blk-core.c (8324aa91d1e11a1fc25f209687a0b2e6c2ed47d0) blk-core.c (86db1e29772372155db08ff48a9ceb76e11a2ad1)
1/*
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 */

--- 6 unchanged lines hidden (view full) ---

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>
1/*
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 */

--- 6 unchanged lines hidden (view full) ---

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>
28#include <linux/task_io_accounting_ops.h>
29#include <linux/interrupt.h>
30#include <linux/cpu.h>
31#include <linux/blktrace_api.h>
32#include <linux/fault-inject.h>
33#include <linux/scatterlist.h>
34
35#include "blk.h"
36
23#include <linux/completion.h>
24#include <linux/slab.h>
25#include <linux/swap.h>
26#include <linux/writeback.h>
27#include <linux/task_io_accounting_ops.h>
28#include <linux/interrupt.h>
29#include <linux/cpu.h>
30#include <linux/blktrace_api.h>
31#include <linux/fault-inject.h>
32#include <linux/scatterlist.h>
33
34#include "blk.h"
35
37/*
38 * for max sense size
39 */
40#include <scsi/scsi_cmnd.h>
41
42static void blk_unplug_work(struct work_struct *work);
43static void blk_unplug_timeout(unsigned long data);
44static void drive_stat_acct(struct request *rq, int new_io);
36static void drive_stat_acct(struct request *rq, int new_io);
45static void init_request_from_bio(struct request *req, struct bio *bio);
46static int __make_request(struct request_queue *q, struct bio *bio);
37static int __make_request(struct request_queue *q, struct bio *bio);
47static struct io_context *current_io_context(gfp_t gfp_flags, int node);
48static void blk_recalc_rq_segments(struct request *rq);
38static void blk_recalc_rq_segments(struct request *rq);
49static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
50 struct bio *bio);
51
52/*
53 * For the allocated request tables
54 */
55struct kmem_cache *request_cachep;
56
57/*
58 * For queue allocation
59 */
60struct kmem_cache *blk_requestq_cachep = NULL;
61
62/*
39
40/*
41 * For the allocated request tables
42 */
43struct kmem_cache *request_cachep;
44
45/*
46 * For queue allocation
47 */
48struct kmem_cache *blk_requestq_cachep = NULL;
49
50/*
63 * For io context allocations
64 */
65static struct kmem_cache *iocontext_cachep;
66
67/*
68 * Controlling structure to kblockd
69 */
70static struct workqueue_struct *kblockd_workqueue;
71
51 * Controlling structure to kblockd
52 */
53static struct workqueue_struct *kblockd_workqueue;
54
72unsigned long blk_max_low_pfn, blk_max_pfn;
73
74EXPORT_SYMBOL(blk_max_low_pfn);
75EXPORT_SYMBOL(blk_max_pfn);
76
77static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
78
55static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
56
79/* Amount of time in which a process may batch requests */
80#define BLK_BATCH_TIME (HZ/50UL)
81
82/* Number of requests a "batching" process may submit */
83#define BLK_BATCH_REQ 32
84
85void blk_queue_congestion_threshold(struct request_queue *q)
86{
87 int nr;
88
89 nr = q->nr_requests - (q->nr_requests / 8) + 1;
90 if (nr > q->nr_requests)
91 nr = q->nr_requests;
92 q->nr_congestion_on = nr;

--- 19 unchanged lines hidden (view full) ---

112 struct request_queue *q = bdev_get_queue(bdev);
113
114 if (q)
115 ret = &q->backing_dev_info;
116 return ret;
117}
118EXPORT_SYMBOL(blk_get_backing_dev_info);
119
57void blk_queue_congestion_threshold(struct request_queue *q)
58{
59 int nr;
60
61 nr = q->nr_requests - (q->nr_requests / 8) + 1;
62 if (nr > q->nr_requests)
63 nr = q->nr_requests;
64 q->nr_congestion_on = nr;

--- 19 unchanged lines hidden (view full) ---

84 struct request_queue *q = bdev_get_queue(bdev);
85
86 if (q)
87 ret = &q->backing_dev_info;
88 return ret;
89}
90EXPORT_SYMBOL(blk_get_backing_dev_info);
91
120/**
121 * blk_queue_prep_rq - set a prepare_request function for queue
122 * @q: queue
123 * @pfn: prepare_request function
124 *
125 * It's possible for a queue to register a prepare_request callback which
126 * is invoked before the request is handed to the request_fn. The goal of
127 * the function is to prepare a request for I/O, it can be used to build a
128 * cdb from the request data for instance.
129 *
130 */
131void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
92void rq_init(struct request_queue *q, struct request *rq)
132{
93{
133 q->prep_rq_fn = pfn;
134}
135
136EXPORT_SYMBOL(blk_queue_prep_rq);
137
138/**
139 * blk_queue_merge_bvec - set a merge_bvec function for queue
140 * @q: queue
141 * @mbfn: merge_bvec_fn
142 *
143 * Usually queues have static limitations on the max sectors or segments that
144 * we can put in a request. Stacking drivers may have some settings that
145 * are dynamic, and thus we have to query the queue whether it is ok to
146 * add a new bio_vec to a bio at a given offset or not. If the block device
147 * has such limitations, it needs to register a merge_bvec_fn to control
148 * the size of bio's sent to it. Note that a block device *must* allow a
149 * single page to be added to an empty bio. The block device driver may want
150 * to use the bio_split() function to deal with these bio's. By default
151 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
152 * honored.
153 */
154void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
155{
156 q->merge_bvec_fn = mbfn;
157}
158
159EXPORT_SYMBOL(blk_queue_merge_bvec);
160
161void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
162{
163 q->softirq_done_fn = fn;
164}
165
166EXPORT_SYMBOL(blk_queue_softirq_done);
167
168/**
169 * blk_queue_make_request - define an alternate make_request function for a device
170 * @q: the request queue for the device to be affected
171 * @mfn: the alternate make_request function
172 *
173 * Description:
174 * The normal way for &struct bios to be passed to a device
175 * driver is for them to be collected into requests on a request
176 * queue, and then to allow the device driver to select requests
177 * off that queue when it is ready. This works well for many block
178 * devices. However some block devices (typically virtual devices
179 * such as md or lvm) do not benefit from the processing on the
180 * request queue, and are served best by having the requests passed
181 * directly to them. This can be achieved by providing a function
182 * to blk_queue_make_request().
183 *
184 * Caveat:
185 * The driver that does this *must* be able to deal appropriately
186 * with buffers in "highmemory". This can be accomplished by either calling
187 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
188 * blk_queue_bounce() to create a buffer in normal memory.
189 **/
190void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn)
191{
192 /*
193 * set defaults
194 */
195 q->nr_requests = BLKDEV_MAX_RQ;
196 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
197 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
198 q->make_request_fn = mfn;
199 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
200 q->backing_dev_info.state = 0;
201 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
202 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
203 blk_queue_hardsect_size(q, 512);
204 blk_queue_dma_alignment(q, 511);
205 blk_queue_congestion_threshold(q);
206 q->nr_batching = BLK_BATCH_REQ;
207
208 q->unplug_thresh = 4; /* hmm */
209 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
210 if (q->unplug_delay == 0)
211 q->unplug_delay = 1;
212
213 INIT_WORK(&q->unplug_work, blk_unplug_work);
214
215 q->unplug_timer.function = blk_unplug_timeout;
216 q->unplug_timer.data = (unsigned long)q;
217
218 /*
219 * by default assume old behaviour and bounce for any highmem page
220 */
221 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
222}
223
224EXPORT_SYMBOL(blk_queue_make_request);
225
226static void rq_init(struct request_queue *q, struct request *rq)
227{
228 INIT_LIST_HEAD(&rq->queuelist);
229 INIT_LIST_HEAD(&rq->donelist);
230
231 rq->errors = 0;
232 rq->bio = rq->biotail = NULL;
233 INIT_HLIST_NODE(&rq->hash);
234 RB_CLEAR_NODE(&rq->rb_node);
235 rq->ioprio = 0;

