blob: 8b1ae69bc5ac0b012fc8985091fad14395c4157e [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;
Mike Christiedefd94b2005-12-05 02:37:06 -0600242 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 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 */
Tejun Heo8ffdc652006-01-06 09:49:03 +0100347static void blk_pre_flush_end_io(struct request *flush_rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
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
Tejun Heo8ffdc652006-01-06 09:49:03 +0100365static void blk_post_flush_end_io(struct request *flush_rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
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
Mike Christiedefd94b2005-12-05 02:37:06 -0600558 if (BLK_DEF_MAX_SECTORS > max_sectors)
559 q->max_hw_sectors = q->max_sectors = max_sectors;
560 else {
561 q->max_sectors = BLK_DEF_MAX_SECTORS;
562 q->max_hw_sectors = max_sectors;
563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564}
565
566EXPORT_SYMBOL(blk_queue_max_sectors);
567
568/**
569 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
570 * @q: the request queue for the device
571 * @max_segments: max number of segments
572 *
573 * Description:
574 * Enables a low level driver to set an upper limit on the number of
575 * physical data segments in a request. This would be the largest sized
576 * scatter list the driver could handle.
577 **/
578void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
579{
580 if (!max_segments) {
581 max_segments = 1;
582 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
583 }
584
585 q->max_phys_segments = max_segments;
586}
587
588EXPORT_SYMBOL(blk_queue_max_phys_segments);
589
590/**
591 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
592 * @q: the request queue for the device
593 * @max_segments: max number of segments
594 *
595 * Description:
596 * Enables a low level driver to set an upper limit on the number of
597 * hw data segments in a request. This would be the largest number of
598 * address/length pairs the host adapter can actually give as once
599 * to the device.
600 **/
601void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
602{
603 if (!max_segments) {
604 max_segments = 1;
605 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
606 }
607
608 q->max_hw_segments = max_segments;
609}
610
611EXPORT_SYMBOL(blk_queue_max_hw_segments);
612
613/**
614 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
615 * @q: the request queue for the device
616 * @max_size: max size of segment in bytes
617 *
618 * Description:
619 * Enables a low level driver to set an upper limit on the size of a
620 * coalesced segment
621 **/
622void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
623{
624 if (max_size < PAGE_CACHE_SIZE) {
625 max_size = PAGE_CACHE_SIZE;
626 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
627 }
628
629 q->max_segment_size = max_size;
630}
631
632EXPORT_SYMBOL(blk_queue_max_segment_size);
633
634/**
635 * blk_queue_hardsect_size - set hardware sector size for the queue
636 * @q: the request queue for the device
637 * @size: the hardware sector size, in bytes
638 *
639 * Description:
640 * This should typically be set to the lowest possible sector size
641 * that the hardware can operate on (possible without reverting to
642 * even internal read-modify-write operations). Usually the default
643 * of 512 covers most hardware.
644 **/
645void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
646{
647 q->hardsect_size = size;
648}
649
650EXPORT_SYMBOL(blk_queue_hardsect_size);
651
652/*
653 * Returns the minimum that is _not_ zero, unless both are zero.
654 */
655#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
656
657/**
658 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
659 * @t: the stacking driver (top)
660 * @b: the underlying device (bottom)
661 **/
662void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
663{
664 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600665 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
666 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
669 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
670 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
671 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
672}
673
674EXPORT_SYMBOL(blk_queue_stack_limits);
675
676/**
677 * blk_queue_segment_boundary - set boundary rules for segment merging
678 * @q: the request queue for the device
679 * @mask: the memory boundary mask
680 **/
681void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
682{
683 if (mask < PAGE_CACHE_SIZE - 1) {
684 mask = PAGE_CACHE_SIZE - 1;
685 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
686 }
687
688 q->seg_boundary_mask = mask;
689}
690
691EXPORT_SYMBOL(blk_queue_segment_boundary);
692
693/**
694 * blk_queue_dma_alignment - set dma length and memory alignment
695 * @q: the request queue for the device
696 * @mask: alignment mask
697 *
698 * description:
699 * set required memory and length aligment for direct dma transactions.
700 * this is used when buiding direct io requests for the queue.
701 *
702 **/
703void blk_queue_dma_alignment(request_queue_t *q, int mask)
704{
705 q->dma_alignment = mask;
706}
707
708EXPORT_SYMBOL(blk_queue_dma_alignment);
709
710/**
711 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 * @q: The request queue for the device
713 * @tag: The tag of the request
714 *
715 * Notes:
716 * Should be used when a device returns a tag and you want to match
717 * it with a request.
718 *
719 * no locks need be held.
720 **/
721struct request *blk_queue_find_tag(request_queue_t *q, int tag)
722{
723 struct blk_queue_tag *bqt = q->queue_tags;
724
Tejun Heoba025082005-08-05 13:28:11 -0700725 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return NULL;
727
728 return bqt->tag_index[tag];
729}
730
731EXPORT_SYMBOL(blk_queue_find_tag);
732
733/**
734 * __blk_queue_free_tags - release tag maintenance info
735 * @q: the request queue for the device
736 *
737 * Notes:
738 * blk_cleanup_queue() will take care of calling this function, if tagging
739 * has been used. So there's no need to call this directly.
740 **/
741static void __blk_queue_free_tags(request_queue_t *q)
742{
743 struct blk_queue_tag *bqt = q->queue_tags;
744
745 if (!bqt)
746 return;
747
748 if (atomic_dec_and_test(&bqt->refcnt)) {
749 BUG_ON(bqt->busy);
750 BUG_ON(!list_empty(&bqt->busy_list));
751
752 kfree(bqt->tag_index);
753 bqt->tag_index = NULL;
754
755 kfree(bqt->tag_map);
756 bqt->tag_map = NULL;
757
758 kfree(bqt);
759 }
760
761 q->queue_tags = NULL;
762 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
763}
764
765/**
766 * blk_queue_free_tags - release tag maintenance info
767 * @q: the request queue for the device
768 *
769 * Notes:
770 * This is used to disabled tagged queuing to a device, yet leave
771 * queue in function.
772 **/
773void blk_queue_free_tags(request_queue_t *q)
774{
775 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
776}
777
778EXPORT_SYMBOL(blk_queue_free_tags);
779
780static int
781init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
782{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 struct request **tag_index;
784 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700785 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 if (depth > q->nr_requests * 2) {
788 depth = q->nr_requests * 2;
789 printk(KERN_ERR "%s: adjusted depth to %d\n",
790 __FUNCTION__, depth);
791 }
792
793 tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
794 if (!tag_index)
795 goto fail;
796
Tejun Heof7d37d02005-06-23 00:08:50 -0700797 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700798 tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 if (!tag_map)
800 goto fail;
801
802 memset(tag_index, 0, depth * sizeof(struct request *));
Tejun Heofa72b902005-06-23 00:08:49 -0700803 memset(tag_map, 0, nr_ulongs * sizeof(unsigned long));
Tejun Heoba025082005-08-05 13:28:11 -0700804 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 tags->tag_index = tag_index;
807 tags->tag_map = tag_map;
808
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return 0;
810fail:
811 kfree(tag_index);
812 return -ENOMEM;
813}
814
815/**
816 * blk_queue_init_tags - initialize the queue tag info
817 * @q: the request queue for the device
818 * @depth: the maximum queue depth supported
819 * @tags: the tag to use
820 **/
821int blk_queue_init_tags(request_queue_t *q, int depth,
822 struct blk_queue_tag *tags)
823{
824 int rc;
825
826 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
827
828 if (!tags && !q->queue_tags) {
829 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
830 if (!tags)
831 goto fail;
832
833 if (init_tag_map(q, tags, depth))
834 goto fail;
835
836 INIT_LIST_HEAD(&tags->busy_list);
837 tags->busy = 0;
838 atomic_set(&tags->refcnt, 1);
839 } else if (q->queue_tags) {
840 if ((rc = blk_queue_resize_tags(q, depth)))
841 return rc;
842 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
843 return 0;
844 } else
845 atomic_inc(&tags->refcnt);
846
847 /*
848 * assign it, all done
849 */
850 q->queue_tags = tags;
851 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
852 return 0;
853fail:
854 kfree(tags);
855 return -ENOMEM;
856}
857
858EXPORT_SYMBOL(blk_queue_init_tags);
859
860/**
861 * blk_queue_resize_tags - change the queueing depth
862 * @q: the request queue for the device
863 * @new_depth: the new max command queueing depth
864 *
865 * Notes:
866 * Must be called with the queue lock held.
867 **/
868int blk_queue_resize_tags(request_queue_t *q, int new_depth)
869{
870 struct blk_queue_tag *bqt = q->queue_tags;
871 struct request **tag_index;
872 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700873 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 if (!bqt)
876 return -ENXIO;
877
878 /*
Tejun Heoba025082005-08-05 13:28:11 -0700879 * if we already have large enough real_max_depth. just
880 * adjust max_depth. *NOTE* as requests with tag value
881 * between new_depth and real_max_depth can be in-flight, tag
882 * map can not be shrunk blindly here.
883 */
884 if (new_depth <= bqt->real_max_depth) {
885 bqt->max_depth = new_depth;
886 return 0;
887 }
888
889 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 * save the old state info, so we can copy it back
891 */
892 tag_index = bqt->tag_index;
893 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -0700894 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
896 if (init_tag_map(q, bqt, new_depth))
897 return -ENOMEM;
898
899 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -0700900 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700901 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 kfree(tag_index);
904 kfree(tag_map);
905 return 0;
906}
907
908EXPORT_SYMBOL(blk_queue_resize_tags);
909
910/**
911 * blk_queue_end_tag - end tag operations for a request
912 * @q: the request queue for the device
913 * @rq: the request that has completed
914 *
915 * Description:
916 * Typically called when end_that_request_first() returns 0, meaning
917 * all transfers have been done for a request. It's important to call
918 * this function before end_that_request_last(), as that will put the
919 * request back on the free list thus corrupting the internal tag list.
920 *
921 * Notes:
922 * queue lock must be held.
923 **/
924void blk_queue_end_tag(request_queue_t *q, struct request *rq)
925{
926 struct blk_queue_tag *bqt = q->queue_tags;
927 int tag = rq->tag;
928
929 BUG_ON(tag == -1);
930
Tejun Heoba025082005-08-05 13:28:11 -0700931 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -0700932 /*
933 * This can happen after tag depth has been reduced.
934 * FIXME: how about a warning or info message here?
935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return;
937
938 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -0700939 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
940 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return;
942 }
943
944 list_del_init(&rq->queuelist);
945 rq->flags &= ~REQ_QUEUED;
946 rq->tag = -1;
947
948 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -0700949 printk(KERN_ERR "%s: tag %d is missing\n",
950 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
952 bqt->tag_index[tag] = NULL;
953 bqt->busy--;
954}
955
956EXPORT_SYMBOL(blk_queue_end_tag);
957
958/**
959 * blk_queue_start_tag - find a free tag and assign it
960 * @q: the request queue for the device
961 * @rq: the block request that needs tagging
962 *
963 * Description:
964 * This can either be used as a stand-alone helper, or possibly be
965 * assigned as the queue &prep_rq_fn (in which case &struct request
966 * automagically gets a tag assigned). Note that this function
967 * assumes that any type of request can be queued! if this is not
968 * true for your device, you must check the request type before
969 * calling this function. The request will also be removed from
970 * the request queue, so it's the drivers responsibility to readd
971 * it if it should need to be restarted for some reason.
