]>
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; |
abae1fde | 259 | rq->next_rq = NULL; |
1da177e4 LT |
260 | } |
261 | ||
262 | /** | |
263 | * blk_queue_ordered - does this queue support ordered writes | |
797e7dbb TH |
264 | * @q: the request queue |
265 | * @ordered: one of QUEUE_ORDERED_* | |
fddfdeaf | 266 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes |
1da177e4 LT |
267 | * |
268 | * Description: | |
269 | * For journalled file systems, doing ordered writes on a commit | |
270 | * block instead of explicitly doing wait_on_buffer (which is bad | |
271 | * for performance) can be a big win. Block drivers supporting this | |
272 | * feature should call this function and indicate so. | |
273 | * | |
274 | **/ | |
797e7dbb TH |
275 | int blk_queue_ordered(request_queue_t *q, unsigned ordered, |
276 | prepare_flush_fn *prepare_flush_fn) | |
277 | { | |
278 | if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && | |
279 | prepare_flush_fn == NULL) { | |
280 | printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); | |
281 | return -EINVAL; | |
282 | } | |
283 | ||
284 | if (ordered != QUEUE_ORDERED_NONE && | |
285 | ordered != QUEUE_ORDERED_DRAIN && | |
286 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
287 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
288 | ordered != QUEUE_ORDERED_TAG && | |
289 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
290 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
291 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
292 | return -EINVAL; | |
1da177e4 | 293 | } |
797e7dbb | 294 | |
60481b12 | 295 | q->ordered = ordered; |
797e7dbb TH |
296 | q->next_ordered = ordered; |
297 | q->prepare_flush_fn = prepare_flush_fn; | |
298 | ||
299 | return 0; | |
1da177e4 LT |
300 | } |
301 | ||
302 | EXPORT_SYMBOL(blk_queue_ordered); | |
303 | ||
304 | /** | |
305 | * blk_queue_issue_flush_fn - set function for issuing a flush | |
306 | * @q: the request queue | |
307 | * @iff: the function to be called issuing the flush | |
308 | * | |
309 | * Description: | |
310 | * If a driver supports issuing a flush command, the support is notified | |
311 | * to the block layer by defining it through this call. | |
312 | * | |
313 | **/ | |
314 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) | |
315 | { | |
316 | q->issue_flush_fn = iff; | |
317 | } | |
318 | ||
319 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); | |
320 | ||
321 | /* | |
322 | * Cache flushing for ordered writes handling | |
323 | */ | |
797e7dbb | 324 | inline unsigned blk_ordered_cur_seq(request_queue_t *q) |
1da177e4 | 325 | { |
797e7dbb TH |
326 | if (!q->ordseq) |
327 | return 0; | |
328 | return 1 << ffz(q->ordseq); | |
1da177e4 LT |
329 | } |
330 | ||
797e7dbb | 331 | unsigned blk_ordered_req_seq(struct request *rq) |
1da177e4 | 332 | { |
1da177e4 LT |
333 | request_queue_t *q = rq->q; |
334 | ||
797e7dbb | 335 | BUG_ON(q->ordseq == 0); |
8922e16c | 336 | |
797e7dbb TH |
337 | if (rq == &q->pre_flush_rq) |
338 | return QUEUE_ORDSEQ_PREFLUSH; | |
339 | if (rq == &q->bar_rq) | |
340 | return QUEUE_ORDSEQ_BAR; | |
341 | if (rq == &q->post_flush_rq) | |
342 | return QUEUE_ORDSEQ_POSTFLUSH; | |
1da177e4 | 343 | |
bc90ba09 TH |
344 | /* |
345 | * !fs requests don't need to follow barrier ordering. Always | |
346 | * put them at the front. This fixes the following deadlock. | |
347 | * | |
348 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
349 | */ | |
350 | if (!blk_fs_request(rq)) | |
351 | return QUEUE_ORDSEQ_DRAIN; | |
352 | ||
4aff5e23 JA |
353 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == |
354 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
797e7dbb TH |
355 | return QUEUE_ORDSEQ_DRAIN; |
356 | else | |
357 | return QUEUE_ORDSEQ_DONE; | |
1da177e4 LT |
358 | } |
359 | ||
797e7dbb | 360 | void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error) |
1da177e4 | 361 | { |
797e7dbb TH |
362 | struct request *rq; |
363 | int uptodate; | |
1da177e4 | 364 | |
797e7dbb TH |
365 | if (error && !q->orderr) |
366 | q->orderr = error; | |
1da177e4 | 367 | |
797e7dbb TH |
368 | BUG_ON(q->ordseq & seq); |
369 | q->ordseq |= seq; | |
1da177e4 | 370 | |
797e7dbb TH |
371 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) |
372 | return; | |
1da177e4 LT |
373 | |
374 | /* | |
797e7dbb | 375 | * Okay, sequence complete. |
1da177e4 | 376 | */ |
797e7dbb TH |
377 | rq = q->orig_bar_rq; |
378 | uptodate = q->orderr ? q->orderr : 1; | |
1da177e4 | 379 | |
797e7dbb | 380 | q->ordseq = 0; |
1da177e4 | 381 | |
797e7dbb TH |
382 | end_that_request_first(rq, uptodate, rq->hard_nr_sectors); |
383 | end_that_request_last(rq, uptodate); | |
1da177e4 LT |
384 | } |
385 | ||
797e7dbb | 386 | static void pre_flush_end_io(struct request *rq, int error) |
1da177e4 | 387 | { |
797e7dbb TH |
388 | elv_completed_request(rq->q, rq); |
389 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
390 | } | |
1da177e4 | 391 | |
797e7dbb TH |
392 | static void bar_end_io(struct request *rq, int error) |
393 | { | |
394 | elv_completed_request(rq->q, rq); | |
395 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
396 | } | |
1da177e4 | 397 | |
797e7dbb TH |
398 | static void post_flush_end_io(struct request *rq, int error) |
399 | { | |
400 | elv_completed_request(rq->q, rq); | |
401 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
402 | } | |
1da177e4 | 403 | |
797e7dbb TH |
404 | static void queue_flush(request_queue_t *q, unsigned which) |
405 | { | |
406 | struct request *rq; | |
407 | rq_end_io_fn *end_io; | |
1da177e4 | 408 | |
797e7dbb TH |
409 | if (which == QUEUE_ORDERED_PREFLUSH) { |
410 | rq = &q->pre_flush_rq; | |
411 | end_io = pre_flush_end_io; | |
412 | } else { | |
413 | rq = &q->post_flush_rq; | |
414 | end_io = post_flush_end_io; | |
1da177e4 | 415 | } |
797e7dbb | 416 | |
4aff5e23 | 417 | rq->cmd_flags = REQ_HARDBARRIER; |
797e7dbb | 418 | rq_init(q, rq); |
797e7dbb | 419 | rq->elevator_private = NULL; |
c00895ab | 420 | rq->elevator_private2 = NULL; |
797e7dbb | 421 | rq->rq_disk = q->bar_rq.rq_disk; |
797e7dbb TH |
422 | rq->end_io = end_io; |
423 | q->prepare_flush_fn(q, rq); | |
424 | ||
30e9656c | 425 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
1da177e4 LT |
426 | } |
427 | ||
797e7dbb TH |
428 | static inline struct request *start_ordered(request_queue_t *q, |
429 | struct request *rq) | |
1da177e4 | 430 | { |
797e7dbb TH |
431 | q->bi_size = 0; |
432 | q->orderr = 0; | |
433 | q->ordered = q->next_ordered; | |
434 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
435 | ||
436 | /* | |
437 | * Prep proxy barrier request. | |
438 | */ | |
439 | blkdev_dequeue_request(rq); | |
440 | q->orig_bar_rq = rq; | |
441 | rq = &q->bar_rq; | |
4aff5e23 | 442 | rq->cmd_flags = 0; |
797e7dbb | 443 | rq_init(q, rq); |
4aff5e23 JA |
444 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) |
445 | rq->cmd_flags |= REQ_RW; | |
446 | rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0; | |
797e7dbb | 447 | rq->elevator_private = NULL; |
c00895ab | 448 | rq->elevator_private2 = NULL; |
797e7dbb TH |
449 | init_request_from_bio(rq, q->orig_bar_rq->bio); |
450 | rq->end_io = bar_end_io; | |
451 | ||
452 | /* | |
453 | * Queue ordered sequence. As we stack them at the head, we | |
454 | * need to queue in reverse order. Note that we rely on that | |
455 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
456 | * request gets inbetween ordered sequence. | |
457 | */ | |
458 | if (q->ordered & QUEUE_ORDERED_POSTFLUSH) | |
459 | queue_flush(q, QUEUE_ORDERED_POSTFLUSH); | |
460 | else | |
461 | q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; | |
462 | ||
30e9656c | 463 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
797e7dbb TH |
464 | |
465 | if (q->ordered & QUEUE_ORDERED_PREFLUSH) { | |
466 | queue_flush(q, QUEUE_ORDERED_PREFLUSH); | |
467 | rq = &q->pre_flush_rq; | |
468 | } else | |
469 | q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; | |
1da177e4 | 470 | |
797e7dbb TH |
471 | if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) |
472 | q->ordseq |= QUEUE_ORDSEQ_DRAIN; | |
473 | else | |
474 | rq = NULL; | |
475 | ||
476 | return rq; | |
1da177e4 LT |
477 | } |
478 | ||
797e7dbb | 479 | int blk_do_ordered(request_queue_t *q, struct request **rqp) |
1da177e4 | 480 | { |
9a7a67af | 481 | struct request *rq = *rqp; |
797e7dbb | 482 | int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); |
1da177e4 | 483 | |
797e7dbb TH |
484 | if (!q->ordseq) { |
485 | if (!is_barrier) | |
486 | return 1; | |
1da177e4 | 487 | |
797e7dbb TH |
488 | if (q->next_ordered != QUEUE_ORDERED_NONE) { |
489 | *rqp = start_ordered(q, rq); | |
490 | return 1; | |
491 | } else { | |
492 | /* | |
493 | * This can happen when the queue switches to | |
494 | * ORDERED_NONE while this request is on it. | |
495 | */ | |
496 | blkdev_dequeue_request(rq); | |
497 | end_that_request_first(rq, -EOPNOTSUPP, | |
498 | rq->hard_nr_sectors); | |
499 | end_that_request_last(rq, -EOPNOTSUPP); | |
500 | *rqp = NULL; | |
501 | return 0; | |
502 | } | |
503 | } | |
1da177e4 | 504 | |
9a7a67af JA |
505 | /* |
506 | * Ordered sequence in progress | |
507 | */ | |
508 | ||
509 | /* Special requests are not subject to ordering rules. */ | |
510 | if (!blk_fs_request(rq) && | |
511 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
512 | return 1; | |
513 | ||
797e7dbb | 514 | if (q->ordered & QUEUE_ORDERED_TAG) { |
9a7a67af | 515 | /* Ordered by tag. Blocking the next barrier is enough. */ |
797e7dbb TH |
516 | if (is_barrier && rq != &q->bar_rq) |
517 | *rqp = NULL; | |
9a7a67af JA |
518 | } else { |
519 | /* Ordered by draining. Wait for turn. */ | |
520 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
521 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
522 | *rqp = NULL; | |
1da177e4 LT |
523 | } |
524 | ||
525 | return 1; | |
526 | } | |
527 | ||
797e7dbb | 528 | static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error) |
1da177e4 | 529 | { |
797e7dbb | 530 | request_queue_t *q = bio->bi_private; |
797e7dbb TH |
531 | |
532 | /* | |
533 | * This is dry run, restore bio_sector and size. We'll finish | |
534 | * this request again with the original bi_end_io after an | |
535 | * error occurs or post flush is complete. | |
536 | */ | |
537 | q->bi_size += bytes; | |
538 | ||
539 | if (bio->bi_size) | |
540 | return 1; | |
541 | ||
797e7dbb TH |
542 | /* Reset bio */ |
543 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
544 | bio->bi_size = q->bi_size; | |
545 | bio->bi_sector -= (q->bi_size >> 9); | |
546 | q->bi_size = 0; | |
547 | ||
548 | return 0; | |
1da177e4 | 549 | } |
1da177e4 | 550 | |
1ea25ecb JA |
551 | static int ordered_bio_endio(struct request *rq, struct bio *bio, |
552 | unsigned int nbytes, int error) | |
1da177e4 | 553 | { |
797e7dbb TH |
554 | request_queue_t *q = rq->q; |
555 | bio_end_io_t *endio; | |
556 | void *private; | |
557 | ||
558 | if (&q->bar_rq != rq) | |
559 | return 0; | |
560 | ||
561 | /* | |
562 | * Okay, this is the barrier request in progress, dry finish it. | |
563 | */ | |
564 | if (error && !q->orderr) | |
565 | q->orderr = error; | |
566 | ||
567 | endio = bio->bi_end_io; | |
568 | private = bio->bi_private; | |
569 | bio->bi_end_io = flush_dry_bio_endio; | |
570 | bio->bi_private = q; | |
571 | ||
572 | bio_endio(bio, nbytes, error); | |
573 | ||
574 | bio->bi_end_io = endio; | |
575 | bio->bi_private = private; | |
576 | ||
577 | return 1; | |
1da177e4 | 578 | } |
1da177e4 LT |
579 | |
580 | /** | |
581 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
582 | * @q: the request queue for the device | |
583 | * @dma_addr: bus address limit | |
584 | * | |
585 | * Description: | |
586 | * Different hardware can have different requirements as to what pages | |
587 | * it can do I/O directly to. A low level driver can call | |
588 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
5ee1af9f | 589 | * buffers for doing I/O to pages residing above @page. |
1da177e4 LT |
590 | **/ |
591 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) | |
592 | { | |
593 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
5ee1af9f AK |
594 | int dma = 0; |
595 | ||
596 | q->bounce_gfp = GFP_NOIO; | |
597 | #if BITS_PER_LONG == 64 | |
598 | /* Assume anything <= 4GB can be handled by IOMMU. | |
599 | Actually some IOMMUs can handle everything, but I don't | |
600 | know of a way to test this here. */ | |
8269730b | 601 | if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
5ee1af9f AK |
602 | dma = 1; |
603 | q->bounce_pfn = max_low_pfn; | |
604 | #else | |
605 | if (bounce_pfn < blk_max_low_pfn) | |
606 | dma = 1; | |
607 | q->bounce_pfn = bounce_pfn; | |
608 | #endif | |
609 | if (dma) { | |
1da177e4 LT |
610 | init_emergency_isa_pool(); |
611 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
5ee1af9f AK |
612 | q->bounce_pfn = bounce_pfn; |
613 | } | |
1da177e4 LT |
614 | } |
615 | ||
616 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
617 | ||
618 | /** | |
619 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
620 | * @q: the request queue for the device | |
621 | * @max_sectors: max sectors in the usual 512b unit | |
622 | * | |
623 | * Description: | |
624 | * Enables a low level driver to set an upper limit on the size of | |
625 | * received requests. | |
626 | **/ | |
2cb2e147 | 627 | void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors) |
1da177e4 LT |
628 | { |
629 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
630 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
631 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
632 | } | |
633 | ||
defd94b7 MC |
634 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
635 | q->max_hw_sectors = q->max_sectors = max_sectors; | |
636 | else { | |
637 | q->max_sectors = BLK_DEF_MAX_SECTORS; | |
638 | q->max_hw_sectors = max_sectors; | |
639 | } | |
1da177e4 LT |
640 | } |
641 | ||
642 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
643 | ||
644 | /** | |
645 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
646 | * @q: the request queue for the device | |
647 | * @max_segments: max number of segments | |
648 | * | |
649 | * Description: | |
650 | * Enables a low level driver to set an upper limit on the number of | |
651 | * physical data segments in a request. This would be the largest sized | |
652 | * scatter list the driver could handle. | |
653 | **/ | |
654 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) | |
655 | { | |
656 | if (!max_segments) { | |
657 | max_segments = 1; | |
658 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
659 | } | |
660 | ||
661 | q->max_phys_segments = max_segments; | |
662 | } | |
663 | ||
664 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
665 | ||
666 | /** | |
667 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
668 | * @q: the request queue for the device | |
669 | * @max_segments: max number of segments | |
670 | * | |
671 | * Description: | |
672 | * Enables a low level driver to set an upper limit on the number of | |
673 | * hw data segments in a request. This would be the largest number of | |
674 | * address/length pairs the host adapter can actually give as once | |
675 | * to the device. | |
676 | **/ | |
677 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) | |
678 | { | |
679 | if (!max_segments) { | |
680 | max_segments = 1; | |
681 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
682 | } | |
683 | ||
684 | q->max_hw_segments = max_segments; | |
685 | } | |
686 | ||
687 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
688 | ||
689 | /** | |
690 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
691 | * @q: the request queue for the device | |
692 | * @max_size: max size of segment in bytes | |
693 | * | |
694 | * Description: | |
695 | * Enables a low level driver to set an upper limit on the size of a | |
696 | * coalesced segment | |
697 | **/ | |
698 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) | |
699 | { | |
700 | if (max_size < PAGE_CACHE_SIZE) { | |
701 | max_size = PAGE_CACHE_SIZE; | |
702 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
703 | } | |
704 | ||
705 | q->max_segment_size = max_size; | |
706 | } | |
707 | ||
708 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
709 | ||
710 | /** | |
711 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
712 | * @q: the request queue for the device | |
713 | * @size: the hardware sector size, in bytes | |
714 | * | |
715 | * Description: | |
716 | * This should typically be set to the lowest possible sector size | |
717 | * that the hardware can operate on (possible without reverting to | |
718 | * even internal read-modify-write operations). Usually the default | |
719 | * of 512 covers most hardware. | |
720 | **/ | |
721 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) | |
722 | { | |
723 | q->hardsect_size = size; | |
724 | } | |
725 | ||
726 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
727 | ||
728 | /* | |
729 | * Returns the minimum that is _not_ zero, unless both are zero. | |
730 | */ | |
731 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
732 | ||
733 | /** | |
734 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
735 | * @t: the stacking driver (top) | |
736 | * @b: the underlying device (bottom) | |
737 | **/ | |
738 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) | |
739 | { | |
