2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
11 * This handles all read/write requests to block devices
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/bio.h>
17 #include <linux/blkdev.h>
18 #include <linux/highmem.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24 #include <linux/completion.h>
25 #include <linux/slab.h>
26 #include <linux/swap.h>
27 #include <linux/writeback.h>
28 #include <linux/interrupt.h>
29 #include <linux/cpu.h>
30 #include <linux/blktrace_api.h>
31 #include <linux/fault-inject.h>
36 #include <scsi/scsi_cmnd.h>
38 static void blk_unplug_work(struct work_struct
*work
);
39 static void blk_unplug_timeout(unsigned long data
);
40 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
41 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
42 static int __make_request(request_queue_t
*q
, struct bio
*bio
);
43 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
46 * For the allocated request tables
48 static struct kmem_cache
*request_cachep
;
51 * For queue allocation
53 static struct kmem_cache
*requestq_cachep
;
56 * For io context allocations
58 static struct kmem_cache
*iocontext_cachep
;
61 * Controlling structure to kblockd
63 static struct workqueue_struct
*kblockd_workqueue
;
65 unsigned long blk_max_low_pfn
, blk_max_pfn
;
67 EXPORT_SYMBOL(blk_max_low_pfn
);
68 EXPORT_SYMBOL(blk_max_pfn
);
70 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
72 /* Amount of time in which a process may batch requests */
73 #define BLK_BATCH_TIME (HZ/50UL)
75 /* Number of requests a "batching" process may submit */
76 #define BLK_BATCH_REQ 32
79 * Return the threshold (number of used requests) at which the queue is
80 * considered to be congested. It include a little hysteresis to keep the
81 * context switch rate down.
83 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
85 return q
->nr_congestion_on
;
89 * The threshold at which a queue is considered to be uncongested
91 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
93 return q
->nr_congestion_off
;
96 static void blk_queue_congestion_threshold(struct request_queue
*q
)
100 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
101 if (nr
> q
->nr_requests
)
103 q
->nr_congestion_on
= nr
;
105 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
108 q
->nr_congestion_off
= nr
;
112 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
115 * Locates the passed device's request queue and returns the address of its
118 * Will return NULL if the request queue cannot be located.
120 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
122 struct backing_dev_info
*ret
= NULL
;
123 request_queue_t
*q
= bdev_get_queue(bdev
);
126 ret
= &q
->backing_dev_info
;
129 EXPORT_SYMBOL(blk_get_backing_dev_info
);
131 void blk_queue_activity_fn(request_queue_t
*q
, activity_fn
*fn
, void *data
)
134 q
->activity_data
= data
;
136 EXPORT_SYMBOL(blk_queue_activity_fn
);
139 * blk_queue_prep_rq - set a prepare_request function for queue
141 * @pfn: prepare_request function
143 * It's possible for a queue to register a prepare_request callback which
144 * is invoked before the request is handed to the request_fn. The goal of
145 * the function is to prepare a request for I/O, it can be used to build a
146 * cdb from the request data for instance.
149 void blk_queue_prep_rq(request_queue_t
*q
, prep_rq_fn
*pfn
)
154 EXPORT_SYMBOL(blk_queue_prep_rq
);
157 * blk_queue_merge_bvec - set a merge_bvec function for queue
159 * @mbfn: merge_bvec_fn
161 * Usually queues have static limitations on the max sectors or segments that
162 * we can put in a request. Stacking drivers may have some settings that
163 * are dynamic, and thus we have to query the queue whether it is ok to
164 * add a new bio_vec to a bio at a given offset or not. If the block device
165 * has such limitations, it needs to register a merge_bvec_fn to control
166 * the size of bio's sent to it. Note that a block device *must* allow a
167 * single page to be added to an empty bio. The block device driver may want
168 * to use the bio_split() function to deal with these bio's. By default
169 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
172 void blk_queue_merge_bvec(request_queue_t
*q
, merge_bvec_fn
*mbfn
)
174 q
->merge_bvec_fn
= mbfn
;
177 EXPORT_SYMBOL(blk_queue_merge_bvec
);
179 void blk_queue_softirq_done(request_queue_t
*q
, softirq_done_fn
*fn
)
181 q
->softirq_done_fn
= fn
;
184 EXPORT_SYMBOL(blk_queue_softirq_done
);
187 * blk_queue_make_request - define an alternate make_request function for a device
188 * @q: the request queue for the device to be affected
189 * @mfn: the alternate make_request function
192 * The normal way for &struct bios to be passed to a device
193 * driver is for them to be collected into requests on a request
194 * queue, and then to allow the device driver to select requests
195 * off that queue when it is ready. This works well for many block
196 * devices. However some block devices (typically virtual devices
197 * such as md or lvm) do not benefit from the processing on the
198 * request queue, and are served best by having the requests passed
199 * directly to them. This can be achieved by providing a function
200 * to blk_queue_make_request().
203 * The driver that does this *must* be able to deal appropriately
204 * with buffers in "highmemory". This can be accomplished by either calling
205 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
206 * blk_queue_bounce() to create a buffer in normal memory.
208 void blk_queue_make_request(request_queue_t
* q
, make_request_fn
* mfn
)
213 q
->nr_requests
= BLKDEV_MAX_RQ
;
214 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
215 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
216 q
->make_request_fn
= mfn
;
217 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
218 q
->backing_dev_info
.state
= 0;
219 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
220 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
221 blk_queue_hardsect_size(q
, 512);
222 blk_queue_dma_alignment(q
, 511);
223 blk_queue_congestion_threshold(q
);
224 q
->nr_batching
= BLK_BATCH_REQ
;
226 q
->unplug_thresh
= 4; /* hmm */
227 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
228 if (q
->unplug_delay
== 0)
231 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
233 q
->unplug_timer
.function
= blk_unplug_timeout
;
234 q
->unplug_timer
.data
= (unsigned long)q
;
237 * by default assume old behaviour and bounce for any highmem page
239 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
241 blk_queue_activity_fn(q
, NULL
, NULL
);
244 EXPORT_SYMBOL(blk_queue_make_request
);
246 static void rq_init(request_queue_t
*q
, struct request
*rq
)
248 INIT_LIST_HEAD(&rq
->queuelist
);
249 INIT_LIST_HEAD(&rq
->donelist
);
252 rq
->bio
= rq
->biotail
= NULL
;
253 INIT_HLIST_NODE(&rq
->hash
);
254 RB_CLEAR_NODE(&rq
->rb_node
);
262 rq
->nr_phys_segments
= 0;
265 rq
->end_io_data
= NULL
;
266 rq
->completion_data
= NULL
;
270 * blk_queue_ordered - does this queue support ordered writes
271 * @q: the request queue
272 * @ordered: one of QUEUE_ORDERED_*
273 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
276 * For journalled file systems, doing ordered writes on a commit
277 * block instead of explicitly doing wait_on_buffer (which is bad
278 * for performance) can be a big win. Block drivers supporting this
279 * feature should call this function and indicate so.
282 int blk_queue_ordered(request_queue_t
*q
, unsigned ordered
,
283 prepare_flush_fn
*prepare_flush_fn
)
285 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
286 prepare_flush_fn
== NULL
) {
287 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
291 if (ordered
!= QUEUE_ORDERED_NONE
&&
292 ordered
!= QUEUE_ORDERED_DRAIN
&&
293 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
294 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
295 ordered
!= QUEUE_ORDERED_TAG
&&
296 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
297 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
298 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
302 q
->ordered
= ordered
;
303 q
->next_ordered
= ordered
;
304 q
->prepare_flush_fn
= prepare_flush_fn
;
309 EXPORT_SYMBOL(blk_queue_ordered
);
312 * blk_queue_issue_flush_fn - set function for issuing a flush
313 * @q: the request queue
314 * @iff: the function to be called issuing the flush
317 * If a driver supports issuing a flush command, the support is notified
318 * to the block layer by defining it through this call.
321 void blk_queue_issue_flush_fn(request_queue_t
*q
, issue_flush_fn
*iff
)
323 q
->issue_flush_fn
= iff
;
326 EXPORT_SYMBOL(blk_queue_issue_flush_fn
);
329 * Cache flushing for ordered writes handling
331 inline unsigned blk_ordered_cur_seq(request_queue_t
*q
)
335 return 1 << ffz(q
->ordseq
);
338 unsigned blk_ordered_req_seq(struct request
*rq
)
340 request_queue_t
*q
= rq
->q
;
342 BUG_ON(q
->ordseq
== 0);
344 if (rq
== &q
->pre_flush_rq
)
345 return QUEUE_ORDSEQ_PREFLUSH
;
346 if (rq
== &q
->bar_rq
)
347 return QUEUE_ORDSEQ_BAR
;
348 if (rq
== &q
->post_flush_rq
)
349 return QUEUE_ORDSEQ_POSTFLUSH
;
351 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
352 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
353 return QUEUE_ORDSEQ_DRAIN
;
355 return QUEUE_ORDSEQ_DONE
;
358 void blk_ordered_complete_seq(request_queue_t
*q
, unsigned seq
, int error
)
363 if (error
&& !q
->orderr
)
366 BUG_ON(q
->ordseq
& seq
);
369 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
373 * Okay, sequence complete.
376 uptodate
= q
->orderr
? q
->orderr
: 1;
380 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
381 end_that_request_last(rq
, uptodate
);
384 static void pre_flush_end_io(struct request
*rq
, int error
)
386 elv_completed_request(rq
->q
, rq
);
387 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
390 static void bar_end_io(struct request
*rq
, int error
)
392 elv_completed_request(rq
->q
, rq
);
393 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
396 static void post_flush_end_io(struct request
*rq
, int error
)
398 elv_completed_request(rq
->q
, rq
);
399 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
402 static void queue_flush(request_queue_t
*q
, unsigned which
)
405 rq_end_io_fn
*end_io
;
407 if (which
== QUEUE_ORDERED_PREFLUSH
) {
408 rq
= &q
->pre_flush_rq
;
409 end_io
= pre_flush_end_io
;
411 rq
= &q
->post_flush_rq
;
412 end_io
= post_flush_end_io
;
415 rq
->cmd_flags
= REQ_HARDBARRIER
;
417 rq
->elevator_private
= NULL
;
418 rq
->elevator_private2
= NULL
;
419 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
421 q
->prepare_flush_fn(q
, rq
);
423 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
426 static inline struct request
*start_ordered(request_queue_t
*q
,
431 q
->ordered
= q
->next_ordered
;
432 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
435 * Prep proxy barrier request.
437 blkdev_dequeue_request(rq
);
442 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
443 rq
->cmd_flags
|= REQ_RW
;
444 rq
->cmd_flags
|= q
->ordered
& QUEUE_ORDERED_FUA
? REQ_FUA
: 0;
445 rq
->elevator_private
= NULL
;
446 rq
->elevator_private2
= NULL
;
447 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
448 rq
->end_io
= bar_end_io
;
451 * Queue ordered sequence. As we stack them at the head, we
452 * need to queue in reverse order. Note that we rely on that
453 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
454 * request gets inbetween ordered sequence.
456 if (q
->ordered
& QUEUE_ORDERED_POSTFLUSH
)
457 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
459 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
461 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
463 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
464 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
465 rq
= &q
->pre_flush_rq
;
467 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
469 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
470 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
477 int blk_do_ordered(request_queue_t
*q
, struct request
**rqp
)
479 struct request
*rq
= *rqp
;
480 int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
486 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
487 *rqp
= start_ordered(q
, rq
);
491 * This can happen when the queue switches to
492 * ORDERED_NONE while this request is on it.
