]>
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> |
f565913e | 33 | #include <linux/scatterlist.h> |
1da177e4 LT |
34 | |
35 | /* | |
36 | * for max sense size | |
37 | */ | |
38 | #include <scsi/scsi_cmnd.h> | |
39 | ||
65f27f38 | 40 | static void blk_unplug_work(struct work_struct *work); |
1da177e4 | 41 | static void blk_unplug_timeout(unsigned long data); |
b238b3d4 | 42 | static void drive_stat_acct(struct request *rq, int new_io); |
52d9e675 | 43 | static void init_request_from_bio(struct request *req, struct bio *bio); |
165125e1 | 44 | static int __make_request(struct request_queue *q, struct bio *bio); |
b5deef90 | 45 | static struct io_context *current_io_context(gfp_t gfp_flags, int node); |
9dfa5283 | 46 | static void blk_recalc_rq_segments(struct request *rq); |
66846572 N |
47 | static void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
48 | struct bio *bio); | |
1da177e4 LT |
49 | |
50 | /* | |
51 | * For the allocated request tables | |
52 | */ | |
e18b890b | 53 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
54 | |
55 | /* | |
56 | * For queue allocation | |
57 | */ | |
e18b890b | 58 | static struct kmem_cache *requestq_cachep; |
1da177e4 LT |
59 | |
60 | /* | |
61 | * For io context allocations | |
62 | */ | |
e18b890b | 63 | static struct kmem_cache *iocontext_cachep; |
1da177e4 | 64 | |
1da177e4 LT |
65 | /* |
66 | * Controlling structure to kblockd | |
67 | */ | |
ff856bad | 68 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 LT |
69 | |
70 | unsigned long blk_max_low_pfn, blk_max_pfn; | |
71 | ||
72 | EXPORT_SYMBOL(blk_max_low_pfn); | |
73 | EXPORT_SYMBOL(blk_max_pfn); | |
74 | ||
ff856bad JA |
75 | static DEFINE_PER_CPU(struct list_head, blk_cpu_done); |
76 | ||
1da177e4 LT |
77 | /* Amount of time in which a process may batch requests */ |
78 | #define BLK_BATCH_TIME (HZ/50UL) | |
79 | ||
80 | /* Number of requests a "batching" process may submit */ | |
81 | #define BLK_BATCH_REQ 32 | |
82 | ||
83 | /* | |
84 | * Return the threshold (number of used requests) at which the queue is | |
85 | * considered to be congested. It include a little hysteresis to keep the | |
86 | * context switch rate down. | |
87 | */ | |
88 | static inline int queue_congestion_on_threshold(struct request_queue *q) | |
89 | { | |
90 | return q->nr_congestion_on; | |
91 | } | |
92 | ||
93 | /* | |
94 | * The threshold at which a queue is considered to be uncongested | |
95 | */ | |
96 | static inline int queue_congestion_off_threshold(struct request_queue *q) | |
97 | { | |
98 | return q->nr_congestion_off; | |
99 | } | |
100 | ||
101 | static void blk_queue_congestion_threshold(struct request_queue *q) | |
102 | { | |
103 | int nr; | |
104 | ||
105 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
106 | if (nr > q->nr_requests) | |
107 | nr = q->nr_requests; | |
108 | q->nr_congestion_on = nr; | |
109 | ||
110 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
111 | if (nr < 1) | |
112 | nr = 1; | |
113 | q->nr_congestion_off = nr; | |
114 | } | |
115 | ||
1da177e4 LT |
116 | /** |
117 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
118 | * @bdev: device | |
119 | * | |
120 | * Locates the passed device's request queue and returns the address of its | |
121 | * backing_dev_info | |
122 | * | |
123 | * Will return NULL if the request queue cannot be located. | |
124 | */ | |
125 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
126 | { | |
127 | struct backing_dev_info *ret = NULL; | |
165125e1 | 128 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
129 | |
130 | if (q) | |
131 | ret = &q->backing_dev_info; | |
132 | return ret; | |
133 | } | |
1da177e4 LT |
134 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
135 | ||
1da177e4 LT |
136 | /** |
137 | * blk_queue_prep_rq - set a prepare_request function for queue | |
138 | * @q: queue | |
139 | * @pfn: prepare_request function | |
140 | * | |
141 | * It's possible for a queue to register a prepare_request callback which | |
142 | * is invoked before the request is handed to the request_fn. The goal of | |
143 | * the function is to prepare a request for I/O, it can be used to build a | |
144 | * cdb from the request data for instance. | |
145 | * | |
146 | */ | |
165125e1 | 147 | void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn) |
1da177e4 LT |
148 | { |
149 | q->prep_rq_fn = pfn; | |
150 | } | |
151 | ||
152 | EXPORT_SYMBOL(blk_queue_prep_rq); | |
153 | ||
154 | /** | |
155 | * blk_queue_merge_bvec - set a merge_bvec function for queue | |
156 | * @q: queue | |
157 | * @mbfn: merge_bvec_fn | |
158 | * | |
159 | * Usually queues have static limitations on the max sectors or segments that | |
160 | * we can put in a request. Stacking drivers may have some settings that | |
161 | * are dynamic, and thus we have to query the queue whether it is ok to | |
162 | * add a new bio_vec to a bio at a given offset or not. If the block device | |
163 | * has such limitations, it needs to register a merge_bvec_fn to control | |
164 | * the size of bio's sent to it. Note that a block device *must* allow a | |
165 | * single page to be added to an empty bio. The block device driver may want | |
166 | * to use the bio_split() function to deal with these bio's. By default | |
167 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | |
168 | * honored. | |
169 | */ | |
165125e1 | 170 | void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn) |
1da177e4 LT |
171 | { |
172 | q->merge_bvec_fn = mbfn; | |
173 | } | |
174 | ||
175 | EXPORT_SYMBOL(blk_queue_merge_bvec); | |
176 | ||
165125e1 | 177 | void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn) |
ff856bad JA |
178 | { |
179 | q->softirq_done_fn = fn; | |
180 | } | |
181 | ||
182 | EXPORT_SYMBOL(blk_queue_softirq_done); | |
183 | ||
1da177e4 LT |
184 | /** |
185 | * blk_queue_make_request - define an alternate make_request function for a device | |
186 | * @q: the request queue for the device to be affected | |
187 | * @mfn: the alternate make_request function | |
188 | * | |
189 | * Description: | |
190 | * The normal way for &struct bios to be passed to a device | |
191 | * driver is for them to be collected into requests on a request | |
192 | * queue, and then to allow the device driver to select requests | |
193 | * off that queue when it is ready. This works well for many block | |
194 | * devices. However some block devices (typically virtual devices | |
195 | * such as md or lvm) do not benefit from the processing on the | |
196 | * request queue, and are served best by having the requests passed | |
197 | * directly to them. This can be achieved by providing a function | |
198 | * to blk_queue_make_request(). | |
199 | * | |
200 | * Caveat: | |
201 | * The driver that does this *must* be able to deal appropriately | |
202 | * with buffers in "highmemory". This can be accomplished by either calling | |
203 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | |
204 | * blk_queue_bounce() to create a buffer in normal memory. | |
205 | **/ | |
165125e1 | 206 | void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn) |
1da177e4 LT |
207 | { |
208 | /* | |
209 | * set defaults | |
210 | */ | |
211 | q->nr_requests = BLKDEV_MAX_RQ; | |
309c0a1d SM |
212 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
213 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1da177e4 LT |
214 | q->make_request_fn = mfn; |
215 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | |
216 | q->backing_dev_info.state = 0; | |
217 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | |
defd94b7 | 218 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
1da177e4 LT |
219 | blk_queue_hardsect_size(q, 512); |
220 | blk_queue_dma_alignment(q, 511); | |
221 | blk_queue_congestion_threshold(q); | |
222 | q->nr_batching = BLK_BATCH_REQ; | |
223 | ||
224 | q->unplug_thresh = 4; /* hmm */ | |
225 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | |
226 | if (q->unplug_delay == 0) | |
227 | q->unplug_delay = 1; | |
228 | ||
65f27f38 | 229 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
1da177e4 LT |
230 | |
231 | q->unplug_timer.function = blk_unplug_timeout; | |
232 | q->unplug_timer.data = (unsigned long)q; | |
233 | ||
234 | /* | |
235 | * by default assume old behaviour and bounce for any highmem page | |
236 | */ | |
237 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | |
1da177e4 LT |
238 | } |
239 | ||
240 | EXPORT_SYMBOL(blk_queue_make_request); | |
241 | ||
165125e1 | 242 | static void rq_init(struct request_queue *q, struct request *rq) |
1da177e4 LT |
243 | { |
244 | INIT_LIST_HEAD(&rq->queuelist); | |
ff856bad | 245 | INIT_LIST_HEAD(&rq->donelist); |
1da177e4 LT |
246 | |
247 | rq->errors = 0; | |
1da177e4 | 248 | rq->bio = rq->biotail = NULL; |
2e662b65 JA |
249 | INIT_HLIST_NODE(&rq->hash); |
250 | RB_CLEAR_NODE(&rq->rb_node); | |
22e2c507 | 251 | rq->ioprio = 0; |
1da177e4 LT |
252 | rq->buffer = NULL; |
253 | rq->ref_count = 1; | |
254 | rq->q = q; | |
1da177e4 LT |
255 | rq->special = NULL; |
256 | rq->data_len = 0; | |
257 | rq->data = NULL; | |
df46b9a4 | 258 | rq->nr_phys_segments = 0; |
1da177e4 LT |
259 | rq->sense = NULL; |
260 | rq->end_io = NULL; | |
261 | rq->end_io_data = NULL; | |
ff856bad | 262 | rq->completion_data = NULL; |
abae1fde | 263 | rq->next_rq = NULL; |
1da177e4 LT |
264 | } |
265 | ||
266 | /** | |
267 | * blk_queue_ordered - does this queue support ordered writes | |
797e7dbb TH |
268 | * @q: the request queue |
269 | * @ordered: one of QUEUE_ORDERED_* | |
fddfdeaf | 270 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes |
1da177e4 LT |
271 | * |
272 | * Description: | |
273 | * For journalled file systems, doing ordered writes on a commit | |
274 | * block instead of explicitly doing wait_on_buffer (which is bad | |
275 | * for performance) can be a big win. Block drivers supporting this | |
276 | * feature should call this function and indicate so. | |
277 | * | |
278 | **/ | |
165125e1 | 279 | int blk_queue_ordered(struct request_queue *q, unsigned ordered, |
797e7dbb TH |
280 | prepare_flush_fn *prepare_flush_fn) |
281 | { | |
282 | if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && | |
283 | prepare_flush_fn == NULL) { | |
284 | printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); | |
285 | return -EINVAL; | |
286 | } | |
287 | ||
288 | if (ordered != QUEUE_ORDERED_NONE && | |
289 | ordered != QUEUE_ORDERED_DRAIN && | |
290 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
291 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
292 | ordered != QUEUE_ORDERED_TAG && | |
293 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
294 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
295 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
296 | return -EINVAL; | |
1da177e4 | 297 | } |
797e7dbb | 298 | |
60481b12 | 299 | q->ordered = ordered; |
797e7dbb TH |
300 | q->next_ordered = ordered; |
301 | q->prepare_flush_fn = prepare_flush_fn; | |
302 | ||
303 | return 0; | |
1da177e4 LT |
304 | } |
305 | ||
306 | EXPORT_SYMBOL(blk_queue_ordered); | |
307 | ||
1da177e4 LT |
308 | /* |
309 | * Cache flushing for ordered writes handling | |
310 | */ | |
165125e1 | 311 | inline unsigned blk_ordered_cur_seq(struct request_queue *q) |
1da177e4 | 312 | { |
797e7dbb TH |
313 | if (!q->ordseq) |
314 | return 0; | |
315 | return 1 << ffz(q->ordseq); | |
1da177e4 LT |
316 | } |
317 | ||
797e7dbb | 318 | unsigned blk_ordered_req_seq(struct request *rq) |
1da177e4 | 319 | { |
165125e1 | 320 | struct request_queue *q = rq->q; |
1da177e4 | 321 | |
797e7dbb | 322 | BUG_ON(q->ordseq == 0); |
8922e16c | 323 | |
797e7dbb TH |
324 | if (rq == &q->pre_flush_rq) |
325 | return QUEUE_ORDSEQ_PREFLUSH; | |
326 | if (rq == &q->bar_rq) | |
327 | return QUEUE_ORDSEQ_BAR; | |
328 | if (rq == &q->post_flush_rq) | |
329 | return QUEUE_ORDSEQ_POSTFLUSH; | |
1da177e4 | 330 | |
bc90ba09 TH |
331 | /* |
332 | * !fs requests don't need to follow barrier ordering. Always | |
333 | * put them at the front. This fixes the following deadlock. | |
334 | * | |
335 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
336 | */ | |
337 | if (!blk_fs_request(rq)) | |
338 | return QUEUE_ORDSEQ_DRAIN; | |
339 | ||
4aff5e23 JA |
340 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == |
341 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
797e7dbb TH |
342 | return QUEUE_ORDSEQ_DRAIN; |
343 | else | |
344 | return QUEUE_ORDSEQ_DONE; | |
1da177e4 LT |
345 | } |
346 | ||
165125e1 | 347 | void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error) |
1da177e4 | 348 | { |
797e7dbb | 349 | struct request *rq; |
1da177e4 | 350 | |
797e7dbb TH |
351 | if (error && !q->orderr) |
352 | q->orderr = error; | |
1da177e4 | 353 | |
797e7dbb TH |
354 | BUG_ON(q->ordseq & seq); |
355 | q->ordseq |= seq; | |
1da177e4 | 356 | |
797e7dbb TH |
357 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) |
358 | return; | |
1da177e4 LT |
359 | |
360 | /* | |
797e7dbb | 361 | * Okay, sequence complete. |
1da177e4 | 362 | */ |
797e7dbb | 363 | q->ordseq = 0; |
4fa253f3 | 364 | rq = q->orig_bar_rq; |
1da177e4 | 365 | |
9e6e39f2 KU |
366 | if (__blk_end_request(rq, q->orderr, blk_rq_bytes(rq))) |
367 | BUG(); | |
1da177e4 LT |
368 | } |
369 | ||
797e7dbb | 370 | static void pre_flush_end_io(struct request *rq, int error) |
1da177e4 | 371 | { |
797e7dbb TH |
372 | elv_completed_request(rq->q, rq); |
373 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
374 | } | |
1da177e4 | 375 | |
797e7dbb TH |
376 | static void bar_end_io(struct request *rq, int error) |
377 | { | |
378 | elv_completed_request(rq->q, rq); | |
379 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
380 | } | |
1da177e4 | 381 | |
797e7dbb TH |
382 | static void post_flush_end_io(struct request *rq, int error) |
383 | { | |
384 | elv_completed_request(rq->q, rq); | |
385 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
386 | } | |
1da177e4 | 387 | |
165125e1 | 388 | static void queue_flush(struct request_queue *q, unsigned which) |
797e7dbb TH |
389 | { |
390 | struct request *rq; | |
391 | rq_end_io_fn *end_io; | |
1da177e4 | 392 | |
797e7dbb TH |
393 | if (which == QUEUE_ORDERED_PREFLUSH) { |
394 | rq = &q->pre_flush_rq; | |
395 | end_io = pre_flush_end_io; | |
396 | } else { | |
397 | rq = &q->post_flush_rq; | |
398 | end_io = post_flush_end_io; | |
1da177e4 | 399 | } |
797e7dbb | 400 | |
4aff5e23 | 401 | rq->cmd_flags = REQ_HARDBARRIER; |
797e7dbb | 402 | rq_init(q, rq); |
797e7dbb | 403 | rq->elevator_private = NULL; |
c00895ab | 404 | rq->elevator_private2 = NULL; |
797e7dbb | 405 | rq->rq_disk = q->bar_rq.rq_disk; |
797e7dbb TH |
406 | rq->end_io = end_io; |
407 | q->prepare_flush_fn(q, rq); | |
408 | ||
30e9656c | 409 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
1da177e4 LT |
410 | } |
411 | ||
165125e1 | 412 | static inline struct request *start_ordered(struct request_queue *q, |
797e7dbb | 413 | struct request *rq) |
1da177e4 | 414 | { |
797e7dbb TH |
415 | q->orderr = 0; |
416 | q->ordered = q->next_ordered; | |
417 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
418 | ||
419 | /* | |
420 | * Prep proxy barrier request. | |
421 | */ | |
422 | blkdev_dequeue_request(rq); | |
423 | q->orig_bar_rq = rq; | |
424 | rq = &q->bar_rq; | |
4aff5e23 | 425 | rq->cmd_flags = 0; |
797e7dbb | 426 | rq_init(q, rq); |
4aff5e23 JA |
427 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) |
428 | rq->cmd_flags |= REQ_RW; | |
4fa253f3 JA |
429 | if (q->ordered & QUEUE_ORDERED_FUA) |
430 | rq->cmd_flags |= REQ_FUA; | |
797e7dbb | 431 | rq->elevator_private = NULL; |
c00895ab | 432 | rq->elevator_private2 = NULL; |
797e7dbb TH |
433 | init_request_from_bio(rq, q->orig_bar_rq->bio); |
434 | rq->end_io = bar_end_io; | |
435 | ||
436 | /* | |
437 | * Queue ordered sequence. As we stack them at the head, we | |
438 | * need to queue in reverse order. Note that we rely on that | |
439 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
bf2de6f5 JA |
440 | * request gets inbetween ordered sequence. If this request is |
441 | * an empty barrier, we don't need to do a postflush ever since | |
442 | * there will be no data written between the pre and post flush. | |
443 | * Hence a single flush will suffice. | |
797e7dbb | 444 | */ |
bf2de6f5 | 445 | if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq)) |
797e7dbb TH |
446 | queue_flush(q, QUEUE_ORDERED_POSTFLUSH); |
447 | else | |
448 | q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; | |
449 | ||
30e9656c | 450 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); |
797e7dbb TH |
451 | |
452 | if (q->ordered & QUEUE_ORDERED_PREFLUSH) { | |
453 | queue_flush(q, QUEUE_ORDERED_PREFLUSH); | |
454 | rq = &q->pre_flush_rq; | |
455 | } else | |
456 | q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; | |
1da177e4 | 457 | |
797e7dbb TH |
458 | if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) |
459 | q->ordseq |= QUEUE_ORDSEQ_DRAIN; | |
460 | else | |
461 | rq = NULL; | |
462 | ||
463 | return rq; | |
1da177e4 LT |
464 | } |
465 | ||
165125e1 | 466 | int blk_do_ordered(struct request_queue *q, struct request **rqp) |
1da177e4 | 467 | { |
9a7a67af | 468 | struct request *rq = *rqp; |
bf2de6f5 | 469 | const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); |
1da177e4 | 470 | |
797e7dbb TH |
471 | if (!q->ordseq) { |
472 | if (!is_barrier) | |
473 | return 1; | |
1da177e4 | 474 | |
797e7dbb TH |
475 | if (q->next_ordered != QUEUE_ORDERED_NONE) { |
476 | *rqp = start_ordered(q, rq); | |
477 | return 1; | |
478 | } else { | |
479 | /* | |
480 | * This can happen when the queue switches to | |
481 | * ORDERED_NONE while this request is on it. | |
482 | */ | |
483 | blkdev_dequeue_request(rq); | |
9e6e39f2 KU |
484 | if (__blk_end_request(rq, -EOPNOTSUPP, |
485 | blk_rq_bytes(rq))) | |
486 | BUG(); | |
797e7dbb TH |
487 | *rqp = NULL; |
488 | return 0; | |
489 | } | |
490 | } | |
1da177e4 | 491 | |
9a7a67af JA |
492 | /* |
493 | * Ordered sequence in progress | |
494 | */ | |
495 | ||
496 | /* Special requests are not subject to ordering rules. */ | |
497 | if (!blk_fs_request(rq) && | |
498 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
499 | return 1; | |
500 | ||
797e7dbb | 501 | if (q->ordered & QUEUE_ORDERED_TAG) { |
9a7a67af | 502 | /* Ordered by tag. Blocking the next barrier is enough. */ |
797e7dbb TH |
503 | if (is_barrier && rq != &q->bar_rq) |
504 | *rqp = NULL; | |
9a7a67af JA |
505 | } else { |
506 | /* Ordered by draining. Wait for turn. */ | |
507 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
508 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
509 | *rqp = NULL; | |
1da177e4 LT |
510 | } |
511 | ||
512 | return 1; | |
513 | } | |
514 | ||
5bb23a68 N |
515 | static void req_bio_endio(struct request *rq, struct bio *bio, |
516 | unsigned int nbytes, int error) | |
1da177e4 | 517 | { |
165125e1 | 518 | struct request_queue *q = rq->q; |
797e7dbb | 519 | |
5bb23a68 N |
520 | if (&q->bar_rq != rq) { |
521 | if (error) | |
522 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
523 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
524 | error = -EIO; | |
797e7dbb | 525 | |
5bb23a68 N |
526 | if (unlikely(nbytes > bio->bi_size)) { |
527 | printk("%s: want %u bytes done, only %u left\n", | |
528 | __FUNCTION__, nbytes, bio->bi_size); | |
529 | nbytes = bio->bi_size; | |
530 | } | |
797e7dbb | 531 | |
5bb23a68 N |
532 | bio->bi_size -= nbytes; |
533 | bio->bi_sector += (nbytes >> 9); | |
534 | if (bio->bi_size == 0) | |
6712ecf8 | 535 | bio_endio(bio, error); |
5bb23a68 N |
536 | } else { |
537 | ||
538 | /* | |
539 | * Okay, this is the barrier request in progress, just | |
540 | * record the error; | |
541 | */ | |
542 | if (error && !q->orderr) | |
543 | q->orderr = error; | |
544 | } | |
1da177e4 | 545 | } |
1da177e4 LT |
546 | |
547 | /** | |
548 | * blk_queue_bounce_limit - set bounce buffer limit for queue | |
549 | * @q: the request queue for the device | |
550 | * @dma_addr: bus address limit | |
551 | * | |
552 | * Description: | |
553 | * Different hardware can have different requirements as to what pages | |
554 | * it can do I/O directly to. A low level driver can call | |
555 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | |
5ee1af9f | 556 | * buffers for doing I/O to pages residing above @page. |
1da177e4 | 557 | **/ |
