** | |
492dfb48 JB |
815 | * __blk_free_tags - release a given set of tag maintenance info |
816 | * @bqt: the tag map to free | |
1da177e4 | 817 | * |
492dfb48 JB |
818 | * Tries to free the specified @bqt@. Returns true if it was |
819 | * actually freed and false if there are still references using it | |
820 | */ | |
821 | static int __blk_free_tags(struct blk_queue_tag *bqt) | |
1da177e4 | 822 | { |
492dfb48 | 823 | int retval; |
1da177e4 | 824 | |
492dfb48 JB |
825 | retval = atomic_dec_and_test(&bqt->refcnt); |
826 | if (retval) { | |
1da177e4 LT |
827 | BUG_ON(bqt->busy); |
828 | BUG_ON(!list_empty(&bqt->busy_list)); | |
829 | ||
830 | kfree(bqt->tag_index); | |
831 | bqt->tag_index = NULL; | |
832 | ||
833 | kfree(bqt->tag_map); | |
834 | bqt->tag_map = NULL; | |
835 | ||
836 | kfree(bqt); | |
492dfb48 | 837 | |
1da177e4 LT |
838 | } |
839 | ||
492dfb48 JB |
840 | return retval; |
841 | } | |
842 | ||
843 | /** | |
844 | * __blk_queue_free_tags - release tag maintenance info | |
845 | * @q: the request queue for the device | |
846 | * | |
847 | * Notes: | |
848 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
849 | * has been used. So there's no need to call this directly. | |
850 | **/ | |
851 | static void __blk_queue_free_tags(request_queue_t *q) | |
852 | { | |
853 | struct blk_queue_tag *bqt = q->queue_tags; | |
854 | ||
855 | if (!bqt) | |
856 | return; | |
857 | ||
858 | __blk_free_tags(bqt); | |
859 | ||
1da177e4 LT |
860 | q->queue_tags = NULL; |
861 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
862 | } | |
863 | ||
492dfb48 JB |
864 | |
865 | /** | |
866 | * blk_free_tags - release a given set of tag maintenance info | |
867 | * @bqt: the tag map to free | |
868 | * | |
869 | * For externally managed @bqt@ frees the map. Callers of this | |
870 | * function must guarantee to have released all the queues that | |
871 | * might have been using this tag map. | |
872 | */ | |
873 | void blk_free_tags(struct blk_queue_tag *bqt) | |
874 | { | |
875 | if (unlikely(!__blk_free_tags(bqt))) | |
876 | BUG(); | |
877 | } | |
878 | EXPORT_SYMBOL(blk_free_tags); | |
879 | ||
1da177e4 LT |
880 | /** |
881 | * blk_queue_free_tags - release tag maintenance info | |
882 | * @q: the request queue for the device | |
883 | * | |
884 | * Notes: | |
885 | * This is used to disabled tagged queuing to a device, yet leave | |
886 | * queue in function. | |
887 | **/ | |
888 | void blk_queue_free_tags(request_queue_t *q) | |
889 | { | |
890 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
891 | } | |
892 | ||
893 | EXPORT_SYMBOL(blk_queue_free_tags); | |
894 | ||
895 | static int | |
896 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) | |
897 | { | |
1da177e4 LT |
898 | struct request **tag_index; |
899 | unsigned long *tag_map; | |
fa72b903 | 900 | int nr_ulongs; |
1da177e4 | 901 | |
492dfb48 | 902 | if (q && depth > q->nr_requests * 2) { |
1da177e4 LT |
903 | depth = q->nr_requests * 2; |
904 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
905 | __FUNCTION__, depth); | |
906 | } | |
907 | ||
f68110fc | 908 | tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
1da177e4 LT |
909 | if (!tag_index) |
910 | goto fail; | |
911 | ||
f7d37d02 | 912 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
f68110fc | 913 | tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
1da177e4 LT |
914 | if (!tag_map) |
915 | goto fail; | |
916 | ||
ba025082 | 917 | tags->real_max_depth = depth; |
1da177e4 | 918 | tags->max_depth = depth; |
1da177e4 LT |
919 | tags->tag_index = tag_index; |
920 | tags->tag_map = tag_map; | |
921 | ||
1da177e4 LT |
922 | return 0; |
923 | fail: | |
924 | kfree(tag_index); | |
925 | return -ENOMEM; | |
926 | } | |
927 | ||
492dfb48 JB |
928 | static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q, |
929 | int depth) | |
930 | { | |
931 | struct blk_queue_tag *tags; | |
932 | ||
933 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
934 | if (!tags) | |
935 | goto fail; | |
936 | ||
937 | if (init_tag_map(q, tags, depth)) | |
938 | goto fail; | |
939 | ||
940 | INIT_LIST_HEAD(&tags->busy_list); | |
941 | tags->busy = 0; | |
942 | atomic_set(&tags->refcnt, 1); | |
943 | return tags; | |
944 | fail: | |
945 | kfree(tags); | |
946 | return NULL; | |
947 | } | |
948 | ||
949 | /** | |
950 | * blk_init_tags - initialize the tag info for an external tag map | |
951 | * @depth: the maximum queue depth supported | |
952 | * @tags: the tag to use | |
953 | **/ | |
954 | struct blk_queue_tag *blk_init_tags(int depth) | |
955 | { | |
956 | return __blk_queue_init_tags(NULL, depth); | |
957 | } | |
958 | EXPORT_SYMBOL(blk_init_tags); | |
959 | ||
1da177e4 LT |
960 | /** |
961 | * blk_queue_init_tags - initialize the queue tag info | |
962 | * @q: the request queue for the device | |
963 | * @depth: the maximum queue depth supported | |
964 | * @tags: the tag to use | |
965 | **/ | |
966 | int blk_queue_init_tags(request_queue_t *q, int depth, | |
967 | struct blk_queue_tag *tags) | |
968 | { | |
969 | int rc; | |
970 | ||
971 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
972 | ||
973 | if (!tags && !q->queue_tags) { | |
492dfb48 | 974 | tags = __blk_queue_init_tags(q, depth); |
1da177e4 | 975 | |
492dfb48 | 976 | if (!tags) |
1da177e4 | 977 | goto fail; |
1da177e4 LT |
978 | } else if (q->queue_tags) { |
979 | if ((rc = blk_queue_resize_tags(q, depth))) | |
980 | return rc; | |
981 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
982 | return 0; | |
983 | } else | |
984 | atomic_inc(&tags->refcnt); | |
985 | ||
986 | /* | |
987 | * assign it, all done | |
988 | */ | |
989 | q->queue_tags = tags; | |
990 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
991 | return 0; | |
992 | fail: | |
993 | kfree(tags); | |
994 | return -ENOMEM; | |
995 | } | |
996 | ||
997 | EXPORT_SYMBOL(blk_queue_init_tags); | |
998 | ||
999 | /** | |
1000 | * blk_queue_resize_tags - change the queueing depth | |
1001 | * @q: the request queue for the device | |
1002 | * @new_depth: the new max command queueing depth | |
1003 | * | |
1004 | * Notes: | |
1005 | * Must be called with the queue lock held. | |
1006 | **/ | |
1007 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) | |
1008 | { | |
1009 | struct blk_queue_tag *bqt = q->queue_tags; | |
1010 | struct request **tag_index; | |
1011 | unsigned long *tag_map; | |
fa72b903 | 1012 | int max_depth, nr_ulongs; |
1da177e4 LT |
1013 | |
1014 | if (!bqt) | |
1015 | return -ENXIO; | |
1016 | ||
ba025082 TH |
1017 | /* |
1018 | * if we already have large enough real_max_depth. just | |
1019 | * adjust max_depth. *NOTE* as requests with tag value | |
1020 | * between new_depth and real_max_depth can be in-flight, tag | |
1021 | * map can not be shrunk blindly here. | |
1022 | */ | |
1023 | if (new_depth <= bqt->real_max_depth) { | |
1024 | bqt->max_depth = new_depth; | |
1025 | return 0; | |
1026 | } | |
1027 | ||
492dfb48 JB |
1028 | /* |
1029 | * Currently cannot replace a shared tag map with a new | |
1030 | * one, so error out if this is the case | |
1031 | */ | |
1032 | if (atomic_read(&bqt->refcnt) != 1) | |
1033 | return -EBUSY; | |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * save the old state info, so we can copy it back | |
1037 | */ | |
1038 | tag_index = bqt->tag_index; | |
1039 | tag_map = bqt->tag_map; | |
ba025082 | 1040 | max_depth = bqt->real_max_depth; |
1da177e4 LT |
1041 | |
1042 | if (init_tag_map(q, bqt, new_depth)) | |
1043 | return -ENOMEM; | |
1044 | ||
1045 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
f7d37d02 | 1046 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 1047 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
1da177e4 LT |
1048 | |
1049 | kfree(tag_index); | |
1050 | kfree(tag_map); | |
1051 | return 0; | |
1052 | } | |
1053 | ||
1054 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
1055 | ||
1056 | /** | |
1057 | * blk_queue_end_tag - end tag operations for a request | |
1058 | * @q: the request queue for the device | |
1059 | * @rq: the request that has completed | |
1060 | * | |
1061 | * Description: | |
1062 | * Typically called when end_that_request_first() returns 0, meaning | |
1063 | * all transfers have been done for a request. It's important to call | |
1064 | * this function before end_that_request_last(), as that will put the | |
1065 | * request back on the free list thus corrupting the internal tag list. | |
1066 | * | |
1067 | * Notes: | |
1068 | * queue lock must be held. | |
1069 | **/ | |
1070 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) | |
1071 | { | |
1072 | struct blk_queue_tag *bqt = q->queue_tags; | |
1073 | int tag = rq->tag; | |
1074 | ||
1075 | BUG_ON(tag == -1); | |
1076 | ||
ba025082 | 1077 | if (unlikely(tag >= bqt->real_max_depth)) |
040c928c TH |
1078 | /* |
1079 | * This can happen after tag depth has been reduced. | |
1080 | * FIXME: how about a warning or info message here? | |
1081 | */ | |
1da177e4 LT |
1082 | return; |
1083 | ||
1084 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { | |
040c928c TH |
1085 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
1086 | __FUNCTION__, tag); | |
1da177e4 LT |
1087 | return; |
1088 | } | |
1089 | ||
1090 | list_del_init(&rq->queuelist); | |
4aff5e23 | 1091 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1092 | rq->tag = -1; |
1093 | ||
1094 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
040c928c TH |
1095 | printk(KERN_ERR "%s: tag %d is missing\n", |
1096 | __FUNCTION__, tag); | |
1da177e4 LT |
1097 | |
1098 | bqt->tag_index[tag] = NULL; | |
1099 | bqt->busy--; | |
1100 | } | |
1101 | ||
1102 | EXPORT_SYMBOL(blk_queue_end_tag); | |
1103 | ||
1104 | /** | |
1105 | * blk_queue_start_tag - find a free tag and assign it | |
1106 | * @q: the request queue for the device | |
1107 | * @rq: the block request that needs tagging | |
1108 | * | |
1109 | * Description: | |
1110 | * This can either be used as a stand-alone helper, or possibly be | |
1111 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
1112 | * automagically gets a tag assigned). Note that this function | |
1113 | * assumes that any type of request can be queued! if this is not | |
1114 | * true for your device, you must check the request type before | |
1115 | * calling this function. The request will also be removed from | |
1116 | * the request queue, so it's the drivers responsibility to readd | |
1117 | * it if it should need to be restarted for some reason. | |
1118 | * | |
1119 | * Notes: | |
1120 | * queue lock must be held. | |
1121 | **/ | |
1122 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) | |
1123 | { | |
1124 | struct blk_queue_tag *bqt = q->queue_tags; | |
2bf0fdad | 1125 | int tag; |
1da177e4 | 1126 | |
4aff5e23 | 1127 | if (unlikely((rq->cmd_flags & REQ_QUEUED))) { |
1da177e4 | 1128 | printk(KERN_ERR |
040c928c TH |
1129 | "%s: request %p for device [%s] already tagged %d", |
1130 | __FUNCTION__, rq, | |
1131 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1da177e4 LT |
1132 | BUG(); |
1133 | } | |
1134 | ||
059af497 JA |
1135 | /* |
1136 | * Protect against shared tag maps, as we may not have exclusive | |
1137 | * access to the tag map. | |
1138 | */ | |
1139 | do { | |
1140 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | |
1141 | if (tag >= bqt->max_depth) | |
1142 | return 1; | |
1da177e4 | 1143 | |
059af497 | 1144 | } while (test_and_set_bit(tag, bqt->tag_map)); |
1da177e4 | 1145 | |
4aff5e23 | 1146 | rq->cmd_flags |= REQ_QUEUED; |
1da177e4 LT |
1147 | rq->tag = tag; |
1148 | bqt->tag_index[tag] = rq; | |
1149 | blkdev_dequeue_request(rq); | |
1150 | list_add(&rq->queuelist, &bqt->busy_list); | |
1151 | bqt->busy++; | |
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1156 | ||
1157 | /** | |
1158 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1159 | * @q: the request queue for the device | |
1160 | * | |
1161 | * Description: | |
1162 | * Hardware conditions may dictate a need to stop all pending requests. | |
1163 | * In this case, we will safely clear the block side of the tag queue and | |
1164 | * readd all requests to the request queue in the right order. | |
1165 | * | |
1166 | * Notes: | |
1167 | * queue lock must be held. | |
1168 | **/ | |
1169 | void blk_queue_invalidate_tags(request_queue_t *q) | |
1170 | { | |
1171 | struct blk_queue_tag *bqt = q->queue_tags; | |
1172 | struct list_head *tmp, *n; | |
1173 | struct request *rq; | |
1174 | ||
1175 | list_for_each_safe(tmp, n, &bqt->busy_list) { | |
1176 | rq = list_entry_rq(tmp); | |
1177 | ||
1178 | if (rq->tag == -1) { | |
040c928c TH |
1179 | printk(KERN_ERR |
1180 | "%s: bad tag found on list\n", __FUNCTION__); | |
1da177e4 | 1181 | list_del_init(&rq->queuelist); |
4aff5e23 | 1182 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1183 | } else |
1184 | blk_queue_end_tag(q, rq); | |
1185 | ||
4aff5e23 | 1186 | rq->cmd_flags &= ~REQ_STARTED; |
1da177e4 LT |
1187 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
1188 | } | |
1189 | } | |
1190 | ||
1191 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1192 | ||
1da177e4 LT |
1193 | void blk_dump_rq_flags(struct request *rq, char *msg) |
1194 | { | |
1195 | int bit; | |
1196 | ||
4aff5e23 JA |
1197 | printk("%s: dev %s: type=%x, flags=%x\n", msg, |
1198 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
1199 | rq->cmd_flags); | |
1da177e4 LT |
1200 | |
1201 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1202 | rq->nr_sectors, | |
1203 | rq->current_nr_sectors); | |
1204 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1205 | ||
4aff5e23 | 1206 | if (blk_pc_request(rq)) { |
1da177e4 LT |
1207 | printk("cdb: "); |
1208 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1209 | printk("%02x ", rq->cmd[bit]); | |
1210 | printk("\n"); | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1215 | ||
1216 | void blk_recount_segments(request_queue_t *q, struct bio *bio) | |
1217 | { | |
1218 | struct bio_vec *bv, *bvprv = NULL; | |
1219 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; | |
1220 | int high, highprv = 1; | |
1221 | ||
1222 | if (unlikely(!bio->bi_io_vec)) | |
1223 | return; | |
1224 | ||
1225 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1226 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; | |
1227 | bio_for_each_segment(bv, bio, i) { | |
1228 | /* | |
1229 | * the trick here is making sure that a high page is never | |
1230 | * considered part of another segment, since that might | |
1231 | * change with the bounce page. | |
1232 | */ | |
f772b3d9 | 1233 | high = page_to_pfn(bv->bv_page) > q->bounce_pfn; |
1da177e4 LT |
1234 | if (high || highprv) |
1235 | goto new_hw_segment; | |
1236 | if (cluster) { | |
1237 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1238 | goto new_segment; | |
1239 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1240 | goto new_segment; | |
1241 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1242 | goto new_segment; | |