494 blkdev_dequeue_request(rq
);
495 end_that_request_first(rq
, -EOPNOTSUPP
,
496 rq
->hard_nr_sectors
);
497 end_that_request_last(rq
, -EOPNOTSUPP
);
504 * Ordered sequence in progress
507 /* Special requests are not subject to ordering rules. */
508 if (!blk_fs_request(rq
) &&
509 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
512 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
513 /* Ordered by tag. Blocking the next barrier is enough. */
514 if (is_barrier
&& rq
!= &q
->bar_rq
)
517 /* Ordered by draining. Wait for turn. */
518 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
519 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
526 static int flush_dry_bio_endio(struct bio
*bio
, unsigned int bytes
, int error
)
528 request_queue_t
*q
= bio
->bi_private
;
529 struct bio_vec
*bvec
;
533 * This is dry run, restore bio_sector and size. We'll finish
534 * this request again with the original bi_end_io after an
535 * error occurs or post flush is complete.
544 bio_for_each_segment(bvec
, bio
, i
) {
545 bvec
->bv_len
+= bvec
->bv_offset
;
550 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
551 bio
->bi_size
= q
->bi_size
;
552 bio
->bi_sector
-= (q
->bi_size
>> 9);
558 static int ordered_bio_endio(struct request
*rq
, struct bio
*bio
,
559 unsigned int nbytes
, int error
)
561 request_queue_t
*q
= rq
->q
;
565 if (&q
->bar_rq
!= rq
)
569 * Okay, this is the barrier request in progress, dry finish it.
571 if (error
&& !q
->orderr
)
574 endio
= bio
->bi_end_io
;
575 private = bio
->bi_private
;
576 bio
->bi_end_io
= flush_dry_bio_endio
;
579 bio_endio(bio
, nbytes
, error
);
581 bio
->bi_end_io
= endio
;
582 bio
->bi_private
= private;
588 * blk_queue_bounce_limit - set bounce buffer limit for queue
589 * @q: the request queue for the device
590 * @dma_addr: bus address limit
593 * Different hardware can have different requirements as to what pages
594 * it can do I/O directly to. A low level driver can call
595 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
596 * buffers for doing I/O to pages residing above @page.
598 void blk_queue_bounce_limit(request_queue_t
*q
, u64 dma_addr
)
600 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
603 q
->bounce_gfp
= GFP_NOIO
;
604 #if BITS_PER_LONG == 64
605 /* Assume anything <= 4GB can be handled by IOMMU.
606 Actually some IOMMUs can handle everything, but I don't
607 know of a way to test this here. */
608 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
610 q
->bounce_pfn
= max_low_pfn
;
612 if (bounce_pfn
< blk_max_low_pfn
)
614 q
->bounce_pfn
= bounce_pfn
;
617 init_emergency_isa_pool();
618 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
619 q
->bounce_pfn
= bounce_pfn
;
623 EXPORT_SYMBOL(blk_queue_bounce_limit
);
626 * blk_queue_max_sectors - set max sectors for a request for this queue
627 * @q: the request queue for the device
628 * @max_sectors: max sectors in the usual 512b unit
631 * Enables a low level driver to set an upper limit on the size of
634 void blk_queue_max_sectors(request_queue_t
*q
, unsigned int max_sectors
)
636 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
637 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
638 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
641 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
642 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
644 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
645 q
->max_hw_sectors
= max_sectors
;
649 EXPORT_SYMBOL(blk_queue_max_sectors
);
652 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
653 * @q: the request queue for the device
654 * @max_segments: max number of segments
657 * Enables a low level driver to set an upper limit on the number of
658 * physical data segments in a request. This would be the largest sized
659 * scatter list the driver could handle.
661 void blk_queue_max_phys_segments(request_queue_t
*q
, unsigned short max_segments
)
665 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
668 q
->max_phys_segments
= max_segments
;
671 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
674 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
675 * @q: the request queue for the device
676 * @max_segments: max number of segments
679 * Enables a low level driver to set an upper limit on the number of
680 * hw data segments in a request. This would be the largest number of
681 * address/length pairs the host adapter can actually give as once
684 void blk_queue_max_hw_segments(request_queue_t
*q
, unsigned short max_segments
)
688 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
691 q
->max_hw_segments
= max_segments
;
694 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
697 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
698 * @q: the request queue for the device
699 * @max_size: max size of segment in bytes
702 * Enables a low level driver to set an upper limit on the size of a
705 void blk_queue_max_segment_size(request_queue_t
*q
, unsigned int max_size
)
707 if (max_size
< PAGE_CACHE_SIZE
) {
708 max_size
= PAGE_CACHE_SIZE
;
709 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
712 q
->max_segment_size
= max_size
;
715 EXPORT_SYMBOL(blk_queue_max_segment_size
);
718 * blk_queue_hardsect_size - set hardware sector size for the queue
719 * @q: the request queue for the device
720 * @size: the hardware sector size, in bytes
723 * This should typically be set to the lowest possible sector size
724 * that the hardware can operate on (possible without reverting to
725 * even internal read-modify-write operations). Usually the default
726 * of 512 covers most hardware.
728 void blk_queue_hardsect_size(request_queue_t
*q
, unsigned short size
)
730 q
->hardsect_size
= size
;
733 EXPORT_SYMBOL(blk_queue_hardsect_size
);
736 * Returns the minimum that is _not_ zero, unless both are zero.
738 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
741 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
742 * @t: the stacking driver (top)
743 * @b: the underlying device (bottom)
745 void blk_queue_stack_limits(request_queue_t
*t
, request_queue_t
*b
)
747 /* zero is "infinity" */
748 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
749 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
751 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
752 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
753 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
754 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
755 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
756 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
759 EXPORT_SYMBOL(blk_queue_stack_limits
);
762 * blk_queue_segment_boundary - set boundary rules for segment merging
763 * @q: the request queue for the device
764 * @mask: the memory boundary mask
766 void blk_queue_segment_boundary(request_queue_t
*q
, unsigned long mask
)
768 if (mask
< PAGE_CACHE_SIZE
- 1) {
769 mask
= PAGE_CACHE_SIZE
- 1;
770 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
773 q
->seg_boundary_mask
= mask
;
776 EXPORT_SYMBOL(blk_queue_segment_boundary
);
779 * blk_queue_dma_alignment - set dma length and memory alignment
780 * @q: the request queue for the device
781 * @mask: alignment mask
784 * set required memory and length aligment for direct dma transactions.
785 * this is used when buiding direct io requests for the queue.
788 void blk_queue_dma_alignment(request_queue_t
*q
, int mask
)
790 q
->dma_alignment
= mask
;
793 EXPORT_SYMBOL(blk_queue_dma_alignment
);
796 * blk_queue_find_tag - find a request by its tag and queue
797 * @q: The request queue for the device
798 * @tag: The tag of the request
801 * Should be used when a device returns a tag and you want to match
804 * no locks need be held.
806 struct request
*blk_queue_find_tag(request_queue_t
*q
, int tag
)
808 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
811 EXPORT_SYMBOL(blk_queue_find_tag
);
814 * __blk_free_tags - release a given set of tag maintenance info
815 * @bqt: the tag map to free
817 * Tries to free the specified @bqt@. Returns true if it was
818 * actually freed and false if there are still references using it
820 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
824 retval
= atomic_dec_and_test(&bqt
->refcnt
);
827 BUG_ON(!list_empty(&bqt
->busy_list
));
829 kfree(bqt
->tag_index
);
830 bqt
->tag_index
= NULL
;
843 * __blk_queue_free_tags - release tag maintenance info
844 * @q: the request queue for the device
847 * blk_cleanup_queue() will take care of calling this function, if tagging
848 * has been used. So there's no need to call this directly.
850 static void __blk_queue_free_tags(request_queue_t
*q
)
852 struct blk_queue_tag
*bqt
= q
->queue_tags
;
857 __blk_free_tags(bqt
);
859 q
->queue_tags
= NULL
;
860 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
865 * blk_free_tags - release a given set of tag maintenance info
866 * @bqt: the tag map to free
868 * For externally managed @bqt@ frees the map. Callers of this
869 * function must guarantee to have released all the queues that
870 * might have been using this tag map.
872 void blk_free_tags(struct blk_queue_tag
*bqt
)
874 if (unlikely(!__blk_free_tags(bqt
)))
877 EXPORT_SYMBOL(blk_free_tags
);
880 * blk_queue_free_tags - release tag maintenance info
881 * @q: the request queue for the device
884 * This is used to disabled tagged queuing to a device, yet leave
887 void blk_queue_free_tags(request_queue_t
*q
)
889 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
892 EXPORT_SYMBOL(blk_queue_free_tags
);
895 init_tag_map(request_queue_t
*q
, struct blk_queue_tag
*tags
, int depth
)
897 struct request
**tag_index
;
898 unsigned long *tag_map
;
901 if (q
&& depth
> q
->nr_requests
* 2) {
902 depth
= q
->nr_requests
* 2;
903 printk(KERN_ERR
"%s: adjusted depth to %d\n",
904 __FUNCTION__
, depth
);
907 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
911 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
912 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
916 tags
->real_max_depth
= depth
;
917 tags
->max_depth
= depth
;
918 tags
->tag_index
= tag_index
;
919 tags
->tag_map
= tag_map
;
927 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
930 struct blk_queue_tag
*tags
;
932 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
936 if (init_tag_map(q
, tags
, depth
))
939 INIT_LIST_HEAD(&tags
->busy_list
);
941 atomic_set(&tags
->refcnt
, 1);
949 * blk_init_tags - initialize the tag info for an external tag map
950 * @depth: the maximum queue depth supported
951 * @tags: the tag to use
953 struct blk_queue_tag
*blk_init_tags(int depth
)
955 return __blk_queue_init_tags(NULL
, depth
);
957 EXPORT_SYMBOL(blk_init_tags
);
960 * blk_queue_init_tags - initialize the queue tag info
961 * @q: the request queue for the device
962 * @depth: the maximum queue depth supported
963 * @tags: the tag to use
965 int blk_queue_init_tags(request_queue_t
*q
, int depth
,
966 struct blk_queue_tag
*tags
)
970 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
972 if (!tags
&& !q
->queue_tags
) {
973 tags
= __blk_queue_init_tags(q
, depth
);
977 } else if (q
->queue_tags
) {
978 if ((rc
= blk_queue_resize_tags(q
, depth
)))
980 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
983 atomic_inc(&tags
->refcnt
);
986 * assign it, all done
988 q
->queue_tags
= tags
;
989 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
996 EXPORT_SYMBOL(blk_queue_init_tags
);
999 * blk_queue_resize_tags - change the queueing depth
1000 * @q: the request queue for the device
1001 * @new_depth: the new max command queueing depth
1004 * Must be called with the queue lock held.
1006 int blk_queue_resize_tags(request_queue_t
*q
, int new_depth
)
1008 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1009 struct request
**tag_index
;
1010 unsigned long *tag_map
;
1011 int max_depth
, nr_ulongs
;
1017 * if we already have large enough real_max_depth. just
1018 * adjust max_depth. *NOTE* as requests with tag value
1019 * between new_depth and real_max_depth can be in-flight, tag
1020 * map can not be shrunk blindly here.