165125e1 | 558 | void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr) |
1da177e4 LT |
559 | { |
560 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | |
5ee1af9f AK |
561 | int dma = 0; |
562 | ||
563 | q->bounce_gfp = GFP_NOIO; | |
564 | #if BITS_PER_LONG == 64 | |
565 | /* Assume anything <= 4GB can be handled by IOMMU. | |
566 | Actually some IOMMUs can handle everything, but I don't | |
567 | know of a way to test this here. */ | |
8269730b | 568 | if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT)) |
5ee1af9f AK |
569 | dma = 1; |
570 | q->bounce_pfn = max_low_pfn; | |
571 | #else | |
572 | if (bounce_pfn < blk_max_low_pfn) | |
573 | dma = 1; | |
574 | q->bounce_pfn = bounce_pfn; | |
575 | #endif | |
576 | if (dma) { | |
1da177e4 LT |
577 | init_emergency_isa_pool(); |
578 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | |
5ee1af9f AK |
579 | q->bounce_pfn = bounce_pfn; |
580 | } | |
1da177e4 LT |
581 | } |
582 | ||
583 | EXPORT_SYMBOL(blk_queue_bounce_limit); | |
584 | ||
585 | /** | |
586 | * blk_queue_max_sectors - set max sectors for a request for this queue | |
587 | * @q: the request queue for the device | |
588 | * @max_sectors: max sectors in the usual 512b unit | |
589 | * | |
590 | * Description: | |
591 | * Enables a low level driver to set an upper limit on the size of | |
592 | * received requests. | |
593 | **/ | |
165125e1 | 594 | void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors) |
1da177e4 LT |
595 | { |
596 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | |
597 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | |
598 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | |
599 | } | |
600 | ||
defd94b7 MC |
601 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
602 | q->max_hw_sectors = q->max_sectors = max_sectors; | |
603 | else { | |
604 | q->max_sectors = BLK_DEF_MAX_SECTORS; | |
605 | q->max_hw_sectors = max_sectors; | |
606 | } | |
1da177e4 LT |
607 | } |
608 | ||
609 | EXPORT_SYMBOL(blk_queue_max_sectors); | |
610 | ||
611 | /** | |
612 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | |
613 | * @q: the request queue for the device | |
614 | * @max_segments: max number of segments | |
615 | * | |
616 | * Description: | |
617 | * Enables a low level driver to set an upper limit on the number of | |
618 | * physical data segments in a request. This would be the largest sized | |
619 | * scatter list the driver could handle. | |
620 | **/ | |
165125e1 JA |
621 | void blk_queue_max_phys_segments(struct request_queue *q, |
622 | unsigned short max_segments) | |
1da177e4 LT |
623 | { |
624 | if (!max_segments) { | |
625 | max_segments = 1; | |
626 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
627 | } | |
628 | ||
629 | q->max_phys_segments = max_segments; | |
630 | } | |
631 | ||
632 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | |
633 | ||
634 | /** | |
635 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | |
636 | * @q: the request queue for the device | |
637 | * @max_segments: max number of segments | |
638 | * | |
639 | * Description: | |
640 | * Enables a low level driver to set an upper limit on the number of | |
641 | * hw data segments in a request. This would be the largest number of | |
642 | * address/length pairs the host adapter can actually give as once | |
643 | * to the device. | |
644 | **/ | |
165125e1 JA |
645 | void blk_queue_max_hw_segments(struct request_queue *q, |
646 | unsigned short max_segments) | |
1da177e4 LT |
647 | { |
648 | if (!max_segments) { | |
649 | max_segments = 1; | |
650 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | |
651 | } | |
652 | ||
653 | q->max_hw_segments = max_segments; | |
654 | } | |
655 | ||
656 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | |
657 | ||
658 | /** | |
659 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | |
660 | * @q: the request queue for the device | |
661 | * @max_size: max size of segment in bytes | |
662 | * | |
663 | * Description: | |
664 | * Enables a low level driver to set an upper limit on the size of a | |
665 | * coalesced segment | |
666 | **/ | |
165125e1 | 667 | void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size) |
1da177e4 LT |
668 | { |
669 | if (max_size < PAGE_CACHE_SIZE) { | |
670 | max_size = PAGE_CACHE_SIZE; | |
671 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | |
672 | } | |
673 | ||
674 | q->max_segment_size = max_size; | |
675 | } | |
676 | ||
677 | EXPORT_SYMBOL(blk_queue_max_segment_size); | |
678 | ||
679 | /** | |
680 | * blk_queue_hardsect_size - set hardware sector size for the queue | |
681 | * @q: the request queue for the device | |
682 | * @size: the hardware sector size, in bytes | |
683 | * | |
684 | * Description: | |
685 | * This should typically be set to the lowest possible sector size | |
686 | * that the hardware can operate on (possible without reverting to | |
687 | * even internal read-modify-write operations). Usually the default | |
688 | * of 512 covers most hardware. | |
689 | **/ | |
165125e1 | 690 | void blk_queue_hardsect_size(struct request_queue *q, unsigned short size) |
1da177e4 LT |
691 | { |
692 | q->hardsect_size = size; | |
693 | } | |
694 | ||
695 | EXPORT_SYMBOL(blk_queue_hardsect_size); | |
696 | ||
697 | /* | |
698 | * Returns the minimum that is _not_ zero, unless both are zero. | |
699 | */ | |
700 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | |
701 | ||
702 | /** | |
703 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | |
704 | * @t: the stacking driver (top) | |
705 | * @b: the underlying device (bottom) | |
706 | **/ | |
165125e1 | 707 | void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) |
1da177e4 LT |
708 | { |
709 | /* zero is "infinity" */ | |
defd94b7 MC |
710 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); |
711 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | |
1da177e4 LT |
712 | |
713 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | |
714 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | |
715 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | |
716 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | |
89e5c8b5 N |
717 | if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) |
718 | clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); | |
1da177e4 LT |
719 | } |
720 | ||
721 | EXPORT_SYMBOL(blk_queue_stack_limits); | |
722 | ||
723 | /** | |
724 | * blk_queue_segment_boundary - set boundary rules for segment merging | |
725 | * @q: the request queue for the device | |
726 | * @mask: the memory boundary mask | |
727 | **/ | |
165125e1 | 728 | void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask) |
1da177e4 LT |
729 | { |
730 | if (mask < PAGE_CACHE_SIZE - 1) { | |
731 | mask = PAGE_CACHE_SIZE - 1; | |
732 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | |
733 | } | |
734 | ||
735 | q->seg_boundary_mask = mask; | |
736 | } | |
737 | ||
738 | EXPORT_SYMBOL(blk_queue_segment_boundary); | |
739 | ||
740 | /** | |
741 | * blk_queue_dma_alignment - set dma length and memory alignment | |
742 | * @q: the request queue for the device | |
743 | * @mask: alignment mask | |
744 | * | |
745 | * description: | |
746 | * set required memory and length aligment for direct dma transactions. | |
747 | * this is used when buiding direct io requests for the queue. | |
748 | * | |
749 | **/ | |
165125e1 | 750 | void blk_queue_dma_alignment(struct request_queue *q, int mask) |
1da177e4 LT |
751 | { |
752 | q->dma_alignment = mask; | |
753 | } | |
754 | ||
755 | EXPORT_SYMBOL(blk_queue_dma_alignment); | |
756 | ||
11c3e689 JB |
757 | /** |
758 | * blk_queue_update_dma_alignment - update dma length and memory alignment | |
759 | * @q: the request queue for the device | |
760 | * @mask: alignment mask | |
761 | * | |
762 | * description: | |
763 | * update required memory and length aligment for direct dma transactions. | |
764 | * If the requested alignment is larger than the current alignment, then | |
765 | * the current queue alignment is updated to the new value, otherwise it | |
766 | * is left alone. The design of this is to allow multiple objects | |
767 | * (driver, device, transport etc) to set their respective | |
768 | * alignments without having them interfere. | |
769 | * | |
770 | **/ | |
771 | void blk_queue_update_dma_alignment(struct request_queue *q, int mask) | |
772 | { | |
773 | BUG_ON(mask > PAGE_SIZE); | |
774 | ||
775 | if (mask > q->dma_alignment) | |
776 | q->dma_alignment = mask; | |
777 | } | |
778 | ||
779 | EXPORT_SYMBOL(blk_queue_update_dma_alignment); | |
780 | ||
1da177e4 LT |
781 | /** |
782 | * blk_queue_find_tag - find a request by its tag and queue | |
1da177e4 LT |
783 | * @q: The request queue for the device |
784 | * @tag: The tag of the request | |
785 | * | |
786 | * Notes: | |
787 | * Should be used when a device returns a tag and you want to match | |
788 | * it with a request. | |
789 | * | |
790 | * no locks need be held. | |
791 | **/ | |
165125e1 | 792 | struct request *blk_queue_find_tag(struct request_queue *q, int tag) |
1da177e4 | 793 | { |
f583f492 | 794 | return blk_map_queue_find_tag(q->queue_tags, tag); |
1da177e4 LT |
795 | } |
796 | ||
797 | EXPORT_SYMBOL(blk_queue_find_tag); | |
798 | ||
799 | /** | |
492dfb48 JB |
800 | * __blk_free_tags - release a given set of tag maintenance info |
801 | * @bqt: the tag map to free | |
1da177e4 | 802 | * |
492dfb48 JB |
803 | * Tries to free the specified @bqt@. Returns true if it was |
804 | * actually freed and false if there are still references using it | |
805 | */ | |
806 | static int __blk_free_tags(struct blk_queue_tag *bqt) | |
1da177e4 | 807 | { |
492dfb48 | 808 | int retval; |
1da177e4 | 809 | |
492dfb48 JB |
810 | retval = atomic_dec_and_test(&bqt->refcnt); |
811 | if (retval) { | |
1da177e4 | 812 | BUG_ON(bqt->busy); |
1da177e4 LT |
813 | |
814 | kfree(bqt->tag_index); | |
815 | bqt->tag_index = NULL; | |
816 | ||
817 | kfree(bqt->tag_map); | |
818 | bqt->tag_map = NULL; | |
819 | ||
820 | kfree(bqt); | |
492dfb48 | 821 | |
1da177e4 LT |
822 | } |
823 | ||
492dfb48 JB |
824 | return retval; |
825 | } | |
826 | ||
827 | /** | |
828 | * __blk_queue_free_tags - release tag maintenance info | |
829 | * @q: the request queue for the device | |
830 | * | |
831 | * Notes: | |
832 | * blk_cleanup_queue() will take care of calling this function, if tagging | |
833 | * has been used. So there's no need to call this directly. | |
834 | **/ | |
165125e1 | 835 | static void __blk_queue_free_tags(struct request_queue *q) |
492dfb48 JB |
836 | { |
837 | struct blk_queue_tag *bqt = q->queue_tags; | |
838 | ||
839 | if (!bqt) | |
840 | return; | |
841 | ||
842 | __blk_free_tags(bqt); | |
843 | ||
1da177e4 LT |
844 | q->queue_tags = NULL; |
845 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | |
846 | } | |
847 | ||
492dfb48 JB |
848 | |
849 | /** | |
850 | * blk_free_tags - release a given set of tag maintenance info | |
851 | * @bqt: the tag map to free | |
852 | * | |
853 | * For externally managed @bqt@ frees the map. Callers of this | |
854 | * function must guarantee to have released all the queues that | |
855 | * might have been using this tag map. | |
856 | */ | |
857 | void blk_free_tags(struct blk_queue_tag *bqt) | |
858 | { | |
859 | if (unlikely(!__blk_free_tags(bqt))) | |
860 | BUG(); | |
861 | } | |
862 | EXPORT_SYMBOL(blk_free_tags); | |
863 | ||
1da177e4 LT |
864 | /** |
865 | * blk_queue_free_tags - release tag maintenance info | |
866 | * @q: the request queue for the device | |
867 | * | |
868 | * Notes: | |
869 | * This is used to disabled tagged queuing to a device, yet leave | |
870 | * queue in function. | |
871 | **/ | |
165125e1 | 872 | void blk_queue_free_tags(struct request_queue *q) |
1da177e4 LT |
873 | { |
874 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
875 | } | |
876 | ||
877 | EXPORT_SYMBOL(blk_queue_free_tags); | |
878 | ||
879 | static int | |
165125e1 | 880 | init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth) |
1da177e4 | 881 | { |
1da177e4 LT |
882 | struct request **tag_index; |
883 | unsigned long *tag_map; | |
fa72b903 | 884 | int nr_ulongs; |
1da177e4 | 885 | |
492dfb48 | 886 | if (q && depth > q->nr_requests * 2) { |
1da177e4 LT |
887 | depth = q->nr_requests * 2; |
888 | printk(KERN_ERR "%s: adjusted depth to %d\n", | |
889 | __FUNCTION__, depth); | |
890 | } | |
891 | ||
f68110fc | 892 | tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
1da177e4 LT |
893 | if (!tag_index) |
894 | goto fail; | |
895 | ||
f7d37d02 | 896 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
f68110fc | 897 | tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
1da177e4 LT |
898 | if (!tag_map) |
899 | goto fail; | |
900 | ||
ba025082 | 901 | tags->real_max_depth = depth; |
1da177e4 | 902 | tags->max_depth = depth; |
1da177e4 LT |
903 | tags->tag_index = tag_index; |
904 | tags->tag_map = tag_map; | |
905 | ||
1da177e4 LT |
906 | return 0; |
907 | fail: | |
908 | kfree(tag_index); | |
909 | return -ENOMEM; | |
910 | } | |
911 | ||
492dfb48 JB |
912 | static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q, |
913 | int depth) | |
914 | { | |
915 | struct blk_queue_tag *tags; | |
916 | ||
917 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | |
918 | if (!tags) | |
919 | goto fail; | |
920 | ||
921 | if (init_tag_map(q, tags, depth)) | |
922 | goto fail; | |
923 | ||
492dfb48 JB |
924 | tags->busy = 0; |
925 | atomic_set(&tags->refcnt, 1); | |
926 | return tags; | |
927 | fail: | |
928 | kfree(tags); | |
929 | return NULL; | |
930 | } | |
931 | ||
932 | /** | |
933 | * blk_init_tags - initialize the tag info for an external tag map | |
934 | * @depth: the maximum queue depth supported | |
935 | * @tags: the tag to use | |
936 | **/ | |
937 | struct blk_queue_tag *blk_init_tags(int depth) | |
938 | { | |
939 | return __blk_queue_init_tags(NULL, depth); | |
940 | } | |
941 | EXPORT_SYMBOL(blk_init_tags); | |
942 | ||
1da177e4 LT |
943 | /** |
944 | * blk_queue_init_tags - initialize the queue tag info | |
945 | * @q: the request queue for the device | |
946 | * @depth: the maximum queue depth supported | |
947 | * @tags: the tag to use | |
948 | **/ | |
165125e1 | 949 | int blk_queue_init_tags(struct request_queue *q, int depth, |
1da177e4 LT |
950 | struct blk_queue_tag *tags) |
951 | { | |
952 | int rc; | |
953 | ||
954 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | |
955 | ||
956 | if (!tags && !q->queue_tags) { | |
492dfb48 | 957 | tags = __blk_queue_init_tags(q, depth); |
1da177e4 | 958 | |
492dfb48 | 959 | if (!tags) |
1da177e4 | 960 | goto fail; |
1da177e4 LT |
961 | } else if (q->queue_tags) { |
962 | if ((rc = blk_queue_resize_tags(q, depth))) | |
963 | return rc; | |
964 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | |
965 | return 0; | |
966 | } else | |
967 | atomic_inc(&tags->refcnt); | |
968 | ||
969 | /* | |
970 | * assign it, all done | |
971 | */ | |
972 | q->queue_tags = tags; | |
973 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | |
6eca9004 | 974 | INIT_LIST_HEAD(&q->tag_busy_list); |
1da177e4 LT |
975 | return 0; |
976 | fail: | |
977 | kfree(tags); | |
978 | return -ENOMEM; | |
979 | } | |
980 | ||
981 | EXPORT_SYMBOL(blk_queue_init_tags); | |
982 | ||
983 | /** | |
984 | * blk_queue_resize_tags - change the queueing depth | |
985 | * @q: the request queue for the device | |
986 | * @new_depth: the new max command queueing depth | |
987 | * | |
988 | * Notes: | |
989 | * Must be called with the queue lock held. | |
990 | **/ | |
165125e1 | 991 | int blk_queue_resize_tags(struct request_queue *q, int new_depth) |
1da177e4 LT |
992 | { |
993 | struct blk_queue_tag *bqt = q->queue_tags; | |
994 | struct request **tag_index; | |
995 | unsigned long *tag_map; | |
fa72b903 | 996 | int max_depth, nr_ulongs; |
1da177e4 LT |
997 | |
998 | if (!bqt) | |
999 | return -ENXIO; | |
1000 | ||
ba025082 TH |
1001 | /* |
1002 | * if we already have large enough real_max_depth. just | |
1003 | * adjust max_depth. *NOTE* as requests with tag value | |
1004 | * between new_depth and real_max_depth can be in-flight, tag | |
1005 | * map can not be shrunk blindly here. | |
1006 | */ | |
1007 | if (new_depth <= bqt->real_max_depth) { | |
1008 | bqt->max_depth = new_depth; | |
1009 | return 0; | |
1010 | } | |
1011 | ||
492dfb48 JB |
1012 | /* |
1013 | * Currently cannot replace a shared tag map with a new | |
1014 | * one, so error out if this is the case | |
1015 | */ | |
1016 | if (atomic_read(&bqt->refcnt) != 1) | |
1017 | return -EBUSY; | |
1018 | ||
1da177e4 LT |
1019 | /* |
1020 | * save the old state info, so we can copy it back | |
1021 | */ | |
1022 | tag_index = bqt->tag_index; | |
1023 | tag_map = bqt->tag_map; | |
ba025082 | 1024 | max_depth = bqt->real_max_depth; |
1da177e4 LT |
1025 | |
1026 | if (init_tag_map(q, bqt, new_depth)) | |
1027 | return -ENOMEM; | |
1028 | ||
1029 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | |
f7d37d02 | 1030 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
fa72b903 | 1031 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
1da177e4 LT |
1032 | |
1033 | kfree(tag_index); | |
1034 | kfree(tag_map); | |
1035 | return 0; | |
1036 | } | |
1037 | ||
1038 | EXPORT_SYMBOL(blk_queue_resize_tags); | |
1039 | ||
1040 | /** | |
1041 | * blk_queue_end_tag - end tag operations for a request | |
1042 | * @q: the request queue for the device | |
1043 | * @rq: the request that has completed | |
1044 | * | |
1045 | * Description: | |
1046 | * Typically called when end_that_request_first() returns 0, meaning | |
1047 | * all transfers have been done for a request. It's important to call | |
1048 | * this function before end_that_request_last(), as that will put the | |
1049 | * request back on the free list thus corrupting the internal tag list. | |
1050 | * | |
1051 | * Notes: | |
1052 | * queue lock must be held. | |
1053 | **/ | |
165125e1 | 1054 | void blk_queue_end_tag(struct request_queue *q, struct request *rq) |
1da177e4 LT |
1055 | { |
1056 | struct blk_queue_tag *bqt = q->queue_tags; | |
1057 | int tag = rq->tag; | |
1058 | ||
1059 | BUG_ON(tag == -1); | |
1060 | ||
ba025082 | 1061 | if (unlikely(tag >= bqt->real_max_depth)) |
040c928c TH |
1062 | /* |
1063 | * This can happen after tag depth has been reduced. | |
1064 | * FIXME: how about a warning or info message here? | |
1065 | */ | |
1da177e4 LT |
1066 | return; |
1067 | ||
1da177e4 | 1068 | list_del_init(&rq->queuelist); |
4aff5e23 | 1069 | rq->cmd_flags &= ~REQ_QUEUED; |
1da177e4 LT |
1070 | rq->tag = -1; |
1071 | ||
1072 | if (unlikely(bqt->tag_index[tag] == NULL)) | |
040c928c TH |
1073 | printk(KERN_ERR "%s: tag %d is missing\n", |
1074 | __FUNCTION__, tag); | |
1da177e4 LT |
1075 | |
1076 | bqt->tag_index[tag] = NULL; | |
f3da54ba | 1077 | |
adb4ddbb | 1078 | if (unlikely(!test_bit(tag, bqt->tag_map))) { |
f3da54ba JA |
1079 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
1080 | __FUNCTION__, tag); | |
1081 | return; | |
1082 | } | |
adb4ddbb NP |
1083 | /* |
1084 | * The tag_map bit acts as a lock for tag_index[bit], so we need | |
1085 | * unlock memory barrier semantics. | |
1086 | */ | |
1087 | clear_bit_unlock(tag, bqt->tag_map); | |
1da177e4 LT |
1088 | bqt->busy--; |
1089 | } | |
1090 | ||
1091 | EXPORT_SYMBOL(blk_queue_end_tag); | |
1092 | ||
1093 | /** | |
1094 | * blk_queue_start_tag - find a free tag and assign it | |
1095 | * @q: the request queue for the device | |
1096 | * @rq: the block request that needs tagging | |
1097 | * | |
1098 | * Description: | |
1099 | * This can either be used as a stand-alone helper, or possibly be | |
1100 | * assigned as the queue &prep_rq_fn (in which case &struct request | |
1101 | * automagically gets a tag assigned). Note that this function | |
1102 | * assumes that any type of request can be queued! if this is not | |
1103 | * true for your device, you must check the request type before | |
1104 | * calling this function. The request will also be removed from | |
1105 | * the request queue, so it's the drivers responsibility to readd | |
1106 | * it if it should need to be restarted for some reason. | |
1107 | * | |
1108 | * Notes: | |
1109 | * queue lock must be held. | |
1110 | **/ | |
165125e1 | 1111 | int blk_queue_start_tag(struct request_queue *q, struct request *rq) |
1da177e4 LT |
1112 | { |
1113 | struct blk_queue_tag *bqt = q->queue_tags; | |