1243 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1244 | goto new_hw_segment; | |
1245 | ||
1246 | seg_size += bv->bv_len; | |
1247 | hw_seg_size += bv->bv_len; | |
1248 | bvprv = bv; | |
1249 | continue; | |
1250 | } | |
1251 | new_segment: | |
1252 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
1253 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { | |
1254 | hw_seg_size += bv->bv_len; | |
1255 | } else { | |
1256 | new_hw_segment: | |
1257 | if (hw_seg_size > bio->bi_hw_front_size) | |
1258 | bio->bi_hw_front_size = hw_seg_size; | |
1259 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | |
1260 | nr_hw_segs++; | |
1261 | } | |
1262 | ||
1263 | nr_phys_segs++; | |
1264 | bvprv = bv; | |
1265 | seg_size = bv->bv_len; | |
1266 | highprv = high; | |
1267 | } | |
1268 | if (hw_seg_size > bio->bi_hw_back_size) | |
1269 | bio->bi_hw_back_size = hw_seg_size; | |
1270 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) | |
1271 | bio->bi_hw_front_size = hw_seg_size; | |
1272 | bio->bi_phys_segments = nr_phys_segs; | |
1273 | bio->bi_hw_segments = nr_hw_segs; | |
1274 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1275 | } | |
387bb173 | 1276 | EXPORT_SYMBOL(blk_recount_segments); |
1da177e4 | 1277 | |
93d17d3d | 1278 | static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1279 | struct bio *nxt) |
1280 | { | |
1281 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1282 | return 0; | |
1283 | ||
1284 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1285 | return 0; | |
1286 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1287 | return 0; | |
1288 | ||
1289 | /* | |
1290 | * bio and nxt are contigous in memory, check if the queue allows | |
1291 | * these two to be merged into one | |
1292 | */ | |
1293 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1294 | return 1; | |
1295 | ||
1296 | return 0; | |
1297 | } | |
1298 | ||
93d17d3d | 1299 | static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1300 | struct bio *nxt) |
1301 | { | |
1302 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1303 | blk_recount_segments(q, bio); | |
1304 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1305 | blk_recount_segments(q, nxt); | |
1306 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
1307 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) | |
1308 | return 0; | |
1309 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1310 | return 0; | |
1311 | ||
1312 | return 1; | |
1313 | } | |
1314 | ||
1da177e4 LT |
1315 | /* |
1316 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1317 | * must make sure sg can hold rq->nr_phys_segments entries | |
1318 | */ | |
1319 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) | |
1320 | { | |
1321 | struct bio_vec *bvec, *bvprv; | |
1322 | struct bio *bio; | |
1323 | int nsegs, i, cluster; | |
1324 | ||
1325 | nsegs = 0; | |
1326 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1327 | ||
1328 | /* | |
1329 | * for each bio in rq | |
1330 | */ | |
1331 | bvprv = NULL; | |
1332 | rq_for_each_bio(bio, rq) { | |
1333 | /* | |
1334 | * for each segment in bio | |
1335 | */ | |
1336 | bio_for_each_segment(bvec, bio, i) { | |
1337 | int nbytes = bvec->bv_len; | |
1338 | ||
1339 | if (bvprv && cluster) { | |
1340 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) | |
1341 | goto new_segment; | |
1342 | ||
1343 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | |
1344 | goto new_segment; | |
1345 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1346 | goto new_segment; | |
1347 | ||
1348 | sg[nsegs - 1].length += nbytes; | |
1349 | } else { | |
1350 | new_segment: | |
1351 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); | |
1352 | sg[nsegs].page = bvec->bv_page; | |
1353 | sg[nsegs].length = nbytes; | |
1354 | sg[nsegs].offset = bvec->bv_offset; | |
1355 | ||
1356 | nsegs++; | |
1357 | } | |
1358 | bvprv = bvec; | |
1359 | } /* segments in bio */ | |
1360 | } /* bios in rq */ | |
1361 | ||
1362 | return nsegs; | |
1363 | } | |
1364 | ||
1365 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1366 | ||
1367 | /* | |
1368 | * the standard queue merge functions, can be overridden with device | |
1369 | * specific ones if so desired | |
1370 | */ | |
1371 | ||
1372 | static inline int ll_new_mergeable(request_queue_t *q, | |
1373 | struct request *req, | |
1374 | struct bio *bio) | |
1375 | { | |
1376 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1377 | ||
1378 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1379 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1380 | if (req == q->last_merge) |
1381 | q->last_merge = NULL; | |
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * A hw segment is just getting larger, bump just the phys | |
1387 | * counter. | |
1388 | */ | |
1389 | req->nr_phys_segments += nr_phys_segs; | |
1390 | return 1; | |
1391 | } | |
1392 | ||
1393 | static inline int ll_new_hw_segment(request_queue_t *q, | |
1394 | struct request *req, | |
1395 | struct bio *bio) | |
1396 | { | |
1397 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1398 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1399 | ||
1400 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1401 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1402 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1403 | if (req == q->last_merge) |
1404 | q->last_merge = NULL; | |
1405 | return 0; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * This will form the start of a new hw segment. Bump both | |
1410 | * counters. | |
1411 | */ | |
1412 | req->nr_hw_segments += nr_hw_segs; | |
1413 | req->nr_phys_segments += nr_phys_segs; | |
1414 | return 1; | |
1415 | } | |
1416 | ||
1aa4f24f | 1417 | int ll_back_merge_fn(request_queue_t *q, struct request *req, struct bio *bio) |
1da177e4 | 1418 | { |
defd94b7 | 1419 | unsigned short max_sectors; |
1da177e4 LT |
1420 | int len; |
1421 | ||
defd94b7 MC |
1422 | if (unlikely(blk_pc_request(req))) |
1423 | max_sectors = q->max_hw_sectors; | |
1424 | else | |
1425 | max_sectors = q->max_sectors; | |
1426 | ||
1427 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1428 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1429 | if (req == q->last_merge) |
1430 | q->last_merge = NULL; | |
1431 | return 0; | |
1432 | } | |
1433 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1434 | blk_recount_segments(q, req->biotail); | |
1435 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1436 | blk_recount_segments(q, bio); | |
1437 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1438 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1439 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1440 | int mergeable = ll_new_mergeable(q, req, bio); | |
1441 | ||
1442 | if (mergeable) { | |
1443 | if (req->nr_hw_segments == 1) | |
1444 | req->bio->bi_hw_front_size = len; | |
1445 | if (bio->bi_hw_segments == 1) | |
1446 | bio->bi_hw_back_size = len; | |
1447 | } | |
1448 | return mergeable; | |
1449 | } | |
1450 | ||
1451 | return ll_new_hw_segment(q, req, bio); | |
1452 | } | |
1aa4f24f | 1453 | EXPORT_SYMBOL(ll_back_merge_fn); |
1da177e4 LT |
1454 | |
1455 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, | |
1456 | struct bio *bio) | |
1457 | { | |
defd94b7 | 1458 | unsigned short max_sectors; |
1da177e4 LT |
1459 | int len; |
1460 | ||
defd94b7 MC |
1461 | if (unlikely(blk_pc_request(req))) |
1462 | max_sectors = q->max_hw_sectors; | |
1463 | else | |
1464 | max_sectors = q->max_sectors; | |
1465 | ||
1466 | ||
1467 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1468 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1469 | if (req == q->last_merge) |
1470 | q->last_merge = NULL; | |
1471 | return 0; | |
1472 | } | |
1473 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1474 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1475 | blk_recount_segments(q, bio); | |
1476 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1477 | blk_recount_segments(q, req->bio); | |
1478 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1479 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1480 | int mergeable = ll_new_mergeable(q, req, bio); | |
1481 | ||
1482 | if (mergeable) { | |
1483 | if (bio->bi_hw_segments == 1) | |
1484 | bio->bi_hw_front_size = len; | |
1485 | if (req->nr_hw_segments == 1) | |
1486 | req->biotail->bi_hw_back_size = len; | |
1487 | } | |
1488 | return mergeable; | |
1489 | } | |
1490 | ||
1491 | return ll_new_hw_segment(q, req, bio); | |
1492 | } | |
1493 | ||
1494 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, | |
1495 | struct request *next) | |
1496 | { | |
dfa1a553 ND |
1497 | int total_phys_segments; |
1498 | int total_hw_segments; | |
1da177e4 LT |
1499 | |
1500 | /* | |
1501 | * First check if the either of the requests are re-queued | |
1502 | * requests. Can't merge them if they are. | |
1503 | */ | |
1504 | if (req->special || next->special) | |
1505 | return 0; | |
1506 | ||
1507 | /* | |
dfa1a553 | 1508 | * Will it become too large? |
1da177e4 LT |
1509 | */ |
1510 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1511 | return 0; | |
1512 | ||
1513 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1514 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1515 | total_phys_segments--; | |
1516 | ||
1517 | if (total_phys_segments > q->max_phys_segments) | |
1518 | return 0; | |
1519 | ||
1520 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1521 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1522 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1523 | /* | |
1524 | * propagate the combined length to the end of the requests | |
1525 | */ | |
1526 | if (req->nr_hw_segments == 1) | |
1527 | req->bio->bi_hw_front_size = len; | |
1528 | if (next->nr_hw_segments == 1) | |
1529 | next->biotail->bi_hw_back_size = len; | |
1530 | total_hw_segments--; | |
1531 | } | |
1532 | ||
1533 | if (total_hw_segments > q->max_hw_segments) | |
1534 | return 0; | |
1535 | ||
1536 | /* Merge is OK... */ | |
1537 | req->nr_phys_segments = total_phys_segments; | |
1538 | req->nr_hw_segments = total_hw_segments; | |
1539 | return 1; | |
1540 | } | |
1541 | ||
1542 | /* | |
1543 | * "plug" the device if there are no outstanding requests: this will | |
1544 | * force the transfer to start only after we have put all the requests | |
1545 | * on the list. | |
1546 | * | |
1547 | * This is called with interrupts off and no requests on the queue and | |
1548 | * with the queue lock held. | |
1549 | */ | |
1550 | void blk_plug_device(request_queue_t *q) | |
1551 | { | |
1552 | WARN_ON(!irqs_disabled()); | |
1553 | ||
1554 | /* | |
1555 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1556 | * which will restart the queueing | |
1557 | */ | |
7daac490 | 1558 | if (blk_queue_stopped(q)) |
1da177e4 LT |
1559 | return; |
1560 | ||
2056a782 | 1561 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 1562 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
1563 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
1564 | } | |
1da177e4 LT |
1565 | } |
1566 | ||
1567 | EXPORT_SYMBOL(blk_plug_device); | |
1568 | ||
1569 | /* | |
1570 | * remove the queue from the plugged list, if present. called with | |
1571 | * queue lock held and interrupts disabled. | |
1572 | */ | |
1573 | int blk_remove_plug(request_queue_t *q) | |
1574 | { | |
1575 | WARN_ON(!irqs_disabled()); | |
1576 | ||
1577 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1578 | return 0; | |
1579 | ||
1580 | del_timer(&q->unplug_timer); | |
1581 | return 1; | |
1582 | } | |
1583 | ||
1584 | EXPORT_SYMBOL(blk_remove_plug); | |
1585 | ||
1586 | /* | |
1587 | * remove the plug and let it rip.. | |
1588 | */ | |
1589 | void __generic_unplug_device(request_queue_t *q) | |
1590 | { | |
7daac490 | 1591 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
1592 | return; |
1593 | ||
1594 | if (!blk_remove_plug(q)) | |
1595 | return; | |
1596 | ||
22e2c507 | 1597 | q->request_fn(q); |
1da177e4 LT |
1598 | } |
1599 | EXPORT_SYMBOL(__generic_unplug_device); | |
1600 | ||
1601 | /** | |
1602 | * generic_unplug_device - fire a request queue | |
1603 | * @q: The &request_queue_t in question | |
1604 | * | |
1605 | * Description: | |
1606 | * Linux uses plugging to build bigger requests queues before letting | |
1607 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1608 | * is still adding and merging requests on the queue. Once the queue | |
1609 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1610 | * transfers started. | |
1611 | **/ | |
1612 | void generic_unplug_device(request_queue_t *q) | |
1613 | { | |
1614 | spin_lock_irq(q->queue_lock); | |
1615 | __generic_unplug_device(q); | |
1616 | spin_unlock_irq(q->queue_lock); | |
1617 | } | |
1618 | EXPORT_SYMBOL(generic_unplug_device); | |
1619 | ||
1620 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1621 | struct page *page) | |
1622 | { | |
1623 | request_queue_t *q = bdi->unplug_io_data; | |
1624 | ||
1625 | /* | |
1626 | * devices don't necessarily have an ->unplug_fn defined | |
1627 | */ | |
2056a782 JA |
1628 | if (q->unplug_fn) { |
1629 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1630 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1631 | ||
1da177e4 | 1632 | q->unplug_fn(q); |
2056a782 | 1633 | } |
1da177e4 LT |
1634 | } |
1635 | ||
65f27f38 | 1636 | static void blk_unplug_work(struct work_struct *work) |
1da177e4 | 1637 | { |
65f27f38 | 1638 | request_queue_t *q = container_of(work, request_queue_t, unplug_work); |
1da177e4 | 1639 | |
2056a782 JA |
1640 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
1641 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1642 | ||
1da177e4 LT |
1643 | q->unplug_fn(q); |
1644 | } | |
1645 | ||
1646 | static void blk_unplug_timeout(unsigned long data) | |
1647 | { | |
1648 | request_queue_t *q = (request_queue_t *)data; | |
1649 | ||
2056a782 JA |
1650 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
1651 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1652 | ||
1da177e4 LT |
1653 | kblockd_schedule_work(&q->unplug_work); |
1654 | } | |
1655 | ||
1656 | /** | |
1657 | * blk_start_queue - restart a previously stopped queue | |
1658 | * @q: The &request_queue_t in question | |
1659 | * | |
1660 | * Description: | |
1661 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1662 | * the request_fn for the queue if it was in a stopped state when | |
1663 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1664 | **/ | |
1665 | void blk_start_queue(request_queue_t *q) | |
1666 | { | |
a038e253 PBG |
1667 | WARN_ON(!irqs_disabled()); |
1668 | ||
1da177e4 LT |
1669 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1670 | ||
1671 | /* | |
1672 | * one level of recursion is ok and is much faster than kicking | |
1673 | * the unplug handling | |
1674 | */ | |
1675 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1676 | q->request_fn(q); | |