972 *
973 * Notes:
974 * queue lock must be held.
975 **/
976int blk_queue_start_tag(request_queue_t *q, struct request *rq)
977{
978 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -0700979 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if (unlikely((rq->flags & REQ_QUEUED))) {
982 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -0700983 "%s: request %p for device [%s] already tagged %d",
984 __FUNCTION__, rq,
985 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 BUG();
987 }
988
Tejun Heo2bf0fda2005-06-23 00:08:48 -0700989 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
990 if (tag >= bqt->max_depth)
991 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 __set_bit(tag, bqt->tag_map);
994
995 rq->flags |= REQ_QUEUED;
996 rq->tag = tag;
997 bqt->tag_index[tag] = rq;
998 blkdev_dequeue_request(rq);
999 list_add(&rq->queuelist, &bqt->busy_list);
1000 bqt->busy++;
1001 return 0;
1002}
1003
1004EXPORT_SYMBOL(blk_queue_start_tag);
1005
1006/**
1007 * blk_queue_invalidate_tags - invalidate all pending tags
1008 * @q: the request queue for the device
1009 *
1010 * Description:
1011 * Hardware conditions may dictate a need to stop all pending requests.
1012 * In this case, we will safely clear the block side of the tag queue and
1013 * readd all requests to the request queue in the right order.
1014 *
1015 * Notes:
1016 * queue lock must be held.
1017 **/
1018void blk_queue_invalidate_tags(request_queue_t *q)
1019{
1020 struct blk_queue_tag *bqt = q->queue_tags;
1021 struct list_head *tmp, *n;
1022 struct request *rq;
1023
1024 list_for_each_safe(tmp, n, &bqt->busy_list) {
1025 rq = list_entry_rq(tmp);
1026
1027 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001028 printk(KERN_ERR
1029 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 list_del_init(&rq->queuelist);
1031 rq->flags &= ~REQ_QUEUED;
1032 } else
1033 blk_queue_end_tag(q, rq);
1034
1035 rq->flags &= ~REQ_STARTED;
1036 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1037 }
1038}
1039
1040EXPORT_SYMBOL(blk_queue_invalidate_tags);
1041
Arjan van de Ven64100092006-01-06 09:46:02 +01001042static const char * const rq_flags[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 "REQ_RW",
1044 "REQ_FAILFAST",
Tejun Heo8922e162005-10-20 16:23:44 +02001045 "REQ_SORTED",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 "REQ_SOFTBARRIER",
1047 "REQ_HARDBARRIER",
1048 "REQ_CMD",
1049 "REQ_NOMERGE",
1050 "REQ_STARTED",
1051 "REQ_DONTPREP",
1052 "REQ_QUEUED",
Tejun Heocb98fc82005-10-28 08:29:39 +02001053 "REQ_ELVPRIV",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 "REQ_PC",
1055 "REQ_BLOCK_PC",
1056 "REQ_SENSE",
1057 "REQ_FAILED",
1058 "REQ_QUIET",
1059 "REQ_SPECIAL",
1060 "REQ_DRIVE_CMD",
1061 "REQ_DRIVE_TASK",
1062 "REQ_DRIVE_TASKFILE",
1063 "REQ_PREEMPT",
1064 "REQ_PM_SUSPEND",
1065 "REQ_PM_RESUME",
1066 "REQ_PM_SHUTDOWN",
1067};
1068
1069void blk_dump_rq_flags(struct request *rq, char *msg)
1070{
1071 int bit;
1072
1073 printk("%s: dev %s: flags = ", msg,
1074 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1075 bit = 0;
1076 do {
1077 if (rq->flags & (1 << bit))
1078 printk("%s ", rq_flags[bit]);
1079 bit++;
1080 } while (bit < __REQ_NR_BITS);
1081
1082 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1083 rq->nr_sectors,
1084 rq->current_nr_sectors);
1085 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1086
1087 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1088 printk("cdb: ");
1089 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1090 printk("%02x ", rq->cmd[bit]);
1091 printk("\n");
1092 }
1093}
1094
1095EXPORT_SYMBOL(blk_dump_rq_flags);
1096
1097void blk_recount_segments(request_queue_t *q, struct bio *bio)
1098{
1099 struct bio_vec *bv, *bvprv = NULL;
1100 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1101 int high, highprv = 1;
1102
1103 if (unlikely(!bio->bi_io_vec))
1104 return;
1105
1106 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1107 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1108 bio_for_each_segment(bv, bio, i) {
1109 /*
1110 * the trick here is making sure that a high page is never
1111 * considered part of another segment, since that might
1112 * change with the bounce page.
1113 */
1114 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1115 if (high || highprv)
1116 goto new_hw_segment;
1117 if (cluster) {
1118 if (seg_size + bv->bv_len > q->max_segment_size)
1119 goto new_segment;
1120 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1121 goto new_segment;
1122 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1123 goto new_segment;
1124 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1125 goto new_hw_segment;
1126
1127 seg_size += bv->bv_len;
1128 hw_seg_size += bv->bv_len;
1129 bvprv = bv;
1130 continue;
1131 }
1132new_segment:
1133 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1134 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1135 hw_seg_size += bv->bv_len;
1136 } else {
1137new_hw_segment:
1138 if (hw_seg_size > bio->bi_hw_front_size)
1139 bio->bi_hw_front_size = hw_seg_size;
1140 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1141 nr_hw_segs++;
1142 }
1143
1144 nr_phys_segs++;
1145 bvprv = bv;
1146 seg_size = bv->bv_len;
1147 highprv = high;
1148 }
1149 if (hw_seg_size > bio->bi_hw_back_size)
1150 bio->bi_hw_back_size = hw_seg_size;
1151 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1152 bio->bi_hw_front_size = hw_seg_size;
1153 bio->bi_phys_segments = nr_phys_segs;
1154 bio->bi_hw_segments = nr_hw_segs;
1155 bio->bi_flags |= (1 << BIO_SEG_VALID);
1156}
1157
1158
Adrian Bunk93d17d32005-06-25 14:59:10 -07001159static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 struct bio *nxt)
1161{
1162 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1163 return 0;
1164
1165 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1166 return 0;
1167 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1168 return 0;
1169
1170 /*
1171 * bio and nxt are contigous in memory, check if the queue allows
1172 * these two to be merged into one
1173 */
1174 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1175 return 1;
1176
1177 return 0;
1178}
1179
Adrian Bunk93d17d32005-06-25 14:59:10 -07001180static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 struct bio *nxt)
1182{
1183 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1184 blk_recount_segments(q, bio);
1185 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1186 blk_recount_segments(q, nxt);
1187 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1188 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1189 return 0;
1190 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1191 return 0;
1192
1193 return 1;
1194}
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196/*
1197 * map a request to scatterlist, return number of sg entries setup. Caller
1198 * must make sure sg can hold rq->nr_phys_segments entries
1199 */
1200int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1201{
1202 struct bio_vec *bvec, *bvprv;
1203 struct bio *bio;
1204 int nsegs, i, cluster;
1205
1206 nsegs = 0;
1207 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1208
1209 /*
1210 * for each bio in rq
1211 */
1212 bvprv = NULL;
1213 rq_for_each_bio(bio, rq) {
1214 /*
1215 * for each segment in bio
1216 */
1217 bio_for_each_segment(bvec, bio, i) {
1218 int nbytes = bvec->bv_len;
1219
1220 if (bvprv && cluster) {
1221 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1222 goto new_segment;
1223
1224 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1225 goto new_segment;
1226 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1227 goto new_segment;
1228
1229 sg[nsegs - 1].length += nbytes;
1230 } else {
1231new_segment:
1232 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1233 sg[nsegs].page = bvec->bv_page;
1234 sg[nsegs].length = nbytes;
1235 sg[nsegs].offset = bvec->bv_offset;
1236
1237 nsegs++;
1238 }
1239 bvprv = bvec;
1240 } /* segments in bio */
1241 } /* bios in rq */
1242
1243 return nsegs;
1244}
1245
1246EXPORT_SYMBOL(blk_rq_map_sg);
1247
1248/*
1249 * the standard queue merge functions, can be overridden with device
1250 * specific ones if so desired
1251 */
1252
1253static inline int ll_new_mergeable(request_queue_t *q,
1254 struct request *req,
1255 struct bio *bio)
1256{
1257 int nr_phys_segs = bio_phys_segments(q, bio);
1258
1259 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1260 req->flags |= REQ_NOMERGE;
1261 if (req == q->last_merge)
1262 q->last_merge = NULL;
1263 return 0;
1264 }
1265
1266 /*
1267 * A hw segment is just getting larger, bump just the phys
1268 * counter.
1269 */
1270 req->nr_phys_segments += nr_phys_segs;
1271 return 1;
1272}
1273
1274static inline int ll_new_hw_segment(request_queue_t *q,
1275 struct request *req,
1276 struct bio *bio)
1277{
1278 int nr_hw_segs = bio_hw_segments(q, bio);
1279 int nr_phys_segs = bio_phys_segments(q, bio);
1280
1281 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1282 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1283 req->flags |= REQ_NOMERGE;
1284 if (req == q->last_merge)
1285 q->last_merge = NULL;
1286 return 0;
1287 }
1288
1289 /*
1290 * This will form the start of a new hw segment. Bump both
1291 * counters.
1292 */
1293 req->nr_hw_segments += nr_hw_segs;
1294 req->nr_phys_segments += nr_phys_segs;
1295 return 1;
1296}
1297
1298static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1299 struct bio *bio)
1300{
Mike Christiedefd94b2005-12-05 02:37:06 -06001301 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 int len;
1303
Mike Christiedefd94b2005-12-05 02:37:06 -06001304 if (unlikely(blk_pc_request(req)))
1305 max_sectors = q->max_hw_sectors;
1306 else
1307 max_sectors = q->max_sectors;
1308
1309 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 req->flags |= REQ_NOMERGE;
1311 if (req == q->last_merge)
1312 q->last_merge = NULL;
1313 return 0;
1314 }
1315 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1316 blk_recount_segments(q, req->biotail);
1317 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1318 blk_recount_segments(q, bio);
1319 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1320 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1321 !BIOVEC_VIRT_OVERSIZE(len)) {
1322 int mergeable = ll_new_mergeable(q, req, bio);
1323
1324 if (mergeable) {
1325 if (req->nr_hw_segments == 1)
1326 req->bio->bi_hw_front_size = len;
1327 if (bio->bi_hw_segments == 1)
1328 bio->bi_hw_back_size = len;
1329 }
1330 return mergeable;
1331 }
1332
1333 return ll_new_hw_segment(q, req, bio);
1334}
1335
1336static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1337 struct bio *bio)
1338{
Mike Christiedefd94b2005-12-05 02:37:06 -06001339 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 int len;
1341
Mike Christiedefd94b2005-12-05 02:37:06 -06001342 if (unlikely(blk_pc_request(req)))
1343 max_sectors = q->max_hw_sectors;
1344 else
1345 max_sectors = q->max_sectors;
1346
1347
1348 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 req->flags |= REQ_NOMERGE;
1350 if (req == q->last_merge)
1351 q->last_merge = NULL;
1352 return 0;
1353 }
1354 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1355 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1356 blk_recount_segments(q, bio);
1357 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1358 blk_recount_segments(q, req->bio);
1359 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1360 !BIOVEC_VIRT_OVERSIZE(len)) {
1361 int mergeable = ll_new_mergeable(q, req, bio);
1362
1363 if (mergeable) {
1364 if (bio->bi_hw_segments == 1)
1365 bio->bi_hw_front_size = len;
1366 if (req->nr_hw_segments == 1)
1367 req->biotail->bi_hw_back_size = len;
1368 }
1369 return mergeable;
1370 }
1371
1372 return ll_new_hw_segment(q, req, bio);
1373}
1374
1375static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1376 struct request *next)
1377{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001378 int total_phys_segments;
1379 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 /*
1382 * First check if the either of the requests are re-queued
1383 * requests. Can't merge them if they are.