--- 6 unchanged lines hidden (view full) ---

242 rq->nr_phys_segments = 0;
243 rq->sense = NULL;
244 rq->end_io = NULL;
245 rq->end_io_data = NULL;
246 rq->completion_data = NULL;
247 rq->next_rq = NULL;
248}
249
94 INIT_LIST_HEAD(&rq->queuelist);
95 INIT_LIST_HEAD(&rq->donelist);
96
97 rq->errors = 0;
98 rq->bio = rq->biotail = NULL;
99 INIT_HLIST_NODE(&rq->hash);
100 RB_CLEAR_NODE(&rq->rb_node);
101 rq->ioprio = 0;

--- 6 unchanged lines hidden (view full) ---

108 rq->nr_phys_segments = 0;
109 rq->sense = NULL;
110 rq->end_io = NULL;
111 rq->end_io_data = NULL;
112 rq->completion_data = NULL;
113 rq->next_rq = NULL;
114}
115
250/**
251 * blk_queue_ordered - does this queue support ordered writes
252 * @q: the request queue
253 * @ordered: one of QUEUE_ORDERED_*
254 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
255 *
256 * Description:
257 * For journalled file systems, doing ordered writes on a commit
258 * block instead of explicitly doing wait_on_buffer (which is bad
259 * for performance) can be a big win. Block drivers supporting this
260 * feature should call this function and indicate so.
261 *
262 **/
263int blk_queue_ordered(struct request_queue *q, unsigned ordered,
264 prepare_flush_fn *prepare_flush_fn)
265{
266 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
267 prepare_flush_fn == NULL) {
268 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
269 return -EINVAL;
270 }
271
272 if (ordered != QUEUE_ORDERED_NONE &&
273 ordered != QUEUE_ORDERED_DRAIN &&
274 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
275 ordered != QUEUE_ORDERED_DRAIN_FUA &&
276 ordered != QUEUE_ORDERED_TAG &&
277 ordered != QUEUE_ORDERED_TAG_FLUSH &&
278 ordered != QUEUE_ORDERED_TAG_FUA) {
279 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
280 return -EINVAL;
281 }
282
283 q->ordered = ordered;
284 q->next_ordered = ordered;
285 q->prepare_flush_fn = prepare_flush_fn;
286
287 return 0;
288}
289
290EXPORT_SYMBOL(blk_queue_ordered);
291
292/*
293 * Cache flushing for ordered writes handling
294 */
295inline unsigned blk_ordered_cur_seq(struct request_queue *q)
296{
297 if (!q->ordseq)
298 return 0;
299 return 1 << ffz(q->ordseq);
300}
301
302unsigned blk_ordered_req_seq(struct request *rq)
303{
304 struct request_queue *q = rq->q;
305
306 BUG_ON(q->ordseq == 0);
307
308 if (rq == &q->pre_flush_rq)
309 return QUEUE_ORDSEQ_PREFLUSH;
310 if (rq == &q->bar_rq)
311 return QUEUE_ORDSEQ_BAR;
312 if (rq == &q->post_flush_rq)
313 return QUEUE_ORDSEQ_POSTFLUSH;
314
315 /*
316 * !fs requests don't need to follow barrier ordering. Always
317 * put them at the front. This fixes the following deadlock.
318 *
319 * http://thread.gmane.org/gmane.linux.kernel/537473
320 */
321 if (!blk_fs_request(rq))
322 return QUEUE_ORDSEQ_DRAIN;
323
324 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
325 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
326 return QUEUE_ORDSEQ_DRAIN;
327 else
328 return QUEUE_ORDSEQ_DONE;
329}
330
331void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
332{
333 struct request *rq;
334
335 if (error && !q->orderr)
336 q->orderr = error;
337
338 BUG_ON(q->ordseq & seq);
339 q->ordseq |= seq;
340
341 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
342 return;
343
344 /*
345 * Okay, sequence complete.
346 */
347 q->ordseq = 0;
348 rq = q->orig_bar_rq;
349
350 if (__blk_end_request(rq, q->orderr, blk_rq_bytes(rq)))
351 BUG();
352}
353
354static void pre_flush_end_io(struct request *rq, int error)
355{
356 elv_completed_request(rq->q, rq);
357 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
358}
359
360static void bar_end_io(struct request *rq, int error)
361{
362 elv_completed_request(rq->q, rq);
363 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
364}
365
366static void post_flush_end_io(struct request *rq, int error)
367{
368 elv_completed_request(rq->q, rq);
369 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
370}
371
372static void queue_flush(struct request_queue *q, unsigned which)
373{
374 struct request *rq;
375 rq_end_io_fn *end_io;
376
377 if (which == QUEUE_ORDERED_PREFLUSH) {
378 rq = &q->pre_flush_rq;
379 end_io = pre_flush_end_io;
380 } else {
381 rq = &q->post_flush_rq;
382 end_io = post_flush_end_io;
383 }
384
385 rq->cmd_flags = REQ_HARDBARRIER;
386 rq_init(q, rq);
387 rq->elevator_private = NULL;
388 rq->elevator_private2 = NULL;
389 rq->rq_disk = q->bar_rq.rq_disk;
390 rq->end_io = end_io;
391 q->prepare_flush_fn(q, rq);
392
393 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
394}
395
396static inline struct request *start_ordered(struct request_queue *q,
397 struct request *rq)
398{
399 q->orderr = 0;
400 q->ordered = q->next_ordered;
401 q->ordseq |= QUEUE_ORDSEQ_STARTED;
402
403 /*
404 * Prep proxy barrier request.
405 */
406 blkdev_dequeue_request(rq);
407 q->orig_bar_rq = rq;
408 rq = &q->bar_rq;
409 rq->cmd_flags = 0;
410 rq_init(q, rq);
411 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
412 rq->cmd_flags |= REQ_RW;
413 if (q->ordered & QUEUE_ORDERED_FUA)
414 rq->cmd_flags |= REQ_FUA;
415 rq->elevator_private = NULL;
416 rq->elevator_private2 = NULL;
417 init_request_from_bio(rq, q->orig_bar_rq->bio);
418 rq->end_io = bar_end_io;
419
420 /*
421 * Queue ordered sequence. As we stack them at the head, we
422 * need to queue in reverse order. Note that we rely on that
423 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
424 * request gets inbetween ordered sequence. If this request is
425 * an empty barrier, we don't need to do a postflush ever since
426 * there will be no data written between the pre and post flush.
427 * Hence a single flush will suffice.
428 */
429 if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq))
430 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
431 else
432 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
433
434 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
435
436 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
437 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
438 rq = &q->pre_flush_rq;
439 } else
440 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
441
442 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
443 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
444 else
445 rq = NULL;
446
447 return rq;
448}
449
450int blk_do_ordered(struct request_queue *q, struct request **rqp)
451{
452 struct request *rq = *rqp;
453 const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
454
455 if (!q->ordseq) {
456 if (!is_barrier)
457 return 1;
458
459 if (q->next_ordered != QUEUE_ORDERED_NONE) {
460 *rqp = start_ordered(q, rq);
461 return 1;
462 } else {
463 /*
464 * This can happen when the queue switches to
465 * ORDERED_NONE while this request is on it.
466 */
467 blkdev_dequeue_request(rq);
468 if (__blk_end_request(rq, -EOPNOTSUPP,
469 blk_rq_bytes(rq)))
470 BUG();
471 *rqp = NULL;
472 return 0;
473 }
474 }
475
476 /*
477 * Ordered sequence in progress
478 */
479
480 /* Special requests are not subject to ordering rules. */
481 if (!blk_fs_request(rq) &&
482 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
483 return 1;
484
485 if (q->ordered & QUEUE_ORDERED_TAG) {
486 /* Ordered by tag. Blocking the next barrier is enough. */
487 if (is_barrier && rq != &q->bar_rq)
488 *rqp = NULL;
489 } else {
490 /* Ordered by draining. Wait for turn. */
491 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
492 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
493 *rqp = NULL;
494 }
495
496 return 1;
497}
498
499static void req_bio_endio(struct request *rq, struct bio *bio,
500 unsigned int nbytes, int error)
501{
502 struct request_queue *q = rq->q;
503
504 if (&q->bar_rq != rq) {
505 if (error)
506 clear_bit(BIO_UPTODATE, &bio->bi_flags);

