blob: c525b5a2b59849d7af6fb3fe45e8a8699711e085 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * 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 */
13#include <linux/config.h>
14#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
24#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
Christoph Lameter19460892005-06-23 00:08:19 -070029#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31/*
32 * for max sense size
33 */
34#include <scsi/scsi_cmnd.h>
35
36static void blk_unplug_work(void *data);
37static void blk_unplug_timeout(unsigned long data);
Adrian Bunk93d17d32005-06-25 14:59:10 -070038static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40/*
41 * For the allocated request tables
42 */
43static kmem_cache_t *request_cachep;
44
45/*
46 * For queue allocation
47 */
48static kmem_cache_t *requestq_cachep;
49
50/*
51 * For io context allocations
52 */
53static kmem_cache_t *iocontext_cachep;
54
55static wait_queue_head_t congestion_wqh[2] = {
56 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
57 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
58 };
59
60/*
61 * Controlling structure to kblockd
62 */
63static struct workqueue_struct *kblockd_workqueue;
64
65unsigned long blk_max_low_pfn, blk_max_pfn;
66
67EXPORT_SYMBOL(blk_max_low_pfn);
68EXPORT_SYMBOL(blk_max_pfn);
69
70/* Amount of time in which a process may batch requests */
71#define BLK_BATCH_TIME (HZ/50UL)
72
73/* Number of requests a "batching" process may submit */
74#define BLK_BATCH_REQ 32
75
76/*
77 * Return the threshold (number of used requests) at which the queue is
78 * considered to be congested. It include a little hysteresis to keep the
79 * context switch rate down.
80 */
81static inline int queue_congestion_on_threshold(struct request_queue *q)
82{
83 return q->nr_congestion_on;
84}
85
86/*
87 * The threshold at which a queue is considered to be uncongested
88 */
89static inline int queue_congestion_off_threshold(struct request_queue *q)
90{
91 return q->nr_congestion_off;
92}
93
94static void blk_queue_congestion_threshold(struct request_queue *q)
95{
96 int nr;
97
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
100 nr = q->nr_requests;
101 q->nr_congestion_on = nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
104 if (nr < 1)
105 nr = 1;
106 q->nr_congestion_off = nr;
107}
108
109/*
110 * A queue has just exitted congestion. Note this in the global counter of
111 * congested queues, and wake up anyone who was waiting for requests to be
112 * put back.
113 */
114static void clear_queue_congested(request_queue_t *q, int rw)
115{
116 enum bdi_state bit;
117 wait_queue_head_t *wqh = &congestion_wqh[rw];
118
119 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
120 clear_bit(bit, &q->backing_dev_info.state);
121 smp_mb__after_clear_bit();
122 if (waitqueue_active(wqh))
123 wake_up(wqh);
124}
125
126/*
127 * A queue has just entered congestion. Flag that in the queue's VM-visible
128 * state flags and increment the global gounter of congested queues.
129 */
130static void set_queue_congested(request_queue_t *q, int rw)
131{
132 enum bdi_state bit;
133
134 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
135 set_bit(bit, &q->backing_dev_info.state);
136}
137
138/**
139 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
140 * @bdev: device
141 *
142 * Locates the passed device's request queue and returns the address of its
143 * backing_dev_info
144 *
145 * Will return NULL if the request queue cannot be located.
146 */
147struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
148{
149 struct backing_dev_info *ret = NULL;
150 request_queue_t *q = bdev_get_queue(bdev);
151
152 if (q)
153 ret = &q->backing_dev_info;
154 return ret;
155}
156
157EXPORT_SYMBOL(blk_get_backing_dev_info);
158
159void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
160{
161 q->activity_fn = fn;
162 q->activity_data = data;
163}
164
165EXPORT_SYMBOL(blk_queue_activity_fn);
166
167/**
168 * blk_queue_prep_rq - set a prepare_request function for queue
169 * @q: queue
170 * @pfn: prepare_request function
171 *
172 * It's possible for a queue to register a prepare_request callback which
173 * is invoked before the request is handed to the request_fn. The goal of
174 * the function is to prepare a request for I/O, it can be used to build a
175 * cdb from the request data for instance.
176 *
177 */
178void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
179{
180 q->prep_rq_fn = pfn;
181}
182
183EXPORT_SYMBOL(blk_queue_prep_rq);
184
185/**
186 * blk_queue_merge_bvec - set a merge_bvec function for queue
187 * @q: queue
188 * @mbfn: merge_bvec_fn
189 *
190 * Usually queues have static limitations on the max sectors or segments that
191 * we can put in a request. Stacking drivers may have some settings that
192 * are dynamic, and thus we have to query the queue whether it is ok to
193 * add a new bio_vec to a bio at a given offset or not. If the block device
194 * has such limitations, it needs to register a merge_bvec_fn to control
195 * the size of bio's sent to it. Note that a block device *must* allow a
196 * single page to be added to an empty bio. The block device driver may want
197 * to use the bio_split() function to deal with these bio's. By default
198 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
199 * honored.
200 */
201void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
202{
203 q->merge_bvec_fn = mbfn;
204}
205
206EXPORT_SYMBOL(blk_queue_merge_bvec);
207
208/**
209 * blk_queue_make_request - define an alternate make_request function for a device
210 * @q: the request queue for the device to be affected
211 * @mfn: the alternate make_request function
212 *
213 * Description:
214 * The normal way for &struct bios to be passed to a device
215 * driver is for them to be collected into requests on a request
216 * queue, and then to allow the device driver to select requests
217 * off that queue when it is ready. This works well for many block
218 * devices. However some block devices (typically virtual devices
219 * such as md or lvm) do not benefit from the processing on the
220 * request queue, and are served best by having the requests passed
221 * directly to them. This can be achieved by providing a function
222 * to blk_queue_make_request().
223 *
224 * Caveat:
225 * The driver that does this *must* be able to deal appropriately
226 * with buffers in "highmemory". This can be accomplished by either calling
227 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
228 * blk_queue_bounce() to create a buffer in normal memory.
229 **/
230void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
231{
232 /*
233 * set defaults
234 */
235 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700236 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
237 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 q->make_request_fn = mfn;
239 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
240 q->backing_dev_info.state = 0;
241 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
242 blk_queue_max_sectors(q, MAX_SECTORS);
243 blk_queue_hardsect_size(q, 512);
244 blk_queue_dma_alignment(q, 511);
245 blk_queue_congestion_threshold(q);
246 q->nr_batching = BLK_BATCH_REQ;
247
248 q->unplug_thresh = 4; /* hmm */
249 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
250 if (q->unplug_delay == 0)
251 q->unplug_delay = 1;
252
253 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
254
255 q->unplug_timer.function = blk_unplug_timeout;
256 q->unplug_timer.data = (unsigned long)q;
257
258 /*
259 * by default assume old behaviour and bounce for any highmem page
260 */
261 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
262
263 blk_queue_activity_fn(q, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266EXPORT_SYMBOL(blk_queue_make_request);
267
268static inline void rq_init(request_queue_t *q, struct request *rq)
269{
270 INIT_LIST_HEAD(&rq->queuelist);
271
272 rq->errors = 0;
273 rq->rq_status = RQ_ACTIVE;
274 rq->bio = rq->biotail = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +0200275 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 rq->buffer = NULL;
277 rq->ref_count = 1;
278 rq->q = q;
279 rq->waiting = NULL;
280 rq->special = NULL;
281 rq->data_len = 0;
282 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200283 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 rq->sense = NULL;
285 rq->end_io = NULL;
286 rq->end_io_data = NULL;
287}
288
289/**
290 * blk_queue_ordered - does this queue support ordered writes
291 * @q: the request queue
292 * @flag: see below
293 *
294 * Description:
295 * For journalled file systems, doing ordered writes on a commit
296 * block instead of explicitly doing wait_on_buffer (which is bad
297 * for performance) can be a big win. Block drivers supporting this
298 * feature should call this function and indicate so.
299 *
300 **/
301void blk_queue_ordered(request_queue_t *q, int flag)
302{
303 switch (flag) {
304 case QUEUE_ORDERED_NONE:
305 if (q->flush_rq)
306 kmem_cache_free(request_cachep, q->flush_rq);
307 q->flush_rq = NULL;
308 q->ordered = flag;
309 break;
310 case QUEUE_ORDERED_TAG:
311 q->ordered = flag;
312 break;
313 case QUEUE_ORDERED_FLUSH:
314 q->ordered = flag;
315 if (!q->flush_rq)
316 q->flush_rq = kmem_cache_alloc(request_cachep,
317 GFP_KERNEL);
318 break;
319 default:
320 printk("blk_queue_ordered: bad value %d\n", flag);
321 break;
322 }
323}
324
325EXPORT_SYMBOL(blk_queue_ordered);
326
327/**
328 * blk_queue_issue_flush_fn - set function for issuing a flush
329 * @q: the request queue
330 * @iff: the function to be called issuing the flush
331 *
332 * Description:
333 * If a driver supports issuing a flush command, the support is notified
334 * to the block layer by defining it through this call.
335 *
336 **/
337void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
338{
339 q->issue_flush_fn = iff;
340}
341
342EXPORT_SYMBOL(blk_queue_issue_flush_fn);
343
344/*
345 * Cache flushing for ordered writes handling
346 */
347static void blk_pre_flush_end_io(struct request *flush_rq)
348{
349 struct request *rq = flush_rq->end_io_data;
350 request_queue_t *q = rq->q;
351
Tejun Heo8922e162005-10-20 16:23:44 +0200352 elv_completed_request(q, flush_rq);
353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 rq->flags |= REQ_BAR_PREFLUSH;
355
356 if (!flush_rq->errors)
357 elv_requeue_request(q, rq);
358 else {
359 q->end_flush_fn(q, flush_rq);
360 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
361 q->request_fn(q);
362 }
363}
364
365static void blk_post_flush_end_io(struct request *flush_rq)
366{
367 struct request *rq = flush_rq->end_io_data;
368 request_queue_t *q = rq->q;
369
Tejun Heo8922e162005-10-20 16:23:44 +0200370 elv_completed_request(q, flush_rq);
371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 rq->flags |= REQ_BAR_POSTFLUSH;
373
374 q->end_flush_fn(q, flush_rq);
375 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
376 q->request_fn(q);
377}
378
379struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq)
380{
381 struct request *flush_rq = q->flush_rq;
382
383 BUG_ON(!blk_barrier_rq(rq));
384
385 if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags))
386 return NULL;
387
388 rq_init(q, flush_rq);
389 flush_rq->elevator_private = NULL;
390 flush_rq->flags = REQ_BAR_FLUSH;
391 flush_rq->rq_disk = rq->rq_disk;
392 flush_rq->rl = NULL;
393
394 /*
395 * prepare_flush returns 0 if no flush is needed, just mark both
396 * pre and post flush as done in that case
397 */
398 if (!q->prepare_flush_fn(q, flush_rq)) {
399 rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH;
400 clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
401 return rq;
402 }
403
404 /*
405 * some drivers dequeue requests right away, some only after io
406 * completion. make sure the request is dequeued.
407 */
408 if (!list_empty(&rq->queuelist))
409 blkdev_dequeue_request(rq);
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 flush_rq->end_io_data = rq;
412 flush_rq->end_io = blk_pre_flush_end_io;
413
414 __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
415 return flush_rq;
416}
417
418static void blk_start_post_flush(request_queue_t *q, struct request *rq)
419{
420 struct request *flush_rq = q->flush_rq;
421
422 BUG_ON(!blk_barrier_rq(rq));
423
424 rq_init(q, flush_rq);
425 flush_rq->elevator_private = NULL;
426 flush_rq->flags = REQ_BAR_FLUSH;
427 flush_rq->rq_disk = rq->rq_disk;
428 flush_rq->rl = NULL;
429
430 if (q->prepare_flush_fn(q, flush_rq)) {
431 flush_rq->end_io_data = rq;
432 flush_rq->end_io = blk_post_flush_end_io;
433
434 __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
435 q->request_fn(q);
436 }
437}
438
439static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq,
440 int sectors)
441{
442 if (sectors > rq->nr_sectors)
443 sectors = rq->nr_sectors;
444
445 rq->nr_sectors -= sectors;
446 return rq->nr_sectors;
447}
448
449static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq,
450 int sectors, int queue_locked)
451{
452 if (q->ordered != QUEUE_ORDERED_FLUSH)
453 return 0;
454 if (!blk_fs_request(rq) || !blk_barrier_rq(rq))
455 return 0;
456 if (blk_barrier_postflush(rq))
457 return 0;
458
459 if (!blk_check_end_barrier(q, rq, sectors)) {
460 unsigned long flags = 0;
461
462 if (!queue_locked)
463 spin_lock_irqsave(q->queue_lock, flags);
464
465 blk_start_post_flush(q, rq);
466
467 if (!queue_locked)
468 spin_unlock_irqrestore(q->queue_lock, flags);
469 }
470
471 return 1;
472}
473
474/**
475 * blk_complete_barrier_rq - complete possible barrier request
476 * @q: the request queue for the device
477 * @rq: the request
478 * @sectors: number of sectors to complete
479 *
480 * Description:
481 * Used in driver end_io handling to determine whether to postpone
482 * completion of a barrier request until a post flush has been done. This
483 * is the unlocked variant, used if the caller doesn't already hold the
484 * queue lock.