1384 */
1385 if (req->special || next->special)
1386 return 0;
1387
1388 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001389 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 */
1391 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1392 return 0;
1393
1394 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1395 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1396 total_phys_segments--;
1397
1398 if (total_phys_segments > q->max_phys_segments)
1399 return 0;
1400
1401 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1402 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1403 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1404 /*
1405 * propagate the combined length to the end of the requests
1406 */
1407 if (req->nr_hw_segments == 1)
1408 req->bio->bi_hw_front_size = len;
1409 if (next->nr_hw_segments == 1)
1410 next->biotail->bi_hw_back_size = len;
1411 total_hw_segments--;
1412 }
1413
1414 if (total_hw_segments > q->max_hw_segments)
1415 return 0;
1416
1417 /* Merge is OK... */
1418 req->nr_phys_segments = total_phys_segments;
1419 req->nr_hw_segments = total_hw_segments;
1420 return 1;
1421}
1422
1423/*
1424 * "plug" the device if there are no outstanding requests: this will
1425 * force the transfer to start only after we have put all the requests
1426 * on the list.
1427 *
1428 * This is called with interrupts off and no requests on the queue and
1429 * with the queue lock held.
1430 */
1431void blk_plug_device(request_queue_t *q)
1432{
1433 WARN_ON(!irqs_disabled());
1434
1435 /*
1436 * don't plug a stopped queue, it must be paired with blk_start_queue()
1437 * which will restart the queueing
1438 */
1439 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1440 return;
1441
1442 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1443 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1444}
1445
1446EXPORT_SYMBOL(blk_plug_device);
1447
1448/*
1449 * remove the queue from the plugged list, if present. called with
1450 * queue lock held and interrupts disabled.
1451 */
1452int blk_remove_plug(request_queue_t *q)
1453{
1454 WARN_ON(!irqs_disabled());
1455
1456 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1457 return 0;
1458
1459 del_timer(&q->unplug_timer);
1460 return 1;
1461}
1462
1463EXPORT_SYMBOL(blk_remove_plug);
1464
1465/*
1466 * remove the plug and let it rip..
1467 */
1468void __generic_unplug_device(request_queue_t *q)
1469{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001470 if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return;
1472
1473 if (!blk_remove_plug(q))
1474 return;
1475
Jens Axboe22e2c502005-06-27 10:55:12 +02001476 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477}
1478EXPORT_SYMBOL(__generic_unplug_device);
1479
1480/**
1481 * generic_unplug_device - fire a request queue
1482 * @q: The &request_queue_t in question
1483 *
1484 * Description:
1485 * Linux uses plugging to build bigger requests queues before letting
1486 * the device have at them. If a queue is plugged, the I/O scheduler
1487 * is still adding and merging requests on the queue. Once the queue
1488 * gets unplugged, the request_fn defined for the queue is invoked and
1489 * transfers started.
1490 **/
1491void generic_unplug_device(request_queue_t *q)
1492{
1493 spin_lock_irq(q->queue_lock);
1494 __generic_unplug_device(q);
1495 spin_unlock_irq(q->queue_lock);
1496}
1497EXPORT_SYMBOL(generic_unplug_device);
1498
1499static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1500 struct page *page)
1501{
1502 request_queue_t *q = bdi->unplug_io_data;
1503
1504 /*
1505 * devices don't necessarily have an ->unplug_fn defined
1506 */
1507 if (q->unplug_fn)
1508 q->unplug_fn(q);
1509}
1510
1511static void blk_unplug_work(void *data)
1512{
1513 request_queue_t *q = data;
1514
1515 q->unplug_fn(q);
1516}
1517
1518static void blk_unplug_timeout(unsigned long data)
1519{
1520 request_queue_t *q = (request_queue_t *)data;
1521
1522 kblockd_schedule_work(&q->unplug_work);
1523}
1524
1525/**
1526 * blk_start_queue - restart a previously stopped queue
1527 * @q: The &request_queue_t in question
1528 *
1529 * Description:
1530 * blk_start_queue() will clear the stop flag on the queue, and call
1531 * the request_fn for the queue if it was in a stopped state when
1532 * entered. Also see blk_stop_queue(). Queue lock must be held.
1533 **/
1534void blk_start_queue(request_queue_t *q)
1535{
1536 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1537
1538 /*
1539 * one level of recursion is ok and is much faster than kicking
1540 * the unplug handling
1541 */
1542 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1543 q->request_fn(q);
1544 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1545 } else {
1546 blk_plug_device(q);
1547 kblockd_schedule_work(&q->unplug_work);
1548 }
1549}
1550
1551EXPORT_SYMBOL(blk_start_queue);
1552
1553/**
1554 * blk_stop_queue - stop a queue
1555 * @q: The &request_queue_t in question
1556 *
1557 * Description:
1558 * The Linux block layer assumes that a block driver will consume all
1559 * entries on the request queue when the request_fn strategy is called.
1560 * Often this will not happen, because of hardware limitations (queue
1561 * depth settings). If a device driver gets a 'queue full' response,
1562 * or if it simply chooses not to queue more I/O at one point, it can
1563 * call this function to prevent the request_fn from being called until
1564 * the driver has signalled it's ready to go again. This happens by calling
1565 * blk_start_queue() to restart queue operations. Queue lock must be held.
1566 **/
1567void blk_stop_queue(request_queue_t *q)
1568{
1569 blk_remove_plug(q);
1570 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1571}
1572EXPORT_SYMBOL(blk_stop_queue);
1573
1574/**
1575 * blk_sync_queue - cancel any pending callbacks on a queue
1576 * @q: the queue
1577 *
1578 * Description:
1579 * The block layer may perform asynchronous callback activity
1580 * on a queue, such as calling the unplug function after a timeout.
1581 * A block device may call blk_sync_queue to ensure that any
1582 * such activity is cancelled, thus allowing it to release resources
1583 * the the callbacks might use. The caller must already have made sure
1584 * that its ->make_request_fn will not re-add plugging prior to calling
1585 * this function.
1586 *
1587 */
1588void blk_sync_queue(struct request_queue *q)
1589{
1590 del_timer_sync(&q->unplug_timer);
1591 kblockd_flush();
1592}
1593EXPORT_SYMBOL(blk_sync_queue);
1594
1595/**
1596 * blk_run_queue - run a single device queue
1597 * @q: The queue to run
1598 */
1599void blk_run_queue(struct request_queue *q)
1600{
1601 unsigned long flags;
1602
1603 spin_lock_irqsave(q->queue_lock, flags);
1604 blk_remove_plug(q);
Ken Chena2997382005-04-16 15:25:43 -07001605 if (!elv_queue_empty(q))
1606 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 spin_unlock_irqrestore(q->queue_lock, flags);
1608}
1609EXPORT_SYMBOL(blk_run_queue);
1610
1611/**
1612 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1613 * @q: the request queue to be released
1614 *
1615 * Description:
1616 * blk_cleanup_queue is the pair to blk_init_queue() or
1617 * blk_queue_make_request(). It should be called when a request queue is
1618 * being released; typically when a block device is being de-registered.
1619 * Currently, its primary task it to free all the &struct request
1620 * structures that were allocated to the queue and the queue itself.
1621 *
1622 * Caveat:
1623 * Hopefully the low level driver will have finished any
1624 * outstanding requests first...
1625 **/
1626void blk_cleanup_queue(request_queue_t * q)
1627{
1628 struct request_list *rl = &q->rq;
1629
1630 if (!atomic_dec_and_test(&q->refcnt))
1631 return;
1632
1633 if (q->elevator)
1634 elevator_exit(q->elevator);
1635
1636 blk_sync_queue(q);
1637
1638 if (rl->rq_pool)
1639 mempool_destroy(rl->rq_pool);
1640
1641 if (q->queue_tags)
1642 __blk_queue_free_tags(q);
1643
1644 blk_queue_ordered(q, QUEUE_ORDERED_NONE);
1645
1646 kmem_cache_free(requestq_cachep, q);
1647}
1648
1649EXPORT_SYMBOL(blk_cleanup_queue);
1650
1651static int blk_init_free_list(request_queue_t *q)
1652{
1653 struct request_list *rl = &q->rq;
1654
1655 rl->count[READ] = rl->count[WRITE] = 0;
1656 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001657 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 init_waitqueue_head(&rl->wait[READ]);
1659 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Christoph Lameter19460892005-06-23 00:08:19 -07001661 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1662 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
1664 if (!rl->rq_pool)
1665 return -ENOMEM;
1666
1667 return 0;
1668}
1669
1670static int __make_request(request_queue_t *, struct bio *);
1671
Al Viro8267e262005-10-21 03:20:53 -04001672request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Christoph Lameter19460892005-06-23 00:08:19 -07001674 return blk_alloc_queue_node(gfp_mask, -1);
1675}
1676EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
Al Viro8267e262005-10-21 03:20:53 -04001678request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001679{
1680 request_queue_t *q;
1681
1682 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 if (!q)
1684 return NULL;
1685
1686 memset(q, 0, sizeof(*q));
1687 init_timer(&q->unplug_timer);
1688 atomic_set(&q->refcnt, 1);
1689
1690 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1691 q->backing_dev_info.unplug_io_data = q;
1692
1693 return q;
1694}
Christoph Lameter19460892005-06-23 00:08:19 -07001695EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
1697/**
1698 * blk_init_queue - prepare a request queue for use with a block device
1699 * @rfn: The function to be called to process requests that have been
1700 * placed on the queue.
1701 * @lock: Request queue spin lock
1702 *
1703 * Description:
1704 * If a block device wishes to use the standard request handling procedures,
1705 * which sorts requests and coalesces adjacent requests, then it must
1706 * call blk_init_queue(). The function @rfn will be called when there
1707 * are requests on the queue that need to be processed. If the device
1708 * supports plugging, then @rfn may not be called immediately when requests
1709 * are available on the queue, but may be called at some time later instead.
1710 * Plugged queues are generally unplugged when a buffer belonging to one
1711 * of the requests on the queue is needed, or due to memory pressure.
1712 *
1713 * @rfn is not required, or even expected, to remove all requests off the
1714 * queue, but only as many as it can handle at a time. If it does leave
1715 * requests on the queue, it is responsible for arranging that the requests
1716 * get dealt with eventually.
1717 *
1718 * The queue spin lock must be held while manipulating the requests on the
1719 * request queue.
1720 *
1721 * Function returns a pointer to the initialized request queue, or NULL if
1722 * it didn't succeed.
1723 *
1724 * Note:
1725 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1726 * when the block device is deactivated (such as at module unload).