--- 16 unchanged lines hidden (view full) ---

523 * Okay, this is the barrier request in progress, just
524 * record the error;
525 */
526 if (error && !q->orderr)
527 q->orderr = error;
528 }
529}
530
116static void req_bio_endio(struct request *rq, struct bio *bio,
117 unsigned int nbytes, int error)
118{
119 struct request_queue *q = rq->q;
120
121 if (&q->bar_rq != rq) {
122 if (error)
123 clear_bit(BIO_UPTODATE, &bio->bi_flags);

--- 16 unchanged lines hidden (view full) ---

140 * Okay, this is the barrier request in progress, just
141 * record the error;
142 */
143 if (error && !q->orderr)
144 q->orderr = error;
145 }
146}
147
531/**
532 * blk_queue_bounce_limit - set bounce buffer limit for queue
533 * @q: the request queue for the device
534 * @dma_addr: bus address limit
535 *
536 * Description:
537 * Different hardware can have different requirements as to what pages
538 * it can do I/O directly to. A low level driver can call
539 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
540 * buffers for doing I/O to pages residing above @page.
541 **/
542void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
543{
544 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
545 int dma = 0;
546
547 q->bounce_gfp = GFP_NOIO;
548#if BITS_PER_LONG == 64
549 /* Assume anything <= 4GB can be handled by IOMMU.
550 Actually some IOMMUs can handle everything, but I don't
551 know of a way to test this here. */
552 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
553 dma = 1;
554 q->bounce_pfn = max_low_pfn;
555#else
556 if (bounce_pfn < blk_max_low_pfn)
557 dma = 1;
558 q->bounce_pfn = bounce_pfn;
559#endif
560 if (dma) {
561 init_emergency_isa_pool();
562 q->bounce_gfp = GFP_NOIO | GFP_DMA;
563 q->bounce_pfn = bounce_pfn;
564 }
565}
566
567EXPORT_SYMBOL(blk_queue_bounce_limit);
568
569/**
570 * blk_queue_max_sectors - set max sectors for a request for this queue
571 * @q: the request queue for the device
572 * @max_sectors: max sectors in the usual 512b unit
573 *
574 * Description:
575 * Enables a low level driver to set an upper limit on the size of
576 * received requests.
577 **/
578void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
579{
580 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
581 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
582 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
583 }
584
585 if (BLK_DEF_MAX_SECTORS > max_sectors)
586 q->max_hw_sectors = q->max_sectors = max_sectors;
587 else {
588 q->max_sectors = BLK_DEF_MAX_SECTORS;
589 q->max_hw_sectors = max_sectors;
590 }
591}
592
593EXPORT_SYMBOL(blk_queue_max_sectors);
594
595/**
596 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
597 * @q: the request queue for the device
598 * @max_segments: max number of segments
599 *
600 * Description:
601 * Enables a low level driver to set an upper limit on the number of
602 * physical data segments in a request. This would be the largest sized
603 * scatter list the driver could handle.
604 **/
605void blk_queue_max_phys_segments(struct request_queue *q,
606 unsigned short max_segments)
607{
608 if (!max_segments) {
609 max_segments = 1;
610 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
611 }
612
613 q->max_phys_segments = max_segments;
614}
615
616EXPORT_SYMBOL(blk_queue_max_phys_segments);
617
618/**
619 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
620 * @q: the request queue for the device
621 * @max_segments: max number of segments
622 *
623 * Description:
624 * Enables a low level driver to set an upper limit on the number of
625 * hw data segments in a request. This would be the largest number of
626 * address/length pairs the host adapter can actually give as once
627 * to the device.
628 **/
629void blk_queue_max_hw_segments(struct request_queue *q,
630 unsigned short max_segments)
631{
632 if (!max_segments) {
633 max_segments = 1;
634 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
635 }
636
637 q->max_hw_segments = max_segments;
638}
639
640EXPORT_SYMBOL(blk_queue_max_hw_segments);
641
642/**
643 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
644 * @q: the request queue for the device
645 * @max_size: max size of segment in bytes
646 *
647 * Description:
648 * Enables a low level driver to set an upper limit on the size of a
649 * coalesced segment
650 **/
651void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
652{
653 if (max_size < PAGE_CACHE_SIZE) {
654 max_size = PAGE_CACHE_SIZE;
655 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
656 }
657
658 q->max_segment_size = max_size;
659}
660
661EXPORT_SYMBOL(blk_queue_max_segment_size);
662
663/**
664 * blk_queue_hardsect_size - set hardware sector size for the queue
665 * @q: the request queue for the device
666 * @size: the hardware sector size, in bytes
667 *
668 * Description:
669 * This should typically be set to the lowest possible sector size
670 * that the hardware can operate on (possible without reverting to
671 * even internal read-modify-write operations). Usually the default
672 * of 512 covers most hardware.
673 **/
674void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
675{
676 q->hardsect_size = size;
677}
678
679EXPORT_SYMBOL(blk_queue_hardsect_size);
680
681/*
682 * Returns the minimum that is _not_ zero, unless both are zero.
683 */
684#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
685
686/**
687 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
688 * @t: the stacking driver (top)
689 * @b: the underlying device (bottom)
690 **/
691void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
692{
693 /* zero is "infinity" */
694 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
695 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
696
697 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
698 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
699 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
700 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
701 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
702 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
703}
704
705EXPORT_SYMBOL(blk_queue_stack_limits);
706
707/**
708 * blk_queue_dma_drain - Set up a drain buffer for excess dma.
709 *
710 * @q: the request queue for the device
711 * @buf: physically contiguous buffer
712 * @size: size of the buffer in bytes
713 *
714 * Some devices have excess DMA problems and can't simply discard (or
715 * zero fill) the unwanted piece of the transfer. They have to have a
716 * real area of memory to transfer it into. The use case for this is
717 * ATAPI devices in DMA mode. If the packet command causes a transfer
718 * bigger than the transfer size some HBAs will lock up if there
719 * aren't DMA elements to contain the excess transfer. What this API
720 * does is adjust the queue so that the buf is always appended
721 * silently to the scatterlist.
722 *
723 * Note: This routine adjusts max_hw_segments to make room for
724 * appending the drain buffer. If you call
725 * blk_queue_max_hw_segments() or blk_queue_max_phys_segments() after
726 * calling this routine, you must set the limit to one fewer than your
727 * device can support otherwise there won't be room for the drain
728 * buffer.
729 */
730int blk_queue_dma_drain(struct request_queue *q, void *buf,
731 unsigned int size)
732{
733 if (q->max_hw_segments < 2 || q->max_phys_segments < 2)
734 return -EINVAL;
735 /* make room for appending the drain */
736 --q->max_hw_segments;
737 --q->max_phys_segments;
738 q->dma_drain_buffer = buf;
739 q->dma_drain_size = size;
740
741 return 0;
742}
743
744EXPORT_SYMBOL_GPL(blk_queue_dma_drain);
745
746/**
747 * blk_queue_segment_boundary - set boundary rules for segment merging
748 * @q: the request queue for the device
749 * @mask: the memory boundary mask
750 **/
751void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
752{
753 if (mask < PAGE_CACHE_SIZE - 1) {
754 mask = PAGE_CACHE_SIZE - 1;
755 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
756 }
757
758 q->seg_boundary_mask = mask;
759}
760
761EXPORT_SYMBOL(blk_queue_segment_boundary);
762
763/**
764 * blk_queue_dma_alignment - set dma length and memory alignment
765 * @q: the request queue for the device
766 * @mask: alignment mask
767 *
768 * description:
769 * set required memory and length aligment for direct dma transactions.
770 * this is used when buiding direct io requests for the queue.
771 *
772 **/
773void blk_queue_dma_alignment(struct request_queue *q, int mask)
774{
775 q->dma_alignment = mask;
776}
777
778EXPORT_SYMBOL(blk_queue_dma_alignment);
779
780/**
781 * blk_queue_update_dma_alignment - update dma length and memory alignment
782 * @q: the request queue for the device
783 * @mask: alignment mask
784 *
785 * description:
786 * update required memory and length aligment for direct dma transactions.
787 * If the requested alignment is larger than the current alignment, then
788 * the current queue alignment is updated to the new value, otherwise it
789 * is left alone. The design of this is to allow multiple objects
790 * (driver, device, transport etc) to set their respective
791 * alignments without having them interfere.
792 *
793 **/
794void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
795{
796 BUG_ON(mask > PAGE_SIZE);
797
798 if (mask > q->dma_alignment)
799 q->dma_alignment = mask;
800}
801
802EXPORT_SYMBOL(blk_queue_update_dma_alignment);
803
804void blk_dump_rq_flags(struct request *rq, char *msg)
805{
806 int bit;
807
808 printk("%s: dev %s: type=%x, flags=%x\n", msg,
809 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
810 rq->cmd_flags);
811

--- 257 unchanged lines hidden (view full) ---

1069 * This will form the start of a new hw segment. Bump both
1070 * counters.
1071 */
1072 req->nr_hw_segments += nr_hw_segs;
1073 req->nr_phys_segments += nr_phys_segs;
1074 return 1;
1075}
1076
148void blk_dump_rq_flags(struct request *rq, char *msg)
149{
150 int bit;
151
152 printk("%s: dev %s: type=%x, flags=%x\n", msg,
153 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
154 rq->cmd_flags);
155