1022 if (new_depth
<= bqt
->real_max_depth
) {
1023 bqt
->max_depth
= new_depth
;
1028 * Currently cannot replace a shared tag map with a new
1029 * one, so error out if this is the case
1031 if (atomic_read(&bqt
->refcnt
) != 1)
1035 * save the old state info, so we can copy it back
1037 tag_index
= bqt
->tag_index
;
1038 tag_map
= bqt
->tag_map
;
1039 max_depth
= bqt
->real_max_depth
;
1041 if (init_tag_map(q
, bqt
, new_depth
))
1044 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1045 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1046 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1053 EXPORT_SYMBOL(blk_queue_resize_tags
);
1056 * blk_queue_end_tag - end tag operations for a request
1057 * @q: the request queue for the device
1058 * @rq: the request that has completed
1061 * Typically called when end_that_request_first() returns 0, meaning
1062 * all transfers have been done for a request. It's important to call
1063 * this function before end_that_request_last(), as that will put the
1064 * request back on the free list thus corrupting the internal tag list.
1067 * queue lock must be held.
1069 void blk_queue_end_tag(request_queue_t
*q
, struct request
*rq
)
1071 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1076 if (unlikely(tag
>= bqt
->real_max_depth
))
1078 * This can happen after tag depth has been reduced.
1079 * FIXME: how about a warning or info message here?
1083 if (unlikely(!__test_and_clear_bit(tag
, bqt
->tag_map
))) {
1084 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1089 list_del_init(&rq
->queuelist
);
1090 rq
->cmd_flags
&= ~REQ_QUEUED
;
1093 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1094 printk(KERN_ERR
"%s: tag %d is missing\n",
1097 bqt
->tag_index
[tag
] = NULL
;
1101 EXPORT_SYMBOL(blk_queue_end_tag
);
1104 * blk_queue_start_tag - find a free tag and assign it
1105 * @q: the request queue for the device
1106 * @rq: the block request that needs tagging
1109 * This can either be used as a stand-alone helper, or possibly be
1110 * assigned as the queue &prep_rq_fn (in which case &struct request
1111 * automagically gets a tag assigned). Note that this function
1112 * assumes that any type of request can be queued! if this is not
1113 * true for your device, you must check the request type before
1114 * calling this function. The request will also be removed from
1115 * the request queue, so it's the drivers responsibility to readd
1116 * it if it should need to be restarted for some reason.
1119 * queue lock must be held.
1121 int blk_queue_start_tag(request_queue_t
*q
, struct request
*rq
)
1123 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1126 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1128 "%s: request %p for device [%s] already tagged %d",
1130 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1135 * Protect against shared tag maps, as we may not have exclusive
1136 * access to the tag map.
1139 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1140 if (tag
>= bqt
->max_depth
)
1143 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1145 rq
->cmd_flags
|= REQ_QUEUED
;
1147 bqt
->tag_index
[tag
] = rq
;
1148 blkdev_dequeue_request(rq
);
1149 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1154 EXPORT_SYMBOL(blk_queue_start_tag
);
1157 * blk_queue_invalidate_tags - invalidate all pending tags
1158 * @q: the request queue for the device
1161 * Hardware conditions may dictate a need to stop all pending requests.
1162 * In this case, we will safely clear the block side of the tag queue and
1163 * readd all requests to the request queue in the right order.
1166 * queue lock must be held.
1168 void blk_queue_invalidate_tags(request_queue_t
*q
)
1170 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1171 struct list_head
*tmp
, *n
;
1174 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1175 rq
= list_entry_rq(tmp
);
1177 if (rq
->tag
== -1) {
1179 "%s: bad tag found on list\n", __FUNCTION__
);
1180 list_del_init(&rq
->queuelist
);
1181 rq
->cmd_flags
&= ~REQ_QUEUED
;
1183 blk_queue_end_tag(q
, rq
);
1185 rq
->cmd_flags
&= ~REQ_STARTED
;
1186 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1190 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1192 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1196 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1197 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1200 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1202 rq
->current_nr_sectors
);
1203 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1205 if (blk_pc_request(rq
)) {
1207 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1208 printk("%02x ", rq
->cmd
[bit
]);
1213 EXPORT_SYMBOL(blk_dump_rq_flags
);
1215 void blk_recount_segments(request_queue_t
*q
, struct bio
*bio
)
1217 struct bio_vec
*bv
, *bvprv
= NULL
;
1218 int i
, nr_phys_segs
, nr_hw_segs
, seg_size
, hw_seg_size
, cluster
;
1219 int high
, highprv
= 1;
1221 if (unlikely(!bio
->bi_io_vec
))
1224 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1225 hw_seg_size
= seg_size
= nr_phys_segs
= nr_hw_segs
= 0;
1226 bio_for_each_segment(bv
, bio
, i
) {
1228 * the trick here is making sure that a high page is never
1229 * considered part of another segment, since that might
1230 * change with the bounce page.
1232 high
= page_to_pfn(bv
->bv_page
) >= q
->bounce_pfn
;
1233 if (high
|| highprv
)
1234 goto new_hw_segment
;
1236 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1238 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1240 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1242 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1243 goto new_hw_segment
;
1245 seg_size
+= bv
->bv_len
;
1246 hw_seg_size
+= bv
->bv_len
;
1251 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1252 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
)) {
1253 hw_seg_size
+= bv
->bv_len
;
1256 if (hw_seg_size
> bio
->bi_hw_front_size
)
1257 bio
->bi_hw_front_size
= hw_seg_size
;
1258 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1264 seg_size
= bv
->bv_len
;
1267 if (hw_seg_size
> bio
->bi_hw_back_size
)
1268 bio
->bi_hw_back_size
= hw_seg_size
;
1269 if (nr_hw_segs
== 1 && hw_seg_size
> bio
->bi_hw_front_size
)
1270 bio
->bi_hw_front_size
= hw_seg_size
;
1271 bio
->bi_phys_segments
= nr_phys_segs
;
1272 bio
->bi_hw_segments
= nr_hw_segs
;
1273 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1277 static int blk_phys_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1280 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1283 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1285 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1289 * bio and nxt are contigous in memory, check if the queue allows
1290 * these two to be merged into one
1292 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1298 static int blk_hw_contig_segment(request_queue_t
*q
, struct bio
*bio
,
1301 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1302 blk_recount_segments(q
, bio
);
1303 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1304 blk_recount_segments(q
, nxt
);
1305 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1306 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_front_size
+ bio
->bi_hw_back_size
))
1308 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1315 * map a request to scatterlist, return number of sg entries setup. Caller
1316 * must make sure sg can hold rq->nr_phys_segments entries
1318 int blk_rq_map_sg(request_queue_t
*q
, struct request
*rq
, struct scatterlist
*sg
)
1320 struct bio_vec
*bvec
, *bvprv
;
1322 int nsegs
, i
, cluster
;
1325 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1328 * for each bio in rq
1331 rq_for_each_bio(bio
, rq
) {
1333 * for each segment in bio
1335 bio_for_each_segment(bvec
, bio
, i
) {
1336 int nbytes
= bvec
->bv_len
;
1338 if (bvprv
&& cluster
) {
1339 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1342 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1344 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1347 sg
[nsegs
- 1].length
+= nbytes
;
1350 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1351 sg
[nsegs
].page
= bvec
->bv_page
;
1352 sg
[nsegs
].length
= nbytes
;
1353 sg
[nsegs
].offset
= bvec
->bv_offset
;
1358 } /* segments in bio */
1364 EXPORT_SYMBOL(blk_rq_map_sg
);
1367 * the standard queue merge functions, can be overridden with device
1368 * specific ones if so desired
1371 static inline int ll_new_mergeable(request_queue_t
*q
,
1372 struct request
*req
,
1375 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1377 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1378 req
->cmd_flags
|= REQ_NOMERGE
;
1379 if (req
== q
->last_merge
)
1380 q
->last_merge
= NULL
;
1385 * A hw segment is just getting larger, bump just the phys
1388 req
->nr_phys_segments
+= nr_phys_segs
;
1392 static inline int ll_new_hw_segment(request_queue_t
*q
,
1393 struct request
*req
,
1396 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1397 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1399 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1400 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1401 req
->cmd_flags
|= REQ_NOMERGE
;
1402 if (req
== q
->last_merge
)
1403 q
->last_merge
= NULL
;
1408 * This will form the start of a new hw segment. Bump both
1411 req
->nr_hw_segments
+= nr_hw_segs
;
1412 req
->nr_phys_segments
+= nr_phys_segs
;
1416 static int ll_back_merge_fn(request_queue_t
*q
, struct request
*req
,
1419 unsigned short max_sectors
;
1422 if (unlikely(blk_pc_request(req
)))
1423 max_sectors
= q
->max_hw_sectors
;
1425 max_sectors
= q
->max_sectors
;
1427 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1428 req
->cmd_flags
|= REQ_NOMERGE
;
1429 if (req
== q
->last_merge
)
1430 q
->last_merge
= NULL
;
1433 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1434 blk_recount_segments(q
, req
->biotail
);
1435 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1436 blk_recount_segments(q
, bio
);
1437 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1438 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1439 !BIOVEC_VIRT_OVERSIZE(len
)) {
1440 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1443 if (req
->nr_hw_segments
== 1)
1444 req
->bio
->bi_hw_front_size
= len
;
1445 if (bio
->bi_hw_segments
== 1)
1446 bio
->bi_hw_back_size
= len
;
1451 return ll_new_hw_segment(q
, req
, bio
);
1454 static int ll_front_merge_fn(request_queue_t
*q
, struct request
*req
,
1457 unsigned short max_sectors
;
1460 if (unlikely(blk_pc_request(req
)))
1461 max_sectors
= q
->max_hw_sectors
;
1463 max_sectors
= q
->max_sectors
;
1466 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1467 req
->cmd_flags
|= REQ_NOMERGE
;
1468 if (req
== q
->last_merge
)
1469 q
->last_merge
= NULL
;
1472 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1473 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1474 blk_recount_segments(q
, bio
);
1475 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1476 blk_recount_segments(q
, req
->bio
);
1477 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1478 !BIOVEC_VIRT_OVERSIZE(len
)) {
1479 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1482 if (bio
->bi_hw_segments
== 1)
1483 bio
->bi_hw_front_size
= len
;
1484 if (req
->nr_hw_segments
== 1)
1485 req
->biotail
->bi_hw_back_size
= len
;
1490 return ll_new_hw_segment(q
, req
, bio
);
1493 static int ll_merge_requests_fn(request_queue_t
*q
, struct request
*req
,
1494 struct request
*next
)
1496 int total_phys_segments
;
1497 int total_hw_segments
;
1500 * First check if the either of the requests are re-queued
1501 * requests. Can't merge them if they are.
1503 if (req
->special
|| next
->special
)
1507 * Will it become too large?
1509 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1512 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1513 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1514 total_phys_segments
--;
1516 if (total_phys_segments
> q
->max_phys_segments
)
1519 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1520 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1521 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1523 * propagate the combined length to the end of the requests
1525 if (req
->nr_hw_segments
== 1)
1526 req
->bio
->bi_hw_front_size
= len
;
1527 if (next
->nr_hw_segments
== 1)
1528 next
->biotail
->bi_hw_back_size
= len
;
1529 total_hw_segments
--;
1532 if (total_hw_segments
> q
->max_hw_segments
)
1535 /* Merge is OK... */
1536 req
->nr_phys_segments
= total_phys_segments
;
1537 req
->nr_hw_segments
= total_hw_segments
;
1542 * "plug" the device if there are no outstanding requests: this will
1543 * force the transfer to start only after we have put all the requests
1546 * This is called with interrupts off and no requests on the queue and
1547 * with the queue lock held.
1549 void blk_plug_device(request_queue_t
*q
)
1551 WARN_ON(!irqs_disabled());
1554 * don't plug a stopped queue, it must be paired with blk_start_queue()
1555 * which will restart the queueing
1557 if (blk_queue_stopped(q
))
1560 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1561 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1562 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1566 EXPORT_SYMBOL(blk_plug_device
);
1569 * remove the queue from the plugged list, if present. called with
1570 * queue lock held and interrupts disabled.