2bf0fdad | 1114 | int tag; |
1da177e4 | 1115 | |
4aff5e23 | 1116 | if (unlikely((rq->cmd_flags & REQ_QUEUED))) { |
1da177e4 | 1117 | printk(KERN_ERR |
040c928c TH |
1118 | "%s: request %p for device [%s] already tagged %d", |
1119 | __FUNCTION__, rq, | |
1120 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | |
1da177e4 LT |
1121 | BUG(); |
1122 | } | |
1123 | ||
059af497 JA |
1124 | /* |
1125 | * Protect against shared tag maps, as we may not have exclusive | |
1126 | * access to the tag map. | |
1127 | */ | |
1128 | do { | |
1129 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | |
1130 | if (tag >= bqt->max_depth) | |
1131 | return 1; | |
1da177e4 | 1132 | |
adb4ddbb | 1133 | } while (test_and_set_bit_lock(tag, bqt->tag_map)); |
dd941252 | 1134 | /* |
adb4ddbb NP |
1135 | * We need lock ordering semantics given by test_and_set_bit_lock. |
1136 | * See blk_queue_end_tag for details. | |
dd941252 | 1137 | */ |
1da177e4 | 1138 | |
4aff5e23 | 1139 | rq->cmd_flags |= REQ_QUEUED; |
1da177e4 LT |
1140 | rq->tag = tag; |
1141 | bqt->tag_index[tag] = rq; | |
1142 | blkdev_dequeue_request(rq); | |
6eca9004 | 1143 | list_add(&rq->queuelist, &q->tag_busy_list); |
1da177e4 LT |
1144 | bqt->busy++; |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | EXPORT_SYMBOL(blk_queue_start_tag); | |
1149 | ||
1150 | /** | |
1151 | * blk_queue_invalidate_tags - invalidate all pending tags | |
1152 | * @q: the request queue for the device | |
1153 | * | |
1154 | * Description: | |
1155 | * Hardware conditions may dictate a need to stop all pending requests. | |
1156 | * In this case, we will safely clear the block side of the tag queue and | |
1157 | * readd all requests to the request queue in the right order. | |
1158 | * | |
1159 | * Notes: | |
1160 | * queue lock must be held. | |
1161 | **/ | |
165125e1 | 1162 | void blk_queue_invalidate_tags(struct request_queue *q) |
1da177e4 | 1163 | { |
1da177e4 | 1164 | struct list_head *tmp, *n; |
1da177e4 | 1165 | |
d85532ed JA |
1166 | list_for_each_safe(tmp, n, &q->tag_busy_list) |
1167 | blk_requeue_request(q, list_entry_rq(tmp)); | |
1da177e4 LT |
1168 | } |
1169 | ||
1170 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | |
1171 | ||
1da177e4 LT |
1172 | void blk_dump_rq_flags(struct request *rq, char *msg) |
1173 | { | |
1174 | int bit; | |
1175 | ||
4aff5e23 JA |
1176 | printk("%s: dev %s: type=%x, flags=%x\n", msg, |
1177 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, | |
1178 | rq->cmd_flags); | |
1da177e4 LT |
1179 | |
1180 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | |
1181 | rq->nr_sectors, | |
1182 | rq->current_nr_sectors); | |
1183 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | |
1184 | ||
4aff5e23 | 1185 | if (blk_pc_request(rq)) { |
1da177e4 LT |
1186 | printk("cdb: "); |
1187 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | |
1188 | printk("%02x ", rq->cmd[bit]); | |
1189 | printk("\n"); | |
1190 | } | |
1191 | } | |
1192 | ||
1193 | EXPORT_SYMBOL(blk_dump_rq_flags); | |
1194 | ||
165125e1 | 1195 | void blk_recount_segments(struct request_queue *q, struct bio *bio) |
1da177e4 | 1196 | { |
9dfa5283 N |
1197 | struct request rq; |
1198 | struct bio *nxt = bio->bi_next; | |
1199 | rq.q = q; | |
1200 | rq.bio = rq.biotail = bio; | |
1201 | bio->bi_next = NULL; | |
1202 | blk_recalc_rq_segments(&rq); | |
1203 | bio->bi_next = nxt; | |
1204 | bio->bi_phys_segments = rq.nr_phys_segments; | |
1205 | bio->bi_hw_segments = rq.nr_hw_segments; | |
1206 | bio->bi_flags |= (1 << BIO_SEG_VALID); | |
1207 | } | |
1208 | EXPORT_SYMBOL(blk_recount_segments); | |
1209 | ||
1210 | static void blk_recalc_rq_segments(struct request *rq) | |
1211 | { | |
1212 | int nr_phys_segs; | |
1213 | int nr_hw_segs; | |
1214 | unsigned int phys_size; | |
1215 | unsigned int hw_size; | |
1da177e4 | 1216 | struct bio_vec *bv, *bvprv = NULL; |
9dfa5283 N |
1217 | int seg_size; |
1218 | int hw_seg_size; | |
1219 | int cluster; | |
5705f702 | 1220 | struct req_iterator iter; |
1da177e4 | 1221 | int high, highprv = 1; |
9dfa5283 | 1222 | struct request_queue *q = rq->q; |
1da177e4 | 1223 | |
9dfa5283 | 1224 | if (!rq->bio) |
1da177e4 LT |
1225 | return; |
1226 | ||
1227 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
9dfa5283 N |
1228 | hw_seg_size = seg_size = 0; |
1229 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | |
5705f702 | 1230 | rq_for_each_segment(bv, rq, iter) { |
1da177e4 LT |
1231 | /* |
1232 | * the trick here is making sure that a high page is never | |
1233 | * considered part of another segment, since that might | |
1234 | * change with the bounce page. | |
1235 | */ | |
f772b3d9 | 1236 | high = page_to_pfn(bv->bv_page) > q->bounce_pfn; |
1da177e4 LT |
1237 | if (high || highprv) |
1238 | goto new_hw_segment; | |
1239 | if (cluster) { | |
1240 | if (seg_size + bv->bv_len > q->max_segment_size) | |
1241 | goto new_segment; | |
1242 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | |
1243 | goto new_segment; | |
1244 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | |
1245 | goto new_segment; | |
1246 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | |
1247 | goto new_hw_segment; | |
1248 | ||
1249 | seg_size += bv->bv_len; | |
1250 | hw_seg_size += bv->bv_len; | |
1251 | bvprv = bv; | |
1252 | continue; | |
1253 | } | |
1254 | new_segment: | |
1255 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | |
9dfa5283 | 1256 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) |
1da177e4 | 1257 | hw_seg_size += bv->bv_len; |
9dfa5283 | 1258 | else { |
1da177e4 | 1259 | new_hw_segment: |
9dfa5283 N |
1260 | if (nr_hw_segs == 1 && |
1261 | hw_seg_size > rq->bio->bi_hw_front_size) | |
1262 | rq->bio->bi_hw_front_size = hw_seg_size; | |
1da177e4 LT |
1263 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; |
1264 | nr_hw_segs++; | |
1265 | } | |
1266 | ||
1267 | nr_phys_segs++; | |
1268 | bvprv = bv; | |
1269 | seg_size = bv->bv_len; | |
1270 | highprv = high; | |
1271 | } | |
9dfa5283 N |
1272 | |
1273 | if (nr_hw_segs == 1 && | |
1274 | hw_seg_size > rq->bio->bi_hw_front_size) | |
1275 | rq->bio->bi_hw_front_size = hw_seg_size; | |
1276 | if (hw_seg_size > rq->biotail->bi_hw_back_size) | |
1277 | rq->biotail->bi_hw_back_size = hw_seg_size; | |
1278 | rq->nr_phys_segments = nr_phys_segs; | |
1279 | rq->nr_hw_segments = nr_hw_segs; | |
1da177e4 | 1280 | } |
1da177e4 | 1281 | |
165125e1 | 1282 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, |
1da177e4 LT |
1283 | struct bio *nxt) |
1284 | { | |
1285 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | |
1286 | return 0; | |
1287 | ||
1288 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | |
1289 | return 0; | |
1290 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | |
1291 | return 0; | |
1292 | ||
1293 | /* | |
1294 | * bio and nxt are contigous in memory, check if the queue allows | |
1295 | * these two to be merged into one | |
1296 | */ | |
1297 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | |
1298 | return 1; | |
1299 | ||
1300 | return 0; | |
1301 | } | |
1302 | ||
165125e1 | 1303 | static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio, |
1da177e4 LT |
1304 | struct bio *nxt) |
1305 | { | |
1306 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1307 | blk_recount_segments(q, bio); | |
1308 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | |
1309 | blk_recount_segments(q, nxt); | |
1310 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | |
32eef964 | 1311 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size)) |
1da177e4 | 1312 | return 0; |
32eef964 | 1313 | if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size) |
1da177e4 LT |
1314 | return 0; |
1315 | ||
1316 | return 1; | |
1317 | } | |
1318 | ||
1da177e4 LT |
1319 | /* |
1320 | * map a request to scatterlist, return number of sg entries setup. Caller | |
1321 | * must make sure sg can hold rq->nr_phys_segments entries | |
1322 | */ | |
165125e1 | 1323 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, |
f565913e | 1324 | struct scatterlist *sglist) |
1da177e4 LT |
1325 | { |
1326 | struct bio_vec *bvec, *bvprv; | |
5705f702 | 1327 | struct req_iterator iter; |
ba951841 | 1328 | struct scatterlist *sg; |
5705f702 | 1329 | int nsegs, cluster; |
1da177e4 LT |
1330 | |
1331 | nsegs = 0; | |
1332 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | |
1333 | ||
1334 | /* | |
1335 | * for each bio in rq | |
1336 | */ | |
1337 | bvprv = NULL; | |
ba951841 | 1338 | sg = NULL; |
5705f702 | 1339 | rq_for_each_segment(bvec, rq, iter) { |
6c92e699 | 1340 | int nbytes = bvec->bv_len; |
1da177e4 | 1341 | |
6c92e699 | 1342 | if (bvprv && cluster) { |
f565913e | 1343 | if (sg->length + nbytes > q->max_segment_size) |
6c92e699 | 1344 | goto new_segment; |
1da177e4 | 1345 | |
6c92e699 JA |
1346 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) |
1347 | goto new_segment; | |
1348 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | |
1349 | goto new_segment; | |
1da177e4 | 1350 | |
f565913e | 1351 | sg->length += nbytes; |
6c92e699 | 1352 | } else { |
1da177e4 | 1353 | new_segment: |
ba951841 JA |
1354 | if (!sg) |
1355 | sg = sglist; | |
7aeacf98 JA |
1356 | else { |
1357 | /* | |
1358 | * If the driver previously mapped a shorter | |
1359 | * list, we could see a termination bit | |
1360 | * prematurely unless it fully inits the sg | |
1361 | * table on each mapping. We KNOW that there | |
1362 | * must be more entries here or the driver | |
1363 | * would be buggy, so force clear the | |
1364 | * termination bit to avoid doing a full | |
1365 | * sg_init_table() in drivers for each command. | |
1366 | */ | |
1367 | sg->page_link &= ~0x02; | |
ba951841 | 1368 | sg = sg_next(sg); |
7aeacf98 | 1369 | } |
6c92e699 | 1370 | |
642f1490 | 1371 | sg_set_page(sg, bvec->bv_page, nbytes, bvec->bv_offset); |
6c92e699 JA |
1372 | nsegs++; |
1373 | } | |
1374 | bvprv = bvec; | |
5705f702 | 1375 | } /* segments in rq */ |
1da177e4 | 1376 | |
9b61764b | 1377 | if (sg) |
c46f2334 | 1378 | sg_mark_end(sg); |
9b61764b | 1379 | |
1da177e4 LT |
1380 | return nsegs; |
1381 | } | |
1382 | ||
1383 | EXPORT_SYMBOL(blk_rq_map_sg); | |
1384 | ||
1385 | /* | |
1386 | * the standard queue merge functions, can be overridden with device | |
1387 | * specific ones if so desired | |
1388 | */ | |
1389 | ||
165125e1 | 1390 | static inline int ll_new_mergeable(struct request_queue *q, |
1da177e4 LT |
1391 | struct request *req, |
1392 | struct bio *bio) | |
1393 | { | |
1394 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1395 | ||
1396 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
4aff5e23 | 1397 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1398 | if (req == q->last_merge) |
1399 | q->last_merge = NULL; | |
1400 | return 0; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * A hw segment is just getting larger, bump just the phys | |
1405 | * counter. | |
1406 | */ | |
1407 | req->nr_phys_segments += nr_phys_segs; | |
1408 | return 1; | |
1409 | } | |
1410 | ||
165125e1 | 1411 | static inline int ll_new_hw_segment(struct request_queue *q, |
1da177e4 LT |
1412 | struct request *req, |
1413 | struct bio *bio) | |
1414 | { | |
1415 | int nr_hw_segs = bio_hw_segments(q, bio); | |
1416 | int nr_phys_segs = bio_phys_segments(q, bio); | |
1417 | ||
1418 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | |
1419 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | |
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 | ||
1426 | /* | |
1427 | * This will form the start of a new hw segment. Bump both | |
1428 | * counters. | |
1429 | */ | |
1430 | req->nr_hw_segments += nr_hw_segs; | |
1431 | req->nr_phys_segments += nr_phys_segs; | |
1432 | return 1; | |
1433 | } | |
1434 | ||
3001ca77 N |
1435 | static int ll_back_merge_fn(struct request_queue *q, struct request *req, |
1436 | struct bio *bio) | |
1da177e4 | 1437 | { |
defd94b7 | 1438 | unsigned short max_sectors; |
1da177e4 LT |
1439 | int len; |
1440 | ||
defd94b7 MC |
1441 | if (unlikely(blk_pc_request(req))) |
1442 | max_sectors = q->max_hw_sectors; | |
1443 | else | |
1444 | max_sectors = q->max_sectors; | |
1445 | ||
1446 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1447 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1448 | if (req == q->last_merge) |
1449 | q->last_merge = NULL; | |
1450 | return 0; | |
1451 | } | |
1452 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | |
1453 | blk_recount_segments(q, req->biotail); | |
1454 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1455 | blk_recount_segments(q, bio); | |
1456 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | |
1457 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | |
1458 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1459 | int mergeable = ll_new_mergeable(q, req, bio); | |
1460 | ||
1461 | if (mergeable) { | |
1462 | if (req->nr_hw_segments == 1) | |
1463 | req->bio->bi_hw_front_size = len; | |
1464 | if (bio->bi_hw_segments == 1) | |
1465 | bio->bi_hw_back_size = len; | |
1466 | } | |
1467 | return mergeable; | |
1468 | } | |
1469 | ||
1470 | return ll_new_hw_segment(q, req, bio); | |
1471 | } | |
1472 | ||
165125e1 | 1473 | static int ll_front_merge_fn(struct request_queue *q, struct request *req, |
1da177e4 LT |
1474 | struct bio *bio) |
1475 | { | |
defd94b7 | 1476 | unsigned short max_sectors; |
1da177e4 LT |
1477 | int len; |
1478 | ||
defd94b7 MC |
1479 | if (unlikely(blk_pc_request(req))) |
1480 | max_sectors = q->max_hw_sectors; | |
1481 | else | |
1482 | max_sectors = q->max_sectors; | |
1483 | ||
1484 | ||
1485 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | |
4aff5e23 | 1486 | req->cmd_flags |= REQ_NOMERGE; |
1da177e4 LT |
1487 | if (req == q->last_merge) |
1488 | q->last_merge = NULL; | |
1489 | return 0; | |
1490 | } | |
1491 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | |
1492 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
1493 | blk_recount_segments(q, bio); | |
1494 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | |
1495 | blk_recount_segments(q, req->bio); | |
1496 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | |
1497 | !BIOVEC_VIRT_OVERSIZE(len)) { | |
1498 | int mergeable = ll_new_mergeable(q, req, bio); | |
1499 | ||
1500 | if (mergeable) { | |
1501 | if (bio->bi_hw_segments == 1) | |
1502 | bio->bi_hw_front_size = len; | |
1503 | if (req->nr_hw_segments == 1) | |
1504 | req->biotail->bi_hw_back_size = len; | |
1505 | } | |
1506 | return mergeable; | |
1507 | } | |
1508 | ||
1509 | return ll_new_hw_segment(q, req, bio); | |
1510 | } | |
1511 | ||
165125e1 | 1512 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
1da177e4 LT |
1513 | struct request *next) |
1514 | { | |
dfa1a553 ND |
1515 | int total_phys_segments; |
1516 | int total_hw_segments; | |
1da177e4 LT |
1517 | |
1518 | /* | |
1519 | * First check if the either of the requests are re-queued | |
1520 | * requests. Can't merge them if they are. | |
1521 | */ | |
1522 | if (req->special || next->special) | |
1523 | return 0; | |
1524 | ||
1525 | /* | |
dfa1a553 | 1526 | * Will it become too large? |
1da177e4 LT |
1527 | */ |
1528 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | |
1529 | return 0; | |
1530 | ||
1531 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
1532 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | |
1533 | total_phys_segments--; | |
1534 | ||
1535 | if (total_phys_segments > q->max_phys_segments) | |
1536 | return 0; | |
1537 | ||
1538 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | |
1539 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | |
1540 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | |
1541 | /* | |
1542 | * propagate the combined length to the end of the requests | |
1543 | */ | |
1544 | if (req->nr_hw_segments == 1) | |
1545 | req->bio->bi_hw_front_size = len; | |
1546 | if (next->nr_hw_segments == 1) | |
1547 | next->biotail->bi_hw_back_size = len; | |
1548 | total_hw_segments--; | |
1549 | } | |
1550 | ||
1551 | if (total_hw_segments > q->max_hw_segments) | |
1552 | return 0; | |
1553 | ||
1554 | /* Merge is OK... */ | |
1555 | req->nr_phys_segments = total_phys_segments; | |
1556 | req->nr_hw_segments = total_hw_segments; | |
1557 | return 1; | |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * "plug" the device if there are no outstanding requests: this will | |
1562 | * force the transfer to start only after we have put all the requests | |
1563 | * on the list. | |
1564 | * | |
1565 | * This is called with interrupts off and no requests on the queue and | |
1566 | * with the queue lock held. | |
1567 | */ | |
165125e1 | 1568 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
1569 | { |
1570 | WARN_ON(!irqs_disabled()); | |
1571 | ||
1572 | /* | |
1573 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
1574 | * which will restart the queueing | |
1575 | */ | |
7daac490 | 1576 | if (blk_queue_stopped(q)) |
1da177e4 LT |
1577 | return; |
1578 | ||
2056a782 | 1579 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { |
1da177e4 | 1580 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
2056a782 JA |
1581 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
1582 | } | |
1da177e4 LT |
1583 | } |
1584 | ||
1585 | EXPORT_SYMBOL(blk_plug_device); | |
1586 | ||
1587 | /* | |
1588 | * remove the queue from the plugged list, if present. called with | |
1589 | * queue lock held and interrupts disabled. | |
1590 | */ | |
165125e1 | 1591 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
1592 | { |
1593 | WARN_ON(!irqs_disabled()); | |
1594 | ||
1595 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | |
1596 | return 0; | |
1597 | ||
1598 | del_timer(&q->unplug_timer); | |
1599 | return 1; | |
1600 | } | |
1601 | ||
1602 | EXPORT_SYMBOL(blk_remove_plug); | |
1603 | ||
1604 | /* | |
1605 | * remove the plug and let it rip.. | |
1606 | */ | |
165125e1 | 1607 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 1608 | { |
7daac490 | 1609 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 LT |
1610 | return; |
1611 | ||
1612 | if (!blk_remove_plug(q)) | |
1613 | return; | |
1614 | ||
22e2c507 | 1615 | q->request_fn(q); |
1da177e4 LT |
1616 | } |
1617 | EXPORT_SYMBOL(__generic_unplug_device); | |
1618 | ||
1619 | /** | |
1620 | * generic_unplug_device - fire a request queue | |
165125e1 | 1621 | * @q: The &struct request_queue in question |
1da177e4 LT |
1622 | * |
1623 | * Description: | |
1624 | * Linux uses plugging to build bigger requests queues before letting | |
1625 | * the device have at them. If a queue is plugged, the I/O scheduler | |
1626 | * is still adding and merging requests on the queue. Once the queue | |
1627 | * gets unplugged, the request_fn defined for the queue is invoked and | |
1628 | * transfers started. | |
1629 | **/ | |
165125e1 | 1630 | void generic_unplug_device(struct request_queue *q) |
1da177e4 LT |
1631 | { |
1632 | spin_lock_irq(q->queue_lock); | |
1633 | __generic_unplug_device(q); | |
1634 | spin_unlock_irq(q->queue_lock); | |
1635 | } | |
1636 | EXPORT_SYMBOL(generic_unplug_device); | |
1637 | ||
1638 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
1639 | struct page *page) | |
1640 | { | |
165125e1 | 1641 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 1642 | |
2ad8b1ef | 1643 | blk_unplug(q); |
1da177e4 LT |
1644 | } |
1645 | ||
65f27f38 | 1646 | static void blk_unplug_work(struct work_struct *work) |
1da177e4 | 1647 | { |
165125e1 JA |
1648 | struct request_queue *q = |
1649 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 1650 | |
2056a782 JA |
1651 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
1652 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1653 | ||
1da177e4 LT |
1654 | q->unplug_fn(q); |
1655 | } | |
1656 | ||
1657 | static void blk_unplug_timeout(unsigned long data) | |
1658 | { | |
165125e1 | 1659 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 1660 | |
2056a782 JA |
1661 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
1662 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1663 | ||
1da177e4 LT |
1664 | kblockd_schedule_work(&q->unplug_work); |
1665 | } | |
1666 | ||
2ad8b1ef AB |
1667 | void blk_unplug(struct request_queue *q) |
1668 | { | |
1669 | /* | |
1670 | * devices don't necessarily have an ->unplug_fn defined | |
1671 | */ | |
1672 | if (q->unplug_fn) { | |
1673 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, | |