1677 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1678 | } else { | |
1679 | blk_plug_device(q); | |
1680 | kblockd_schedule_work(&q->unplug_work); | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | EXPORT_SYMBOL(blk_start_queue); | |
1685 | ||
1686 | /** | |
1687 | * blk_stop_queue - stop a queue | |
1688 | * @q: The &request_queue_t in question | |
1689 | * | |
1690 | * Description: | |
1691 | * The Linux block layer assumes that a block driver will consume all | |
1692 | * entries on the request queue when the request_fn strategy is called. | |
1693 | * Often this will not happen, because of hardware limitations (queue | |
1694 | * depth settings). If a device driver gets a 'queue full' response, | |
1695 | * or if it simply chooses not to queue more I/O at one point, it can | |
1696 | * call this function to prevent the request_fn from being called until | |
1697 | * the driver has signalled it's ready to go again. This happens by calling | |
1698 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1699 | **/ | |
1700 | void blk_stop_queue(request_queue_t *q) | |
1701 | { | |
1702 | blk_remove_plug(q); | |
1703 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1704 | } | |
1705 | EXPORT_SYMBOL(blk_stop_queue); | |
1706 | ||
1707 | /** | |
1708 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1709 | * @q: the queue | |
1710 | * | |
1711 | * Description: | |
1712 | * The block layer may perform asynchronous callback activity | |
1713 | * on a queue, such as calling the unplug function after a timeout. | |
1714 | * A block device may call blk_sync_queue to ensure that any | |
1715 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 1716 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
1717 | * that its ->make_request_fn will not re-add plugging prior to calling |
1718 | * this function. | |
1719 | * | |
1720 | */ | |
1721 | void blk_sync_queue(struct request_queue *q) | |
1722 | { | |
1723 | del_timer_sync(&q->unplug_timer); | |
1da177e4 LT |
1724 | } |
1725 | EXPORT_SYMBOL(blk_sync_queue); | |
1726 | ||
1727 | /** | |
1728 | * blk_run_queue - run a single device queue | |
1729 | * @q: The queue to run | |
1730 | */ | |
1731 | void blk_run_queue(struct request_queue *q) | |
1732 | { | |
1733 | unsigned long flags; | |
1734 | ||
1735 | spin_lock_irqsave(q->queue_lock, flags); | |
1736 | blk_remove_plug(q); | |
dac07ec1 JA |
1737 | |
1738 | /* | |
1739 | * Only recurse once to avoid overrunning the stack, let the unplug | |
1740 | * handling reinvoke the handler shortly if we already got there. | |
1741 | */ | |
1742 | if (!elv_queue_empty(q)) { | |
1743 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1744 | q->request_fn(q); | |
1745 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1746 | } else { | |
1747 | blk_plug_device(q); | |
1748 | kblockd_schedule_work(&q->unplug_work); | |
1749 | } | |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | spin_unlock_irqrestore(q->queue_lock, flags); |
1753 | } | |
1754 | EXPORT_SYMBOL(blk_run_queue); | |
1755 | ||
1756 | /** | |
1757 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed | |
a580290c | 1758 | * @kobj: the kobj belonging of the request queue to be released |
1da177e4 LT |
1759 | * |
1760 | * Description: | |
1761 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1762 | * blk_queue_make_request(). It should be called when a request queue is | |
1763 | * being released; typically when a block device is being de-registered. | |
1764 | * Currently, its primary task it to free all the &struct request | |
1765 | * structures that were allocated to the queue and the queue itself. | |
1766 | * | |
1767 | * Caveat: | |
1768 | * Hopefully the low level driver will have finished any | |
1769 | * outstanding requests first... | |
1770 | **/ | |
483f4afc | 1771 | static void blk_release_queue(struct kobject *kobj) |
1da177e4 | 1772 | { |
483f4afc | 1773 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
1da177e4 LT |
1774 | struct request_list *rl = &q->rq; |
1775 | ||
1da177e4 LT |
1776 | blk_sync_queue(q); |
1777 | ||
1778 | if (rl->rq_pool) | |
1779 | mempool_destroy(rl->rq_pool); | |
1780 | ||
1781 | if (q->queue_tags) | |
1782 | __blk_queue_free_tags(q); | |
1783 | ||
6c5c9341 | 1784 | blk_trace_shutdown(q); |
2056a782 | 1785 | |
1da177e4 LT |
1786 | kmem_cache_free(requestq_cachep, q); |
1787 | } | |
1788 | ||
483f4afc AV |
1789 | void blk_put_queue(request_queue_t *q) |
1790 | { | |
1791 | kobject_put(&q->kobj); | |
1792 | } | |
1793 | EXPORT_SYMBOL(blk_put_queue); | |
1794 | ||
1795 | void blk_cleanup_queue(request_queue_t * q) | |
1796 | { | |
1797 | mutex_lock(&q->sysfs_lock); | |
1798 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
1799 | mutex_unlock(&q->sysfs_lock); | |
1800 | ||
1801 | if (q->elevator) | |
1802 | elevator_exit(q->elevator); | |
1803 | ||
1804 | blk_put_queue(q); | |
1805 | } | |
1806 | ||
1da177e4 LT |
1807 | EXPORT_SYMBOL(blk_cleanup_queue); |
1808 | ||
1809 | static int blk_init_free_list(request_queue_t *q) | |
1810 | { | |
1811 | struct request_list *rl = &q->rq; | |
1812 | ||
1813 | rl->count[READ] = rl->count[WRITE] = 0; | |
1814 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 1815 | rl->elvpriv = 0; |
1da177e4 LT |
1816 | init_waitqueue_head(&rl->wait[READ]); |
1817 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 1818 | |
1946089a CL |
1819 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1820 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
1821 | |
1822 | if (!rl->rq_pool) | |
1823 | return -ENOMEM; | |
1824 | ||
1825 | return 0; | |
1826 | } | |
1827 | ||
8267e268 | 1828 | request_queue_t *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 1829 | { |
1946089a CL |
1830 | return blk_alloc_queue_node(gfp_mask, -1); |
1831 | } | |
1832 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 1833 | |
483f4afc AV |
1834 | static struct kobj_type queue_ktype; |
1835 | ||
8267e268 | 1836 | request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a CL |
1837 | { |
1838 | request_queue_t *q; | |
1839 | ||
1840 | q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id); | |
1da177e4 LT |
1841 | if (!q) |
1842 | return NULL; | |
1843 | ||
1844 | memset(q, 0, sizeof(*q)); | |
1845 | init_timer(&q->unplug_timer); | |
483f4afc AV |
1846 | |
1847 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); | |
1848 | q->kobj.ktype = &queue_ktype; | |
1849 | kobject_init(&q->kobj); | |
1da177e4 LT |
1850 | |
1851 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
1852 | q->backing_dev_info.unplug_io_data = q; | |
1853 | ||
483f4afc AV |
1854 | mutex_init(&q->sysfs_lock); |
1855 | ||
1da177e4 LT |
1856 | return q; |
1857 | } | |
1946089a | 1858 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1859 | |
1860 | /** | |
1861 | * blk_init_queue - prepare a request queue for use with a block device | |
1862 | * @rfn: The function to be called to process requests that have been | |
1863 | * placed on the queue. | |
1864 | * @lock: Request queue spin lock | |
1865 | * | |
1866 | * Description: | |
1867 | * If a block device wishes to use the standard request handling procedures, | |
1868 | * which sorts requests and coalesces adjacent requests, then it must | |
1869 | * call blk_init_queue(). The function @rfn will be called when there | |
1870 | * are requests on the queue that need to be processed. If the device | |
1871 | * supports plugging, then @rfn may not be called immediately when requests | |
1872 | * are available on the queue, but may be called at some time later instead. | |
1873 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1874 | * of the requests on the queue is needed, or due to memory pressure. | |
1875 | * | |
1876 | * @rfn is not required, or even expected, to remove all requests off the | |
1877 | * queue, but only as many as it can handle at a time. If it does leave | |
1878 | * requests on the queue, it is responsible for arranging that the requests | |
1879 | * get dealt with eventually. | |
1880 | * | |
1881 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1882 | * request queue; this lock will be taken also from interrupt context, so irq |
1883 | * disabling is needed for it. | |
1da177e4 LT |
1884 | * |
1885 | * Function returns a pointer to the initialized request queue, or NULL if | |
1886 | * it didn't succeed. | |
1887 | * | |
1888 | * Note: | |
1889 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1890 | * when the block device is deactivated (such as at module unload). | |
1891 | **/ | |
1946089a | 1892 | |
1da177e4 LT |
1893 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1894 | { | |
1946089a CL |
1895 | return blk_init_queue_node(rfn, lock, -1); |
1896 | } | |
1897 | EXPORT_SYMBOL(blk_init_queue); | |
1898 | ||
1899 | request_queue_t * | |
1900 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
1901 | { | |
1902 | request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
1da177e4 LT |
1903 | |
1904 | if (!q) | |
1905 | return NULL; | |
1906 | ||
1946089a | 1907 | q->node = node_id; |
8669aafd AV |
1908 | if (blk_init_free_list(q)) { |
1909 | kmem_cache_free(requestq_cachep, q); | |
1910 | return NULL; | |
1911 | } | |
1da177e4 | 1912 | |
152587de | 1913 | /* |
1914 | * if caller didn't supply a lock, they get per-queue locking with | |
1915 | * our embedded lock | |
1916 | */ | |
1917 | if (!lock) { | |
1918 | spin_lock_init(&q->__queue_lock); | |
1919 | lock = &q->__queue_lock; | |
1920 | } | |
1921 | ||
1da177e4 | 1922 | q->request_fn = rfn; |
1da177e4 LT |
1923 | q->prep_rq_fn = NULL; |
1924 | q->unplug_fn = generic_unplug_device; | |
1925 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1926 | q->queue_lock = lock; | |
1927 | ||
1928 | blk_queue_segment_boundary(q, 0xffffffff); | |
1929 | ||
1930 | blk_queue_make_request(q, __make_request); | |
1931 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1932 | ||
1933 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1934 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1935 | ||
44ec9542 AS |
1936 | q->sg_reserved_size = INT_MAX; |
1937 | ||
1da177e4 LT |
1938 | /* |
1939 | * all done | |
1940 | */ | |
1941 | if (!elevator_init(q, NULL)) { | |
1942 | blk_queue_congestion_threshold(q); | |
1943 | return q; | |
1944 | } | |
1945 | ||
8669aafd | 1946 | blk_put_queue(q); |
1da177e4 LT |
1947 | return NULL; |
1948 | } | |
1946089a | 1949 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 LT |
1950 | |
1951 | int blk_get_queue(request_queue_t *q) | |
1952 | { | |
fde6ad22 | 1953 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 1954 | kobject_get(&q->kobj); |
1da177e4 LT |
1955 | return 0; |
1956 | } | |
1957 | ||
1958 | return 1; | |
1959 | } | |
1960 | ||
1961 | EXPORT_SYMBOL(blk_get_queue); | |
1962 | ||
1963 | static inline void blk_free_request(request_queue_t *q, struct request *rq) | |
1964 | { | |
4aff5e23 | 1965 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 1966 | elv_put_request(q, rq); |
1da177e4 LT |
1967 | mempool_free(rq, q->rq.rq_pool); |
1968 | } | |
1969 | ||
1ea25ecb | 1970 | static struct request * |
cb78b285 | 1971 | blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
1972 | { |
1973 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
1974 | ||
1975 | if (!rq) | |
1976 | return NULL; | |
1977 | ||
1978 | /* | |
4aff5e23 | 1979 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
1980 | * see bio.h and blkdev.h |
1981 | */ | |
49171e5c | 1982 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 1983 | |
cb98fc8b | 1984 | if (priv) { |
cb78b285 | 1985 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
1986 | mempool_free(rq, q->rq.rq_pool); |
1987 | return NULL; | |
1988 | } | |
4aff5e23 | 1989 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 1990 | } |
1da177e4 | 1991 | |
cb98fc8b | 1992 | return rq; |
1da177e4 LT |
1993 | } |
1994 | ||
1995 | /* | |
1996 | * ioc_batching returns true if the ioc is a valid batching request and | |
1997 | * should be given priority access to a request. | |
1998 | */ | |
1999 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) | |
2000 | { | |
2001 | if (!ioc) | |
2002 | return 0; | |
2003 | ||
2004 | /* | |
2005 | * Make sure the process is able to allocate at least 1 request | |
2006 | * even if the batch times out, otherwise we could theoretically | |
2007 | * lose wakeups. | |
2008 | */ | |
2009 | return ioc->nr_batch_requests == q->nr_batching || | |
2010 | (ioc->nr_batch_requests > 0 | |
2011 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
2016 | * will cause the process to be a "batcher" on all queues in the system. This | |
2017 | * is the behaviour we want though - once it gets a wakeup it should be given | |
2018 | * a nice run. | |
2019 | */ | |
93d17d3d | 2020 | static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) |
1da177e4 LT |
2021 | { |
2022 | if (!ioc || ioc_batching(q, ioc)) | |
2023 | return; | |
2024 | ||
2025 | ioc->nr_batch_requests = q->nr_batching; | |
2026 | ioc->last_waited = jiffies; | |
2027 | } | |
2028 | ||
2029 | static void __freed_request(request_queue_t *q, int rw) | |
2030 | { | |
2031 | struct request_list *rl = &q->rq; | |
2032 | ||
2033 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 2034 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
2035 | |
2036 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
2037 | if (waitqueue_active(&rl->wait[rw])) |
2038 | wake_up(&rl->wait[rw]); | |
2039 | ||
2040 | blk_clear_queue_full(q, rw); | |
2041 | } | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * A request has just been released. Account for it, update the full and | |
2046 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
2047 | */ | |
cb98fc8b | 2048 | static void freed_request(request_queue_t *q, int rw, int priv) |
1da177e4 LT |
2049 | { |
2050 | struct request_list *rl = &q->rq; | |
2051 | ||
2052 | rl->count[rw]--; | |
cb98fc8b TH |
2053 | if (priv) |
2054 | rl->elvpriv--; | |
1da177e4 LT |
2055 | |
2056 | __freed_request(q, rw); | |
2057 | ||
2058 | if (unlikely(rl->starved[rw ^ 1])) | |
2059 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
2060 | } |
2061 | ||
2062 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
2063 | /* | |
d6344532 NP |
2064 | * Get a free request, queue_lock must be held. |
2065 | * Returns NULL on failure, with queue_lock held. | |
2066 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 2067 | */ |
7749a8d4 JA |
2068 | static struct request *get_request(request_queue_t *q, int rw_flags, |
2069 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 LT |
2070 | { |
2071 | struct request *rq = NULL; | |
2072 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 2073 | struct io_context *ioc = NULL; |
7749a8d4 | 2074 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
2075 | int may_queue, priv; |
2076 | ||
7749a8d4 | 2077 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