740 | /* zero is "infinity" */ | |
defd94b7 MC |
741 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); |
742 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | |
1da177e4 LT |
743 | |
744 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
745 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
746 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
747 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
89e5c8b5 N |
748 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
749 | clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); | |
1da177e4 LT |
750 | } |
751 | ||
752 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
753 | ||
754 | /** | |
755 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
756 | * @q: the request queue for the device | |
757 | * @mask: the memory boundary mask | |
758 | **/ | |
759 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) | |
760 | { | |
761 | if (mask < PAGE_CACHE_SIZE - 1) { | |
762 | mask = PAGE_CACHE_SIZE - 1; | |
763 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
764 | } | |
765 | ||
766 | q->seg_boundary_mask = mask; | |
767 | } | |
768 | ||
769 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
770 | ||
771 | /** | |
772 | * blk_queue_dma_alignment - set dma length and memory alignment | |
773 | * @q: the request queue for the device | |
774 | * @mask: alignment mask | |
775 | * | |
776 | * description: | |
777 | * set required memory and length aligment for direct dma transactions. | |
778 | * this is used when buiding direct io requests for the queue. | |
779 | * | |
780 | **/ | |
781 | void blk_queue_dma_alignment(request_queue_t *q, int mask) | |
782 | { | |
783 | q->dma_alignment = mask; | |
784 | } | |
785 | ||
786 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
787 | ||
788 | /** | |
789 | * blk_queue_find_tag - find a request by its tag and queue | |
1da177e4 LT |
790 | * @q: The request queue for the device |
791 | * @tag: The tag of the request | |
792 | * | |
793 | * Notes: | |
794 | * Should be used when a device returns a tag and you want to match | |
795 | * it with a request. | |
796 | * | |
797 | * no locks need be held. | |
798 | **/ | |
799 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) | |
800 | { | |
f583f492 | 801 | return blk_map_queue_find_tag(q->queue_tags, tag); |
1da177e4 LT |
802 | } |
803 | ||
804 | EXPORT_SYMBOL(blk_queue_find_tag); | |
805 | ||
806 | /** | |
492dfb48 JB |
807 | * __blk_free_tags - release a given set of tag maintenance info |
808 | * @bqt: the tag map to free | |
1da177e4 | 809 | * |
492dfb48 JB |
810 | * Tries to free the specified @bqt@. Returns true if it was |
811 | * actually freed and false if there are still references using it | |
812 | */ | |
813 | static int __blk_free_tags(struct blk_queue_tag *bqt) | |
1da177e4 | 814 | { |
492dfb48 | 815 | int retval; |
1da177e4 | 816 | |
492dfb48 JB |
817 | retval = atomic_dec_and_test(&bqt->refcnt); |
818 | if (retval) { | |
1da177e4 LT |
819 | BUG_ON(bqt->busy); |
820 | BUG_ON(!list_empty(&bqt->busy_list)); | |
821 | ||
822 | kfree(bqt->tag_index); | |
823 | bqt->tag_index = NULL; | |
824 | ||
825 | kfree(bqt->tag_map); | |
826 | bqt->tag_map = NULL; | |
827 | ||
828 | kfree(bqt); | |
492dfb48 | 829 | |
1da177e4 LT |
830 | } |
831 | ||
492dfb48 JB |
832 | return retval; |
833 | } | |
834 | ||
835 | /** | |
836 | * __blk_queue_free_tags - release tag maintenance info | |
837 | * @q: the request queue for the device | |
838 | * | |
839 | * Notes: | |
840 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
841 | * has been used. So there's no need to call this directly. | |
842 | **/ | |
843 | static void __blk_queue_free_tags(request_queue_t *q) | |
844 | { | |
845 | struct blk_queue_tag *bqt = q->queue_tags; | |
846 | ||
847 | if (!bqt) | |
848 | return; | |
849 | ||
850 | __blk_free_tags(bqt); | |
851 | ||
1da177e4 LT |
852 | q->queue_tags = NULL; |
853 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
854 | } | |
855 | ||
492dfb48 JB |
856 | |
857 | /** | |
858 | * blk_free_tags - release a given set of tag maintenance info | |
859 | * @bqt: the tag map to free | |
860 | * | |
861 | * For externally managed @bqt@ frees the map. Callers of this | |
862 | * function must guarantee to have released all the queues that | |
863 | * might have been using this tag map. | |
864 | */ | |
865 | void blk_free_tags(struct blk_queue_tag *bqt) | |
866 | { | |
867 | if (unlikely(!__blk_free_tags(bqt))) | |
868 | BUG(); | |
869 | } | |
870 | EXPORT_SYMBOL(blk_free_tags); | |
871 | ||
1da177e4 LT |
872 | /** |
873 | * blk_queue_free_tags - release tag maintenance info | |
874 | * @q: the request queue for the device | |
875 | * | |
876 | * Notes: | |
877 | * This is used to disabled tagged queuing to a device, yet leave | |
878 | * queue in function. | |
879 | **/ | |
880 | void blk_queue_free_tags(request_queue_t *q) | |
881 | { | |
882 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
883 | } | |
884 | ||
885 | EXPORT_SYMBOL(blk_queue_free_tags); | |
886 | ||
887 | static int | |
888 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) | |
889 | { | |
1da177e4 LT |
890 | struct request **tag_index; |
891 | unsigned long *tag_map; | |
fa72b903 | 892 | int nr_ulongs; |
1da177e4 | 893 | |
492dfb48 | 894 | if (q && depth > q->nr_requests * 2) { |
1da177e4 LT |
895 | depth = q->nr_requests * 2; |
896 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
897 | __FUNCTION__, depth); | |
898 | } | |
899 | ||
f68110fc | 900 | tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
1da177e4 LT |
901 | if (!tag_index) |
902 | goto fail; | |
903 | ||
f7d37d02 | 904 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
f68110fc | 905 | tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
1da177e4 LT |
906 | if (!tag_map) |
907 | goto fail; | |
908 | ||
ba025082 | 909 | tags->real_max_depth = depth; |
1da177e4 | 910 | tags->max_depth = depth; |
1da177e4 LT |
911 | tags->tag_index = tag_index; |
912 | tags->tag_map = tag_map; | |
913 | ||
1da177e4 LT |
914 | return 0; |
915 | fail: | |
916 | kfree(tag_index); | |
917 | return -ENOMEM; | |
918 | } | |
919 | ||
492dfb48 JB |
920 | static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q, |
921 | int depth) | |
922 | { | |
923 | struct blk_queue_tag *tags; | |
924 | ||
925 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
926 | if (!tags) | |
927 | goto fail; | |
928 | ||
929 | if (init_tag_map(q, tags, depth)) | |
930 | goto fail; | |
931 | ||
932 | INIT_LIST_HEAD(&tags->busy_list); | |
933 | tags->busy = 0; | |
934 | atomic_set(&tags->refcnt, 1); | |
935 | return tags; | |
936 | fail: | |
937 | kfree(tags); | |
938 | return NULL; | |
939 | } | |
940 | ||
941 | /** | |
942 | * blk_init_tags - initialize the tag info for an external tag map | |
943 | * @depth: the maximum queue depth supported | |
944 | * @tags: the tag to use | |
945 | **/ | |
946 | struct blk_queue_tag *blk_init_tags(int depth) | |
947 | { | |
948 | return __blk_queue_init_tags(NULL, depth); | |
949 | } | |
950 | EXPORT_SYMBOL(blk_init_tags); | |
951 | ||
1da177e4 LT |
952 | /** |
953 | * blk_queue_init_tags - initialize the queue tag info | |
954 | * @q: the request queue for the device | |
955 | * @depth: the maximum queue depth supported | |
956 | * @tags: the tag to use | |
957 | **/ | |
958 | int blk_queue_init_tags(request_queue_t *q, int depth, | |
959 | struct blk_queue_tag *tags) | |
960 | { | |
961 | int rc; | |
962 | ||
963 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
964 | ||
965 | if (!tags && !q->queue_tags) { | |
492dfb48 | 966 | tags = __blk_queue_init_tags(q, depth); |
1da177e4 | 967 | |
492dfb48 | 968 | if (!tags) |
1da177e4 | 969 | goto fail; |
1da177e4 LT |
970 | } else if (q->queue_tags) { |
971 | if ((rc = blk_queue_resize_tags(q, depth))) | |
972 | return rc; | |
973 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
974 | return 0; | |
975 | } else | |
976 | atomic_inc(&tags->refcnt); | |
977 | ||
978 | /* | |
979 | * assign it, all done | |
980 | */ | |
981 | q->queue_tags = tags; | |
982 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
983 | return 0; | |
984 | fail: | |
985 | kfree(tags); | |
986 | return -ENOMEM; | |
987 | } | |
988 | ||
989 | EXPORT_SYMBOL(blk_queue_init_tags); | |
990 | ||
991 | /** | |
992 | * blk_queue_resize_tags - change the queueing depth | |
993 | * @q: the request queue for the device | |
994 | * @new_depth: the new max command queueing depth | |
995 | * | |
996 | * Notes: | |
997 | * Must be called with the queue lock held. | |
998 | **/ | |
999 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) | |
1000 | { | |
1001 | struct blk_queue_tag *bqt = q->queue_tags; | |
1002 | struct request **tag_index; | |
1003 | unsigned long *tag_map; | |
fa72b903 | 1004 | int max_depth, nr_ulongs; |
1da177e4 LT |
1005 | |
1006 | if (!bqt) | |
1007 | return -ENXIO; | |
1008 | ||
ba025082 TH |
1009 | /* |
1010 | * if we already have large enough real_max_depth. just | |
1011 | * adjust max_depth. *NOTE* as requests with tag value | |
1012 | * between new_depth and real_max_depth can be in-flight, tag | |
1013 | * map can not be shrunk blindly here. | |
1014 | */ | |
1015 | if (new_depth <= bqt->real_max_depth) { | |
1016 | bqt->max_depth = new_depth; | |
1017 | return 0; | |
1018 | } | |
1019 | ||
492dfb48 JB |
1020 | /* |
1021 | * Currently cannot replace a shared tag map with a new | |
1022 | * one, so error out if this is the case | |
1023 | */ | |
1024 | if (atomic_read(&bqt->refcnt) != 1) | |
1025 | return -EBUSY; | |
1026 | ||
1da177e4 LT |
1027 | /* |
1028 | * save the old state info, so we can copy it back | |
1029 | */ | |
1030 | tag_index = bqt->tag_index; | |
1031 | tag_map = bqt->tag_map; | |
ba025082 | 1032 | max_depth = bqt->real_max_depth; |
1da177e4 LT |
1033 | |
1034 | if (init_tag_map(q, bqt, new_depth)) | |
1035 | return -ENOMEM; | |
1036 | ||
1037 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
f7d37d02 | 1038 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 1039 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
1da177e4 LT |
1040 | |
1041 | kfree(tag_index); | |
1042 | kfree(tag_map); | |
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
1047 | ||
1048 | /** | |
1049 | * blk_queue_end_tag - end tag operations for a request | |
1050 | * @q: the request queue for the device | |
1051 | * @rq: the request that has completed | |
1052 | * | |
1053 | * Description: | |
1054 | * Typically called when end_that_request_first() returns 0, meaning | |
1055 | * all transfers have been done for a request. It's important to call | |
1056 | * this function before end_that_request_last(), as that will put the | |
1057 | * request back on the free list thus corrupting the internal tag list. | |
1058 | * | |
1059 | * Notes: | |
1060 | * queue lock must be held. | |
1061 | **/ | |
1062 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) | |
1063 | { | |
1064 | struct blk_queue_tag *bqt = q->queue_tags; | |
1065 | int tag = rq->tag; | |
1066 | ||
1067 | BUG_ON(tag == -1); | |
1068 | ||
ba025082 | 1069 | if (unlikely(tag >= bqt->real_max_depth)) |
040c928c TH |
1070 | /* |
1071 | * This can happen after tag depth has been reduced. | |
1072 | * FIXME: how about a warning or info message here? | |
1073 | */ | |
1da177e4 LT |
1074 | return; |
1075 | ||
1076 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { | |
040c928c TH |
1077 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
1078 | __FUNCTION__, tag); | |
1da177e4 LT |
1079 | return; |
1080 | } | |
1081 | ||
1082 | list_del_init(&rq->queuelist); | |
4aff5e23 | 1083 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1084 | rq->tag = -1; |
1085 | ||
1086 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
040c928c TH |
1087 | printk(KERN_ERR "%s: tag %d is missing\n", |
1088 | __FUNCTION__, tag); | |
1da177e4 LT |
1089 | |
1090 | bqt->tag_index[tag] = NULL; | |
1091 | bqt->busy--; | |
1092 | } | |
1093 | ||
1094 | EXPORT_SYMBOL(blk_queue_end_tag); | |
1095 | ||
1096 | /** | |
1097 | * blk_queue_start_tag - find a free tag and assign it | |
1098 | * @q: the request queue for the device | |
1099 | * @rq: the block request that needs tagging | |
1100 | * | |
1101 | * Description: | |
1102 | * This can either be used as a stand-alone helper, or possibly be | |
1103 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
1104 | * automagically gets a tag assigned). Note that this function | |
1105 | * assumes that any type of request can be queued! if this is not | |
1106 | * true for your device, you must check the request type before | |
1107 | * calling this function. The request will also be removed from | |
1108 | * the request queue, so it's the drivers responsibility to readd | |
1109 | * it if it should need to be restarted for some reason. | |
1110 | * | |
1111 | * Notes: | |
1112 | * queue lock must be held. | |
1113 | **/ | |
1114 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) | |
1115 | { | |
1116 | struct blk_queue_tag *bqt = q->queue_tags; | |
2bf0fdad | 1117 | int tag; |
1da177e4 | 1118 | |
4aff5e23 | 1119 | if (unlikely((rq->cmd_flags & REQ_QUEUED))) { |
1da177e4 | 1120 | printk(KERN_ERR |
040c928c TH |
1121 | "%s: request %p for device [%s] already tagged %d", |
1122 | __FUNCTION__, rq, | |
1123 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1da177e4 LT |
1124 | BUG(); |
1125 | } | |
1126 | ||
059af497 JA |
1127 | /* |
1128 | * Protect against shared tag maps, as we may not have exclusive | |
1129 | * access to the tag map. | |
1130 | */ | |
1131 | do { | |
1132 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | |
1133 | if (tag >= bqt->max_depth) | |
1134 | return 1; | |
1da177e4 | 1135 | |
059af497 | 1136 | } while (test_and_set_bit(tag, bqt->tag_map)); |
1da177e4 | 1137 | |
4aff5e23 | 1138 | rq->cmd_flags |= REQ_QUEUED; |
1da177e4 LT |
1139 | rq->tag = tag; |
1140 | bqt->tag_index[tag] = rq; | |
1141 | blkdev_dequeue_request(rq); | |
1142 | list_add(&rq->queuelist, &bqt->busy_list); | |
1143 | bqt->busy++; | |
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1148 | ||
1149 | /** | |
1150 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1151 | * @q: the request queue for the device | |
1152 | * | |
1153 | * Description: | |
1154 | * Hardware conditions may dictate a need to stop all pending requests. | |
1155 | * In this case, we will safely clear the block side of the tag queue and | |
1156 | * readd all requests to the request queue in the right order. | |
1157 | * | |
1158 | * Notes: | |
1159 | * queue lock must be held. | |
1160 | **/ | |
1161 | void blk_queue_invalidate_tags(request_queue_t *q) | |
1162 | { | |
1163 | struct blk_queue_tag *bqt = q->queue_tags; | |
1164 | struct list_head *tmp, *n; | |
1165 | struct request *rq; | |
1166 | ||
1167 | list_for_each_safe(tmp, n, &bqt->busy_list) { | |
1168 | rq = list_entry_rq(tmp); | |
1169 | ||
1170 | if (rq->tag == -1) { | |
040c928c TH |
1171 | printk(KERN_ERR |
1172 | "%s: bad tag found on list\n", __FUNCTION__); | |
1da177e4 | 1173 | list_del_init(&rq->queuelist); |
4aff5e23 | 1174 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1175 | } else |
1176 | blk_queue_end_tag(q, rq); | |
1177 | ||
4aff5e23 | 1178 | rq->cmd_flags &= ~REQ_STARTED; |
1da177e4 LT |
1179 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
1180 | } | |
1181 | } | |
1182 | ||
1183 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1184 | ||
1da177e4 LT |
1185 | void blk_dump_rq_flags(struct request *rq, char *msg) |
1186 | { | |
1187 | int bit; | |
1188 | ||
4aff5e23 JA |
1189 | printk("%s: dev %s: type=%x, flags=%x\n", msg, |
1190 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
1191 | rq->cmd_flags); | |
1da177e4 LT |
1192 | |
1193 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1194 | rq->nr_sectors, | |
1195 | rq->current_nr_sectors); | |
1196 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1197 | ||
4aff5e23 | 1198 | if (blk_pc_request(rq)) { |
1da177e4 LT |
1199 | printk("cdb: "); |
1200 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1201 | printk("%02x ", rq->cmd[bit]); | |
1202 | printk("\n"); | |
1203 | } | |
1204 | } | |
1205 | ||
1206 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1207 | ||
1208 | void blk_recount_segments(request_queue_t *q, struct bio *bio) | |
1209 | { | |
1210 | struct bio_vec *bv, *bvprv = NULL; | |
1211 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; | |
1212 | int high, highprv = 1; | |
1213 | ||
1214 | if (unlikely(!bio->bi_io_vec)) | |
1215 | return; | |
1216 | ||
1217 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1218 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; | |
1219 | bio_for_each_segment(bv, bio, i) { | |
1220 | /* | |
1221 | * the trick here is making sure that a high page is never | |
1222 | * considered part of another segment, since that might | |
1223 | * change with the bounce page. | |
1224 | */ | |
f772b3d9 | 1225 | high = page_to_pfn(bv->bv_page) > q->bounce_pfn; |
1da177e4 LT |
1226 | if (high || highprv) |
1227 | goto new_hw_segment; | |
1228 | if (cluster) { | |
1229 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1230 | goto new_segment; | |
1231 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1232 | goto new_segment; | |
1233 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1234 | goto new_segment; | |
1235 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1236 | goto new_hw_segment; | |