1572 int blk_remove_plug(request_queue_t
*q
)
1574 WARN_ON(!irqs_disabled());
1576 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1579 del_timer(&q
->unplug_timer
);
1583 EXPORT_SYMBOL(blk_remove_plug
);
1586 * remove the plug and let it rip..
1588 void __generic_unplug_device(request_queue_t
*q
)
1590 if (unlikely(blk_queue_stopped(q
)))
1593 if (!blk_remove_plug(q
))
1598 EXPORT_SYMBOL(__generic_unplug_device
);
1601 * generic_unplug_device - fire a request queue
1602 * @q: The &request_queue_t in question
1605 * Linux uses plugging to build bigger requests queues before letting
1606 * the device have at them. If a queue is plugged, the I/O scheduler
1607 * is still adding and merging requests on the queue. Once the queue
1608 * gets unplugged, the request_fn defined for the queue is invoked and
1609 * transfers started.
1611 void generic_unplug_device(request_queue_t
*q
)
1613 spin_lock_irq(q
->queue_lock
);
1614 __generic_unplug_device(q
);
1615 spin_unlock_irq(q
->queue_lock
);
1617 EXPORT_SYMBOL(generic_unplug_device
);
1619 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1622 request_queue_t
*q
= bdi
->unplug_io_data
;
1625 * devices don't necessarily have an ->unplug_fn defined
1628 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1629 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1635 static void blk_unplug_work(struct work_struct
*work
)
1637 request_queue_t
*q
= container_of(work
, request_queue_t
, unplug_work
);
1639 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1640 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1645 static void blk_unplug_timeout(unsigned long data
)
1647 request_queue_t
*q
= (request_queue_t
*)data
;
1649 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1650 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1652 kblockd_schedule_work(&q
->unplug_work
);
1656 * blk_start_queue - restart a previously stopped queue
1657 * @q: The &request_queue_t in question
1660 * blk_start_queue() will clear the stop flag on the queue, and call
1661 * the request_fn for the queue if it was in a stopped state when
1662 * entered. Also see blk_stop_queue(). Queue lock must be held.
1664 void blk_start_queue(request_queue_t
*q
)
1666 WARN_ON(!irqs_disabled());
1668 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1671 * one level of recursion is ok and is much faster than kicking
1672 * the unplug handling
1674 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1676 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1679 kblockd_schedule_work(&q
->unplug_work
);
1683 EXPORT_SYMBOL(blk_start_queue
);
1686 * blk_stop_queue - stop a queue
1687 * @q: The &request_queue_t in question
1690 * The Linux block layer assumes that a block driver will consume all
1691 * entries on the request queue when the request_fn strategy is called.
1692 * Often this will not happen, because of hardware limitations (queue
1693 * depth settings). If a device driver gets a 'queue full' response,
1694 * or if it simply chooses not to queue more I/O at one point, it can
1695 * call this function to prevent the request_fn from being called until
1696 * the driver has signalled it's ready to go again. This happens by calling
1697 * blk_start_queue() to restart queue operations. Queue lock must be held.
1699 void blk_stop_queue(request_queue_t
*q
)
1702 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1704 EXPORT_SYMBOL(blk_stop_queue
);
1707 * blk_sync_queue - cancel any pending callbacks on a queue
1711 * The block layer may perform asynchronous callback activity
1712 * on a queue, such as calling the unplug function after a timeout.
1713 * A block device may call blk_sync_queue to ensure that any
1714 * such activity is cancelled, thus allowing it to release resources
1715 * the the callbacks might use. The caller must already have made sure
1716 * that its ->make_request_fn will not re-add plugging prior to calling
1720 void blk_sync_queue(struct request_queue
*q
)
1722 del_timer_sync(&q
->unplug_timer
);
1725 EXPORT_SYMBOL(blk_sync_queue
);
1728 * blk_run_queue - run a single device queue
1729 * @q: The queue to run
1731 void blk_run_queue(struct request_queue
*q
)
1733 unsigned long flags
;
1735 spin_lock_irqsave(q
->queue_lock
, flags
);
1739 * Only recurse once to avoid overrunning the stack, let the unplug
1740 * handling reinvoke the handler shortly if we already got there.
1742 if (!elv_queue_empty(q
)) {
1743 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1745 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1748 kblockd_schedule_work(&q
->unplug_work
);
1752 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1754 EXPORT_SYMBOL(blk_run_queue
);
1757 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1758 * @kobj: the kobj belonging of the request queue to be released
1761 * blk_cleanup_queue is the pair to blk_init_queue() or
1762 * blk_queue_make_request(). It should be called when a request queue is
1763 * being released; typically when a block device is being de-registered.
1764 * Currently, its primary task it to free all the &struct request
1765 * structures that were allocated to the queue and the queue itself.
1768 * Hopefully the low level driver will have finished any
1769 * outstanding requests first...
1771 static void blk_release_queue(struct kobject
*kobj
)
1773 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
1774 struct request_list
*rl
= &q
->rq
;
1779 mempool_destroy(rl
->rq_pool
);
1782 __blk_queue_free_tags(q
);
1784 blk_trace_shutdown(q
);
1786 kmem_cache_free(requestq_cachep
, q
);
1789 void blk_put_queue(request_queue_t
*q
)
1791 kobject_put(&q
->kobj
);
1793 EXPORT_SYMBOL(blk_put_queue
);
1795 void blk_cleanup_queue(request_queue_t
* q
)
1797 mutex_lock(&q
->sysfs_lock
);
1798 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1799 mutex_unlock(&q
->sysfs_lock
);
1802 elevator_exit(q
->elevator
);
1807 EXPORT_SYMBOL(blk_cleanup_queue
);
1809 static int blk_init_free_list(request_queue_t
*q
)
1811 struct request_list
*rl
= &q
->rq
;
1813 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1814 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1816 init_waitqueue_head(&rl
->wait
[READ
]);
1817 init_waitqueue_head(&rl
->wait
[WRITE
]);
1819 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1820 mempool_free_slab
, request_cachep
, q
->node
);
1828 request_queue_t
*blk_alloc_queue(gfp_t gfp_mask
)
1830 return blk_alloc_queue_node(gfp_mask
, -1);
1832 EXPORT_SYMBOL(blk_alloc_queue
);
1834 static struct kobj_type queue_ktype
;
1836 request_queue_t
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1840 q
= kmem_cache_alloc_node(requestq_cachep
, gfp_mask
, node_id
);
1844 memset(q
, 0, sizeof(*q
));
1845 init_timer(&q
->unplug_timer
);
1847 snprintf(q
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "queue");
1848 q
->kobj
.ktype
= &queue_ktype
;
1849 kobject_init(&q
->kobj
);
1851 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1852 q
->backing_dev_info
.unplug_io_data
= q
;
1854 mutex_init(&q
->sysfs_lock
);
1858 EXPORT_SYMBOL(blk_alloc_queue_node
);
1861 * blk_init_queue - prepare a request queue for use with a block device
1862 * @rfn: The function to be called to process requests that have been
1863 * placed on the queue.
1864 * @lock: Request queue spin lock
1867 * If a block device wishes to use the standard request handling procedures,
1868 * which sorts requests and coalesces adjacent requests, then it must
1869 * call blk_init_queue(). The function @rfn will be called when there
1870 * are requests on the queue that need to be processed. If the device
1871 * supports plugging, then @rfn may not be called immediately when requests
1872 * are available on the queue, but may be called at some time later instead.
1873 * Plugged queues are generally unplugged when a buffer belonging to one
1874 * of the requests on the queue is needed, or due to memory pressure.
1876 * @rfn is not required, or even expected, to remove all requests off the
1877 * queue, but only as many as it can handle at a time. If it does leave
1878 * requests on the queue, it is responsible for arranging that the requests
1879 * get dealt with eventually.
1881 * The queue spin lock must be held while manipulating the requests on the
1882 * request queue; this lock will be taken also from interrupt context, so irq
1883 * disabling is needed for it.
1885 * Function returns a pointer to the initialized request queue, or NULL if
1886 * it didn't succeed.
1889 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1890 * when the block device is deactivated (such as at module unload).
1893 request_queue_t
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1895 return blk_init_queue_node(rfn
, lock
, -1);
1897 EXPORT_SYMBOL(blk_init_queue
);
1900 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1902 request_queue_t
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1908 if (blk_init_free_list(q
)) {
1909 kmem_cache_free(requestq_cachep
, q
);
1914 * if caller didn't supply a lock, they get per-queue locking with
1918 spin_lock_init(&q
->__queue_lock
);
1919 lock
= &q
->__queue_lock
;
1922 q
->request_fn
= rfn
;
1923 q
->back_merge_fn
= ll_back_merge_fn
;
1924 q
->front_merge_fn
= ll_front_merge_fn
;
1925 q
->merge_requests_fn
= ll_merge_requests_fn
;
1926 q
->prep_rq_fn
= NULL
;
1927 q
->unplug_fn
= generic_unplug_device
;
1928 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1929 q
->queue_lock
= lock
;
1931 blk_queue_segment_boundary(q
, 0xffffffff);
1933 blk_queue_make_request(q
, __make_request
);
1934 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1936 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1937 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1942 if (!elevator_init(q
, NULL
)) {
1943 blk_queue_congestion_threshold(q
);
1950 EXPORT_SYMBOL(blk_init_queue_node
);
1952 int blk_get_queue(request_queue_t
*q
)
1954 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1955 kobject_get(&q
->kobj
);
1962 EXPORT_SYMBOL(blk_get_queue
);
1964 static inline void blk_free_request(request_queue_t
*q
, struct request
*rq
)
1966 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1967 elv_put_request(q
, rq
);
1968 mempool_free(rq
, q
->rq
.rq_pool
);
1971 static struct request
*
1972 blk_alloc_request(request_queue_t
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1974 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1980 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1981 * see bio.h and blkdev.h
1983 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
1986 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
1987 mempool_free(rq
, q
->rq
.rq_pool
);
1990 rq
->cmd_flags
|= REQ_ELVPRIV
;
1997 * ioc_batching returns true if the ioc is a valid batching request and
1998 * should be given priority access to a request.
2000 static inline int ioc_batching(request_queue_t
*q
, struct io_context
*ioc
)
2006 * Make sure the process is able to allocate at least 1 request
2007 * even if the batch times out, otherwise we could theoretically
2010 return ioc
->nr_batch_requests
== q
->nr_batching
||
2011 (ioc
->nr_batch_requests
> 0
2012 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2016 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2017 * will cause the process to be a "batcher" on all queues in the system. This
2018 * is the behaviour we want though - once it gets a wakeup it should be given
2021 static void ioc_set_batching(request_queue_t
*q
, struct io_context
*ioc
)
2023 if (!ioc
|| ioc_batching(q
, ioc
))
2026 ioc
->nr_batch_requests
= q
->nr_batching
;
2027 ioc
->last_waited
= jiffies
;
2030 static void __freed_request(request_queue_t
*q
, int rw
)
2032 struct request_list
*rl
= &q
->rq
;
2034 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2035 blk_clear_queue_congested(q
, rw
);
2037 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2038 if (waitqueue_active(&rl
->wait
[rw
]))
2039 wake_up(&rl
->wait
[rw
]);
2041 blk_clear_queue_full(q
, rw
);
2046 * A request has just been released. Account for it, update the full and
2047 * congestion status, wake up any waiters. Called under q->queue_lock.