485 **/
486int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors)
487{
488 return __blk_complete_barrier_rq(q, rq, sectors, 0);
489}
490EXPORT_SYMBOL(blk_complete_barrier_rq);
491
492/**
493 * blk_complete_barrier_rq_locked - complete possible barrier request
494 * @q: the request queue for the device
495 * @rq: the request
496 * @sectors: number of sectors to complete
497 *
498 * Description:
499 * See blk_complete_barrier_rq(). This variant must be used if the caller
500 * holds the queue lock.
501 **/
502int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq,
503 int sectors)
504{
505 return __blk_complete_barrier_rq(q, rq, sectors, 1);
506}
507EXPORT_SYMBOL(blk_complete_barrier_rq_locked);
508
509/**
510 * blk_queue_bounce_limit - set bounce buffer limit for queue
511 * @q: the request queue for the device
512 * @dma_addr: bus address limit
513 *
514 * Description:
515 * Different hardware can have different requirements as to what pages
516 * it can do I/O directly to. A low level driver can call
517 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
518 * buffers for doing I/O to pages residing above @page. By default
519 * the block layer sets this to the highest numbered "low" memory page.
520 **/
521void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
522{
523 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
524
525 /*
526 * set appropriate bounce gfp mask -- unfortunately we don't have a
527 * full 4GB zone, so we have to resort to low memory for any bounces.
528 * ISA has its own < 16MB zone.
529 */
530 if (bounce_pfn < blk_max_low_pfn) {
531 BUG_ON(dma_addr < BLK_BOUNCE_ISA);
532 init_emergency_isa_pool();
533 q->bounce_gfp = GFP_NOIO | GFP_DMA;
534 } else
535 q->bounce_gfp = GFP_NOIO;
536
537 q->bounce_pfn = bounce_pfn;
538}
539
540EXPORT_SYMBOL(blk_queue_bounce_limit);
541
542/**
543 * blk_queue_max_sectors - set max sectors for a request for this queue
544 * @q: the request queue for the device
545 * @max_sectors: max sectors in the usual 512b unit
546 *
547 * Description:
548 * Enables a low level driver to set an upper limit on the size of
549 * received requests.
550 **/
551void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors)
552{
553 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
554 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
555 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
556 }
557
558 q->max_sectors = q->max_hw_sectors = max_sectors;
559}
560
561EXPORT_SYMBOL(blk_queue_max_sectors);
562
563/**
564 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
565 * @q: the request queue for the device
566 * @max_segments: max number of segments
567 *
568 * Description:
569 * Enables a low level driver to set an upper limit on the number of
570 * physical data segments in a request. This would be the largest sized
571 * scatter list the driver could handle.
572 **/
573void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
574{
575 if (!max_segments) {
576 max_segments = 1;
577 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
578 }
579
580 q->max_phys_segments = max_segments;
581}
582
583EXPORT_SYMBOL(blk_queue_max_phys_segments);
584
585/**
586 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
587 * @q: the request queue for the device
588 * @max_segments: max number of segments
589 *
590 * Description:
591 * Enables a low level driver to set an upper limit on the number of
592 * hw data segments in a request. This would be the largest number of
593 * address/length pairs the host adapter can actually give as once
594 * to the device.
595 **/
596void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
597{
598 if (!max_segments) {
599 max_segments = 1;
600 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
601 }
602
603 q->max_hw_segments = max_segments;
604}
605
606EXPORT_SYMBOL(blk_queue_max_hw_segments);
607
608/**
609 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
610 * @q: the request queue for the device
611 * @max_size: max size of segment in bytes
612 *
613 * Description:
614 * Enables a low level driver to set an upper limit on the size of a
615 * coalesced segment
616 **/
617void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
618{
619 if (max_size < PAGE_CACHE_SIZE) {
620 max_size = PAGE_CACHE_SIZE;
621 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
622 }
623
624 q->max_segment_size = max_size;
625}
626
627EXPORT_SYMBOL(blk_queue_max_segment_size);
628
629/**
630 * blk_queue_hardsect_size - set hardware sector size for the queue
631 * @q: the request queue for the device
632 * @size: the hardware sector size, in bytes
633 *
634 * Description:
635 * This should typically be set to the lowest possible sector size
636 * that the hardware can operate on (possible without reverting to
637 * even internal read-modify-write operations). Usually the default
638 * of 512 covers most hardware.
639 **/
640void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
641{
642 q->hardsect_size = size;
643}
644
645EXPORT_SYMBOL(blk_queue_hardsect_size);
646
647/*
648 * Returns the minimum that is _not_ zero, unless both are zero.
649 */
650#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
651
652/**
653 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
654 * @t: the stacking driver (top)
655 * @b: the underlying device (bottom)
656 **/
657void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
658{
659 /* zero is "infinity" */
660 t->max_sectors = t->max_hw_sectors =
661 min_not_zero(t->max_sectors,b->max_sectors);
662
663 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
664 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
665 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
666 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
667}
668
669EXPORT_SYMBOL(blk_queue_stack_limits);
670
671/**
672 * blk_queue_segment_boundary - set boundary rules for segment merging
673 * @q: the request queue for the device
674 * @mask: the memory boundary mask
675 **/
676void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
677{
678 if (mask < PAGE_CACHE_SIZE - 1) {
679 mask = PAGE_CACHE_SIZE - 1;
680 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
681 }
682
683 q->seg_boundary_mask = mask;
684}
685
686EXPORT_SYMBOL(blk_queue_segment_boundary);
687
688/**
689 * blk_queue_dma_alignment - set dma length and memory alignment
690 * @q: the request queue for the device
691 * @mask: alignment mask
692 *
693 * description:
694 * set required memory and length aligment for direct dma transactions.
695 * this is used when buiding direct io requests for the queue.
696 *
697 **/
698void blk_queue_dma_alignment(request_queue_t *q, int mask)
699{
700 q->dma_alignment = mask;
701}
702
703EXPORT_SYMBOL(blk_queue_dma_alignment);
704
705/**
706 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 * @q: The request queue for the device
708 * @tag: The tag of the request
709 *
710 * Notes:
711 * Should be used when a device returns a tag and you want to match
712 * it with a request.
713 *
714 * no locks need be held.
715 **/
716struct request *blk_queue_find_tag(request_queue_t *q, int tag)
717{
718 struct blk_queue_tag *bqt = q->queue_tags;
719
Tejun Heoba025082005-08-05 13:28:11 -0700720 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 return NULL;
722
723 return bqt->tag_index[tag];
724}
725
726EXPORT_SYMBOL(blk_queue_find_tag);
727
728/**
729 * __blk_queue_free_tags - release tag maintenance info
730 * @q: the request queue for the device
731 *
732 * Notes:
733 * blk_cleanup_queue() will take care of calling this function, if tagging
734 * has been used. So there's no need to call this directly.
735 **/
736static void __blk_queue_free_tags(request_queue_t *q)
737{
738 struct blk_queue_tag *bqt = q->queue_tags;
739
740 if (!bqt)
741 return;
742
743 if (atomic_dec_and_test(&bqt->refcnt)) {
744 BUG_ON(bqt->busy);
745 BUG_ON(!list_empty(&bqt->busy_list));
746
747 kfree(bqt->tag_index);
748 bqt->tag_index = NULL;
749
750 kfree(bqt->tag_map);
751 bqt->tag_map = NULL;
752
753 kfree(bqt);
754 }
755
756 q->queue_tags = NULL;
757 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
758}
759
760/**
761 * blk_queue_free_tags - release tag maintenance info
762 * @q: the request queue for the device
763 *
764 * Notes:
765 * This is used to disabled tagged queuing to a device, yet leave
766 * queue in function.
767 **/
768void blk_queue_free_tags(request_queue_t *q)
769{
770 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
771}
772
773EXPORT_SYMBOL(blk_queue_free_tags);
774
775static int
776init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
777{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 struct request **tag_index;
779 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700780 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
782 if (depth > q->nr_requests * 2) {
783 depth = q->nr_requests * 2;
784 printk(KERN_ERR "%s: adjusted depth to %d\n",
785 __FUNCTION__, depth);
786 }
787
788 tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
789 if (!tag_index)
790 goto fail;
791
Tejun Heof7d37d02005-06-23 00:08:50 -0700792 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700793 tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 if (!tag_map)
795 goto fail;
796
797 memset(tag_index, 0, depth * sizeof(struct request *));
Tejun Heofa72b902005-06-23 00:08:49 -0700798 memset(tag_map, 0, nr_ulongs * sizeof(unsigned long));
Tejun Heoba025082005-08-05 13:28:11 -0700799 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 tags->tag_index = tag_index;
802 tags->tag_map = tag_map;
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return 0;
805fail:
806 kfree(tag_index);
807 return -ENOMEM;
808}
809
810/**
811 * blk_queue_init_tags - initialize the queue tag info
812 * @q: the request queue for the device
813 * @depth: the maximum queue depth supported
814 * @tags: the tag to use
815 **/
816int blk_queue_init_tags(request_queue_t *q, int depth,
817 struct blk_queue_tag *tags)
818{
819 int rc;
820
821 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
822
823 if (!tags && !q->queue_tags) {
824 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
825 if (!tags)
826 goto fail;
827
828 if (init_tag_map(q, tags, depth))
829 goto fail;
830
831 INIT_LIST_HEAD(&tags->busy_list);
832 tags->busy = 0;
833 atomic_set(&tags->refcnt, 1);
834 } else if (q->queue_tags) {
835 if ((rc = blk_queue_resize_tags(q, depth)))
836 return rc;
837 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
838 return 0;
839 } else
840 atomic_inc(&tags->refcnt);
841
842 /*
843 * assign it, all done
844 */
845 q->queue_tags = tags;
846 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
847 return 0;
848fail:
849 kfree(tags);
850 return -ENOMEM;
851}
852
853EXPORT_SYMBOL(blk_queue_init_tags);
854
855/**
856 * blk_queue_resize_tags - change the queueing depth
857 * @q: the request queue for the device
858 * @new_depth: the new max command queueing depth
859 *
860 * Notes:
861 * Must be called with the queue lock held.
862 **/
863int blk_queue_resize_tags(request_queue_t *q, int new_depth)
864{
865 struct blk_queue_tag *bqt = q->queue_tags;
866 struct request **tag_index;
867 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700868 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870 if (!bqt)
871 return -ENXIO;
872
873 /*
Tejun Heoba025082005-08-05 13:28:11 -0700874 * if we already have large enough real_max_depth. just
875 * adjust max_depth. *NOTE* as requests with tag value
876 * between new_depth and real_max_depth can be in-flight, tag
877 * map can not be shrunk blindly here.
878 */
879 if (new_depth <= bqt->real_max_depth) {
880 bqt->max_depth = new_depth;
881 return 0;
882 }
883
884 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 * save the old state info, so we can copy it back
886 */
887 tag_index = bqt->tag_index;
888 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -0700889 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 if (init_tag_map(q, bqt, new_depth))
892 return -ENOMEM;
893
894 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -0700895 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700896 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 kfree(tag_index);
899 kfree(tag_map);
900 return 0;
901}
902
903EXPORT_SYMBOL(blk_queue_resize_tags);
904
905/**
906 * blk_queue_end_tag - end tag operations for a request
907 * @q: the request queue for the device
908 * @rq: the request that has completed
909 *
910 * Description:
911 * Typically called when end_that_request_first() returns 0, meaning
912 * all transfers have been done for a request. It's important to call
913 * this function before end_that_request_last(), as that will put the
914 * request back on the free list thus corrupting the internal tag list.
915 *
916 * Notes:
917 * queue lock must be held.
918 **/
919void blk_queue_end_tag(request_queue_t *q, struct request *rq)
920{
921 struct blk_queue_tag *bqt = q->queue_tags;
922 int tag = rq->tag;
923
924 BUG_ON(tag == -1);
925
Tejun Heoba025082005-08-05 13:28:11 -0700926 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -0700927 /*
928 * This can happen after tag depth has been reduced.
929 * FIXME: how about a warning or info message here?
930 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 return;
932
933 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -0700934 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
935 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return;
937 }
938
939 list_del_init(&rq->queuelist);
940 rq->flags &= ~REQ_QUEUED;
941 rq->tag = -1;
942
943 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -0700944 printk(KERN_ERR "%s: tag %d is missing\n",
945 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
947 bqt->tag_index[tag] = NULL;
948 bqt->busy--;
949}
950
951EXPORT_SYMBOL(blk_queue_end_tag);
952
953/**
954 * blk_queue_start_tag - find a free tag and assign it
955 * @q: the request queue for the device
956 * @rq: the block request that needs tagging
957 *
958 * Description:
959 * This can either be used as a stand-alone helper, or possibly be
960 * assigned as the queue &prep_rq_fn (in which case &struct request
961 * automagically gets a tag assigned). Note that this function
962 * assumes that any type of request can be queued! if this is not
963 * true for your device, you must check the request type before
964 * calling this function. The request will also be removed from
965 * the request queue, so it's the drivers responsibility to readd
966 * it if it should need to be restarted for some reason.