1727 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001728
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1730{
Christoph Lameter19460892005-06-23 00:08:19 -07001731 return blk_init_queue_node(rfn, lock, -1);
1732}
1733EXPORT_SYMBOL(blk_init_queue);
1734
1735request_queue_t *
1736blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1737{
1738 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740 if (!q)
1741 return NULL;
1742
Christoph Lameter19460892005-06-23 00:08:19 -07001743 q->node = node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if (blk_init_free_list(q))
1745 goto out_init;
1746
152587d2005-04-12 16:22:06 -05001747 /*
1748 * if caller didn't supply a lock, they get per-queue locking with
1749 * our embedded lock
1750 */
1751 if (!lock) {
1752 spin_lock_init(&q->__queue_lock);
1753 lock = &q->__queue_lock;
1754 }
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 q->request_fn = rfn;
1757 q->back_merge_fn = ll_back_merge_fn;
1758 q->front_merge_fn = ll_front_merge_fn;
1759 q->merge_requests_fn = ll_merge_requests_fn;
1760 q->prep_rq_fn = NULL;
1761 q->unplug_fn = generic_unplug_device;
1762 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1763 q->queue_lock = lock;
1764
1765 blk_queue_segment_boundary(q, 0xffffffff);
1766
1767 blk_queue_make_request(q, __make_request);
1768 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1769
1770 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1771 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1772
1773 /*
1774 * all done
1775 */
1776 if (!elevator_init(q, NULL)) {
1777 blk_queue_congestion_threshold(q);
1778 return q;
1779 }
1780
1781 blk_cleanup_queue(q);
1782out_init:
1783 kmem_cache_free(requestq_cachep, q);
1784 return NULL;
1785}
Christoph Lameter19460892005-06-23 00:08:19 -07001786EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788int blk_get_queue(request_queue_t *q)
1789{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001790 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 atomic_inc(&q->refcnt);
1792 return 0;
1793 }
1794
1795 return 1;
1796}
1797
1798EXPORT_SYMBOL(blk_get_queue);
1799
1800static inline void blk_free_request(request_queue_t *q, struct request *rq)
1801{
Tejun Heocb98fc82005-10-28 08:29:39 +02001802 if (rq->flags & REQ_ELVPRIV)
1803 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 mempool_free(rq, q->rq.rq_pool);
1805}
1806
Jens Axboe22e2c502005-06-27 10:55:12 +02001807static inline struct request *
Tejun Heocb98fc82005-10-28 08:29:39 +02001808blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
Linus Torvalds5dd96242005-10-28 08:56:34 -07001809 int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810{
1811 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1812
1813 if (!rq)
1814 return NULL;
1815
1816 /*
1817 * first three bits are identical in rq->flags and bio->bi_rw,
1818 * see bio.h and blkdev.h
1819 */
1820 rq->flags = rw;
1821
Tejun Heocb98fc82005-10-28 08:29:39 +02001822 if (priv) {
1823 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
1824 mempool_free(rq, q->rq.rq_pool);
1825 return NULL;
1826 }
1827 rq->flags |= REQ_ELVPRIV;
1828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Tejun Heocb98fc82005-10-28 08:29:39 +02001830 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831}
1832
1833/*
1834 * ioc_batching returns true if the ioc is a valid batching request and
1835 * should be given priority access to a request.
1836 */
1837static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1838{
1839 if (!ioc)
1840 return 0;
1841
1842 /*
1843 * Make sure the process is able to allocate at least 1 request
1844 * even if the batch times out, otherwise we could theoretically
1845 * lose wakeups.
1846 */
1847 return ioc->nr_batch_requests == q->nr_batching ||
1848 (ioc->nr_batch_requests > 0
1849 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1850}
1851
1852/*
1853 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1854 * will cause the process to be a "batcher" on all queues in the system. This
1855 * is the behaviour we want though - once it gets a wakeup it should be given
1856 * a nice run.
1857 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07001858static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859{
1860 if (!ioc || ioc_batching(q, ioc))
1861 return;
1862
1863 ioc->nr_batch_requests = q->nr_batching;
1864 ioc->last_waited = jiffies;
1865}
1866
1867static void __freed_request(request_queue_t *q, int rw)
1868{
1869 struct request_list *rl = &q->rq;
1870
1871 if (rl->count[rw] < queue_congestion_off_threshold(q))
1872 clear_queue_congested(q, rw);
1873
1874 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (waitqueue_active(&rl->wait[rw]))
1876 wake_up(&rl->wait[rw]);
1877
1878 blk_clear_queue_full(q, rw);
1879 }
1880}
1881
1882/*
1883 * A request has just been released. Account for it, update the full and
1884 * congestion status, wake up any waiters. Called under q->queue_lock.
1885 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001886static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
1888 struct request_list *rl = &q->rq;
1889
1890 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02001891 if (priv)
1892 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894 __freed_request(q, rw);
1895
1896 if (unlikely(rl->starved[rw ^ 1]))
1897 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898}
1899
1900#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1901/*
Nick Piggind6344532005-06-28 20:45:14 -07001902 * Get a free request, queue_lock must be held.
1903 * Returns NULL on failure, with queue_lock held.
1904 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 */
Jens Axboe22e2c502005-06-27 10:55:12 +02001906static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04001907 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908{
1909 struct request *rq = NULL;
1910 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001911 struct io_context *ioc = NULL;
1912 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001914 may_queue = elv_may_queue(q, rw, bio);
1915 if (may_queue == ELV_MQUEUE_NO)
1916 goto rq_starved;
1917
1918 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
1919 if (rl->count[rw]+1 >= q->nr_requests) {
1920 ioc = current_io_context(GFP_ATOMIC);
1921 /*
1922 * The queue will fill after this allocation, so set
1923 * it as full, and mark this process as "batching".
1924 * This process will be allowed to complete a batch of
1925 * requests, others will be blocked.
1926 */
1927 if (!blk_queue_full(q, rw)) {
1928 ioc_set_batching(q, ioc);
1929 blk_set_queue_full(q, rw);
1930 } else {
1931 if (may_queue != ELV_MQUEUE_MUST
1932 && !ioc_batching(q, ioc)) {
1933 /*
1934 * The queue is full and the allocating
1935 * process is not a "batcher", and not
1936 * exempted by the IO scheduler
1937 */
1938 goto out;
1939 }
1940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 }
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001942 set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 }
1944
Jens Axboe082cf692005-06-28 16:35:11 +02001945 /*
1946 * Only allow batching queuers to allocate up to 50% over the defined
1947 * limit of requests, otherwise we could have thousands of requests
1948 * allocated with any setting of ->nr_requests
1949 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001950 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02001951 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01001952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 rl->count[rw]++;
1954 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001955
Jens Axboe64521d12005-10-28 08:30:39 +02001956 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02001957 if (priv)
1958 rl->elvpriv++;
1959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 spin_unlock_irq(q->queue_lock);
1961
Tejun Heocb98fc82005-10-28 08:29:39 +02001962 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001963 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 /*
1965 * Allocation failed presumably due to memory. Undo anything
1966 * we might have messed up.
1967 *
1968 * Allocating task should really be put onto the front of the
1969 * wait queue, but this is pretty rare.
1970 */
1971 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02001972 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
1974 /*
1975 * in the very unlikely event that allocation failed and no
1976 * requests for this direction was pending, mark us starved
1977 * so that freeing of a request in the other direction will
1978 * notice us. another possible fix would be to split the
1979 * rq mempool into READ and WRITE
1980 */
1981rq_starved:
1982 if (unlikely(rl->count[rw] == 0))
1983 rl->starved[rw] = 1;
1984
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 goto out;
1986 }
1987
Jens Axboe88ee5ef2005-11-12 11:09:12 +01001988 /*
1989 * ioc may be NULL here, and ioc_batching will be false. That's
1990 * OK, if the queue is under the request limit then requests need
1991 * not count toward the nr_batch_requests limit. There will always
1992 * be some limit enforced by BLK_BATCH_TIME.
1993 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 if (ioc_batching(q, ioc))
1995 ioc->nr_batch_requests--;
1996
1997 rq_init(q, rq);
1998 rq->rl = rl;
1999out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 return rq;
2001}
2002
2003/*
2004 * No available requests for this queue, unplug the device and wait for some
2005 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002006 *
2007 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002009static struct request *get_request_wait(request_queue_t *q, int rw,
2010 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 struct request *rq;
2013
Nick Piggin450991b2005-06-28 20:45:13 -07002014 rq = get_request(q, rw, bio, GFP_NOIO);
2015 while (!rq) {
2016 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 struct request_list *rl = &q->rq;
2018
2019 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2020 TASK_UNINTERRUPTIBLE);
2021
Jens Axboe22e2c502005-06-27 10:55:12 +02002022 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
2024 if (!rq) {
2025 struct io_context *ioc;
2026
Nick Piggind6344532005-06-28 20:45:14 -07002027 __generic_unplug_device(q);
2028 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 io_schedule();
2030
2031 /*
2032 * After sleeping, we become a "batching" process and
2033 * will be able to allocate at least one request, and
2034 * up to a big batch of them for a small period time.
2035 * See ioc_batching, ioc_set_batching
2036 */
Nick Pigginfb3cc432005-06-28 20:45:15 -07002037 ioc = current_io_context(GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002039
2040 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 return rq;
2046}
2047
Al Viro8267e262005-10-21 03:20:53 -04002048struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049{
2050 struct request *rq;
2051
2052 BUG_ON(rw != READ && rw != WRITE);
2053
Nick Piggind6344532005-06-28 20:45:14 -07002054 spin_lock_irq(q->queue_lock);
2055 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002056 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002057 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002058 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002059 if (!rq)
2060 spin_unlock_irq(q->queue_lock);
2061 }
2062 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
2064 return rq;
2065}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066EXPORT_SYMBOL(blk_get_request);
2067
2068/**
2069 * blk_requeue_request - put a request back on queue
2070 * @q: request queue where request should be inserted
2071 * @rq: request to be inserted
2072 *
2073 * Description:
2074 * Drivers often keep queueing requests until the hardware cannot accept
2075 * more, when that condition happens we need to put the request back
2076 * on the queue. Must be called with queue lock held.
2077 */
2078void blk_requeue_request(request_queue_t *q, struct request *rq)
2079{
2080 if (blk_rq_tagged(rq))
2081 blk_queue_end_tag(q, rq);
2082
2083 elv_requeue_request(q, rq);
2084}
2085
2086EXPORT_SYMBOL(blk_requeue_request);
2087
2088/**
2089 * blk_insert_request - insert a special request in to a request queue
2090 * @q: request queue where request should be inserted
2091 * @rq: request to be inserted
2092 * @at_head: insert request at head or tail of queue
2093 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 *
2095 * Description:
2096 * Many block devices need to execute commands asynchronously, so they don't
2097 * block the whole kernel from preemption during request execution. This is
2098 * accomplished normally by inserting aritficial requests tagged as
2099 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2100 * scheduled for actual execution by the request queue.
2101 *
2102 * We have the option of inserting the head or the tail of the queue.
2103 * Typically we use the tail for new ioctls and so forth. We use the head
2104 * of the queue for things like a QUEUE_FULL message from a device, or a
2105 * host that is unable to accept a particular command.