--- 257 unchanged lines hidden (view full) ---

413 * This will form the start of a new hw segment. Bump both
414 * counters.
415 */
416 req->nr_hw_segments += nr_hw_segs;
417 req->nr_phys_segments += nr_phys_segs;
418 return 1;
419}
420
1077static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1078 struct bio *bio)
421int ll_back_merge_fn(struct request_queue *q, struct request *req,
422 struct bio *bio)
1079{
1080 unsigned short max_sectors;
1081 int len;
1082
1083 if (unlikely(blk_pc_request(req)))
1084 max_sectors = q->max_hw_sectors;
1085 else
1086 max_sectors = q->max_sectors;

--- 193 unchanged lines hidden (view full) ---

1280static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1281 struct page *page)
1282{
1283 struct request_queue *q = bdi->unplug_io_data;
1284
1285 blk_unplug(q);
1286}
1287
423{
424 unsigned short max_sectors;
425 int len;
426
427 if (unlikely(blk_pc_request(req)))
428 max_sectors = q->max_hw_sectors;
429 else
430 max_sectors = q->max_sectors;

--- 193 unchanged lines hidden (view full) ---

624static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
625 struct page *page)
626{
627 struct request_queue *q = bdi->unplug_io_data;
628
629 blk_unplug(q);
630}
631
1288static void blk_unplug_work(struct work_struct *work)
632void blk_unplug_work(struct work_struct *work)
1289{
1290 struct request_queue *q =
1291 container_of(work, struct request_queue, unplug_work);
1292
1293 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1294 q->rq.count[READ] + q->rq.count[WRITE]);
1295
1296 q->unplug_fn(q);
1297}
1298
633{
634 struct request_queue *q =
635 container_of(work, struct request_queue, unplug_work);
636
637 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
638 q->rq.count[READ] + q->rq.count[WRITE]);
639
640 q->unplug_fn(q);
641}
642
1299static void blk_unplug_timeout(unsigned long data)
643void blk_unplug_timeout(unsigned long data)
1300{
1301 struct request_queue *q = (struct request_queue *)data;
1302
1303 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1304 q->rq.count[READ] + q->rq.count[WRITE]);
1305
1306 kblockd_schedule_work(&q->unplug_work);
1307}

--- 648 unchanged lines hidden (view full) ---

1956 drive_stat_acct(rq, 1);
1957 __elv_add_request(q, rq, where, 0);
1958 blk_start_queueing(q);
1959 spin_unlock_irqrestore(q->queue_lock, flags);
1960}
1961
1962EXPORT_SYMBOL(blk_insert_request);
1963
644{
645 struct request_queue *q = (struct request_queue *)data;
646
647 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
648 q->rq.count[READ] + q->rq.count[WRITE]);
649
650 kblockd_schedule_work(&q->unplug_work);
651}