967 *
968 * Notes:
969 * queue lock must be held.
970 **/
971int blk_queue_start_tag(request_queue_t *q, struct request *rq)
972{
973 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -0700974 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 if (unlikely((rq->flags & REQ_QUEUED))) {
977 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -0700978 "%s: request %p for device [%s] already tagged %d",
979 __FUNCTION__, rq,
980 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 BUG();
982 }
983
Tejun Heo2bf0fda2005-06-23 00:08:48 -0700984 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
985 if (tag >= bqt->max_depth)
986 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 __set_bit(tag, bqt->tag_map);
989
990 rq->flags |= REQ_QUEUED;
991 rq->tag = tag;
992 bqt->tag_index[tag] = rq;
993 blkdev_dequeue_request(rq);
994 list_add(&rq->queuelist, &bqt->busy_list);
995 bqt->busy++;
996 return 0;
997}
998
999EXPORT_SYMBOL(blk_queue_start_tag);
1000
1001/**
1002 * blk_queue_invalidate_tags - invalidate all pending tags
1003 * @q: the request queue for the device
1004 *
1005 * Description:
1006 * Hardware conditions may dictate a need to stop all pending requests.
1007 * In this case, we will safely clear the block side of the tag queue and
1008 * readd all requests to the request queue in the right order.
1009 *
1010 * Notes:
1011 * queue lock must be held.
1012 **/
1013void blk_queue_invalidate_tags(request_queue_t *q)
1014{
1015 struct blk_queue_tag *bqt = q->queue_tags;
1016 struct list_head *tmp, *n;
1017 struct request *rq;
1018
1019 list_for_each_safe(tmp, n, &bqt->busy_list) {
1020 rq = list_entry_rq(tmp);
1021
1022 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001023 printk(KERN_ERR
1024 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 list_del_init(&rq->queuelist);
1026 rq->flags &= ~REQ_QUEUED;
1027 } else
1028 blk_queue_end_tag(q, rq);
1029
1030 rq->flags &= ~REQ_STARTED;
1031 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1032 }
1033}
1034
1035EXPORT_SYMBOL(blk_queue_invalidate_tags);
1036
1037static char *rq_flags[] = {
1038 "REQ_RW",
1039 "REQ_FAILFAST",
Tejun Heo8922e162005-10-20 16:23:44 +02001040 "REQ_SORTED",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 "REQ_SOFTBARRIER",
1042 "REQ_HARDBARRIER",
1043 "REQ_CMD",
1044 "REQ_NOMERGE",
1045 "REQ_STARTED",
1046 "REQ_DONTPREP",
1047 "REQ_QUEUED",
Tejun Heocb98fc82005-10-28 08:29:39 +02001048 "REQ_ELVPRIV",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 "REQ_PC",
1050 "REQ_BLOCK_PC",
1051 "REQ_SENSE",
1052 "REQ_FAILED",
1053 "REQ_QUIET",
1054 "REQ_SPECIAL",
1055 "REQ_DRIVE_CMD",
1056 "REQ_DRIVE_TASK",
1057 "REQ_DRIVE_TASKFILE",
1058 "REQ_PREEMPT",
1059 "REQ_PM_SUSPEND",
1060 "REQ_PM_RESUME",
1061 "REQ_PM_SHUTDOWN",
1062};
1063
1064void blk_dump_rq_flags(struct request *rq, char *msg)
1065{
1066 int bit;
1067
1068 printk("%s: dev %s: flags = ", msg,
1069 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1070 bit = 0;
1071 do {
1072 if (rq->flags & (1 << bit))
1073 printk("%s ", rq_flags[bit]);
1074 bit++;
1075 } while (bit < __REQ_NR_BITS);
1076
1077 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1078 rq->nr_sectors,
1079 rq->current_nr_sectors);
1080 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1081
1082 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1083 printk("cdb: ");
1084 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1085 printk("%02x ", rq->cmd[bit]);
1086 printk("\n");
1087 }
1088}
1089
1090EXPORT_SYMBOL(blk_dump_rq_flags);
1091
1092void blk_recount_segments(request_queue_t *q, struct bio *bio)
1093{
1094 struct bio_vec *bv, *bvprv = NULL;
1095 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1096 int high, highprv = 1;
1097
1098 if (unlikely(!bio->bi_io_vec))
1099 return;
1100
1101 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1102 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1103 bio_for_each_segment(bv, bio, i) {
1104 /*
1105 * the trick here is making sure that a high page is never
1106 * considered part of another segment, since that might
1107 * change with the bounce page.
1108 */
1109 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1110 if (high || highprv)
1111 goto new_hw_segment;
1112 if (cluster) {
1113 if (seg_size + bv->bv_len > q->max_segment_size)
1114 goto new_segment;
1115 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1116 goto new_segment;
1117 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1118 goto new_segment;
1119 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1120 goto new_hw_segment;
1121
1122 seg_size += bv->bv_len;
1123 hw_seg_size += bv->bv_len;
1124 bvprv = bv;
1125 continue;
1126 }
1127new_segment:
1128 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1129 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1130 hw_seg_size += bv->bv_len;
1131 } else {
1132new_hw_segment:
1133 if (hw_seg_size > bio->bi_hw_front_size)
1134 bio->bi_hw_front_size = hw_seg_size;
1135 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1136 nr_hw_segs++;
1137 }
1138
1139 nr_phys_segs++;
1140 bvprv = bv;
1141 seg_size = bv->bv_len;
1142 highprv = high;
1143 }
1144 if (hw_seg_size > bio->bi_hw_back_size)
1145 bio->bi_hw_back_size = hw_seg_size;
1146 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1147 bio->bi_hw_front_size = hw_seg_size;
1148 bio->bi_phys_segments = nr_phys_segs;
1149 bio->bi_hw_segments = nr_hw_segs;
1150 bio->bi_flags |= (1 << BIO_SEG_VALID);
1151}
1152
1153
Adrian Bunk93d17d32005-06-25 14:59:10 -07001154static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 struct bio *nxt)
1156{
1157 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1158 return 0;
1159
1160 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1161 return 0;
1162 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1163 return 0;
1164
1165 /*
1166 * bio and nxt are contigous in memory, check if the queue allows
1167 * these two to be merged into one
1168 */
1169 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1170 return 1;
1171
1172 return 0;
1173}
1174
Adrian Bunk93d17d32005-06-25 14:59:10 -07001175static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 struct bio *nxt)
1177{
1178 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1179 blk_recount_segments(q, bio);
1180 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1181 blk_recount_segments(q, nxt);
1182 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1183 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1184 return 0;
1185 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1186 return 0;
1187
1188 return 1;
1189}
1190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191/*
1192 * map a request to scatterlist, return number of sg entries setup. Caller
1193 * must make sure sg can hold rq->nr_phys_segments entries
1194 */
1195int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1196{
1197 struct bio_vec *bvec, *bvprv;
1198 struct bio *bio;
1199 int nsegs, i, cluster;
1200
1201 nsegs = 0;
1202 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1203
1204 /*
1205 * for each bio in rq
1206 */
1207 bvprv = NULL;
1208 rq_for_each_bio(bio, rq) {
1209 /*
1210 * for each segment in bio
1211 */
1212 bio_for_each_segment(bvec, bio, i) {
1213 int nbytes = bvec->bv_len;
1214
1215 if (bvprv && cluster) {
1216 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1217 goto new_segment;
1218
1219 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1220 goto new_segment;
1221 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1222 goto new_segment;
1223
1224 sg[nsegs - 1].length += nbytes;
1225 } else {
1226new_segment:
1227 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1228 sg[nsegs].page = bvec->bv_page;
1229 sg[nsegs].length = nbytes;
1230 sg[nsegs].offset = bvec->bv_offset;
1231
1232 nsegs++;
1233 }
1234 bvprv = bvec;
1235 } /* segments in bio */
1236 } /* bios in rq */
1237
1238 return nsegs;
1239}
1240
1241EXPORT_SYMBOL(blk_rq_map_sg);
1242
1243/*
1244 * the standard queue merge functions, can be overridden with device
1245 * specific ones if so desired
1246 */
1247
1248static inline int ll_new_mergeable(request_queue_t *q,
1249 struct request *req,
1250 struct bio *bio)
1251{
1252 int nr_phys_segs = bio_phys_segments(q, bio);
1253
1254 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1255 req->flags |= REQ_NOMERGE;
1256 if (req == q->last_merge)
1257 q->last_merge = NULL;
1258 return 0;
1259 }
1260
1261 /*
1262 * A hw segment is just getting larger, bump just the phys
1263 * counter.
1264 */
1265 req->nr_phys_segments += nr_phys_segs;
1266 return 1;
1267}
1268
1269static inline int ll_new_hw_segment(request_queue_t *q,
1270 struct request *req,
1271 struct bio *bio)
1272{
1273 int nr_hw_segs = bio_hw_segments(q, bio);
1274 int nr_phys_segs = bio_phys_segments(q, bio);
1275
1276 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1277 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1278 req->flags |= REQ_NOMERGE;
1279 if (req == q->last_merge)
1280 q->last_merge = NULL;
1281 return 0;
1282 }
1283
1284 /*
1285 * This will form the start of a new hw segment. Bump both
1286 * counters.
1287 */
1288 req->nr_hw_segments += nr_hw_segs;
1289 req->nr_phys_segments += nr_phys_segs;
1290 return 1;
1291}
1292
1293static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1294 struct bio *bio)
1295{
1296 int len;
1297
1298 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1299 req->flags |= REQ_NOMERGE;
1300 if (req == q->last_merge)
1301 q->last_merge = NULL;
1302 return 0;
1303 }
1304 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1305 blk_recount_segments(q, req->biotail);
1306 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1307 blk_recount_segments(q, bio);
1308 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1309 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1310 !BIOVEC_VIRT_OVERSIZE(len)) {
1311 int mergeable = ll_new_mergeable(q, req, bio);
1312
1313 if (mergeable) {
1314 if (req->nr_hw_segments == 1)
1315 req->bio->bi_hw_front_size = len;
1316 if (bio->bi_hw_segments == 1)
1317 bio->bi_hw_back_size = len;
1318 }
1319 return mergeable;
1320 }
1321
1322 return ll_new_hw_segment(q, req, bio);
1323}
1324
1325static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1326 struct bio *bio)
1327{
1328 int len;
1329
1330 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1331 req->flags |= REQ_NOMERGE;
1332 if (req == q->last_merge)
1333 q->last_merge = NULL;
1334 return 0;
1335 }
1336 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1337 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1338 blk_recount_segments(q, bio);
1339 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1340 blk_recount_segments(q, req->bio);
1341 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1342 !BIOVEC_VIRT_OVERSIZE(len)) {
1343 int mergeable = ll_new_mergeable(q, req, bio);
1344
1345 if (mergeable) {
1346 if (bio->bi_hw_segments == 1)
1347 bio->bi_hw_front_size = len;
1348 if (req->nr_hw_segments == 1)
1349 req->biotail->bi_hw_back_size = len;
1350 }
1351 return mergeable;
1352 }
1353
1354 return ll_new_hw_segment(q, req, bio);
1355}
1356
1357static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1358 struct request *next)
1359{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001360 int total_phys_segments;
1361 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363 /*
1364 * First check if the either of the requests are re-queued
1365 * requests. Can't merge them if they are.
1366 */
1367 if (req->special || next->special)
1368 return 0;
1369
1370 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001371 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 */
1373 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1374 return 0;
1375
1376 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1377 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1378 total_phys_segments--;
1379
1380 if (total_phys_segments > q->max_phys_segments)
1381 return 0;
1382
1383 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1384 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1385 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1386 /*
1387 * propagate the combined length to the end of the requests
1388 */
1389 if (req->nr_hw_segments == 1)
1390 req->bio->bi_hw_front_size = len;
1391 if (next->nr_hw_segments == 1)
1392 next->biotail->bi_hw_back_size = len;
1393 total_hw_segments--;
1394 }
1395
1396 if (total_hw_segments > q->max_hw_segments)
1397 return 0;
1398
1399 /* Merge is OK... */
1400 req->nr_phys_segments = total_phys_segments;
1401 req->nr_hw_segments = total_hw_segments;
1402 return 1;
1403}
1404
1405/*
1406 * "plug" the device if there are no outstanding requests: this will
1407 * force the transfer to start only after we have put all the requests
1408 * on the list.
1409 *
1410 * This is called with interrupts off and no requests on the queue and
1411 * with the queue lock held.
1412 */
1413void blk_plug_device(request_queue_t *q)
1414{
1415 WARN_ON(!irqs_disabled());
1416
1417 /*
1418 * don't plug a stopped queue, it must be paired with blk_start_queue()
1419 * which will restart the queueing
1420 */
1421 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1422 return;
1423
1424 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1425 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1426}
1427
1428EXPORT_SYMBOL(blk_plug_device);
1429
1430/*
1431 * remove the queue from the plugged list, if present. called with
1432 * queue lock held and interrupts disabled.