2078 | if (may_queue == ELV_MQUEUE_NO) |
2079 | goto rq_starved; | |
2080 | ||
2081 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
2082 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 2083 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
2084 | /* |
2085 | * The queue will fill after this allocation, so set | |
2086 | * it as full, and mark this process as "batching". | |
2087 | * This process will be allowed to complete a batch of | |
2088 | * requests, others will be blocked. | |
2089 | */ | |
2090 | if (!blk_queue_full(q, rw)) { | |
2091 | ioc_set_batching(q, ioc); | |
2092 | blk_set_queue_full(q, rw); | |
2093 | } else { | |
2094 | if (may_queue != ELV_MQUEUE_MUST | |
2095 | && !ioc_batching(q, ioc)) { | |
2096 | /* | |
2097 | * The queue is full and the allocating | |
2098 | * process is not a "batcher", and not | |
2099 | * exempted by the IO scheduler | |
2100 | */ | |
2101 | goto out; | |
2102 | } | |
2103 | } | |
1da177e4 | 2104 | } |
79e2de4b | 2105 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
2106 | } |
2107 | ||
082cf69e JA |
2108 | /* |
2109 | * Only allow batching queuers to allocate up to 50% over the defined | |
2110 | * limit of requests, otherwise we could have thousands of requests | |
2111 | * allocated with any setting of ->nr_requests | |
2112 | */ | |
fd782a4a | 2113 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 2114 | goto out; |
fd782a4a | 2115 | |
1da177e4 LT |
2116 | rl->count[rw]++; |
2117 | rl->starved[rw] = 0; | |
cb98fc8b | 2118 | |
64521d1a | 2119 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
2120 | if (priv) |
2121 | rl->elvpriv++; | |
2122 | ||
1da177e4 LT |
2123 | spin_unlock_irq(q->queue_lock); |
2124 | ||
7749a8d4 | 2125 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 2126 | if (unlikely(!rq)) { |
1da177e4 LT |
2127 | /* |
2128 | * Allocation failed presumably due to memory. Undo anything | |
2129 | * we might have messed up. | |
2130 | * | |
2131 | * Allocating task should really be put onto the front of the | |
2132 | * wait queue, but this is pretty rare. | |
2133 | */ | |
2134 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 2135 | freed_request(q, rw, priv); |
1da177e4 LT |
2136 | |
2137 | /* | |
2138 | * in the very unlikely event that allocation failed and no | |
2139 | * requests for this direction was pending, mark us starved | |
2140 | * so that freeing of a request in the other direction will | |
2141 | * notice us. another possible fix would be to split the | |
2142 | * rq mempool into READ and WRITE | |
2143 | */ | |
2144 | rq_starved: | |
2145 | if (unlikely(rl->count[rw] == 0)) | |
2146 | rl->starved[rw] = 1; | |
2147 | ||
1da177e4 LT |
2148 | goto out; |
2149 | } | |
2150 | ||
88ee5ef1 JA |
2151 | /* |
2152 | * ioc may be NULL here, and ioc_batching will be false. That's | |
2153 | * OK, if the queue is under the request limit then requests need | |
2154 | * not count toward the nr_batch_requests limit. There will always | |
2155 | * be some limit enforced by BLK_BATCH_TIME. | |
2156 | */ | |
1da177e4 LT |
2157 | if (ioc_batching(q, ioc)) |
2158 | ioc->nr_batch_requests--; | |
2159 | ||
2160 | rq_init(q, rq); | |
2056a782 JA |
2161 | |
2162 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 2163 | out: |
1da177e4 LT |
2164 | return rq; |
2165 | } | |
2166 | ||
2167 | /* | |
2168 | * No available requests for this queue, unplug the device and wait for some | |
2169 | * requests to become available. | |
d6344532 NP |
2170 | * |
2171 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 2172 | */ |
7749a8d4 | 2173 | static struct request *get_request_wait(request_queue_t *q, int rw_flags, |
22e2c507 | 2174 | struct bio *bio) |
1da177e4 | 2175 | { |
7749a8d4 | 2176 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
2177 | struct request *rq; |
2178 | ||
7749a8d4 | 2179 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
2180 | while (!rq) { |
2181 | DEFINE_WAIT(wait); | |
1da177e4 LT |
2182 | struct request_list *rl = &q->rq; |
2183 | ||
2184 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
2185 | TASK_UNINTERRUPTIBLE); | |
2186 | ||
7749a8d4 | 2187 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
2188 | |
2189 | if (!rq) { | |
2190 | struct io_context *ioc; | |
2191 | ||
2056a782 JA |
2192 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
2193 | ||
d6344532 NP |
2194 | __generic_unplug_device(q); |
2195 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
2196 | io_schedule(); |
2197 | ||
2198 | /* | |
2199 | * After sleeping, we become a "batching" process and | |
2200 | * will be able to allocate at least one request, and | |
2201 | * up to a big batch of them for a small period time. | |
2202 | * See ioc_batching, ioc_set_batching | |
2203 | */ | |
b5deef90 | 2204 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 2205 | ioc_set_batching(q, ioc); |
d6344532 NP |
2206 | |
2207 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
2208 | } |
2209 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 2210 | } |
1da177e4 LT |
2211 | |
2212 | return rq; | |
2213 | } | |
2214 | ||
8267e268 | 2215 | struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
2216 | { |
2217 | struct request *rq; | |
2218 | ||
2219 | BUG_ON(rw != READ && rw != WRITE); | |
2220 | ||
d6344532 NP |
2221 | spin_lock_irq(q->queue_lock); |
2222 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 2223 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 2224 | } else { |
22e2c507 | 2225 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
2226 | if (!rq) |
2227 | spin_unlock_irq(q->queue_lock); | |
2228 | } | |
2229 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
2230 | |
2231 | return rq; | |
2232 | } | |
1da177e4 LT |
2233 | EXPORT_SYMBOL(blk_get_request); |
2234 | ||
dc72ef4a JA |
2235 | /** |
2236 | * blk_start_queueing - initiate dispatch of requests to device | |
2237 | * @q: request queue to kick into gear | |
2238 | * | |
2239 | * This is basically a helper to remove the need to know whether a queue | |
2240 | * is plugged or not if someone just wants to initiate dispatch of requests | |
2241 | * for this queue. | |
2242 | * | |
2243 | * The queue lock must be held with interrupts disabled. | |
2244 | */ | |
2245 | void blk_start_queueing(request_queue_t *q) | |
2246 | { | |
2247 | if (!blk_queue_plugged(q)) | |
2248 | q->request_fn(q); | |
2249 | else | |
2250 | __generic_unplug_device(q); | |
2251 | } | |
2252 | EXPORT_SYMBOL(blk_start_queueing); | |
2253 | ||
1da177e4 LT |
2254 | /** |
2255 | * blk_requeue_request - put a request back on queue | |
2256 | * @q: request queue where request should be inserted | |
2257 | * @rq: request to be inserted | |
2258 | * | |
2259 | * Description: | |
2260 | * Drivers often keep queueing requests until the hardware cannot accept | |
2261 | * more, when that condition happens we need to put the request back | |
2262 | * on the queue. Must be called with queue lock held. | |
2263 | */ | |
2264 | void blk_requeue_request(request_queue_t *q, struct request *rq) | |
2265 | { | |
2056a782 JA |
2266 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
2267 | ||
1da177e4 LT |
2268 | if (blk_rq_tagged(rq)) |
2269 | blk_queue_end_tag(q, rq); | |
2270 | ||
2271 | elv_requeue_request(q, rq); | |
2272 | } | |
2273 | ||
2274 | EXPORT_SYMBOL(blk_requeue_request); | |
2275 | ||
2276 | /** | |
2277 | * blk_insert_request - insert a special request in to a request queue | |
2278 | * @q: request queue where request should be inserted | |
2279 | * @rq: request to be inserted | |
2280 | * @at_head: insert request at head or tail of queue | |
2281 | * @data: private data | |
1da177e4 LT |
2282 | * |
2283 | * Description: | |
2284 | * Many block devices need to execute commands asynchronously, so they don't | |
2285 | * block the whole kernel from preemption during request execution. This is | |
2286 | * accomplished normally by inserting aritficial requests tagged as | |
2287 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2288 | * scheduled for actual execution by the request queue. | |
2289 | * | |
2290 | * We have the option of inserting the head or the tail of the queue. | |
2291 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2292 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2293 | * host that is unable to accept a particular command. | |
2294 | */ | |
2295 | void blk_insert_request(request_queue_t *q, struct request *rq, | |
867d1191 | 2296 | int at_head, void *data) |
1da177e4 | 2297 | { |
867d1191 | 2298 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
2299 | unsigned long flags; |
2300 | ||
2301 | /* | |
2302 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2303 | * must not attempt merges on this) and that it acts as a soft | |
2304 | * barrier | |
2305 | */ | |
4aff5e23 JA |
2306 | rq->cmd_type = REQ_TYPE_SPECIAL; |
2307 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
2308 | |
2309 | rq->special = data; | |
2310 | ||
2311 | spin_lock_irqsave(q->queue_lock, flags); | |
2312 | ||
2313 | /* | |
2314 | * If command is tagged, release the tag | |
2315 | */ | |
867d1191 TH |
2316 | if (blk_rq_tagged(rq)) |
2317 | blk_queue_end_tag(q, rq); | |
1da177e4 | 2318 | |
867d1191 TH |
2319 | drive_stat_acct(rq, rq->nr_sectors, 1); |
2320 | __elv_add_request(q, rq, where, 0); | |
dc72ef4a | 2321 | blk_start_queueing(q); |
1da177e4 LT |
2322 | spin_unlock_irqrestore(q->queue_lock, flags); |
2323 | } | |
2324 | ||
2325 | EXPORT_SYMBOL(blk_insert_request); | |
2326 | ||
0e75f906 MC |
2327 | static int __blk_rq_unmap_user(struct bio *bio) |
2328 | { | |
2329 | int ret = 0; | |
2330 | ||
2331 | if (bio) { | |
2332 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2333 | bio_unmap_user(bio); | |
2334 | else | |
2335 | ret = bio_uncopy_user(bio); | |
2336 | } | |
2337 | ||
2338 | return ret; | |
2339 | } | |
2340 | ||
2341 | static int __blk_rq_map_user(request_queue_t *q, struct request *rq, | |
2342 | void __user *ubuf, unsigned int len) | |
2343 | { | |
2344 | unsigned long uaddr; | |
2345 | struct bio *bio, *orig_bio; | |
2346 | int reading, ret; | |
2347 | ||
2348 | reading = rq_data_dir(rq) == READ; | |
2349 | ||
2350 | /* | |
2351 | * if alignment requirement is satisfied, map in user pages for | |
2352 | * direct dma. else, set up kernel bounce buffers | |
2353 | */ | |
2354 | uaddr = (unsigned long) ubuf; | |
2355 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
2356 | bio = bio_map_user(q, NULL, uaddr, len, reading); | |
2357 | else | |
2358 | bio = bio_copy_user(q, uaddr, len, reading); | |
2359 | ||
2985259b | 2360 | if (IS_ERR(bio)) |
0e75f906 | 2361 | return PTR_ERR(bio); |
0e75f906 MC |
2362 | |
2363 | orig_bio = bio; | |
2364 | blk_queue_bounce(q, &bio); | |
2985259b | 2365 | |
0e75f906 MC |
2366 | /* |
2367 | * We link the bounce buffer in and could have to traverse it | |
2368 | * later so we have to get a ref to prevent it from being freed | |
2369 | */ | |
2370 | bio_get(bio); | |
2371 | ||
0e75f906 MC |
2372 | if (!rq->bio) |
2373 | blk_rq_bio_prep(q, rq, bio); | |
1aa4f24f | 2374 | else if (!ll_back_merge_fn(q, rq, bio)) { |
0e75f906 | 2375 | ret = -EINVAL; |
0e75f906 MC |
2376 | goto unmap_bio; |
2377 | } else { | |
2378 | rq->biotail->bi_next = bio; | |
2379 | rq->biotail = bio; | |
2380 | ||
0e75f906 MC |
2381 | rq->data_len += bio->bi_size; |
2382 | } | |
0e75f906 MC |
2383 | |
2384 | return bio->bi_size; | |
2385 | ||
2386 | unmap_bio: | |
2387 | /* if it was boucned we must call the end io function */ | |
2388 | bio_endio(bio, bio->bi_size, 0); | |
2389 | __blk_rq_unmap_user(orig_bio); | |
2390 | bio_put(bio); | |
2391 | return ret; | |
2392 | } | |
2393 | ||
1da177e4 LT |
2394 | /** |
2395 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2396 | * @q: request queue where request should be inserted | |
73747aed | 2397 | * @rq: request structure to fill |
1da177e4 LT |
2398 | * @ubuf: the user buffer |
2399 | * @len: length of user data | |
2400 | * | |
2401 | * Description: | |
2402 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2403 | * a kernel bounce buffer is used. | |
2404 | * | |
2405 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2406 | * still in process context. | |
2407 | * | |
2408 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2409 | * before being submitted to the device, as pages mapped may be out of | |
2410 | * reach. It's the callers responsibility to make sure this happens. The | |
2411 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2412 | * unmapping. | |
2413 | */ | |
dd1cab95 | 2414 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, |
0e75f906 | 2415 | unsigned long len) |
1da177e4 | 2416 | { |
0e75f906 | 2417 | unsigned long bytes_read = 0; |
8e5cfc45 | 2418 | struct bio *bio = NULL; |
0e75f906 | 2419 | int ret; |
1da177e4 | 2420 | |
defd94b7 | 2421 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2422 | return -EINVAL; |
2423 | if (!len || !ubuf) | |
2424 | return -EINVAL; | |
1da177e4 | 2425 | |
0e75f906 MC |
2426 | while (bytes_read != len) { |
2427 | unsigned long map_len, end, start; | |
1da177e4 | 2428 | |
0e75f906 MC |
2429 | map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE); |
2430 | end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1) | |
2431 | >> PAGE_SHIFT; | |
2432 | start = (unsigned long)ubuf >> PAGE_SHIFT; | |
1da177e4 | 2433 | |
0e75f906 MC |
2434 | /* |
2435 | * A bad offset could cause us to require BIO_MAX_PAGES + 1 | |
2436 | * pages. If this happens we just lower the requested | |
2437 | * mapping len by a page so that we can fit | |
2438 | */ | |
2439 | if (end - start > BIO_MAX_PAGES) | |
2440 | map_len -= PAGE_SIZE; | |
1da177e4 | 2441 | |
0e75f906 MC |
2442 | ret = __blk_rq_map_user(q, rq, ubuf, map_len); |
2443 | if (ret < 0) | |
2444 | goto unmap_rq; | |
8e5cfc45 JA |
2445 | if (!bio) |
2446 | bio = rq->bio; | |
0e75f906 MC |
2447 | bytes_read += ret; |
2448 | ubuf += ret; | |
1da177e4 LT |
2449 | } |
2450 | ||
0e75f906 MC |
2451 | rq->buffer = rq->data = NULL; |
2452 | return 0; | |
2453 | unmap_rq: | |
8e5cfc45 | 2454 | blk_rq_unmap_user(bio); |
0e75f906 | 2455 | return ret; |
1da177e4 LT |
2456 | } |
2457 | ||
2458 | EXPORT_SYMBOL(blk_rq_map_user); | |
2459 | ||
f1970baf JB |
2460 | /** |
2461 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2462 | * @q: request queue where request should be inserted | |
2463 | * @rq: request to map data to | |
2464 | * @iov: pointer to the iovec | |
2465 | * @iov_count: number of elements in the iovec | |
af9997e4 | 2466 | * @len: I/O byte count |
f1970baf JB |
2467 | * |
2468 | * Description: | |
2469 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2470 | * a kernel bounce buffer is used. | |
2471 | * | |
2472 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2473 | * still in process context. | |
2474 | * | |
2475 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2476 | * before being submitted to the device, as pages mapped may be out of | |