--- 648 unchanged lines hidden (view full) ---

1300 drive_stat_acct(rq, 1);
1301 __elv_add_request(q, rq, where, 0);
1302 blk_start_queueing(q);
1303 spin_unlock_irqrestore(q->queue_lock, flags);
1304}
1305
1306EXPORT_SYMBOL(blk_insert_request);
1307
1964static int __blk_rq_unmap_user(struct bio *bio)
1965{
1966 int ret = 0;
1967
1968 if (bio) {
1969 if (bio_flagged(bio, BIO_USER_MAPPED))
1970 bio_unmap_user(bio);
1971 else
1972 ret = bio_uncopy_user(bio);
1973 }
1974
1975 return ret;
1976}
1977
1978int blk_rq_append_bio(struct request_queue *q, struct request *rq,
1979 struct bio *bio)
1980{
1981 if (!rq->bio)
1982 blk_rq_bio_prep(q, rq, bio);
1983 else if (!ll_back_merge_fn(q, rq, bio))
1984 return -EINVAL;
1985 else {
1986 rq->biotail->bi_next = bio;
1987 rq->biotail = bio;
1988
1989 rq->data_len += bio->bi_size;
1990 }
1991 return 0;
1992}
1993EXPORT_SYMBOL(blk_rq_append_bio);
1994
1995static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
1996 void __user *ubuf, unsigned int len)
1997{
1998 unsigned long uaddr;
1999 struct bio *bio, *orig_bio;
2000 int reading, ret;
2001
2002 reading = rq_data_dir(rq) == READ;
2003
2004 /*
2005 * if alignment requirement is satisfied, map in user pages for
2006 * direct dma. else, set up kernel bounce buffers
2007 */
2008 uaddr = (unsigned long) ubuf;
2009 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2010 bio = bio_map_user(q, NULL, uaddr, len, reading);
2011 else
2012 bio = bio_copy_user(q, uaddr, len, reading);
2013
2014 if (IS_ERR(bio))
2015 return PTR_ERR(bio);
2016
2017 orig_bio = bio;
2018 blk_queue_bounce(q, &bio);
2019
2020 /*
2021 * We link the bounce buffer in and could have to traverse it
2022 * later so we have to get a ref to prevent it from being freed
2023 */
2024 bio_get(bio);
2025
2026 ret = blk_rq_append_bio(q, rq, bio);
2027 if (!ret)
2028 return bio->bi_size;
2029
2030 /* if it was boucned we must call the end io function */
2031 bio_endio(bio, 0);
2032 __blk_rq_unmap_user(orig_bio);
2033 bio_put(bio);
2034 return ret;
2035}
2036
2037/**
2038 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2039 * @q: request queue where request should be inserted
2040 * @rq: request structure to fill
2041 * @ubuf: the user buffer
2042 * @len: length of user data
2043 *
2044 * Description:
2045 * Data will be mapped directly for zero copy io, if possible. Otherwise
2046 * a kernel bounce buffer is used.
2047 *
2048 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2049 * still in process context.
2050 *
2051 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2052 * before being submitted to the device, as pages mapped may be out of
2053 * reach. It's the callers responsibility to make sure this happens. The
2054 * original bio must be passed back in to blk_rq_unmap_user() for proper
2055 * unmapping.
2056 */
2057int blk_rq_map_user(struct request_queue *q, struct request *rq,
2058 void __user *ubuf, unsigned long len)
2059{
2060 unsigned long bytes_read = 0;
2061 struct bio *bio = NULL;
2062 int ret;
2063
2064 if (len > (q->max_hw_sectors << 9))
2065 return -EINVAL;
2066 if (!len || !ubuf)
2067 return -EINVAL;
2068
2069 while (bytes_read != len) {
2070 unsigned long map_len, end, start;
2071
2072 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2073 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2074 >> PAGE_SHIFT;
2075 start = (unsigned long)ubuf >> PAGE_SHIFT;
2076
2077 /*
2078 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2079 * pages. If this happens we just lower the requested
2080 * mapping len by a page so that we can fit
2081 */
2082 if (end - start > BIO_MAX_PAGES)
2083 map_len -= PAGE_SIZE;
2084
2085 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2086 if (ret < 0)
2087 goto unmap_rq;
2088 if (!bio)
2089 bio = rq->bio;
2090 bytes_read += ret;
2091 ubuf += ret;
2092 }
2093
2094 rq->buffer = rq->data = NULL;
2095 return 0;
2096unmap_rq:
2097 blk_rq_unmap_user(bio);
2098 return ret;
2099}
2100
2101EXPORT_SYMBOL(blk_rq_map_user);
2102
2103/**
2104 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2105 * @q: request queue where request should be inserted
2106 * @rq: request to map data to
2107 * @iov: pointer to the iovec
2108 * @iov_count: number of elements in the iovec
2109 * @len: I/O byte count
2110 *
2111 * Description:
2112 * Data will be mapped directly for zero copy io, if possible. Otherwise
2113 * a kernel bounce buffer is used.
2114 *
2115 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2116 * still in process context.
2117 *
2118 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2119 * before being submitted to the device, as pages mapped may be out of
2120 * reach. It's the callers responsibility to make sure this happens. The
2121 * original bio must be passed back in to blk_rq_unmap_user() for proper
2122 * unmapping.
2123 */
2124int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
2125 struct sg_iovec *iov, int iov_count, unsigned int len)
2126{
2127 struct bio *bio;
2128
2129 if (!iov || iov_count <= 0)
2130 return -EINVAL;
2131
2132 /* we don't allow misaligned data like bio_map_user() does. If the
2133 * user is using sg, they're expected to know the alignment constraints
2134 * and respect them accordingly */
2135 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2136 if (IS_ERR(bio))
2137 return PTR_ERR(bio);
2138
2139 if (bio->bi_size != len) {
2140 bio_endio(bio, 0);
2141 bio_unmap_user(bio);
2142 return -EINVAL;
2143 }
2144
2145 bio_get(bio);
2146 blk_rq_bio_prep(q, rq, bio);
2147 rq->buffer = rq->data = NULL;