1433 */
1434int blk_remove_plug(request_queue_t *q)
1435{
1436 WARN_ON(!irqs_disabled());
1437
1438 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1439 return 0;
1440
1441 del_timer(&q->unplug_timer);
1442 return 1;
1443}
1444
1445EXPORT_SYMBOL(blk_remove_plug);
1446
1447/*
1448 * remove the plug and let it rip..
1449 */
1450void __generic_unplug_device(request_queue_t *q)
1451{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001452 if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 return;
1454
1455 if (!blk_remove_plug(q))
1456 return;
1457
Jens Axboe22e2c502005-06-27 10:55:12 +02001458 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459}
1460EXPORT_SYMBOL(__generic_unplug_device);
1461
1462/**
1463 * generic_unplug_device - fire a request queue
1464 * @q: The &request_queue_t in question
1465 *
1466 * Description:
1467 * Linux uses plugging to build bigger requests queues before letting
1468 * the device have at them. If a queue is plugged, the I/O scheduler
1469 * is still adding and merging requests on the queue. Once the queue
1470 * gets unplugged, the request_fn defined for the queue is invoked and
1471 * transfers started.
1472 **/
1473void generic_unplug_device(request_queue_t *q)
1474{
1475 spin_lock_irq(q->queue_lock);
1476 __generic_unplug_device(q);
1477 spin_unlock_irq(q->queue_lock);
1478}
1479EXPORT_SYMBOL(generic_unplug_device);
1480
1481static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1482 struct page *page)
1483{
1484 request_queue_t *q = bdi->unplug_io_data;
1485
1486 /*
1487 * devices don't necessarily have an ->unplug_fn defined
1488 */
1489 if (q->unplug_fn)
1490 q->unplug_fn(q);
1491}
1492
1493static void blk_unplug_work(void *data)
1494{
1495 request_queue_t *q = data;
1496
1497 q->unplug_fn(q);
1498}
1499
1500static void blk_unplug_timeout(unsigned long data)
1501{
1502 request_queue_t *q = (request_queue_t *)data;
1503
1504 kblockd_schedule_work(&q->unplug_work);
1505}
1506
1507/**
1508 * blk_start_queue - restart a previously stopped queue
1509 * @q: The &request_queue_t in question
1510 *
1511 * Description:
1512 * blk_start_queue() will clear the stop flag on the queue, and call
1513 * the request_fn for the queue if it was in a stopped state when
1514 * entered. Also see blk_stop_queue(). Queue lock must be held.
1515 **/
1516void blk_start_queue(request_queue_t *q)
1517{
1518 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1519
1520 /*
1521 * one level of recursion is ok and is much faster than kicking
1522 * the unplug handling
1523 */
1524 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1525 q->request_fn(q);
1526 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1527 } else {
1528 blk_plug_device(q);
1529 kblockd_schedule_work(&q->unplug_work);
1530 }
1531}
1532
1533EXPORT_SYMBOL(blk_start_queue);
1534
1535/**
1536 * blk_stop_queue - stop a queue
1537 * @q: The &request_queue_t in question
1538 *
1539 * Description:
1540 * The Linux block layer assumes that a block driver will consume all
1541 * entries on the request queue when the request_fn strategy is called.
1542 * Often this will not happen, because of hardware limitations (queue
1543 * depth settings). If a device driver gets a 'queue full' response,
1544 * or if it simply chooses not to queue more I/O at one point, it can
1545 * call this function to prevent the request_fn from being called until
1546 * the driver has signalled it's ready to go again. This happens by calling
1547 * blk_start_queue() to restart queue operations. Queue lock must be held.
1548 **/
1549void blk_stop_queue(request_queue_t *q)
1550{
1551 blk_remove_plug(q);
1552 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1553}
1554EXPORT_SYMBOL(blk_stop_queue);
1555
1556/**
1557 * blk_sync_queue - cancel any pending callbacks on a queue
1558 * @q: the queue
1559 *
1560 * Description:
1561 * The block layer may perform asynchronous callback activity
1562 * on a queue, such as calling the unplug function after a timeout.
1563 * A block device may call blk_sync_queue to ensure that any
1564 * such activity is cancelled, thus allowing it to release resources
1565 * the the callbacks might use. The caller must already have made sure
1566 * that its ->make_request_fn will not re-add plugging prior to calling
1567 * this function.
1568 *
1569 */
1570void blk_sync_queue(struct request_queue *q)
1571{
1572 del_timer_sync(&q->unplug_timer);
1573 kblockd_flush();
1574}
1575EXPORT_SYMBOL(blk_sync_queue);
1576
1577/**
1578 * blk_run_queue - run a single device queue
1579 * @q: The queue to run
1580 */
1581void blk_run_queue(struct request_queue *q)
1582{
1583 unsigned long flags;
1584
1585 spin_lock_irqsave(q->queue_lock, flags);
1586 blk_remove_plug(q);
Ken Chena2997382005-04-16 15:25:43 -07001587 if (!elv_queue_empty(q))
1588 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 spin_unlock_irqrestore(q->queue_lock, flags);
1590}
1591EXPORT_SYMBOL(blk_run_queue);
1592
1593/**
1594 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1595 * @q: the request queue to be released
1596 *
1597 * Description:
1598 * blk_cleanup_queue is the pair to blk_init_queue() or
1599 * blk_queue_make_request(). It should be called when a request queue is
1600 * being released; typically when a block device is being de-registered.
1601 * Currently, its primary task it to free all the &struct request
1602 * structures that were allocated to the queue and the queue itself.
1603 *
1604 * Caveat:
1605 * Hopefully the low level driver will have finished any
1606 * outstanding requests first...
1607 **/
1608void blk_cleanup_queue(request_queue_t * q)
1609{
1610 struct request_list *rl = &q->rq;
1611
1612 if (!atomic_dec_and_test(&q->refcnt))
1613 return;
1614
1615 if (q->elevator)
1616 elevator_exit(q->elevator);
1617
1618 blk_sync_queue(q);
1619
1620 if (rl->rq_pool)
1621 mempool_destroy(rl->rq_pool);
1622
1623 if (q->queue_tags)
1624 __blk_queue_free_tags(q);
1625
1626 blk_queue_ordered(q, QUEUE_ORDERED_NONE);
1627
1628 kmem_cache_free(requestq_cachep, q);
1629}
1630
1631EXPORT_SYMBOL(blk_cleanup_queue);
1632
1633static int blk_init_free_list(request_queue_t *q)
1634{
1635 struct request_list *rl = &q->rq;
1636
1637 rl->count[READ] = rl->count[WRITE] = 0;
1638 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001639 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 init_waitqueue_head(&rl->wait[READ]);
1641 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
Christoph Lameter19460892005-06-23 00:08:19 -07001643 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1644 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646 if (!rl->rq_pool)
1647 return -ENOMEM;
1648
1649 return 0;
1650}
1651
1652static int __make_request(request_queue_t *, struct bio *);
1653
Al Viro8267e262005-10-21 03:20:53 -04001654request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655{
Christoph Lameter19460892005-06-23 00:08:19 -07001656 return blk_alloc_queue_node(gfp_mask, -1);
1657}
1658EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Al Viro8267e262005-10-21 03:20:53 -04001660request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001661{
1662 request_queue_t *q;
1663
1664 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (!q)
1666 return NULL;
1667
1668 memset(q, 0, sizeof(*q));
1669 init_timer(&q->unplug_timer);
1670 atomic_set(&q->refcnt, 1);
1671
1672 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1673 q->backing_dev_info.unplug_io_data = q;
1674
1675 return q;
1676}
Christoph Lameter19460892005-06-23 00:08:19 -07001677EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
1679/**
1680 * blk_init_queue - prepare a request queue for use with a block device
1681 * @rfn: The function to be called to process requests that have been
1682 * placed on the queue.
1683 * @lock: Request queue spin lock
1684 *
1685 * Description:
1686 * If a block device wishes to use the standard request handling procedures,
1687 * which sorts requests and coalesces adjacent requests, then it must
1688 * call blk_init_queue(). The function @rfn will be called when there
1689 * are requests on the queue that need to be processed. If the device
1690 * supports plugging, then @rfn may not be called immediately when requests
1691 * are available on the queue, but may be called at some time later instead.
1692 * Plugged queues are generally unplugged when a buffer belonging to one
1693 * of the requests on the queue is needed, or due to memory pressure.
1694 *
1695 * @rfn is not required, or even expected, to remove all requests off the
1696 * queue, but only as many as it can handle at a time. If it does leave
1697 * requests on the queue, it is responsible for arranging that the requests
1698 * get dealt with eventually.
1699 *
1700 * The queue spin lock must be held while manipulating the requests on the
1701 * request queue.
1702 *
1703 * Function returns a pointer to the initialized request queue, or NULL if
1704 * it didn't succeed.
1705 *
1706 * Note:
1707 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1708 * when the block device is deactivated (such as at module unload).
1709 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1712{
Christoph Lameter19460892005-06-23 00:08:19 -07001713 return blk_init_queue_node(rfn, lock, -1);
1714}
1715EXPORT_SYMBOL(blk_init_queue);
1716
1717request_queue_t *
1718blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1719{
1720 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 if (!q)
1723 return NULL;
1724
Christoph Lameter19460892005-06-23 00:08:19 -07001725 q->node = node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (blk_init_free_list(q))
1727 goto out_init;
1728
152587d2005-04-12 16:22:06 -05001729 /*
1730 * if caller didn't supply a lock, they get per-queue locking with
1731 * our embedded lock
1732 */
1733 if (!lock) {
1734 spin_lock_init(&q->__queue_lock);
1735 lock = &q->__queue_lock;
1736 }
1737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 q->request_fn = rfn;
1739 q->back_merge_fn = ll_back_merge_fn;
1740 q->front_merge_fn = ll_front_merge_fn;
1741 q->merge_requests_fn = ll_merge_requests_fn;
1742 q->prep_rq_fn = NULL;
1743 q->unplug_fn = generic_unplug_device;
1744 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1745 q->queue_lock = lock;
1746
1747 blk_queue_segment_boundary(q, 0xffffffff);
1748
1749 blk_queue_make_request(q, __make_request);
1750 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1751
1752 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1753 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1754
1755 /*
1756 * all done
1757 */
1758 if (!elevator_init(q, NULL)) {
1759 blk_queue_congestion_threshold(q);
1760 return q;
1761 }
1762
1763 blk_cleanup_queue(q);
1764out_init:
1765 kmem_cache_free(requestq_cachep, q);
1766 return NULL;
1767}
Christoph Lameter19460892005-06-23 00:08:19 -07001768EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770int blk_get_queue(request_queue_t *q)
1771{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001772 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 atomic_inc(&q->refcnt);
1774 return 0;
1775 }
1776
1777 return 1;
1778}
1779
1780EXPORT_SYMBOL(blk_get_queue);
1781
1782static inline void blk_free_request(request_queue_t *q, struct request *rq)
1783{
Tejun Heocb98fc82005-10-28 08:29:39 +02001784 if (rq->flags & REQ_ELVPRIV)
1785 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 mempool_free(rq, q->rq.rq_pool);
1787}
1788
Jens Axboe22e2c502005-06-27 10:55:12 +02001789static inline struct request *
Tejun Heocb98fc82005-10-28 08:29:39 +02001790blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
Linus Torvalds5dd96242005-10-28 08:56:34 -07001791 int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
1793 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1794
1795 if (!rq)
1796 return NULL;
1797
1798 /*
1799 * first three bits are identical in rq->flags and bio->bi_rw,
1800 * see bio.h and blkdev.h
1801 */
1802 rq->flags = rw;
1803
Tejun Heocb98fc82005-10-28 08:29:39 +02001804 if (priv) {
1805 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
1806 mempool_free(rq, q->rq.rq_pool);
1807 return NULL;
1808 }
1809 rq->flags |= REQ_ELVPRIV;
1810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Tejun Heocb98fc82005-10-28 08:29:39 +02001812 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
1814
1815/*
1816 * ioc_batching returns true if the ioc is a valid batching request and
1817 * should be given priority access to a request.
1818 */
1819static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1820{
1821 if (!ioc)
1822 return 0;
1823
1824 /*
1825 * Make sure the process is able to allocate at least 1 request
1826 * even if the batch times out, otherwise we could theoretically
1827 * lose wakeups.
1828 */
1829 return ioc->nr_batch_requests == q->nr_batching ||
1830 (ioc->nr_batch_requests > 0
1831 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1832}
1833
1834/*
1835 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1836 * will cause the process to be a "batcher" on all queues in the system. This
1837 * is the behaviour we want though - once it gets a wakeup it should be given
1838 * a nice run.
1839 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07001840static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841{
1842 if (!ioc || ioc_batching(q, ioc))
1843 return;
1844
1845 ioc->nr_batch_requests = q->nr_batching;
1846 ioc->last_waited = jiffies;
1847}
1848
1849static void __freed_request(request_queue_t *q, int rw)
1850{
1851 struct request_list *rl = &q->rq;
1852
1853 if (rl->count[rw] < queue_congestion_off_threshold(q))
1854 clear_queue_congested(q, rw);
1855
1856 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 if (waitqueue_active(&rl->wait[rw]))
1858 wake_up(&rl->wait[rw]);
1859
1860 blk_clear_queue_full(q, rw);
1861 }
1862}
1863
1864/*
1865 * A request has just been released. Account for it, update the full and
1866 * congestion status, wake up any waiters. Called under q->queue_lock.