2477 | * reach. It's the callers responsibility to make sure this happens. The | |
2478 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2479 | * unmapping. | |
2480 | */ | |
2481 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, | |
0e75f906 | 2482 | struct sg_iovec *iov, int iov_count, unsigned int len) |
f1970baf JB |
2483 | { |
2484 | struct bio *bio; | |
2485 | ||
2486 | if (!iov || iov_count <= 0) | |
2487 | return -EINVAL; | |
2488 | ||
2489 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2490 | * user is using sg, they're expected to know the alignment constraints | |
2491 | * and respect them accordingly */ | |
2492 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2493 | if (IS_ERR(bio)) | |
2494 | return PTR_ERR(bio); | |
2495 | ||
0e75f906 MC |
2496 | if (bio->bi_size != len) { |
2497 | bio_endio(bio, bio->bi_size, 0); | |
2498 | bio_unmap_user(bio); | |
2499 | return -EINVAL; | |
2500 | } | |
2501 | ||
2502 | bio_get(bio); | |
f1970baf JB |
2503 | blk_rq_bio_prep(q, rq, bio); |
2504 | rq->buffer = rq->data = NULL; | |
f1970baf JB |
2505 | return 0; |
2506 | } | |
2507 | ||
2508 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2509 | ||
1da177e4 LT |
2510 | /** |
2511 | * blk_rq_unmap_user - unmap a request with user data | |
8e5cfc45 | 2512 | * @bio: start of bio list |
1da177e4 LT |
2513 | * |
2514 | * Description: | |
8e5cfc45 JA |
2515 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must |
2516 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
2517 | * the io completion may have changed rq->bio. | |
1da177e4 | 2518 | */ |
8e5cfc45 | 2519 | int blk_rq_unmap_user(struct bio *bio) |
1da177e4 | 2520 | { |
8e5cfc45 | 2521 | struct bio *mapped_bio; |
48785bb9 | 2522 | int ret = 0, ret2; |
1da177e4 | 2523 | |
8e5cfc45 JA |
2524 | while (bio) { |
2525 | mapped_bio = bio; | |
2526 | if (unlikely(bio_flagged(bio, BIO_BOUNCED))) | |
0e75f906 | 2527 | mapped_bio = bio->bi_private; |
1da177e4 | 2528 | |
48785bb9 JA |
2529 | ret2 = __blk_rq_unmap_user(mapped_bio); |
2530 | if (ret2 && !ret) | |
2531 | ret = ret2; | |
2532 | ||
8e5cfc45 JA |
2533 | mapped_bio = bio; |
2534 | bio = bio->bi_next; | |
2535 | bio_put(mapped_bio); | |
0e75f906 | 2536 | } |
48785bb9 JA |
2537 | |
2538 | return ret; | |
1da177e4 LT |
2539 | } |
2540 | ||
2541 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2542 | ||
df46b9a4 MC |
2543 | /** |
2544 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2545 | * @q: request queue where request should be inserted | |
73747aed | 2546 | * @rq: request to fill |
df46b9a4 MC |
2547 | * @kbuf: the kernel buffer |
2548 | * @len: length of user data | |
73747aed | 2549 | * @gfp_mask: memory allocation flags |
df46b9a4 | 2550 | */ |
dd1cab95 | 2551 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, |
8267e268 | 2552 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 | 2553 | { |
df46b9a4 MC |
2554 | struct bio *bio; |
2555 | ||
defd94b7 | 2556 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2557 | return -EINVAL; |
2558 | if (!len || !kbuf) | |
2559 | return -EINVAL; | |
df46b9a4 MC |
2560 | |
2561 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
dd1cab95 JA |
2562 | if (IS_ERR(bio)) |
2563 | return PTR_ERR(bio); | |
df46b9a4 | 2564 | |
dd1cab95 JA |
2565 | if (rq_data_dir(rq) == WRITE) |
2566 | bio->bi_rw |= (1 << BIO_RW); | |
df46b9a4 | 2567 | |
dd1cab95 | 2568 | blk_rq_bio_prep(q, rq, bio); |
821de3a2 | 2569 | blk_queue_bounce(q, &rq->bio); |
dd1cab95 | 2570 | rq->buffer = rq->data = NULL; |
dd1cab95 | 2571 | return 0; |
df46b9a4 MC |
2572 | } |
2573 | ||
2574 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2575 | ||
73747aed CH |
2576 | /** |
2577 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2578 | * @q: queue to insert the request in | |
2579 | * @bd_disk: matching gendisk | |
2580 | * @rq: request to insert | |
2581 | * @at_head: insert request at head or tail of queue | |
2582 | * @done: I/O completion handler | |
2583 | * | |
2584 | * Description: | |
2585 | * Insert a fully prepared request at the back of the io scheduler queue | |
2586 | * for execution. Don't wait for completion. | |
2587 | */ | |
f1970baf JB |
2588 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, |
2589 | struct request *rq, int at_head, | |
8ffdc655 | 2590 | rq_end_io_fn *done) |
f1970baf JB |
2591 | { |
2592 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2593 | ||
2594 | rq->rq_disk = bd_disk; | |
4aff5e23 | 2595 | rq->cmd_flags |= REQ_NOMERGE; |
f1970baf | 2596 | rq->end_io = done; |
4c5d0bbd AM |
2597 | WARN_ON(irqs_disabled()); |
2598 | spin_lock_irq(q->queue_lock); | |
2599 | __elv_add_request(q, rq, where, 1); | |
2600 | __generic_unplug_device(q); | |
2601 | spin_unlock_irq(q->queue_lock); | |
f1970baf | 2602 | } |
6e39b69e MC |
2603 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
2604 | ||
1da177e4 LT |
2605 | /** |
2606 | * blk_execute_rq - insert a request into queue for execution | |
2607 | * @q: queue to insert the request in | |
2608 | * @bd_disk: matching gendisk | |
2609 | * @rq: request to insert | |
994ca9a1 | 2610 | * @at_head: insert request at head or tail of queue |
1da177e4 LT |
2611 | * |
2612 | * Description: | |
2613 | * Insert a fully prepared request at the back of the io scheduler queue | |
73747aed | 2614 | * for execution and wait for completion. |
1da177e4 LT |
2615 | */ |
2616 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, | |
994ca9a1 | 2617 | struct request *rq, int at_head) |
1da177e4 | 2618 | { |
60be6b9a | 2619 | DECLARE_COMPLETION_ONSTACK(wait); |
1da177e4 LT |
2620 | char sense[SCSI_SENSE_BUFFERSIZE]; |
2621 | int err = 0; | |
2622 | ||
1da177e4 LT |
2623 | /* |
2624 | * we need an extra reference to the request, so we can look at | |
2625 | * it after io completion | |
2626 | */ | |
2627 | rq->ref_count++; | |
2628 | ||
2629 | if (!rq->sense) { | |
2630 | memset(sense, 0, sizeof(sense)); | |
2631 | rq->sense = sense; | |
2632 | rq->sense_len = 0; | |
2633 | } | |
2634 | ||
c00895ab | 2635 | rq->end_io_data = &wait; |
994ca9a1 | 2636 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
1da177e4 | 2637 | wait_for_completion(&wait); |
1da177e4 LT |
2638 | |
2639 | if (rq->errors) | |
2640 | err = -EIO; | |
2641 | ||
2642 | return err; | |
2643 | } | |
2644 | ||
2645 | EXPORT_SYMBOL(blk_execute_rq); | |
2646 | ||
2647 | /** | |
2648 | * blkdev_issue_flush - queue a flush | |
2649 | * @bdev: blockdev to issue flush for | |
2650 | * @error_sector: error sector | |
2651 | * | |
2652 | * Description: | |
2653 | * Issue a flush for the block device in question. Caller can supply | |
2654 | * room for storing the error offset in case of a flush error, if they | |
2655 | * wish to. Caller must run wait_for_completion() on its own. | |
2656 | */ | |
2657 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2658 | { | |
2659 | request_queue_t *q; | |
2660 | ||
2661 | if (bdev->bd_disk == NULL) | |
2662 | return -ENXIO; | |
2663 | ||
2664 | q = bdev_get_queue(bdev); | |
2665 | if (!q) | |
2666 | return -ENXIO; | |
2667 | if (!q->issue_flush_fn) | |
2668 | return -EOPNOTSUPP; | |
2669 | ||
2670 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); | |
2671 | } | |
2672 | ||
2673 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2674 | ||
93d17d3d | 2675 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
1da177e4 LT |
2676 | { |
2677 | int rw = rq_data_dir(rq); | |
2678 | ||
2679 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2680 | return; | |
2681 | ||
d72d904a | 2682 | if (!new_io) { |
a362357b | 2683 | __disk_stat_inc(rq->rq_disk, merges[rw]); |
d72d904a | 2684 | } else { |
1da177e4 LT |
2685 | disk_round_stats(rq->rq_disk); |
2686 | rq->rq_disk->in_flight++; | |
2687 | } | |
2688 | } | |
2689 | ||
2690 | /* | |
2691 | * add-request adds a request to the linked list. | |
2692 | * queue lock is held and interrupts disabled, as we muck with the | |
2693 | * request queue list. | |
2694 | */ | |
2695 | static inline void add_request(request_queue_t * q, struct request * req) | |
2696 | { | |
2697 | drive_stat_acct(req, req->nr_sectors, 1); | |
2698 | ||
1da177e4 LT |
2699 | /* |
2700 | * elevator indicated where it wants this request to be | |
2701 | * inserted at elevator_merge time | |
2702 | */ | |
2703 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2704 | } | |
2705 | ||
2706 | /* | |
2707 | * disk_round_stats() - Round off the performance stats on a struct | |
2708 | * disk_stats. | |
2709 | * | |
2710 | * The average IO queue length and utilisation statistics are maintained | |
2711 | * by observing the current state of the queue length and the amount of | |
2712 | * time it has been in this state for. | |
2713 | * | |
2714 | * Normally, that accounting is done on IO completion, but that can result | |
2715 | * in more than a second's worth of IO being accounted for within any one | |
2716 | * second, leading to >100% utilisation. To deal with that, we call this | |
2717 | * function to do a round-off before returning the results when reading | |
2718 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2719 | * the current jiffies and restarts the counters again. | |
2720 | */ | |
2721 | void disk_round_stats(struct gendisk *disk) | |
2722 | { | |
2723 | unsigned long now = jiffies; | |
2724 | ||
b2982649 CK |
2725 | if (now == disk->stamp) |
2726 | return; | |
1da177e4 | 2727 | |
20e5c81f CK |
2728 | if (disk->in_flight) { |
2729 | __disk_stat_add(disk, time_in_queue, | |
2730 | disk->in_flight * (now - disk->stamp)); | |
2731 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
2732 | } | |
1da177e4 | 2733 | disk->stamp = now; |
1da177e4 LT |
2734 | } |
2735 | ||
3eaf840e JNN |
2736 | EXPORT_SYMBOL_GPL(disk_round_stats); |
2737 | ||
1da177e4 LT |
2738 | /* |
2739 | * queue lock must be held | |
2740 | */ | |
6e39b69e | 2741 | void __blk_put_request(request_queue_t *q, struct request *req) |
1da177e4 | 2742 | { |
1da177e4 LT |
2743 | if (unlikely(!q)) |
2744 | return; | |
2745 | if (unlikely(--req->ref_count)) | |
2746 | return; | |
2747 | ||
8922e16c TH |
2748 | elv_completed_request(q, req); |
2749 | ||
1da177e4 LT |
2750 | /* |
2751 | * Request may not have originated from ll_rw_blk. if not, | |
2752 | * it didn't come out of our reserved rq pools | |
2753 | */ | |
49171e5c | 2754 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 2755 | int rw = rq_data_dir(req); |
4aff5e23 | 2756 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 2757 | |
1da177e4 | 2758 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 2759 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
2760 | |
2761 | blk_free_request(q, req); | |
cb98fc8b | 2762 | freed_request(q, rw, priv); |
1da177e4 LT |
2763 | } |
2764 | } | |
2765 | ||
6e39b69e MC |
2766 | EXPORT_SYMBOL_GPL(__blk_put_request); |
2767 | ||
1da177e4 LT |
2768 | void blk_put_request(struct request *req) |
2769 | { | |
8922e16c TH |
2770 | unsigned long flags; |
2771 | request_queue_t *q = req->q; | |
2772 | ||
1da177e4 | 2773 | /* |
8922e16c TH |
2774 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
2775 | * following if (q) test. | |
1da177e4 | 2776 | */ |
8922e16c | 2777 | if (q) { |
1da177e4 LT |
2778 | spin_lock_irqsave(q->queue_lock, flags); |
2779 | __blk_put_request(q, req); | |
2780 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | EXPORT_SYMBOL(blk_put_request); | |
2785 | ||
2786 | /** | |
2787 | * blk_end_sync_rq - executes a completion event on a request | |
2788 | * @rq: request to complete | |
fddfdeaf | 2789 | * @error: end io status of the request |
1da177e4 | 2790 | */ |
8ffdc655 | 2791 | void blk_end_sync_rq(struct request *rq, int error) |
1da177e4 | 2792 | { |
c00895ab | 2793 | struct completion *waiting = rq->end_io_data; |
1da177e4 | 2794 | |
c00895ab | 2795 | rq->end_io_data = NULL; |
1da177e4 LT |
2796 | __blk_put_request(rq->q, rq); |
2797 | ||
2798 | /* | |
2799 | * complete last, if this is a stack request the process (and thus | |
2800 | * the rq pointer) could be invalid right after this complete() | |
2801 | */ | |
2802 | complete(waiting); | |
2803 | } | |
2804 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2805 | ||
1da177e4 LT |
2806 | /* |
2807 | * Has to be called with the request spinlock acquired | |
2808 | */ | |
2809 | static int attempt_merge(request_queue_t *q, struct request *req, | |
2810 | struct request *next) | |
2811 | { | |
2812 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2813 | return 0; | |
2814 | ||
2815 | /* | |
d6e05edc | 2816 | * not contiguous |
1da177e4 LT |
2817 | */ |
2818 | if (req->sector + req->nr_sectors != next->sector) | |
2819 | return 0; | |
2820 | ||
2821 | if (rq_data_dir(req) != rq_data_dir(next) | |
2822 | || req->rq_disk != next->rq_disk | |
c00895ab | 2823 | || next->special) |
1da177e4 LT |
2824 | return 0; |
2825 | ||
2826 | /* | |
2827 | * If we are allowed to merge, then append bio list | |
2828 | * from next to rq and release next. merge_requests_fn | |
2829 | * will have updated segment counts, update sector | |
2830 | * counts here. | |
2831 | */ | |
1aa4f24f | 2832 | if (!ll_merge_requests_fn(q, req, next)) |
1da177e4 LT |
2833 | return 0; |
2834 | ||
2835 | /* | |
2836 | * At this point we have either done a back merge | |
2837 | * or front merge. We need the smaller start_time of | |
2838 | * the merged requests to be the current request | |
2839 | * for accounting purposes. | |
2840 | */ | |
2841 | if (time_after(req->start_time, next->start_time)) | |
2842 | req->start_time = next->start_time; | |
2843 | ||
2844 | req->biotail->bi_next = next->bio; | |
2845 | req->biotail = next->biotail; | |
2846 | ||
2847 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2848 | ||
2849 | elv_merge_requests(q, req, next); | |
2850 | ||
2851 | if (req->rq_disk) { | |
2852 | disk_round_stats(req->rq_disk); | |
2853 | req->rq_disk->in_flight--; | |
2854 | } | |
2855 | ||
22e2c507 JA |
2856 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2857 | ||
1da177e4 LT |
2858 | __blk_put_request(q, next); |
2859 | return 1; | |
2860 | } | |
2861 | ||
2862 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) | |
2863 | { | |
2864 | struct request *next = elv_latter_request(q, rq); | |
2865 | ||
2866 | if (next) | |
2867 | return attempt_merge(q, rq, next); | |
2868 | ||
2869 | return 0; | |
2870 | } | |
2871 | ||
2872 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) | |
2873 | { | |
2874 | struct request *prev = elv_former_request(q, rq); | |
2875 | ||
2876 | if (prev) | |
2877 | return attempt_merge(q, prev, rq); | |
2878 | ||
2879 | return 0; | |
2880 | } | |
2881 | ||
52d9e675 TH |
2882 | static void init_request_from_bio(struct request *req, struct bio *bio) |
2883 | { | |
4aff5e23 | 2884 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
2885 | |
2886 | /* | |
2887 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2888 | */ | |