2148 return 0;
2149}
2150
2151EXPORT_SYMBOL(blk_rq_map_user_iov);
2152
2153/**
2154 * blk_rq_unmap_user - unmap a request with user data
2155 * @bio: start of bio list
2156 *
2157 * Description:
2158 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2159 * supply the original rq->bio from the blk_rq_map_user() return, since
2160 * the io completion may have changed rq->bio.
2161 */
2162int blk_rq_unmap_user(struct bio *bio)
2163{
2164 struct bio *mapped_bio;
2165 int ret = 0, ret2;
2166
2167 while (bio) {
2168 mapped_bio = bio;
2169 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
2170 mapped_bio = bio->bi_private;
2171
2172 ret2 = __blk_rq_unmap_user(mapped_bio);
2173 if (ret2 && !ret)
2174 ret = ret2;
2175
2176 mapped_bio = bio;
2177 bio = bio->bi_next;
2178 bio_put(mapped_bio);
2179 }
2180
2181 return ret;
2182}
2183
2184EXPORT_SYMBOL(blk_rq_unmap_user);
2185
2186/**
2187 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2188 * @q: request queue where request should be inserted
2189 * @rq: request to fill
2190 * @kbuf: the kernel buffer
2191 * @len: length of user data
2192 * @gfp_mask: memory allocation flags
2193 */
2194int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
2195 unsigned int len, gfp_t gfp_mask)
2196{
2197 struct bio *bio;
2198
2199 if (len > (q->max_hw_sectors << 9))
2200 return -EINVAL;
2201 if (!len || !kbuf)
2202 return -EINVAL;
2203
2204 bio = bio_map_kern(q, kbuf, len, gfp_mask);
2205 if (IS_ERR(bio))
2206 return PTR_ERR(bio);
2207
2208 if (rq_data_dir(rq) == WRITE)
2209 bio->bi_rw |= (1 << BIO_RW);
2210
2211 blk_rq_bio_prep(q, rq, bio);
2212 blk_queue_bounce(q, &rq->bio);
2213 rq->buffer = rq->data = NULL;
2214 return 0;
2215}
2216
2217EXPORT_SYMBOL(blk_rq_map_kern);
2218
2219/**
2220 * blk_execute_rq_nowait - insert a request into queue for execution
2221 * @q: queue to insert the request in
2222 * @bd_disk: matching gendisk
2223 * @rq: request to insert
2224 * @at_head: insert request at head or tail of queue
2225 * @done: I/O completion handler
2226 *
2227 * Description:
2228 * Insert a fully prepared request at the back of the io scheduler queue
2229 * for execution. Don't wait for completion.
2230 */
2231void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
2232 struct request *rq, int at_head,
2233 rq_end_io_fn *done)
2234{
2235 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2236
2237 rq->rq_disk = bd_disk;
2238 rq->cmd_flags |= REQ_NOMERGE;
2239 rq->end_io = done;
2240 WARN_ON(irqs_disabled());
2241 spin_lock_irq(q->queue_lock);
2242 __elv_add_request(q, rq, where, 1);
2243 __generic_unplug_device(q);
2244 spin_unlock_irq(q->queue_lock);
2245}
2246EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2247
2248/**
2249 * blk_execute_rq - insert a request into queue for execution
2250 * @q: queue to insert the request in
2251 * @bd_disk: matching gendisk
2252 * @rq: request to insert
2253 * @at_head: insert request at head or tail of queue
2254 *
2255 * Description:
2256 * Insert a fully prepared request at the back of the io scheduler queue
2257 * for execution and wait for completion.
2258 */
2259int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
2260 struct request *rq, int at_head)
2261{
2262 DECLARE_COMPLETION_ONSTACK(wait);
2263 char sense[SCSI_SENSE_BUFFERSIZE];
2264 int err = 0;
2265
2266 /*
2267 * we need an extra reference to the request, so we can look at
2268 * it after io completion
2269 */
2270 rq->ref_count++;
2271
2272 if (!rq->sense) {
2273 memset(sense, 0, sizeof(sense));
2274 rq->sense = sense;
2275 rq->sense_len = 0;
2276 }
2277
2278 rq->end_io_data = &wait;
2279 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
2280 wait_for_completion(&wait);
2281
2282 if (rq->errors)
2283 err = -EIO;
2284
2285 return err;
2286}
2287
2288EXPORT_SYMBOL(blk_execute_rq);
2289
2290static void bio_end_empty_barrier(struct bio *bio, int err)
2291{
2292 if (err)
2293 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2294
2295 complete(bio->bi_private);
2296}
2297
2298/**
2299 * blkdev_issue_flush - queue a flush
2300 * @bdev: blockdev to issue flush for
2301 * @error_sector: error sector
2302 *
2303 * Description:
2304 * Issue a flush for the block device in question. Caller can supply
2305 * room for storing the error offset in case of a flush error, if they
2306 * wish to. Caller must run wait_for_completion() on its own.
2307 */
2308int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2309{
2310 DECLARE_COMPLETION_ONSTACK(wait);
2311 struct request_queue *q;
2312 struct bio *bio;
2313 int ret;
2314
2315 if (bdev->bd_disk == NULL)
2316 return -ENXIO;
2317
2318 q = bdev_get_queue(bdev);
2319 if (!q)
2320 return -ENXIO;
2321
2322 bio = bio_alloc(GFP_KERNEL, 0);
2323 if (!bio)
2324 return -ENOMEM;
2325
2326 bio->bi_end_io = bio_end_empty_barrier;
2327 bio->bi_private = &wait;
2328 bio->bi_bdev = bdev;
2329 submit_bio(1 << BIO_RW_BARRIER, bio);
2330
2331 wait_for_completion(&wait);
2332
2333 /*
2334 * The driver must store the error location in ->bi_sector, if
2335 * it supports it. For non-stacked drivers, this should be copied
2336 * from rq->sector.
2337 */
2338 if (error_sector)
2339 *error_sector = bio->bi_sector;
2340
2341 ret = 0;
2342 if (!bio_flagged(bio, BIO_UPTODATE))
2343 ret = -EIO;
2344
2345 bio_put(bio);
2346 return ret;
2347}
2348
2349EXPORT_SYMBOL(blkdev_issue_flush);
2350
2351static void drive_stat_acct(struct request *rq, int new_io)
2352{
2353 int rw = rq_data_dir(rq);
2354
2355 if (!blk_fs_request(rq) || !rq->rq_disk)
2356 return;
2357
2358 if (!new_io) {