1674 | q->rq.count[READ] + q->rq.count[WRITE]); | |
1675 | ||
1676 | q->unplug_fn(q); | |
1677 | } | |
1678 | } | |
1679 | EXPORT_SYMBOL(blk_unplug); | |
1680 | ||
1da177e4 LT |
1681 | /** |
1682 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 1683 | * @q: The &struct request_queue in question |
1da177e4 LT |
1684 | * |
1685 | * Description: | |
1686 | * blk_start_queue() will clear the stop flag on the queue, and call | |
1687 | * the request_fn for the queue if it was in a stopped state when | |
1688 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
1689 | **/ | |
165125e1 | 1690 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 1691 | { |
a038e253 PBG |
1692 | WARN_ON(!irqs_disabled()); |
1693 | ||
1da177e4 LT |
1694 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1695 | ||
1696 | /* | |
1697 | * one level of recursion is ok and is much faster than kicking | |
1698 | * the unplug handling | |
1699 | */ | |
1700 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1701 | q->request_fn(q); | |
1702 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1703 | } else { | |
1704 | blk_plug_device(q); | |
1705 | kblockd_schedule_work(&q->unplug_work); | |
1706 | } | |
1707 | } | |
1708 | ||
1709 | EXPORT_SYMBOL(blk_start_queue); | |
1710 | ||
1711 | /** | |
1712 | * blk_stop_queue - stop a queue | |
165125e1 | 1713 | * @q: The &struct request_queue in question |
1da177e4 LT |
1714 | * |
1715 | * Description: | |
1716 | * The Linux block layer assumes that a block driver will consume all | |
1717 | * entries on the request queue when the request_fn strategy is called. | |
1718 | * Often this will not happen, because of hardware limitations (queue | |
1719 | * depth settings). If a device driver gets a 'queue full' response, | |
1720 | * or if it simply chooses not to queue more I/O at one point, it can | |
1721 | * call this function to prevent the request_fn from being called until | |
1722 | * the driver has signalled it's ready to go again. This happens by calling | |
1723 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
1724 | **/ | |
165125e1 | 1725 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
1726 | { |
1727 | blk_remove_plug(q); | |
1728 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | |
1729 | } | |
1730 | EXPORT_SYMBOL(blk_stop_queue); | |
1731 | ||
1732 | /** | |
1733 | * blk_sync_queue - cancel any pending callbacks on a queue | |
1734 | * @q: the queue | |
1735 | * | |
1736 | * Description: | |
1737 | * The block layer may perform asynchronous callback activity | |
1738 | * on a queue, such as calling the unplug function after a timeout. | |
1739 | * A block device may call blk_sync_queue to ensure that any | |
1740 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 1741 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
1742 | * that its ->make_request_fn will not re-add plugging prior to calling |
1743 | * this function. | |
1744 | * | |
1745 | */ | |
1746 | void blk_sync_queue(struct request_queue *q) | |
1747 | { | |
1748 | del_timer_sync(&q->unplug_timer); | |
abbeb88d | 1749 | kblockd_flush_work(&q->unplug_work); |
1da177e4 LT |
1750 | } |
1751 | EXPORT_SYMBOL(blk_sync_queue); | |
1752 | ||
1753 | /** | |
1754 | * blk_run_queue - run a single device queue | |
1755 | * @q: The queue to run | |
1756 | */ | |
1757 | void blk_run_queue(struct request_queue *q) | |
1758 | { | |
1759 | unsigned long flags; | |
1760 | ||
1761 | spin_lock_irqsave(q->queue_lock, flags); | |
1762 | blk_remove_plug(q); | |
dac07ec1 JA |
1763 | |
1764 | /* | |
1765 | * Only recurse once to avoid overrunning the stack, let the unplug | |
1766 | * handling reinvoke the handler shortly if we already got there. | |
1767 | */ | |
1768 | if (!elv_queue_empty(q)) { | |
1769 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | |
1770 | q->request_fn(q); | |
1771 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | |
1772 | } else { | |
1773 | blk_plug_device(q); | |
1774 | kblockd_schedule_work(&q->unplug_work); | |
1775 | } | |
1776 | } | |
1777 | ||
1da177e4 LT |
1778 | spin_unlock_irqrestore(q->queue_lock, flags); |
1779 | } | |
1780 | EXPORT_SYMBOL(blk_run_queue); | |
1781 | ||
1782 | /** | |
165125e1 | 1783 | * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed |
a580290c | 1784 | * @kobj: the kobj belonging of the request queue to be released |
1da177e4 LT |
1785 | * |
1786 | * Description: | |
1787 | * blk_cleanup_queue is the pair to blk_init_queue() or | |
1788 | * blk_queue_make_request(). It should be called when a request queue is | |
1789 | * being released; typically when a block device is being de-registered. | |
1790 | * Currently, its primary task it to free all the &struct request | |
1791 | * structures that were allocated to the queue and the queue itself. | |
1792 | * | |
1793 | * Caveat: | |
1794 | * Hopefully the low level driver will have finished any | |
1795 | * outstanding requests first... | |
1796 | **/ | |
483f4afc | 1797 | static void blk_release_queue(struct kobject *kobj) |
1da177e4 | 1798 | { |
165125e1 JA |
1799 | struct request_queue *q = |
1800 | container_of(kobj, struct request_queue, kobj); | |
1da177e4 LT |
1801 | struct request_list *rl = &q->rq; |
1802 | ||
1da177e4 LT |
1803 | blk_sync_queue(q); |
1804 | ||
1805 | if (rl->rq_pool) | |
1806 | mempool_destroy(rl->rq_pool); | |
1807 | ||
1808 | if (q->queue_tags) | |
1809 | __blk_queue_free_tags(q); | |
1810 | ||
6c5c9341 | 1811 | blk_trace_shutdown(q); |
2056a782 | 1812 | |
e0bf68dd | 1813 | bdi_destroy(&q->backing_dev_info); |
1da177e4 LT |
1814 | kmem_cache_free(requestq_cachep, q); |
1815 | } | |
1816 | ||
165125e1 | 1817 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
1818 | { |
1819 | kobject_put(&q->kobj); | |
1820 | } | |
1821 | EXPORT_SYMBOL(blk_put_queue); | |
1822 | ||
165125e1 | 1823 | void blk_cleanup_queue(struct request_queue * q) |
483f4afc AV |
1824 | { |
1825 | mutex_lock(&q->sysfs_lock); | |
1826 | set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); | |
1827 | mutex_unlock(&q->sysfs_lock); | |
1828 | ||
1829 | if (q->elevator) | |
1830 | elevator_exit(q->elevator); | |
1831 | ||
1832 | blk_put_queue(q); | |
1833 | } | |
1834 | ||
1da177e4 LT |
1835 | EXPORT_SYMBOL(blk_cleanup_queue); |
1836 | ||
165125e1 | 1837 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
1838 | { |
1839 | struct request_list *rl = &q->rq; | |
1840 | ||
1841 | rl->count[READ] = rl->count[WRITE] = 0; | |
1842 | rl->starved[READ] = rl->starved[WRITE] = 0; | |
cb98fc8b | 1843 | rl->elvpriv = 0; |
1da177e4 LT |
1844 | init_waitqueue_head(&rl->wait[READ]); |
1845 | init_waitqueue_head(&rl->wait[WRITE]); | |
1da177e4 | 1846 | |
1946089a CL |
1847 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1848 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
1849 | |
1850 | if (!rl->rq_pool) | |
1851 | return -ENOMEM; | |
1852 | ||
1853 | return 0; | |
1854 | } | |
1855 | ||
165125e1 | 1856 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 1857 | { |
1946089a CL |
1858 | return blk_alloc_queue_node(gfp_mask, -1); |
1859 | } | |
1860 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 1861 | |
483f4afc AV |
1862 | static struct kobj_type queue_ktype; |
1863 | ||
165125e1 | 1864 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 1865 | { |
165125e1 | 1866 | struct request_queue *q; |
e0bf68dd | 1867 | int err; |
1946089a | 1868 | |
94f6030c CL |
1869 | q = kmem_cache_alloc_node(requestq_cachep, |
1870 | gfp_mask | __GFP_ZERO, node_id); | |
1da177e4 LT |
1871 | if (!q) |
1872 | return NULL; | |
1873 | ||
e0bf68dd PZ |
1874 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
1875 | q->backing_dev_info.unplug_io_data = q; | |
1876 | err = bdi_init(&q->backing_dev_info); | |
1877 | if (err) { | |
1878 | kmem_cache_free(requestq_cachep, q); | |
1879 | return NULL; | |
1880 | } | |
1881 | ||
1da177e4 | 1882 | init_timer(&q->unplug_timer); |
483f4afc | 1883 | |
f9cb074b | 1884 | kobject_init(&q->kobj, &queue_ktype); |
1da177e4 | 1885 | |
483f4afc AV |
1886 | mutex_init(&q->sysfs_lock); |
1887 | ||
1da177e4 LT |
1888 | return q; |
1889 | } | |
1946089a | 1890 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
1891 | |
1892 | /** | |
1893 | * blk_init_queue - prepare a request queue for use with a block device | |
1894 | * @rfn: The function to be called to process requests that have been | |
1895 | * placed on the queue. | |
1896 | * @lock: Request queue spin lock | |
1897 | * | |
1898 | * Description: | |
1899 | * If a block device wishes to use the standard request handling procedures, | |
1900 | * which sorts requests and coalesces adjacent requests, then it must | |
1901 | * call blk_init_queue(). The function @rfn will be called when there | |
1902 | * are requests on the queue that need to be processed. If the device | |
1903 | * supports plugging, then @rfn may not be called immediately when requests | |
1904 | * are available on the queue, but may be called at some time later instead. | |
1905 | * Plugged queues are generally unplugged when a buffer belonging to one | |
1906 | * of the requests on the queue is needed, or due to memory pressure. | |
1907 | * | |
1908 | * @rfn is not required, or even expected, to remove all requests off the | |
1909 | * queue, but only as many as it can handle at a time. If it does leave | |
1910 | * requests on the queue, it is responsible for arranging that the requests | |
1911 | * get dealt with eventually. | |
1912 | * | |
1913 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
1914 | * request queue; this lock will be taken also from interrupt context, so irq |
1915 | * disabling is needed for it. | |
1da177e4 LT |
1916 | * |
1917 | * Function returns a pointer to the initialized request queue, or NULL if | |
1918 | * it didn't succeed. | |
1919 | * | |
1920 | * Note: | |
1921 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
1922 | * when the block device is deactivated (such as at module unload). | |
1923 | **/ | |
1946089a | 1924 | |
165125e1 | 1925 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 1926 | { |
1946089a CL |
1927 | return blk_init_queue_node(rfn, lock, -1); |
1928 | } | |
1929 | EXPORT_SYMBOL(blk_init_queue); | |
1930 | ||
165125e1 | 1931 | struct request_queue * |
1946089a CL |
1932 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
1933 | { | |
165125e1 | 1934 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
1935 | |
1936 | if (!q) | |
1937 | return NULL; | |
1938 | ||
1946089a | 1939 | q->node = node_id; |
8669aafd AV |
1940 | if (blk_init_free_list(q)) { |
1941 | kmem_cache_free(requestq_cachep, q); | |
1942 | return NULL; | |
1943 | } | |
1da177e4 | 1944 | |
152587de JA |
1945 | /* |
1946 | * if caller didn't supply a lock, they get per-queue locking with | |
1947 | * our embedded lock | |
1948 | */ | |
1949 | if (!lock) { | |
1950 | spin_lock_init(&q->__queue_lock); | |
1951 | lock = &q->__queue_lock; | |
1952 | } | |
1953 | ||
1da177e4 | 1954 | q->request_fn = rfn; |
1da177e4 LT |
1955 | q->prep_rq_fn = NULL; |
1956 | q->unplug_fn = generic_unplug_device; | |
1957 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | |
1958 | q->queue_lock = lock; | |
1959 | ||
1960 | blk_queue_segment_boundary(q, 0xffffffff); | |
1961 | ||
1962 | blk_queue_make_request(q, __make_request); | |
1963 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | |
1964 | ||
1965 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | |
1966 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | |
1967 | ||
44ec9542 AS |
1968 | q->sg_reserved_size = INT_MAX; |
1969 | ||
1da177e4 LT |
1970 | /* |
1971 | * all done | |
1972 | */ | |
1973 | if (!elevator_init(q, NULL)) { | |
1974 | blk_queue_congestion_threshold(q); | |
1975 | return q; | |
1976 | } | |
1977 | ||
8669aafd | 1978 | blk_put_queue(q); |
1da177e4 LT |
1979 | return NULL; |
1980 | } | |
1946089a | 1981 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 1982 | |
165125e1 | 1983 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 1984 | { |
fde6ad22 | 1985 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 1986 | kobject_get(&q->kobj); |
1da177e4 LT |
1987 | return 0; |
1988 | } | |
1989 | ||
1990 | return 1; | |
1991 | } | |
1992 | ||
1993 | EXPORT_SYMBOL(blk_get_queue); | |
1994 | ||
165125e1 | 1995 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 1996 | { |
4aff5e23 | 1997 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 1998 | elv_put_request(q, rq); |
1da177e4 LT |
1999 | mempool_free(rq, q->rq.rq_pool); |
2000 | } | |
2001 | ||
1ea25ecb | 2002 | static struct request * |
165125e1 | 2003 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
1da177e4 LT |
2004 | { |
2005 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
2006 | ||
2007 | if (!rq) | |
2008 | return NULL; | |
2009 | ||
2010 | /* | |
4aff5e23 | 2011 | * first three bits are identical in rq->cmd_flags and bio->bi_rw, |
1da177e4 LT |
2012 | * see bio.h and blkdev.h |
2013 | */ | |
49171e5c | 2014 | rq->cmd_flags = rw | REQ_ALLOCED; |
1da177e4 | 2015 | |
cb98fc8b | 2016 | if (priv) { |
cb78b285 | 2017 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
2018 | mempool_free(rq, q->rq.rq_pool); |
2019 | return NULL; | |
2020 | } | |
4aff5e23 | 2021 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 2022 | } |
1da177e4 | 2023 | |
cb98fc8b | 2024 | return rq; |
1da177e4 LT |
2025 | } |
2026 | ||
2027 | /* | |
2028 | * ioc_batching returns true if the ioc is a valid batching request and | |
2029 | * should be given priority access to a request. | |
2030 | */ | |
165125e1 | 2031 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
2032 | { |
2033 | if (!ioc) | |
2034 | return 0; | |
2035 | ||
2036 | /* | |
2037 | * Make sure the process is able to allocate at least 1 request | |
2038 | * even if the batch times out, otherwise we could theoretically | |
2039 | * lose wakeups. | |
2040 | */ | |
2041 | return ioc->nr_batch_requests == q->nr_batching || | |
2042 | (ioc->nr_batch_requests > 0 | |
2043 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
2048 | * will cause the process to be a "batcher" on all queues in the system. This | |
2049 | * is the behaviour we want though - once it gets a wakeup it should be given | |
2050 | * a nice run. | |
2051 | */ | |
165125e1 | 2052 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
2053 | { |
2054 | if (!ioc || ioc_batching(q, ioc)) | |
2055 | return; | |
2056 | ||
2057 | ioc->nr_batch_requests = q->nr_batching; | |
2058 | ioc->last_waited = jiffies; | |
2059 | } | |
2060 | ||
165125e1 | 2061 | static void __freed_request(struct request_queue *q, int rw) |
1da177e4 LT |
2062 | { |
2063 | struct request_list *rl = &q->rq; | |
2064 | ||
2065 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | |
79e2de4b | 2066 | blk_clear_queue_congested(q, rw); |
1da177e4 LT |
2067 | |
2068 | if (rl->count[rw] + 1 <= q->nr_requests) { | |
1da177e4 LT |
2069 | if (waitqueue_active(&rl->wait[rw])) |
2070 | wake_up(&rl->wait[rw]); | |
2071 | ||
2072 | blk_clear_queue_full(q, rw); | |
2073 | } | |
2074 | } | |
2075 | ||
2076 | /* | |
2077 | * A request has just been released. Account for it, update the full and | |
2078 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
2079 | */ | |
165125e1 | 2080 | static void freed_request(struct request_queue *q, int rw, int priv) |
1da177e4 LT |
2081 | { |
2082 | struct request_list *rl = &q->rq; | |
2083 | ||
2084 | rl->count[rw]--; | |
cb98fc8b TH |
2085 | if (priv) |
2086 | rl->elvpriv--; | |
1da177e4 LT |
2087 | |
2088 | __freed_request(q, rw); | |
2089 | ||
2090 | if (unlikely(rl->starved[rw ^ 1])) | |
2091 | __freed_request(q, rw ^ 1); | |
1da177e4 LT |
2092 | } |
2093 | ||
2094 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | |
2095 | /* | |
d6344532 NP |
2096 | * Get a free request, queue_lock must be held. |
2097 | * Returns NULL on failure, with queue_lock held. | |
2098 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 2099 | */ |
165125e1 | 2100 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 2101 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
2102 | { |
2103 | struct request *rq = NULL; | |
2104 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 2105 | struct io_context *ioc = NULL; |
7749a8d4 | 2106 | const int rw = rw_flags & 0x01; |
88ee5ef1 JA |
2107 | int may_queue, priv; |
2108 | ||
7749a8d4 | 2109 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
2110 | if (may_queue == ELV_MQUEUE_NO) |
2111 | goto rq_starved; | |
2112 | ||
2113 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { | |
2114 | if (rl->count[rw]+1 >= q->nr_requests) { | |
b5deef90 | 2115 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
2116 | /* |
2117 | * The queue will fill after this allocation, so set | |
2118 | * it as full, and mark this process as "batching". | |
2119 | * This process will be allowed to complete a batch of | |
2120 | * requests, others will be blocked. | |
2121 | */ | |
2122 | if (!blk_queue_full(q, rw)) { | |
2123 | ioc_set_batching(q, ioc); | |
2124 | blk_set_queue_full(q, rw); | |
2125 | } else { | |
2126 | if (may_queue != ELV_MQUEUE_MUST | |
2127 | && !ioc_batching(q, ioc)) { | |
2128 | /* | |
2129 | * The queue is full and the allocating | |
2130 | * process is not a "batcher", and not | |
2131 | * exempted by the IO scheduler | |
2132 | */ | |
2133 | goto out; | |
2134 | } | |
2135 | } | |
1da177e4 | 2136 | } |
79e2de4b | 2137 | blk_set_queue_congested(q, rw); |
1da177e4 LT |
2138 | } |
2139 | ||
082cf69e JA |
2140 | /* |
2141 | * Only allow batching queuers to allocate up to 50% over the defined | |
2142 | * limit of requests, otherwise we could have thousands of requests | |
2143 | * allocated with any setting of ->nr_requests | |
2144 | */ | |
fd782a4a | 2145 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
082cf69e | 2146 | goto out; |
fd782a4a | 2147 | |
1da177e4 LT |
2148 | rl->count[rw]++; |
2149 | rl->starved[rw] = 0; | |
cb98fc8b | 2150 | |
64521d1a | 2151 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
2152 | if (priv) |
2153 | rl->elvpriv++; | |
2154 | ||
1da177e4 LT |
2155 | spin_unlock_irq(q->queue_lock); |
2156 | ||
7749a8d4 | 2157 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 2158 | if (unlikely(!rq)) { |
1da177e4 LT |
2159 | /* |
2160 | * Allocation failed presumably due to memory. Undo anything | |
2161 | * we might have messed up. | |
2162 | * | |
2163 | * Allocating task should really be put onto the front of the | |
2164 | * wait queue, but this is pretty rare. | |
2165 | */ | |
2166 | spin_lock_irq(q->queue_lock); | |
cb98fc8b | 2167 | freed_request(q, rw, priv); |
1da177e4 LT |
2168 | |
2169 | /* | |
2170 | * in the very unlikely event that allocation failed and no | |
2171 | * requests for this direction was pending, mark us starved | |
2172 | * so that freeing of a request in the other direction will | |
2173 | * notice us. another possible fix would be to split the | |
2174 | * rq mempool into READ and WRITE | |
2175 | */ | |
2176 | rq_starved: | |
2177 | if (unlikely(rl->count[rw] == 0)) | |
2178 | rl->starved[rw] = 1; | |
2179 | ||
1da177e4 LT |
2180 | goto out; |
2181 | } | |
2182 | ||
88ee5ef1 JA |
2183 | /* |
2184 | * ioc may be NULL here, and ioc_batching will be false. That's | |
2185 | * OK, if the queue is under the request limit then requests need | |
2186 | * not count toward the nr_batch_requests limit. There will always | |
2187 | * be some limit enforced by BLK_BATCH_TIME. | |
2188 | */ | |
1da177e4 LT |
2189 | if (ioc_batching(q, ioc)) |
2190 | ioc->nr_batch_requests--; | |
2191 | ||
2192 | rq_init(q, rq); | |
2056a782 JA |
2193 | |
2194 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); | |
1da177e4 | 2195 | out: |
1da177e4 LT |
2196 | return rq; |
2197 | } | |
2198 | ||
2199 | /* | |
2200 | * No available requests for this queue, unplug the device and wait for some | |
2201 | * requests to become available. | |
d6344532 NP |
2202 | * |
2203 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 2204 | */ |
165125e1 | 2205 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 2206 | struct bio *bio) |
1da177e4 | 2207 | { |
7749a8d4 | 2208 | const int rw = rw_flags & 0x01; |
1da177e4 LT |
2209 | struct request *rq; |
2210 | ||
7749a8d4 | 2211 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
2212 | while (!rq) { |
2213 | DEFINE_WAIT(wait); | |
1da177e4 LT |