1237 | ||
1238 | seg_size += bv->bv_len; | |
1239 | hw_seg_size += bv->bv_len; | |
1240 | bvprv = bv; | |
1241 | continue; | |
1242 | } | |
1243 | new_segment: | |
1244 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
1245 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { | |
1246 | hw_seg_size += bv->bv_len; | |
1247 | } else { | |
1248 | new_hw_segment: | |
1249 | if (hw_seg_size > bio->bi_hw_front_size) | |
1250 | bio->bi_hw_front_size = hw_seg_size; | |
1251 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | |
1252 | nr_hw_segs++; | |
1253 | } | |
1254 | ||
1255 | nr_phys_segs++; | |
1256 | bvprv = bv; | |
1257 | seg_size = bv->bv_len; | |
1258 | highprv = high; | |
1259 | } | |
1260 | if (hw_seg_size > bio->bi_hw_back_size) | |
1261 | bio->bi_hw_back_size = hw_seg_size; | |
1262 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) | |
1263 | bio->bi_hw_front_size = hw_seg_size; | |
1264 | bio->bi_phys_segments = nr_phys_segs; | |
1265 | bio->bi_hw_segments = nr_hw_segs; | |
1266 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1267 | } | |
387bb173 | 1268 | EXPORT_SYMBOL(blk_recount_segments); |
1da177e4 | 1269 | |
93d17d3d | 1270 | static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1271 | struct bio *nxt) |
1272 | { | |
1273 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1274 | return 0; | |
1275 | ||
1276 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1277 | return 0; | |
1278 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1279 | return 0; | |
1280 | ||
1281 | /* | |
1282 | * bio and nxt are contigous in memory, check if the queue allows | |
1283 | * these two to be merged into one | |
1284 | */ | |
1285 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1286 | return 1; | |
1287 | ||
1288 | return 0; | |
1289 | } | |
1290 | ||
93d17d3d | 1291 | static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, |
1da177e4 LT |
1292 | struct bio *nxt) |
1293 | { | |
1294 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1295 | blk_recount_segments(q, bio); | |
1296 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1297 | blk_recount_segments(q, nxt); | |
1298 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
32eef964 | 1299 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size)) |
1da177e4 | 1300 | return 0; |
32eef964 | 1301 | if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size) |
1da177e4 LT |
1302 | return 0; |
1303 | ||
1304 | return 1; | |
1305 | } | |
1306 | ||
1da177e4 LT |
1307 | /* |
1308 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1309 | * must make sure sg can hold rq->nr_phys_segments entries | |
1310 | */ | |
1311 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) | |
1312 | { | |
1313 | struct bio_vec *bvec, *bvprv; | |
1314 | struct bio *bio; | |
1315 | int nsegs, i, cluster; | |
1316 | ||
1317 | nsegs = 0; | |
1318 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1319 | ||
1320 | /* | |
1321 | * for each bio in rq | |
1322 | */ | |
1323 | bvprv = NULL; | |
1324 | rq_for_each_bio(bio, rq) { | |
1325 | /* | |
1326 | * for each segment in bio | |
1327 | */ | |
1328 | bio_for_each_segment(bvec, bio, i) { | |
1329 | int nbytes = bvec->bv_len; | |
1330 | ||
1331 | if (bvprv && cluster) { | |
1332 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) | |
1333 | goto new_segment; | |
1334 | ||
1335 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | |
1336 | goto new_segment; | |
1337 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1338 | goto new_segment; | |
1339 | ||
1340 | sg[nsegs - 1].length += nbytes; | |
1341 | } else { | |
1342 | new_segment: | |
1343 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); | |
1344 | sg[nsegs].page = bvec->bv_page; | |
1345 | sg[nsegs].length = nbytes; | |
1346 | sg[nsegs].offset = bvec->bv_offset; | |
1347 | ||
1348 | nsegs++; | |
1349 | } | |
1350 | bvprv = bvec; | |
1351 | } /* segments in bio */ | |
1352 | } /* bios in rq */ | |
1353 | ||
1354 | return nsegs; | |
1355 | } | |
1356 | ||
1357 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1358 | ||
1359 | /* | |
1360 | * the standard queue merge functions, can be overridden with device | |
1361 | * specific ones if so desired | |
1362 | */ | |
1363 | ||
1364 | static inline int ll_new_mergeable(request_queue_t *q, | |
1365 | struct request *req, | |
1366 | struct bio *bio) | |
1367 | { | |
1368 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1369 | ||
1370 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1371 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1372 | if (req == q->last_merge) |
1373 | q->last_merge = NULL; | |
1374 | return 0; | |
1375 | } | |
1376 | ||
1377 | /* | |
1378 | * A hw segment is just getting larger, bump just the phys | |
1379 | * counter. | |
1380 | */ | |
1381 | req->nr_phys_segments += nr_phys_segs; | |
1382 | return 1; | |
1383 | } | |
1384 | ||
1385 | static inline int ll_new_hw_segment(request_queue_t *q, | |
1386 | struct request *req, | |
1387 | struct bio *bio) | |
1388 | { | |
1389 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1390 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1391 | ||
1392 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1393 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1394 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1395 | if (req == q->last_merge) |
1396 | q->last_merge = NULL; | |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | /* | |
1401 | * This will form the start of a new hw segment. Bump both | |
1402 | * counters. | |
1403 | */ | |
1404 | req->nr_hw_segments += nr_hw_segs; | |
1405 | req->nr_phys_segments += nr_phys_segs; | |
1406 | return 1; | |
1407 | } | |
1408 | ||
1aa4f24f | 1409 | int ll_back_merge_fn(request_queue_t *q, struct request *req, struct bio *bio) |
1da177e4 | 1410 | { |
defd94b7 | 1411 | unsigned short max_sectors; |
1da177e4 LT |
1412 | int len; |
1413 | ||
defd94b7 MC |
1414 | if (unlikely(blk_pc_request(req))) |
1415 | max_sectors = q->max_hw_sectors; | |
1416 | else | |
1417 | max_sectors = q->max_sectors; | |
1418 | ||
1419 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1420 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1421 | if (req == q->last_merge) |
1422 | q->last_merge = NULL; | |
1423 | return 0; | |
1424 | } | |
1425 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1426 | blk_recount_segments(q, req->biotail); | |
1427 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1428 | blk_recount_segments(q, bio); | |
1429 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1430 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1431 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1432 | int mergeable = ll_new_mergeable(q, req, bio); | |
1433 | ||
1434 | if (mergeable) { | |
1435 | if (req->nr_hw_segments == 1) | |
1436 | req->bio->bi_hw_front_size = len; | |
1437 | if (bio->bi_hw_segments == 1) | |
1438 | bio->bi_hw_back_size = len; | |
1439 | } | |
1440 | return mergeable; | |
1441 | } | |
1442 | ||
1443 | return ll_new_hw_segment(q, req, bio); | |
1444 | } | |
1aa4f24f | 1445 | EXPORT_SYMBOL(ll_back_merge_fn); |
1da177e4 LT |
1446 | |
1447 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, | |
1448 | struct bio *bio) | |
1449 | { | |
defd94b7 | 1450 | unsigned short max_sectors; |
1da177e4 LT |
1451 | int len; |
1452 | ||
defd94b7 MC |
1453 | if (unlikely(blk_pc_request(req))) |
1454 | max_sectors = q->max_hw_sectors; | |
1455 | else | |
1456 | max_sectors = q->max_sectors; | |
1457 | ||
1458 | ||
1459 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1460 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1461 | if (req == q->last_merge) |
1462 | q->last_merge = NULL; | |
1463 | return 0; | |
1464 | } | |
1465 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1466 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1467 | blk_recount_segments(q, bio); | |
1468 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1469 | blk_recount_segments(q, req->bio); | |
1470 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1471 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1472 | int mergeable = ll_new_mergeable(q, req, bio); | |
1473 | ||
1474 | if (mergeable) { | |
1475 | if (bio->bi_hw_segments == 1) | |
1476 | bio->bi_hw_front_size = len; | |
1477 | if (req->nr_hw_segments == 1) | |
1478 | req->biotail->bi_hw_back_size = len; | |
1479 | } | |
1480 | return mergeable; | |
1481 | } | |
1482 | ||
1483 | return ll_new_hw_segment(q, req, bio); | |
1484 | } | |
1485 | ||
1486 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, | |
1487 | struct request *next) | |
1488 | { | |
dfa1a553 ND |
1489 | int total_phys_segments; |
1490 | int total_hw_segments; | |
1da177e4 LT |
1491 | |
1492 | /* | |
1493 | * First check if the either of the requests are re-queued | |
1494 | * requests. Can't merge them if they are. | |
1495 | */ | |
1496 | if (req->special || next->special) | |
1497 | return 0; | |
1498 | ||
1499 | /* | |
dfa1a553 | 1500 | * Will it become too large? |
1da177e4 LT |
1501 | */ |
1502 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1503 | return 0; | |
1504 | ||
1505 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1506 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1507 | total_phys_segments--; | |
1508 | ||
1509 | if (total_phys_segments > q->max_phys_segments) | |
1510 | return 0; | |
1511 | ||
1512 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1513 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1514 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1515 | /* | |
1516 | * propagate the combined length to the end of the requests | |
1517 | */ | |
1518 | if (req->nr_hw_segments == 1) | |
1519 | req->bio->bi_hw_front_size = len; | |
1520 | if (next->nr_hw_segments == 1) | |
1521 | next->biotail->bi_hw_back_size = len; | |
1522 | total_hw_segments--; | |
1523 | } | |
1524 | ||
1525 | if (total_hw_segments > q->max_hw_segments) | |
1526 | return 0; | |
1527 | ||
1528 | /* Merge is OK... */ | |
1529 | req->nr_phys_segments = total_phys_segments; | |
1530 | req->nr_hw_segments = total_hw_segments; | |
1531 | return 1; | |
1532 | } | |
1533 | ||
1534 | /* | |
1535 | * "plug" the device if there are no outstanding requests: this will | |
1536 | * force the transfer to start only after we have put all the requests | |
1537 | * on the list. | |
1538 | * | |
1539 | * This is called with interrupts off and no requests on the queue and | |
1540 | * with the queue lock held. | |
1541 | */ | |
1542 | void blk_plug_device(request_queue_t *q) | |
1543 | { | |
1544 | WARN_ON(!irqs_disabled()); | |
1545 | ||
1546 | /* | |
1547 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1548 | * which will restart the queueing | |
1549 | */ | |
7daac490 | 1550 | if (blk_queue_stopped(q)) |
1da177e4 LT |
1551 | return; |
1552 | ||
2056a782 | 1553 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 1554 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
1555 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
1556 | } | |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | EXPORT_SYMBOL(blk_plug_device); | |
1560 | ||
1561 | /* | |
1562 | * remove the queue from the plugged list, if present. called with | |
1563 | * queue lock held and interrupts disabled. | |
1564 | */ | |
1565 | int blk_remove_plug(request_queue_t *q) | |
1566 | { | |
1567 | WARN_ON(!irqs_disabled()); | |
1568 | ||
1569 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1570 | return 0; | |
1571 | ||
1572 | del_timer(&q->unplug_timer); | |
1573 | return 1; | |
1574 | } | |
1575 | ||
1576 | EXPORT_SYMBOL(blk_remove_plug); | |
1577 | ||
1578 | /* | |
1579 | * remove the plug and let it rip.. | |
1580 | */ | |
1581 | void __generic_unplug_device(request_queue_t *q) | |
1582 | { | |
7daac490 | 1583 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
1584 | return; |
1585 | ||
1586 | if (!blk_remove_plug(q)) | |
1587 | return; | |
1588 | ||
22e2c507 | 1589 | q->request_fn(q); |
1da177e4 LT |
1590 | } |
1591 | EXPORT_SYMBOL(__generic_unplug_device); | |
1592 | ||
1593 | /** | |
1594 | * generic_unplug_device - fire a request queue | |
1595 | * @q: The &request_queue_t in question | |
1596 | * | |
1597 | * Description: | |
1598 | * Linux uses plugging to build bigger requests queues before letting | |
1599 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1600 | * is still adding and merging requests on the queue. Once the queue | |
1601 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1602 | * transfers started. | |
1603 | **/ | |
1604 | void generic_unplug_device(request_queue_t *q) | |
1605 | { | |
1606 | spin_lock_irq(q->queue_lock); | |
1607 | __generic_unplug_device(q); | |
1608 | spin_unlock_irq(q->queue_lock); | |
1609 | } | |
1610 | EXPORT_SYMBOL(generic_unplug_device); | |
1611 | ||
1612 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1613 | struct page *page) | |
1614 | { | |
1615 | request_queue_t *q = bdi->unplug_io_data; | |
1616 | ||
1617 | /* | |
1618 | * devices don't necessarily have an ->unplug_fn defined | |
1619 | */ | |
2056a782 JA |
1620 | if (q->unplug_fn) { |
1621 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1622 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1623 | ||
1da177e4 | 1624 | q->unplug_fn(q); |
2056a782 | 1625 | } |
1da177e4 LT |
1626 | } |
1627 | ||
65f27f38 | 1628 | static void blk_unplug_work(struct work_struct *work) |
1da177e4 | 1629 | { |
65f27f38 | 1630 | request_queue_t *q = container_of(work, request_queue_t, unplug_work); |
1da177e4 | 1631 | |
2056a782 JA |
1632 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
1633 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1634 | ||
1da177e4 LT |
1635 | q->unplug_fn(q); |
1636 | } | |
1637 | ||
1638 | static void blk_unplug_timeout(unsigned long data) | |
1639 | { | |
1640 | request_queue_t *q = (request_queue_t *)data; | |
1641 | ||
2056a782 JA |
1642 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
1643 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1644 | ||
1da177e4 LT |
1645 | kblockd_schedule_work(&q->unplug_work); |
1646 | } | |
1647 | ||
1648 | /** | |
1649 | * blk_start_queue - restart a previously stopped queue | |
1650 | * @q: The &request_queue_t in question | |
1651 | * | |
1652 | * Description: | |
1653 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1654 | * the request_fn for the queue if it was in a stopped state when | |
1655 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1656 | **/ | |
1657 | void blk_start_queue(request_queue_t *q) | |
1658 | { | |
a038e253 PBG |
1659 | WARN_ON(!irqs_disabled()); |
1660 | ||
1da177e4 LT |
1661 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1662 | ||
1663 | /* | |
1664 | * one level of recursion is ok and is much faster than kicking | |
1665 | * the unplug handling | |
1666 | */ | |
1667 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1668 | q->request_fn(q); | |
1669 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1670 | } else { | |
1671 | blk_plug_device(q); | |
1672 | kblockd_schedule_work(&q->unplug_work); | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | EXPORT_SYMBOL(blk_start_queue); | |
1677 | ||
1678 | /** | |
1679 | * blk_stop_queue - stop a queue | |
1680 | * @q: The &request_queue_t in question | |
1681 | * | |
1682 | * Description: | |
1683 | * The Linux block layer assumes that a block driver will consume all | |
1684 | * entries on the request queue when the request_fn strategy is called. | |
1685 | * Often this will not happen, because of hardware limitations (queue | |
1686 | * depth settings). If a device driver gets a 'queue full' response, | |
1687 | * or if it simply chooses not to queue more I/O at one point, it can | |
1688 | * call this function to prevent the request_fn from being called until | |
1689 | * the driver has signalled it's ready to go again. This happens by calling | |
1690 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1691 | **/ | |
1692 | void blk_stop_queue(request_queue_t *q) | |
1693 | { | |
1694 | blk_remove_plug(q); | |
1695 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1696 | } | |
1697 | EXPORT_SYMBOL(blk_stop_queue); | |
1698 | ||
1699 | /** | |
1700 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1701 | * @q: the queue | |
1702 | * | |
1703 | * Description: | |
1704 | * The block layer may perform asynchronous callback activity | |
1705 | * on a queue, such as calling the unplug function after a timeout. | |
1706 | * A block device may call blk_sync_queue to ensure that any | |
1707 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 1708 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
1709 | * that its ->make_request_fn will not re-add plugging prior to calling |
1710 | * this function. | |
1711 | * | |
1712 | */ | |
1713 | void blk_sync_queue(struct request_queue *q) | |
1714 | { | |
1715 | del_timer_sync(&q->unplug_timer); | |
1da177e4 LT |
1716 | } |
1717 | EXPORT_SYMBOL(blk_sync_queue); | |
1718 | ||
1719 | /** | |
1720 | * blk_run_queue - run a single device queue | |
1721 | * @q: The queue to run | |
1722 | */ | |
1723 | void blk_run_queue(struct request_queue *q) | |
1724 | { | |
1725 | unsigned long flags; | |
1726 | ||
1727 | spin_lock_irqsave(q->queue_lock, flags); | |
1728 | blk_remove_plug(q); | |
dac07ec1 JA |
1729 | |
1730 | /* | |
1731 | * Only recurse once to avoid overrunning the stack, let the unplug | |