2049 static void freed_request(request_queue_t
*q
, int rw
, int priv
)
2051 struct request_list
*rl
= &q
->rq
;
2057 __freed_request(q
, rw
);
2059 if (unlikely(rl
->starved
[rw
^ 1]))
2060 __freed_request(q
, rw
^ 1);
2063 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2065 * Get a free request, queue_lock must be held.
2066 * Returns NULL on failure, with queue_lock held.
2067 * Returns !NULL on success, with queue_lock *not held*.
2069 static struct request
*get_request(request_queue_t
*q
, int rw
, struct bio
*bio
,
2072 struct request
*rq
= NULL
;
2073 struct request_list
*rl
= &q
->rq
;
2074 struct io_context
*ioc
= NULL
;
2075 int may_queue
, priv
;
2077 may_queue
= elv_may_queue(q
, rw
);
2078 if (may_queue
== ELV_MQUEUE_NO
)
2081 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2082 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2083 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2085 * The queue will fill after this allocation, so set
2086 * it as full, and mark this process as "batching".
2087 * This process will be allowed to complete a batch of
2088 * requests, others will be blocked.
2090 if (!blk_queue_full(q
, rw
)) {
2091 ioc_set_batching(q
, ioc
);
2092 blk_set_queue_full(q
, rw
);
2094 if (may_queue
!= ELV_MQUEUE_MUST
2095 && !ioc_batching(q
, ioc
)) {
2097 * The queue is full and the allocating
2098 * process is not a "batcher", and not
2099 * exempted by the IO scheduler
2105 blk_set_queue_congested(q
, rw
);
2109 * Only allow batching queuers to allocate up to 50% over the defined
2110 * limit of requests, otherwise we could have thousands of requests
2111 * allocated with any setting of ->nr_requests
2113 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2117 rl
->starved
[rw
] = 0;
2119 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2123 spin_unlock_irq(q
->queue_lock
);
2125 rq
= blk_alloc_request(q
, rw
, priv
, gfp_mask
);
2126 if (unlikely(!rq
)) {
2128 * Allocation failed presumably due to memory. Undo anything
2129 * we might have messed up.
2131 * Allocating task should really be put onto the front of the
2132 * wait queue, but this is pretty rare.
2134 spin_lock_irq(q
->queue_lock
);
2135 freed_request(q
, rw
, priv
);
2138 * in the very unlikely event that allocation failed and no
2139 * requests for this direction was pending, mark us starved
2140 * so that freeing of a request in the other direction will
2141 * notice us. another possible fix would be to split the
2142 * rq mempool into READ and WRITE
2145 if (unlikely(rl
->count
[rw
] == 0))
2146 rl
->starved
[rw
] = 1;
2152 * ioc may be NULL here, and ioc_batching will be false. That's
2153 * OK, if the queue is under the request limit then requests need
2154 * not count toward the nr_batch_requests limit. There will always
2155 * be some limit enforced by BLK_BATCH_TIME.
2157 if (ioc_batching(q
, ioc
))
2158 ioc
->nr_batch_requests
--;
2162 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2168 * No available requests for this queue, unplug the device and wait for some
2169 * requests to become available.
2171 * Called with q->queue_lock held, and returns with it unlocked.
2173 static struct request
*get_request_wait(request_queue_t
*q
, int rw
,
2178 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2181 struct request_list
*rl
= &q
->rq
;
2183 prepare_to_wait_exclusive(&rl
->wait
[rw
], &wait
,
2184 TASK_UNINTERRUPTIBLE
);
2186 rq
= get_request(q
, rw
, bio
, GFP_NOIO
);
2189 struct io_context
*ioc
;
2191 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_SLEEPRQ
);
2193 __generic_unplug_device(q
);
2194 spin_unlock_irq(q
->queue_lock
);
2198 * After sleeping, we become a "batching" process and
2199 * will be able to allocate at least one request, and
2200 * up to a big batch of them for a small period time.
2201 * See ioc_batching, ioc_set_batching
2203 ioc
= current_io_context(GFP_NOIO
, q
->node
);
2204 ioc_set_batching(q
, ioc
);
2206 spin_lock_irq(q
->queue_lock
);
2208 finish_wait(&rl
->wait
[rw
], &wait
);
2214 struct request
*blk_get_request(request_queue_t
*q
, int rw
, gfp_t gfp_mask
)
2218 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
2220 spin_lock_irq(q
->queue_lock
);
2221 if (gfp_mask
& __GFP_WAIT
) {
2222 rq
= get_request_wait(q
, rw
, NULL
);
2224 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
2226 spin_unlock_irq(q
->queue_lock
);
2228 /* q->queue_lock is unlocked at this point */
2232 EXPORT_SYMBOL(blk_get_request
);
2235 * blk_start_queueing - initiate dispatch of requests to device
2236 * @q: request queue to kick into gear
2238 * This is basically a helper to remove the need to know whether a queue
2239 * is plugged or not if someone just wants to initiate dispatch of requests
2242 * The queue lock must be held with interrupts disabled.
2244 void blk_start_queueing(request_queue_t
*q
)
2246 if (!blk_queue_plugged(q
))
2249 __generic_unplug_device(q
);
2251 EXPORT_SYMBOL(blk_start_queueing
);
2254 * blk_requeue_request - put a request back on queue
2255 * @q: request queue where request should be inserted
2256 * @rq: request to be inserted
2259 * Drivers often keep queueing requests until the hardware cannot accept
2260 * more, when that condition happens we need to put the request back
2261 * on the queue. Must be called with queue lock held.
2263 void blk_requeue_request(request_queue_t
*q
, struct request
*rq
)
2265 blk_add_trace_rq(q
, rq
, BLK_TA_REQUEUE
);
2267 if (blk_rq_tagged(rq
))
2268 blk_queue_end_tag(q
, rq
);
2270 elv_requeue_request(q
, rq
);
2273 EXPORT_SYMBOL(blk_requeue_request
);
2276 * blk_insert_request - insert a special request in to a request queue
2277 * @q: request queue where request should be inserted
2278 * @rq: request to be inserted
2279 * @at_head: insert request at head or tail of queue
2280 * @data: private data
2283 * Many block devices need to execute commands asynchronously, so they don't
2284 * block the whole kernel from preemption during request execution. This is
2285 * accomplished normally by inserting aritficial requests tagged as
2286 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2287 * scheduled for actual execution by the request queue.
2289 * We have the option of inserting the head or the tail of the queue.
2290 * Typically we use the tail for new ioctls and so forth. We use the head
2291 * of the queue for things like a QUEUE_FULL message from a device, or a
2292 * host that is unable to accept a particular command.
2294 void blk_insert_request(request_queue_t
*q
, struct request
*rq
,
2295 int at_head
, void *data
)
2297 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2298 unsigned long flags
;
2301 * tell I/O scheduler that this isn't a regular read/write (ie it
2302 * must not attempt merges on this) and that it acts as a soft
2305 rq
->cmd_type
= REQ_TYPE_SPECIAL
;
2306 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
2310 spin_lock_irqsave(q
->queue_lock
, flags
);
2313 * If command is tagged, release the tag
2315 if (blk_rq_tagged(rq
))
2316 blk_queue_end_tag(q
, rq
);
2318 drive_stat_acct(rq
, rq
->nr_sectors
, 1);
2319 __elv_add_request(q
, rq
, where
, 0);
2320 blk_start_queueing(q
);
2321 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2324 EXPORT_SYMBOL(blk_insert_request
);
2326 static int __blk_rq_unmap_user(struct bio
*bio
)
2331 if (bio_flagged(bio
, BIO_USER_MAPPED
))
2332 bio_unmap_user(bio
);
2334 ret
= bio_uncopy_user(bio
);
2340 static int __blk_rq_map_user(request_queue_t
*q
, struct request
*rq
,
2341 void __user
*ubuf
, unsigned int len
)
2343 unsigned long uaddr
;
2344 struct bio
*bio
, *orig_bio
;
2347 reading
= rq_data_dir(rq
) == READ
;
2350 * if alignment requirement is satisfied, map in user pages for
2351 * direct dma. else, set up kernel bounce buffers
2353 uaddr
= (unsigned long) ubuf
;
2354 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2355 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2357 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2360 return PTR_ERR(bio
);
2364 blk_queue_bounce(q
, &bio
);
2366 * We link the bounce buffer in and could have to traverse it
2367 * later so we have to get a ref to prevent it from being freed
2372 * for most (all? don't know of any) queues we could
2373 * skip grabbing the queue lock here. only drivers with
2374 * funky private ->back_merge_fn() function could be
2377 spin_lock_irq(q
->queue_lock
);
2379 blk_rq_bio_prep(q
, rq
, bio
);
2380 else if (!q
->back_merge_fn(q
, rq
, bio
)) {
2382 spin_unlock_irq(q
->queue_lock
);
2385 rq
->biotail
->bi_next
= bio
;
2388 rq
->nr_sectors
+= bio_sectors(bio
);
2389 rq
->hard_nr_sectors
= rq
->nr_sectors
;
2390 rq
->data_len
+= bio
->bi_size
;
2392 spin_unlock_irq(q
->queue_lock
);
2394 return bio
->bi_size
;
2397 /* if it was boucned we must call the end io function */
2398 bio_endio(bio
, bio
->bi_size
, 0);
2399 __blk_rq_unmap_user(orig_bio
);
2405 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2406 * @q: request queue where request should be inserted
2407 * @rq: request structure to fill
2408 * @ubuf: the user buffer
2409 * @len: length of user data
2412 * Data will be mapped directly for zero copy io, if possible. Otherwise
2413 * a kernel bounce buffer is used.
2415 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2416 * still in process context.
2418 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2419 * before being submitted to the device, as pages mapped may be out of
2420 * reach. It's the callers responsibility to make sure this happens. The
2421 * original bio must be passed back in to blk_rq_unmap_user() for proper
2424 int blk_rq_map_user(request_queue_t
*q
, struct request
*rq
, void __user
*ubuf
,
2427 unsigned long bytes_read
= 0;
2430 if (len
> (q
->max_hw_sectors
<< 9))
2435 while (bytes_read
!= len
) {
2436 unsigned long map_len
, end
, start
;
2438 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2439 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2441 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2444 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2445 * pages. If this happens we just lower the requested
2446 * mapping len by a page so that we can fit
2448 if (end
- start
> BIO_MAX_PAGES
)
2449 map_len
-= PAGE_SIZE
;
2451 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2458 rq
->buffer
= rq
->data
= NULL
;
2461 blk_rq_unmap_user(rq
);
2465 EXPORT_SYMBOL(blk_rq_map_user
);
2468 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2469 * @q: request queue where request should be inserted
2470 * @rq: request to map data to
2471 * @iov: pointer to the iovec
2472 * @iov_count: number of elements in the iovec
2475 * Data will be mapped directly for zero copy io, if possible. Otherwise
2476 * a kernel bounce buffer is used.
2478 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2479 * still in process context.
2481 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2482 * before being submitted to the device, as pages mapped may be out of
2483 * reach. It's the callers responsibility to make sure this happens. The
2484 * original bio must be passed back in to blk_rq_unmap_user() for proper
2487 int blk_rq_map_user_iov(request_queue_t
*q
, struct request
*rq
,
2488 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2492 if (!iov
|| iov_count
<= 0)
2495 /* we don't allow misaligned data like bio_map_user() does. If the
2496 * user is using sg, they're expected to know the alignment constraints
2497 * and respect them accordingly */
2498 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2500 return PTR_ERR(bio
);
2502 if (bio
->bi_size
!= len
) {
2503 bio_endio(bio
, bio
->bi_size
, 0);
2504 bio_unmap_user(bio
);
2509 blk_rq_bio_prep(q
, rq
, bio
);
2510 rq
->buffer
= rq
->data
= NULL
;
2514 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2517 * blk_rq_unmap_user - unmap a request with user data
2518 * @rq: rq to be unmapped
2521 * Unmap a rq previously mapped by blk_rq_map_user().