2106 */
2107void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002108 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109{
Tejun Heo 867d1192005-04-24 02:06:05 -05002110 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 unsigned long flags;
2112
2113 /*
2114 * tell I/O scheduler that this isn't a regular read/write (ie it
2115 * must not attempt merges on this) and that it acts as a soft
2116 * barrier
2117 */
2118 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2119
2120 rq->special = data;
2121
2122 spin_lock_irqsave(q->queue_lock, flags);
2123
2124 /*
2125 * If command is tagged, release the tag
2126 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002127 if (blk_rq_tagged(rq))
2128 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Tejun Heo 867d1192005-04-24 02:06:05 -05002130 drive_stat_acct(rq, rq->nr_sectors, 1);
2131 __elv_add_request(q, rq, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 if (blk_queue_plugged(q))
2134 __generic_unplug_device(q);
2135 else
2136 q->request_fn(q);
2137 spin_unlock_irqrestore(q->queue_lock, flags);
2138}
2139
2140EXPORT_SYMBOL(blk_insert_request);
2141
2142/**
2143 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2144 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002145 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 * @ubuf: the user buffer
2147 * @len: length of user data
2148 *
2149 * Description:
2150 * Data will be mapped directly for zero copy io, if possible. Otherwise
2151 * a kernel bounce buffer is used.
2152 *
2153 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2154 * still in process context.
2155 *
2156 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2157 * before being submitted to the device, as pages mapped may be out of
2158 * reach. It's the callers responsibility to make sure this happens. The
2159 * original bio must be passed back in to blk_rq_unmap_user() for proper
2160 * unmapping.
2161 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002162int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2163 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164{
2165 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002167 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168
Mike Christiedefd94b2005-12-05 02:37:06 -06002169 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002170 return -EINVAL;
2171 if (!len || !ubuf)
2172 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Jens Axboedd1cab92005-06-20 14:06:01 +02002174 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175
2176 /*
2177 * if alignment requirement is satisfied, map in user pages for
2178 * direct dma. else, set up kernel bounce buffers
2179 */
2180 uaddr = (unsigned long) ubuf;
2181 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002182 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002184 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185
2186 if (!IS_ERR(bio)) {
2187 rq->bio = rq->biotail = bio;
2188 blk_rq_bio_prep(q, rq, bio);
2189
2190 rq->buffer = rq->data = NULL;
2191 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002192 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 }
2194
2195 /*
2196 * bio is the err-ptr
2197 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002198 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199}
2200
2201EXPORT_SYMBOL(blk_rq_map_user);
2202
2203/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002204 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2205 * @q: request queue where request should be inserted
2206 * @rq: request to map data to
2207 * @iov: pointer to the iovec
2208 * @iov_count: number of elements in the iovec
2209 *
2210 * Description:
2211 * Data will be mapped directly for zero copy io, if possible. Otherwise
2212 * a kernel bounce buffer is used.
2213 *
2214 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2215 * still in process context.
2216 *
2217 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2218 * before being submitted to the device, as pages mapped may be out of
2219 * reach. It's the callers responsibility to make sure this happens. The
2220 * original bio must be passed back in to blk_rq_unmap_user() for proper
2221 * unmapping.
2222 */
2223int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2224 struct sg_iovec *iov, int iov_count)
2225{
2226 struct bio *bio;
2227
2228 if (!iov || iov_count <= 0)
2229 return -EINVAL;
2230
2231 /* we don't allow misaligned data like bio_map_user() does. If the
2232 * user is using sg, they're expected to know the alignment constraints
2233 * and respect them accordingly */
2234 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2235 if (IS_ERR(bio))
2236 return PTR_ERR(bio);
2237
2238 rq->bio = rq->biotail = bio;
2239 blk_rq_bio_prep(q, rq, bio);
2240 rq->buffer = rq->data = NULL;
2241 rq->data_len = bio->bi_size;
2242 return 0;
2243}
2244
2245EXPORT_SYMBOL(blk_rq_map_user_iov);
2246
2247/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002249 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 * @ulen: length of user buffer
2251 *
2252 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002253 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002255int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256{
2257 int ret = 0;
2258
2259 if (bio) {
2260 if (bio_flagged(bio, BIO_USER_MAPPED))
2261 bio_unmap_user(bio);
2262 else
2263 ret = bio_uncopy_user(bio);
2264 }
2265
Jens Axboedd1cab92005-06-20 14:06:01 +02002266 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267}
2268
2269EXPORT_SYMBOL(blk_rq_unmap_user);
2270
2271/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002272 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2273 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002274 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002275 * @kbuf: the kernel buffer
2276 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002277 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002278 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002279int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002280 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002281{
Mike Christie df46b9a2005-06-20 14:04:44 +02002282 struct bio *bio;
2283
Mike Christiedefd94b2005-12-05 02:37:06 -06002284 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002285 return -EINVAL;
2286 if (!len || !kbuf)
2287 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002288
2289 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002290 if (IS_ERR(bio))
2291 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002292
Jens Axboedd1cab92005-06-20 14:06:01 +02002293 if (rq_data_dir(rq) == WRITE)
2294 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002295
Jens Axboedd1cab92005-06-20 14:06:01 +02002296 rq->bio = rq->biotail = bio;
2297 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002298
Jens Axboedd1cab92005-06-20 14:06:01 +02002299 rq->buffer = rq->data = NULL;
2300 rq->data_len = len;
2301 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002302}
2303
2304EXPORT_SYMBOL(blk_rq_map_kern);
2305
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002306/**
2307 * blk_execute_rq_nowait - insert a request into queue for execution
2308 * @q: queue to insert the request in
2309 * @bd_disk: matching gendisk
2310 * @rq: request to insert
2311 * @at_head: insert request at head or tail of queue
2312 * @done: I/O completion handler
2313 *
2314 * Description:
2315 * Insert a fully prepared request at the back of the io scheduler queue
2316 * for execution. Don't wait for completion.
2317 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002318void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2319 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002320 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002321{
2322 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2323
2324 rq->rq_disk = bd_disk;
2325 rq->flags |= REQ_NOMERGE;
2326 rq->end_io = done;
2327 elv_add_request(q, rq, where, 1);
2328 generic_unplug_device(q);
2329}
2330
Mike Christie6e39b692005-11-11 05:30:24 -06002331EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333/**
2334 * blk_execute_rq - insert a request into queue for execution
2335 * @q: queue to insert the request in
2336 * @bd_disk: matching gendisk
2337 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002338 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 *
2340 * Description:
2341 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002342 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 */
2344int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002345 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346{
2347 DECLARE_COMPLETION(wait);
2348 char sense[SCSI_SENSE_BUFFERSIZE];
2349 int err = 0;
2350
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 /*
2352 * we need an extra reference to the request, so we can look at
2353 * it after io completion
2354 */
2355 rq->ref_count++;
2356
2357 if (!rq->sense) {
2358 memset(sense, 0, sizeof(sense));
2359 rq->sense = sense;
2360 rq->sense_len = 0;
2361 }
2362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 rq->waiting = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002364 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 wait_for_completion(&wait);
2366 rq->waiting = NULL;
2367
2368 if (rq->errors)
2369 err = -EIO;
2370
2371 return err;
2372}
2373
2374EXPORT_SYMBOL(blk_execute_rq);
2375
2376/**
2377 * blkdev_issue_flush - queue a flush
2378 * @bdev: blockdev to issue flush for
2379 * @error_sector: error sector
2380 *
2381 * Description:
2382 * Issue a flush for the block device in question. Caller can supply
2383 * room for storing the error offset in case of a flush error, if they
2384 * wish to. Caller must run wait_for_completion() on its own.
2385 */
2386int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2387{
2388 request_queue_t *q;
2389
2390 if (bdev->bd_disk == NULL)
2391 return -ENXIO;
2392
2393 q = bdev_get_queue(bdev);
2394 if (!q)
2395 return -ENXIO;
2396 if (!q->issue_flush_fn)
2397 return -EOPNOTSUPP;
2398
2399 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2400}
2401
2402EXPORT_SYMBOL(blkdev_issue_flush);
2403
Adrian Bunk93d17d32005-06-25 14:59:10 -07002404static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
2406 int rw = rq_data_dir(rq);
2407
2408 if (!blk_fs_request(rq) || !rq->rq_disk)
2409 return;
2410
Jens Axboed72d9042005-11-01 08:35:42 +01002411 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002412 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002413 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 disk_round_stats(rq->rq_disk);
2415 rq->rq_disk->in_flight++;
2416 }
2417}
2418
2419/*
2420 * add-request adds a request to the linked list.
2421 * queue lock is held and interrupts disabled, as we muck with the
2422 * request queue list.
2423 */
2424static inline void add_request(request_queue_t * q, struct request * req)
2425{
2426 drive_stat_acct(req, req->nr_sectors, 1);
2427
2428 if (q->activity_fn)
2429 q->activity_fn(q->activity_data, rq_data_dir(req));
2430
2431 /*
2432 * elevator indicated where it wants this request to be
2433 * inserted at elevator_merge time
2434 */
2435 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2436}
2437
2438/*
2439 * disk_round_stats() - Round off the performance stats on a struct
2440 * disk_stats.
2441 *
2442 * The average IO queue length and utilisation statistics are maintained
2443 * by observing the current state of the queue length and the amount of
2444 * time it has been in this state for.
2445 *
2446 * Normally, that accounting is done on IO completion, but that can result
2447 * in more than a second's worth of IO being accounted for within any one
2448 * second, leading to >100% utilisation. To deal with that, we call this
2449 * function to do a round-off before returning the results when reading
2450 * /proc/diskstats. This accounts immediately for all queue usage up to
2451 * the current jiffies and restarts the counters again.
2452 */
2453void disk_round_stats(struct gendisk *disk)
2454{
2455 unsigned long now = jiffies;
2456
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002457 if (now == disk->stamp)
2458 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002460 if (disk->in_flight) {
2461 __disk_stat_add(disk, time_in_queue,
2462 disk->in_flight * (now - disk->stamp));
2463 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466}
2467
2468/*
2469 * queue lock must be held
2470 */
Mike Christie6e39b692005-11-11 05:30:24 -06002471void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472{
2473 struct request_list *rl = req->rl;
2474
2475 if (unlikely(!q))
2476 return;
2477 if (unlikely(--req->ref_count))
2478 return;
2479
Tejun Heo8922e162005-10-20 16:23:44 +02002480 elv_completed_request(q, req);
2481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 req->rq_status = RQ_INACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 req->rl = NULL;
2484
2485 /*
2486 * Request may not have originated from ll_rw_blk. if not,
2487 * it didn't come out of our reserved rq pools
2488 */
2489 if (rl) {
2490 int rw = rq_data_dir(req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002491 int priv = req->flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 BUG_ON(!list_empty(&req->queuelist));
2494
2495 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002496 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 }
2498}
2499
Mike Christie6e39b692005-11-11 05:30:24 -06002500EXPORT_SYMBOL_GPL(__blk_put_request);
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502void blk_put_request(struct request *req)
2503{
Tejun Heo8922e162005-10-20 16:23:44 +02002504 unsigned long flags;
2505 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506
Tejun Heo8922e162005-10-20 16:23:44 +02002507 /*
2508 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2509 * following if (q) test.