2214 | struct request_list *rl = &q->rq; |
2215 | ||
2216 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | |
2217 | TASK_UNINTERRUPTIBLE); | |
2218 | ||
7749a8d4 | 2219 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
1da177e4 LT |
2220 | |
2221 | if (!rq) { | |
2222 | struct io_context *ioc; | |
2223 | ||
2056a782 JA |
2224 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
2225 | ||
d6344532 NP |
2226 | __generic_unplug_device(q); |
2227 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
2228 | io_schedule(); |
2229 | ||
2230 | /* | |
2231 | * After sleeping, we become a "batching" process and | |
2232 | * will be able to allocate at least one request, and | |
2233 | * up to a big batch of them for a small period time. | |
2234 | * See ioc_batching, ioc_set_batching | |
2235 | */ | |
b5deef90 | 2236 | ioc = current_io_context(GFP_NOIO, q->node); |
1da177e4 | 2237 | ioc_set_batching(q, ioc); |
d6344532 NP |
2238 | |
2239 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
2240 | } |
2241 | finish_wait(&rl->wait[rw], &wait); | |
450991bc | 2242 | } |
1da177e4 LT |
2243 | |
2244 | return rq; | |
2245 | } | |
2246 | ||
165125e1 | 2247 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
2248 | { |
2249 | struct request *rq; | |
2250 | ||
2251 | BUG_ON(rw != READ && rw != WRITE); | |
2252 | ||
d6344532 NP |
2253 | spin_lock_irq(q->queue_lock); |
2254 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 2255 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 2256 | } else { |
22e2c507 | 2257 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
2258 | if (!rq) |
2259 | spin_unlock_irq(q->queue_lock); | |
2260 | } | |
2261 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
2262 | |
2263 | return rq; | |
2264 | } | |
1da177e4 LT |
2265 | EXPORT_SYMBOL(blk_get_request); |
2266 | ||
dc72ef4a JA |
2267 | /** |
2268 | * blk_start_queueing - initiate dispatch of requests to device | |
2269 | * @q: request queue to kick into gear | |
2270 | * | |
2271 | * This is basically a helper to remove the need to know whether a queue | |
2272 | * is plugged or not if someone just wants to initiate dispatch of requests | |
2273 | * for this queue. | |
2274 | * | |
2275 | * The queue lock must be held with interrupts disabled. | |
2276 | */ | |
165125e1 | 2277 | void blk_start_queueing(struct request_queue *q) |
dc72ef4a JA |
2278 | { |
2279 | if (!blk_queue_plugged(q)) | |
2280 | q->request_fn(q); | |
2281 | else | |
2282 | __generic_unplug_device(q); | |
2283 | } | |
2284 | EXPORT_SYMBOL(blk_start_queueing); | |
2285 | ||
1da177e4 LT |
2286 | /** |
2287 | * blk_requeue_request - put a request back on queue | |
2288 | * @q: request queue where request should be inserted | |
2289 | * @rq: request to be inserted | |
2290 | * | |
2291 | * Description: | |
2292 | * Drivers often keep queueing requests until the hardware cannot accept | |
2293 | * more, when that condition happens we need to put the request back | |
2294 | * on the queue. Must be called with queue lock held. | |
2295 | */ | |
165125e1 | 2296 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 2297 | { |
2056a782 JA |
2298 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
2299 | ||
1da177e4 LT |
2300 | if (blk_rq_tagged(rq)) |
2301 | blk_queue_end_tag(q, rq); | |
2302 | ||
2303 | elv_requeue_request(q, rq); | |
2304 | } | |
2305 | ||
2306 | EXPORT_SYMBOL(blk_requeue_request); | |
2307 | ||
2308 | /** | |
2309 | * blk_insert_request - insert a special request in to a request queue | |
2310 | * @q: request queue where request should be inserted | |
2311 | * @rq: request to be inserted | |
2312 | * @at_head: insert request at head or tail of queue | |
2313 | * @data: private data | |
1da177e4 LT |
2314 | * |
2315 | * Description: | |
2316 | * Many block devices need to execute commands asynchronously, so they don't | |
2317 | * block the whole kernel from preemption during request execution. This is | |
2318 | * accomplished normally by inserting aritficial requests tagged as | |
2319 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | |
2320 | * scheduled for actual execution by the request queue. | |
2321 | * | |
2322 | * We have the option of inserting the head or the tail of the queue. | |
2323 | * Typically we use the tail for new ioctls and so forth. We use the head | |
2324 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
2325 | * host that is unable to accept a particular command. | |
2326 | */ | |
165125e1 | 2327 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 2328 | int at_head, void *data) |
1da177e4 | 2329 | { |
867d1191 | 2330 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
2331 | unsigned long flags; |
2332 | ||
2333 | /* | |
2334 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
2335 | * must not attempt merges on this) and that it acts as a soft | |
2336 | * barrier | |
2337 | */ | |
4aff5e23 JA |
2338 | rq->cmd_type = REQ_TYPE_SPECIAL; |
2339 | rq->cmd_flags |= REQ_SOFTBARRIER; | |
1da177e4 LT |
2340 | |
2341 | rq->special = data; | |
2342 | ||
2343 | spin_lock_irqsave(q->queue_lock, flags); | |
2344 | ||
2345 | /* | |
2346 | * If command is tagged, release the tag | |
2347 | */ | |
867d1191 TH |
2348 | if (blk_rq_tagged(rq)) |
2349 | blk_queue_end_tag(q, rq); | |
1da177e4 | 2350 | |
b238b3d4 | 2351 | drive_stat_acct(rq, 1); |
867d1191 | 2352 | __elv_add_request(q, rq, where, 0); |
dc72ef4a | 2353 | blk_start_queueing(q); |
1da177e4 LT |
2354 | spin_unlock_irqrestore(q->queue_lock, flags); |
2355 | } | |
2356 | ||
2357 | EXPORT_SYMBOL(blk_insert_request); | |
2358 | ||
0e75f906 MC |
2359 | static int __blk_rq_unmap_user(struct bio *bio) |
2360 | { | |
2361 | int ret = 0; | |
2362 | ||
2363 | if (bio) { | |
2364 | if (bio_flagged(bio, BIO_USER_MAPPED)) | |
2365 | bio_unmap_user(bio); | |
2366 | else | |
2367 | ret = bio_uncopy_user(bio); | |
2368 | } | |
2369 | ||
2370 | return ret; | |
2371 | } | |
2372 | ||
3001ca77 N |
2373 | int blk_rq_append_bio(struct request_queue *q, struct request *rq, |
2374 | struct bio *bio) | |
2375 | { | |
2376 | if (!rq->bio) | |
2377 | blk_rq_bio_prep(q, rq, bio); | |
2378 | else if (!ll_back_merge_fn(q, rq, bio)) | |
2379 | return -EINVAL; | |
2380 | else { | |
2381 | rq->biotail->bi_next = bio; | |
2382 | rq->biotail = bio; | |
2383 | ||
2384 | rq->data_len += bio->bi_size; | |
2385 | } | |
2386 | return 0; | |
2387 | } | |
2388 | EXPORT_SYMBOL(blk_rq_append_bio); | |
2389 | ||
165125e1 | 2390 | static int __blk_rq_map_user(struct request_queue *q, struct request *rq, |
0e75f906 MC |
2391 | void __user *ubuf, unsigned int len) |
2392 | { | |
2393 | unsigned long uaddr; | |
2394 | struct bio *bio, *orig_bio; | |
2395 | int reading, ret; | |
2396 | ||
2397 | reading = rq_data_dir(rq) == READ; | |
2398 | ||
2399 | /* | |
2400 | * if alignment requirement is satisfied, map in user pages for | |
2401 | * direct dma. else, set up kernel bounce buffers | |
2402 | */ | |
2403 | uaddr = (unsigned long) ubuf; | |
2404 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | |
2405 | bio = bio_map_user(q, NULL, uaddr, len, reading); | |
2406 | else | |
2407 | bio = bio_copy_user(q, uaddr, len, reading); | |
2408 | ||
2985259b | 2409 | if (IS_ERR(bio)) |
0e75f906 | 2410 | return PTR_ERR(bio); |
0e75f906 MC |
2411 | |
2412 | orig_bio = bio; | |
2413 | blk_queue_bounce(q, &bio); | |
2985259b | 2414 | |
0e75f906 MC |
2415 | /* |
2416 | * We link the bounce buffer in and could have to traverse it | |
2417 | * later so we have to get a ref to prevent it from being freed | |
2418 | */ | |
2419 | bio_get(bio); | |
2420 | ||
3001ca77 N |
2421 | ret = blk_rq_append_bio(q, rq, bio); |
2422 | if (!ret) | |
2423 | return bio->bi_size; | |
0e75f906 | 2424 | |
0e75f906 | 2425 | /* if it was boucned we must call the end io function */ |
6712ecf8 | 2426 | bio_endio(bio, 0); |
0e75f906 MC |
2427 | __blk_rq_unmap_user(orig_bio); |
2428 | bio_put(bio); | |
2429 | return ret; | |
2430 | } | |
2431 | ||
1da177e4 LT |
2432 | /** |
2433 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | |
2434 | * @q: request queue where request should be inserted | |
73747aed | 2435 | * @rq: request structure to fill |
1da177e4 LT |
2436 | * @ubuf: the user buffer |
2437 | * @len: length of user data | |
2438 | * | |
2439 | * Description: | |
2440 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2441 | * a kernel bounce buffer is used. | |
2442 | * | |
2443 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2444 | * still in process context. | |
2445 | * | |
2446 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2447 | * before being submitted to the device, as pages mapped may be out of | |
2448 | * reach. It's the callers responsibility to make sure this happens. The | |
2449 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2450 | * unmapping. | |
2451 | */ | |
165125e1 JA |
2452 | int blk_rq_map_user(struct request_queue *q, struct request *rq, |
2453 | void __user *ubuf, unsigned long len) | |
1da177e4 | 2454 | { |
0e75f906 | 2455 | unsigned long bytes_read = 0; |
8e5cfc45 | 2456 | struct bio *bio = NULL; |
0e75f906 | 2457 | int ret; |
1da177e4 | 2458 | |
defd94b7 | 2459 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2460 | return -EINVAL; |
2461 | if (!len || !ubuf) | |
2462 | return -EINVAL; | |
1da177e4 | 2463 | |
0e75f906 MC |
2464 | while (bytes_read != len) { |
2465 | unsigned long map_len, end, start; | |
1da177e4 | 2466 | |
0e75f906 MC |
2467 | map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE); |
2468 | end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1) | |
2469 | >> PAGE_SHIFT; | |
2470 | start = (unsigned long)ubuf >> PAGE_SHIFT; | |
1da177e4 | 2471 | |
0e75f906 MC |
2472 | /* |
2473 | * A bad offset could cause us to require BIO_MAX_PAGES + 1 | |
2474 | * pages. If this happens we just lower the requested | |
2475 | * mapping len by a page so that we can fit | |
2476 | */ | |
2477 | if (end - start > BIO_MAX_PAGES) | |
2478 | map_len -= PAGE_SIZE; | |
1da177e4 | 2479 | |
0e75f906 MC |
2480 | ret = __blk_rq_map_user(q, rq, ubuf, map_len); |
2481 | if (ret < 0) | |
2482 | goto unmap_rq; | |
8e5cfc45 JA |
2483 | if (!bio) |
2484 | bio = rq->bio; | |
0e75f906 MC |
2485 | bytes_read += ret; |
2486 | ubuf += ret; | |
1da177e4 LT |
2487 | } |
2488 | ||
0e75f906 MC |
2489 | rq->buffer = rq->data = NULL; |
2490 | return 0; | |
2491 | unmap_rq: | |
8e5cfc45 | 2492 | blk_rq_unmap_user(bio); |
0e75f906 | 2493 | return ret; |
1da177e4 LT |
2494 | } |
2495 | ||
2496 | EXPORT_SYMBOL(blk_rq_map_user); | |
2497 | ||
f1970baf JB |
2498 | /** |
2499 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | |
2500 | * @q: request queue where request should be inserted | |
2501 | * @rq: request to map data to | |
2502 | * @iov: pointer to the iovec | |
2503 | * @iov_count: number of elements in the iovec | |
af9997e4 | 2504 | * @len: I/O byte count |
f1970baf JB |
2505 | * |
2506 | * Description: | |
2507 | * Data will be mapped directly for zero copy io, if possible. Otherwise | |
2508 | * a kernel bounce buffer is used. | |
2509 | * | |
2510 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | |
2511 | * still in process context. | |
2512 | * | |
2513 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | |
2514 | * before being submitted to the device, as pages mapped may be out of | |
2515 | * reach. It's the callers responsibility to make sure this happens. The | |
2516 | * original bio must be passed back in to blk_rq_unmap_user() for proper | |
2517 | * unmapping. | |
2518 | */ | |
165125e1 | 2519 | int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, |
0e75f906 | 2520 | struct sg_iovec *iov, int iov_count, unsigned int len) |
f1970baf JB |
2521 | { |
2522 | struct bio *bio; | |
2523 | ||
2524 | if (!iov || iov_count <= 0) | |
2525 | return -EINVAL; | |
2526 | ||
2527 | /* we don't allow misaligned data like bio_map_user() does. If the | |
2528 | * user is using sg, they're expected to know the alignment constraints | |
2529 | * and respect them accordingly */ | |
2530 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | |
2531 | if (IS_ERR(bio)) | |
2532 | return PTR_ERR(bio); | |
2533 | ||
0e75f906 | 2534 | if (bio->bi_size != len) { |
6712ecf8 | 2535 | bio_endio(bio, 0); |
0e75f906 MC |
2536 | bio_unmap_user(bio); |
2537 | return -EINVAL; | |
2538 | } | |
2539 | ||
2540 | bio_get(bio); | |
f1970baf JB |
2541 | blk_rq_bio_prep(q, rq, bio); |
2542 | rq->buffer = rq->data = NULL; | |
f1970baf JB |
2543 | return 0; |
2544 | } | |
2545 | ||
2546 | EXPORT_SYMBOL(blk_rq_map_user_iov); | |
2547 | ||
1da177e4 LT |
2548 | /** |
2549 | * blk_rq_unmap_user - unmap a request with user data | |
8e5cfc45 | 2550 | * @bio: start of bio list |
1da177e4 LT |
2551 | * |
2552 | * Description: | |
8e5cfc45 JA |
2553 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must |
2554 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
2555 | * the io completion may have changed rq->bio. | |
1da177e4 | 2556 | */ |
8e5cfc45 | 2557 | int blk_rq_unmap_user(struct bio *bio) |
1da177e4 | 2558 | { |
8e5cfc45 | 2559 | struct bio *mapped_bio; |
48785bb9 | 2560 | int ret = 0, ret2; |
1da177e4 | 2561 | |
8e5cfc45 JA |
2562 | while (bio) { |
2563 | mapped_bio = bio; | |
2564 | if (unlikely(bio_flagged(bio, BIO_BOUNCED))) | |
0e75f906 | 2565 | mapped_bio = bio->bi_private; |
1da177e4 | 2566 | |
48785bb9 JA |
2567 | ret2 = __blk_rq_unmap_user(mapped_bio); |
2568 | if (ret2 && !ret) | |
2569 | ret = ret2; | |
2570 | ||
8e5cfc45 JA |
2571 | mapped_bio = bio; |
2572 | bio = bio->bi_next; | |
2573 | bio_put(mapped_bio); | |
0e75f906 | 2574 | } |
48785bb9 JA |
2575 | |
2576 | return ret; | |
1da177e4 LT |
2577 | } |
2578 | ||
2579 | EXPORT_SYMBOL(blk_rq_unmap_user); | |
2580 | ||
df46b9a4 MC |
2581 | /** |
2582 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | |
2583 | * @q: request queue where request should be inserted | |
73747aed | 2584 | * @rq: request to fill |
df46b9a4 MC |
2585 | * @kbuf: the kernel buffer |
2586 | * @len: length of user data | |
73747aed | 2587 | * @gfp_mask: memory allocation flags |
df46b9a4 | 2588 | */ |
165125e1 | 2589 | int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf, |
8267e268 | 2590 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 | 2591 | { |
df46b9a4 MC |
2592 | struct bio *bio; |
2593 | ||
defd94b7 | 2594 | if (len > (q->max_hw_sectors << 9)) |
dd1cab95 JA |
2595 | return -EINVAL; |
2596 | if (!len || !kbuf) | |
2597 | return -EINVAL; | |
df46b9a4 MC |
2598 | |
2599 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
dd1cab95 JA |
2600 | if (IS_ERR(bio)) |
2601 | return PTR_ERR(bio); | |
df46b9a4 | 2602 | |
dd1cab95 JA |
2603 | if (rq_data_dir(rq) == WRITE) |
2604 | bio->bi_rw |= (1 << BIO_RW); | |
df46b9a4 | 2605 | |
dd1cab95 | 2606 | blk_rq_bio_prep(q, rq, bio); |
821de3a2 | 2607 | blk_queue_bounce(q, &rq->bio); |
dd1cab95 | 2608 | rq->buffer = rq->data = NULL; |
dd1cab95 | 2609 | return 0; |
df46b9a4 MC |
2610 | } |
2611 | ||
2612 | EXPORT_SYMBOL(blk_rq_map_kern); | |
2613 | ||
73747aed CH |
2614 | /** |
2615 | * blk_execute_rq_nowait - insert a request into queue for execution | |
2616 | * @q: queue to insert the request in | |
2617 | * @bd_disk: matching gendisk | |
2618 | * @rq: request to insert | |
2619 | * @at_head: insert request at head or tail of queue | |
2620 | * @done: I/O completion handler | |
2621 | * | |
2622 | * Description: | |
2623 | * Insert a fully prepared request at the back of the io scheduler queue | |
2624 | * for execution. Don't wait for completion. | |
2625 | */ | |
165125e1 | 2626 | void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk, |
f1970baf | 2627 | struct request *rq, int at_head, |
8ffdc655 | 2628 | rq_end_io_fn *done) |
f1970baf JB |
2629 | { |
2630 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | |
2631 | ||
2632 | rq->rq_disk = bd_disk; | |
4aff5e23 | 2633 | rq->cmd_flags |= REQ_NOMERGE; |
f1970baf | 2634 | rq->end_io = done; |
4c5d0bbd AM |
2635 | WARN_ON(irqs_disabled()); |
2636 | spin_lock_irq(q->queue_lock); | |
2637 | __elv_add_request(q, rq, where, 1); | |
2638 | __generic_unplug_device(q); | |
2639 | spin_unlock_irq(q->queue_lock); | |
f1970baf | 2640 | } |
6e39b69e MC |
2641 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
2642 | ||
1da177e4 LT |
2643 | /** |
2644 | * blk_execute_rq - insert a request into queue for execution | |
2645 | * @q: queue to insert the request in | |
2646 | * @bd_disk: matching gendisk | |
2647 | * @rq: request to insert | |
994ca9a1 | 2648 | * @at_head: insert request at head or tail of queue |
1da177e4 LT |
2649 | * |
2650 | * Description: | |
2651 | * Insert a fully prepared request at the back of the io scheduler queue | |
73747aed | 2652 | * for execution and wait for completion. |
1da177e4 | 2653 | */ |
165125e1 | 2654 | int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk, |
994ca9a1 | 2655 | struct request *rq, int at_head) |
1da177e4 | 2656 | { |
60be6b9a | 2657 | DECLARE_COMPLETION_ONSTACK(wait); |
1da177e4 LT |
2658 | char sense[SCSI_SENSE_BUFFERSIZE]; |
2659 | int err = 0; | |
2660 | ||
1da177e4 LT |
2661 | /* |
2662 | * we need an extra reference to the request, so we can look at | |
2663 | * it after io completion | |
2664 | */ | |
2665 | rq->ref_count++; | |
2666 | ||
2667 | if (!rq->sense) { | |
2668 | memset(sense, 0, sizeof(sense)); | |
2669 | rq->sense = sense; | |
2670 | rq->sense_len = 0; | |
2671 | } | |
2672 | ||
c00895ab | 2673 | rq->end_io_data = &wait; |
994ca9a1 | 2674 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
1da177e4 | 2675 | wait_for_completion(&wait); |
1da177e4 LT |
2676 | |
2677 | if (rq->errors) | |
2678 | err = -EIO; | |
2679 | ||
2680 | return err; | |
2681 | } | |
2682 | ||
2683 | EXPORT_SYMBOL(blk_execute_rq); | |
2684 | ||
fd5d8062 JA |
2685 | static void bio_end_empty_barrier(struct bio *bio, int err) |
2686 | { | |
2687 | if (err) | |
2688 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
2689 | ||
2690 | complete(bio->bi_private); | |
2691 | } | |
2692 | ||
1da177e4 LT |
2693 | /** |
2694 | * blkdev_issue_flush - queue a flush | |
2695 | * @bdev: blockdev to issue flush for | |
2696 | * @error_sector: error sector | |
2697 | * | |
2698 | * Description: | |
2699 | * Issue a flush for the block device in question. Caller can supply | |
2700 | * room for storing the error offset in case of a flush error, if they | |
2701 | * wish to. Caller must run wait_for_completion() on its own. | |
2702 | */ | |
2703 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | |
2704 | { | |
fd5d8062 | 2705 | DECLARE_COMPLETION_ONSTACK(wait); |
165125e1 | 2706 | struct request_queue *q; |
fd5d8062 JA |
2707 | struct bio *bio; |
2708 | int ret; | |
1da177e4 LT |
2709 | |
2710 | if (bdev->bd_disk == NULL) | |
2711 | return -ENXIO; | |
2712 | ||
2713 | q = bdev_get_queue(bdev); | |
2714 | if (!q) | |
2715 | return -ENXIO; | |
1da177e4 | 2716 | |
fd5d8062 JA |
2717 | bio = bio_alloc(GFP_KERNEL, 0); |
2718 | if (!bio) | |
2719 | return -ENOMEM; | |
2720 | ||
2721 | bio->bi_end_io = bio_end_empty_barrier; | |
2722 | bio->bi_private = &wait; | |
2723 | bio->bi_bdev = bdev; | |
2724 | submit_bio(1 << BIO_RW_BARRIER, bio); | |
2725 | ||
2726 | wait_for_completion(&wait); | |
2727 | ||
2728 | /* | |
2729 | * The driver must store the error location in ->bi_sector, if | |
2730 | * it supports it. For non-stacked drivers, this should be copied | |
2731 | * from rq->sector. | |
2732 | */ | |
2733 | if (error_sector) | |
2734 | *error_sector = bio->bi_sector; | |
2735 | ||
2736 | ret = 0; | |
2737 | if (!bio_flagged(bio, BIO_UPTODATE)) | |
2738 | ret = -EIO; | |
2739 | ||
2740 | bio_put(bio); | |
2741 | return ret; | |
1da177e4 LT |
2742 | } |
2743 | ||
2744 | EXPORT_SYMBOL(blkdev_issue_flush); | |
2745 | ||
b238b3d4 | 2746 | static void drive_stat_acct(struct request *rq, int new_io) |
1da177e4 LT |
2747 | { |
2748 | int rw = rq_data_dir(rq); | |
2749 | ||
2750 | if (!blk_fs_request(rq) || !rq->rq_disk) | |
2751 | return; | |
2752 | ||
d72d904a | 2753 | if (!new_io) { |
a362357b | 2754 | __disk_stat_inc(rq->rq_disk, merges[rw]); |
d72d904a | 2755 | } else { |
1da177e4 LT |
2756 | disk_round_stats(rq->rq_disk); |
2757 | rq->rq_disk->in_flight++; | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | /* | |
2762 | * add-request adds a request to the linked list. | |
2763 | * queue lock is held and interrupts disabled, as we muck with the | |