1867 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001868static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869{
1870 struct request_list *rl = &q->rq;
1871
1872 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02001873 if (priv)
1874 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876 __freed_request(q, rw);
1877
1878 if (unlikely(rl->starved[rw ^ 1]))
1879 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
1882#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1883/*
Nick Piggind6344532005-06-28 20:45:14 -07001884 * Get a free request, queue_lock must be held.
1885 * Returns NULL on failure, with queue_lock held.
1886 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Jens Axboe22e2c502005-06-27 10:55:12 +02001888static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04001889 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890{
1891 struct request *rq = NULL;
1892 struct request_list *rl = &q->rq;
Nick Pigginfb3cc432005-06-28 20:45:15 -07001893 struct io_context *ioc = current_io_context(GFP_ATOMIC);
Tejun Heocb98fc82005-10-28 08:29:39 +02001894 int priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (rl->count[rw]+1 >= q->nr_requests) {
1897 /*
1898 * The queue will fill after this allocation, so set it as
1899 * full, and mark this process as "batching". This process
1900 * will be allowed to complete a batch of requests, others
1901 * will be blocked.
1902 */
1903 if (!blk_queue_full(q, rw)) {
1904 ioc_set_batching(q, ioc);
1905 blk_set_queue_full(q, rw);
1906 }
1907 }
1908
Jens Axboe22e2c502005-06-27 10:55:12 +02001909 switch (elv_may_queue(q, rw, bio)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 case ELV_MQUEUE_NO:
1911 goto rq_starved;
1912 case ELV_MQUEUE_MAY:
1913 break;
1914 case ELV_MQUEUE_MUST:
1915 goto get_rq;
1916 }
1917
1918 if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) {
1919 /*
1920 * The queue is full and the allocating process is not a
1921 * "batcher", and not exempted by the IO scheduler
1922 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 goto out;
1924 }
1925
1926get_rq:
Jens Axboe082cf692005-06-28 16:35:11 +02001927 /*
1928 * Only allow batching queuers to allocate up to 50% over the defined
1929 * limit of requests, otherwise we could have thousands of requests
1930 * allocated with any setting of ->nr_requests
1931 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001932 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02001933 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 rl->count[rw]++;
1936 rl->starved[rw] = 0;
1937 if (rl->count[rw] >= queue_congestion_on_threshold(q))
1938 set_queue_congested(q, rw);
Tejun Heocb98fc82005-10-28 08:29:39 +02001939
Jens Axboe64521d12005-10-28 08:30:39 +02001940 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02001941 if (priv)
1942 rl->elvpriv++;
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 spin_unlock_irq(q->queue_lock);
1945
Tejun Heocb98fc82005-10-28 08:29:39 +02001946 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 if (!rq) {
1948 /*
1949 * Allocation failed presumably due to memory. Undo anything
1950 * we might have messed up.
1951 *
1952 * Allocating task should really be put onto the front of the
1953 * wait queue, but this is pretty rare.
1954 */
1955 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02001956 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958 /*
1959 * in the very unlikely event that allocation failed and no
1960 * requests for this direction was pending, mark us starved
1961 * so that freeing of a request in the other direction will
1962 * notice us. another possible fix would be to split the
1963 * rq mempool into READ and WRITE
1964 */
1965rq_starved:
1966 if (unlikely(rl->count[rw] == 0))
1967 rl->starved[rw] = 1;
1968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 goto out;
1970 }
1971
1972 if (ioc_batching(q, ioc))
1973 ioc->nr_batch_requests--;
1974
1975 rq_init(q, rq);
1976 rq->rl = rl;
1977out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 return rq;
1979}
1980
1981/*
1982 * No available requests for this queue, unplug the device and wait for some
1983 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07001984 *
1985 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 */
Jens Axboe22e2c502005-06-27 10:55:12 +02001987static struct request *get_request_wait(request_queue_t *q, int rw,
1988 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 struct request *rq;
1991
Nick Piggin450991b2005-06-28 20:45:13 -07001992 rq = get_request(q, rw, bio, GFP_NOIO);
1993 while (!rq) {
1994 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 struct request_list *rl = &q->rq;
1996
1997 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1998 TASK_UNINTERRUPTIBLE);
1999
Jens Axboe22e2c502005-06-27 10:55:12 +02002000 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 if (!rq) {
2003 struct io_context *ioc;
2004
Nick Piggind6344532005-06-28 20:45:14 -07002005 __generic_unplug_device(q);
2006 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 io_schedule();
2008
2009 /*
2010 * After sleeping, we become a "batching" process and
2011 * will be able to allocate at least one request, and
2012 * up to a big batch of them for a small period time.
2013 * See ioc_batching, ioc_set_batching
2014 */
Nick Pigginfb3cc432005-06-28 20:45:15 -07002015 ioc = current_io_context(GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002017
2018 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
2020 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002021 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
2023 return rq;
2024}
2025
Al Viro8267e262005-10-21 03:20:53 -04002026struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
2028 struct request *rq;
2029
2030 BUG_ON(rw != READ && rw != WRITE);
2031
Nick Piggind6344532005-06-28 20:45:14 -07002032 spin_lock_irq(q->queue_lock);
2033 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002034 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002035 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002036 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002037 if (!rq)
2038 spin_unlock_irq(q->queue_lock);
2039 }
2040 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 return rq;
2043}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044EXPORT_SYMBOL(blk_get_request);
2045
2046/**
2047 * blk_requeue_request - put a request back on queue
2048 * @q: request queue where request should be inserted
2049 * @rq: request to be inserted
2050 *
2051 * Description:
2052 * Drivers often keep queueing requests until the hardware cannot accept
2053 * more, when that condition happens we need to put the request back
2054 * on the queue. Must be called with queue lock held.
2055 */
2056void blk_requeue_request(request_queue_t *q, struct request *rq)
2057{
2058 if (blk_rq_tagged(rq))
2059 blk_queue_end_tag(q, rq);
2060
2061 elv_requeue_request(q, rq);
2062}
2063
2064EXPORT_SYMBOL(blk_requeue_request);
2065
2066/**
2067 * blk_insert_request - insert a special request in to a request queue
2068 * @q: request queue where request should be inserted
2069 * @rq: request to be inserted
2070 * @at_head: insert request at head or tail of queue
2071 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 *
2073 * Description:
2074 * Many block devices need to execute commands asynchronously, so they don't
2075 * block the whole kernel from preemption during request execution. This is
2076 * accomplished normally by inserting aritficial requests tagged as
2077 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2078 * scheduled for actual execution by the request queue.
2079 *
2080 * We have the option of inserting the head or the tail of the queue.
2081 * Typically we use the tail for new ioctls and so forth. We use the head
2082 * of the queue for things like a QUEUE_FULL message from a device, or a
2083 * host that is unable to accept a particular command.
2084 */
2085void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002086 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087{
Tejun Heo 867d1192005-04-24 02:06:05 -05002088 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 unsigned long flags;
2090
2091 /*
2092 * tell I/O scheduler that this isn't a regular read/write (ie it
2093 * must not attempt merges on this) and that it acts as a soft
2094 * barrier
2095 */
2096 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2097
2098 rq->special = data;
2099
2100 spin_lock_irqsave(q->queue_lock, flags);
2101
2102 /*
2103 * If command is tagged, release the tag
2104 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002105 if (blk_rq_tagged(rq))
2106 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Tejun Heo 867d1192005-04-24 02:06:05 -05002108 drive_stat_acct(rq, rq->nr_sectors, 1);
2109 __elv_add_request(q, rq, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 if (blk_queue_plugged(q))
2112 __generic_unplug_device(q);
2113 else
2114 q->request_fn(q);
2115 spin_unlock_irqrestore(q->queue_lock, flags);
2116}
2117
2118EXPORT_SYMBOL(blk_insert_request);
2119
2120/**
2121 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2122 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002123 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 * @ubuf: the user buffer
2125 * @len: length of user data
2126 *
2127 * Description:
2128 * Data will be mapped directly for zero copy io, if possible. Otherwise
2129 * a kernel bounce buffer is used.
2130 *
2131 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2132 * still in process context.
2133 *
2134 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2135 * before being submitted to the device, as pages mapped may be out of
2136 * reach. It's the callers responsibility to make sure this happens. The
2137 * original bio must be passed back in to blk_rq_unmap_user() for proper
2138 * unmapping.
2139 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002140int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2141 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
2143 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002145 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
2147 if (len > (q->max_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002148 return -EINVAL;
2149 if (!len || !ubuf)
2150 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
Jens Axboedd1cab92005-06-20 14:06:01 +02002152 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 /*
2155 * if alignment requirement is satisfied, map in user pages for
2156 * direct dma. else, set up kernel bounce buffers
2157 */
2158 uaddr = (unsigned long) ubuf;
2159 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002160 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002162 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163
2164 if (!IS_ERR(bio)) {
2165 rq->bio = rq->biotail = bio;
2166 blk_rq_bio_prep(q, rq, bio);
2167
2168 rq->buffer = rq->data = NULL;
2169 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002170 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 }
2172
2173 /*
2174 * bio is the err-ptr
2175 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002176 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177}
2178
2179EXPORT_SYMBOL(blk_rq_map_user);
2180
2181/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002182 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2183 * @q: request queue where request should be inserted
2184 * @rq: request to map data to
2185 * @iov: pointer to the iovec
2186 * @iov_count: number of elements in the iovec
2187 *
2188 * Description:
2189 * Data will be mapped directly for zero copy io, if possible. Otherwise
2190 * a kernel bounce buffer is used.
2191 *
2192 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2193 * still in process context.
2194 *
2195 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2196 * before being submitted to the device, as pages mapped may be out of
2197 * reach. It's the callers responsibility to make sure this happens. The
2198 * original bio must be passed back in to blk_rq_unmap_user() for proper
2199 * unmapping.
2200 */
2201int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2202 struct sg_iovec *iov, int iov_count)
2203{
2204 struct bio *bio;
2205
2206 if (!iov || iov_count <= 0)
2207 return -EINVAL;
2208
2209 /* we don't allow misaligned data like bio_map_user() does. If the
2210 * user is using sg, they're expected to know the alignment constraints
2211 * and respect them accordingly */
2212 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2213 if (IS_ERR(bio))
2214 return PTR_ERR(bio);
2215
2216 rq->bio = rq->biotail = bio;
2217 blk_rq_bio_prep(q, rq, bio);
2218 rq->buffer = rq->data = NULL;
2219 rq->data_len = bio->bi_size;
2220 return 0;
2221}
2222
2223EXPORT_SYMBOL(blk_rq_map_user_iov);
2224
2225/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002227 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 * @ulen: length of user buffer
2229 *
2230 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002231 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002233int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
2235 int ret = 0;
2236
2237 if (bio) {
2238 if (bio_flagged(bio, BIO_USER_MAPPED))
2239 bio_unmap_user(bio);
2240 else
2241 ret = bio_uncopy_user(bio);
2242 }
2243
Jens Axboedd1cab92005-06-20 14:06:01 +02002244 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245}
2246
2247EXPORT_SYMBOL(blk_rq_unmap_user);
2248
2249/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002250 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2251 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002252 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002253 * @kbuf: the kernel buffer
2254 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002255 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002256 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002257int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002258 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002259{
Mike Christie df46b9a2005-06-20 14:04:44 +02002260 struct bio *bio;
2261
2262 if (len > (q->max_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002263 return -EINVAL;
2264 if (!len || !kbuf)
2265 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002266
2267 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002268 if (IS_ERR(bio))
2269 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002270
Jens Axboedd1cab92005-06-20 14:06:01 +02002271 if (rq_data_dir(rq) == WRITE)
2272 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002273
Jens Axboedd1cab92005-06-20 14:06:01 +02002274 rq->bio = rq->biotail = bio;
2275 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002276
Jens Axboedd1cab92005-06-20 14:06:01 +02002277 rq->buffer = rq->data = NULL;
2278 rq->data_len = len;
2279 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002280}
2281
2282EXPORT_SYMBOL(blk_rq_map_kern);
2283
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002284/**
2285 * blk_execute_rq_nowait - insert a request into queue for execution
2286 * @q: queue to insert the request in
2287 * @bd_disk: matching gendisk
2288 * @rq: request to insert
2289 * @at_head: insert request at head or tail of queue
2290 * @done: I/O completion handler
2291 *
2292 * Description:
2293 * Insert a fully prepared request at the back of the io scheduler queue
2294 * for execution. Don't wait for completion.