--- 95 unchanged lines hidden (view full) ---

2454 spin_lock_irqsave(q->queue_lock, flags);
2455 __blk_put_request(q, req);
2456 spin_unlock_irqrestore(q->queue_lock, flags);
2457 }
2458}
2459
2460EXPORT_SYMBOL(blk_put_request);
2461
1308static void drive_stat_acct(struct request *rq, int new_io)
1309{
1310 int rw = rq_data_dir(rq);
1311
1312 if (!blk_fs_request(rq) || !rq->rq_disk)
1313 return;
1314
1315 if (!new_io) {

--- 95 unchanged lines hidden (view full) ---

1411 spin_lock_irqsave(q->queue_lock, flags);
1412 __blk_put_request(q, req);
1413 spin_unlock_irqrestore(q->queue_lock, flags);
1414 }
1415}
1416
1417EXPORT_SYMBOL(blk_put_request);
1418
2462/**
2463 * blk_end_sync_rq - executes a completion event on a request
2464 * @rq: request to complete
2465 * @error: end io status of the request
2466 */
2467void blk_end_sync_rq(struct request *rq, int error)
2468{
2469 struct completion *waiting = rq->end_io_data;
2470
2471 rq->end_io_data = NULL;
2472 __blk_put_request(rq->q, rq);
2473
2474 /*
2475 * complete last, if this is a stack request the process (and thus
2476 * the rq pointer) could be invalid right after this complete()
2477 */
2478 complete(waiting);
2479}
2480EXPORT_SYMBOL(blk_end_sync_rq);
2481
2482/*
2483 * Has to be called with the request spinlock acquired
2484 */
2485static int attempt_merge(struct request_queue *q, struct request *req,
2486 struct request *next)
2487{
2488 if (!rq_mergeable(req) || !rq_mergeable(next))
2489 return 0;

--- 62 unchanged lines hidden (view full) ---

2552 struct request *prev = elv_former_request(q, rq);
2553
2554 if (prev)
2555 return attempt_merge(q, prev, rq);
2556
2557 return 0;
2558}
2559
1419/*
1420 * Has to be called with the request spinlock acquired
1421 */
1422static int attempt_merge(struct request_queue *q, struct request *req,
1423 struct request *next)
1424{
1425 if (!rq_mergeable(req) || !rq_mergeable(next))
1426 return 0;

--- 62 unchanged lines hidden (view full) ---

1489 struct request *prev = elv_former_request(q, rq);
1490
1491 if (prev)
1492 return attempt_merge(q, prev, rq);
1493
1494 return 0;
1495}
1496
2560static void init_request_from_bio(struct request *req, struct bio *bio)
1497void init_request_from_bio(struct request *req, struct bio *bio)
2561{
2562 req->cmd_type = REQ_TYPE_FS;
2563
2564 /*
2565 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2566 */
2567 if (bio_rw_ahead(bio) || bio_failfast(bio))
2568 req->cmd_flags |= REQ_FAILFAST;

--- 950 unchanged lines hidden (view full) ---

3519 **/
3520int blk_end_request_callback(struct request *rq, int error, int nr_bytes,
3521 int (drv_callback)(struct request *))
3522{
3523 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
3524}
3525EXPORT_SYMBOL_GPL(blk_end_request_callback);
3526
1498{
1499 req->cmd_type = REQ_TYPE_FS;
1500
1501 /*
1502 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
1503 */
1504 if (bio_rw_ahead(bio) || bio_failfast(bio))
1505 req->cmd_flags |= REQ_FAILFAST;

--- 950 unchanged lines hidden (view full) ---

2456 **/
2457int blk_end_request_callback(struct request *rq, int error, int nr_bytes,
2458 int (drv_callback)(struct request *))
2459{
2460 return blk_end_io(rq, error, nr_bytes, 0, drv_callback);
2461}
2462EXPORT_SYMBOL_GPL(blk_end_request_callback);
2463
3527static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3528 struct bio *bio)
2464void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2465 struct bio *bio)
3529{
3530 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3531 rq->cmd_flags |= (bio->bi_rw & 3);
3532
3533 rq->nr_phys_segments = bio_phys_segments(q, bio);
3534 rq->nr_hw_segments = bio_hw_segments(q, bio);
3535 rq->current_nr_sectors = bio_cur_sectors(bio);
3536 rq->hard_cur_sectors = rq->current_nr_sectors;