2510 */
2511 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 spin_lock_irqsave(q->queue_lock, flags);
2513 __blk_put_request(q, req);
2514 spin_unlock_irqrestore(q->queue_lock, flags);
2515 }
2516}
2517
2518EXPORT_SYMBOL(blk_put_request);
2519
2520/**
2521 * blk_end_sync_rq - executes a completion event on a request
2522 * @rq: request to complete
2523 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002524void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525{
2526 struct completion *waiting = rq->waiting;
2527
2528 rq->waiting = NULL;
2529 __blk_put_request(rq->q, rq);
2530
2531 /*
2532 * complete last, if this is a stack request the process (and thus
2533 * the rq pointer) could be invalid right after this complete()
2534 */
2535 complete(waiting);
2536}
2537EXPORT_SYMBOL(blk_end_sync_rq);
2538
2539/**
2540 * blk_congestion_wait - wait for a queue to become uncongested
2541 * @rw: READ or WRITE
2542 * @timeout: timeout in jiffies
2543 *
2544 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2545 * If no queues are congested then just wait for the next request to be
2546 * returned.
2547 */
2548long blk_congestion_wait(int rw, long timeout)
2549{
2550 long ret;
2551 DEFINE_WAIT(wait);
2552 wait_queue_head_t *wqh = &congestion_wqh[rw];
2553
2554 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2555 ret = io_schedule_timeout(timeout);
2556 finish_wait(wqh, &wait);
2557 return ret;
2558}
2559
2560EXPORT_SYMBOL(blk_congestion_wait);
2561
2562/*
2563 * Has to be called with the request spinlock acquired
2564 */
2565static int attempt_merge(request_queue_t *q, struct request *req,
2566 struct request *next)
2567{
2568 if (!rq_mergeable(req) || !rq_mergeable(next))
2569 return 0;
2570
2571 /*
2572 * not contigious
2573 */
2574 if (req->sector + req->nr_sectors != next->sector)
2575 return 0;
2576
2577 if (rq_data_dir(req) != rq_data_dir(next)
2578 || req->rq_disk != next->rq_disk
2579 || next->waiting || next->special)
2580 return 0;
2581
2582 /*
2583 * If we are allowed to merge, then append bio list
2584 * from next to rq and release next. merge_requests_fn
2585 * will have updated segment counts, update sector
2586 * counts here.
2587 */
2588 if (!q->merge_requests_fn(q, req, next))
2589 return 0;
2590
2591 /*
2592 * At this point we have either done a back merge
2593 * or front merge. We need the smaller start_time of
2594 * the merged requests to be the current request
2595 * for accounting purposes.
2596 */
2597 if (time_after(req->start_time, next->start_time))
2598 req->start_time = next->start_time;
2599
2600 req->biotail->bi_next = next->bio;
2601 req->biotail = next->biotail;
2602
2603 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2604
2605 elv_merge_requests(q, req, next);
2606
2607 if (req->rq_disk) {
2608 disk_round_stats(req->rq_disk);
2609 req->rq_disk->in_flight--;
2610 }
2611
Jens Axboe22e2c502005-06-27 10:55:12 +02002612 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2613
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 __blk_put_request(q, next);
2615 return 1;
2616}
2617
2618static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2619{
2620 struct request *next = elv_latter_request(q, rq);
2621
2622 if (next)
2623 return attempt_merge(q, rq, next);
2624
2625 return 0;
2626}
2627
2628static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2629{
2630 struct request *prev = elv_former_request(q, rq);
2631
2632 if (prev)
2633 return attempt_merge(q, prev, rq);
2634
2635 return 0;
2636}
2637
2638/**
2639 * blk_attempt_remerge - attempt to remerge active head with next request
2640 * @q: The &request_queue_t belonging to the device
2641 * @rq: The head request (usually)
2642 *
2643 * Description:
2644 * For head-active devices, the queue can easily be unplugged so quickly
2645 * that proper merging is not done on the front request. This may hurt
2646 * performance greatly for some devices. The block layer cannot safely
2647 * do merging on that first request for these queues, but the driver can
2648 * call this function and make it happen any way. Only the driver knows
2649 * when it is safe to do so.
2650 **/
2651void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2652{
2653 unsigned long flags;
2654
2655 spin_lock_irqsave(q->queue_lock, flags);
2656 attempt_back_merge(q, rq);
2657 spin_unlock_irqrestore(q->queue_lock, flags);
2658}
2659
2660EXPORT_SYMBOL(blk_attempt_remerge);
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662static int __make_request(request_queue_t *q, struct bio *bio)
2663{
Nick Piggin450991b2005-06-28 20:45:13 -07002664 struct request *req;
Jens Axboe4a534f92005-04-16 15:25:40 -07002665 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
Jens Axboe22e2c502005-06-27 10:55:12 +02002666 unsigned short prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 sector_t sector;
2668
2669 sector = bio->bi_sector;
2670 nr_sectors = bio_sectors(bio);
2671 cur_nr_sectors = bio_cur_sectors(bio);
Jens Axboe22e2c502005-06-27 10:55:12 +02002672 prio = bio_prio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673
2674 rw = bio_data_dir(bio);
Jens Axboe4a534f92005-04-16 15:25:40 -07002675 sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
2677 /*
2678 * low level driver can indicate that it wants pages above a
2679 * certain limit bounced to low memory (ie for highmem, or even
2680 * ISA dma in theory)
2681 */
2682 blk_queue_bounce(q, &bio);
2683
2684 spin_lock_prefetch(q->queue_lock);
2685
2686 barrier = bio_barrier(bio);
Nick Pigginfde6ad22005-06-23 00:08:53 -07002687 if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 err = -EOPNOTSUPP;
2689 goto end_io;
2690 }
2691
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 spin_lock_irq(q->queue_lock);
2693
Nick Piggin450991b2005-06-28 20:45:13 -07002694 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 goto get_rq;
2696
2697 el_ret = elv_merge(q, &req, bio);
2698 switch (el_ret) {
2699 case ELEVATOR_BACK_MERGE:
2700 BUG_ON(!rq_mergeable(req));
2701
2702 if (!q->back_merge_fn(q, req, bio))
2703 break;
2704
2705 req->biotail->bi_next = bio;
2706 req->biotail = bio;
2707 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002708 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 drive_stat_acct(req, nr_sectors, 0);
2710 if (!attempt_back_merge(q, req))
2711 elv_merged_request(q, req);
2712 goto out;
2713
2714 case ELEVATOR_FRONT_MERGE:
2715 BUG_ON(!rq_mergeable(req));
2716
2717 if (!q->front_merge_fn(q, req, bio))
2718 break;
2719
2720 bio->bi_next = req->bio;
2721 req->bio = bio;
2722
2723 /*
2724 * may not be valid. if the low level driver said
2725 * it didn't need a bounce buffer then it better
2726 * not touch req->buffer either...
2727 */
2728 req->buffer = bio_data(bio);
2729 req->current_nr_sectors = cur_nr_sectors;
2730 req->hard_cur_sectors = cur_nr_sectors;
2731 req->sector = req->hard_sector = sector;
2732 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002733 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 drive_stat_acct(req, nr_sectors, 0);
2735 if (!attempt_front_merge(q, req))
2736 elv_merged_request(q, req);
2737 goto out;
2738
Nick Piggin450991b2005-06-28 20:45:13 -07002739 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002741 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
2743
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002745 /*
2746 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002747 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002748 */
Nick Piggin450991b2005-06-28 20:45:13 -07002749 req = get_request_wait(q, rw, bio);
Nick Piggind6344532005-06-28 20:45:14 -07002750
Nick Piggin450991b2005-06-28 20:45:13 -07002751 /*
2752 * After dropping the lock and possibly sleeping here, our request
2753 * may now be mergeable after it had proven unmergeable (above).
2754 * We don't worry about that case for efficiency. It won't happen
2755 * often, and the elevators are able to handle it.
2756 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
2758 req->flags |= REQ_CMD;
2759
2760 /*
2761 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2762 */
2763 if (bio_rw_ahead(bio) || bio_failfast(bio))
2764 req->flags |= REQ_FAILFAST;
2765
2766 /*
2767 * REQ_BARRIER implies no merging, but lets make it explicit
2768 */
Nick Pigginfde6ad22005-06-23 00:08:53 -07002769 if (unlikely(barrier))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2771
2772 req->errors = 0;
2773 req->hard_sector = req->sector = sector;
2774 req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2775 req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2776 req->nr_phys_segments = bio_phys_segments(q, bio);
2777 req->nr_hw_segments = bio_hw_segments(q, bio);
2778 req->buffer = bio_data(bio); /* see ->buffer comment above */
2779 req->waiting = NULL;
2780 req->bio = req->biotail = bio;
Jens Axboe22e2c502005-06-27 10:55:12 +02002781 req->ioprio = prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 req->rq_disk = bio->bi_bdev->bd_disk;
2783 req->start_time = jiffies;
2784
Nick Piggin450991b2005-06-28 20:45:13 -07002785 spin_lock_irq(q->queue_lock);
2786 if (elv_queue_empty(q))
2787 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 add_request(q, req);
2789out:
Jens Axboe4a534f92005-04-16 15:25:40 -07002790 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 __generic_unplug_device(q);
2792
2793 spin_unlock_irq(q->queue_lock);
2794 return 0;
2795
2796end_io:
2797 bio_endio(bio, nr_sectors << 9, err);
2798 return 0;
2799}
2800
2801/*
2802 * If bio->bi_dev is a partition, remap the location
2803 */
2804static inline void blk_partition_remap(struct bio *bio)
2805{
2806 struct block_device *bdev = bio->bi_bdev;
2807
2808 if (bdev != bdev->bd_contains) {
2809 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01002810 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811
Jens Axboea3623572005-11-01 09:26:16 +01002812 p->sectors[rw] += bio_sectors(bio);
2813 p->ios[rw]++;
2814
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 bio->bi_sector += p->start_sect;
2816 bio->bi_bdev = bdev->bd_contains;
2817 }
2818}
2819
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820static void handle_bad_sector(struct bio *bio)
2821{
2822 char b[BDEVNAME_SIZE];
2823
2824 printk(KERN_INFO "attempt to access beyond end of device\n");
2825 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2826 bdevname(bio->bi_bdev, b),
2827 bio->bi_rw,
2828 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2829 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2830
2831 set_bit(BIO_EOF, &bio->bi_flags);
2832}
2833
2834/**
2835 * generic_make_request: hand a buffer to its device driver for I/O
2836 * @bio: The bio describing the location in memory and on the device.
2837 *
2838 * generic_make_request() is used to make I/O requests of block
2839 * devices. It is passed a &struct bio, which describes the I/O that needs
2840 * to be done.
2841 *
2842 * generic_make_request() does not return any status. The
2843 * success/failure status of the request, along with notification of
2844 * completion, is delivered asynchronously through the bio->bi_end_io
2845 * function described (one day) else where.
2846 *
2847 * The caller of generic_make_request must make sure that bi_io_vec
2848 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2849 * set to describe the device address, and the
2850 * bi_end_io and optionally bi_private are set to describe how
2851 * completion notification should be signaled.
2852 *
2853 * generic_make_request and the drivers it calls may use bi_next if this
2854 * bio happens to be merged with someone else, and may change bi_dev and
2855 * bi_sector for remaps as it sees fit. So the values of these fields
2856 * should NOT be depended on after the call to generic_make_request.