2889 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 2890 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
2891 | |
2892 | /* | |
2893 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2894 | */ | |
2895 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 2896 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 2897 | |
b31dc66a | 2898 | if (bio_sync(bio)) |
4aff5e23 | 2899 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
2900 | if (bio_rw_meta(bio)) |
2901 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 2902 | |
52d9e675 TH |
2903 | req->errors = 0; |
2904 | req->hard_sector = req->sector = bio->bi_sector; | |
2905 | req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio); | |
2906 | req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio); | |
2907 | req->nr_phys_segments = bio_phys_segments(req->q, bio); | |
2908 | req->nr_hw_segments = bio_hw_segments(req->q, bio); | |
2909 | req->buffer = bio_data(bio); /* see ->buffer comment above */ | |
52d9e675 TH |
2910 | req->bio = req->biotail = bio; |
2911 | req->ioprio = bio_prio(bio); | |
2912 | req->rq_disk = bio->bi_bdev->bd_disk; | |
2913 | req->start_time = jiffies; | |
2914 | } | |
2915 | ||
1da177e4 LT |
2916 | static int __make_request(request_queue_t *q, struct bio *bio) |
2917 | { | |
450991bc | 2918 | struct request *req; |
51da90fc JA |
2919 | int el_ret, nr_sectors, barrier, err; |
2920 | const unsigned short prio = bio_prio(bio); | |
2921 | const int sync = bio_sync(bio); | |
7749a8d4 | 2922 | int rw_flags; |
1da177e4 | 2923 | |
1da177e4 | 2924 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
2925 | |
2926 | /* | |
2927 | * low level driver can indicate that it wants pages above a | |
2928 | * certain limit bounced to low memory (ie for highmem, or even | |
2929 | * ISA dma in theory) | |
2930 | */ | |
2931 | blk_queue_bounce(q, &bio); | |
2932 | ||
1da177e4 | 2933 | barrier = bio_barrier(bio); |
797e7dbb | 2934 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
2935 | err = -EOPNOTSUPP; |
2936 | goto end_io; | |
2937 | } | |
2938 | ||
1da177e4 LT |
2939 | spin_lock_irq(q->queue_lock); |
2940 | ||
450991bc | 2941 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
2942 | goto get_rq; |
2943 | ||
2944 | el_ret = elv_merge(q, &req, bio); | |
2945 | switch (el_ret) { | |
2946 | case ELEVATOR_BACK_MERGE: | |
2947 | BUG_ON(!rq_mergeable(req)); | |
2948 | ||
1aa4f24f | 2949 | if (!ll_back_merge_fn(q, req, bio)) |
1da177e4 LT |
2950 | break; |
2951 | ||
2056a782 JA |
2952 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2953 | ||
1da177e4 LT |
2954 | req->biotail->bi_next = bio; |
2955 | req->biotail = bio; | |
2956 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 2957 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2958 | drive_stat_acct(req, nr_sectors, 0); |
2959 | if (!attempt_back_merge(q, req)) | |
2e662b65 | 2960 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2961 | goto out; |
2962 | ||
2963 | case ELEVATOR_FRONT_MERGE: | |
2964 | BUG_ON(!rq_mergeable(req)); | |
2965 | ||
1aa4f24f | 2966 | if (!ll_front_merge_fn(q, req, bio)) |
1da177e4 LT |
2967 | break; |
2968 | ||
2056a782 JA |
2969 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2970 | ||
1da177e4 LT |
2971 | bio->bi_next = req->bio; |
2972 | req->bio = bio; | |
2973 | ||
2974 | /* | |
2975 | * may not be valid. if the low level driver said | |
2976 | * it didn't need a bounce buffer then it better | |
2977 | * not touch req->buffer either... | |
2978 | */ | |
2979 | req->buffer = bio_data(bio); | |
51da90fc JA |
2980 | req->current_nr_sectors = bio_cur_sectors(bio); |
2981 | req->hard_cur_sectors = req->current_nr_sectors; | |
2982 | req->sector = req->hard_sector = bio->bi_sector; | |
1da177e4 | 2983 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
22e2c507 | 2984 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2985 | drive_stat_acct(req, nr_sectors, 0); |
2986 | if (!attempt_front_merge(q, req)) | |
2e662b65 | 2987 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2988 | goto out; |
2989 | ||
450991bc | 2990 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
1da177e4 | 2991 | default: |
450991bc | 2992 | ; |
1da177e4 LT |
2993 | } |
2994 | ||
450991bc | 2995 | get_rq: |
7749a8d4 JA |
2996 | /* |
2997 | * This sync check and mask will be re-done in init_request_from_bio(), | |
2998 | * but we need to set it earlier to expose the sync flag to the | |
2999 | * rq allocator and io schedulers. | |
3000 | */ | |
3001 | rw_flags = bio_data_dir(bio); | |
3002 | if (sync) | |
3003 | rw_flags |= REQ_RW_SYNC; | |
3004 | ||
1da177e4 | 3005 | /* |
450991bc | 3006 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 3007 | * Returns with the queue unlocked. |
450991bc | 3008 | */ |
7749a8d4 | 3009 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 3010 | |
450991bc NP |
3011 | /* |
3012 | * After dropping the lock and possibly sleeping here, our request | |
3013 | * may now be mergeable after it had proven unmergeable (above). | |
3014 | * We don't worry about that case for efficiency. It won't happen | |
3015 | * often, and the elevators are able to handle it. | |
1da177e4 | 3016 | */ |
52d9e675 | 3017 | init_request_from_bio(req, bio); |
1da177e4 | 3018 | |
450991bc NP |
3019 | spin_lock_irq(q->queue_lock); |
3020 | if (elv_queue_empty(q)) | |
3021 | blk_plug_device(q); | |
1da177e4 LT |
3022 | add_request(q, req); |
3023 | out: | |
4a534f93 | 3024 | if (sync) |
1da177e4 LT |
3025 | __generic_unplug_device(q); |
3026 | ||
3027 | spin_unlock_irq(q->queue_lock); | |
3028 | return 0; | |
3029 | ||
3030 | end_io: | |
3031 | bio_endio(bio, nr_sectors << 9, err); | |
3032 | return 0; | |
3033 | } | |
3034 | ||
3035 | /* | |
3036 | * If bio->bi_dev is a partition, remap the location | |
3037 | */ | |
3038 | static inline void blk_partition_remap(struct bio *bio) | |
3039 | { | |
3040 | struct block_device *bdev = bio->bi_bdev; | |
3041 | ||
3042 | if (bdev != bdev->bd_contains) { | |
3043 | struct hd_struct *p = bdev->bd_part; | |
a362357b JA |
3044 | const int rw = bio_data_dir(bio); |
3045 | ||
3046 | p->sectors[rw] += bio_sectors(bio); | |
3047 | p->ios[rw]++; | |
1da177e4 | 3048 | |
1da177e4 LT |
3049 | bio->bi_sector += p->start_sect; |
3050 | bio->bi_bdev = bdev->bd_contains; | |
3051 | } | |
3052 | } | |
3053 | ||
1da177e4 LT |
3054 | static void handle_bad_sector(struct bio *bio) |
3055 | { | |
3056 | char b[BDEVNAME_SIZE]; | |
3057 | ||
3058 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
3059 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
3060 | bdevname(bio->bi_bdev, b), | |
3061 | bio->bi_rw, | |
3062 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
3063 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
3064 | ||
3065 | set_bit(BIO_EOF, &bio->bi_flags); | |
3066 | } | |
3067 | ||
c17bb495 AM |
3068 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
3069 | ||
3070 | static DECLARE_FAULT_ATTR(fail_make_request); | |
3071 | ||
3072 | static int __init setup_fail_make_request(char *str) | |
3073 | { | |
3074 | return setup_fault_attr(&fail_make_request, str); | |
3075 | } | |
3076 | __setup("fail_make_request=", setup_fail_make_request); | |
3077 | ||
3078 | static int should_fail_request(struct bio *bio) | |
3079 | { | |
3080 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
3081 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
3082 | return should_fail(&fail_make_request, bio->bi_size); | |
3083 | ||
3084 | return 0; | |
3085 | } | |
3086 | ||
3087 | static int __init fail_make_request_debugfs(void) | |
3088 | { | |
3089 | return init_fault_attr_dentries(&fail_make_request, | |
3090 | "fail_make_request"); | |
3091 | } | |
3092 | ||
3093 | late_initcall(fail_make_request_debugfs); | |
3094 | ||
3095 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
3096 | ||
3097 | static inline int should_fail_request(struct bio *bio) | |
3098 | { | |
3099 | return 0; | |
3100 | } | |
3101 | ||
3102 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
3103 | ||
1da177e4 LT |
3104 | /** |
3105 | * generic_make_request: hand a buffer to its device driver for I/O | |
3106 | * @bio: The bio describing the location in memory and on the device. | |
3107 | * | |
3108 | * generic_make_request() is used to make I/O requests of block | |
3109 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
3110 | * to be done. | |
3111 | * | |
3112 | * generic_make_request() does not return any status. The | |
3113 | * success/failure status of the request, along with notification of | |
3114 | * completion, is delivered asynchronously through the bio->bi_end_io | |
3115 | * function described (one day) else where. | |
3116 | * | |
3117 | * The caller of generic_make_request must make sure that bi_io_vec | |
3118 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
3119 | * set to describe the device address, and the | |
3120 | * bi_end_io and optionally bi_private are set to describe how | |
3121 | * completion notification should be signaled. | |
3122 | * | |
3123 | * generic_make_request and the drivers it calls may use bi_next if this | |
3124 | * bio happens to be merged with someone else, and may change bi_dev and | |
3125 | * bi_sector for remaps as it sees fit. So the values of these fields | |
3126 | * should NOT be depended on after the call to generic_make_request. | |
3127 | */ | |
d89d8796 | 3128 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 LT |
3129 | { |
3130 | request_queue_t *q; | |
3131 | sector_t maxsector; | |
5ddfe969 | 3132 | sector_t old_sector; |
1da177e4 | 3133 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 3134 | dev_t old_dev; |
1da177e4 LT |
3135 | |
3136 | might_sleep(); | |
3137 | /* Test device or partition size, when known. */ | |
3138 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
3139 | if (maxsector) { | |
3140 | sector_t sector = bio->bi_sector; | |
3141 | ||
3142 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
3143 | /* | |
3144 | * This may well happen - the kernel calls bread() | |
3145 | * without checking the size of the device, e.g., when | |
3146 | * mounting a device. | |
3147 | */ | |
3148 | handle_bad_sector(bio); | |
3149 | goto end_io; | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Resolve the mapping until finished. (drivers are | |
3155 | * still free to implement/resolve their own stacking | |
3156 | * by explicitly returning 0) | |
3157 | * | |
3158 | * NOTE: we don't repeat the blk_size check for each new device. | |
3159 | * Stacking drivers are expected to know what they are doing. | |
3160 | */ | |
5ddfe969 | 3161 | old_sector = -1; |
2056a782 | 3162 | old_dev = 0; |
1da177e4 LT |
3163 | do { |
3164 | char b[BDEVNAME_SIZE]; | |
3165 | ||
3166 | q = bdev_get_queue(bio->bi_bdev); | |
3167 | if (!q) { | |
3168 | printk(KERN_ERR | |
3169 | "generic_make_request: Trying to access " | |
3170 | "nonexistent block-device %s (%Lu)\n", | |
3171 | bdevname(bio->bi_bdev, b), | |
3172 | (long long) bio->bi_sector); | |
3173 | end_io: | |
3174 | bio_endio(bio, bio->bi_size, -EIO); | |
3175 | break; | |
3176 | } | |
3177 | ||
3178 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { | |
3179 | printk("bio too big device %s (%u > %u)\n", | |
3180 | bdevname(bio->bi_bdev, b), | |
3181 | bio_sectors(bio), | |
3182 | q->max_hw_sectors); | |
3183 | goto end_io; | |
3184 | } | |
3185 | ||
fde6ad22 | 3186 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
3187 | goto end_io; |
3188 | ||
c17bb495 AM |
3189 | if (should_fail_request(bio)) |
3190 | goto end_io; | |
3191 | ||
1da177e4 LT |
3192 | /* |
3193 | * If this device has partitions, remap block n | |
3194 | * of partition p to block n+start(p) of the disk. | |
3195 | */ | |
3196 | blk_partition_remap(bio); | |
3197 | ||
5ddfe969 | 3198 | if (old_sector != -1) |
2056a782 | 3199 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 3200 | old_sector); |
2056a782 JA |
3201 | |
3202 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
3203 | ||
5ddfe969 | 3204 | old_sector = bio->bi_sector; |
2056a782 JA |
3205 | old_dev = bio->bi_bdev->bd_dev; |
3206 | ||
5ddfe969 N |
3207 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
3208 | if (maxsector) { | |
3209 | sector_t sector = bio->bi_sector; | |
3210 | ||
df66b855 AM |
3211 | if (maxsector < nr_sectors || |
3212 | maxsector - nr_sectors < sector) { | |
5ddfe969 | 3213 | /* |
df66b855 AM |
3214 | * This may well happen - partitions are not |
3215 | * checked to make sure they are within the size | |
3216 | * of the whole device. | |
5ddfe969 N |
3217 | */ |
3218 | handle_bad_sector(bio); | |
3219 | goto end_io; | |
3220 | } | |
3221 | } | |
3222 | ||
1da177e4 LT |
3223 | ret = q->make_request_fn(q, bio); |
3224 | } while (ret); | |
3225 | } | |
3226 | ||
d89d8796 NB |
3227 | /* |
3228 | * We only want one ->make_request_fn to be active at a time, | |
3229 | * else stack usage with stacked devices could be a problem. | |
3230 | * So use current->bio_{list,tail} to keep a list of requests | |
3231 | * submited by a make_request_fn function. | |
3232 | * current->bio_tail is also used as a flag to say if | |
3233 | * generic_make_request is currently active in this task or not. | |
3234 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
3235 | * then a make_request is active, and new requests should be added | |
3236 | * at the tail | |
3237 | */ | |
3238 | void generic_make_request(struct bio *bio) | |
3239 | { | |
3240 | if (current->bio_tail) { | |
3241 | /* make_request is active */ | |
3242 | *(current->bio_tail) = bio; | |
3243 | bio->bi_next = NULL; | |
3244 | current->bio_tail = &bio->bi_next; | |
3245 | return; | |
3246 | } | |
3247 | /* following loop may be a bit non-obvious, and so deserves some | |
3248 | * explanation. | |
3249 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
3250 | * ensure that) so we have a list with a single bio. | |
3251 | * We pretend that we have just taken it off a longer list, so | |
3252 | * we assign bio_list to the next (which is NULL) and bio_tail | |
3253 | * to &bio_list, thus initialising the bio_list of new bios to be | |
3254 | * added. __generic_make_request may indeed add some more bios | |
3255 | * through a recursive call to generic_make_request. If it | |
3256 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
3257 | * from the top. In this case we really did just take the bio | |
3258 | * of the top of the list (no pretending) and so fixup bio_list and | |
3259 | * bio_tail or bi_next, and call into __generic_make_request again. | |
3260 | * | |
3261 | * The loop was structured like this to make only one call to | |
3262 | * __generic_make_request (which is important as it is large and | |
3263 | * inlined) and to keep the structure simple. | |
3264 | */ | |
3265 | BUG_ON(bio->bi_next); | |
3266 | do { | |
3267 | current->bio_list = bio->bi_next; | |
3268 | if (bio->bi_next == NULL) | |
3269 | current->bio_tail = ¤t->bio_list; | |