2764 | * request queue list. | |
2765 | */ | |
165125e1 | 2766 | static inline void add_request(struct request_queue * q, struct request * req) |
1da177e4 | 2767 | { |
b238b3d4 | 2768 | drive_stat_acct(req, 1); |
1da177e4 | 2769 | |
1da177e4 LT |
2770 | /* |
2771 | * elevator indicated where it wants this request to be | |
2772 | * inserted at elevator_merge time | |
2773 | */ | |
2774 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
2775 | } | |
2776 | ||
2777 | /* | |
2778 | * disk_round_stats() - Round off the performance stats on a struct | |
2779 | * disk_stats. | |
2780 | * | |
2781 | * The average IO queue length and utilisation statistics are maintained | |
2782 | * by observing the current state of the queue length and the amount of | |
2783 | * time it has been in this state for. | |
2784 | * | |
2785 | * Normally, that accounting is done on IO completion, but that can result | |
2786 | * in more than a second's worth of IO being accounted for within any one | |
2787 | * second, leading to >100% utilisation. To deal with that, we call this | |
2788 | * function to do a round-off before returning the results when reading | |
2789 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
2790 | * the current jiffies and restarts the counters again. | |
2791 | */ | |
2792 | void disk_round_stats(struct gendisk *disk) | |
2793 | { | |
2794 | unsigned long now = jiffies; | |
2795 | ||
b2982649 KC |
2796 | if (now == disk->stamp) |
2797 | return; | |
1da177e4 | 2798 | |
20e5c81f KC |
2799 | if (disk->in_flight) { |
2800 | __disk_stat_add(disk, time_in_queue, | |
2801 | disk->in_flight * (now - disk->stamp)); | |
2802 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | |
2803 | } | |
1da177e4 | 2804 | disk->stamp = now; |
1da177e4 LT |
2805 | } |
2806 | ||
3eaf840e JNN |
2807 | EXPORT_SYMBOL_GPL(disk_round_stats); |
2808 | ||
1da177e4 LT |
2809 | /* |
2810 | * queue lock must be held | |
2811 | */ | |
165125e1 | 2812 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 2813 | { |
1da177e4 LT |
2814 | if (unlikely(!q)) |
2815 | return; | |
2816 | if (unlikely(--req->ref_count)) | |
2817 | return; | |
2818 | ||
8922e16c TH |
2819 | elv_completed_request(q, req); |
2820 | ||
1da177e4 LT |
2821 | /* |
2822 | * Request may not have originated from ll_rw_blk. if not, | |
2823 | * it didn't come out of our reserved rq pools | |
2824 | */ | |
49171e5c | 2825 | if (req->cmd_flags & REQ_ALLOCED) { |
1da177e4 | 2826 | int rw = rq_data_dir(req); |
4aff5e23 | 2827 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 2828 | |
1da177e4 | 2829 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 2830 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
2831 | |
2832 | blk_free_request(q, req); | |
cb98fc8b | 2833 | freed_request(q, rw, priv); |
1da177e4 LT |
2834 | } |
2835 | } | |
2836 | ||
6e39b69e MC |
2837 | EXPORT_SYMBOL_GPL(__blk_put_request); |
2838 | ||
1da177e4 LT |
2839 | void blk_put_request(struct request *req) |
2840 | { | |
8922e16c | 2841 | unsigned long flags; |
165125e1 | 2842 | struct request_queue *q = req->q; |
8922e16c | 2843 | |
1da177e4 | 2844 | /* |
8922e16c TH |
2845 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
2846 | * following if (q) test. | |
1da177e4 | 2847 | */ |
8922e16c | 2848 | if (q) { |
1da177e4 LT |
2849 | spin_lock_irqsave(q->queue_lock, flags); |
2850 | __blk_put_request(q, req); | |
2851 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2852 | } | |
2853 | } | |
2854 | ||
2855 | EXPORT_SYMBOL(blk_put_request); | |
2856 | ||
2857 | /** | |
2858 | * blk_end_sync_rq - executes a completion event on a request | |
2859 | * @rq: request to complete | |
fddfdeaf | 2860 | * @error: end io status of the request |
1da177e4 | 2861 | */ |
8ffdc655 | 2862 | void blk_end_sync_rq(struct request *rq, int error) |
1da177e4 | 2863 | { |
c00895ab | 2864 | struct completion *waiting = rq->end_io_data; |
1da177e4 | 2865 | |
c00895ab | 2866 | rq->end_io_data = NULL; |
1da177e4 LT |
2867 | __blk_put_request(rq->q, rq); |
2868 | ||
2869 | /* | |
2870 | * complete last, if this is a stack request the process (and thus | |
2871 | * the rq pointer) could be invalid right after this complete() | |
2872 | */ | |
2873 | complete(waiting); | |
2874 | } | |
2875 | EXPORT_SYMBOL(blk_end_sync_rq); | |
2876 | ||
1da177e4 LT |
2877 | /* |
2878 | * Has to be called with the request spinlock acquired | |
2879 | */ | |
165125e1 | 2880 | static int attempt_merge(struct request_queue *q, struct request *req, |
1da177e4 LT |
2881 | struct request *next) |
2882 | { | |
2883 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
2884 | return 0; | |
2885 | ||
2886 | /* | |
d6e05edc | 2887 | * not contiguous |
1da177e4 LT |
2888 | */ |
2889 | if (req->sector + req->nr_sectors != next->sector) | |
2890 | return 0; | |
2891 | ||
2892 | if (rq_data_dir(req) != rq_data_dir(next) | |
2893 | || req->rq_disk != next->rq_disk | |
c00895ab | 2894 | || next->special) |
1da177e4 LT |
2895 | return 0; |
2896 | ||
2897 | /* | |
2898 | * If we are allowed to merge, then append bio list | |
2899 | * from next to rq and release next. merge_requests_fn | |
2900 | * will have updated segment counts, update sector | |
2901 | * counts here. | |
2902 | */ | |
1aa4f24f | 2903 | if (!ll_merge_requests_fn(q, req, next)) |
1da177e4 LT |
2904 | return 0; |
2905 | ||
2906 | /* | |
2907 | * At this point we have either done a back merge | |
2908 | * or front merge. We need the smaller start_time of | |
2909 | * the merged requests to be the current request | |
2910 | * for accounting purposes. | |
2911 | */ | |
2912 | if (time_after(req->start_time, next->start_time)) | |
2913 | req->start_time = next->start_time; | |
2914 | ||
2915 | req->biotail->bi_next = next->bio; | |
2916 | req->biotail = next->biotail; | |
2917 | ||
2918 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | |
2919 | ||
2920 | elv_merge_requests(q, req, next); | |
2921 | ||
2922 | if (req->rq_disk) { | |
2923 | disk_round_stats(req->rq_disk); | |
2924 | req->rq_disk->in_flight--; | |
2925 | } | |
2926 | ||
22e2c507 JA |
2927 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2928 | ||
1da177e4 LT |
2929 | __blk_put_request(q, next); |
2930 | return 1; | |
2931 | } | |
2932 | ||
165125e1 JA |
2933 | static inline int attempt_back_merge(struct request_queue *q, |
2934 | struct request *rq) | |
1da177e4 LT |
2935 | { |
2936 | struct request *next = elv_latter_request(q, rq); | |
2937 | ||
2938 | if (next) | |
2939 | return attempt_merge(q, rq, next); | |
2940 | ||
2941 | return 0; | |
2942 | } | |
2943 | ||
165125e1 JA |
2944 | static inline int attempt_front_merge(struct request_queue *q, |
2945 | struct request *rq) | |
1da177e4 LT |
2946 | { |
2947 | struct request *prev = elv_former_request(q, rq); | |
2948 | ||
2949 | if (prev) | |
2950 | return attempt_merge(q, prev, rq); | |
2951 | ||
2952 | return 0; | |
2953 | } | |
2954 | ||
52d9e675 TH |
2955 | static void init_request_from_bio(struct request *req, struct bio *bio) |
2956 | { | |
4aff5e23 | 2957 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
2958 | |
2959 | /* | |
2960 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
2961 | */ | |
2962 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | |
4aff5e23 | 2963 | req->cmd_flags |= REQ_FAILFAST; |
52d9e675 TH |
2964 | |
2965 | /* | |
2966 | * REQ_BARRIER implies no merging, but lets make it explicit | |
2967 | */ | |
2968 | if (unlikely(bio_barrier(bio))) | |
4aff5e23 | 2969 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
52d9e675 | 2970 | |
b31dc66a | 2971 | if (bio_sync(bio)) |
4aff5e23 | 2972 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
2973 | if (bio_rw_meta(bio)) |
2974 | req->cmd_flags |= REQ_RW_META; | |
b31dc66a | 2975 | |
52d9e675 TH |
2976 | req->errors = 0; |
2977 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 2978 | req->ioprio = bio_prio(bio); |
52d9e675 | 2979 | req->start_time = jiffies; |
bc1c56fd | 2980 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
2981 | } |
2982 | ||
165125e1 | 2983 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 2984 | { |
450991bc | 2985 | struct request *req; |
51da90fc JA |
2986 | int el_ret, nr_sectors, barrier, err; |
2987 | const unsigned short prio = bio_prio(bio); | |
2988 | const int sync = bio_sync(bio); | |
7749a8d4 | 2989 | int rw_flags; |
1da177e4 | 2990 | |
1da177e4 | 2991 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
2992 | |
2993 | /* | |
2994 | * low level driver can indicate that it wants pages above a | |
2995 | * certain limit bounced to low memory (ie for highmem, or even | |
2996 | * ISA dma in theory) | |
2997 | */ | |
2998 | blk_queue_bounce(q, &bio); | |
2999 | ||
1da177e4 | 3000 | barrier = bio_barrier(bio); |
797e7dbb | 3001 | if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { |
1da177e4 LT |
3002 | err = -EOPNOTSUPP; |
3003 | goto end_io; | |
3004 | } | |
3005 | ||
1da177e4 LT |
3006 | spin_lock_irq(q->queue_lock); |
3007 | ||
450991bc | 3008 | if (unlikely(barrier) || elv_queue_empty(q)) |
1da177e4 LT |
3009 | goto get_rq; |
3010 | ||
3011 | el_ret = elv_merge(q, &req, bio); | |
3012 | switch (el_ret) { | |
3013 | case ELEVATOR_BACK_MERGE: | |
3014 | BUG_ON(!rq_mergeable(req)); | |
3015 | ||
1aa4f24f | 3016 | if (!ll_back_merge_fn(q, req, bio)) |
1da177e4 LT |
3017 | break; |
3018 | ||
2056a782 JA |
3019 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
3020 | ||
1da177e4 LT |
3021 | req->biotail->bi_next = bio; |
3022 | req->biotail = bio; | |
3023 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
22e2c507 | 3024 | req->ioprio = ioprio_best(req->ioprio, prio); |
b238b3d4 | 3025 | drive_stat_acct(req, 0); |
1da177e4 | 3026 | if (!attempt_back_merge(q, req)) |
2e662b65 | 3027 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
3028 | goto out; |
3029 | ||
3030 | case ELEVATOR_FRONT_MERGE: | |
3031 | BUG_ON(!rq_mergeable(req)); | |
3032 | ||
1aa4f24f | 3033 | if (!ll_front_merge_fn(q, req, bio)) |
1da177e4 LT |
3034 | break; |
3035 | ||
2056a782 JA |
3036 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
3037 | ||
1da177e4 LT |
3038 | bio->bi_next = req->bio; |
3039 | req->bio = bio; | |
3040 | ||
3041 | /* | |
3042 | * may not be valid. if the low level driver said | |
3043 | * it didn't need a bounce buffer then it better | |
3044 | * not touch req->buffer either... | |
3045 | */ | |
3046 | req->buffer = bio_data(bio); | |
51da90fc JA |
3047 | req->current_nr_sectors = bio_cur_sectors(bio); |
3048 | req->hard_cur_sectors = req->current_nr_sectors; | |
3049 | req->sector = req->hard_sector = bio->bi_sector; | |
1da177e4 | 3050 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
22e2c507 | 3051 | req->ioprio = ioprio_best(req->ioprio, prio); |
b238b3d4 | 3052 | drive_stat_acct(req, 0); |
1da177e4 | 3053 | if (!attempt_front_merge(q, req)) |
2e662b65 | 3054 | elv_merged_request(q, req, el_ret); |
1da177e4 LT |
3055 | goto out; |
3056 | ||
450991bc | 3057 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
1da177e4 | 3058 | default: |
450991bc | 3059 | ; |
1da177e4 LT |
3060 | } |
3061 | ||
450991bc | 3062 | get_rq: |
7749a8d4 JA |
3063 | /* |
3064 | * This sync check and mask will be re-done in init_request_from_bio(), | |
3065 | * but we need to set it earlier to expose the sync flag to the | |
3066 | * rq allocator and io schedulers. | |
3067 | */ | |
3068 | rw_flags = bio_data_dir(bio); | |
3069 | if (sync) | |
3070 | rw_flags |= REQ_RW_SYNC; | |
3071 | ||
1da177e4 | 3072 | /* |
450991bc | 3073 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 3074 | * Returns with the queue unlocked. |
450991bc | 3075 | */ |
7749a8d4 | 3076 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 3077 | |
450991bc NP |
3078 | /* |
3079 | * After dropping the lock and possibly sleeping here, our request | |
3080 | * may now be mergeable after it had proven unmergeable (above). | |
3081 | * We don't worry about that case for efficiency. It won't happen | |
3082 | * often, and the elevators are able to handle it. | |
1da177e4 | 3083 | */ |
52d9e675 | 3084 | init_request_from_bio(req, bio); |
1da177e4 | 3085 | |
450991bc NP |
3086 | spin_lock_irq(q->queue_lock); |
3087 | if (elv_queue_empty(q)) | |
3088 | blk_plug_device(q); | |
1da177e4 LT |
3089 | add_request(q, req); |
3090 | out: | |
4a534f93 | 3091 | if (sync) |
1da177e4 LT |
3092 | __generic_unplug_device(q); |
3093 | ||
3094 | spin_unlock_irq(q->queue_lock); | |
3095 | return 0; | |
3096 | ||
3097 | end_io: | |
6712ecf8 | 3098 | bio_endio(bio, err); |
1da177e4 LT |
3099 | return 0; |
3100 | } | |
3101 | ||
3102 | /* | |
3103 | * If bio->bi_dev is a partition, remap the location | |
3104 | */ | |
3105 | static inline void blk_partition_remap(struct bio *bio) | |
3106 | { | |
3107 | struct block_device *bdev = bio->bi_bdev; | |
3108 | ||
bf2de6f5 | 3109 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 | 3110 | struct hd_struct *p = bdev->bd_part; |
a362357b JA |
3111 | const int rw = bio_data_dir(bio); |
3112 | ||
3113 | p->sectors[rw] += bio_sectors(bio); | |
3114 | p->ios[rw]++; | |
1da177e4 | 3115 | |
1da177e4 LT |
3116 | bio->bi_sector += p->start_sect; |
3117 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b AB |
3118 | |
3119 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, | |
3120 | bdev->bd_dev, bio->bi_sector, | |
3121 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
3122 | } |
3123 | } | |
3124 | ||
1da177e4 LT |
3125 | static void handle_bad_sector(struct bio *bio) |
3126 | { | |
3127 | char b[BDEVNAME_SIZE]; | |
3128 | ||
3129 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
3130 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
3131 | bdevname(bio->bi_bdev, b), | |
3132 | bio->bi_rw, | |
3133 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
3134 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
3135 | ||
3136 | set_bit(BIO_EOF, &bio->bi_flags); | |
3137 | } | |
3138 | ||
c17bb495 AM |
3139 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
3140 | ||
3141 | static DECLARE_FAULT_ATTR(fail_make_request); | |
3142 | ||
3143 | static int __init setup_fail_make_request(char *str) | |
3144 | { | |
3145 | return setup_fault_attr(&fail_make_request, str); | |
3146 | } | |
3147 | __setup("fail_make_request=", setup_fail_make_request); | |
3148 | ||
3149 | static int should_fail_request(struct bio *bio) | |
3150 | { | |
3151 | if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) || | |
3152 | (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail)) | |
3153 | return should_fail(&fail_make_request, bio->bi_size); | |
3154 | ||
3155 | return 0; | |
3156 | } | |
3157 | ||
3158 | static int __init fail_make_request_debugfs(void) | |
3159 | { | |
3160 | return init_fault_attr_dentries(&fail_make_request, | |
3161 | "fail_make_request"); | |
3162 | } | |
3163 | ||
3164 | late_initcall(fail_make_request_debugfs); | |
3165 | ||
3166 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
3167 | ||
3168 | static inline int should_fail_request(struct bio *bio) | |
3169 | { | |
3170 | return 0; | |
3171 | } | |
3172 | ||
3173 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
3174 | ||
c07e2b41 JA |
3175 | /* |
3176 | * Check whether this bio extends beyond the end of the device. | |
3177 | */ | |
3178 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
3179 | { | |
3180 | sector_t maxsector; | |
3181 | ||
3182 | if (!nr_sectors) | |
3183 | return 0; | |
3184 | ||
3185 | /* Test device or partition size, when known. */ | |
3186 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
3187 | if (maxsector) { | |
3188 | sector_t sector = bio->bi_sector; | |
3189 | ||
3190 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
3191 | /* | |
3192 | * This may well happen - the kernel calls bread() | |
3193 | * without checking the size of the device, e.g., when | |
3194 | * mounting a device. | |
3195 | */ | |
3196 | handle_bad_sector(bio); | |
3197 | return 1; | |
3198 | } | |
3199 | } | |
3200 | ||
3201 | return 0; | |
3202 | } | |
3203 | ||
1da177e4 LT |
3204 | /** |
3205 | * generic_make_request: hand a buffer to its device driver for I/O | |
3206 | * @bio: The bio describing the location in memory and on the device. | |
3207 | * | |
3208 | * generic_make_request() is used to make I/O requests of block | |
3209 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
3210 | * to be done. | |
3211 | * | |
3212 | * generic_make_request() does not return any status. The | |
3213 | * success/failure status of the request, along with notification of | |
3214 | * completion, is delivered asynchronously through the bio->bi_end_io | |
3215 | * function described (one day) else where. | |
3216 | * | |
3217 | * The caller of generic_make_request must make sure that bi_io_vec | |
3218 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
3219 | * set to describe the device address, and the | |
3220 | * bi_end_io and optionally bi_private are set to describe how | |
3221 | * completion notification should be signaled. | |
3222 | * | |
3223 | * generic_make_request and the drivers it calls may use bi_next if this | |
3224 | * bio happens to be merged with someone else, and may change bi_dev and | |
3225 | * bi_sector for remaps as it sees fit. So the values of these fields | |
3226 | * should NOT be depended on after the call to generic_make_request. | |
3227 | */ | |
d89d8796 | 3228 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 3229 | { |
165125e1 | 3230 | struct request_queue *q; |
5ddfe969 | 3231 | sector_t old_sector; |
1da177e4 | 3232 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 3233 | dev_t old_dev; |
51fd77bd | 3234 | int err = -EIO; |
1da177e4 LT |
3235 | |
3236 | might_sleep(); | |
1da177e4 | 3237 | |
c07e2b41 JA |
3238 | if (bio_check_eod(bio, nr_sectors)) |
3239 | goto end_io; | |
1da177e4 LT |
3240 | |
3241 | /* | |
3242 | * Resolve the mapping until finished. (drivers are | |
3243 | * still free to implement/resolve their own stacking | |
3244 | * by explicitly returning 0) | |
3245 | * | |
3246 | * NOTE: we don't repeat the blk_size check for each new device. | |
3247 | * Stacking drivers are expected to know what they are doing. | |
3248 | */ | |
5ddfe969 | 3249 | old_sector = -1; |
2056a782 | 3250 | old_dev = 0; |
1da177e4 LT |
3251 | do { |
3252 | char b[BDEVNAME_SIZE]; | |
3253 | ||
3254 | q = bdev_get_queue(bio->bi_bdev); | |
3255 | if (!q) { | |
3256 | printk(KERN_ERR | |
3257 | "generic_make_request: Trying to access " | |
3258 | "nonexistent block-device %s (%Lu)\n", | |
3259 | bdevname(bio->bi_bdev, b), | |
3260 | (long long) bio->bi_sector); | |
3261 | end_io: | |
51fd77bd | 3262 | bio_endio(bio, err); |
1da177e4 LT |
3263 | break; |
3264 | } | |
3265 | ||
4fa253f3 | 3266 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
1da177e4 LT |
3267 | printk("bio too big device %s (%u > %u)\n", |
3268 | bdevname(bio->bi_bdev, b), | |
3269 | bio_sectors(bio), | |
3270 | q->max_hw_sectors); | |
3271 | goto end_io; | |
3272 | } | |
3273 | ||
fde6ad22 | 3274 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
3275 | goto end_io; |
3276 | ||
c17bb495 AM |
3277 | if (should_fail_request(bio)) |
3278 | goto end_io; | |
3279 | ||
1da177e4 LT |
3280 | /* |
3281 | * If this device has partitions, remap block n | |
3282 | * of partition p to block n+start(p) of the disk. | |
3283 | */ | |
3284 | blk_partition_remap(bio); | |
3285 | ||
5ddfe969 | 3286 | if (old_sector != -1) |
4fa253f3 | 3287 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 3288 | old_sector); |
2056a782 JA |
3289 | |
3290 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); | |
3291 | ||
5ddfe969 | 3292 | old_sector = bio->bi_sector; |
2056a782 JA |
3293 | old_dev = bio->bi_bdev->bd_dev; |
3294 | ||
c07e2b41 JA |
3295 | if (bio_check_eod(bio, nr_sectors)) |
3296 | goto end_io; | |
51fd77bd JA |
3297 | if (bio_empty_barrier(bio) && !q->prepare_flush_fn) { |
3298 | err = -EOPNOTSUPP; | |
3299 | goto end_io; | |
3300 | } | |
5ddfe969 | 3301 | |
1da177e4 LT |
3302 | ret = q->make_request_fn(q, bio); |
3303 | } while (ret); | |
3304 | } | |
3305 | ||
d89d8796 NB |
3306 | /* |
3307 | * We only want one ->make_request_fn to be active at a time, | |
3308 | * else stack usage with stacked devices could be a problem. | |
3309 | * So use current->bio_{list,tail} to keep a list of requests | |
3310 | * submited by a make_request_fn function. | |