1732 | * handling reinvoke the handler shortly if we already got there. | |
1733 | */ | |
1734 | if (!elv_queue_empty(q)) { | |
1735 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1736 | q->request_fn(q); | |
1737 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1738 | } else { | |
1739 | blk_plug_device(q); | |
1740 | kblockd_schedule_work(&q->unplug_work); | |
1741 | } | |
1742 | } | |
1743 | ||
1da177e4 LT |
1744 | spin_unlock_irqrestore(q->queue_lock, flags); |
1745 | } | |
1746 | EXPORT_SYMBOL(blk_run_queue); | |
1747 | ||
1748 | /** | |
1749 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed | |
a580290c | 1750 | * @kobj: the kobj belonging of the request queue to be released |
1da177e4 LT |
1751 | * |
1752 | * Description: | |
1753 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1754 | * blk_queue_make_request(). It should be called when a request queue is | |
1755 | * being released; typically when a block device is being de-registered. | |
1756 | * Currently, its primary task it to free all the &struct request | |
1757 | * structures that were allocated to the queue and the queue itself. | |
1758 | * | |
1759 | * Caveat: | |
1760 | * Hopefully the low level driver will have finished any | |
1761 | * outstanding requests first... | |
1762 | **/ | |
483f4afc | 1763 | static void blk_release_queue(struct kobject *kobj) |
1da177e4 | 1764 | { |
483f4afc | 1765 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
1da177e4 LT |
1766 | struct request_list *rl = &q->rq; |
1767 | ||
1da177e4 LT |
1768 | blk_sync_queue(q); |
1769 | ||
1770 | if (rl->rq_pool) | |
1771 | mempool_destroy(rl->rq_pool); | |
1772 | ||
1773 | if (q->queue_tags) | |
1774 | __blk_queue_free_tags(q); | |
1775 | ||
6c5c9341 | 1776 | blk_trace_shutdown(q); |
2056a782 | 1777 | |
1da177e4 LT |
1778 | kmem_cache_free(requestq_cachep, q); |
1779 | } | |
1780 | ||
483f4afc AV |
1781 | void blk_put_queue(request_queue_t *q) |
1782 | { | |
1783 | kobject_put(&q->kobj); | |
1784 | } | |
1785 | EXPORT_SYMBOL(blk_put_queue); | |
1786 | ||
1787 | void blk_cleanup_queue(request_queue_t * q) | |
1788 | { | |
1789 | mutex_lock(&q->sysfs_lock); | |
1790 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
1791 | mutex_unlock(&q->sysfs_lock); | |
1792 | ||
1793 | if (q->elevator) | |
1794 | elevator_exit(q->elevator); | |
1795 | ||
1796 | blk_put_queue(q); | |
1797 | } | |
1798 | ||
1da177e4 LT |
1799 | EXPORT_SYMBOL(blk_cleanup_queue); |
1800 | ||
1801 | static int blk_init_free_list(request_queue_t *q) | |
1802 | { | |
1803 | struct request_list *rl = &q->rq; | |
1804 | ||
1805 | rl->count[READ] = rl->count[WRITE] = 0; | |
1806 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 1807 | rl->elvpriv = 0; |
1da177e4 LT |
1808 | init_waitqueue_head(&rl->wait[READ]); |
1809 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 1810 | |
1946089a CL |
1811 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1812 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
1813 | |
1814 | if (!rl->rq_pool) | |
1815 | return -ENOMEM; | |
1816 | ||
1817 | return 0; | |
1818 | } | |
1819 | ||
8267e268 | 1820 | request_queue_t *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 1821 | { |
1946089a CL |
1822 | return blk_alloc_queue_node(gfp_mask, -1); |
1823 | } | |
1824 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 1825 | |
483f4afc AV |
1826 | static struct kobj_type queue_ktype; |
1827 | ||
8267e268 | 1828 | request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a CL |
1829 | { |
1830 | request_queue_t *q; | |
1831 | ||
94f6030c CL |
1832 | q = kmem_cache_alloc_node(requestq_cachep, |
1833 | gfp_mask | __GFP_ZERO, node_id); | |
1da177e4 LT |
1834 | if (!q) |
1835 | return NULL; | |
1836 | ||
1da177e4 | 1837 | init_timer(&q->unplug_timer); |
483f4afc AV |
1838 | |
1839 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); | |
1840 | q->kobj.ktype = &queue_ktype; | |
1841 | kobject_init(&q->kobj); | |
1da177e4 LT |
1842 | |
1843 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | |
1844 | q->backing_dev_info.unplug_io_data = q; | |
1845 | ||
483f4afc AV |
1846 | mutex_init(&q->sysfs_lock); |
1847 | ||
1da177e4 LT |
1848 | return q; |
1849 | } | |
1946089a | 1850 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1851 | |
1852 | /** | |
1853 | * blk_init_queue - prepare a request queue for use with a block device | |
1854 | * @rfn: The function to be called to process requests that have been | |
1855 | * placed on the queue. | |
1856 | * @lock: Request queue spin lock | |
1857 | * | |
1858 | * Description: | |
1859 | * If a block device wishes to use the standard request handling procedures, | |
1860 | * which sorts requests and coalesces adjacent requests, then it must | |
1861 | * call blk_init_queue(). The function @rfn will be called when there | |
1862 | * are requests on the queue that need to be processed. If the device | |
1863 | * supports plugging, then @rfn may not be called immediately when requests | |
1864 | * are available on the queue, but may be called at some time later instead. | |
1865 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1866 | * of the requests on the queue is needed, or due to memory pressure. | |
1867 | * | |
1868 | * @rfn is not required, or even expected, to remove all requests off the | |
1869 | * queue, but only as many as it can handle at a time. If it does leave | |
1870 | * requests on the queue, it is responsible for arranging that the requests | |
1871 | * get dealt with eventually. | |
1872 | * | |
1873 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1874 | * request queue; this lock will be taken also from interrupt context, so irq |
1875 | * disabling is needed for it. | |
1da177e4 LT |
1876 | * |
1877 | * Function returns a pointer to the initialized request queue, or NULL if | |
1878 | * it didn't succeed. | |
1879 | * | |
1880 | * Note: | |
1881 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1882 | * when the block device is deactivated (such as at module unload). | |
1883 | **/ | |
1946089a | 1884 | |
1da177e4 LT |
1885 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1886 | { | |
1946089a CL |
1887 | return blk_init_queue_node(rfn, lock, -1); |
1888 | } | |
1889 | EXPORT_SYMBOL(blk_init_queue); | |
1890 | ||
1891 | request_queue_t * | |
1892 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | |
1893 | { | |
1894 | request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | |
1da177e4 LT |
1895 | |
1896 | if (!q) | |
1897 | return NULL; | |
1898 | ||
1946089a | 1899 | q->node = node_id; |
8669aafd AV |
1900 | if (blk_init_free_list(q)) { |
1901 | kmem_cache_free(requestq_cachep, q); | |
1902 | return NULL; | |
1903 | } | |
1da177e4 | 1904 | |
152587de JA |
1905 | /* |
1906 | * if caller didn't supply a lock, they get per-queue locking with | |
1907 | * our embedded lock | |
1908 | */ | |
1909 | if (!lock) { | |
1910 | spin_lock_init(&q->__queue_lock); | |
1911 | lock = &q->__queue_lock; | |
1912 | } | |
1913 | ||
1da177e4 | 1914 | q->request_fn = rfn; |
1da177e4 LT |
1915 | q->prep_rq_fn = NULL; |
1916 | q->unplug_fn = generic_unplug_device; | |
1917 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1918 | q->queue_lock = lock; | |
1919 | ||
1920 | blk_queue_segment_boundary(q, 0xffffffff); | |
1921 | ||
1922 | blk_queue_make_request(q, __make_request); | |
1923 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1924 | ||
1925 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1926 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1927 | ||
44ec9542 AS |
1928 | q->sg_reserved_size = INT_MAX; |
1929 | ||
1da177e4 LT |
1930 | /* |
1931 | * all done | |
1932 | */ | |
1933 | if (!elevator_init(q, NULL)) { | |
1934 | blk_queue_congestion_threshold(q); | |
1935 | return q; | |
1936 | } | |
1937 | ||
8669aafd | 1938 | blk_put_queue(q); |
1da177e4 LT |
1939 | return NULL; |
1940 | } | |
1946089a | 1941 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 LT |
1942 | |
1943 | int blk_get_queue(request_queue_t *q) | |
1944 | { | |
fde6ad22 | 1945 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 1946 | kobject_get(&q->kobj); |
1da177e4 LT |
1947 | return 0; |
1948 | } | |
1949 | ||
1950 | return 1; | |
1951 | } | |
1952 | ||
1953 | EXPORT_SYMBOL(blk_get_queue); | |
1954 | ||
1955 | static inline void blk_free_request(request_queue_t *q, struct request *rq) | |
1956 | { | |
4aff5e23 | 1957 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 1958 | elv_put_request(q, rq); |
1da177e4 LT |
1959 | mempool_free(rq, q->rq.rq_pool); |
1960 | } | |
1961 | ||
1ea25ecb | 1962 | static struct request * |
cb78b285 | 1963 | blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
1964 | { |
1965 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
1966 | ||
1967 | if (!rq) | |
1968 | return NULL; | |
1969 | ||
1970 | /* | |
4aff5e23 | 1971 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
1972 | * see bio.h and blkdev.h |
1973 | */ | |
49171e5c | 1974 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 1975 | |
cb98fc8b | 1976 | if (priv) { |
cb78b285 | 1977 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
1978 | mempool_free(rq, q->rq.rq_pool); |
1979 | return NULL; | |
1980 | } | |
4aff5e23 | 1981 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 1982 | } |
1da177e4 | 1983 | |
cb98fc8b | 1984 | return rq; |
1da177e4 LT |
1985 | } |
1986 | ||
1987 | /* | |
1988 | * ioc_batching returns true if the ioc is a valid batching request and | |
1989 | * should be given priority access to a request. | |
1990 | */ | |
1991 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) | |
1992 | { | |
1993 | if (!ioc) | |
1994 | return 0; | |
1995 | ||
1996 | /* | |
1997 | * Make sure the process is able to allocate at least 1 request | |
1998 | * even if the batch times out, otherwise we could theoretically | |
1999 | * lose wakeups. | |
2000 | */ | |
2001 | return ioc->nr_batch_requests == q->nr_batching || | |
2002 | (ioc->nr_batch_requests > 0 | |
2003 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
2008 | * will cause the process to be a "batcher" on all queues in the system. This | |
2009 | * is the behaviour we want though - once it gets a wakeup it should be given | |
2010 | * a nice run. | |
2011 | */ | |
93d17d3d | 2012 | static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) |
1da177e4 LT |
2013 | { |
2014 | if (!ioc || ioc_batching(q, ioc)) | |
2015 | return; | |
2016 | ||
2017 | ioc->nr_batch_requests = q->nr_batching; | |
2018 | ioc->last_waited = jiffies; | |
2019 | } | |
2020 | ||
2021 | static void __freed_request(request_queue_t *q, int rw) | |
2022 | { | |
2023 | struct request_list *rl = &q->rq; | |
2024 | ||
2025 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 2026 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
2027 | |
2028 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
2029 | if (waitqueue_active(&rl->wait[rw])) |
2030 | wake_up(&rl->wait[rw]); | |
2031 | ||
2032 | blk_clear_queue_full(q, rw); | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | /* | |
2037 | * A request has just been released. Account for it, update the full and | |
2038 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
2039 | */ | |
cb98fc8b | 2040 | static void freed_request(request_queue_t *q, int rw, int priv) |
1da177e4 LT |
2041 | { |
2042 | struct request_list *rl = &q->rq; | |
2043 | ||
2044 | rl->count[rw]--; | |
cb98fc8b TH |
2045 | if (priv) |
2046 | rl->elvpriv--; | |
1da177e4 LT |
2047 | |
2048 | __freed_request(q, rw); | |
2049 | ||
2050 | if (unlikely(rl->starved[rw ^ 1])) | |
2051 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
2052 | } |
2053 | ||
2054 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
2055 | /* | |
d6344532 NP |
2056 | * Get a free request, queue_lock must be held. |
2057 | * Returns NULL on failure, with queue_lock held. | |
2058 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 2059 | */ |
7749a8d4 JA |
2060 | static struct request *get_request(request_queue_t *q, int rw_flags, |
2061 | struct bio *bio, gfp_t gfp_mask) | |
1da177e4 LT |
2062 | { |
2063 | struct request *rq = NULL; | |
2064 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 2065 | struct io_context *ioc = NULL; |
7749a8d4 | 2066 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
2067 | int may_queue, priv; |
2068 | ||
7749a8d4 | 2069 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
2070 | if (may_queue == ELV_MQUEUE_NO) |
2071 | goto rq_starved; | |
2072 | ||
2073 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
2074 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 2075 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
2076 | /* |
2077 | * The queue will fill after this allocation, so set | |
2078 | * it as full, and mark this process as "batching". | |
2079 | * This process will be allowed to complete a batch of | |
2080 | * requests, others will be blocked. | |
2081 | */ | |
2082 | if (!blk_queue_full(q, rw)) { | |
2083 | ioc_set_batching(q, ioc); | |
2084 | blk_set_queue_full(q, rw); | |
2085 | } else { | |
2086 | if (may_queue != ELV_MQUEUE_MUST | |
2087 | && !ioc_batching(q, ioc)) { | |
2088 | /* | |
2089 | * The queue is full and the allocating | |
2090 | * process is not a "batcher", and not | |
2091 | * exempted by the IO scheduler | |
2092 | */ | |
2093 | goto out; | |
2094 | } | |
2095 | } | |
1da177e4 | 2096 | } |
79e2de4b | 2097 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
2098 | } |
2099 | ||
082cf69e JA |
2100 | /* |
2101 | * Only allow batching queuers to allocate up to 50% over the defined | |
2102 | * limit of requests, otherwise we could have thousands of requests | |
2103 | * allocated with any setting of ->nr_requests | |
2104 | */ | |
fd782a4a | 2105 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 2106 | goto out; |
fd782a4a | 2107 | |
1da177e4 LT |
2108 | rl->count[rw]++; |
2109 | rl->starved[rw] = 0; | |
cb98fc8b | 2110 | |
64521d1a | 2111 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
2112 | if (priv) |
2113 | rl->elvpriv++; | |
2114 | ||
1da177e4 LT |
2115 | spin_unlock_irq(q->queue_lock); |
2116 | ||
7749a8d4 | 2117 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 2118 | if (unlikely(!rq)) { |
1da177e4 LT |
2119 | /* |
2120 | * Allocation failed presumably due to memory. Undo anything | |
2121 | * we might have messed up. | |
2122 | * | |
2123 | * Allocating task should really be put onto the front of the | |
2124 | * wait queue, but this is pretty rare. | |
2125 | */ | |
2126 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 2127 | freed_request(q, rw, priv); |
1da177e4 LT |
2128 | |
2129 | /* | |
2130 | * in the very unlikely event that allocation failed and no | |
2131 | * requests for this direction was pending, mark us starved | |
2132 | * so that freeing of a request in the other direction will | |
2133 | * notice us. another possible fix would be to split the | |
2134 | * rq mempool into READ and WRITE | |
2135 | */ | |
2136 | rq_starved: | |
2137 | if (unlikely(rl->count[rw] == 0)) | |
2138 | rl->starved[rw] = 1; | |
2139 | ||
1da177e4 LT |
2140 | goto out; |
2141 | } | |
2142 | ||
88ee5ef1 JA |
2143 | /* |
2144 | * ioc may be NULL here, and ioc_batching will be false. That's | |
2145 | * OK, if the queue is under the request limit then requests need | |
2146 | * not count toward the nr_batch_requests limit. There will always | |
2147 | * be some limit enforced by BLK_BATCH_TIME. | |
2148 | */ | |
1da177e4 LT |
2149 | if (ioc_batching(q, ioc)) |
2150 | ioc->nr_batch_requests--; | |
2151 | ||
2152 | rq_init(q, rq); | |
2056a782 JA |
2153 | |
2154 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 2155 | out: |
1da177e4 LT |
2156 | return rq; |
2157 | } | |
2158 | ||
2159 | /* | |
2160 | * No available requests for this queue, unplug the device and wait for some | |
2161 | * requests to become available. | |
d6344532 NP |
2162 | * |
2163 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 2164 | */ |
7749a8d4 | 2165 | static struct request *get_request_wait(request_queue_t *q, int rw_flags, |
22e2c507 | 2166 | struct bio *bio) |
1da177e4 | 2167 | { |
7749a8d4 | 2168 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
2169 | struct request *rq; |
2170 | ||
7749a8d4 | 2171 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
2172 | while (!rq) { |
2173 | DEFINE_WAIT(wait); | |
1da177e4 LT |
2174 | struct request_list *rl = &q->rq; |
2175 | ||
2176 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
2177 | TASK_UNINTERRUPTIBLE); | |
2178 | ||
7749a8d4 | 2179 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
2180 | |
2181 | if (!rq) { | |
2182 | struct io_context *ioc; | |
2183 | ||
2056a782 JA |
2184 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
2185 | ||
d6344532 NP |
2186 | __generic_unplug_device(q); |
2187 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
2188 | io_schedule(); |
2189 | ||
2190 | /* | |
2191 | * After sleeping, we become a "batching" process and | |
2192 | * will be able to allocate at least one request, and | |
2193 | * up to a big batch of them for a small period time. | |
2194 | * See ioc_batching, ioc_set_batching | |
2195 | */ | |
b5deef90 | 2196 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 2197 | ioc_set_batching(q, ioc); |
d6344532 NP |
2198 | |
2199 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
2200 | } |
2201 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 2202 | } |