2522 * rq->bio must be set to the original head of the request.
2524 int blk_rq_unmap_user(struct request
*rq
)
2526 struct bio
*bio
, *mapped_bio
;
2528 while ((bio
= rq
->bio
)) {
2529 if (bio_flagged(bio
, BIO_BOUNCED
))
2530 mapped_bio
= bio
->bi_private
;
2534 __blk_rq_unmap_user(mapped_bio
);
2535 rq
->bio
= bio
->bi_next
;
2541 EXPORT_SYMBOL(blk_rq_unmap_user
);
2544 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2545 * @q: request queue where request should be inserted
2546 * @rq: request to fill
2547 * @kbuf: the kernel buffer
2548 * @len: length of user data
2549 * @gfp_mask: memory allocation flags
2551 int blk_rq_map_kern(request_queue_t
*q
, struct request
*rq
, void *kbuf
,
2552 unsigned int len
, gfp_t gfp_mask
)
2556 if (len
> (q
->max_hw_sectors
<< 9))
2561 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2563 return PTR_ERR(bio
);
2565 if (rq_data_dir(rq
) == WRITE
)
2566 bio
->bi_rw
|= (1 << BIO_RW
);
2568 blk_rq_bio_prep(q
, rq
, bio
);
2569 rq
->buffer
= rq
->data
= NULL
;
2573 EXPORT_SYMBOL(blk_rq_map_kern
);
2576 * blk_execute_rq_nowait - insert a request into queue for execution
2577 * @q: queue to insert the request in
2578 * @bd_disk: matching gendisk
2579 * @rq: request to insert
2580 * @at_head: insert request at head or tail of queue
2581 * @done: I/O completion handler
2584 * Insert a fully prepared request at the back of the io scheduler queue
2585 * for execution. Don't wait for completion.
2587 void blk_execute_rq_nowait(request_queue_t
*q
, struct gendisk
*bd_disk
,
2588 struct request
*rq
, int at_head
,
2591 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2593 rq
->rq_disk
= bd_disk
;
2594 rq
->cmd_flags
|= REQ_NOMERGE
;
2596 WARN_ON(irqs_disabled());
2597 spin_lock_irq(q
->queue_lock
);
2598 __elv_add_request(q
, rq
, where
, 1);
2599 __generic_unplug_device(q
);
2600 spin_unlock_irq(q
->queue_lock
);
2602 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2605 * blk_execute_rq - insert a request into queue for execution
2606 * @q: queue to insert the request in
2607 * @bd_disk: matching gendisk
2608 * @rq: request to insert
2609 * @at_head: insert request at head or tail of queue
2612 * Insert a fully prepared request at the back of the io scheduler queue
2613 * for execution and wait for completion.
2615 int blk_execute_rq(request_queue_t
*q
, struct gendisk
*bd_disk
,
2616 struct request
*rq
, int at_head
)
2618 DECLARE_COMPLETION_ONSTACK(wait
);
2619 char sense
[SCSI_SENSE_BUFFERSIZE
];
2623 * we need an extra reference to the request, so we can look at
2624 * it after io completion
2629 memset(sense
, 0, sizeof(sense
));
2634 rq
->end_io_data
= &wait
;
2635 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2636 wait_for_completion(&wait
);
2644 EXPORT_SYMBOL(blk_execute_rq
);
2647 * blkdev_issue_flush - queue a flush
2648 * @bdev: blockdev to issue flush for
2649 * @error_sector: error sector
2652 * Issue a flush for the block device in question. Caller can supply
2653 * room for storing the error offset in case of a flush error, if they
2654 * wish to. Caller must run wait_for_completion() on its own.
2656 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2660 if (bdev
->bd_disk
== NULL
)
2663 q
= bdev_get_queue(bdev
);
2666 if (!q
->issue_flush_fn
)
2669 return q
->issue_flush_fn(q
, bdev
->bd_disk
, error_sector
);
2672 EXPORT_SYMBOL(blkdev_issue_flush
);
2674 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2676 int rw
= rq_data_dir(rq
);
2678 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2682 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2684 disk_round_stats(rq
->rq_disk
);
2685 rq
->rq_disk
->in_flight
++;
2690 * add-request adds a request to the linked list.
2691 * queue lock is held and interrupts disabled, as we muck with the
2692 * request queue list.
2694 static inline void add_request(request_queue_t
* q
, struct request
* req
)
2696 drive_stat_acct(req
, req
->nr_sectors
, 1);
2699 q
->activity_fn(q
->activity_data
, rq_data_dir(req
));
2702 * elevator indicated where it wants this request to be
2703 * inserted at elevator_merge time
2705 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2709 * disk_round_stats() - Round off the performance stats on a struct
2712 * The average IO queue length and utilisation statistics are maintained
2713 * by observing the current state of the queue length and the amount of
2714 * time it has been in this state for.
2716 * Normally, that accounting is done on IO completion, but that can result
2717 * in more than a second's worth of IO being accounted for within any one
2718 * second, leading to >100% utilisation. To deal with that, we call this
2719 * function to do a round-off before returning the results when reading
2720 * /proc/diskstats. This accounts immediately for all queue usage up to
2721 * the current jiffies and restarts the counters again.
2723 void disk_round_stats(struct gendisk
*disk
)
2725 unsigned long now
= jiffies
;
2727 if (now
== disk
->stamp
)
2730 if (disk
->in_flight
) {
2731 __disk_stat_add(disk
, time_in_queue
,
2732 disk
->in_flight
* (now
- disk
->stamp
));
2733 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2738 EXPORT_SYMBOL_GPL(disk_round_stats
);
2741 * queue lock must be held
2743 void __blk_put_request(request_queue_t
*q
, struct request
*req
)
2747 if (unlikely(--req
->ref_count
))
2750 elv_completed_request(q
, req
);
2753 * Request may not have originated from ll_rw_blk. if not,
2754 * it didn't come out of our reserved rq pools
2756 if (req
->cmd_flags
& REQ_ALLOCED
) {
2757 int rw
= rq_data_dir(req
);
2758 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2760 BUG_ON(!list_empty(&req
->queuelist
));
2761 BUG_ON(!hlist_unhashed(&req
->hash
));
2763 blk_free_request(q
, req
);
2764 freed_request(q
, rw
, priv
);
2768 EXPORT_SYMBOL_GPL(__blk_put_request
);
2770 void blk_put_request(struct request
*req
)
2772 unsigned long flags
;
2773 request_queue_t
*q
= req
->q
;
2776 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2777 * following if (q) test.
2780 spin_lock_irqsave(q
->queue_lock
, flags
);
2781 __blk_put_request(q
, req
);
2782 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2786 EXPORT_SYMBOL(blk_put_request
);
2789 * blk_end_sync_rq - executes a completion event on a request
2790 * @rq: request to complete
2791 * @error: end io status of the request
2793 void blk_end_sync_rq(struct request
*rq
, int error
)
2795 struct completion
*waiting
= rq
->end_io_data
;
2797 rq
->end_io_data
= NULL
;
2798 __blk_put_request(rq
->q
, rq
);
2801 * complete last, if this is a stack request the process (and thus
2802 * the rq pointer) could be invalid right after this complete()
2806 EXPORT_SYMBOL(blk_end_sync_rq
);
2809 * Has to be called with the request spinlock acquired
2811 static int attempt_merge(request_queue_t
*q
, struct request
*req
,
2812 struct request
*next
)
2814 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2820 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2823 if (rq_data_dir(req
) != rq_data_dir(next
)
2824 || req
->rq_disk
!= next
->rq_disk
2829 * If we are allowed to merge, then append bio list
2830 * from next to rq and release next. merge_requests_fn
2831 * will have updated segment counts, update sector
2834 if (!q
->merge_requests_fn(q
, req
, next
))
2838 * At this point we have either done a back merge
2839 * or front merge. We need the smaller start_time of
2840 * the merged requests to be the current request
2841 * for accounting purposes.
2843 if (time_after(req
->start_time
, next
->start_time
))
2844 req
->start_time
= next
->start_time
;
2846 req
->biotail
->bi_next
= next
->bio
;
2847 req
->biotail
= next
->biotail
;
2849 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2851 elv_merge_requests(q
, req
, next
);
2854 disk_round_stats(req
->rq_disk
);
2855 req
->rq_disk
->in_flight
--;
2858 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2860 __blk_put_request(q
, next
);
2864 static inline int attempt_back_merge(request_queue_t
*q
, struct request
*rq
)
2866 struct request
*next
= elv_latter_request(q
, rq
);
2869 return attempt_merge(q
, rq
, next
);
2874 static inline int attempt_front_merge(request_queue_t
*q
, struct request
*rq
)
2876 struct request
*prev
= elv_former_request(q
, rq
);
2879 return attempt_merge(q
, prev
, rq
);
2884 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2886 req
->cmd_type
= REQ_TYPE_FS
;
2889 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2891 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2892 req
->cmd_flags
|= REQ_FAILFAST
;
2895 * REQ_BARRIER implies no merging, but lets make it explicit
2897 if (unlikely(bio_barrier(bio
)))
2898 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2901 req
->cmd_flags
|= REQ_RW_SYNC
;
2902 if (bio_rw_meta(bio
))
2903 req
->cmd_flags
|= REQ_RW_META
;
2906 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2907 req
->hard_nr_sectors
= req
->nr_sectors
= bio_sectors(bio
);
2908 req
->current_nr_sectors
= req
->hard_cur_sectors
= bio_cur_sectors(bio
);
2909 req
->nr_phys_segments
= bio_phys_segments(req
->q
, bio
);
2910 req
->nr_hw_segments
= bio_hw_segments(req
->q
, bio
);
2911 req
->buffer
= bio_data(bio
); /* see ->buffer comment above */
2912 req
->bio
= req
->biotail
= bio
;
2913 req
->ioprio
= bio_prio(bio
);
2914 req
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2915 req
->start_time
= jiffies
;
2918 static int __make_request(request_queue_t
*q
, struct bio
*bio
)
2920 struct request
*req
;
2921 int el_ret
, nr_sectors
, barrier
, err
;
2922 const unsigned short prio
= bio_prio(bio
);
2923 const int sync
= bio_sync(bio
);
2925 nr_sectors
= bio_sectors(bio
);
2928 * low level driver can indicate that it wants pages above a
2929 * certain limit bounced to low memory (ie for highmem, or even
2930 * ISA dma in theory)
2932 blk_queue_bounce(q
, &bio
);
2934 barrier
= bio_barrier(bio
);
2935 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2940 spin_lock_irq(q
->queue_lock
);
2942 if (unlikely(barrier
) || elv_queue_empty(q
))
2945 el_ret
= elv_merge(q
, &req
, bio
);
2947 case ELEVATOR_BACK_MERGE
:
2948 BUG_ON(!rq_mergeable(req
));
2950 if (!q
->back_merge_fn(q
, req
, bio
))
2953 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2955 req
->biotail
->bi_next
= bio
;
2957 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2958 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2959 drive_stat_acct(req
, nr_sectors
, 0);
2960 if (!attempt_back_merge(q
, req
))
2961 elv_merged_request(q
, req
, el_ret
);
2964 case ELEVATOR_FRONT_MERGE
:
2965 BUG_ON(!rq_mergeable(req
));
2967 if (!q
->front_merge_fn(q
, req
, bio
))
2970 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
2972 bio
->bi_next
= req
->bio
;
2976 * may not be valid. if the low level driver said
2977 * it didn't need a bounce buffer then it better
2978 * not touch req->buffer either...