2857 */
2858void generic_make_request(struct bio *bio)
2859{
2860 request_queue_t *q;
2861 sector_t maxsector;
2862 int ret, nr_sectors = bio_sectors(bio);
2863
2864 might_sleep();
2865 /* Test device or partition size, when known. */
2866 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2867 if (maxsector) {
2868 sector_t sector = bio->bi_sector;
2869
2870 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2871 /*
2872 * This may well happen - the kernel calls bread()
2873 * without checking the size of the device, e.g., when
2874 * mounting a device.
2875 */
2876 handle_bad_sector(bio);
2877 goto end_io;
2878 }
2879 }
2880
2881 /*
2882 * Resolve the mapping until finished. (drivers are
2883 * still free to implement/resolve their own stacking
2884 * by explicitly returning 0)
2885 *
2886 * NOTE: we don't repeat the blk_size check for each new device.
2887 * Stacking drivers are expected to know what they are doing.
2888 */
2889 do {
2890 char b[BDEVNAME_SIZE];
2891
2892 q = bdev_get_queue(bio->bi_bdev);
2893 if (!q) {
2894 printk(KERN_ERR
2895 "generic_make_request: Trying to access "
2896 "nonexistent block-device %s (%Lu)\n",
2897 bdevname(bio->bi_bdev, b),
2898 (long long) bio->bi_sector);
2899end_io:
2900 bio_endio(bio, bio->bi_size, -EIO);
2901 break;
2902 }
2903
2904 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
2905 printk("bio too big device %s (%u > %u)\n",
2906 bdevname(bio->bi_bdev, b),
2907 bio_sectors(bio),
2908 q->max_hw_sectors);
2909 goto end_io;
2910 }
2911
Nick Pigginfde6ad22005-06-23 00:08:53 -07002912 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 goto end_io;
2914
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 /*
2916 * If this device has partitions, remap block n
2917 * of partition p to block n+start(p) of the disk.
2918 */
2919 blk_partition_remap(bio);
2920
2921 ret = q->make_request_fn(q, bio);
2922 } while (ret);
2923}
2924
2925EXPORT_SYMBOL(generic_make_request);
2926
2927/**
2928 * submit_bio: submit a bio to the block device layer for I/O
2929 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2930 * @bio: The &struct bio which describes the I/O
2931 *
2932 * submit_bio() is very similar in purpose to generic_make_request(), and
2933 * uses that function to do most of the work. Both are fairly rough
2934 * interfaces, @bio must be presetup and ready for I/O.
2935 *
2936 */
2937void submit_bio(int rw, struct bio *bio)
2938{
2939 int count = bio_sectors(bio);
2940
2941 BIO_BUG_ON(!bio->bi_size);
2942 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02002943 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 if (rw & WRITE)
2945 mod_page_state(pgpgout, count);
2946 else
2947 mod_page_state(pgpgin, count);
2948
2949 if (unlikely(block_dump)) {
2950 char b[BDEVNAME_SIZE];
2951 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
2952 current->comm, current->pid,
2953 (rw & WRITE) ? "WRITE" : "READ",
2954 (unsigned long long)bio->bi_sector,
2955 bdevname(bio->bi_bdev,b));
2956 }
2957
2958 generic_make_request(bio);
2959}
2960
2961EXPORT_SYMBOL(submit_bio);
2962
Adrian Bunk93d17d32005-06-25 14:59:10 -07002963static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964{
2965 struct bio *bio, *prevbio = NULL;
2966 int nr_phys_segs, nr_hw_segs;
2967 unsigned int phys_size, hw_size;
2968 request_queue_t *q = rq->q;
2969
2970 if (!rq->bio)
2971 return;
2972
2973 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
2974 rq_for_each_bio(bio, rq) {
2975 /* Force bio hw/phys segs to be recalculated. */
2976 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
2977
2978 nr_phys_segs += bio_phys_segments(q, bio);
2979 nr_hw_segs += bio_hw_segments(q, bio);
2980 if (prevbio) {
2981 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
2982 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
2983
2984 if (blk_phys_contig_segment(q, prevbio, bio) &&
2985 pseg <= q->max_segment_size) {
2986 nr_phys_segs--;
2987 phys_size += prevbio->bi_size + bio->bi_size;
2988 } else
2989 phys_size = 0;
2990
2991 if (blk_hw_contig_segment(q, prevbio, bio) &&
2992 hseg <= q->max_segment_size) {
2993 nr_hw_segs--;
2994 hw_size += prevbio->bi_size + bio->bi_size;
2995 } else
2996 hw_size = 0;
2997 }
2998 prevbio = bio;
2999 }
3000
3001 rq->nr_phys_segments = nr_phys_segs;
3002 rq->nr_hw_segments = nr_hw_segs;
3003}
3004
Adrian Bunk93d17d32005-06-25 14:59:10 -07003005static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
3007 if (blk_fs_request(rq)) {
3008 rq->hard_sector += nsect;
3009 rq->hard_nr_sectors -= nsect;
3010
3011 /*
3012 * Move the I/O submission pointers ahead if required.
3013 */
3014 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3015 (rq->sector <= rq->hard_sector)) {
3016 rq->sector = rq->hard_sector;
3017 rq->nr_sectors = rq->hard_nr_sectors;
3018 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3019 rq->current_nr_sectors = rq->hard_cur_sectors;
3020 rq->buffer = bio_data(rq->bio);
3021 }
3022
3023 /*
3024 * if total number of sectors is less than the first segment
3025 * size, something has gone terribly wrong
3026 */
3027 if (rq->nr_sectors < rq->current_nr_sectors) {
3028 printk("blk: request botched\n");
3029 rq->nr_sectors = rq->current_nr_sectors;
3030 }
3031 }
3032}
3033
3034static int __end_that_request_first(struct request *req, int uptodate,
3035 int nr_bytes)
3036{
3037 int total_bytes, bio_nbytes, error, next_idx = 0;
3038 struct bio *bio;
3039
3040 /*
3041 * extend uptodate bool to allow < 0 value to be direct io error
3042 */
3043 error = 0;
3044 if (end_io_error(uptodate))
3045 error = !uptodate ? -EIO : uptodate;
3046
3047 /*
3048 * for a REQ_BLOCK_PC request, we want to carry any eventual
3049 * sense key with us all the way through
3050 */
3051 if (!blk_pc_request(req))
3052 req->errors = 0;
3053
3054 if (!uptodate) {
3055 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3056 printk("end_request: I/O error, dev %s, sector %llu\n",
3057 req->rq_disk ? req->rq_disk->disk_name : "?",
3058 (unsigned long long)req->sector);
3059 }
3060
Jens Axboed72d9042005-11-01 08:35:42 +01003061 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003062 const int rw = rq_data_dir(req);
3063
3064 __disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003065 }
3066
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 total_bytes = bio_nbytes = 0;
3068 while ((bio = req->bio) != NULL) {
3069 int nbytes;
3070
3071 if (nr_bytes >= bio->bi_size) {
3072 req->bio = bio->bi_next;
3073 nbytes = bio->bi_size;
3074 bio_endio(bio, nbytes, error);
3075 next_idx = 0;
3076 bio_nbytes = 0;
3077 } else {
3078 int idx = bio->bi_idx + next_idx;
3079
3080 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3081 blk_dump_rq_flags(req, "__end_that");
3082 printk("%s: bio idx %d >= vcnt %d\n",
3083 __FUNCTION__,
3084 bio->bi_idx, bio->bi_vcnt);
3085 break;
3086 }
3087
3088 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3089 BIO_BUG_ON(nbytes > bio->bi_size);
3090
3091 /*
3092 * not a complete bvec done
3093 */
3094 if (unlikely(nbytes > nr_bytes)) {
3095 bio_nbytes += nr_bytes;
3096 total_bytes += nr_bytes;
3097 break;
3098 }
3099
3100 /*
3101 * advance to the next vector
3102 */
3103 next_idx++;
3104 bio_nbytes += nbytes;
3105 }
3106
3107 total_bytes += nbytes;
3108 nr_bytes -= nbytes;
3109
3110 if ((bio = req->bio)) {
3111 /*
3112 * end more in this run, or just return 'not-done'
3113 */
3114 if (unlikely(nr_bytes <= 0))
3115 break;
3116 }
3117 }
3118
3119 /*
3120 * completely done
3121 */
3122 if (!req->bio)
3123 return 0;
3124
3125 /*
3126 * if the request wasn't completed, update state
3127 */
3128 if (bio_nbytes) {
3129 bio_endio(bio, bio_nbytes, error);
3130 bio->bi_idx += next_idx;
3131 bio_iovec(bio)->bv_offset += nr_bytes;
3132 bio_iovec(bio)->bv_len -= nr_bytes;
3133 }
3134
3135 blk_recalc_rq_sectors(req, total_bytes >> 9);
3136 blk_recalc_rq_segments(req);
3137 return 1;
3138}
3139
3140/**
3141 * end_that_request_first - end I/O on a request
3142 * @req: the request being processed
3143 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3144 * @nr_sectors: number of sectors to end I/O on
3145 *
3146 * Description:
3147 * Ends I/O on a number of sectors attached to @req, and sets it up
3148 * for the next range of segments (if any) in the cluster.
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_first(struct request *req, int uptodate, int nr_sectors)
3155{
3156 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3157}
3158
3159EXPORT_SYMBOL(end_that_request_first);
3160
3161/**
3162 * end_that_request_chunk - end I/O on a request
3163 * @req: the request being processed
3164 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3165 * @nr_bytes: number of bytes to complete
3166 *
3167 * Description:
3168 * Ends I/O on a number of bytes attached to @req, and sets it up
3169 * for the next range of segments (if any). Like end_that_request_first(),
3170 * but deals with bytes instead of sectors.