1da177e4 LT |
2203 | |
2204 | return rq; | |
2205 | } | |
2206 | ||
8267e268 | 2207 | struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
2208 | { |
2209 | struct request *rq; | |
2210 | ||
2211 | BUG_ON(rw != READ && rw != WRITE); | |
2212 | ||
d6344532 NP |
2213 | spin_lock_irq(q->queue_lock); |
2214 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 2215 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 2216 | } else { |
22e2c507 | 2217 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
2218 | if (!rq) |
2219 | spin_unlock_irq(q->queue_lock); | |
2220 | } | |
2221 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
2222 | |
2223 | return rq; | |
2224 | } | |
1da177e4 LT |
2225 | EXPORT_SYMBOL(blk_get_request); |
2226 | ||
dc72ef4a JA |
2227 | /** |
2228 | * blk_start_queueing - initiate dispatch of requests to device | |
2229 | * @q: request queue to kick into gear | |
2230 | * | |
2231 | * This is basically a helper to remove the need to know whether a queue | |
2232 | * is plugged or not if someone just wants to initiate dispatch of requests | |
2233 | * for this queue. | |
2234 | * | |
2235 | * The queue lock must be held with interrupts disabled. | |
2236 | */ | |
2237 | void blk_start_queueing(request_queue_t *q) | |
2238 | { | |
2239 | if (!blk_queue_plugged(q)) | |
2240 | q->request_fn(q); | |
2241 | else | |
2242 | __generic_unplug_device(q); | |
2243 | } | |
2244 | EXPORT_SYMBOL(blk_start_queueing); | |
2245 | ||
1da177e4 LT |
2246 | /** |
2247 | * blk_requeue_request - put a request back on queue | |
2248 | * @q: request queue where request should be inserted | |
2249 | * @rq: request to be inserted | |
2250 | * | |
2251 | * Description: | |
2252 | * Drivers often keep queueing requests until the hardware cannot accept | |
2253 | * more, when that condition happens we need to put the request back | |
2254 | * on the queue. Must be called with queue lock held. | |
2255 | */ | |
2256 | void blk_requeue_request(request_queue_t *q, struct request *rq) | |
2257 | { | |
2056a782 JA |
2258 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
2259 | ||
1da177e4 LT |
2260 | if (blk_rq_tagged(rq)) |
2261 | blk_queue_end_tag(q, rq); | |
2262 | ||
2263 | elv_requeue_request(q, rq); | |
2264 | } | |
2265 | ||
2266 | EXPORT_SYMBOL(blk_requeue_request); | |
2267 | ||
2268 | /** | |
2269 | * blk_insert_request - insert a special request in to a request queue | |
2270 | * @q: request queue where request should be inserted | |
2271 | * @rq: request to be inserted | |
2272 | * @at_head: insert request at head or tail of queue | |
2273 | * @data: private data | |
1da177e4 LT |
2274 | * |
2275 | * Description: | |
2276 | * Many block devices need to execute commands asynchronously, so they don't | |
2277 | * block the whole kernel from preemption during request execution. This is | |
2278 | * accomplished normally by inserting aritficial requests tagged as | |
2279 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2280 | * scheduled for actual execution by the request queue. | |
2281 | * | |
2282 | * We have the option of inserting the head or the tail of the queue. | |
2283 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2284 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2285 | * host that is unable to accept a particular command. | |
2286 | */ | |
2287 | void blk_insert_request(request_queue_t *q, struct request *rq, | |
867d1191 | 2288 | int at_head, void *data) |
1da177e4 | 2289 | { |
867d1191 | 2290 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
2291 | unsigned long flags; |
2292 | ||
2293 | /* | |
2294 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2295 | * must not attempt merges on this) and that it acts as a soft | |
2296 | * barrier | |
2297 | */ | |
4aff5e23 JA |
2298 | rq->cmd_type = REQ_TYPE_SPECIAL; |
2299 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
2300 | |
2301 | rq->special = data; | |
2302 | ||
2303 | spin_lock_irqsave(q->queue_lock, flags); | |
2304 | ||
2305 | /* | |
2306 | * If command is tagged, release the tag | |
2307 | */ | |
867d1191 TH |
2308 | if (blk_rq_tagged(rq)) |
2309 | blk_queue_end_tag(q, rq); | |
1da177e4 | 2310 | |
867d1191 TH |
2311 | drive_stat_acct(rq, rq->nr_sectors, 1); |
2312 | __elv_add_request(q, rq, where, 0); | |
dc72ef4a | 2313 | blk_start_queueing(q); |
1da177e4 LT |
2314 | spin_unlock_irqrestore(q->queue_lock, flags); |
2315 | } | |
2316 | ||
2317 | EXPORT_SYMBOL(blk_insert_request); | |
2318 | ||
0e75f906 MC |
2319 | static int __blk_rq_unmap_user(struct bio *bio) |
2320 | { | |
2321 | int ret = 0; | |
2322 | ||
2323 | if (bio) { | |
2324 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2325 | bio_unmap_user(bio); | |
2326 | else | |
2327 | ret = bio_uncopy_user(bio); | |
2328 | } | |
2329 | ||
2330 | return ret; | |
2331 | } | |
2332 | ||
2333 | static int __blk_rq_map_user(request_queue_t *q, struct request *rq, | |
2334 | void __user *ubuf, unsigned int len) | |
2335 | { | |
2336 | unsigned long uaddr; | |
2337 | struct bio *bio, *orig_bio; | |
2338 | int reading, ret; | |
2339 | ||
2340 | reading = rq_data_dir(rq) == READ; | |
2341 | ||
2342 | /* | |
2343 | * if alignment requirement is satisfied, map in user pages for | |
2344 | * direct dma. else, set up kernel bounce buffers | |
2345 | */ | |
2346 | uaddr = (unsigned long) ubuf; | |
2347 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
2348 | bio = bio_map_user(q, NULL, uaddr, len, reading); | |
2349 | else | |
2350 | bio = bio_copy_user(q, uaddr, len, reading); | |
2351 | ||
2985259b | 2352 | if (IS_ERR(bio)) |
0e75f906 | 2353 | return PTR_ERR(bio); |
0e75f906 MC |
2354 | |
2355 | orig_bio = bio; | |
2356 | blk_queue_bounce(q, &bio); | |
2985259b | 2357 | |
0e75f906 MC |
2358 | /* |
2359 | * We link the bounce buffer in and could have to traverse it | |
2360 | * later so we have to get a ref to prevent it from being freed | |
2361 | */ | |
2362 | bio_get(bio); | |
2363 | ||
0e75f906 MC |
2364 | if (!rq->bio) |
2365 | blk_rq_bio_prep(q, rq, bio); | |
1aa4f24f | 2366 | else if (!ll_back_merge_fn(q, rq, bio)) { |
0e75f906 | 2367 | ret = -EINVAL; |
0e75f906 MC |
2368 | goto unmap_bio; |
2369 | } else { | |
2370 | rq->biotail->bi_next = bio; | |
2371 | rq->biotail = bio; | |
2372 | ||
0e75f906 MC |
2373 | rq->data_len += bio->bi_size; |
2374 | } | |
0e75f906 MC |
2375 | |
2376 | return bio->bi_size; | |
2377 | ||
2378 | unmap_bio: | |
2379 | /* if it was boucned we must call the end io function */ | |
2380 | bio_endio(bio, bio->bi_size, 0); | |
2381 | __blk_rq_unmap_user(orig_bio); | |
2382 | bio_put(bio); | |
2383 | return ret; | |
2384 | } | |
2385 | ||
1da177e4 LT |
2386 | /** |
2387 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2388 | * @q: request queue where request should be inserted | |
73747aed | 2389 | * @rq: request structure to fill |
1da177e4 LT |
2390 | * @ubuf: the user buffer |
2391 | * @len: length of user data | |
2392 | * | |
2393 | * Description: | |
2394 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2395 | * a kernel bounce buffer is used. | |
2396 | * | |
2397 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2398 | * still in process context. | |
2399 | * | |
2400 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2401 | * before being submitted to the device, as pages mapped may be out of | |
2402 | * reach. It's the callers responsibility to make sure this happens. The | |
2403 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2404 | * unmapping. | |
2405 | */ | |
dd1cab95 | 2406 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, |
0e75f906 | 2407 | unsigned long len) |
1da177e4 | 2408 | { |
0e75f906 | 2409 | unsigned long bytes_read = 0; |
8e5cfc45 | 2410 | struct bio *bio = NULL; |
0e75f906 | 2411 | int ret; |
1da177e4 | 2412 | |
defd94b7 | 2413 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2414 | return -EINVAL; |
2415 | if (!len || !ubuf) | |
2416 | return -EINVAL; | |
1da177e4 | 2417 | |
0e75f906 MC |
2418 | while (bytes_read != len) { |
2419 | unsigned long map_len, end, start; | |
1da177e4 | 2420 | |
0e75f906 MC |
2421 | map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE); |
2422 | end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1) | |
2423 | >> PAGE_SHIFT; | |
2424 | start = (unsigned long)ubuf >> PAGE_SHIFT; | |
1da177e4 | 2425 | |
0e75f906 MC |
2426 | /* |
2427 | * A bad offset could cause us to require BIO_MAX_PAGES + 1 | |
2428 | * pages. If this happens we just lower the requested | |
2429 | * mapping len by a page so that we can fit | |
2430 | */ | |
2431 | if (end - start > BIO_MAX_PAGES) | |
2432 | map_len -= PAGE_SIZE; | |
1da177e4 | 2433 | |
0e75f906 MC |
2434 | ret = __blk_rq_map_user(q, rq, ubuf, map_len); |
2435 | if (ret < 0) | |
2436 | goto unmap_rq; | |
8e5cfc45 JA |
2437 | if (!bio) |
2438 | bio = rq->bio; | |
0e75f906 MC |
2439 | bytes_read += ret; |
2440 | ubuf += ret; | |
1da177e4 LT |
2441 | } |
2442 | ||
0e75f906 MC |
2443 | rq->buffer = rq->data = NULL; |
2444 | return 0; | |
2445 | unmap_rq: | |
8e5cfc45 | 2446 | blk_rq_unmap_user(bio); |
0e75f906 | 2447 | return ret; |
1da177e4 LT |
2448 | } |
2449 | ||
2450 | EXPORT_SYMBOL(blk_rq_map_user); | |
2451 | ||
f1970baf JB |
2452 | /** |
2453 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2454 | * @q: request queue where request should be inserted | |
2455 | * @rq: request to map data to | |
2456 | * @iov: pointer to the iovec | |
2457 | * @iov_count: number of elements in the iovec | |
af9997e4 | 2458 | * @len: I/O byte count |
f1970baf JB |
2459 | * |
2460 | * Description: | |
2461 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2462 | * a kernel bounce buffer is used. | |
2463 | * | |
2464 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2465 | * still in process context. | |
2466 | * | |
2467 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2468 | * before being submitted to the device, as pages mapped may be out of | |
2469 | * reach. It's the callers responsibility to make sure this happens. The | |
2470 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2471 | * unmapping. | |
2472 | */ | |
2473 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, | |
0e75f906 | 2474 | struct sg_iovec *iov, int iov_count, unsigned int len) |
f1970baf JB |
2475 | { |
2476 | struct bio *bio; | |
2477 | ||
2478 | if (!iov || iov_count <= 0) | |
2479 | return -EINVAL; | |
2480 | ||
2481 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2482 | * user is using sg, they're expected to know the alignment constraints | |
2483 | * and respect them accordingly */ | |
2484 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2485 | if (IS_ERR(bio)) | |
2486 | return PTR_ERR(bio); | |
2487 | ||
0e75f906 MC |
2488 | if (bio->bi_size != len) { |
2489 | bio_endio(bio, bio->bi_size, 0); | |
2490 | bio_unmap_user(bio); | |
2491 | return -EINVAL; | |
2492 | } | |
2493 | ||
2494 | bio_get(bio); | |
f1970baf JB |
2495 | blk_rq_bio_prep(q, rq, bio); |
2496 | rq->buffer = rq->data = NULL; | |
f1970baf JB |
2497 | return 0; |
2498 | } | |
2499 | ||
2500 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2501 | ||
1da177e4 LT |
2502 | /** |
2503 | * blk_rq_unmap_user - unmap a request with user data | |
8e5cfc45 | 2504 | * @bio: start of bio list |
1da177e4 LT |
2505 | * |
2506 | * Description: | |
8e5cfc45 JA |
2507 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must |
2508 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
2509 | * the io completion may have changed rq->bio. | |
1da177e4 | 2510 | */ |
8e5cfc45 | 2511 | int blk_rq_unmap_user(struct bio *bio) |
1da177e4 | 2512 | { |
8e5cfc45 | 2513 | struct bio *mapped_bio; |
48785bb9 | 2514 | int ret = 0, ret2; |
1da177e4 | 2515 | |
8e5cfc45 JA |
2516 | while (bio) { |
2517 | mapped_bio = bio; | |
2518 | if (unlikely(bio_flagged(bio, BIO_BOUNCED))) | |
0e75f906 | 2519 | mapped_bio = bio->bi_private; |
1da177e4 | 2520 | |
48785bb9 JA |
2521 | ret2 = __blk_rq_unmap_user(mapped_bio); |
2522 | if (ret2 && !ret) | |
2523 | ret = ret2; | |
2524 | ||
8e5cfc45 JA |
2525 | mapped_bio = bio; |
2526 | bio = bio->bi_next; | |
2527 | bio_put(mapped_bio); | |
0e75f906 | 2528 | } |
48785bb9 JA |
2529 | |
2530 | return ret; | |
1da177e4 LT |
2531 | } |
2532 | ||
2533 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2534 | ||
df46b9a4 MC |
2535 | /** |
2536 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2537 | * @q: request queue where request should be inserted | |
73747aed | 2538 | * @rq: request to fill |
df46b9a4 MC |
2539 | * @kbuf: the kernel buffer |
2540 | * @len: length of user data | |
73747aed | 2541 | * @gfp_mask: memory allocation flags |
df46b9a4 | 2542 | */ |
dd1cab95 | 2543 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, |
8267e268 | 2544 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 | 2545 | { |
df46b9a4 MC |
2546 | struct bio *bio; |
2547 | ||
defd94b7 | 2548 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2549 | return -EINVAL; |
2550 | if (!len || !kbuf) | |
2551 | return -EINVAL; | |
df46b9a4 MC |
2552 | |
2553 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
dd1cab95 JA |
2554 | if (IS_ERR(bio)) |
2555 | return PTR_ERR(bio); | |
df46b9a4 | 2556 | |
dd1cab95 JA |
2557 | if (rq_data_dir(rq) == WRITE) |
2558 | bio->bi_rw |= (1 << BIO_RW); | |
df46b9a4 | 2559 | |
dd1cab95 | 2560 | blk_rq_bio_prep(q, rq, bio); |
821de3a2 | 2561 | blk_queue_bounce(q, &rq->bio); |
dd1cab95 | 2562 | rq->buffer = rq->data = NULL; |
dd1cab95 | 2563 | return 0; |
df46b9a4 MC |
2564 | } |
2565 | ||
2566 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2567 | ||
73747aed CH |
2568 | /** |
2569 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2570 | * @q: queue to insert the request in | |
2571 | * @bd_disk: matching gendisk | |
2572 | * @rq: request to insert | |
2573 | * @at_head: insert request at head or tail of queue | |
2574 | * @done: I/O completion handler | |
2575 | * | |
2576 | * Description: | |
2577 | * Insert a fully prepared request at the back of the io scheduler queue | |
2578 | * for execution. Don't wait for completion. | |
2579 | */ | |
f1970baf JB |
2580 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, |
2581 | struct request *rq, int at_head, | |
8ffdc655 | 2582 | rq_end_io_fn *done) |
f1970baf JB |
2583 | { |
2584 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2585 | ||
2586 | rq->rq_disk = bd_disk; | |
4aff5e23 | 2587 | rq->cmd_flags |= REQ_NOMERGE; |
f1970baf | 2588 | rq->end_io = done; |
4c5d0bbd AM |
2589 | WARN_ON(irqs_disabled()); |
2590 | spin_lock_irq(q->queue_lock); | |
2591 | __elv_add_request(q, rq, where, 1); | |
2592 | __generic_unplug_device(q); | |
2593 | spin_unlock_irq(q->queue_lock); | |
f1970baf | 2594 | } |
6e39b69e MC |
2595 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
2596 | ||
1da177e4 LT |
2597 | /** |
2598 | * blk_execute_rq - insert a request into queue for execution | |
2599 | * @q: queue to insert the request in | |
2600 | * @bd_disk: matching gendisk | |
2601 | * @rq: request to insert | |
994ca9a1 | 2602 | * @at_head: insert request at head or tail of queue |
1da177e4 LT |
2603 | * |
2604 | * Description: | |
2605 | * Insert a fully prepared request at the back of the io scheduler queue | |
73747aed | 2606 | * for execution and wait for completion. |
1da177e4 LT |
2607 | */ |
2608 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, | |
994ca9a1 | 2609 | struct request *rq, int at_head) |
1da177e4 | 2610 | { |
60be6b9a | 2611 | DECLARE_COMPLETION_ONSTACK(wait); |
1da177e4 LT |
2612 | char sense[SCSI_SENSE_BUFFERSIZE]; |
2613 | int err = 0; | |
2614 | ||
1da177e4 LT |
2615 | /* |
2616 | * we need an extra reference to the request, so we can look at | |
2617 | * it after io completion | |
2618 | */ | |
2619 | rq->ref_count++; | |
2620 | ||
2621 | if (!rq->sense) { | |
2622 | memset(sense, 0, sizeof(sense)); | |
2623 | rq->sense = sense; | |
2624 | rq->sense_len = 0; | |
2625 | } | |
2626 | ||
c00895ab | 2627 | rq->end_io_data = &wait; |
994ca9a1 | 2628 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
1da177e4 | 2629 | wait_for_completion(&wait); |
1da177e4 LT |
2630 | |
2631 | if (rq->errors) | |
2632 | err = -EIO; | |
2633 | ||
2634 | return err; | |
2635 | } | |
2636 | ||
2637 | EXPORT_SYMBOL(blk_execute_rq); | |
2638 | ||
2639 | /** | |
2640 | * blkdev_issue_flush - queue a flush | |
2641 | * @bdev: blockdev to issue flush for | |
2642 | * @error_sector: error sector | |
2643 | * | |
2644 | * Description: | |
2645 | * Issue a flush for the block device in question. Caller can supply | |
2646 | * room for storing the error offset in case of a flush error, if they | |
2647 | * wish to. Caller must run wait_for_completion() on its own. | |
2648 | */ | |
2649 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2650 | { | |
2651 | request_queue_t *q; | |
2652 | ||
2653 | if (bdev->bd_disk == NULL) | |
2654 | return -ENXIO; | |
2655 | ||
2656 | q = bdev_get_queue(bdev); | |
2657 | if (!q) | |
2658 | return -ENXIO; | |
2659 | if (!q->issue_flush_fn) | |
2660 | return -EOPNOTSUPP; | |
2661 | ||
2662 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); | |
2663 | } | |
2664 | ||
2665 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2666 | ||
93d17d3d | 2667 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
1da177e4 LT |
2668 | { |
2669 | int rw = rq_data_dir(rq); | |
2670 | ||
2671 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2672 | return; | |