2980 req
->buffer
= bio_data(bio
);
2981 req
->current_nr_sectors
= bio_cur_sectors(bio
);
2982 req
->hard_cur_sectors
= req
->current_nr_sectors
;
2983 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
2984 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2985 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2986 drive_stat_acct(req
, nr_sectors
, 0);
2987 if (!attempt_front_merge(q
, req
))
2988 elv_merged_request(q
, req
, el_ret
);
2991 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2998 * Grab a free request. This is might sleep but can not fail.
2999 * Returns with the queue unlocked.
3001 req
= get_request_wait(q
, bio_data_dir(bio
), bio
);
3004 * After dropping the lock and possibly sleeping here, our request
3005 * may now be mergeable after it had proven unmergeable (above).
3006 * We don't worry about that case for efficiency. It won't happen
3007 * often, and the elevators are able to handle it.
3009 init_request_from_bio(req
, bio
);
3011 spin_lock_irq(q
->queue_lock
);
3012 if (elv_queue_empty(q
))
3014 add_request(q
, req
);
3017 __generic_unplug_device(q
);
3019 spin_unlock_irq(q
->queue_lock
);
3023 bio_endio(bio
, nr_sectors
<< 9, err
);
3028 * If bio->bi_dev is a partition, remap the location
3030 static inline void blk_partition_remap(struct bio
*bio
)
3032 struct block_device
*bdev
= bio
->bi_bdev
;
3034 if (bdev
!= bdev
->bd_contains
) {
3035 struct hd_struct
*p
= bdev
->bd_part
;
3036 const int rw
= bio_data_dir(bio
);
3038 p
->sectors
[rw
] += bio_sectors(bio
);
3041 bio
->bi_sector
+= p
->start_sect
;
3042 bio
->bi_bdev
= bdev
->bd_contains
;
3046 static void handle_bad_sector(struct bio
*bio
)
3048 char b
[BDEVNAME_SIZE
];
3050 printk(KERN_INFO
"attempt to access beyond end of device\n");
3051 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3052 bdevname(bio
->bi_bdev
, b
),
3054 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3055 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3057 set_bit(BIO_EOF
, &bio
->bi_flags
);
3060 #ifdef CONFIG_FAIL_MAKE_REQUEST
3062 static DECLARE_FAULT_ATTR(fail_make_request
);
3064 static int __init
setup_fail_make_request(char *str
)
3066 return setup_fault_attr(&fail_make_request
, str
);
3068 __setup("fail_make_request=", setup_fail_make_request
);
3070 static int should_fail_request(struct bio
*bio
)
3072 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3073 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3074 return should_fail(&fail_make_request
, bio
->bi_size
);
3079 static int __init
fail_make_request_debugfs(void)
3081 return init_fault_attr_dentries(&fail_make_request
,
3082 "fail_make_request");
3085 late_initcall(fail_make_request_debugfs
);
3087 #else /* CONFIG_FAIL_MAKE_REQUEST */
3089 static inline int should_fail_request(struct bio
*bio
)
3094 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3097 * generic_make_request: hand a buffer to its device driver for I/O
3098 * @bio: The bio describing the location in memory and on the device.
3100 * generic_make_request() is used to make I/O requests of block
3101 * devices. It is passed a &struct bio, which describes the I/O that needs
3104 * generic_make_request() does not return any status. The
3105 * success/failure status of the request, along with notification of
3106 * completion, is delivered asynchronously through the bio->bi_end_io
3107 * function described (one day) else where.
3109 * The caller of generic_make_request must make sure that bi_io_vec
3110 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3111 * set to describe the device address, and the
3112 * bi_end_io and optionally bi_private are set to describe how
3113 * completion notification should be signaled.
3115 * generic_make_request and the drivers it calls may use bi_next if this
3116 * bio happens to be merged with someone else, and may change bi_dev and
3117 * bi_sector for remaps as it sees fit. So the values of these fields
3118 * should NOT be depended on after the call to generic_make_request.
3120 void generic_make_request(struct bio
*bio
)
3124 sector_t old_sector
;
3125 int ret
, nr_sectors
= bio_sectors(bio
);
3129 /* Test device or partition size, when known. */
3130 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3132 sector_t sector
= bio
->bi_sector
;
3134 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3136 * This may well happen - the kernel calls bread()
3137 * without checking the size of the device, e.g., when
3138 * mounting a device.
3140 handle_bad_sector(bio
);
3146 * Resolve the mapping until finished. (drivers are
3147 * still free to implement/resolve their own stacking
3148 * by explicitly returning 0)
3150 * NOTE: we don't repeat the blk_size check for each new device.
3151 * Stacking drivers are expected to know what they are doing.
3156 char b
[BDEVNAME_SIZE
];
3158 q
= bdev_get_queue(bio
->bi_bdev
);
3161 "generic_make_request: Trying to access "
3162 "nonexistent block-device %s (%Lu)\n",
3163 bdevname(bio
->bi_bdev
, b
),
3164 (long long) bio
->bi_sector
);
3166 bio_endio(bio
, bio
->bi_size
, -EIO
);
3170 if (unlikely(bio_sectors(bio
) > q
->max_hw_sectors
)) {
3171 printk("bio too big device %s (%u > %u)\n",
3172 bdevname(bio
->bi_bdev
, b
),
3178 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3181 if (should_fail_request(bio
))
3185 * If this device has partitions, remap block n
3186 * of partition p to block n+start(p) of the disk.
3188 blk_partition_remap(bio
);
3190 if (old_sector
!= -1)
3191 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3194 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3196 old_sector
= bio
->bi_sector
;
3197 old_dev
= bio
->bi_bdev
->bd_dev
;
3199 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3201 sector_t sector
= bio
->bi_sector
;
3203 if (maxsector
< nr_sectors
||
3204 maxsector
- nr_sectors
< sector
) {
3206 * This may well happen - partitions are not
3207 * checked to make sure they are within the size
3208 * of the whole device.
3210 handle_bad_sector(bio
);
3215 ret
= q
->make_request_fn(q
, bio
);
3219 EXPORT_SYMBOL(generic_make_request
);
3222 * submit_bio: submit a bio to the block device layer for I/O
3223 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3224 * @bio: The &struct bio which describes the I/O
3226 * submit_bio() is very similar in purpose to generic_make_request(), and
3227 * uses that function to do most of the work. Both are fairly rough
3228 * interfaces, @bio must be presetup and ready for I/O.
3231 void submit_bio(int rw
, struct bio
*bio
)
3233 int count
= bio_sectors(bio
);
3235 BIO_BUG_ON(!bio
->bi_size
);
3236 BIO_BUG_ON(!bio
->bi_io_vec
);
3239 count_vm_events(PGPGOUT
, count
);
3241 count_vm_events(PGPGIN
, count
);
3243 if (unlikely(block_dump
)) {
3244 char b
[BDEVNAME_SIZE
];
3245 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3246 current
->comm
, current
->pid
,
3247 (rw
& WRITE
) ? "WRITE" : "READ",
3248 (unsigned long long)bio
->bi_sector
,
3249 bdevname(bio
->bi_bdev
,b
));
3252 generic_make_request(bio
);
3255 EXPORT_SYMBOL(submit_bio
);
3257 static void blk_recalc_rq_segments(struct request
*rq
)
3259 struct bio
*bio
, *prevbio
= NULL
;
3260 int nr_phys_segs
, nr_hw_segs
;
3261 unsigned int phys_size
, hw_size
;
3262 request_queue_t
*q
= rq
->q
;
3267 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
3268 rq_for_each_bio(bio
, rq
) {
3269 /* Force bio hw/phys segs to be recalculated. */
3270 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
3272 nr_phys_segs
+= bio_phys_segments(q
, bio
);
3273 nr_hw_segs
+= bio_hw_segments(q
, bio
);
3275 int pseg
= phys_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3276 int hseg
= hw_size
+ prevbio
->bi_size
+ bio
->bi_size
;
3278 if (blk_phys_contig_segment(q
, prevbio
, bio
) &&
3279 pseg
<= q
->max_segment_size
) {
3281 phys_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3285 if (blk_hw_contig_segment(q
, prevbio
, bio
) &&
3286 hseg
<= q
->max_segment_size
) {
3288 hw_size
+= prevbio
->bi_size
+ bio
->bi_size
;
3295 rq
->nr_phys_segments
= nr_phys_segs
;
3296 rq
->nr_hw_segments
= nr_hw_segs
;
3299 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3301 if (blk_fs_request(rq
)) {
3302 rq
->hard_sector
+= nsect
;
3303 rq
->hard_nr_sectors
-= nsect
;
3306 * Move the I/O submission pointers ahead if required.
3308 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3309 (rq
->sector
<= rq
->hard_sector
)) {
3310 rq
->sector
= rq
->hard_sector
;
3311 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3312 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3313 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3314 rq
->buffer
= bio_data(rq
->bio
);
3318 * if total number of sectors is less than the first segment
3319 * size, something has gone terribly wrong
3321 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3322 printk("blk: request botched\n");
3323 rq
->nr_sectors
= rq
->current_nr_sectors
;
3328 static int __end_that_request_first(struct request
*req
, int uptodate
,
3331 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3334 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3337 * extend uptodate bool to allow < 0 value to be direct io error
3340 if (end_io_error(uptodate
))
3341 error
= !uptodate
? -EIO
: uptodate
;
3344 * for a REQ_BLOCK_PC request, we want to carry any eventual
3345 * sense key with us all the way through
3347 if (!blk_pc_request(req
))
3351 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3352 printk("end_request: I/O error, dev %s, sector %llu\n",
3353 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3354 (unsigned long long)req
->sector
);
3357 if (blk_fs_request(req
) && req
->rq_disk
) {
3358 const int rw
= rq_data_dir(req
);
3360 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3363 total_bytes
= bio_nbytes
= 0;
3364 while ((bio
= req
->bio
) != NULL
) {
3367 if (nr_bytes
>= bio
->bi_size
) {
3368 req
->bio
= bio
->bi_next
;
3369 nbytes
= bio
->bi_size
;
3370 if (!ordered_bio_endio(req
, bio
, nbytes
, error
))
3371 bio_endio(bio
, nbytes
, error
);
3375 int idx
= bio
->bi_idx
+ next_idx
;
3377 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3378 blk_dump_rq_flags(req
, "__end_that");
3379 printk("%s: bio idx %d >= vcnt %d\n",
3381 bio
->bi_idx
, bio
->bi_vcnt
);
3385 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3386 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3389 * not a complete bvec done
3391 if (unlikely(nbytes
> nr_bytes
)) {
3392 bio_nbytes
+= nr_bytes
;
3393 total_bytes
+= nr_bytes
;
3398 * advance to the next vector
3401 bio_nbytes
+= nbytes
;
3404 total_bytes
+= nbytes
;
3407 if ((bio
= req
->bio
)) {
3409 * end more in this run, or just return 'not-done'
3411 if (unlikely(nr_bytes
<= 0))
3423 * if the request wasn't completed, update state
3426 if (!ordered_bio_endio(req
, bio
, bio_nbytes
, error
))
3427 bio_endio(bio
, bio_nbytes
, error
);
3428 bio
->bi_idx
+= next_idx
;
3429 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3430 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3433 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3434 blk_recalc_rq_segments(req
);
3439 * end_that_request_first - end I/O on a request
3440 * @req: the request being processed
3441 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3442 * @nr_sectors: number of sectors to end I/O on
3445 * Ends I/O on a number of sectors attached to @req, and sets it up
3446 * for the next range of segments (if any) in the cluster.