3270 | else | |
3271 | bio->bi_next = NULL; | |
3272 | __generic_make_request(bio); | |
3273 | bio = current->bio_list; | |
3274 | } while (bio); | |
3275 | current->bio_tail = NULL; /* deactivate */ | |
3276 | } | |
3277 | ||
1da177e4 LT |
3278 | EXPORT_SYMBOL(generic_make_request); |
3279 | ||
3280 | /** | |
3281 | * submit_bio: submit a bio to the block device layer for I/O | |
3282 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3283 | * @bio: The &struct bio which describes the I/O | |
3284 | * | |
3285 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3286 | * uses that function to do most of the work. Both are fairly rough | |
3287 | * interfaces, @bio must be presetup and ready for I/O. | |
3288 | * | |
3289 | */ | |
3290 | void submit_bio(int rw, struct bio *bio) | |
3291 | { | |
3292 | int count = bio_sectors(bio); | |
3293 | ||
3294 | BIO_BUG_ON(!bio->bi_size); | |
3295 | BIO_BUG_ON(!bio->bi_io_vec); | |
22e2c507 | 3296 | bio->bi_rw |= rw; |
faccbd4b | 3297 | if (rw & WRITE) { |
f8891e5e | 3298 | count_vm_events(PGPGOUT, count); |
faccbd4b AM |
3299 | } else { |
3300 | task_io_account_read(bio->bi_size); | |
f8891e5e | 3301 | count_vm_events(PGPGIN, count); |
faccbd4b | 3302 | } |
1da177e4 LT |
3303 | |
3304 | if (unlikely(block_dump)) { | |
3305 | char b[BDEVNAME_SIZE]; | |
3306 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
3307 | current->comm, current->pid, | |
3308 | (rw & WRITE) ? "WRITE" : "READ", | |
3309 | (unsigned long long)bio->bi_sector, | |
3310 | bdevname(bio->bi_bdev,b)); | |
3311 | } | |
3312 | ||
3313 | generic_make_request(bio); | |
3314 | } | |
3315 | ||
3316 | EXPORT_SYMBOL(submit_bio); | |
3317 | ||
93d17d3d | 3318 | static void blk_recalc_rq_segments(struct request *rq) |
1da177e4 LT |
3319 | { |
3320 | struct bio *bio, *prevbio = NULL; | |
3321 | int nr_phys_segs, nr_hw_segs; | |
3322 | unsigned int phys_size, hw_size; | |
3323 | request_queue_t *q = rq->q; | |
3324 | ||
3325 | if (!rq->bio) | |
3326 | return; | |
3327 | ||
3328 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
3329 | rq_for_each_bio(bio, rq) { | |
3330 | /* Force bio hw/phys segs to be recalculated. */ | |
3331 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3332 | ||
3333 | nr_phys_segs += bio_phys_segments(q, bio); | |
3334 | nr_hw_segs += bio_hw_segments(q, bio); | |
3335 | if (prevbio) { | |
3336 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; | |
3337 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; | |
3338 | ||
3339 | if (blk_phys_contig_segment(q, prevbio, bio) && | |
3340 | pseg <= q->max_segment_size) { | |
3341 | nr_phys_segs--; | |
3342 | phys_size += prevbio->bi_size + bio->bi_size; | |
3343 | } else | |
3344 | phys_size = 0; | |
3345 | ||
3346 | if (blk_hw_contig_segment(q, prevbio, bio) && | |
3347 | hseg <= q->max_segment_size) { | |
3348 | nr_hw_segs--; | |
3349 | hw_size += prevbio->bi_size + bio->bi_size; | |
3350 | } else | |
3351 | hw_size = 0; | |
3352 | } | |
3353 | prevbio = bio; | |
3354 | } | |
3355 | ||
3356 | rq->nr_phys_segments = nr_phys_segs; | |
3357 | rq->nr_hw_segments = nr_hw_segs; | |
3358 | } | |
3359 | ||
93d17d3d | 3360 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
1da177e4 LT |
3361 | { |
3362 | if (blk_fs_request(rq)) { | |
3363 | rq->hard_sector += nsect; | |
3364 | rq->hard_nr_sectors -= nsect; | |
3365 | ||
3366 | /* | |
3367 | * Move the I/O submission pointers ahead if required. | |
3368 | */ | |
3369 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3370 | (rq->sector <= rq->hard_sector)) { | |
3371 | rq->sector = rq->hard_sector; | |
3372 | rq->nr_sectors = rq->hard_nr_sectors; | |
3373 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3374 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3375 | rq->buffer = bio_data(rq->bio); | |
3376 | } | |
3377 | ||
3378 | /* | |
3379 | * if total number of sectors is less than the first segment | |
3380 | * size, something has gone terribly wrong | |
3381 | */ | |
3382 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3383 | printk("blk: request botched\n"); | |
3384 | rq->nr_sectors = rq->current_nr_sectors; | |
3385 | } | |
3386 | } | |
3387 | } | |
3388 | ||
3389 | static int __end_that_request_first(struct request *req, int uptodate, | |
3390 | int nr_bytes) | |
3391 | { | |
3392 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3393 | struct bio *bio; | |
3394 | ||
2056a782 JA |
3395 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
3396 | ||
1da177e4 LT |
3397 | /* |
3398 | * extend uptodate bool to allow < 0 value to be direct io error | |
3399 | */ | |
3400 | error = 0; | |
3401 | if (end_io_error(uptodate)) | |
3402 | error = !uptodate ? -EIO : uptodate; | |
3403 | ||
3404 | /* | |
3405 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3406 | * sense key with us all the way through | |
3407 | */ | |
3408 | if (!blk_pc_request(req)) | |
3409 | req->errors = 0; | |
3410 | ||
3411 | if (!uptodate) { | |
4aff5e23 | 3412 | if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET)) |
1da177e4 LT |
3413 | printk("end_request: I/O error, dev %s, sector %llu\n", |
3414 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3415 | (unsigned long long)req->sector); | |
3416 | } | |
3417 | ||
d72d904a | 3418 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
3419 | const int rw = rq_data_dir(req); |
3420 | ||
53e86061 | 3421 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
d72d904a JA |
3422 | } |
3423 | ||
1da177e4 LT |
3424 | total_bytes = bio_nbytes = 0; |
3425 | while ((bio = req->bio) != NULL) { | |
3426 | int nbytes; | |
3427 | ||
3428 | if (nr_bytes >= bio->bi_size) { | |
3429 | req->bio = bio->bi_next; | |
3430 | nbytes = bio->bi_size; | |
797e7dbb TH |
3431 | if (!ordered_bio_endio(req, bio, nbytes, error)) |
3432 | bio_endio(bio, nbytes, error); | |
1da177e4 LT |
3433 | next_idx = 0; |
3434 | bio_nbytes = 0; | |
3435 | } else { | |
3436 | int idx = bio->bi_idx + next_idx; | |
3437 | ||
3438 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3439 | blk_dump_rq_flags(req, "__end_that"); | |
3440 | printk("%s: bio idx %d >= vcnt %d\n", | |
3441 | __FUNCTION__, | |
3442 | bio->bi_idx, bio->bi_vcnt); | |
3443 | break; | |
3444 | } | |
3445 | ||
3446 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3447 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3448 | ||
3449 | /* | |
3450 | * not a complete bvec done | |
3451 | */ | |
3452 | if (unlikely(nbytes > nr_bytes)) { | |
3453 | bio_nbytes += nr_bytes; | |
3454 | total_bytes += nr_bytes; | |
3455 | break; | |
3456 | } | |
3457 | ||
3458 | /* | |
3459 | * advance to the next vector | |
3460 | */ | |
3461 | next_idx++; | |
3462 | bio_nbytes += nbytes; | |
3463 | } | |
3464 | ||
3465 | total_bytes += nbytes; | |
3466 | nr_bytes -= nbytes; | |
3467 | ||
3468 | if ((bio = req->bio)) { | |
3469 | /* | |
3470 | * end more in this run, or just return 'not-done' | |
3471 | */ | |
3472 | if (unlikely(nr_bytes <= 0)) | |
3473 | break; | |
3474 | } | |
3475 | } | |
3476 | ||
3477 | /* | |
3478 | * completely done | |
3479 | */ | |
3480 | if (!req->bio) | |
3481 | return 0; | |
3482 | ||
3483 | /* | |
3484 | * if the request wasn't completed, update state | |
3485 | */ | |
3486 | if (bio_nbytes) { | |
797e7dbb TH |
3487 | if (!ordered_bio_endio(req, bio, bio_nbytes, error)) |
3488 | bio_endio(bio, bio_nbytes, error); | |
1da177e4 LT |
3489 | bio->bi_idx += next_idx; |
3490 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3491 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3492 | } | |
3493 | ||
3494 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3495 | blk_recalc_rq_segments(req); | |
3496 | return 1; | |
3497 | } | |
3498 | ||
3499 | /** | |
3500 | * end_that_request_first - end I/O on a request | |
3501 | * @req: the request being processed | |
3502 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3503 | * @nr_sectors: number of sectors to end I/O on | |
3504 | * | |
3505 | * Description: | |
3506 | * Ends I/O on a number of sectors attached to @req, and sets it up | |
3507 | * for the next range of segments (if any) in the cluster. | |
3508 | * | |
3509 | * Return: | |
3510 | * 0 - we are done with this request, call end_that_request_last() | |
3511 | * 1 - still buffers pending for this request | |
3512 | **/ | |
3513 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | |
3514 | { | |
3515 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | |
3516 | } | |
3517 | ||
3518 | EXPORT_SYMBOL(end_that_request_first); | |
3519 | ||
3520 | /** | |
3521 | * end_that_request_chunk - end I/O on a request | |
3522 | * @req: the request being processed | |
3523 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3524 | * @nr_bytes: number of bytes to complete | |
3525 | * | |
3526 | * Description: | |
3527 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3528 | * for the next range of segments (if any). Like end_that_request_first(), | |
3529 | * but deals with bytes instead of sectors. | |
3530 | * | |
3531 | * Return: | |
3532 | * 0 - we are done with this request, call end_that_request_last() | |
3533 | * 1 - still buffers pending for this request | |
3534 | **/ | |
3535 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | |
3536 | { | |
3537 | return __end_that_request_first(req, uptodate, nr_bytes); | |
3538 | } | |
3539 | ||
3540 | EXPORT_SYMBOL(end_that_request_chunk); | |
3541 | ||
ff856bad JA |
3542 | /* |
3543 | * splice the completion data to a local structure and hand off to | |
3544 | * process_completion_queue() to complete the requests | |
3545 | */ | |
3546 | static void blk_done_softirq(struct softirq_action *h) | |
3547 | { | |
626ab0e6 | 3548 | struct list_head *cpu_list, local_list; |
ff856bad JA |
3549 | |
3550 | local_irq_disable(); | |
3551 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 3552 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
3553 | local_irq_enable(); |
3554 | ||
3555 | while (!list_empty(&local_list)) { | |
3556 | struct request *rq = list_entry(local_list.next, struct request, donelist); | |
3557 | ||
3558 | list_del_init(&rq->donelist); | |
3559 | rq->q->softirq_done_fn(rq); | |
3560 | } | |
3561 | } | |
3562 | ||
ff856bad JA |
3563 | static int blk_cpu_notify(struct notifier_block *self, unsigned long action, |
3564 | void *hcpu) | |
3565 | { | |
3566 | /* | |
3567 | * If a CPU goes away, splice its entries to the current CPU | |
3568 | * and trigger a run of the softirq | |
3569 | */ | |
8bb78442 | 3570 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
3571 | int cpu = (unsigned long) hcpu; |
3572 | ||
3573 | local_irq_disable(); | |
3574 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
3575 | &__get_cpu_var(blk_cpu_done)); | |
3576 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3577 | local_irq_enable(); | |
3578 | } | |
3579 | ||
3580 | return NOTIFY_OK; | |
3581 | } | |
3582 | ||
3583 | ||
054cc8a2 | 3584 | static struct notifier_block __devinitdata blk_cpu_notifier = { |
ff856bad JA |
3585 | .notifier_call = blk_cpu_notify, |
3586 | }; | |
3587 | ||
ff856bad JA |
3588 | /** |
3589 | * blk_complete_request - end I/O on a request | |
3590 | * @req: the request being processed | |
3591 | * | |
3592 | * Description: | |
3593 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 3594 | * unless the driver actually implements this in its completion callback |
ff856bad JA |
3595 | * through requeueing. Theh actual completion happens out-of-order, |
3596 | * through a softirq handler. The user must have registered a completion | |
3597 | * callback through blk_queue_softirq_done(). | |
3598 | **/ | |
3599 | ||
3600 | void blk_complete_request(struct request *req) | |
3601 | { | |
3602 | struct list_head *cpu_list; | |
3603 | unsigned long flags; | |
3604 | ||
3605 | BUG_ON(!req->q->softirq_done_fn); | |
3606 | ||
3607 | local_irq_save(flags); | |
3608 | ||
3609 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3610 | list_add_tail(&req->donelist, cpu_list); | |
3611 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3612 | ||
3613 | local_irq_restore(flags); | |
3614 | } | |
3615 | ||
3616 | EXPORT_SYMBOL(blk_complete_request); | |
3617 | ||
1da177e4 LT |
3618 | /* |
3619 | * queue lock must be held | |
3620 | */ | |
8ffdc655 | 3621 | void end_that_request_last(struct request *req, int uptodate) |
1da177e4 LT |
3622 | { |
3623 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 TH |
3624 | int error; |
3625 | ||
3626 | /* | |
3627 | * extend uptodate bool to allow < 0 value to be direct io error | |
3628 | */ | |
3629 | error = 0; | |
3630 | if (end_io_error(uptodate)) | |
3631 | error = !uptodate ? -EIO : uptodate; | |
1da177e4 LT |
3632 | |
3633 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3634 | laptop_io_completion(); | |
3635 | ||
fd0ff8aa JA |
3636 | /* |
3637 | * Account IO completion. bar_rq isn't accounted as a normal | |
3638 | * IO on queueing nor completion. Accounting the containing | |
3639 | * request is enough. | |
3640 | */ | |
3641 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 3642 | unsigned long duration = jiffies - req->start_time; |
a362357b JA |
3643 | const int rw = rq_data_dir(req); |
3644 | ||
3645 | __disk_stat_inc(disk, ios[rw]); | |
3646 | __disk_stat_add(disk, ticks[rw], duration); | |
1da177e4 LT |
3647 | disk_round_stats(disk); |
3648 | disk->in_flight--; | |
3649 | } | |
3650 | if (req->end_io) | |
8ffdc655 | 3651 | req->end_io(req, error); |
1da177e4 LT |
3652 | else |
3653 | __blk_put_request(req->q, req); | |
3654 | } | |
3655 | ||
3656 | EXPORT_SYMBOL(end_that_request_last); | |
3657 | ||
3658 | void end_request(struct request *req, int uptodate) | |
3659 | { | |
3660 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { | |
3661 | add_disk_randomness(req->rq_disk); | |
3662 | blkdev_dequeue_request(req); | |
8ffdc655 | 3663 | end_that_request_last(req, uptodate); |
1da177e4 LT |
3664 | } |
3665 | } | |
3666 | ||
3667 | EXPORT_SYMBOL(end_request); | |
3668 | ||
3669 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) | |
3670 | { | |
4aff5e23 JA |
3671 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
3672 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
3673 | |
3674 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3675 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3676 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3677 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3678 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3679 | rq->buffer = bio_data(bio); | |
0e75f906 | 3680 | rq->data_len = bio->bi_size; |
1da177e4 LT |
3681 | |
3682 | rq->bio = rq->biotail = bio; | |
3683 | } | |
3684 | ||
3685 | EXPORT_SYMBOL(blk_rq_bio_prep); | |
3686 | ||
3687 | int kblockd_schedule_work(struct work_struct *work) | |
3688 | { | |
3689 | return queue_work(kblockd_workqueue, work); | |
3690 | } | |
3691 | ||
3692 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3693 | ||