2673 | ||
d72d904a | 2674 | if (!new_io) { |
a362357b | 2675 | __disk_stat_inc(rq->rq_disk, merges[rw]); |
d72d904a | 2676 | } else { |
1da177e4 LT |
2677 | disk_round_stats(rq->rq_disk); |
2678 | rq->rq_disk->in_flight++; | |
2679 | } | |
2680 | } | |
2681 | ||
2682 | /* | |
2683 | * add-request adds a request to the linked list. | |
2684 | * queue lock is held and interrupts disabled, as we muck with the | |
2685 | * request queue list. | |
2686 | */ | |
2687 | static inline void add_request(request_queue_t * q, struct request * req) | |
2688 | { | |
2689 | drive_stat_acct(req, req->nr_sectors, 1); | |
2690 | ||
1da177e4 LT |
2691 | /* |
2692 | * elevator indicated where it wants this request to be | |
2693 | * inserted at elevator_merge time | |
2694 | */ | |
2695 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2696 | } | |
2697 | ||
2698 | /* | |
2699 | * disk_round_stats() - Round off the performance stats on a struct | |
2700 | * disk_stats. | |
2701 | * | |
2702 | * The average IO queue length and utilisation statistics are maintained | |
2703 | * by observing the current state of the queue length and the amount of | |
2704 | * time it has been in this state for. | |
2705 | * | |
2706 | * Normally, that accounting is done on IO completion, but that can result | |
2707 | * in more than a second's worth of IO being accounted for within any one | |
2708 | * second, leading to >100% utilisation. To deal with that, we call this | |
2709 | * function to do a round-off before returning the results when reading | |
2710 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2711 | * the current jiffies and restarts the counters again. | |
2712 | */ | |
2713 | void disk_round_stats(struct gendisk *disk) | |
2714 | { | |
2715 | unsigned long now = jiffies; | |
2716 | ||
b2982649 KC |
2717 | if (now == disk->stamp) |
2718 | return; | |
1da177e4 | 2719 | |
20e5c81f KC |
2720 | if (disk->in_flight) { |
2721 | __disk_stat_add(disk, time_in_queue, | |
2722 | disk->in_flight * (now - disk->stamp)); | |
2723 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
2724 | } | |
1da177e4 | 2725 | disk->stamp = now; |
1da177e4 LT |
2726 | } |
2727 | ||
3eaf840e JNN |
2728 | EXPORT_SYMBOL_GPL(disk_round_stats); |
2729 | ||
1da177e4 LT |
2730 | /* |
2731 | * queue lock must be held | |
2732 | */ | |
6e39b69e | 2733 | void __blk_put_request(request_queue_t *q, struct request *req) |
1da177e4 | 2734 | { |
1da177e4 LT |
2735 | if (unlikely(!q)) |
2736 | return; | |
2737 | if (unlikely(--req->ref_count)) | |
2738 | return; | |
2739 | ||
8922e16c TH |
2740 | elv_completed_request(q, req); |
2741 | ||
1da177e4 LT |
2742 | /* |
2743 | * Request may not have originated from ll_rw_blk. if not, | |
2744 | * it didn't come out of our reserved rq pools | |
2745 | */ | |
49171e5c | 2746 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 2747 | int rw = rq_data_dir(req); |
4aff5e23 | 2748 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 2749 | |
1da177e4 | 2750 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 2751 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
2752 | |
2753 | blk_free_request(q, req); | |
cb98fc8b | 2754 | freed_request(q, rw, priv); |
1da177e4 LT |
2755 | } |
2756 | } | |
2757 | ||
6e39b69e MC |
2758 | EXPORT_SYMBOL_GPL(__blk_put_request); |
2759 | ||
1da177e4 LT |
2760 | void blk_put_request(struct request *req) |
2761 | { | |
8922e16c TH |
2762 | unsigned long flags; |
2763 | request_queue_t *q = req->q; | |
2764 | ||
1da177e4 | 2765 | /* |
8922e16c TH |
2766 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
2767 | * following if (q) test. | |
1da177e4 | 2768 | */ |
8922e16c | 2769 | if (q) { |
1da177e4 LT |
2770 | spin_lock_irqsave(q->queue_lock, flags); |
2771 | __blk_put_request(q, req); | |
2772 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2773 | } | |
2774 | } | |
2775 | ||
2776 | EXPORT_SYMBOL(blk_put_request); | |
2777 | ||
2778 | /** | |
2779 | * blk_end_sync_rq - executes a completion event on a request | |
2780 | * @rq: request to complete | |
fddfdeaf | 2781 | * @error: end io status of the request |
1da177e4 | 2782 | */ |
8ffdc655 | 2783 | void blk_end_sync_rq(struct request *rq, int error) |
1da177e4 | 2784 | { |
c00895ab | 2785 | struct completion *waiting = rq->end_io_data; |
1da177e4 | 2786 | |
c00895ab | 2787 | rq->end_io_data = NULL; |
1da177e4 LT |
2788 | __blk_put_request(rq->q, rq); |
2789 | ||
2790 | /* | |
2791 | * complete last, if this is a stack request the process (and thus | |
2792 | * the rq pointer) could be invalid right after this complete() | |
2793 | */ | |
2794 | complete(waiting); | |
2795 | } | |
2796 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2797 | ||
1da177e4 LT |
2798 | /* |
2799 | * Has to be called with the request spinlock acquired | |
2800 | */ | |
2801 | static int attempt_merge(request_queue_t *q, struct request *req, | |
2802 | struct request *next) | |
2803 | { | |
2804 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2805 | return 0; | |
2806 | ||
2807 | /* | |
d6e05edc | 2808 | * not contiguous |
1da177e4 LT |
2809 | */ |
2810 | if (req->sector + req->nr_sectors != next->sector) | |
2811 | return 0; | |
2812 | ||
2813 | if (rq_data_dir(req) != rq_data_dir(next) | |
2814 | || req->rq_disk != next->rq_disk | |
c00895ab | 2815 | || next->special) |
1da177e4 LT |
2816 | return 0; |
2817 | ||
2818 | /* | |
2819 | * If we are allowed to merge, then append bio list | |
2820 | * from next to rq and release next. merge_requests_fn | |
2821 | * will have updated segment counts, update sector | |
2822 | * counts here. | |
2823 | */ | |
1aa4f24f | 2824 | if (!ll_merge_requests_fn(q, req, next)) |
1da177e4 LT |
2825 | return 0; |
2826 | ||
2827 | /* | |
2828 | * At this point we have either done a back merge | |
2829 | * or front merge. We need the smaller start_time of | |
2830 | * the merged requests to be the current request | |
2831 | * for accounting purposes. | |
2832 | */ | |
2833 | if (time_after(req->start_time, next->start_time)) | |
2834 | req->start_time = next->start_time; | |
2835 | ||
2836 | req->biotail->bi_next = next->bio; | |
2837 | req->biotail = next->biotail; | |
2838 | ||
2839 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2840 | ||
2841 | elv_merge_requests(q, req, next); | |
2842 | ||
2843 | if (req->rq_disk) { | |
2844 | disk_round_stats(req->rq_disk); | |
2845 | req->rq_disk->in_flight--; | |
2846 | } | |
2847 | ||
22e2c507 JA |
2848 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2849 | ||
1da177e4 LT |
2850 | __blk_put_request(q, next); |
2851 | return 1; | |
2852 | } | |
2853 | ||
2854 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) | |
2855 | { | |
2856 | struct request *next = elv_latter_request(q, rq); | |
2857 | ||
2858 | if (next) | |
2859 | return attempt_merge(q, rq, next); | |
2860 | ||
2861 | return 0; | |
2862 | } | |
2863 | ||
2864 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) | |
2865 | { | |
2866 | struct request *prev = elv_former_request(q, rq); | |
2867 | ||
2868 | if (prev) | |
2869 | return attempt_merge(q, prev, rq); | |
2870 | ||
2871 | return 0; | |
2872 | } | |
2873 | ||
52d9e675 TH |
2874 | static void init_request_from_bio(struct request *req, struct bio *bio) |
2875 | { | |
4aff5e23 | 2876 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
2877 | |
2878 | /* | |
2879 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2880 | */ | |
2881 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 2882 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
2883 | |
2884 | /* | |
2885 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2886 | */ | |
2887 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 2888 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 2889 | |
b31dc66a | 2890 | if (bio_sync(bio)) |
4aff5e23 | 2891 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
2892 | if (bio_rw_meta(bio)) |
2893 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 2894 | |
52d9e675 TH |
2895 | req->errors = 0; |
2896 | req->hard_sector = req->sector = bio->bi_sector; | |
2897 | req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio); | |
2898 | req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio); | |
2899 | req->nr_phys_segments = bio_phys_segments(req->q, bio); | |
2900 | req->nr_hw_segments = bio_hw_segments(req->q, bio); | |
2901 | req->buffer = bio_data(bio); /* see ->buffer comment above */ | |
52d9e675 TH |
2902 | req->bio = req->biotail = bio; |
2903 | req->ioprio = bio_prio(bio); | |
2904 | req->rq_disk = bio->bi_bdev->bd_disk; | |
2905 | req->start_time = jiffies; | |
2906 | } | |
2907 | ||
1da177e4 LT |
2908 | static int __make_request(request_queue_t *q, struct bio *bio) |
2909 | { | |
450991bc | 2910 | struct request *req; |
51da90fc JA |
2911 | int el_ret, nr_sectors, barrier, err; |
2912 | const unsigned short prio = bio_prio(bio); | |
2913 | const int sync = bio_sync(bio); | |
7749a8d4 | 2914 | int rw_flags; |
1da177e4 | 2915 | |
1da177e4 | 2916 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
2917 | |
2918 | /* | |
2919 | * low level driver can indicate that it wants pages above a | |
2920 | * certain limit bounced to low memory (ie for highmem, or even | |
2921 | * ISA dma in theory) | |
2922 | */ | |
2923 | blk_queue_bounce(q, &bio); | |
2924 | ||
1da177e4 | 2925 | barrier = bio_barrier(bio); |
797e7dbb | 2926 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
2927 | err = -EOPNOTSUPP; |
2928 | goto end_io; | |
2929 | } | |
2930 | ||
1da177e4 LT |
2931 | spin_lock_irq(q->queue_lock); |
2932 | ||
450991bc | 2933 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
2934 | goto get_rq; |
2935 | ||
2936 | el_ret = elv_merge(q, &req, bio); | |
2937 | switch (el_ret) { | |
2938 | case ELEVATOR_BACK_MERGE: | |
2939 | BUG_ON(!rq_mergeable(req)); | |
2940 | ||
1aa4f24f | 2941 | if (!ll_back_merge_fn(q, req, bio)) |
1da177e4 LT |
2942 | break; |
2943 | ||
2056a782 JA |
2944 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
2945 | ||
1da177e4 LT |
2946 | req->biotail->bi_next = bio; |
2947 | req->biotail = bio; | |
2948 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 2949 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2950 | drive_stat_acct(req, nr_sectors, 0); |
2951 | if (!attempt_back_merge(q, req)) | |
2e662b65 | 2952 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2953 | goto out; |
2954 | ||
2955 | case ELEVATOR_FRONT_MERGE: | |
2956 | BUG_ON(!rq_mergeable(req)); | |
2957 | ||
1aa4f24f | 2958 | if (!ll_front_merge_fn(q, req, bio)) |
1da177e4 LT |
2959 | break; |
2960 | ||
2056a782 JA |
2961 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
2962 | ||
1da177e4 LT |
2963 | bio->bi_next = req->bio; |
2964 | req->bio = bio; | |
2965 | ||
2966 | /* | |
2967 | * may not be valid. if the low level driver said | |
2968 | * it didn't need a bounce buffer then it better | |
2969 | * not touch req->buffer either... | |
2970 | */ | |
2971 | req->buffer = bio_data(bio); | |
51da90fc JA |
2972 | req->current_nr_sectors = bio_cur_sectors(bio); |
2973 | req->hard_cur_sectors = req->current_nr_sectors; | |
2974 | req->sector = req->hard_sector = bio->bi_sector; | |
1da177e4 | 2975 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
22e2c507 | 2976 | req->ioprio = ioprio_best(req->ioprio, prio); |
1da177e4 LT |
2977 | drive_stat_acct(req, nr_sectors, 0); |
2978 | if (!attempt_front_merge(q, req)) | |
2e662b65 | 2979 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
2980 | goto out; |
2981 | ||
450991bc | 2982 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
1da177e4 | 2983 | default: |
450991bc | 2984 | ; |
1da177e4 LT |
2985 | } |
2986 | ||
450991bc | 2987 | get_rq: |
7749a8d4 JA |
2988 | /* |
2989 | * This sync check and mask will be re-done in init_request_from_bio(), | |
2990 | * but we need to set it earlier to expose the sync flag to the | |
2991 | * rq allocator and io schedulers. | |
2992 | */ | |
2993 | rw_flags = bio_data_dir(bio); | |
2994 | if (sync) | |
2995 | rw_flags |= REQ_RW_SYNC; | |
2996 | ||
1da177e4 | 2997 | /* |
450991bc | 2998 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 2999 | * Returns with the queue unlocked. |
450991bc | 3000 | */ |
7749a8d4 | 3001 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 3002 | |
450991bc NP |
3003 | /* |
3004 | * After dropping the lock and possibly sleeping here, our request | |
3005 | * may now be mergeable after it had proven unmergeable (above). | |
3006 | * We don't worry about that case for efficiency. It won't happen | |
3007 | * often, and the elevators are able to handle it. | |
1da177e4 | 3008 | */ |
52d9e675 | 3009 | init_request_from_bio(req, bio); |
1da177e4 | 3010 | |
450991bc NP |
3011 | spin_lock_irq(q->queue_lock); |
3012 | if (elv_queue_empty(q)) | |
3013 | blk_plug_device(q); | |
1da177e4 LT |
3014 | add_request(q, req); |
3015 | out: | |
4a534f93 | 3016 | if (sync) |
1da177e4 LT |
3017 | __generic_unplug_device(q); |
3018 | ||
3019 | spin_unlock_irq(q->queue_lock); | |
3020 | return 0; | |
3021 | ||
3022 | end_io: | |
3023 | bio_endio(bio, nr_sectors << 9, err); | |
3024 | return 0; | |
3025 | } | |
3026 | ||
3027 | /* | |
3028 | * If bio->bi_dev is a partition, remap the location | |
3029 | */ | |
3030 | static inline void blk_partition_remap(struct bio *bio) | |
3031 | { | |
3032 | struct block_device *bdev = bio->bi_bdev; | |
3033 | ||
3034 | if (bdev != bdev->bd_contains) { | |
3035 | struct hd_struct *p = bdev->bd_part; | |
a362357b JA |
3036 | const int rw = bio_data_dir(bio); |
3037 | ||
3038 | p->sectors[rw] += bio_sectors(bio); | |
3039 | p->ios[rw]++; | |
1da177e4 | 3040 | |
1da177e4 LT |
3041 | bio->bi_sector += p->start_sect; |
3042 | bio->bi_bdev = bdev->bd_contains; | |
3043 | } | |
3044 | } | |
3045 | ||
1da177e4 LT |
3046 | static void handle_bad_sector(struct bio *bio) |
3047 | { | |
3048 | char b[BDEVNAME_SIZE]; | |
3049 | ||
3050 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
3051 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
3052 | bdevname(bio->bi_bdev, b), | |
3053 | bio->bi_rw, | |
3054 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
3055 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
3056 | ||
3057 | set_bit(BIO_EOF, &bio->bi_flags); | |
3058 | } | |
3059 | ||
c17bb495 AM |
3060 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
3061 | ||
3062 | static DECLARE_FAULT_ATTR(fail_make_request); | |
3063 | ||
3064 | static int __init setup_fail_make_request(char *str) | |
3065 | { | |
3066 | return setup_fault_attr(&fail_make_request, str); | |
3067 | } | |
3068 | __setup("fail_make_request=", setup_fail_make_request); | |
3069 | ||
3070 | static int should_fail_request(struct bio *bio) | |
3071 | { | |
3072 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
3073 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
3074 | return should_fail(&fail_make_request, bio->bi_size); | |
3075 | ||
3076 | return 0; | |
3077 | } | |
3078 | ||
3079 | static int __init fail_make_request_debugfs(void) | |
3080 | { | |
3081 | return init_fault_attr_dentries(&fail_make_request, | |
3082 | "fail_make_request"); | |
3083 | } | |
3084 | ||
3085 | late_initcall(fail_make_request_debugfs); | |
3086 | ||
3087 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
3088 | ||
3089 | static inline int should_fail_request(struct bio *bio) | |
3090 | { | |
3091 | return 0; | |
3092 | } | |
3093 | ||
3094 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
3095 | ||
1da177e4 LT |
3096 | /** |
3097 | * generic_make_request: hand a buffer to its device driver for I/O | |
3098 | * @bio: The bio describing the location in memory and on the device. | |
3099 | * | |
3100 | * generic_make_request() is used to make I/O requests of block | |
3101 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
3102 | * to be done. | |
3103 | * | |
3104 | * generic_make_request() does not return any status. The | |
3105 | * success/failure status of the request, along with notification of | |
3106 | * completion, is delivered asynchronously through the bio->bi_end_io | |
3107 | * function described (one day) else where. | |
3108 | * | |
3109 | * The caller of generic_make_request must make sure that bi_io_vec | |
3110 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
3111 | * set to describe the device address, and the | |
3112 | * bi_end_io and optionally bi_private are set to describe how | |
3113 | * completion notification should be signaled. | |
3114 | * | |
3115 | * generic_make_request and the drivers it calls may use bi_next if this | |
3116 | * bio happens to be merged with someone else, and may change bi_dev and | |
3117 | * bi_sector for remaps as it sees fit. So the values of these fields | |
3118 | * should NOT be depended on after the call to generic_make_request. | |
3119 | */ | |
d89d8796 | 3120 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 LT |
3121 | { |
3122 | request_queue_t *q; | |
3123 | sector_t maxsector; | |
5ddfe969 | 3124 | sector_t old_sector; |
1da177e4 | 3125 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 3126 | dev_t old_dev; |
1da177e4 LT |
3127 | |
3128 | might_sleep(); | |
3129 | /* Test device or partition size, when known. */ | |
3130 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
3131 | if (maxsector) { | |
3132 | sector_t sector = bio->bi_sector; | |
3133 | ||
3134 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
3135 | /* | |