--- 29 unchanged lines hidden (view full) ---

3566 panic("Failed to create kblockd\n");
3567
3568 request_cachep = kmem_cache_create("blkdev_requests",
3569 sizeof(struct request), 0, SLAB_PANIC, NULL);
3570
3571 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3572 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3573
2466{
2467 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
2468 rq->cmd_flags |= (bio->bi_rw & 3);
2469
2470 rq->nr_phys_segments = bio_phys_segments(q, bio);
2471 rq->nr_hw_segments = bio_hw_segments(q, bio);
2472 rq->current_nr_sectors = bio_cur_sectors(bio);
2473 rq->hard_cur_sectors = rq->current_nr_sectors;

--- 29 unchanged lines hidden (view full) ---

2503 panic("Failed to create kblockd\n");
2504
2505 request_cachep = kmem_cache_create("blkdev_requests",
2506 sizeof(struct request), 0, SLAB_PANIC, NULL);
2507
2508 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
2509 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
2510
3574 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3575 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
3576
3577 for_each_possible_cpu(i)
3578 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3579
3580 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
3581 register_hotcpu_notifier(&blk_cpu_notifier);
3582
2511 for_each_possible_cpu(i)
2512 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
2513
2514 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
2515 register_hotcpu_notifier(&blk_cpu_notifier);
2516
3583 blk_max_low_pfn = max_low_pfn - 1;
3584 blk_max_pfn = max_pfn - 1;
3585
3586 return 0;
3587}
3588
2517 return 0;
2518}
2519
3589static void cfq_dtor(struct io_context *ioc)
3590{
3591 struct cfq_io_context *cic[1];
3592 int r;
3593
3594 /*
3595 * We don't have a specific key to lookup with, so use the gang
3596 * lookup to just retrieve the first item stored. The cfq exit
3597 * function will iterate the full tree, so any member will do.
3598 */
3599 r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
3600 if (r > 0)
3601 cic[0]->dtor(ioc);
3602}
3603
3604/*
3605 * IO Context helper functions. put_io_context() returns 1 if there are no
3606 * more users of this io context, 0 otherwise.
3607 */
3608int put_io_context(struct io_context *ioc)
3609{
3610 if (ioc == NULL)
3611 return 1;
3612
3613 BUG_ON(atomic_read(&ioc->refcount) == 0);
3614
3615 if (atomic_dec_and_test(&ioc->refcount)) {
3616 rcu_read_lock();
3617 if (ioc->aic && ioc->aic->dtor)
3618 ioc->aic->dtor(ioc->aic);
3619 rcu_read_unlock();
3620 cfq_dtor(ioc);
3621
3622 kmem_cache_free(iocontext_cachep, ioc);
3623 return 1;
3624 }
3625 return 0;
3626}
3627EXPORT_SYMBOL(put_io_context);
3628
3629static void cfq_exit(struct io_context *ioc)
3630{
3631 struct cfq_io_context *cic[1];
3632 int r;
3633
3634 rcu_read_lock();
3635 /*
3636 * See comment for cfq_dtor()
3637 */
3638 r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
3639 rcu_read_unlock();
3640
3641 if (r > 0)
3642 cic[0]->exit(ioc);
3643}
3644
3645/* Called by the exitting task */
3646void exit_io_context(void)
3647{
3648 struct io_context *ioc;
3649
3650 task_lock(current);
3651 ioc = current->io_context;
3652 current->io_context = NULL;
3653 task_unlock(current);
3654
3655 if (atomic_dec_and_test(&ioc->nr_tasks)) {
3656 if (ioc->aic && ioc->aic->exit)
3657 ioc->aic->exit(ioc->aic);
3658 cfq_exit(ioc);
3659
3660 put_io_context(ioc);
3661 }
3662}
3663
3664struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
3665{
3666 struct io_context *ret;
3667
3668 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
3669 if (ret) {
3670 atomic_set(&ret->refcount, 1);
3671 atomic_set(&ret->nr_tasks, 1);
3672 spin_lock_init(&ret->lock);
3673 ret->ioprio_changed = 0;
3674 ret->ioprio = 0;
3675 ret->last_waited = jiffies; /* doesn't matter... */
3676 ret->nr_batch_requests = 0; /* because this is 0 */
3677 ret->aic = NULL;
3678 INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
3679 ret->ioc_data = NULL;
3680 }
3681
3682 return ret;
3683}
3684
3685/*
3686 * If the current task has no IO context then create one and initialise it.
3687 * Otherwise, return its existing IO context.
3688 *
3689 * This returned IO context doesn't have a specifically elevated refcount,
3690 * but since the current task itself holds a reference, the context can be
3691 * used in general code, so long as it stays within `current` context.
3692 */
3693static struct io_context *current_io_context(gfp_t gfp_flags, int node)
3694{
3695 struct task_struct *tsk = current;
3696 struct io_context *ret;
3697
3698 ret = tsk->io_context;
3699 if (likely(ret))
3700 return ret;
3701
3702 ret = alloc_io_context(gfp_flags, node);
3703 if (ret) {
3704 /* make sure set_task_ioprio() sees the settings above */
3705 smp_wmb();
3706 tsk->io_context = ret;
3707 }
3708
3709 return ret;
3710}
3711
3712/*
3713 * If the current task has no IO context then create one and initialise it.
3714 * If it does have a context, take a ref on it.
3715 *
3716 * This is always called in the context of the task which submitted the I/O.
3717 */
3718struct io_context *get_io_context(gfp_t gfp_flags, int node)
3719{
3720 struct io_context *ret = NULL;
3721
3722 /*
3723 * Check for unlikely race with exiting task. ioc ref count is
3724 * zero when ioc is being detached.
3725 */
3726 do {
3727 ret = current_io_context(gfp_flags, node);
3728 if (unlikely(!ret))
3729 break;
3730 } while (!atomic_inc_not_zero(&ret->refcount));
3731
3732 return ret;
3733}
3734EXPORT_SYMBOL(get_io_context);
3735
3736void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3737{
3738 struct io_context *src = *psrc;
3739 struct io_context *dst = *pdst;
3740
3741 if (src) {
3742 BUG_ON(atomic_read(&src->refcount) == 0);
3743 atomic_inc(&src->refcount);
3744 put_io_context(dst);
3745 *pdst = src;
3746 }
3747}
3748EXPORT_SYMBOL(copy_io_context);
3749
3750void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3751{
3752 struct io_context *temp;
3753 temp = *ioc1;
3754 *ioc1 = *ioc2;
3755 *ioc2 = temp;
3756}
3757EXPORT_SYMBOL(swap_io_context);
3758