3311 | * current->bio_tail is also used as a flag to say if | |
3312 | * generic_make_request is currently active in this task or not. | |
3313 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
3314 | * then a make_request is active, and new requests should be added | |
3315 | * at the tail | |
3316 | */ | |
3317 | void generic_make_request(struct bio *bio) | |
3318 | { | |
3319 | if (current->bio_tail) { | |
3320 | /* make_request is active */ | |
3321 | *(current->bio_tail) = bio; | |
3322 | bio->bi_next = NULL; | |
3323 | current->bio_tail = &bio->bi_next; | |
3324 | return; | |
3325 | } | |
3326 | /* following loop may be a bit non-obvious, and so deserves some | |
3327 | * explanation. | |
3328 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
3329 | * ensure that) so we have a list with a single bio. | |
3330 | * We pretend that we have just taken it off a longer list, so | |
3331 | * we assign bio_list to the next (which is NULL) and bio_tail | |
3332 | * to &bio_list, thus initialising the bio_list of new bios to be | |
3333 | * added. __generic_make_request may indeed add some more bios | |
3334 | * through a recursive call to generic_make_request. If it | |
3335 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
3336 | * from the top. In this case we really did just take the bio | |
3337 | * of the top of the list (no pretending) and so fixup bio_list and | |
3338 | * bio_tail or bi_next, and call into __generic_make_request again. | |
3339 | * | |
3340 | * The loop was structured like this to make only one call to | |
3341 | * __generic_make_request (which is important as it is large and | |
3342 | * inlined) and to keep the structure simple. | |
3343 | */ | |
3344 | BUG_ON(bio->bi_next); | |
3345 | do { | |
3346 | current->bio_list = bio->bi_next; | |
3347 | if (bio->bi_next == NULL) | |
3348 | current->bio_tail = ¤t->bio_list; | |
3349 | else | |
3350 | bio->bi_next = NULL; | |
3351 | __generic_make_request(bio); | |
3352 | bio = current->bio_list; | |
3353 | } while (bio); | |
3354 | current->bio_tail = NULL; /* deactivate */ | |
3355 | } | |
3356 | ||
1da177e4 LT |
3357 | EXPORT_SYMBOL(generic_make_request); |
3358 | ||
3359 | /** | |
3360 | * submit_bio: submit a bio to the block device layer for I/O | |
3361 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
3362 | * @bio: The &struct bio which describes the I/O | |
3363 | * | |
3364 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
3365 | * uses that function to do most of the work. Both are fairly rough | |
3366 | * interfaces, @bio must be presetup and ready for I/O. | |
3367 | * | |
3368 | */ | |
3369 | void submit_bio(int rw, struct bio *bio) | |
3370 | { | |
3371 | int count = bio_sectors(bio); | |
3372 | ||
22e2c507 | 3373 | bio->bi_rw |= rw; |
1da177e4 | 3374 | |
bf2de6f5 JA |
3375 | /* |
3376 | * If it's a regular read/write or a barrier with data attached, | |
3377 | * go through the normal accounting stuff before submission. | |
3378 | */ | |
3379 | if (!bio_empty_barrier(bio)) { | |
3380 | ||
3381 | BIO_BUG_ON(!bio->bi_size); | |
3382 | BIO_BUG_ON(!bio->bi_io_vec); | |
3383 | ||
3384 | if (rw & WRITE) { | |
3385 | count_vm_events(PGPGOUT, count); | |
3386 | } else { | |
3387 | task_io_account_read(bio->bi_size); | |
3388 | count_vm_events(PGPGIN, count); | |
3389 | } | |
3390 | ||
3391 | if (unlikely(block_dump)) { | |
3392 | char b[BDEVNAME_SIZE]; | |
3393 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 3394 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
3395 | (rw & WRITE) ? "WRITE" : "READ", |
3396 | (unsigned long long)bio->bi_sector, | |
3397 | bdevname(bio->bi_bdev,b)); | |
3398 | } | |
1da177e4 LT |
3399 | } |
3400 | ||
3401 | generic_make_request(bio); | |
3402 | } | |
3403 | ||
3404 | EXPORT_SYMBOL(submit_bio); | |
3405 | ||
93d17d3d | 3406 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
1da177e4 LT |
3407 | { |
3408 | if (blk_fs_request(rq)) { | |
3409 | rq->hard_sector += nsect; | |
3410 | rq->hard_nr_sectors -= nsect; | |
3411 | ||
3412 | /* | |
3413 | * Move the I/O submission pointers ahead if required. | |
3414 | */ | |
3415 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | |
3416 | (rq->sector <= rq->hard_sector)) { | |
3417 | rq->sector = rq->hard_sector; | |
3418 | rq->nr_sectors = rq->hard_nr_sectors; | |
3419 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | |
3420 | rq->current_nr_sectors = rq->hard_cur_sectors; | |
3421 | rq->buffer = bio_data(rq->bio); | |
3422 | } | |
3423 | ||
3424 | /* | |
3425 | * if total number of sectors is less than the first segment | |
3426 | * size, something has gone terribly wrong | |
3427 | */ | |
3428 | if (rq->nr_sectors < rq->current_nr_sectors) { | |
3429 | printk("blk: request botched\n"); | |
3430 | rq->nr_sectors = rq->current_nr_sectors; | |
3431 | } | |
3432 | } | |
3433 | } | |
3434 | ||
3bcddeac KU |
3435 | /** |
3436 | * __end_that_request_first - end I/O on a request | |
3437 | * @req: the request being processed | |
3438 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | |
3439 | * @nr_bytes: number of bytes to complete | |
3440 | * | |
3441 | * Description: | |
3442 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
3443 | * for the next range of segments (if any) in the cluster. | |
3444 | * | |
3445 | * Return: | |
3446 | * 0 - we are done with this request, call end_that_request_last() | |
3447 | * 1 - still buffers pending for this request | |
3448 | **/ | |
1da177e4 LT |
3449 | static int __end_that_request_first(struct request *req, int uptodate, |
3450 | int nr_bytes) | |
3451 | { | |
3452 | int total_bytes, bio_nbytes, error, next_idx = 0; | |
3453 | struct bio *bio; | |
3454 | ||
2056a782 JA |
3455 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
3456 | ||
1da177e4 LT |
3457 | /* |
3458 | * extend uptodate bool to allow < 0 value to be direct io error | |
3459 | */ | |
3460 | error = 0; | |
3461 | if (end_io_error(uptodate)) | |
3462 | error = !uptodate ? -EIO : uptodate; | |
3463 | ||
3464 | /* | |
3465 | * for a REQ_BLOCK_PC request, we want to carry any eventual | |
3466 | * sense key with us all the way through | |
3467 | */ | |
3468 | if (!blk_pc_request(req)) | |
3469 | req->errors = 0; | |
3470 | ||
3471 | if (!uptodate) { | |
4aff5e23 | 3472 | if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET)) |
1da177e4 LT |
3473 | printk("end_request: I/O error, dev %s, sector %llu\n", |
3474 | req->rq_disk ? req->rq_disk->disk_name : "?", | |
3475 | (unsigned long long)req->sector); | |
3476 | } | |
3477 | ||
d72d904a | 3478 | if (blk_fs_request(req) && req->rq_disk) { |
a362357b JA |
3479 | const int rw = rq_data_dir(req); |
3480 | ||
53e86061 | 3481 | disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
d72d904a JA |
3482 | } |
3483 | ||
1da177e4 LT |
3484 | total_bytes = bio_nbytes = 0; |
3485 | while ((bio = req->bio) != NULL) { | |
3486 | int nbytes; | |
3487 | ||
bf2de6f5 JA |
3488 | /* |
3489 | * For an empty barrier request, the low level driver must | |
3490 | * store a potential error location in ->sector. We pass | |
3491 | * that back up in ->bi_sector. | |
3492 | */ | |
3493 | if (blk_empty_barrier(req)) | |
3494 | bio->bi_sector = req->sector; | |
3495 | ||
1da177e4 LT |
3496 | if (nr_bytes >= bio->bi_size) { |
3497 | req->bio = bio->bi_next; | |
3498 | nbytes = bio->bi_size; | |
5bb23a68 | 3499 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
3500 | next_idx = 0; |
3501 | bio_nbytes = 0; | |
3502 | } else { | |
3503 | int idx = bio->bi_idx + next_idx; | |
3504 | ||
3505 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
3506 | blk_dump_rq_flags(req, "__end_that"); | |
3507 | printk("%s: bio idx %d >= vcnt %d\n", | |
3508 | __FUNCTION__, | |
3509 | bio->bi_idx, bio->bi_vcnt); | |
3510 | break; | |
3511 | } | |
3512 | ||
3513 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
3514 | BIO_BUG_ON(nbytes > bio->bi_size); | |
3515 | ||
3516 | /* | |
3517 | * not a complete bvec done | |
3518 | */ | |
3519 | if (unlikely(nbytes > nr_bytes)) { | |
3520 | bio_nbytes += nr_bytes; | |
3521 | total_bytes += nr_bytes; | |
3522 | break; | |
3523 | } | |
3524 | ||
3525 | /* | |
3526 | * advance to the next vector | |
3527 | */ | |
3528 | next_idx++; | |
3529 | bio_nbytes += nbytes; | |
3530 | } | |
3531 | ||
3532 | total_bytes += nbytes; | |
3533 | nr_bytes -= nbytes; | |
3534 | ||
3535 | if ((bio = req->bio)) { | |
3536 | /* | |
3537 | * end more in this run, or just return 'not-done' | |
3538 | */ | |
3539 | if (unlikely(nr_bytes <= 0)) | |
3540 | break; | |
3541 | } | |
3542 | } | |
3543 | ||
3544 | /* | |
3545 | * completely done | |
3546 | */ | |
3547 | if (!req->bio) | |
3548 | return 0; | |
3549 | ||
3550 | /* | |
3551 | * if the request wasn't completed, update state | |
3552 | */ | |
3553 | if (bio_nbytes) { | |
5bb23a68 | 3554 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
3555 | bio->bi_idx += next_idx; |
3556 | bio_iovec(bio)->bv_offset += nr_bytes; | |
3557 | bio_iovec(bio)->bv_len -= nr_bytes; | |
3558 | } | |
3559 | ||
3560 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
3561 | blk_recalc_rq_segments(req); | |
3562 | return 1; | |
3563 | } | |
3564 | ||
ff856bad JA |
3565 | /* |
3566 | * splice the completion data to a local structure and hand off to | |
3567 | * process_completion_queue() to complete the requests | |
3568 | */ | |
3569 | static void blk_done_softirq(struct softirq_action *h) | |
3570 | { | |
626ab0e6 | 3571 | struct list_head *cpu_list, local_list; |
ff856bad JA |
3572 | |
3573 | local_irq_disable(); | |
3574 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
626ab0e6 | 3575 | list_replace_init(cpu_list, &local_list); |
ff856bad JA |
3576 | local_irq_enable(); |
3577 | ||
3578 | while (!list_empty(&local_list)) { | |
3579 | struct request *rq = list_entry(local_list.next, struct request, donelist); | |
3580 | ||
3581 | list_del_init(&rq->donelist); | |
3582 | rq->q->softirq_done_fn(rq); | |
3583 | } | |
3584 | } | |
3585 | ||
db47d475 | 3586 | static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action, |
ff856bad JA |
3587 | void *hcpu) |
3588 | { | |
3589 | /* | |
3590 | * If a CPU goes away, splice its entries to the current CPU | |
3591 | * and trigger a run of the softirq | |
3592 | */ | |
8bb78442 | 3593 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
ff856bad JA |
3594 | int cpu = (unsigned long) hcpu; |
3595 | ||
3596 | local_irq_disable(); | |
3597 | list_splice_init(&per_cpu(blk_cpu_done, cpu), | |
3598 | &__get_cpu_var(blk_cpu_done)); | |
3599 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3600 | local_irq_enable(); | |
3601 | } | |
3602 | ||
3603 | return NOTIFY_OK; | |
3604 | } | |
3605 | ||
3606 | ||
db47d475 | 3607 | static struct notifier_block blk_cpu_notifier __cpuinitdata = { |
ff856bad JA |
3608 | .notifier_call = blk_cpu_notify, |
3609 | }; | |
3610 | ||
ff856bad JA |
3611 | /** |
3612 | * blk_complete_request - end I/O on a request | |
3613 | * @req: the request being processed | |
3614 | * | |
3615 | * Description: | |
3616 | * Ends all I/O on a request. It does not handle partial completions, | |
d6e05edc | 3617 | * unless the driver actually implements this in its completion callback |
4fa253f3 | 3618 | * through requeueing. The actual completion happens out-of-order, |
ff856bad JA |
3619 | * through a softirq handler. The user must have registered a completion |
3620 | * callback through blk_queue_softirq_done(). | |
3621 | **/ | |
3622 | ||
3623 | void blk_complete_request(struct request *req) | |
3624 | { | |
3625 | struct list_head *cpu_list; | |
3626 | unsigned long flags; | |
3627 | ||
3628 | BUG_ON(!req->q->softirq_done_fn); | |
3629 | ||
3630 | local_irq_save(flags); | |
3631 | ||
3632 | cpu_list = &__get_cpu_var(blk_cpu_done); | |
3633 | list_add_tail(&req->donelist, cpu_list); | |
3634 | raise_softirq_irqoff(BLOCK_SOFTIRQ); | |
3635 | ||
3636 | local_irq_restore(flags); | |
3637 | } | |
3638 | ||
3639 | EXPORT_SYMBOL(blk_complete_request); | |
3640 | ||
1da177e4 LT |
3641 | /* |
3642 | * queue lock must be held | |
3643 | */ | |
3bcddeac | 3644 | static void end_that_request_last(struct request *req, int uptodate) |
1da177e4 LT |
3645 | { |
3646 | struct gendisk *disk = req->rq_disk; | |
8ffdc655 TH |
3647 | int error; |
3648 | ||
3649 | /* | |
3650 | * extend uptodate bool to allow < 0 value to be direct io error | |
3651 | */ | |
3652 | error = 0; | |
3653 | if (end_io_error(uptodate)) | |
3654 | error = !uptodate ? -EIO : uptodate; | |
1da177e4 LT |
3655 | |
3656 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
3657 | laptop_io_completion(); | |
3658 | ||
fd0ff8aa JA |
3659 | /* |
3660 | * Account IO completion. bar_rq isn't accounted as a normal | |
3661 | * IO on queueing nor completion. Accounting the containing | |
3662 | * request is enough. | |
3663 | */ | |
3664 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { | |
1da177e4 | 3665 | unsigned long duration = jiffies - req->start_time; |
a362357b JA |
3666 | const int rw = rq_data_dir(req); |
3667 | ||
3668 | __disk_stat_inc(disk, ios[rw]); | |
3669 | __disk_stat_add(disk, ticks[rw], duration); | |
1da177e4 LT |
3670 | disk_round_stats(disk); |
3671 | disk->in_flight--; | |
3672 | } | |
3673 | if (req->end_io) | |
8ffdc655 | 3674 | req->end_io(req, error); |
1da177e4 LT |
3675 | else |
3676 | __blk_put_request(req->q, req); | |
3677 | } | |
3678 | ||
a0cd1285 | 3679 | static inline void __end_request(struct request *rq, int uptodate, |
9e6e39f2 | 3680 | unsigned int nr_bytes) |
1da177e4 | 3681 | { |
9e6e39f2 KU |
3682 | int error = 0; |
3683 | ||
3684 | if (uptodate <= 0) | |
3685 | error = uptodate ? uptodate : -EIO; | |
3686 | ||
3687 | __blk_end_request(rq, error, nr_bytes); | |
1da177e4 LT |
3688 | } |
3689 | ||
3b11313a KU |
3690 | /** |
3691 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
3692 | **/ | |
3693 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
3694 | { |
3695 | if (blk_fs_request(rq)) | |
3696 | return rq->hard_nr_sectors << 9; | |
3697 | ||
3698 | return rq->data_len; | |
3699 | } | |
3b11313a KU |
3700 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
3701 | ||
3702 | /** | |
3703 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
3704 | **/ | |
3705 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
3706 | { | |
3707 | if (blk_fs_request(rq)) | |
3708 | return rq->current_nr_sectors << 9; | |
3709 | ||
3710 | if (rq->bio) | |
3711 | return rq->bio->bi_size; | |
3712 | ||
3713 | return rq->data_len; | |
3714 | } | |
3715 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 JA |
3716 | |
3717 | /** | |
3718 | * end_queued_request - end all I/O on a queued request | |
3719 | * @rq: the request being processed | |
3720 | * @uptodate: error value or 0/1 uptodate flag | |
3721 | * | |
3722 | * Description: | |
3723 | * Ends all I/O on a request, and removes it from the block layer queues. | |
3724 | * Not suitable for normal IO completion, unless the driver still has | |
3725 | * the request attached to the block layer. | |
3726 | * | |
3727 | **/ | |
3728 | void end_queued_request(struct request *rq, int uptodate) | |
3729 | { | |
9e6e39f2 | 3730 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
3731 | } |
3732 | EXPORT_SYMBOL(end_queued_request); | |
3733 | ||
3734 | /** | |
3735 | * end_dequeued_request - end all I/O on a dequeued request | |
3736 | * @rq: the request being processed | |
3737 | * @uptodate: error value or 0/1 uptodate flag | |
3738 | * | |
3739 | * Description: | |
3740 | * Ends all I/O on a request. The request must already have been | |
3741 | * dequeued using blkdev_dequeue_request(), as is normally the case | |
3742 | * for most drivers. | |
3743 | * | |
3744 | **/ | |
3745 | void end_dequeued_request(struct request *rq, int uptodate) | |
3746 | { | |
9e6e39f2 | 3747 | __end_request(rq, uptodate, blk_rq_bytes(rq)); |
a0cd1285 JA |
3748 | } |
3749 | EXPORT_SYMBOL(end_dequeued_request); | |
3750 | ||
3751 | ||
3752 | /** | |
3753 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 3754 | * @req: the request being processed |
a0cd1285 JA |
3755 | * @uptodate: error value or 0/1 uptodate flag |
3756 | * | |
3757 | * Description: | |
3758 | * Ends I/O on the current segment of a request. If that is the only | |
3759 | * remaining segment, the request is also completed and freed. | |
3760 | * | |
3761 | * This is a remnant of how older block drivers handled IO completions. | |
3762 | * Modern drivers typically end IO on the full request in one go, unless | |
3763 | * they have a residual value to account for. For that case this function | |
3764 | * isn't really useful, unless the residual just happens to be the | |
3765 | * full current segment. In other words, don't use this function in new | |
3766 | * code. Either use end_request_completely(), or the | |
3767 | * end_that_request_chunk() (along with end_that_request_last()) for | |
3768 | * partial completions. | |
3769 | * | |
3770 | **/ | |
3771 | void end_request(struct request *req, int uptodate) | |
3772 | { | |
9e6e39f2 | 3773 | __end_request(req, uptodate, req->hard_cur_sectors << 9); |
a0cd1285 | 3774 | } |
1da177e4 LT |
3775 | EXPORT_SYMBOL(end_request); |
3776 | ||
336cdb40 KU |
3777 | static void complete_request(struct request *rq, int error) |
3778 | { | |
3779 | /* | |
3780 | * REMOVEME: This conversion is transitional and will be removed | |
3781 | * when old end_that_request_* are unexported. | |
3782 | */ | |
3783 | int uptodate = 1; | |
3784 | if (error) | |
3785 | uptodate = (error == -EIO) ? 0 : error; | |
3786 | ||
3787 | if (blk_rq_tagged(rq)) | |
3788 | blk_queue_end_tag(rq->q, rq); | |
3789 | ||
3790 | if (blk_queued_rq(rq)) | |
3791 | blkdev_dequeue_request(rq); | |
3792 | ||
e3a04fe3 KU |
3793 | if (blk_bidi_rq(rq) && !rq->end_io) |
3794 | __blk_put_request(rq->next_rq->q, rq->next_rq); | |
3795 | ||
336cdb40 KU |
3796 | end_that_request_last(rq, uptodate); |
3797 | } | |
3798 | ||
3799 | /** | |
e19a3ab0 KU |
3800 | * blk_end_io - Generic end_io function to complete a request. |
3801 | * @rq: the request being processed | |
3802 | * @error: 0 for success, < 0 for error | |
e3a04fe3 KU |
3803 | * @nr_bytes: number of bytes to complete @rq |
3804 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
3805 | * @drv_callback: function called between completion of bios in the request |
3806 | * and completion of the request. | |
3807 | * If the callback returns non 0, this helper returns without | |
3808 | * completion of the request. | |
336cdb40 KU |
3809 | * |
3810 | * Description: | |
e3a04fe3 | 3811 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
3812 | * If @rq has leftover, sets it up for the next range of segments. |
3813 | * | |
3814 | * Return: | |
3815 | * 0 - we are done with this request | |
e19a3ab0 | 3816 | * 1 - this request is not freed yet, it still has pending buffers. |
336cdb40 | 3817 | **/ |
e19a3ab0 | 3818 | static int blk_end_io(struct request *rq, int error, int nr_bytes, |
e3a04fe3 | 3819 | int bidi_bytes, int (drv_callback)(struct request *)) |
336cdb40 KU |
3820 | { |
3821 | struct request_queue *q = rq->q; | |
3822 | unsigned long flags = 0UL; | |
3823 | /* | |
3824 | * REMOVEME: This conversion is transitional and will be removed | |
3825 | * when old end_that_request_* are unexported. | |
3826 | */ | |
3827 | int uptodate = 1; | |
3828 | if (error) | |
3829 | uptodate = (error == -EIO) ? 0 : error; | |
3830 | ||
3831 | if (blk_fs_request(rq) || blk_pc_request(rq)) { | |
3832 | if (__end_that_request_first(rq, uptodate, nr_bytes)) | |
3833 | return 1; | |
e3a04fe3 KU |
3834 | |
3835 | /* Bidi request must be completed as a whole */ | |
3836 | if (blk_bidi_rq(rq) && | |
3837 | __end_that_request_first(rq->next_rq, uptodate, bidi_bytes)) | |
3838 | return 1; | |
336cdb40 KU |
3839 | } |
3840 | ||
e19a3ab0 KU |
3841 | /* Special feature for tricky drivers */ |
3842 | if (drv_callback && drv_callback(rq)) | |
3843 | return 1; | |
3844 | ||
336cdb40 KU |
3845 | add_disk_randomness(rq->rq_disk); |
3846 | ||
3847 | spin_lock_irqsave(q->queue_lock, flags); | |
3848 | complete_request(rq, error); | |
3849 | spin_unlock_irqrestore(q->queue_lock, flags); | |
3850 | ||
3851 | return 0; | |
3852 | } | |
e19a3ab0 KU |
3853 | |
3854 | /** | |
3855 | * blk_end_request - Helper function for drivers to complete the request. | |