3136 | * This may well happen - the kernel calls bread() | |
3137 | * without checking the size of the device, e.g., when | |
3138 | * mounting a device. | |
3139 | */ | |
3140 | handle_bad_sector(bio); | |
3141 | goto end_io; | |
3142 | } | |
3143 | } | |
3144 | ||
3145 | /* | |
3146 | * Resolve the mapping until finished. (drivers are | |
3147 | * still free to implement/resolve their own stacking | |
3148 | * by explicitly returning 0) | |
3149 | * | |
3150 | * NOTE: we don't repeat the blk_size check for each new device. | |
3151 | * Stacking drivers are expected to know what they are doing. | |
3152 | */ | |
5ddfe969 | 3153 | old_sector = -1; |
2056a782 | 3154 | old_dev = 0; |
1da177e4 LT |
3155 | do { |
3156 | char b[BDEVNAME_SIZE]; | |
3157 | ||
3158 | q = bdev_get_queue(bio->bi_bdev); | |
3159 | if (!q) { | |
3160 | printk(KERN_ERR | |
3161 | "generic_make_request: Trying to access " | |
3162 | "nonexistent block-device %s (%Lu)\n", | |
3163 | bdevname(bio->bi_bdev, b), | |
3164 | (long long) bio->bi_sector); | |
3165 | end_io: | |
3166 | bio_endio(bio, bio->bi_size, -EIO); | |
3167 | break; | |
3168 | } | |
3169 | ||
3170 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { | |
3171 | printk("bio too big device %s (%u > %u)\n", | |
3172 | bdevname(bio->bi_bdev, b), | |
3173 | bio_sectors(bio), | |
3174 | q->max_hw_sectors); | |
3175 | goto end_io; | |
3176 | } | |
3177 | ||
fde6ad22 | 3178 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
3179 | goto end_io; |
3180 | ||
c17bb495 AM |
3181 | if (should_fail_request(bio)) |
3182 | goto end_io; | |
3183 | ||
1da177e4 LT |
3184 | /* |
3185 | * If this device has partitions, remap block n | |
3186 | * of partition p to block n+start(p) of the disk. | |
3187 | */ | |
3188 | blk_partition_remap(bio); | |
3189 | ||
5ddfe969 | 3190 | if (old_sector != -1) |
2056a782 | 3191 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 3192 | old_sector); |
2056a782 JA |
3193 | |
3194 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
3195 | ||
5ddfe969 | 3196 | old_sector = bio->bi_sector; |
2056a782 JA |
3197 | old_dev = bio->bi_bdev->bd_dev; |
3198 | ||
5ddfe969 N |
3199 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
3200 | if (maxsector) { | |
3201 | sector_t sector = bio->bi_sector; | |
3202 | ||
df66b855 AM |
3203 | if (maxsector < nr_sectors || |
3204 | maxsector - nr_sectors < sector) { | |
5ddfe969 | 3205 | /* |
df66b855 AM |
3206 | * This may well happen - partitions are not |
3207 | * checked to make sure they are within the size | |
3208 | * of the whole device. | |
5ddfe969 N |
3209 | */ |
3210 | handle_bad_sector(bio); | |
3211 | goto end_io; | |
3212 | } | |
3213 | } | |
3214 | ||
1da177e4 LT |
3215 | ret = q->make_request_fn(q, bio); |
3216 | } while (ret); | |
3217 | } | |
3218 | ||
d89d8796 NB |
3219 | /* |
3220 | * We only want one ->make_request_fn to be active at a time, | |
3221 | * else stack usage with stacked devices could be a problem. | |
3222 | * So use current->bio_{list,tail} to keep a list of requests | |
3223 | * submited by a make_request_fn function. | |
3224 | * current->bio_tail is also used as a flag to say if | |
3225 | * generic_make_request is currently active in this task or not. | |
3226 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
3227 | * then a make_request is active, and new requests should be added | |
3228 | * at the tail | |
3229 | */ | |
3230 | void generic_make_request(struct bio *bio) | |
3231 | { | |
3232 | if (current->bio_tail) { | |
3233 | /* make_request is active */ | |
3234 | *(current->bio_tail) = bio; | |
3235 | bio->bi_next = NULL; | |
3236 | current->bio_tail = &bio->bi_next; | |
3237 | return; | |
3238 | } | |
3239 | /* following loop may be a bit non-obvious, and so deserves some | |
3240 | * explanation. | |
3241 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
3242 | * ensure that) so we have a list with a single bio. | |
3243 | * We pretend that we have just taken it off a longer list, so | |
3244 | * we assign bio_list to the next (which is NULL) and bio_tail | |
3245 | * to &bio_list, thus initialising the bio_list of new bios to be | |
3246 | * added. __generic_make_request may indeed add some more bios | |
3247 | * through a recursive call to generic_make_request. If it | |
3248 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
3249 | * from the top. In this case we really did just take the bio | |
3250 | * of the top of the list (no pretending) and so fixup bio_list and | |
3251 | * bio_tail or bi_next, and call into __generic_make_request again. | |
3252 | * | |
3253 | * The loop was structured like this to make only one call to | |
3254 | * __generic_make_request (which is important as it is large and | |
3255 | * inlined) and to keep the structure simple. | |
3256 | */ | |
3257 | BUG_ON(bio->bi_next); | |
3258 | do { | |
3259 | current->bio_list = bio->bi_next; | |
3260 | if (bio->bi_next == NULL) | |
3261 | current->bio_tail = ¤t->bio_list; | |
3262 | else | |
3263 | bio->bi_next = NULL; | |
3264 | __generic_make_request(bio); | |
3265 | bio = current->bio_list; | |
3266 | } while (bio); | |
3267 | current->bio_tail = NULL; /* deactivate */ | |
3268 | } | |
3269 | ||
1da177e4 LT |
3270 | EXPORT_SYMBOL(generic_make_request); |
3271 | ||
3272 | /** | |
3273 | * submit_bio: submit a bio to the block device layer for I/O | |
3274 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3275 | * @bio: The &struct bio which describes the I/O | |
3276 | * | |
3277 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3278 | * uses that function to do most of the work. Both are fairly rough | |
3279 | * interfaces, @bio must be presetup and ready for I/O. | |
3280 | * | |
3281 | */ | |
3282 | void submit_bio(int rw, struct bio *bio) | |
3283 | { | |
3284 | int count = bio_sectors(bio); | |
3285 | ||
3286 | BIO_BUG_ON(!bio->bi_size); | |
3287 | BIO_BUG_ON(!bio->bi_io_vec); | |
22e2c507 | 3288 | bio->bi_rw |= rw; |
faccbd4b | 3289 | if (rw & WRITE) { |
f8891e5e | 3290 | count_vm_events(PGPGOUT, count); |
faccbd4b AM |
3291 | } else { |
3292 | task_io_account_read(bio->bi_size); | |
f8891e5e | 3293 | count_vm_events(PGPGIN, count); |
faccbd4b | 3294 | } |
1da177e4 LT |
3295 | |
3296 | if (unlikely(block_dump)) { | |
3297 | char b[BDEVNAME_SIZE]; | |
3298 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
3299 | current->comm, current->pid, | |
3300 | (rw & WRITE) ? "WRITE" : "READ", | |
3301 | (unsigned long long)bio->bi_sector, | |
3302 | bdevname(bio->bi_bdev,b)); | |
3303 | } | |
3304 | ||
3305 | generic_make_request(bio); | |
3306 | } | |
3307 | ||
3308 | EXPORT_SYMBOL(submit_bio); | |
3309 | ||
93d17d3d | 3310 | static void blk_recalc_rq_segments(struct request *rq) |
1da177e4 LT |
3311 | { |
3312 | struct bio *bio, *prevbio = NULL; | |
3313 | int nr_phys_segs, nr_hw_segs; | |
3314 | unsigned int phys_size, hw_size; | |
3315 | request_queue_t *q = rq->q; | |
3316 | ||
3317 | if (!rq->bio) | |
3318 | return; | |
3319 | ||
3320 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
3321 | rq_for_each_bio(bio, rq) { | |
3322 | /* Force bio hw/phys segs to be recalculated. */ | |
3323 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
3324 | ||
3325 | nr_phys_segs += bio_phys_segments(q, bio); | |
3326 | nr_hw_segs += bio_hw_segments(q, bio); | |
3327 | if (prevbio) { | |
3328 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; | |
3329 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; | |
3330 | ||
3331 | if (blk_phys_contig_segment(q, prevbio, bio) && | |
3332 | pseg <= q->max_segment_size) { | |
3333 | nr_phys_segs--; | |
3334 | phys_size += prevbio->bi_size + bio->bi_size; | |
3335 | } else | |
3336 | phys_size = 0; | |
3337 | ||
3338 | if (blk_hw_contig_segment(q, prevbio, bio) && | |
3339 | hseg <= q->max_segment_size) { | |
3340 | nr_hw_segs--; | |
3341 | hw_size += prevbio->bi_size + bio->bi_size; | |
3342 | } else | |
3343 | hw_size = 0; | |
3344 | } | |
3345 | prevbio = bio; | |
3346 | } | |
3347 | ||
3348 | rq->nr_phys_segments = nr_phys_segs; | |
3349 | rq->nr_hw_segments = nr_hw_segs; | |
3350 | } | |
3351 | ||
93d17d3d | 3352 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
1da177e4 LT |
3353 | { |
3354 | if (blk_fs_request(rq)) { | |
3355 | rq->hard_sector += nsect; | |
3356 | rq->hard_nr_sectors -= nsect; | |
3357 | ||
3358 | /* | |
3359 | * Move the I/O submission pointers ahead if required. | |
3360 | */ | |
3361 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3362 | (rq->sector <= rq->hard_sector)) { | |
3363 | rq->sector = rq->hard_sector; | |
3364 | rq->nr_sectors = rq->hard_nr_sectors; | |
3365 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3366 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3367 | rq->buffer = bio_data(rq->bio); | |
3368 | } | |
3369 | ||
3370 | /* | |
3371 | * if total number of sectors is less than the first segment | |
3372 | * size, something has gone terribly wrong | |
3373 | */ | |
3374 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3375 | printk("blk: request botched\n"); | |
3376 | rq->nr_sectors = rq->current_nr_sectors; | |
3377 | } | |
3378 | } | |
3379 | } | |
3380 | ||
3381 | static int __end_that_request_first(struct request *req, int uptodate, | |
3382 | int nr_bytes) | |
3383 | { | |
3384 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3385 | struct bio *bio; | |
3386 | ||
2056a782 JA |
3387 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
3388 | ||
1da177e4 LT |
3389 | /* |
3390 | * extend uptodate bool to allow < 0 value to be direct io error | |
3391 | */ | |
3392 | error = 0; | |
3393 | if (end_io_error(uptodate)) | |
3394 | error = !uptodate ? -EIO : uptodate; | |
3395 | ||
3396 | /* | |
3397 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3398 | * sense key with us all the way through | |
3399 | */ | |
3400 | if (!blk_pc_request(req)) | |
3401 | req->errors = 0; | |
3402 | ||
3403 | if (!uptodate) { | |
4aff5e23 | 3404 | if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET)) |
1da177e4 LT |
3405 | printk("end_request: I/O error, dev %s, sector %llu\n", |
3406 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3407 | (unsigned long long)req->sector); | |
3408 | } | |
3409 | ||
d72d904a | 3410 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
3411 | const int rw = rq_data_dir(req); |
3412 | ||
53e86061 | 3413 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
d72d904a JA |
3414 | } |
3415 | ||
1da177e4 LT |
3416 | total_bytes = bio_nbytes = 0; |
3417 | while ((bio = req->bio) != NULL) { | |
3418 | int nbytes; | |
3419 | ||
3420 | if (nr_bytes >= bio->bi_size) { | |
3421 | req->bio = bio->bi_next; | |
3422 | nbytes = bio->bi_size; | |
797e7dbb TH |
3423 | if (!ordered_bio_endio(req, bio, nbytes, error)) |
3424 | bio_endio(bio, nbytes, error); | |
1da177e4 LT |
3425 | next_idx = 0; |
3426 | bio_nbytes = 0; | |
3427 | } else { | |
3428 | int idx = bio->bi_idx + next_idx; | |
3429 | ||
3430 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3431 | blk_dump_rq_flags(req, "__end_that"); | |
3432 | printk("%s: bio idx %d >= vcnt %d\n", | |
3433 | __FUNCTION__, | |
3434 | bio->bi_idx, bio->bi_vcnt); | |
3435 | break; | |
3436 | } | |
3437 | ||
3438 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3439 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3440 | ||
3441 | /* | |
3442 | * not a complete bvec done | |
3443 | */ | |
3444 | if (unlikely(nbytes > nr_bytes)) { | |
3445 | bio_nbytes += nr_bytes; | |
3446 | total_bytes += nr_bytes; | |
3447 | break; | |
3448 | } | |
3449 | ||
3450 | /* | |
3451 | * advance to the next vector | |
3452 | */ | |
3453 | next_idx++; | |
3454 | bio_nbytes += nbytes; | |
3455 | } | |
3456 | ||
3457 | total_bytes += nbytes; | |
3458 | nr_bytes -= nbytes; | |
3459 | ||
3460 | if ((bio = req->bio)) { | |
3461 | /* | |
3462 | * end more in this run, or just return 'not-done' | |
3463 | */ | |
3464 | if (unlikely(nr_bytes <= 0)) | |
3465 | break; | |
3466 | } | |
3467 | } | |
3468 | ||
3469 | /* | |
3470 | * completely done | |
3471 | */ | |
3472 | if (!req->bio) | |
3473 | return 0; | |
3474 | ||
3475 | /* | |
3476 | * if the request wasn't completed, update state | |
3477 | */ | |
3478 | if (bio_nbytes) { | |
797e7dbb TH |
3479 | if (!ordered_bio_endio(req, bio, bio_nbytes, error)) |
3480 | bio_endio(bio, bio_nbytes, error); | |
1da177e4 LT |
3481 | bio->bi_idx += next_idx; |
3482 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3483 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3484 | } | |
3485 | ||
3486 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3487 | blk_recalc_rq_segments(req); | |
3488 | return 1; | |
3489 | } | |
3490 | ||
3491 | /** | |
3492 | * end_that_request_first - end I/O on a request | |
3493 | * @req: the request being processed | |
3494 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3495 | * @nr_sectors: number of sectors to end I/O on | |
3496 | * | |
3497 | * Description: | |
3498 | * Ends I/O on a number of sectors attached to @req, and sets it up | |
3499 | * for the next range of segments (if any) in the cluster. | |
3500 | * | |
3501 | * Return: | |
3502 | * 0 - we are done with this request, call end_that_request_last() | |
3503 | * 1 - still buffers pending for this request | |
3504 | **/ | |
3505 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | |
3506 | { | |
3507 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | |
3508 | } | |
3509 | ||
3510 | EXPORT_SYMBOL(end_that_request_first); | |
3511 | ||
3512 | /** | |
3513 | * end_that_request_chunk - end I/O on a request | |
3514 | * @req: the request being processed | |
3515 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3516 | * @nr_bytes: number of bytes to complete | |
3517 | * | |
3518 | * Description: | |
3519 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3520 | * for the next range of segments (if any). Like end_that_request_first(), | |
3521 | * but deals with bytes instead of sectors. | |
3522 | * | |
3523 | * Return: | |
3524 | * 0 - we are done with this request, call end_that_request_last() | |
3525 | * 1 - still buffers pending for this request | |
3526 | **/ | |
3527 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | |
3528 | { | |
3529 | return __end_that_request_first(req, uptodate, nr_bytes); | |
3530 | } | |
3531 | ||
3532 | EXPORT_SYMBOL(end_that_request_chunk); | |
3533 | ||
ff856bad JA |
3534 | /* |
3535 | * splice the completion data to a local structure and hand off to | |
3536 | * process_completion_queue() to complete the requests | |
3537 | */ | |
3538 | static void blk_done_softirq(struct softirq_action *h) | |
3539 | { | |
626ab0e6 | 3540 | struct list_head *cpu_list, local_list; |
ff856bad JA |
3541 | |
3542 | local_irq_disable(); | |
3543 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 3544 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
3545 | local_irq_enable(); |
3546 | ||
3547 | while (!list_empty(&local_list)) { | |
3548 | struct request *rq = list_entry(local_list.next, struct request, donelist); | |
3549 | ||
3550 | list_del_init(&rq->donelist); | |
3551 | rq->q->softirq_done_fn(rq); | |
3552 | } | |
3553 | } | |
3554 | ||
ff856bad JA |
3555 | static int blk_cpu_notify(struct notifier_block *self, unsigned long action, |
3556 | void *hcpu) | |
3557 | { | |
3558 | /* | |
3559 | * If a CPU goes away, splice its entries to the current CPU | |
3560 | * and trigger a run of the softirq | |
3561 | */ | |
8bb78442 | 3562 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
3563 | int cpu = (unsigned long) hcpu; |
3564 | ||
3565 | local_irq_disable(); | |
3566 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
3567 | &__get_cpu_var(blk_cpu_done)); | |
3568 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3569 | local_irq_enable(); | |
3570 | } | |
3571 | ||
3572 | return NOTIFY_OK; | |
3573 | } | |
3574 | ||
3575 | ||
054cc8a2 | 3576 | static struct notifier_block __devinitdata blk_cpu_notifier = { |
ff856bad JA |
3577 | .notifier_call = blk_cpu_notify, |
3578 | }; | |
3579 | ||
ff856bad JA |
3580 | /** |
3581 | * blk_complete_request - end I/O on a request | |
3582 | * @req: the request being processed | |
3583 | * | |
3584 | * Description: | |
3585 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 3586 | * unless the driver actually implements this in its completion callback |
ff856bad JA |
3587 | * through requeueing. Theh actual completion happens out-of-order, |
3588 | * through a softirq handler. The user must have registered a completion | |
3589 | * callback through blk_queue_softirq_done(). | |
3590 | **/ | |
3591 | ||
3592 | void blk_complete_request(struct request *req) | |
3593 | { | |
3594 | struct list_head *cpu_list; | |
3595 | unsigned long flags; | |
3596 | ||
3597 | BUG_ON(!req->q->softirq_done_fn); | |
3598 | ||
3599 | local_irq_save(flags); | |
3600 | ||
3601 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3602 | list_add_tail(&req->donelist, cpu_list); | |
3603 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3604 | ||
3605 | local_irq_restore(flags); | |
3606 | } | |
3607 | ||
3608 | EXPORT_SYMBOL(blk_complete_request); | |
3609 | ||
1da177e4 LT |
3610 | /* |
3611 | * queue lock must be held | |
3612 | */ | |
8ffdc655 | 3613 | void end_that_request_last(struct request *req, int uptodate) |
1da177e4 LT |
3614 | { |
3615 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 TH |
3616 | int error; |