3171 *
3172 * Return:
3173 * 0 - we are done with this request, call end_that_request_last()
3174 * 1 - still buffers pending for this request
3175 **/
3176int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3177{
3178 return __end_that_request_first(req, uptodate, nr_bytes);
3179}
3180
3181EXPORT_SYMBOL(end_that_request_chunk);
3182
3183/*
3184 * queue lock must be held
3185 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003186void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187{
3188 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003189 int error;
3190
3191 /*
3192 * extend uptodate bool to allow < 0 value to be direct io error
3193 */
3194 error = 0;
3195 if (end_io_error(uptodate))
3196 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197
3198 if (unlikely(laptop_mode) && blk_fs_request(req))
3199 laptop_io_completion();
3200
3201 if (disk && blk_fs_request(req)) {
3202 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003203 const int rw = rq_data_dir(req);
3204
3205 __disk_stat_inc(disk, ios[rw]);
3206 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 disk_round_stats(disk);
3208 disk->in_flight--;
3209 }
3210 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003211 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 else
3213 __blk_put_request(req->q, req);
3214}
3215
3216EXPORT_SYMBOL(end_that_request_last);
3217
3218void end_request(struct request *req, int uptodate)
3219{
3220 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3221 add_disk_randomness(req->rq_disk);
3222 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003223 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 }
3225}
3226
3227EXPORT_SYMBOL(end_request);
3228
3229void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3230{
3231 /* first three bits are identical in rq->flags and bio->bi_rw */
3232 rq->flags |= (bio->bi_rw & 7);
3233
3234 rq->nr_phys_segments = bio_phys_segments(q, bio);
3235 rq->nr_hw_segments = bio_hw_segments(q, bio);
3236 rq->current_nr_sectors = bio_cur_sectors(bio);
3237 rq->hard_cur_sectors = rq->current_nr_sectors;
3238 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3239 rq->buffer = bio_data(bio);
3240
3241 rq->bio = rq->biotail = bio;
3242}
3243
3244EXPORT_SYMBOL(blk_rq_bio_prep);
3245
3246int kblockd_schedule_work(struct work_struct *work)
3247{
3248 return queue_work(kblockd_workqueue, work);
3249}
3250
3251EXPORT_SYMBOL(kblockd_schedule_work);
3252
3253void kblockd_flush(void)
3254{
3255 flush_workqueue(kblockd_workqueue);
3256}
3257EXPORT_SYMBOL(kblockd_flush);
3258
3259int __init blk_dev_init(void)
3260{
3261 kblockd_workqueue = create_workqueue("kblockd");
3262 if (!kblockd_workqueue)
3263 panic("Failed to create kblockd\n");
3264
3265 request_cachep = kmem_cache_create("blkdev_requests",
3266 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3267
3268 requestq_cachep = kmem_cache_create("blkdev_queue",
3269 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3270
3271 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3272 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3273
3274 blk_max_low_pfn = max_low_pfn;
3275 blk_max_pfn = max_pfn;
3276
3277 return 0;
3278}
3279
3280/*
3281 * IO Context helper functions
3282 */
3283void put_io_context(struct io_context *ioc)
3284{
3285 if (ioc == NULL)
3286 return;
3287
3288 BUG_ON(atomic_read(&ioc->refcount) == 0);
3289
3290 if (atomic_dec_and_test(&ioc->refcount)) {
3291 if (ioc->aic && ioc->aic->dtor)
3292 ioc->aic->dtor(ioc->aic);
3293 if (ioc->cic && ioc->cic->dtor)
3294 ioc->cic->dtor(ioc->cic);
3295
3296 kmem_cache_free(iocontext_cachep, ioc);
3297 }
3298}
3299EXPORT_SYMBOL(put_io_context);
3300
3301/* Called by the exitting task */
3302void exit_io_context(void)
3303{
3304 unsigned long flags;
3305 struct io_context *ioc;
3306
3307 local_irq_save(flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02003308 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 ioc = current->io_context;
3310 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003311 ioc->task = NULL;
3312 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 local_irq_restore(flags);
3314
3315 if (ioc->aic && ioc->aic->exit)
3316 ioc->aic->exit(ioc->aic);
3317 if (ioc->cic && ioc->cic->exit)
3318 ioc->cic->exit(ioc->cic);
3319
3320 put_io_context(ioc);
3321}
3322
3323/*
3324 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003325 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003327 * This returned IO context doesn't have a specifically elevated refcount,
3328 * but since the current task itself holds a reference, the context can be
3329 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 */
Al Viro8267e262005-10-21 03:20:53 -04003331struct io_context *current_io_context(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332{
3333 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 struct io_context *ret;
3335
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003337 if (likely(ret))
3338 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
3340 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3341 if (ret) {
3342 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003343 ret->task = current;
3344 ret->set_ioprio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 ret->last_waited = jiffies; /* doesn't matter... */
3346 ret->nr_batch_requests = 0; /* because this is 0 */
3347 ret->aic = NULL;
3348 ret->cic = NULL;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003349 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 }
3351
3352 return ret;
3353}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003354EXPORT_SYMBOL(current_io_context);
3355
3356/*
3357 * If the current task has no IO context then create one and initialise it.
3358 * If it does have a context, take a ref on it.
3359 *
3360 * This is always called in the context of the task which submitted the I/O.
3361 */
Al Viro8267e262005-10-21 03:20:53 -04003362struct io_context *get_io_context(gfp_t gfp_flags)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003363{
3364 struct io_context *ret;
3365 ret = current_io_context(gfp_flags);
3366 if (likely(ret))
3367 atomic_inc(&ret->refcount);
3368 return ret;
3369}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370EXPORT_SYMBOL(get_io_context);
3371
3372void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3373{
3374 struct io_context *src = *psrc;
3375 struct io_context *dst = *pdst;
3376
3377 if (src) {
3378 BUG_ON(atomic_read(&src->refcount) == 0);
3379 atomic_inc(&src->refcount);
3380 put_io_context(dst);
3381 *pdst = src;
3382 }
3383}
3384EXPORT_SYMBOL(copy_io_context);
3385
3386void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3387{
3388 struct io_context *temp;
3389 temp = *ioc1;
3390 *ioc1 = *ioc2;
3391 *ioc2 = temp;
3392}
3393EXPORT_SYMBOL(swap_io_context);
3394
3395/*
3396 * sysfs parts below
3397 */
3398struct queue_sysfs_entry {
3399 struct attribute attr;
3400 ssize_t (*show)(struct request_queue *, char *);
3401 ssize_t (*store)(struct request_queue *, const char *, size_t);
3402};
3403
3404static ssize_t
3405queue_var_show(unsigned int var, char *page)
3406{
3407 return sprintf(page, "%d\n", var);
3408}
3409
3410static ssize_t
3411queue_var_store(unsigned long *var, const char *page, size_t count)
3412{
3413 char *p = (char *) page;
3414
3415 *var = simple_strtoul(p, &p, 10);
3416 return count;
3417}
3418
3419static ssize_t queue_requests_show(struct request_queue *q, char *page)
3420{
3421 return queue_var_show(q->nr_requests, (page));
3422}
3423
3424static ssize_t
3425queue_requests_store(struct request_queue *q, const char *page, size_t count)
3426{
3427 struct request_list *rl = &q->rq;
3428
3429 int ret = queue_var_store(&q->nr_requests, page, count);
3430 if (q->nr_requests < BLKDEV_MIN_RQ)
3431 q->nr_requests = BLKDEV_MIN_RQ;
3432 blk_queue_congestion_threshold(q);
3433
3434 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3435 set_queue_congested(q, READ);
3436 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3437 clear_queue_congested(q, READ);
3438
3439 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3440 set_queue_congested(q, WRITE);
3441 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3442 clear_queue_congested(q, WRITE);
3443
3444 if (rl->count[READ] >= q->nr_requests) {
3445 blk_set_queue_full(q, READ);
3446 } else if (rl->count[READ]+1 <= q->nr_requests) {
3447 blk_clear_queue_full(q, READ);
3448 wake_up(&rl->wait[READ]);
3449 }
3450
3451 if (rl->count[WRITE] >= q->nr_requests) {
3452 blk_set_queue_full(q, WRITE);
3453 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3454 blk_clear_queue_full(q, WRITE);
3455 wake_up(&rl->wait[WRITE]);
3456 }
3457 return ret;
3458}
3459
3460static ssize_t queue_ra_show(struct request_queue *q, char *page)
3461{
3462 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3463
3464 return queue_var_show(ra_kb, (page));
3465}
3466
3467static ssize_t
3468queue_ra_store(struct request_queue *q, const char *page, size_t count)
3469{
3470 unsigned long ra_kb;
3471 ssize_t ret = queue_var_store(&ra_kb, page, count);
3472
3473 spin_lock_irq(q->queue_lock);
3474 if (ra_kb > (q->max_sectors >> 1))
3475 ra_kb = (q->max_sectors >> 1);
3476
3477 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3478 spin_unlock_irq(q->queue_lock);
3479
3480 return ret;
3481}
3482
3483static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3484{
3485 int max_sectors_kb = q->max_sectors >> 1;
3486
3487 return queue_var_show(max_sectors_kb, (page));
3488}
3489
3490static ssize_t
3491queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3492{
3493 unsigned long max_sectors_kb,
3494 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3495 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3496 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3497 int ra_kb;
3498
3499 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3500 return -EINVAL;
3501 /*
3502 * Take the queue lock to update the readahead and max_sectors
3503 * values synchronously:
3504 */
3505 spin_lock_irq(q->queue_lock);
3506 /*
3507 * Trim readahead window as well, if necessary:
3508 */
3509 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3510 if (ra_kb > max_sectors_kb)
3511 q->backing_dev_info.ra_pages =
3512 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3513
3514 q->max_sectors = max_sectors_kb << 1;
3515 spin_unlock_irq(q->queue_lock);
3516
3517 return ret;
3518}
3519
3520static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3521{
3522 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3523
3524 return queue_var_show(max_hw_sectors_kb, (page));
3525}
3526
3527
3528static struct queue_sysfs_entry queue_requests_entry = {
3529 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3530 .show = queue_requests_show,
3531 .store = queue_requests_store,
3532};
3533
3534static struct queue_sysfs_entry queue_ra_entry = {
3535 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3536 .show = queue_ra_show,
3537 .store = queue_ra_store,
3538};
3539
3540static struct queue_sysfs_entry queue_max_sectors_entry = {
3541 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3542 .show = queue_max_sectors_show,
3543 .store = queue_max_sectors_store,
3544};
3545
3546static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3547 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3548 .show = queue_max_hw_sectors_show,
3549};
3550
3551static struct queue_sysfs_entry queue_iosched_entry = {
3552 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3553 .show = elv_iosched_show,
3554 .store = elv_iosched_store,
3555};
3556
3557static struct attribute *default_attrs[] = {
3558 &queue_requests_entry.attr,
3559 &queue_ra_entry.attr,
3560 &queue_max_hw_sectors_entry.attr,
3561 &queue_max_sectors_entry.attr,
3562 &queue_iosched_entry.attr,
3563 NULL,
3564};
3565
3566#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3567
3568static ssize_t
3569queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3570{
3571 struct queue_sysfs_entry *entry = to_queue(attr);
3572 struct request_queue *q;
3573
3574 q = container_of(kobj, struct request_queue, kobj);
3575 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003576 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003577
3578 return entry->show(q, page);
3579}
3580
3581static ssize_t
3582queue_attr_store(struct kobject *kobj, struct attribute *attr,
3583 const char *page, size_t length)
3584{
3585 struct queue_sysfs_entry *entry = to_queue(attr);
3586 struct request_queue *q;
3587
3588 q = container_of(kobj, struct request_queue, kobj);
3589 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003590 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591
3592 return entry->store(q, page, length);
3593}
3594
3595static struct sysfs_ops queue_sysfs_ops = {
3596 .show = queue_attr_show,
3597 .store = queue_attr_store,
3598};
3599
Adrian Bunk93d17d32005-06-25 14:59:10 -07003600static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 .sysfs_ops = &queue_sysfs_ops,
3602 .default_attrs = default_attrs,
3603};
3604
3605int blk_register_queue(struct gendisk *disk)
3606{
3607 int ret;
3608
3609 request_queue_t *q = disk->queue;
3610
3611 if (!q || !q->request_fn)
3612 return -ENXIO;
3613
3614 q->kobj.parent = kobject_get(&disk->kobj);
3615 if (!q->kobj.parent)
3616 return -EBUSY;
3617
3618 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3619 q->kobj.ktype = &queue_ktype;
3620
3621 ret = kobject_register(&q->kobj);
3622 if (ret < 0)
3623 return ret;
3624
3625 ret = elv_register_queue(q);
3626 if (ret) {
3627 kobject_unregister(&q->kobj);
3628 return ret;
3629 }
3630
3631 return 0;
3632}
3633
3634void blk_unregister_queue(struct gendisk *disk)
3635{
3636 request_queue_t *q = disk->queue;
3637
3638 if (q && q->request_fn) {
3639 elv_unregister_queue(q);
3640
3641 kobject_unregister(&q->kobj);
3642 kobject_put(&disk->kobj);
3643 }
3644}