3856 | * @rq: the request being processed | |
3857 | * @error: 0 for success, < 0 for error | |
3858 | * @nr_bytes: number of bytes to complete | |
3859 | * | |
3860 | * Description: | |
3861 | * Ends I/O on a number of bytes attached to @rq. | |
3862 | * If @rq has leftover, sets it up for the next range of segments. | |
3863 | * | |
3864 | * Return: | |
3865 | * 0 - we are done with this request | |
3866 | * 1 - still buffers pending for this request | |
3867 | **/ | |
3868 | int blk_end_request(struct request *rq, int error, int nr_bytes) | |
3869 | { | |
e3a04fe3 | 3870 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 3871 | } |
336cdb40 KU |
3872 | EXPORT_SYMBOL_GPL(blk_end_request); |
3873 | ||
3874 | /** | |
3875 | * __blk_end_request - Helper function for drivers to complete the request. | |
3876 | * @rq: the request being processed | |
3877 | * @error: 0 for success, < 0 for error | |
3878 | * @nr_bytes: number of bytes to complete | |
3879 | * | |
3880 | * Description: | |
3881 | * Must be called with queue lock held unlike blk_end_request(). | |
3882 | * | |
3883 | * Return: | |
3884 | * 0 - we are done with this request | |
3885 | * 1 - still buffers pending for this request | |
3886 | **/ | |
3887 | int __blk_end_request(struct request *rq, int error, int nr_bytes) | |
3888 | { | |
3889 | /* | |
3890 | * REMOVEME: This conversion is transitional and will be removed | |
3891 | * when old end_that_request_* are unexported. | |
3892 | */ | |
3893 | int uptodate = 1; | |
3894 | if (error) | |
3895 | uptodate = (error == -EIO) ? 0 : error; | |
3896 | ||
3897 | if (blk_fs_request(rq) || blk_pc_request(rq)) { | |
3898 | if (__end_that_request_first(rq, uptodate, nr_bytes)) | |
3899 | return 1; | |
3900 | } | |
3901 | ||
3902 | add_disk_randomness(rq->rq_disk); | |
3903 | ||
3904 | complete_request(rq, error); | |
3905 | ||
3906 | return 0; | |
3907 | } | |
3908 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
3909 | ||
e3a04fe3 KU |
3910 | /** |
3911 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
3912 | * @rq: the bidi request being processed | |
3913 | * @error: 0 for success, < 0 for error | |
3914 | * @nr_bytes: number of bytes to complete @rq | |
3915 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
3916 | * | |
3917 | * Description: | |
3918 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
3919 | * | |
3920 | * Return: | |
3921 | * 0 - we are done with this request | |
3922 | * 1 - still buffers pending for this request | |
3923 | **/ | |
3924 | int blk_end_bidi_request(struct request *rq, int error, int nr_bytes, | |
3925 | int bidi_bytes) | |
3926 | { | |
3927 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
3928 | } | |
3929 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
3930 | ||
e19a3ab0 KU |
3931 | /** |
3932 | * blk_end_request_callback - Special helper function for tricky drivers | |
3933 | * @rq: the request being processed | |
3934 | * @error: 0 for success, < 0 for error | |
3935 | * @nr_bytes: number of bytes to complete | |
3936 | * @drv_callback: function called between completion of bios in the request | |
3937 | * and completion of the request. | |
3938 | * If the callback returns non 0, this helper returns without | |
3939 | * completion of the request. | |
3940 | * | |
3941 | * Description: | |
3942 | * Ends I/O on a number of bytes attached to @rq. | |
3943 | * If @rq has leftover, sets it up for the next range of segments. | |
3944 | * | |
3945 | * This special helper function is used only for existing tricky drivers. | |
3946 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
3947 | * This interface will be removed when such drivers are rewritten. | |
3948 | * Don't use this interface in other places anymore. | |
3949 | * | |
3950 | * Return: | |
3951 | * 0 - we are done with this request | |
3952 | * 1 - this request is not freed yet. | |
3953 | * this request still has pending buffers or | |
3954 | * the driver doesn't want to finish this request yet. | |
3955 | **/ | |
3956 | int blk_end_request_callback(struct request *rq, int error, int nr_bytes, | |
3957 | int (drv_callback)(struct request *)) | |
3958 | { | |
e3a04fe3 | 3959 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
3960 | } |
3961 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
3962 | ||
66846572 N |
3963 | static void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
3964 | struct bio *bio) | |
1da177e4 | 3965 | { |
4aff5e23 JA |
3966 | /* first two bits are identical in rq->cmd_flags and bio->bi_rw */ |
3967 | rq->cmd_flags |= (bio->bi_rw & 3); | |
1da177e4 LT |
3968 | |
3969 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
3970 | rq->nr_hw_segments = bio_hw_segments(q, bio); | |
3971 | rq->current_nr_sectors = bio_cur_sectors(bio); | |
3972 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
3973 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
3974 | rq->buffer = bio_data(bio); | |
0e75f906 | 3975 | rq->data_len = bio->bi_size; |
1da177e4 LT |
3976 | |
3977 | rq->bio = rq->biotail = bio; | |
1da177e4 | 3978 | |
66846572 N |
3979 | if (bio->bi_bdev) |
3980 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
3981 | } | |
1da177e4 LT |
3982 | |
3983 | int kblockd_schedule_work(struct work_struct *work) | |
3984 | { | |
3985 | return queue_work(kblockd_workqueue, work); | |
3986 | } | |
3987 | ||
3988 | EXPORT_SYMBOL(kblockd_schedule_work); | |
3989 | ||
19a75d83 | 3990 | void kblockd_flush_work(struct work_struct *work) |
1da177e4 | 3991 | { |
28e53bdd | 3992 | cancel_work_sync(work); |
1da177e4 | 3993 | } |
19a75d83 | 3994 | EXPORT_SYMBOL(kblockd_flush_work); |
1da177e4 LT |
3995 | |
3996 | int __init blk_dev_init(void) | |
3997 | { | |
ff856bad JA |
3998 | int i; |
3999 | ||
1da177e4 LT |
4000 | kblockd_workqueue = create_workqueue("kblockd"); |
4001 | if (!kblockd_workqueue) | |
4002 | panic("Failed to create kblockd\n"); | |
4003 | ||
4004 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 4005 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
4006 | |
4007 | requestq_cachep = kmem_cache_create("blkdev_queue", | |
165125e1 | 4008 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
4009 | |
4010 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | |
20c2df83 | 4011 | sizeof(struct io_context), 0, SLAB_PANIC, NULL); |
1da177e4 | 4012 | |
0a945022 | 4013 | for_each_possible_cpu(i) |
ff856bad JA |
4014 | INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); |
4015 | ||
4016 | open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); | |
5a67e4c5 | 4017 | register_hotcpu_notifier(&blk_cpu_notifier); |
ff856bad | 4018 | |
f772b3d9 VT |
4019 | blk_max_low_pfn = max_low_pfn - 1; |
4020 | blk_max_pfn = max_pfn - 1; | |
1da177e4 LT |
4021 | |
4022 | return 0; | |
4023 | } | |
4024 | ||
4025 | /* | |
4026 | * IO Context helper functions | |
4027 | */ | |
4028 | void put_io_context(struct io_context *ioc) | |
4029 | { | |
4030 | if (ioc == NULL) | |
4031 | return; | |
4032 | ||
4033 | BUG_ON(atomic_read(&ioc->refcount) == 0); | |
4034 | ||
4035 | if (atomic_dec_and_test(&ioc->refcount)) { | |
e2d74ac0 JA |
4036 | struct cfq_io_context *cic; |
4037 | ||
334e94de | 4038 | rcu_read_lock(); |
1da177e4 LT |
4039 | if (ioc->aic && ioc->aic->dtor) |
4040 | ioc->aic->dtor(ioc->aic); | |
e2d74ac0 | 4041 | if (ioc->cic_root.rb_node != NULL) { |
7143dd4b JA |
4042 | struct rb_node *n = rb_first(&ioc->cic_root); |
4043 | ||
4044 | cic = rb_entry(n, struct cfq_io_context, rb_node); | |
e2d74ac0 JA |
4045 | cic->dtor(ioc); |
4046 | } | |
334e94de | 4047 | rcu_read_unlock(); |
1da177e4 LT |
4048 | |
4049 | kmem_cache_free(iocontext_cachep, ioc); | |
4050 | } | |
4051 | } | |
4052 | EXPORT_SYMBOL(put_io_context); | |
4053 | ||
4054 | /* Called by the exitting task */ | |
4055 | void exit_io_context(void) | |
4056 | { | |
1da177e4 | 4057 | struct io_context *ioc; |
e2d74ac0 | 4058 | struct cfq_io_context *cic; |
1da177e4 | 4059 | |
22e2c507 | 4060 | task_lock(current); |
1da177e4 LT |
4061 | ioc = current->io_context; |
4062 | current->io_context = NULL; | |
22e2c507 | 4063 | task_unlock(current); |
1da177e4 | 4064 | |
25034d7a | 4065 | ioc->task = NULL; |
1da177e4 LT |
4066 | if (ioc->aic && ioc->aic->exit) |
4067 | ioc->aic->exit(ioc->aic); | |
e2d74ac0 JA |
4068 | if (ioc->cic_root.rb_node != NULL) { |
4069 | cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); | |
4070 | cic->exit(ioc); | |
4071 | } | |
25034d7a | 4072 | |
1da177e4 LT |
4073 | put_io_context(ioc); |
4074 | } | |
4075 | ||
4076 | /* | |
4077 | * If the current task has no IO context then create one and initialise it. | |
fb3cc432 | 4078 | * Otherwise, return its existing IO context. |
1da177e4 | 4079 | * |
fb3cc432 NP |
4080 | * This returned IO context doesn't have a specifically elevated refcount, |
4081 | * but since the current task itself holds a reference, the context can be | |
4082 | * used in general code, so long as it stays within `current` context. | |
1da177e4 | 4083 | */ |
b5deef90 | 4084 | static struct io_context *current_io_context(gfp_t gfp_flags, int node) |
1da177e4 LT |
4085 | { |
4086 | struct task_struct *tsk = current; | |
1da177e4 LT |
4087 | struct io_context *ret; |
4088 | ||
1da177e4 | 4089 | ret = tsk->io_context; |
fb3cc432 NP |
4090 | if (likely(ret)) |
4091 | return ret; | |
1da177e4 | 4092 | |
b5deef90 | 4093 | ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node); |
1da177e4 LT |
4094 | if (ret) { |
4095 | atomic_set(&ret->refcount, 1); | |
22e2c507 | 4096 | ret->task = current; |
fc46379d | 4097 | ret->ioprio_changed = 0; |
1da177e4 LT |
4098 | ret->last_waited = jiffies; /* doesn't matter... */ |
4099 | ret->nr_batch_requests = 0; /* because this is 0 */ | |
4100 | ret->aic = NULL; | |
e2d74ac0 | 4101 | ret->cic_root.rb_node = NULL; |
4e521c27 | 4102 | ret->ioc_data = NULL; |
9f83e45e ON |
4103 | /* make sure set_task_ioprio() sees the settings above */ |
4104 | smp_wmb(); | |
fb3cc432 NP |
4105 | tsk->io_context = ret; |
4106 | } | |
1da177e4 | 4107 | |
fb3cc432 NP |
4108 | return ret; |
4109 | } | |
1da177e4 | 4110 | |
fb3cc432 NP |
4111 | /* |
4112 | * If the current task has no IO context then create one and initialise it. | |
4113 | * If it does have a context, take a ref on it. | |
4114 | * | |
4115 | * This is always called in the context of the task which submitted the I/O. | |
4116 | */ | |
b5deef90 | 4117 | struct io_context *get_io_context(gfp_t gfp_flags, int node) |
fb3cc432 NP |
4118 | { |
4119 | struct io_context *ret; | |
b5deef90 | 4120 | ret = current_io_context(gfp_flags, node); |
fb3cc432 | 4121 | if (likely(ret)) |
1da177e4 | 4122 | atomic_inc(&ret->refcount); |
1da177e4 LT |
4123 | return ret; |
4124 | } | |
4125 | EXPORT_SYMBOL(get_io_context); | |
4126 | ||
4127 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | |
4128 | { | |
4129 | struct io_context *src = *psrc; | |
4130 | struct io_context *dst = *pdst; | |
4131 | ||
4132 | if (src) { | |
4133 | BUG_ON(atomic_read(&src->refcount) == 0); | |
4134 | atomic_inc(&src->refcount); | |
4135 | put_io_context(dst); | |
4136 | *pdst = src; | |
4137 | } | |
4138 | } | |
4139 | EXPORT_SYMBOL(copy_io_context); | |
4140 | ||
4141 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | |
4142 | { | |
4143 | struct io_context *temp; | |
4144 | temp = *ioc1; | |
4145 | *ioc1 = *ioc2; | |
4146 | *ioc2 = temp; | |
4147 | } | |
4148 | EXPORT_SYMBOL(swap_io_context); | |
4149 | ||
4150 | /* | |
4151 | * sysfs parts below | |
4152 | */ | |
4153 | struct queue_sysfs_entry { | |
4154 | struct attribute attr; | |
4155 | ssize_t (*show)(struct request_queue *, char *); | |
4156 | ssize_t (*store)(struct request_queue *, const char *, size_t); | |
4157 | }; | |
4158 | ||
4159 | static ssize_t | |
4160 | queue_var_show(unsigned int var, char *page) | |
4161 | { | |
4162 | return sprintf(page, "%d\n", var); | |
4163 | } | |
4164 | ||
4165 | static ssize_t | |
4166 | queue_var_store(unsigned long *var, const char *page, size_t count) | |
4167 | { | |
4168 | char *p = (char *) page; | |
4169 | ||
4170 | *var = simple_strtoul(p, &p, 10); | |
4171 | return count; | |
4172 | } | |
4173 | ||
4174 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | |
4175 | { | |
4176 | return queue_var_show(q->nr_requests, (page)); | |
4177 | } | |
4178 | ||
4179 | static ssize_t | |
4180 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | |
4181 | { | |
4182 | struct request_list *rl = &q->rq; | |
c981ff9f AV |
4183 | unsigned long nr; |
4184 | int ret = queue_var_store(&nr, page, count); | |
4185 | if (nr < BLKDEV_MIN_RQ) | |
4186 | nr = BLKDEV_MIN_RQ; | |
1da177e4 | 4187 | |
c981ff9f AV |
4188 | spin_lock_irq(q->queue_lock); |
4189 | q->nr_requests = nr; | |
1da177e4 LT |
4190 | blk_queue_congestion_threshold(q); |
4191 | ||
4192 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 4193 | blk_set_queue_congested(q, READ); |
1da177e4 | 4194 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
79e2de4b | 4195 | blk_clear_queue_congested(q, READ); |
1da177e4 LT |
4196 | |
4197 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | |
79e2de4b | 4198 | blk_set_queue_congested(q, WRITE); |
1da177e4 | 4199 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
79e2de4b | 4200 | blk_clear_queue_congested(q, WRITE); |
1da177e4 LT |
4201 | |
4202 | if (rl->count[READ] >= q->nr_requests) { | |
4203 | blk_set_queue_full(q, READ); | |
4204 | } else if (rl->count[READ]+1 <= q->nr_requests) { | |
4205 | blk_clear_queue_full(q, READ); | |
4206 | wake_up(&rl->wait[READ]); | |
4207 | } | |
4208 | ||
4209 | if (rl->count[WRITE] >= q->nr_requests) { | |
4210 | blk_set_queue_full(q, WRITE); | |
4211 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | |
4212 | blk_clear_queue_full(q, WRITE); | |
4213 | wake_up(&rl->wait[WRITE]); | |
4214 | } | |
c981ff9f | 4215 | spin_unlock_irq(q->queue_lock); |
1da177e4 LT |
4216 | return ret; |
4217 | } | |
4218 | ||
4219 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | |
4220 | { | |
4221 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | |
4222 | ||
4223 | return queue_var_show(ra_kb, (page)); | |
4224 | } | |
4225 | ||
4226 | static ssize_t | |
4227 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | |
4228 | { | |
4229 | unsigned long ra_kb; | |
4230 | ssize_t ret = queue_var_store(&ra_kb, page, count); | |
4231 | ||
4232 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
4233 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
4234 | spin_unlock_irq(q->queue_lock); | |
4235 | ||
4236 | return ret; | |
4237 | } | |
4238 | ||
4239 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | |
4240 | { | |
4241 | int max_sectors_kb = q->max_sectors >> 1; | |
4242 | ||
4243 | return queue_var_show(max_sectors_kb, (page)); | |
4244 | } | |
4245 | ||
4246 | static ssize_t | |
4247 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | |
4248 | { | |
4249 | unsigned long max_sectors_kb, | |
4250 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | |
4251 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | |
4252 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | |
1da177e4 LT |
4253 | |
4254 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | |
4255 | return -EINVAL; | |
4256 | /* | |
4257 | * Take the queue lock to update the readahead and max_sectors | |
4258 | * values synchronously: | |
4259 | */ | |
4260 | spin_lock_irq(q->queue_lock); | |
1da177e4 LT |
4261 | q->max_sectors = max_sectors_kb << 1; |
4262 | spin_unlock_irq(q->queue_lock); | |
4263 | ||
4264 | return ret; | |
4265 | } | |
4266 | ||
4267 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | |
4268 | { | |
4269 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | |
4270 | ||
4271 | return queue_var_show(max_hw_sectors_kb, (page)); | |
4272 | } | |
4273 | ||
4274 | ||
4275 | static struct queue_sysfs_entry queue_requests_entry = { | |
4276 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | |
4277 | .show = queue_requests_show, | |
4278 | .store = queue_requests_store, | |
4279 | }; | |
4280 | ||
4281 | static struct queue_sysfs_entry queue_ra_entry = { | |
4282 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | |
4283 | .show = queue_ra_show, | |
4284 | .store = queue_ra_store, | |
4285 | }; | |
4286 | ||
4287 | static struct queue_sysfs_entry queue_max_sectors_entry = { | |
4288 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | |
4289 | .show = queue_max_sectors_show, | |
4290 | .store = queue_max_sectors_store, | |
4291 | }; | |
4292 | ||
4293 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | |
4294 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | |
4295 | .show = queue_max_hw_sectors_show, | |
4296 | }; | |
4297 | ||
4298 | static struct queue_sysfs_entry queue_iosched_entry = { | |
4299 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | |
4300 | .show = elv_iosched_show, | |
4301 | .store = elv_iosched_store, | |
4302 | }; | |
4303 | ||
4304 | static struct attribute *default_attrs[] = { | |
4305 | &queue_requests_entry.attr, | |
4306 | &queue_ra_entry.attr, | |
4307 | &queue_max_hw_sectors_entry.attr, | |
4308 | &queue_max_sectors_entry.attr, | |
4309 | &queue_iosched_entry.attr, | |
4310 | NULL, | |
4311 | }; | |
4312 | ||
4313 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | |
4314 | ||
4315 | static ssize_t | |
4316 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
4317 | { | |
4318 | struct queue_sysfs_entry *entry = to_queue(attr); | |
165125e1 JA |
4319 | struct request_queue *q = |
4320 | container_of(kobj, struct request_queue, kobj); | |
483f4afc | 4321 | ssize_t res; |
1da177e4 | 4322 | |
1da177e4 | 4323 | if (!entry->show) |
6c1852a0 | 4324 | return -EIO; |
483f4afc AV |
4325 | mutex_lock(&q->sysfs_lock); |
4326 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4327 | mutex_unlock(&q->sysfs_lock); | |
4328 | return -ENOENT; | |
4329 | } | |
4330 | res = entry->show(q, page); | |
4331 | mutex_unlock(&q->sysfs_lock); | |
4332 | return res; | |
1da177e4 LT |
4333 | } |
4334 | ||
4335 | static ssize_t | |
4336 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | |
4337 | const char *page, size_t length) | |
4338 | { | |
4339 | struct queue_sysfs_entry *entry = to_queue(attr); | |
165125e1 | 4340 | struct request_queue *q = container_of(kobj, struct request_queue, kobj); |
483f4afc AV |
4341 | |
4342 | ssize_t res; | |
1da177e4 | 4343 | |
1da177e4 | 4344 | if (!entry->store) |
6c1852a0 | 4345 | return -EIO; |
483f4afc AV |
4346 | mutex_lock(&q->sysfs_lock); |
4347 | if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { | |
4348 | mutex_unlock(&q->sysfs_lock); | |
4349 | return -ENOENT; | |
4350 | } | |
4351 | res = entry->store(q, page, length); | |
4352 | mutex_unlock(&q->sysfs_lock); | |
4353 | return res; | |
1da177e4 LT |
4354 | } |
4355 | ||
4356 | static struct sysfs_ops queue_sysfs_ops = { | |
4357 | .show = queue_attr_show, | |
4358 | .store = queue_attr_store, | |
4359 | }; | |
4360 | ||
93d17d3d | 4361 | static struct kobj_type queue_ktype = { |
1da177e4 LT |
4362 | .sysfs_ops = &queue_sysfs_ops, |
4363 | .default_attrs = default_attrs, | |
483f4afc | 4364 | .release = blk_release_queue, |
1da177e4 LT |
4365 | }; |
4366 | ||
4367 | int blk_register_queue(struct gendisk *disk) | |
4368 | { | |
4369 | int ret; | |
4370 | ||
165125e1 | 4371 | struct request_queue *q = disk->queue; |
1da177e4 LT |
4372 | |
4373 | if (!q || !q->request_fn) | |
4374 | return -ENXIO; | |
4375 | ||
b2d6db58 GKH |
4376 | ret = kobject_add(&q->kobj, kobject_get(&disk->dev.kobj), |
4377 | "%s", "queue"); | |
1da177e4 LT |
4378 | if (ret < 0) |
4379 | return ret; | |
4380 | ||
483f4afc AV |
4381 | kobject_uevent(&q->kobj, KOBJ_ADD); |
4382 | ||
1da177e4 LT |
4383 | ret = elv_register_queue(q); |
4384 | if (ret) { | |
483f4afc AV |
4385 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4386 | kobject_del(&q->kobj); | |
1da177e4 LT |
4387 | return ret; |
4388 | } | |
4389 | ||
4390 | return 0; | |
4391 | } | |
4392 | ||
4393 | void blk_unregister_queue(struct gendisk *disk) | |
4394 | { | |
165125e1 | 4395 | struct request_queue *q = disk->queue; |
1da177e4 LT |
4396 | |
4397 | if (q && q->request_fn) { | |
4398 | elv_unregister_queue(q); | |
4399 | ||
483f4afc AV |
4400 | kobject_uevent(&q->kobj, KOBJ_REMOVE); |
4401 | kobject_del(&q->kobj); | |
edfaa7c3 | 4402 | kobject_put(&disk->dev.kobj); |
1da177e4 LT |
4403 | } |
4404 | } |