2295 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002296void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2297 struct request *rq, int at_head,
2298 void (*done)(struct request *))
2299{
2300 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2301
2302 rq->rq_disk = bd_disk;
2303 rq->flags |= REQ_NOMERGE;
2304 rq->end_io = done;
2305 elv_add_request(q, rq, where, 1);
2306 generic_unplug_device(q);
2307}
2308
Mike Christie6e39b692005-11-11 05:30:24 -06002309EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311/**
2312 * blk_execute_rq - insert a request into queue for execution
2313 * @q: queue to insert the request in
2314 * @bd_disk: matching gendisk
2315 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002316 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 *
2318 * Description:
2319 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002320 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 */
2322int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002323 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324{
2325 DECLARE_COMPLETION(wait);
2326 char sense[SCSI_SENSE_BUFFERSIZE];
2327 int err = 0;
2328
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 /*
2330 * we need an extra reference to the request, so we can look at
2331 * it after io completion
2332 */
2333 rq->ref_count++;
2334
2335 if (!rq->sense) {
2336 memset(sense, 0, sizeof(sense));
2337 rq->sense = sense;
2338 rq->sense_len = 0;
2339 }
2340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 rq->waiting = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002342 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 wait_for_completion(&wait);
2344 rq->waiting = NULL;
2345
2346 if (rq->errors)
2347 err = -EIO;
2348
2349 return err;
2350}
2351
2352EXPORT_SYMBOL(blk_execute_rq);
2353
2354/**
2355 * blkdev_issue_flush - queue a flush
2356 * @bdev: blockdev to issue flush for
2357 * @error_sector: error sector
2358 *
2359 * Description:
2360 * Issue a flush for the block device in question. Caller can supply
2361 * room for storing the error offset in case of a flush error, if they
2362 * wish to. Caller must run wait_for_completion() on its own.
2363 */
2364int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2365{
2366 request_queue_t *q;
2367
2368 if (bdev->bd_disk == NULL)
2369 return -ENXIO;
2370
2371 q = bdev_get_queue(bdev);
2372 if (!q)
2373 return -ENXIO;
2374 if (!q->issue_flush_fn)
2375 return -EOPNOTSUPP;
2376
2377 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2378}
2379
2380EXPORT_SYMBOL(blkdev_issue_flush);
2381
Adrian Bunk93d17d32005-06-25 14:59:10 -07002382static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383{
2384 int rw = rq_data_dir(rq);
2385
2386 if (!blk_fs_request(rq) || !rq->rq_disk)
2387 return;
2388
Jens Axboed72d9042005-11-01 08:35:42 +01002389 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002390 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002391 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 disk_round_stats(rq->rq_disk);
2393 rq->rq_disk->in_flight++;
2394 }
2395}
2396
2397/*
2398 * add-request adds a request to the linked list.
2399 * queue lock is held and interrupts disabled, as we muck with the
2400 * request queue list.
2401 */
2402static inline void add_request(request_queue_t * q, struct request * req)
2403{
2404 drive_stat_acct(req, req->nr_sectors, 1);
2405
2406 if (q->activity_fn)
2407 q->activity_fn(q->activity_data, rq_data_dir(req));
2408
2409 /*
2410 * elevator indicated where it wants this request to be
2411 * inserted at elevator_merge time
2412 */
2413 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2414}
2415
2416/*
2417 * disk_round_stats() - Round off the performance stats on a struct
2418 * disk_stats.
2419 *
2420 * The average IO queue length and utilisation statistics are maintained
2421 * by observing the current state of the queue length and the amount of
2422 * time it has been in this state for.
2423 *
2424 * Normally, that accounting is done on IO completion, but that can result
2425 * in more than a second's worth of IO being accounted for within any one
2426 * second, leading to >100% utilisation. To deal with that, we call this
2427 * function to do a round-off before returning the results when reading
2428 * /proc/diskstats. This accounts immediately for all queue usage up to
2429 * the current jiffies and restarts the counters again.
2430 */
2431void disk_round_stats(struct gendisk *disk)
2432{
2433 unsigned long now = jiffies;
2434
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002435 if (now == disk->stamp)
2436 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002438 if (disk->in_flight) {
2439 __disk_stat_add(disk, time_in_queue,
2440 disk->in_flight * (now - disk->stamp));
2441 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444}
2445
2446/*
2447 * queue lock must be held
2448 */
Mike Christie6e39b692005-11-11 05:30:24 -06002449void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450{
2451 struct request_list *rl = req->rl;
2452
2453 if (unlikely(!q))
2454 return;
2455 if (unlikely(--req->ref_count))
2456 return;
2457
Tejun Heo8922e162005-10-20 16:23:44 +02002458 elv_completed_request(q, req);
2459
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 req->rq_status = RQ_INACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 req->rl = NULL;
2462
2463 /*
2464 * Request may not have originated from ll_rw_blk. if not,
2465 * it didn't come out of our reserved rq pools
2466 */
2467 if (rl) {
2468 int rw = rq_data_dir(req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002469 int priv = req->flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 BUG_ON(!list_empty(&req->queuelist));
2472
2473 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002474 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 }
2476}
2477
Mike Christie6e39b692005-11-11 05:30:24 -06002478EXPORT_SYMBOL_GPL(__blk_put_request);
2479
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480void blk_put_request(struct request *req)
2481{
Tejun Heo8922e162005-10-20 16:23:44 +02002482 unsigned long flags;
2483 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Tejun Heo8922e162005-10-20 16:23:44 +02002485 /*
2486 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2487 * following if (q) test.
2488 */
2489 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 spin_lock_irqsave(q->queue_lock, flags);
2491 __blk_put_request(q, req);
2492 spin_unlock_irqrestore(q->queue_lock, flags);
2493 }
2494}
2495
2496EXPORT_SYMBOL(blk_put_request);
2497
2498/**
2499 * blk_end_sync_rq - executes a completion event on a request
2500 * @rq: request to complete
2501 */
2502void blk_end_sync_rq(struct request *rq)
2503{
2504 struct completion *waiting = rq->waiting;
2505
2506 rq->waiting = NULL;
2507 __blk_put_request(rq->q, rq);
2508
2509 /*
2510 * complete last, if this is a stack request the process (and thus
2511 * the rq pointer) could be invalid right after this complete()
2512 */
2513 complete(waiting);
2514}
2515EXPORT_SYMBOL(blk_end_sync_rq);
2516
2517/**
2518 * blk_congestion_wait - wait for a queue to become uncongested
2519 * @rw: READ or WRITE
2520 * @timeout: timeout in jiffies
2521 *
2522 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2523 * If no queues are congested then just wait for the next request to be
2524 * returned.
2525 */
2526long blk_congestion_wait(int rw, long timeout)
2527{
2528 long ret;
2529 DEFINE_WAIT(wait);
2530 wait_queue_head_t *wqh = &congestion_wqh[rw];
2531
2532 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2533 ret = io_schedule_timeout(timeout);
2534 finish_wait(wqh, &wait);
2535 return ret;
2536}
2537
2538EXPORT_SYMBOL(blk_congestion_wait);
2539
2540/*
2541 * Has to be called with the request spinlock acquired
2542 */
2543static int attempt_merge(request_queue_t *q, struct request *req,
2544 struct request *next)
2545{
2546 if (!rq_mergeable(req) || !rq_mergeable(next))
2547 return 0;
2548
2549 /*
2550 * not contigious
2551 */
2552 if (req->sector + req->nr_sectors != next->sector)
2553 return 0;
2554
2555 if (rq_data_dir(req) != rq_data_dir(next)
2556 || req->rq_disk != next->rq_disk
2557 || next->waiting || next->special)
2558 return 0;
2559
2560 /*
2561 * If we are allowed to merge, then append bio list
2562 * from next to rq and release next. merge_requests_fn
2563 * will have updated segment counts, update sector
2564 * counts here.
2565 */
2566 if (!q->merge_requests_fn(q, req, next))
2567 return 0;
2568
2569 /*
2570 * At this point we have either done a back merge
2571 * or front merge. We need the smaller start_time of
2572 * the merged requests to be the current request
2573 * for accounting purposes.
2574 */
2575 if (time_after(req->start_time, next->start_time))
2576 req->start_time = next->start_time;
2577
2578 req->biotail->bi_next = next->bio;
2579 req->biotail = next->biotail;
2580
2581 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2582
2583 elv_merge_requests(q, req, next);
2584
2585 if (req->rq_disk) {
2586 disk_round_stats(req->rq_disk);
2587 req->rq_disk->in_flight--;
2588 }
2589
Jens Axboe22e2c502005-06-27 10:55:12 +02002590 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2591
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 __blk_put_request(q, next);
2593 return 1;
2594}
2595
2596static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2597{
2598 struct request *next = elv_latter_request(q, rq);
2599
2600 if (next)
2601 return attempt_merge(q, rq, next);
2602
2603 return 0;
2604}
2605
2606static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2607{
2608 struct request *prev = elv_former_request(q, rq);
2609
2610 if (prev)
2611 return attempt_merge(q, prev, rq);
2612
2613 return 0;
2614}
2615
2616/**
2617 * blk_attempt_remerge - attempt to remerge active head with next request
2618 * @q: The &request_queue_t belonging to the device
2619 * @rq: The head request (usually)
2620 *
2621 * Description:
2622 * For head-active devices, the queue can easily be unplugged so quickly
2623 * that proper merging is not done on the front request. This may hurt
2624 * performance greatly for some devices. The block layer cannot safely
2625 * do merging on that first request for these queues, but the driver can
2626 * call this function and make it happen any way. Only the driver knows
2627 * when it is safe to do so.
2628 **/
2629void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2630{
2631 unsigned long flags;
2632
2633 spin_lock_irqsave(q->queue_lock, flags);
2634 attempt_back_merge(q, rq);
2635 spin_unlock_irqrestore(q->queue_lock, flags);
2636}
2637
2638EXPORT_SYMBOL(blk_attempt_remerge);
2639
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640static int __make_request(request_queue_t *q, struct bio *bio)
2641{
Nick Piggin450991b2005-06-28 20:45:13 -07002642 struct request *req;
Jens Axboe4a534f92005-04-16 15:25:40 -07002643 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
Jens Axboe22e2c502005-06-27 10:55:12 +02002644 unsigned short prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 sector_t sector;
2646
2647 sector = bio->bi_sector;
2648 nr_sectors = bio_sectors(bio);
2649 cur_nr_sectors = bio_cur_sectors(bio);
Jens Axboe22e2c502005-06-27 10:55:12 +02002650 prio = bio_prio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
2652 rw = bio_data_dir(bio);
Jens Axboe4a534f92005-04-16 15:25:40 -07002653 sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
2655 /*
2656 * low level driver can indicate that it wants pages above a
2657 * certain limit bounced to low memory (ie for highmem, or even
2658 * ISA dma in theory)
2659 */
2660 blk_queue_bounce(q, &bio);
2661
2662 spin_lock_prefetch(q->queue_lock);
2663
2664 barrier = bio_barrier(bio);
Nick Pigginfde6ad22005-06-23 00:08:53 -07002665 if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 err = -EOPNOTSUPP;
2667 goto end_io;
2668 }
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 spin_lock_irq(q->queue_lock);
2671
Nick Piggin450991b2005-06-28 20:45:13 -07002672 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 goto get_rq;
2674
2675 el_ret = elv_merge(q, &req, bio);
2676 switch (el_ret) {
2677 case ELEVATOR_BACK_MERGE:
2678 BUG_ON(!rq_mergeable(req));
2679
2680 if (!q->back_merge_fn(q, req, bio))
2681 break;
2682
2683 req->biotail->bi_next = bio;
2684 req->biotail = bio;
2685 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002686 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 drive_stat_acct(req, nr_sectors, 0);
2688 if (!attempt_back_merge(q, req))
2689 elv_merged_request(q, req);
2690 goto out;
2691
2692 case ELEVATOR_FRONT_MERGE:
2693 BUG_ON(!rq_mergeable(req));
2694
2695 if (!q->front_merge_fn(q, req, bio))
2696 break;
2697
2698 bio->bi_next = req->bio;
2699 req->bio = bio;
2700
2701 /*
2702 * may not be valid. if the low level driver said
2703 * it didn't need a bounce buffer then it better
2704 * not touch req->buffer either...
2705 */
2706 req->buffer = bio_data(bio);
2707 req->current_nr_sectors = cur_nr_sectors;
2708 req->hard_cur_sectors = cur_nr_sectors;
2709 req->sector = req->hard_sector = sector;
2710 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002711 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 drive_stat_acct(req, nr_sectors, 0);
2713 if (!attempt_front_merge(q, req))
2714 elv_merged_request(q, req);
2715 goto out;
2716
Nick Piggin450991b2005-06-28 20:45:13 -07002717 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002719 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
2721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002723 /*
2724 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002725 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002726 */
Nick Piggin450991b2005-06-28 20:45:13 -07002727 req = get_request_wait(q, rw, bio);
Nick Piggind6344532005-06-28 20:45:14 -07002728
Nick Piggin450991b2005-06-28 20:45:13 -07002729 /*
2730 * After dropping the lock and possibly sleeping here, our request
2731 * may now be mergeable after it had proven unmergeable (above).
2732 * We don't worry about that case for efficiency. It won't happen
2733 * often, and the elevators are able to handle it.