19a75d83 | 3694 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 3695 | { |
28e53bdd | 3696 | cancel_work_sync(work); |
1da177e4 | 3697 | } |
19a75d83 | 3698 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
3699 | |
3700 | int __init blk_dev_init(void) | |
3701 | { | |
ff856bad JA |
3702 | int i; |
3703 | ||
1da177e4 LT |
3704 | kblockd_workqueue = create_workqueue("kblockd"); |
3705 | if (!kblockd_workqueue) | |
3706 | panic("Failed to create kblockd\n"); | |
3707 | ||
3708 | request_cachep = kmem_cache_create("blkdev_requests", | |
3709 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); | |
3710 | ||
3711 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
3712 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); | |
3713 | ||
3714 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
3715 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); | |
3716 | ||
0a945022 | 3717 | for_each_possible_cpu(i) |
ff856bad JA |
3718 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
3719 | ||
3720 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 3721 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 3722 | |
f772b3d9 VT |
3723 | blk_max_low_pfn = max_low_pfn - 1; |
3724 | blk_max_pfn = max_pfn - 1; | |
1da177e4 LT |
3725 | |
3726 | return 0; | |
3727 | } | |
3728 | ||
3729 | /* | |
3730 | * IO Context helper functions | |
3731 | */ | |
3732 | void put_io_context(struct io_context *ioc) | |
3733 | { | |
3734 | if (ioc == NULL) | |
3735 | return; | |
3736 | ||
3737 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
3738 | ||
3739 | if (atomic_dec_and_test(&ioc->refcount)) { | |
e2d74ac0 JA |
3740 | struct cfq_io_context *cic; |
3741 | ||
334e94de | 3742 | rcu_read_lock(); |
1da177e4 LT |
3743 | if (ioc->aic && ioc->aic->dtor) |
3744 | ioc->aic->dtor(ioc->aic); | |
e2d74ac0 | 3745 | if (ioc->cic_root.rb_node != NULL) { |
7143dd4b JA |
3746 | struct rb_node *n = rb_first(&ioc->cic_root); |
3747 | ||
3748 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
e2d74ac0 JA |
3749 | cic->dtor(ioc); |
3750 | } | |
334e94de | 3751 | rcu_read_unlock(); |
1da177e4 LT |
3752 | |
3753 | kmem_cache_free(iocontext_cachep, ioc); | |
3754 | } | |
3755 | } | |
3756 | EXPORT_SYMBOL(put_io_context); | |
3757 | ||
3758 | /* Called by the exitting task */ | |
3759 | void exit_io_context(void) | |
3760 | { | |
1da177e4 | 3761 | struct io_context *ioc; |
e2d74ac0 | 3762 | struct cfq_io_context *cic; |
1da177e4 | 3763 | |
22e2c507 | 3764 | task_lock(current); |
1da177e4 LT |
3765 | ioc = current->io_context; |
3766 | current->io_context = NULL; | |
22e2c507 | 3767 | task_unlock(current); |
1da177e4 | 3768 | |
25034d7a | 3769 | ioc->task = NULL; |
1da177e4 LT |
3770 | if (ioc->aic && ioc->aic->exit) |
3771 | ioc->aic->exit(ioc->aic); | |
e2d74ac0 JA |
3772 | if (ioc->cic_root.rb_node != NULL) { |
3773 | cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); | |
3774 | cic->exit(ioc); | |
3775 | } | |
25034d7a | 3776 | |
1da177e4 LT |
3777 | put_io_context(ioc); |
3778 | } | |
3779 | ||
3780 | /* | |
3781 | * If the current task has no IO context then create one and initialise it. | |
fb3cc432 | 3782 | * Otherwise, return its existing IO context. |
1da177e4 | 3783 | * |
fb3cc432 NP |
3784 | * This returned IO context doesn't have a specifically elevated refcount, |
3785 | * but since the current task itself holds a reference, the context can be | |
3786 | * used in general code, so long as it stays within `current` context. | |
1da177e4 | 3787 | */ |
b5deef90 | 3788 | static struct io_context *current_io_context(gfp_t gfp_flags, int node) |
1da177e4 LT |
3789 | { |
3790 | struct task_struct *tsk = current; | |
1da177e4 LT |
3791 | struct io_context *ret; |
3792 | ||
1da177e4 | 3793 | ret = tsk->io_context; |
fb3cc432 NP |
3794 | if (likely(ret)) |
3795 | return ret; | |
1da177e4 | 3796 | |
b5deef90 | 3797 | ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node); |
1da177e4 LT |
3798 | if (ret) { |
3799 | atomic_set(&ret->refcount, 1); | |
22e2c507 | 3800 | ret->task = current; |
fc46379d | 3801 | ret->ioprio_changed = 0; |
1da177e4 LT |
3802 | ret->last_waited = jiffies; /* doesn't matter... */ |
3803 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
3804 | ret->aic = NULL; | |
e2d74ac0 | 3805 | ret->cic_root.rb_node = NULL; |
4e521c27 | 3806 | ret->ioc_data = NULL; |
9f83e45e ON |
3807 | /* make sure set_task_ioprio() sees the settings above */ |
3808 | smp_wmb(); | |
fb3cc432 NP |
3809 | tsk->io_context = ret; |
3810 | } | |
1da177e4 | 3811 | |
fb3cc432 NP |
3812 | return ret; |
3813 | } | |
1da177e4 | 3814 | |
fb3cc432 NP |
3815 | /* |
3816 | * If the current task has no IO context then create one and initialise it. | |
3817 | * If it does have a context, take a ref on it. | |
3818 | * | |
3819 | * This is always called in the context of the task which submitted the I/O. | |
3820 | */ | |
b5deef90 | 3821 | struct io_context *get_io_context(gfp_t gfp_flags, int node) |
fb3cc432 NP |
3822 | { |
3823 | struct io_context *ret; | |
b5deef90 | 3824 | ret = current_io_context(gfp_flags, node); |
fb3cc432 | 3825 | if (likely(ret)) |
1da177e4 | 3826 | atomic_inc(&ret->refcount); |
1da177e4 LT |
3827 | return ret; |
3828 | } | |
3829 | EXPORT_SYMBOL(get_io_context); | |
3830 | ||
3831 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
3832 | { | |
3833 | struct io_context *src = *psrc; | |
3834 | struct io_context *dst = *pdst; | |
3835 | ||
3836 | if (src) { | |
3837 | BUG_ON(atomic_read(&src->refcount) == 0); | |
3838 | atomic_inc(&src->refcount); | |
3839 | put_io_context(dst); | |
3840 | *pdst = src; | |
3841 | } | |
3842 | } | |
3843 | EXPORT_SYMBOL(copy_io_context); | |
3844 | ||
3845 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
3846 | { | |
3847 | struct io_context *temp; | |
3848 | temp = *ioc1; | |
3849 | *ioc1 = *ioc2; | |
3850 | *ioc2 = temp; | |
3851 | } | |
3852 | EXPORT_SYMBOL(swap_io_context); | |
3853 | ||
3854 | /* | |
3855 | * sysfs parts below | |
3856 | */ | |
3857 | struct queue_sysfs_entry { | |
3858 | struct attribute attr; | |
3859 | ssize_t (*show)(struct request_queue *, char *); | |
3860 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
3861 | }; | |
3862 | ||
3863 | static ssize_t | |
3864 | queue_var_show(unsigned int var, char *page) | |
3865 | { | |
3866 | return sprintf(page, "%d\n", var); | |
3867 | } | |
3868 | ||
3869 | static ssize_t | |
3870 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
3871 | { | |
3872 | char *p = (char *) page; | |
3873 | ||
3874 | *var = simple_strtoul(p, &p, 10); | |
3875 | return count; | |
3876 | } | |
3877 | ||
3878 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
3879 | { | |
3880 | return queue_var_show(q->nr_requests, (page)); | |
3881 | } | |
3882 | ||
3883 | static ssize_t | |
3884 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
3885 | { | |
3886 | struct request_list *rl = &q->rq; | |
c981ff9f AV |
3887 | unsigned long nr; |
3888 | int ret = queue_var_store(&nr, page, count); | |
3889 | if (nr < BLKDEV_MIN_RQ) | |
3890 | nr = BLKDEV_MIN_RQ; | |
1da177e4 | 3891 | |
c981ff9f AV |
3892 | spin_lock_irq(q->queue_lock); |
3893 | q->nr_requests = nr; | |
1da177e4 LT |
3894 | blk_queue_congestion_threshold(q); |
3895 | ||
3896 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3897 | blk_set_queue_congested(q, READ); |
1da177e4 | 3898 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
79e2de4b | 3899 | blk_clear_queue_congested(q, READ); |
1da177e4 LT |
3900 | |
3901 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3902 | blk_set_queue_congested(q, WRITE); |
1da177e4 | 3903 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
79e2de4b | 3904 | blk_clear_queue_congested(q, WRITE); |
1da177e4 LT |
3905 | |
3906 | if (rl->count[READ] >= q->nr_requests) { | |
3907 | blk_set_queue_full(q, READ); | |
3908 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
3909 | blk_clear_queue_full(q, READ); | |
3910 | wake_up(&rl->wait[READ]); | |
3911 | } | |
3912 | ||
3913 | if (rl->count[WRITE] >= q->nr_requests) { | |
3914 | blk_set_queue_full(q, WRITE); | |
3915 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
3916 | blk_clear_queue_full(q, WRITE); | |
3917 | wake_up(&rl->wait[WRITE]); | |
3918 | } | |
c981ff9f | 3919 | spin_unlock_irq(q->queue_lock); |
1da177e4 LT |
3920 | return ret; |
3921 | } | |
3922 | ||
3923 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
3924 | { | |
3925 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3926 | ||
3927 | return queue_var_show(ra_kb, (page)); | |
3928 | } | |
3929 | ||
3930 | static ssize_t | |
3931 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
3932 | { | |
3933 | unsigned long ra_kb; | |
3934 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
3935 | ||
3936 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
3937 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
3938 | spin_unlock_irq(q->queue_lock); | |
3939 | ||
3940 | return ret; | |
3941 | } | |
3942 | ||
3943 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
3944 | { | |
3945 | int max_sectors_kb = q->max_sectors >> 1; | |
3946 | ||
3947 | return queue_var_show(max_sectors_kb, (page)); | |
3948 | } | |
3949 | ||
3950 | static ssize_t | |
3951 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
3952 | { | |
3953 | unsigned long max_sectors_kb, | |
3954 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
3955 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
3956 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
3957 | int ra_kb; | |
3958 | ||
3959 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
3960 | return -EINVAL; | |
3961 | /* | |
3962 | * Take the queue lock to update the readahead and max_sectors | |
3963 | * values synchronously: | |
3964 | */ | |
3965 | spin_lock_irq(q->queue_lock); | |
3966 | /* | |
3967 | * Trim readahead window as well, if necessary: | |
3968 | */ | |
3969 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3970 | if (ra_kb > max_sectors_kb) | |
3971 | q->backing_dev_info.ra_pages = | |
3972 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); | |
3973 | ||
3974 | q->max_sectors = max_sectors_kb << 1; | |
3975 | spin_unlock_irq(q->queue_lock); | |
3976 | ||
3977 | return ret; | |
3978 | } | |
3979 | ||
3980 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
3981 | { | |
3982 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
3983 | ||
3984 | return queue_var_show(max_hw_sectors_kb, (page)); | |
3985 | } | |
3986 | ||
3987 | ||
3988 | static struct queue_sysfs_entry queue_requests_entry = { | |
3989 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
3990 | .show = queue_requests_show, | |
3991 | .store = queue_requests_store, | |
3992 | }; | |
3993 | ||
3994 | static struct queue_sysfs_entry queue_ra_entry = { | |
3995 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
3996 | .show = queue_ra_show, | |
3997 | .store = queue_ra_store, | |
3998 | }; | |
3999 | ||
4000 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
4001 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
4002 | .show = queue_max_sectors_show, | |
4003 | .store = queue_max_sectors_store, | |
4004 | }; | |
4005 | ||
4006 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
4007 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
4008 | .show = queue_max_hw_sectors_show, | |
4009 | }; | |
4010 | ||
4011 | static struct queue_sysfs_entry queue_iosched_entry = { | |
4012 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
4013 | .show = elv_iosched_show, | |
4014 | .store = elv_iosched_store, | |
4015 | }; | |
4016 | ||
4017 | static struct attribute *default_attrs[] = { | |
4018 | &queue_requests_entry.attr, | |
4019 | &queue_ra_entry.attr, | |
4020 | &queue_max_hw_sectors_entry.attr, | |
4021 | &queue_max_sectors_entry.attr, | |
4022 | &queue_iosched_entry.attr, | |
4023 | NULL, | |
4024 | }; | |
4025 | ||
4026 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
4027 | ||
4028 | static ssize_t | |
4029 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4030 | { | |
4031 | struct queue_sysfs_entry *entry = to_queue(attr); | |
483f4afc AV |
4032 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
4033 | ssize_t res; | |
1da177e4 | 4034 | |
1da177e4 | 4035 | if (!entry->show) |
6c1852a0 | 4036 | return -EIO; |
483f4afc AV |
4037 | mutex_lock(&q->sysfs_lock); |
4038 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4039 | mutex_unlock(&q->sysfs_lock); | |
4040 | return -ENOENT; | |
4041 | } | |
4042 | res = entry->show(q, page); | |
4043 | mutex_unlock(&q->sysfs_lock); | |
4044 | return res; | |
1da177e4 LT |
4045 | } |
4046 | ||
4047 | static ssize_t | |
4048 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
4049 | const char *page, size_t length) | |
4050 | { | |
4051 | struct queue_sysfs_entry *entry = to_queue(attr); | |
483f4afc AV |
4052 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
4053 | ||
4054 | ssize_t res; | |
1da177e4 | 4055 | |
1da177e4 | 4056 | if (!entry->store) |
6c1852a0 | 4057 | return -EIO; |
483f4afc AV |
4058 | mutex_lock(&q->sysfs_lock); |
4059 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4060 | mutex_unlock(&q->sysfs_lock); | |
4061 | return -ENOENT; | |
4062 | } | |
4063 | res = entry->store(q, page, length); | |
4064 | mutex_unlock(&q->sysfs_lock); | |
4065 | return res; | |
1da177e4 LT |
4066 | } |
4067 | ||
4068 | static struct sysfs_ops queue_sysfs_ops = { | |
4069 | .show = queue_attr_show, | |
4070 | .store = queue_attr_store, | |
4071 | }; | |
4072 | ||
93d17d3d | 4073 | static struct kobj_type queue_ktype = { |
1da177e4 LT |
4074 | .sysfs_ops = &queue_sysfs_ops, |
4075 | .default_attrs = default_attrs, | |
483f4afc | 4076 | .release = blk_release_queue, |
1da177e4 LT |
4077 | }; |
4078 | ||
4079 | int blk_register_queue(struct gendisk *disk) | |
4080 | { | |
4081 | int ret; | |
4082 | ||
4083 | request_queue_t *q = disk->queue; | |
4084 | ||
4085 | if (!q || !q->request_fn) | |
4086 | return -ENXIO; | |
4087 | ||
4088 | q->kobj.parent = kobject_get(&disk->kobj); | |
1da177e4 | 4089 | |
483f4afc | 4090 | ret = kobject_add(&q->kobj); |
1da177e4 LT |
4091 | if (ret < 0) |
4092 | return ret; | |
4093 | ||
483f4afc AV |
4094 | kobject_uevent(&q->kobj, KOBJ_ADD); |
4095 | ||
1da177e4 LT |
4096 | ret = elv_register_queue(q); |
4097 | if (ret) { | |
483f4afc AV |
4098 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4099 | kobject_del(&q->kobj); | |
1da177e4 LT |
4100 | return ret; |
4101 | } | |
4102 | ||
4103 | return 0; | |
4104 | } | |
4105 | ||
4106 | void blk_unregister_queue(struct gendisk *disk) | |
4107 | { | |
4108 | request_queue_t *q = disk->queue; | |
4109 | ||
4110 | if (q && q->request_fn) { | |
4111 | elv_unregister_queue(q); | |
4112 | ||
483f4afc AV |
4113 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4114 | kobject_del(&q->kobj); | |
1da177e4 LT |
4115 | kobject_put(&disk->kobj); |
4116 | } | |
4117 | } |