3617 | ||
3618 | /* | |
3619 | * extend uptodate bool to allow < 0 value to be direct io error | |
3620 | */ | |
3621 | error = 0; | |
3622 | if (end_io_error(uptodate)) | |
3623 | error = !uptodate ? -EIO : uptodate; | |
1da177e4 LT |
3624 | |
3625 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3626 | laptop_io_completion(); | |
3627 | ||
fd0ff8aa JA |
3628 | /* |
3629 | * Account IO completion. bar_rq isn't accounted as a normal | |
3630 | * IO on queueing nor completion. Accounting the containing | |
3631 | * request is enough. | |
3632 | */ | |
3633 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 3634 | unsigned long duration = jiffies - req->start_time; |
a362357b JA |
3635 | const int rw = rq_data_dir(req); |
3636 | ||
3637 | __disk_stat_inc(disk, ios[rw]); | |
3638 | __disk_stat_add(disk, ticks[rw], duration); | |
1da177e4 LT |
3639 | disk_round_stats(disk); |
3640 | disk->in_flight--; | |
3641 | } | |
3642 | if (req->end_io) | |
8ffdc655 | 3643 | req->end_io(req, error); |
1da177e4 LT |
3644 | else |
3645 | __blk_put_request(req->q, req); | |
3646 | } | |
3647 | ||
3648 | EXPORT_SYMBOL(end_that_request_last); | |
3649 | ||
3650 | void end_request(struct request *req, int uptodate) | |
3651 | { | |
3652 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { | |
3653 | add_disk_randomness(req->rq_disk); | |
3654 | blkdev_dequeue_request(req); | |
8ffdc655 | 3655 | end_that_request_last(req, uptodate); |
1da177e4 LT |
3656 | } |
3657 | } | |
3658 | ||
3659 | EXPORT_SYMBOL(end_request); | |
3660 | ||
3661 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) | |
3662 | { | |
4aff5e23 JA |
3663 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
3664 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
3665 | |
3666 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3667 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3668 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3669 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3670 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3671 | rq->buffer = bio_data(bio); | |
0e75f906 | 3672 | rq->data_len = bio->bi_size; |
1da177e4 LT |
3673 | |
3674 | rq->bio = rq->biotail = bio; | |
3675 | } | |
3676 | ||
3677 | EXPORT_SYMBOL(blk_rq_bio_prep); | |
3678 | ||
3679 | int kblockd_schedule_work(struct work_struct *work) | |
3680 | { | |
3681 | return queue_work(kblockd_workqueue, work); | |
3682 | } | |
3683 | ||
3684 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3685 | ||
19a75d83 | 3686 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 3687 | { |
28e53bdd | 3688 | cancel_work_sync(work); |
1da177e4 | 3689 | } |
19a75d83 | 3690 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
3691 | |
3692 | int __init blk_dev_init(void) | |
3693 | { | |
ff856bad JA |
3694 | int i; |
3695 | ||
1da177e4 LT |
3696 | kblockd_workqueue = create_workqueue("kblockd"); |
3697 | if (!kblockd_workqueue) | |
3698 | panic("Failed to create kblockd\n"); | |
3699 | ||
3700 | request_cachep = kmem_cache_create("blkdev_requests", | |
3701 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); | |
3702 | ||
3703 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
3704 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); | |
3705 | ||
3706 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
3707 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); | |
3708 | ||
0a945022 | 3709 | for_each_possible_cpu(i) |
ff856bad JA |
3710 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
3711 | ||
3712 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 3713 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 3714 | |
f772b3d9 VT |
3715 | blk_max_low_pfn = max_low_pfn - 1; |
3716 | blk_max_pfn = max_pfn - 1; | |
1da177e4 LT |
3717 | |
3718 | return 0; | |
3719 | } | |
3720 | ||
3721 | /* | |
3722 | * IO Context helper functions | |
3723 | */ | |
3724 | void put_io_context(struct io_context *ioc) | |
3725 | { | |
3726 | if (ioc == NULL) | |
3727 | return; | |
3728 | ||
3729 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
3730 | ||
3731 | if (atomic_dec_and_test(&ioc->refcount)) { | |
e2d74ac0 JA |
3732 | struct cfq_io_context *cic; |
3733 | ||
334e94de | 3734 | rcu_read_lock(); |
1da177e4 LT |
3735 | if (ioc->aic && ioc->aic->dtor) |
3736 | ioc->aic->dtor(ioc->aic); | |
e2d74ac0 | 3737 | if (ioc->cic_root.rb_node != NULL) { |
7143dd4b JA |
3738 | struct rb_node *n = rb_first(&ioc->cic_root); |
3739 | ||
3740 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
e2d74ac0 JA |
3741 | cic->dtor(ioc); |
3742 | } | |
334e94de | 3743 | rcu_read_unlock(); |
1da177e4 LT |
3744 | |
3745 | kmem_cache_free(iocontext_cachep, ioc); | |
3746 | } | |
3747 | } | |
3748 | EXPORT_SYMBOL(put_io_context); | |
3749 | ||
3750 | /* Called by the exitting task */ | |
3751 | void exit_io_context(void) | |
3752 | { | |
1da177e4 | 3753 | struct io_context *ioc; |
e2d74ac0 | 3754 | struct cfq_io_context *cic; |
1da177e4 | 3755 | |
22e2c507 | 3756 | task_lock(current); |
1da177e4 LT |
3757 | ioc = current->io_context; |
3758 | current->io_context = NULL; | |
22e2c507 | 3759 | task_unlock(current); |
1da177e4 | 3760 | |
25034d7a | 3761 | ioc->task = NULL; |
1da177e4 LT |
3762 | if (ioc->aic && ioc->aic->exit) |
3763 | ioc->aic->exit(ioc->aic); | |
e2d74ac0 JA |
3764 | if (ioc->cic_root.rb_node != NULL) { |
3765 | cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); | |
3766 | cic->exit(ioc); | |
3767 | } | |
25034d7a | 3768 | |
1da177e4 LT |
3769 | put_io_context(ioc); |
3770 | } | |
3771 | ||
3772 | /* | |
3773 | * If the current task has no IO context then create one and initialise it. | |
fb3cc432 | 3774 | * Otherwise, return its existing IO context. |
1da177e4 | 3775 | * |
fb3cc432 NP |
3776 | * This returned IO context doesn't have a specifically elevated refcount, |
3777 | * but since the current task itself holds a reference, the context can be | |
3778 | * used in general code, so long as it stays within `current` context. | |
1da177e4 | 3779 | */ |
b5deef90 | 3780 | static struct io_context *current_io_context(gfp_t gfp_flags, int node) |
1da177e4 LT |
3781 | { |
3782 | struct task_struct *tsk = current; | |
1da177e4 LT |
3783 | struct io_context *ret; |
3784 | ||
1da177e4 | 3785 | ret = tsk->io_context; |
fb3cc432 NP |
3786 | if (likely(ret)) |
3787 | return ret; | |
1da177e4 | 3788 | |
b5deef90 | 3789 | ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node); |
1da177e4 LT |
3790 | if (ret) { |
3791 | atomic_set(&ret->refcount, 1); | |
22e2c507 | 3792 | ret->task = current; |
fc46379d | 3793 | ret->ioprio_changed = 0; |
1da177e4 LT |
3794 | ret->last_waited = jiffies; /* doesn't matter... */ |
3795 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
3796 | ret->aic = NULL; | |
e2d74ac0 | 3797 | ret->cic_root.rb_node = NULL; |
4e521c27 | 3798 | ret->ioc_data = NULL; |
9f83e45e ON |
3799 | /* make sure set_task_ioprio() sees the settings above */ |
3800 | smp_wmb(); | |
fb3cc432 NP |
3801 | tsk->io_context = ret; |
3802 | } | |
1da177e4 | 3803 | |
fb3cc432 NP |
3804 | return ret; |
3805 | } | |
1da177e4 | 3806 | |
fb3cc432 NP |
3807 | /* |
3808 | * If the current task has no IO context then create one and initialise it. | |
3809 | * If it does have a context, take a ref on it. | |
3810 | * | |
3811 | * This is always called in the context of the task which submitted the I/O. | |
3812 | */ | |
b5deef90 | 3813 | struct io_context *get_io_context(gfp_t gfp_flags, int node) |
fb3cc432 NP |
3814 | { |
3815 | struct io_context *ret; | |
b5deef90 | 3816 | ret = current_io_context(gfp_flags, node); |
fb3cc432 | 3817 | if (likely(ret)) |
1da177e4 | 3818 | atomic_inc(&ret->refcount); |
1da177e4 LT |
3819 | return ret; |
3820 | } | |
3821 | EXPORT_SYMBOL(get_io_context); | |
3822 | ||
3823 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
3824 | { | |
3825 | struct io_context *src = *psrc; | |
3826 | struct io_context *dst = *pdst; | |
3827 | ||
3828 | if (src) { | |
3829 | BUG_ON(atomic_read(&src->refcount) == 0); | |
3830 | atomic_inc(&src->refcount); | |
3831 | put_io_context(dst); | |
3832 | *pdst = src; | |
3833 | } | |
3834 | } | |
3835 | EXPORT_SYMBOL(copy_io_context); | |
3836 | ||
3837 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
3838 | { | |
3839 | struct io_context *temp; | |
3840 | temp = *ioc1; | |
3841 | *ioc1 = *ioc2; | |
3842 | *ioc2 = temp; | |
3843 | } | |
3844 | EXPORT_SYMBOL(swap_io_context); | |
3845 | ||
3846 | /* | |
3847 | * sysfs parts below | |
3848 | */ | |
3849 | struct queue_sysfs_entry { | |
3850 | struct attribute attr; | |
3851 | ssize_t (*show)(struct request_queue *, char *); | |
3852 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
3853 | }; | |
3854 | ||
3855 | static ssize_t | |
3856 | queue_var_show(unsigned int var, char *page) | |
3857 | { | |
3858 | return sprintf(page, "%d\n", var); | |
3859 | } | |
3860 | ||
3861 | static ssize_t | |
3862 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
3863 | { | |
3864 | char *p = (char *) page; | |
3865 | ||
3866 | *var = simple_strtoul(p, &p, 10); | |
3867 | return count; | |
3868 | } | |
3869 | ||
3870 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
3871 | { | |
3872 | return queue_var_show(q->nr_requests, (page)); | |
3873 | } | |
3874 | ||
3875 | static ssize_t | |
3876 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
3877 | { | |
3878 | struct request_list *rl = &q->rq; | |
c981ff9f AV |
3879 | unsigned long nr; |
3880 | int ret = queue_var_store(&nr, page, count); | |
3881 | if (nr < BLKDEV_MIN_RQ) | |
3882 | nr = BLKDEV_MIN_RQ; | |
1da177e4 | 3883 | |
c981ff9f AV |
3884 | spin_lock_irq(q->queue_lock); |
3885 | q->nr_requests = nr; | |
1da177e4 LT |
3886 | blk_queue_congestion_threshold(q); |
3887 | ||
3888 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3889 | blk_set_queue_congested(q, READ); |
1da177e4 | 3890 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
79e2de4b | 3891 | blk_clear_queue_congested(q, READ); |
1da177e4 LT |
3892 | |
3893 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 3894 | blk_set_queue_congested(q, WRITE); |
1da177e4 | 3895 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
79e2de4b | 3896 | blk_clear_queue_congested(q, WRITE); |
1da177e4 LT |
3897 | |
3898 | if (rl->count[READ] >= q->nr_requests) { | |
3899 | blk_set_queue_full(q, READ); | |
3900 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
3901 | blk_clear_queue_full(q, READ); | |
3902 | wake_up(&rl->wait[READ]); | |
3903 | } | |
3904 | ||
3905 | if (rl->count[WRITE] >= q->nr_requests) { | |
3906 | blk_set_queue_full(q, WRITE); | |
3907 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
3908 | blk_clear_queue_full(q, WRITE); | |
3909 | wake_up(&rl->wait[WRITE]); | |
3910 | } | |
c981ff9f | 3911 | spin_unlock_irq(q->queue_lock); |
1da177e4 LT |
3912 | return ret; |
3913 | } | |
3914 | ||
3915 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
3916 | { | |
3917 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3918 | ||
3919 | return queue_var_show(ra_kb, (page)); | |
3920 | } | |
3921 | ||
3922 | static ssize_t | |
3923 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
3924 | { | |
3925 | unsigned long ra_kb; | |
3926 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
3927 | ||
3928 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
3929 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
3930 | spin_unlock_irq(q->queue_lock); | |
3931 | ||
3932 | return ret; | |
3933 | } | |
3934 | ||
3935 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
3936 | { | |
3937 | int max_sectors_kb = q->max_sectors >> 1; | |
3938 | ||
3939 | return queue_var_show(max_sectors_kb, (page)); | |
3940 | } | |
3941 | ||
3942 | static ssize_t | |
3943 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
3944 | { | |
3945 | unsigned long max_sectors_kb, | |
3946 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
3947 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
3948 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
3949 | int ra_kb; | |
3950 | ||
3951 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
3952 | return -EINVAL; | |
3953 | /* | |
3954 | * Take the queue lock to update the readahead and max_sectors | |
3955 | * values synchronously: | |
3956 | */ | |
3957 | spin_lock_irq(q->queue_lock); | |
3958 | /* | |
3959 | * Trim readahead window as well, if necessary: | |
3960 | */ | |
3961 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
3962 | if (ra_kb > max_sectors_kb) | |
3963 | q->backing_dev_info.ra_pages = | |
3964 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); | |
3965 | ||
3966 | q->max_sectors = max_sectors_kb << 1; | |
3967 | spin_unlock_irq(q->queue_lock); | |
3968 | ||
3969 | return ret; | |
3970 | } | |
3971 | ||
3972 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
3973 | { | |
3974 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
3975 | ||
3976 | return queue_var_show(max_hw_sectors_kb, (page)); | |
3977 | } | |
3978 | ||
3979 | ||
3980 | static struct queue_sysfs_entry queue_requests_entry = { | |
3981 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
3982 | .show = queue_requests_show, | |
3983 | .store = queue_requests_store, | |
3984 | }; | |
3985 | ||
3986 | static struct queue_sysfs_entry queue_ra_entry = { | |
3987 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
3988 | .show = queue_ra_show, | |
3989 | .store = queue_ra_store, | |
3990 | }; | |
3991 | ||
3992 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
3993 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
3994 | .show = queue_max_sectors_show, | |
3995 | .store = queue_max_sectors_store, | |
3996 | }; | |
3997 | ||
3998 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
3999 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
4000 | .show = queue_max_hw_sectors_show, | |
4001 | }; | |
4002 | ||
4003 | static struct queue_sysfs_entry queue_iosched_entry = { | |
4004 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
4005 | .show = elv_iosched_show, | |
4006 | .store = elv_iosched_store, | |
4007 | }; | |
4008 | ||
4009 | static struct attribute *default_attrs[] = { | |
4010 | &queue_requests_entry.attr, | |
4011 | &queue_ra_entry.attr, | |
4012 | &queue_max_hw_sectors_entry.attr, | |
4013 | &queue_max_sectors_entry.attr, | |
4014 | &queue_iosched_entry.attr, | |
4015 | NULL, | |
4016 | }; | |
4017 | ||
4018 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
4019 | ||
4020 | static ssize_t | |
4021 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4022 | { | |
4023 | struct queue_sysfs_entry *entry = to_queue(attr); | |
483f4afc AV |
4024 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
4025 | ssize_t res; | |
1da177e4 | 4026 | |
1da177e4 | 4027 | if (!entry->show) |
6c1852a0 | 4028 | return -EIO; |
483f4afc AV |
4029 | mutex_lock(&q->sysfs_lock); |
4030 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4031 | mutex_unlock(&q->sysfs_lock); | |
4032 | return -ENOENT; | |
4033 | } | |
4034 | res = entry->show(q, page); | |
4035 | mutex_unlock(&q->sysfs_lock); | |
4036 | return res; | |
1da177e4 LT |
4037 | } |
4038 | ||
4039 | static ssize_t | |
4040 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
4041 | const char *page, size_t length) | |
4042 | { | |
4043 | struct queue_sysfs_entry *entry = to_queue(attr); | |
483f4afc AV |
4044 | request_queue_t *q = container_of(kobj, struct request_queue, kobj); |
4045 | ||
4046 | ssize_t res; | |
1da177e4 | 4047 | |
1da177e4 | 4048 | if (!entry->store) |
6c1852a0 | 4049 | return -EIO; |
483f4afc AV |
4050 | mutex_lock(&q->sysfs_lock); |
4051 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4052 | mutex_unlock(&q->sysfs_lock); | |
4053 | return -ENOENT; | |
4054 | } | |
4055 | res = entry->store(q, page, length); | |
4056 | mutex_unlock(&q->sysfs_lock); | |
4057 | return res; | |
1da177e4 LT |
4058 | } |
4059 | ||
4060 | static struct sysfs_ops queue_sysfs_ops = { | |
4061 | .show = queue_attr_show, | |
4062 | .store = queue_attr_store, | |
4063 | }; | |
4064 | ||
93d17d3d | 4065 | static struct kobj_type queue_ktype = { |
1da177e4 LT |
4066 | .sysfs_ops = &queue_sysfs_ops, |
4067 | .default_attrs = default_attrs, | |
483f4afc | 4068 | .release = blk_release_queue, |
1da177e4 LT |
4069 | }; |
4070 | ||
4071 | int blk_register_queue(struct gendisk *disk) | |
4072 | { | |
4073 | int ret; | |
4074 | ||
4075 | request_queue_t *q = disk->queue; | |
4076 | ||
4077 | if (!q || !q->request_fn) | |
4078 | return -ENXIO; | |
4079 | ||
4080 | q->kobj.parent = kobject_get(&disk->kobj); | |
1da177e4 | 4081 | |
483f4afc | 4082 | ret = kobject_add(&q->kobj); |
1da177e4 LT |
4083 | if (ret < 0) |
4084 | return ret; | |
4085 | ||
483f4afc AV |
4086 | kobject_uevent(&q->kobj, KOBJ_ADD); |
4087 | ||
1da177e4 LT |
4088 | ret = elv_register_queue(q); |
4089 | if (ret) { | |
483f4afc AV |
4090 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4091 | kobject_del(&q->kobj); | |
1da177e4 LT |
4092 | return ret; |
4093 | } | |
4094 | ||
4095 | return 0; | |
4096 | } | |
4097 | ||
4098 | void blk_unregister_queue(struct gendisk *disk) | |
4099 | { | |
4100 | request_queue_t *q = disk->queue; | |
4101 | ||
4102 | if (q && q->request_fn) { | |
4103 | elv_unregister_queue(q); | |
4104 | ||
483f4afc AV |
4105 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4106 | kobject_del(&q->kobj); | |
1da177e4 LT |
4107 | kobject_put(&disk->kobj); |
4108 | } | |
4109 | } |