2734 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
2736 req->flags |= REQ_CMD;
2737
2738 /*
2739 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2740 */
2741 if (bio_rw_ahead(bio) || bio_failfast(bio))
2742 req->flags |= REQ_FAILFAST;
2743
2744 /*
2745 * REQ_BARRIER implies no merging, but lets make it explicit
2746 */
Nick Pigginfde6ad22005-06-23 00:08:53 -07002747 if (unlikely(barrier))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2749
2750 req->errors = 0;
2751 req->hard_sector = req->sector = sector;
2752 req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2753 req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2754 req->nr_phys_segments = bio_phys_segments(q, bio);
2755 req->nr_hw_segments = bio_hw_segments(q, bio);
2756 req->buffer = bio_data(bio); /* see ->buffer comment above */
2757 req->waiting = NULL;
2758 req->bio = req->biotail = bio;
Jens Axboe22e2c502005-06-27 10:55:12 +02002759 req->ioprio = prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 req->rq_disk = bio->bi_bdev->bd_disk;
2761 req->start_time = jiffies;
2762
Nick Piggin450991b2005-06-28 20:45:13 -07002763 spin_lock_irq(q->queue_lock);
2764 if (elv_queue_empty(q))
2765 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 add_request(q, req);
2767out:
Jens Axboe4a534f92005-04-16 15:25:40 -07002768 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769 __generic_unplug_device(q);
2770
2771 spin_unlock_irq(q->queue_lock);
2772 return 0;
2773
2774end_io:
2775 bio_endio(bio, nr_sectors << 9, err);
2776 return 0;
2777}
2778
2779/*
2780 * If bio->bi_dev is a partition, remap the location
2781 */
2782static inline void blk_partition_remap(struct bio *bio)
2783{
2784 struct block_device *bdev = bio->bi_bdev;
2785
2786 if (bdev != bdev->bd_contains) {
2787 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01002788 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789
Jens Axboea3623572005-11-01 09:26:16 +01002790 p->sectors[rw] += bio_sectors(bio);
2791 p->ios[rw]++;
2792
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 bio->bi_sector += p->start_sect;
2794 bio->bi_bdev = bdev->bd_contains;
2795 }
2796}
2797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798static void handle_bad_sector(struct bio *bio)
2799{
2800 char b[BDEVNAME_SIZE];
2801
2802 printk(KERN_INFO "attempt to access beyond end of device\n");
2803 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2804 bdevname(bio->bi_bdev, b),
2805 bio->bi_rw,
2806 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2807 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2808
2809 set_bit(BIO_EOF, &bio->bi_flags);
2810}
2811
2812/**
2813 * generic_make_request: hand a buffer to its device driver for I/O
2814 * @bio: The bio describing the location in memory and on the device.
2815 *
2816 * generic_make_request() is used to make I/O requests of block
2817 * devices. It is passed a &struct bio, which describes the I/O that needs
2818 * to be done.
2819 *
2820 * generic_make_request() does not return any status. The
2821 * success/failure status of the request, along with notification of
2822 * completion, is delivered asynchronously through the bio->bi_end_io
2823 * function described (one day) else where.
2824 *
2825 * The caller of generic_make_request must make sure that bi_io_vec
2826 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2827 * set to describe the device address, and the
2828 * bi_end_io and optionally bi_private are set to describe how
2829 * completion notification should be signaled.
2830 *
2831 * generic_make_request and the drivers it calls may use bi_next if this
2832 * bio happens to be merged with someone else, and may change bi_dev and
2833 * bi_sector for remaps as it sees fit. So the values of these fields
2834 * should NOT be depended on after the call to generic_make_request.
2835 */
2836void generic_make_request(struct bio *bio)
2837{
2838 request_queue_t *q;
2839 sector_t maxsector;
2840 int ret, nr_sectors = bio_sectors(bio);
2841
2842 might_sleep();
2843 /* Test device or partition size, when known. */
2844 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2845 if (maxsector) {
2846 sector_t sector = bio->bi_sector;
2847
2848 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2849 /*
2850 * This may well happen - the kernel calls bread()
2851 * without checking the size of the device, e.g., when
2852 * mounting a device.
2853 */
2854 handle_bad_sector(bio);
2855 goto end_io;
2856 }
2857 }
2858
2859 /*
2860 * Resolve the mapping until finished. (drivers are
2861 * still free to implement/resolve their own stacking
2862 * by explicitly returning 0)
2863 *
2864 * NOTE: we don't repeat the blk_size check for each new device.
2865 * Stacking drivers are expected to know what they are doing.
2866 */
2867 do {
2868 char b[BDEVNAME_SIZE];
2869
2870 q = bdev_get_queue(bio->bi_bdev);
2871 if (!q) {
2872 printk(KERN_ERR
2873 "generic_make_request: Trying to access "
2874 "nonexistent block-device %s (%Lu)\n",
2875 bdevname(bio->bi_bdev, b),
2876 (long long) bio->bi_sector);
2877end_io:
2878 bio_endio(bio, bio->bi_size, -EIO);
2879 break;
2880 }
2881
2882 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
2883 printk("bio too big device %s (%u > %u)\n",
2884 bdevname(bio->bi_bdev, b),
2885 bio_sectors(bio),
2886 q->max_hw_sectors);
2887 goto end_io;
2888 }
2889
Nick Pigginfde6ad22005-06-23 00:08:53 -07002890 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 goto end_io;
2892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 /*
2894 * If this device has partitions, remap block n
2895 * of partition p to block n+start(p) of the disk.
2896 */
2897 blk_partition_remap(bio);
2898
2899 ret = q->make_request_fn(q, bio);
2900 } while (ret);
2901}
2902
2903EXPORT_SYMBOL(generic_make_request);
2904
2905/**
2906 * submit_bio: submit a bio to the block device layer for I/O
2907 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2908 * @bio: The &struct bio which describes the I/O
2909 *
2910 * submit_bio() is very similar in purpose to generic_make_request(), and
2911 * uses that function to do most of the work. Both are fairly rough
2912 * interfaces, @bio must be presetup and ready for I/O.
2913 *
2914 */
2915void submit_bio(int rw, struct bio *bio)
2916{
2917 int count = bio_sectors(bio);
2918
2919 BIO_BUG_ON(!bio->bi_size);
2920 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02002921 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 if (rw & WRITE)
2923 mod_page_state(pgpgout, count);
2924 else
2925 mod_page_state(pgpgin, count);
2926
2927 if (unlikely(block_dump)) {
2928 char b[BDEVNAME_SIZE];
2929 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
2930 current->comm, current->pid,
2931 (rw & WRITE) ? "WRITE" : "READ",
2932 (unsigned long long)bio->bi_sector,
2933 bdevname(bio->bi_bdev,b));
2934 }
2935
2936 generic_make_request(bio);
2937}
2938
2939EXPORT_SYMBOL(submit_bio);
2940
Adrian Bunk93d17d32005-06-25 14:59:10 -07002941static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942{
2943 struct bio *bio, *prevbio = NULL;
2944 int nr_phys_segs, nr_hw_segs;
2945 unsigned int phys_size, hw_size;
2946 request_queue_t *q = rq->q;
2947
2948 if (!rq->bio)
2949 return;
2950
2951 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
2952 rq_for_each_bio(bio, rq) {
2953 /* Force bio hw/phys segs to be recalculated. */
2954 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
2955
2956 nr_phys_segs += bio_phys_segments(q, bio);
2957 nr_hw_segs += bio_hw_segments(q, bio);
2958 if (prevbio) {
2959 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
2960 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
2961
2962 if (blk_phys_contig_segment(q, prevbio, bio) &&
2963 pseg <= q->max_segment_size) {
2964 nr_phys_segs--;
2965 phys_size += prevbio->bi_size + bio->bi_size;
2966 } else
2967 phys_size = 0;
2968
2969 if (blk_hw_contig_segment(q, prevbio, bio) &&
2970 hseg <= q->max_segment_size) {
2971 nr_hw_segs--;
2972 hw_size += prevbio->bi_size + bio->bi_size;
2973 } else
2974 hw_size = 0;
2975 }
2976 prevbio = bio;
2977 }
2978
2979 rq->nr_phys_segments = nr_phys_segs;
2980 rq->nr_hw_segments = nr_hw_segs;
2981}
2982
Adrian Bunk93d17d32005-06-25 14:59:10 -07002983static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984{
2985 if (blk_fs_request(rq)) {
2986 rq->hard_sector += nsect;
2987 rq->hard_nr_sectors -= nsect;
2988
2989 /*
2990 * Move the I/O submission pointers ahead if required.
2991 */
2992 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
2993 (rq->sector <= rq->hard_sector)) {
2994 rq->sector = rq->hard_sector;
2995 rq->nr_sectors = rq->hard_nr_sectors;
2996 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
2997 rq->current_nr_sectors = rq->hard_cur_sectors;
2998 rq->buffer = bio_data(rq->bio);
2999 }
3000
3001 /*
3002 * if total number of sectors is less than the first segment
3003 * size, something has gone terribly wrong
3004 */
3005 if (rq->nr_sectors < rq->current_nr_sectors) {
3006 printk("blk: request botched\n");
3007 rq->nr_sectors = rq->current_nr_sectors;
3008 }
3009 }
3010}
3011
3012static int __end_that_request_first(struct request *req, int uptodate,
3013 int nr_bytes)
3014{
3015 int total_bytes, bio_nbytes, error, next_idx = 0;
3016 struct bio *bio;
3017
3018 /*
3019 * extend uptodate bool to allow < 0 value to be direct io error
3020 */
3021 error = 0;
3022 if (end_io_error(uptodate))
3023 error = !uptodate ? -EIO : uptodate;
3024
3025 /*
3026 * for a REQ_BLOCK_PC request, we want to carry any eventual
3027 * sense key with us all the way through
3028 */
3029 if (!blk_pc_request(req))
3030 req->errors = 0;
3031
3032 if (!uptodate) {
3033 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3034 printk("end_request: I/O error, dev %s, sector %llu\n",
3035 req->rq_disk ? req->rq_disk->disk_name : "?",
3036 (unsigned long long)req->sector);
3037 }
3038
Jens Axboed72d9042005-11-01 08:35:42 +01003039 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003040 const int rw = rq_data_dir(req);
3041
3042 __disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003043 }
3044
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 total_bytes = bio_nbytes = 0;
3046 while ((bio = req->bio) != NULL) {
3047 int nbytes;
3048
3049 if (nr_bytes >= bio->bi_size) {
3050 req->bio = bio->bi_next;
3051 nbytes = bio->bi_size;
3052 bio_endio(bio, nbytes, error);
3053 next_idx = 0;
3054 bio_nbytes = 0;
3055 } else {
3056 int idx = bio->bi_idx + next_idx;
3057
3058 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3059 blk_dump_rq_flags(req, "__end_that");
3060 printk("%s: bio idx %d >= vcnt %d\n",
3061 __FUNCTION__,
3062 bio->bi_idx, bio->bi_vcnt);
3063 break;
3064 }
3065
3066 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3067 BIO_BUG_ON(nbytes > bio->bi_size);
3068
3069 /*
3070 * not a complete bvec done
3071 */
3072 if (unlikely(nbytes > nr_bytes)) {
3073 bio_nbytes += nr_bytes;
3074 total_bytes += nr_bytes;
3075 break;
3076 }
3077
3078 /*
3079 * advance to the next vector
3080 */
3081 next_idx++;
3082 bio_nbytes += nbytes;
3083 }
3084
3085 total_bytes += nbytes;
3086 nr_bytes -= nbytes;
3087
3088 if ((bio = req->bio)) {
3089 /*
3090 * end more in this run, or just return 'not-done'
3091 */
3092 if (unlikely(nr_bytes <= 0))
3093 break;
3094 }
3095 }
3096
3097 /*
3098 * completely done
3099 */
3100 if (!req->bio)
3101 return 0;
3102
3103 /*
3104 * if the request wasn't completed, update state
3105 */
3106 if (bio_nbytes) {
3107 bio_endio(bio, bio_nbytes, error);
3108 bio->bi_idx += next_idx;
3109 bio_iovec(bio)->bv_offset += nr_bytes;
3110 bio_iovec(bio)->bv_len -= nr_bytes;
3111 }
3112
3113 blk_recalc_rq_sectors(req, total_bytes >> 9);
3114 blk_recalc_rq_segments(req);
3115 return 1;
3116}
3117
3118/**
3119 * end_that_request_first - end I/O on a request
3120 * @req: the request being processed
3121 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3122 * @nr_sectors: number of sectors to end I/O on
3123 *
3124 * Description:
3125 * Ends I/O on a number of sectors attached to @req, and sets it up
3126 * for the next range of segments (if any) in the cluster.
3127 *
3128 * Return:
3129 * 0 - we are done with this request, call end_that_request_last()
3130 * 1 - still buffers pending for this request
3131 **/
3132int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3133{
3134 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3135}
3136
3137EXPORT_SYMBOL(end_that_request_first);
3138
3139/**
3140 * end_that_request_chunk - end I/O on a request
3141 * @req: the request being processed
3142 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3143 * @nr_bytes: number of bytes to complete
3144 *
3145 * Description:
3146 * Ends I/O on a number of bytes attached to @req, and sets it up
3147 * for the next range of segments (if any). Like end_that_request_first(),
3148 * but deals with bytes instead of sectors.
3149 *
3150 * Return:
3151 * 0 - we are done with this request, call end_that_request_last()
3152 * 1 - still buffers pending for this request
3153 **/
3154int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3155{
3156 return __end_that_request_first(req, uptodate, nr_bytes);
3157}
3158
3159EXPORT_SYMBOL(end_that_request_chunk);
3160
3161/*
3162 * queue lock must be held
3163 */
3164void end_that_request_last(struct request *req)
3165{
3166 struct gendisk *disk = req->rq_disk;
3167
3168 if (unlikely(laptop_mode) && blk_fs_request(req))
3169 laptop_io_completion();
3170
3171 if (disk && blk_fs_request(req)) {
3172 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003173 const int rw = rq_data_dir(req);
3174
3175 __disk_stat_inc(disk, ios[rw]);
3176 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 disk_round_stats(disk);
3178 disk->in_flight--;
3179 }
3180 if (req->end_io)
3181 req->end_io(req);
3182 else
3183 __blk_put_request(req->q, req);
3184}
3185
3186EXPORT_SYMBOL(end_that_request_last);
3187
3188void end_request(struct request *req, int uptodate)
3189{
3190 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3191 add_disk_randomness(req->rq_disk);
3192 blkdev_dequeue_request(req);
3193 end_that_request_last(req);
3194 }
3195}
3196
3197EXPORT_SYMBOL(end_request);
3198
3199void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3200{
3201 /* first three bits are identical in rq->flags and bio->bi_rw */
3202 rq->flags |= (bio->bi_rw & 7);
3203
3204 rq->nr_phys_segments = bio_phys_segments(q, bio);
3205 rq->nr_hw_segments = bio_hw_segments(q, bio);
3206 rq->current_nr_sectors = bio_cur_sectors(bio);
3207 rq->hard_cur_sectors = rq->current_nr_sectors;
3208 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3209 rq->buffer = bio_data(bio);
3210
3211 rq->bio = rq->biotail = bio;
3212}
3213
3214EXPORT_SYMBOL(blk_rq_bio_prep);
3215
3216int kblockd_schedule_work(struct work_struct *work)
3217{
3218 return queue_work(kblockd_workqueue, work);
3219}
3220
3221EXPORT_SYMBOL(kblockd_schedule_work);
3222
3223void kblockd_flush(void)
3224{
3225 flush_workqueue(kblockd_workqueue);
3226}
3227EXPORT_SYMBOL(kblockd_flush);
3228
3229int __init blk_dev_init(void)
3230{
3231 kblockd_workqueue = create_workqueue("kblockd");
3232 if (!kblockd_workqueue)
3233 panic("Failed to create kblockd\n");
3234
3235 request_cachep = kmem_cache_create("blkdev_requests",
3236 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3237
3238 requestq_cachep = kmem_cache_create("blkdev_queue",
3239 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3240
3241 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3242 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3243
3244 blk_max_low_pfn = max_low_pfn;
3245 blk_max_pfn = max_pfn;
3246
3247 return 0;
3248}
3249
3250/*
3251 * IO Context helper functions
3252 */
3253void put_io_context(struct io_context *ioc)
3254{
3255 if (ioc == NULL)
3256 return;
3257
3258 BUG_ON(atomic_read(&ioc->refcount) == 0);
3259
3260 if (atomic_dec_and_test(&ioc->refcount)) {
3261 if (ioc->aic && ioc->aic->dtor)
3262 ioc->aic->dtor(ioc->aic);
3263 if (ioc->cic && ioc->cic->dtor)
3264 ioc->cic->dtor(ioc->cic);
3265
3266 kmem_cache_free(iocontext_cachep, ioc);
3267 }
3268}
3269EXPORT_SYMBOL(put_io_context);
3270
3271/* Called by the exitting task */
3272void exit_io_context(void)
3273{
3274 unsigned long flags;
3275 struct io_context *ioc;
3276
3277 local_irq_save(flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02003278 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 ioc = current->io_context;
3280 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003281 ioc->task = NULL;
3282 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 local_irq_restore(flags);
3284
3285 if (ioc->aic && ioc->aic->exit)
3286 ioc->aic->exit(ioc->aic);
3287 if (ioc->cic && ioc->cic->exit)
3288 ioc->cic->exit(ioc->cic);
3289
3290 put_io_context(ioc);
3291}
3292
3293/*
3294 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003295 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003297 * This returned IO context doesn't have a specifically elevated refcount,
3298 * but since the current task itself holds a reference, the context can be
3299 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 */
Al Viro8267e262005-10-21 03:20:53 -04003301struct io_context *current_io_context(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302{
3303 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 struct io_context *ret;
3305
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003307 if (likely(ret))
3308 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309
3310 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3311 if (ret) {
3312 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003313 ret->task = current;
3314 ret->set_ioprio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 ret->last_waited = jiffies; /* doesn't matter... */
3316 ret->nr_batch_requests = 0; /* because this is 0 */
3317 ret->aic = NULL;
3318 ret->cic = NULL;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003319 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 }
3321
3322 return ret;
3323}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003324EXPORT_SYMBOL(current_io_context);
3325
3326/*
3327 * If the current task has no IO context then create one and initialise it.
3328 * If it does have a context, take a ref on it.
3329 *
3330 * This is always called in the context of the task which submitted the I/O.
3331 */
Al Viro8267e262005-10-21 03:20:53 -04003332struct io_context *get_io_context(gfp_t gfp_flags)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003333{
3334 struct io_context *ret;
3335 ret = current_io_context(gfp_flags);
3336 if (likely(ret))
3337 atomic_inc(&ret->refcount);
3338 return ret;
3339}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340EXPORT_SYMBOL(get_io_context);
3341
3342void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3343{
3344 struct io_context *src = *psrc;
3345 struct io_context *dst = *pdst;
3346
3347 if (src) {
3348 BUG_ON(atomic_read(&src->refcount) == 0);
3349 atomic_inc(&src->refcount);
3350 put_io_context(dst);
3351 *pdst = src;
3352 }
3353}
3354EXPORT_SYMBOL(copy_io_context);
3355
3356void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3357{
3358 struct io_context *temp;
3359 temp = *ioc1;
3360 *ioc1 = *ioc2;
3361 *ioc2 = temp;
3362}
3363EXPORT_SYMBOL(swap_io_context);
3364
3365/*
3366 * sysfs parts below
3367 */
3368struct queue_sysfs_entry {
3369 struct attribute attr;
3370 ssize_t (*show)(struct request_queue *, char *);
3371 ssize_t (*store)(struct request_queue *, const char *, size_t);
3372};
3373
3374static ssize_t
3375queue_var_show(unsigned int var, char *page)
3376{
3377 return sprintf(page, "%d\n", var);
3378}
3379
3380static ssize_t
3381queue_var_store(unsigned long *var, const char *page, size_t count)
3382{
3383 char *p = (char *) page;
3384
3385 *var = simple_strtoul(p, &p, 10);
3386 return count;
3387}
3388
3389static ssize_t queue_requests_show(struct request_queue *q, char *page)
3390{
3391 return queue_var_show(q->nr_requests, (page));
3392}
3393
3394static ssize_t
3395queue_requests_store(struct request_queue *q, const char *page, size_t count)
3396{
3397 struct request_list *rl = &q->rq;
3398
3399 int ret = queue_var_store(&q->nr_requests, page, count);
3400 if (q->nr_requests < BLKDEV_MIN_RQ)
3401 q->nr_requests = BLKDEV_MIN_RQ;
3402 blk_queue_congestion_threshold(q);
3403
3404 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3405 set_queue_congested(q, READ);
3406 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3407 clear_queue_congested(q, READ);
3408
3409 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3410 set_queue_congested(q, WRITE);
3411 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3412 clear_queue_congested(q, WRITE);
3413
3414 if (rl->count[READ] >= q->nr_requests) {
3415 blk_set_queue_full(q, READ);
3416 } else if (rl->count[READ]+1 <= q->nr_requests) {
3417 blk_clear_queue_full(q, READ);
3418 wake_up(&rl->wait[READ]);
3419 }
3420
3421 if (rl->count[WRITE] >= q->nr_requests) {
3422 blk_set_queue_full(q, WRITE);
3423 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3424 blk_clear_queue_full(q, WRITE);
3425 wake_up(&rl->wait[WRITE]);
3426 }
3427 return ret;
3428}
3429
3430static ssize_t queue_ra_show(struct request_queue *q, char *page)
3431{
3432 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3433
3434 return queue_var_show(ra_kb, (page));
3435}
3436
3437static ssize_t
3438queue_ra_store(struct request_queue *q, const char *page, size_t count)
3439{
3440 unsigned long ra_kb;
3441 ssize_t ret = queue_var_store(&ra_kb, page, count);
3442
3443 spin_lock_irq(q->queue_lock);
3444 if (ra_kb > (q->max_sectors >> 1))
3445 ra_kb = (q->max_sectors >> 1);
3446
3447 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3448 spin_unlock_irq(q->queue_lock);
3449
3450 return ret;
3451}
3452
3453static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3454{
3455 int max_sectors_kb = q->max_sectors >> 1;
3456
3457 return queue_var_show(max_sectors_kb, (page));
3458}
3459
3460static ssize_t
3461queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3462{
3463 unsigned long max_sectors_kb,
3464 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3465 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3466 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3467 int ra_kb;
3468
3469 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3470 return -EINVAL;
3471 /*
3472 * Take the queue lock to update the readahead and max_sectors
3473 * values synchronously:
3474 */
3475 spin_lock_irq(q->queue_lock);
3476 /*
3477 * Trim readahead window as well, if necessary:
3478 */
3479 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3480 if (ra_kb > max_sectors_kb)
3481 q->backing_dev_info.ra_pages =
3482 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3483
3484 q->max_sectors = max_sectors_kb << 1;
3485 spin_unlock_irq(q->queue_lock);
3486
3487 return ret;
3488}
3489
3490static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3491{
3492 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3493
3494 return queue_var_show(max_hw_sectors_kb, (page));
3495}
3496
3497
3498static struct queue_sysfs_entry queue_requests_entry = {
3499 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3500 .show = queue_requests_show,
3501 .store = queue_requests_store,
3502};
3503
3504static struct queue_sysfs_entry queue_ra_entry = {
3505 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3506 .show = queue_ra_show,
3507 .store = queue_ra_store,
3508};
3509
3510static struct queue_sysfs_entry queue_max_sectors_entry = {
3511 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3512 .show = queue_max_sectors_show,
3513 .store = queue_max_sectors_store,
3514};
3515
3516static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3517 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3518 .show = queue_max_hw_sectors_show,
3519};
3520
3521static struct queue_sysfs_entry queue_iosched_entry = {
3522 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3523 .show = elv_iosched_show,
3524 .store = elv_iosched_store,
3525};
3526
3527static struct attribute *default_attrs[] = {
3528 &queue_requests_entry.attr,
3529 &queue_ra_entry.attr,
3530 &queue_max_hw_sectors_entry.attr,
3531 &queue_max_sectors_entry.attr,
3532 &queue_iosched_entry.attr,
3533 NULL,
3534};
3535
3536#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3537
3538static ssize_t
3539queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3540{
3541 struct queue_sysfs_entry *entry = to_queue(attr);
3542 struct request_queue *q;
3543
3544 q = container_of(kobj, struct request_queue, kobj);
3545 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003546 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547
3548 return entry->show(q, page);
3549}
3550
3551static ssize_t
3552queue_attr_store(struct kobject *kobj, struct attribute *attr,
3553 const char *page, size_t length)
3554{
3555 struct queue_sysfs_entry *entry = to_queue(attr);
3556 struct request_queue *q;
3557
3558 q = container_of(kobj, struct request_queue, kobj);
3559 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003560 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
3562 return entry->store(q, page, length);
3563}
3564
3565static struct sysfs_ops queue_sysfs_ops = {
3566 .show = queue_attr_show,
3567 .store = queue_attr_store,
3568};
3569
Adrian Bunk93d17d32005-06-25 14:59:10 -07003570static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 .sysfs_ops = &queue_sysfs_ops,
3572 .default_attrs = default_attrs,
3573};
3574
3575int blk_register_queue(struct gendisk *disk)
3576{
3577 int ret;
3578
3579 request_queue_t *q = disk->queue;
3580
3581 if (!q || !q->request_fn)
3582 return -ENXIO;
3583
3584 q->kobj.parent = kobject_get(&disk->kobj);
3585 if (!q->kobj.parent)
3586 return -EBUSY;
3587
3588 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3589 q->kobj.ktype = &queue_ktype;
3590
3591 ret = kobject_register(&q->kobj);
3592 if (ret < 0)
3593 return ret;
3594
3595 ret = elv_register_queue(q);
3596 if (ret) {
3597 kobject_unregister(&q->kobj);
3598 return ret;
3599 }
3600
3601 return 0;
3602}
3603
3604void blk_unregister_queue(struct gendisk *disk)
3605{
3606 request_queue_t *q = disk->queue;
3607
3608 if (q && q->request_fn) {
3609 elv_unregister_queue(q);
3610
3611 kobject_unregister(&q->kobj);
3612 kobject_put(&disk->kobj);
3613 }
3614}