3449 * 0 - we are done with this request, call end_that_request_last()
3450 * 1 - still buffers pending for this request
3452 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3454 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3457 EXPORT_SYMBOL(end_that_request_first
);
3460 * end_that_request_chunk - end I/O on a request
3461 * @req: the request being processed
3462 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3463 * @nr_bytes: number of bytes to complete
3466 * Ends I/O on a number of bytes attached to @req, and sets it up
3467 * for the next range of segments (if any). Like end_that_request_first(),
3468 * but deals with bytes instead of sectors.
3471 * 0 - we are done with this request, call end_that_request_last()
3472 * 1 - still buffers pending for this request
3474 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3476 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3479 EXPORT_SYMBOL(end_that_request_chunk
);
3482 * splice the completion data to a local structure and hand off to
3483 * process_completion_queue() to complete the requests
3485 static void blk_done_softirq(struct softirq_action
*h
)
3487 struct list_head
*cpu_list
, local_list
;
3489 local_irq_disable();
3490 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3491 list_replace_init(cpu_list
, &local_list
);
3494 while (!list_empty(&local_list
)) {
3495 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3497 list_del_init(&rq
->donelist
);
3498 rq
->q
->softirq_done_fn(rq
);
3502 static int blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3506 * If a CPU goes away, splice its entries to the current CPU
3507 * and trigger a run of the softirq
3509 if (action
== CPU_DEAD
) {
3510 int cpu
= (unsigned long) hcpu
;
3512 local_irq_disable();
3513 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3514 &__get_cpu_var(blk_cpu_done
));
3515 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3523 static struct notifier_block __devinitdata blk_cpu_notifier
= {
3524 .notifier_call
= blk_cpu_notify
,
3528 * blk_complete_request - end I/O on a request
3529 * @req: the request being processed
3532 * Ends all I/O on a request. It does not handle partial completions,
3533 * unless the driver actually implements this in its completion callback
3534 * through requeueing. Theh actual completion happens out-of-order,
3535 * through a softirq handler. The user must have registered a completion
3536 * callback through blk_queue_softirq_done().
3539 void blk_complete_request(struct request
*req
)
3541 struct list_head
*cpu_list
;
3542 unsigned long flags
;
3544 BUG_ON(!req
->q
->softirq_done_fn
);
3546 local_irq_save(flags
);
3548 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3549 list_add_tail(&req
->donelist
, cpu_list
);
3550 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3552 local_irq_restore(flags
);
3555 EXPORT_SYMBOL(blk_complete_request
);
3558 * queue lock must be held
3560 void end_that_request_last(struct request
*req
, int uptodate
)
3562 struct gendisk
*disk
= req
->rq_disk
;
3566 * extend uptodate bool to allow < 0 value to be direct io error
3569 if (end_io_error(uptodate
))
3570 error
= !uptodate
? -EIO
: uptodate
;
3572 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3573 laptop_io_completion();
3576 * Account IO completion. bar_rq isn't accounted as a normal
3577 * IO on queueing nor completion. Accounting the containing
3578 * request is enough.
3580 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3581 unsigned long duration
= jiffies
- req
->start_time
;
3582 const int rw
= rq_data_dir(req
);
3584 __disk_stat_inc(disk
, ios
[rw
]);
3585 __disk_stat_add(disk
, ticks
[rw
], duration
);
3586 disk_round_stats(disk
);
3590 req
->end_io(req
, error
);
3592 __blk_put_request(req
->q
, req
);
3595 EXPORT_SYMBOL(end_that_request_last
);
3597 void end_request(struct request
*req
, int uptodate
)
3599 if (!end_that_request_first(req
, uptodate
, req
->hard_cur_sectors
)) {
3600 add_disk_randomness(req
->rq_disk
);
3601 blkdev_dequeue_request(req
);
3602 end_that_request_last(req
, uptodate
);
3606 EXPORT_SYMBOL(end_request
);
3608 void blk_rq_bio_prep(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
)
3610 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3611 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3613 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3614 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3615 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3616 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3617 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3618 rq
->buffer
= bio_data(bio
);
3619 rq
->data_len
= bio
->bi_size
;
3621 rq
->bio
= rq
->biotail
= bio
;
3624 EXPORT_SYMBOL(blk_rq_bio_prep
);
3626 int kblockd_schedule_work(struct work_struct
*work
)
3628 return queue_work(kblockd_workqueue
, work
);
3631 EXPORT_SYMBOL(kblockd_schedule_work
);
3633 void kblockd_flush(void)
3635 flush_workqueue(kblockd_workqueue
);
3637 EXPORT_SYMBOL(kblockd_flush
);
3639 int __init
blk_dev_init(void)
3643 kblockd_workqueue
= create_workqueue("kblockd");
3644 if (!kblockd_workqueue
)
3645 panic("Failed to create kblockd\n");
3647 request_cachep
= kmem_cache_create("blkdev_requests",
3648 sizeof(struct request
), 0, SLAB_PANIC
, NULL
, NULL
);
3650 requestq_cachep
= kmem_cache_create("blkdev_queue",
3651 sizeof(request_queue_t
), 0, SLAB_PANIC
, NULL
, NULL
);
3653 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3654 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
, NULL
);
3656 for_each_possible_cpu(i
)
3657 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3659 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3660 register_hotcpu_notifier(&blk_cpu_notifier
);
3662 blk_max_low_pfn
= max_low_pfn
;
3663 blk_max_pfn
= max_pfn
;
3669 * IO Context helper functions
3671 void put_io_context(struct io_context
*ioc
)
3676 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3678 if (atomic_dec_and_test(&ioc
->refcount
)) {
3679 struct cfq_io_context
*cic
;
3682 if (ioc
->aic
&& ioc
->aic
->dtor
)
3683 ioc
->aic
->dtor(ioc
->aic
);
3684 if (ioc
->cic_root
.rb_node
!= NULL
) {
3685 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3687 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3692 kmem_cache_free(iocontext_cachep
, ioc
);
3695 EXPORT_SYMBOL(put_io_context
);
3697 /* Called by the exitting task */
3698 void exit_io_context(void)
3700 struct io_context
*ioc
;
3701 struct cfq_io_context
*cic
;
3704 ioc
= current
->io_context
;
3705 current
->io_context
= NULL
;
3706 task_unlock(current
);
3709 if (ioc
->aic
&& ioc
->aic
->exit
)
3710 ioc
->aic
->exit(ioc
->aic
);
3711 if (ioc
->cic_root
.rb_node
!= NULL
) {
3712 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3716 put_io_context(ioc
);
3720 * If the current task has no IO context then create one and initialise it.
3721 * Otherwise, return its existing IO context.
3723 * This returned IO context doesn't have a specifically elevated refcount,
3724 * but since the current task itself holds a reference, the context can be
3725 * used in general code, so long as it stays within `current` context.
3727 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3729 struct task_struct
*tsk
= current
;
3730 struct io_context
*ret
;
3732 ret
= tsk
->io_context
;
3736 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3738 atomic_set(&ret
->refcount
, 1);
3739 ret
->task
= current
;
3740 ret
->ioprio_changed
= 0;
3741 ret
->last_waited
= jiffies
; /* doesn't matter... */
3742 ret
->nr_batch_requests
= 0; /* because this is 0 */
3744 ret
->cic_root
.rb_node
= NULL
;
3745 /* make sure set_task_ioprio() sees the settings above */
3747 tsk
->io_context
= ret
;
3752 EXPORT_SYMBOL(current_io_context
);
3755 * If the current task has no IO context then create one and initialise it.
3756 * If it does have a context, take a ref on it.
3758 * This is always called in the context of the task which submitted the I/O.
3760 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3762 struct io_context
*ret
;
3763 ret
= current_io_context(gfp_flags
, node
);
3765 atomic_inc(&ret
->refcount
);
3768 EXPORT_SYMBOL(get_io_context
);
3770 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3772 struct io_context
*src
= *psrc
;
3773 struct io_context
*dst
= *pdst
;
3776 BUG_ON(atomic_read(&src
->refcount
) == 0);
3777 atomic_inc(&src
->refcount
);
3778 put_io_context(dst
);
3782 EXPORT_SYMBOL(copy_io_context
);
3784 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3786 struct io_context
*temp
;
3791 EXPORT_SYMBOL(swap_io_context
);
3796 struct queue_sysfs_entry
{
3797 struct attribute attr
;
3798 ssize_t (*show
)(struct request_queue
*, char *);
3799 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3803 queue_var_show(unsigned int var
, char *page
)
3805 return sprintf(page
, "%d\n", var
);
3809 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3811 char *p
= (char *) page
;
3813 *var
= simple_strtoul(p
, &p
, 10);
3817 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3819 return queue_var_show(q
->nr_requests
, (page
));
3823 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3825 struct request_list
*rl
= &q
->rq
;
3827 int ret
= queue_var_store(&nr
, page
, count
);
3828 if (nr
< BLKDEV_MIN_RQ
)
3831 spin_lock_irq(q
->queue_lock
);
3832 q
->nr_requests
= nr
;
3833 blk_queue_congestion_threshold(q
);
3835 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3836 blk_set_queue_congested(q
, READ
);
3837 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3838 blk_clear_queue_congested(q
, READ
);
3840 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3841 blk_set_queue_congested(q
, WRITE
);
3842 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3843 blk_clear_queue_congested(q
, WRITE
);
3845 if (rl
->count
[READ
] >= q
->nr_requests
) {
3846 blk_set_queue_full(q
, READ
);
3847 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3848 blk_clear_queue_full(q
, READ
);
3849 wake_up(&rl
->wait
[READ
]);
3852 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3853 blk_set_queue_full(q
, WRITE
);
3854 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
3855 blk_clear_queue_full(q
, WRITE
);
3856 wake_up(&rl
->wait
[WRITE
]);
3858 spin_unlock_irq(q
->queue_lock
);
3862 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
3864 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3866 return queue_var_show(ra_kb
, (page
));
3870 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
3872 unsigned long ra_kb
;
3873 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
3875 spin_lock_irq(q
->queue_lock
);
3876 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
3877 spin_unlock_irq(q
->queue_lock
);
3882 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
3884 int max_sectors_kb
= q
->max_sectors
>> 1;
3886 return queue_var_show(max_sectors_kb
, (page
));
3890 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
3892 unsigned long max_sectors_kb
,
3893 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
3894 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
3895 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
3898 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
3901 * Take the queue lock to update the readahead and max_sectors
3902 * values synchronously:
3904 spin_lock_irq(q
->queue_lock
);
3906 * Trim readahead window as well, if necessary:
3908 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
3909 if (ra_kb
> max_sectors_kb
)
3910 q
->backing_dev_info
.ra_pages
=
3911 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
3913 q
->max_sectors
= max_sectors_kb
<< 1;
3914 spin_unlock_irq(q
->queue_lock
);
3919 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
3921 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
3923 return queue_var_show(max_hw_sectors_kb
, (page
));
3927 static struct queue_sysfs_entry queue_requests_entry
= {
3928 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
3929 .show
= queue_requests_show
,
3930 .store
= queue_requests_store
,
3933 static struct queue_sysfs_entry queue_ra_entry
= {
3934 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
3935 .show
= queue_ra_show
,
3936 .store
= queue_ra_store
,
3939 static struct queue_sysfs_entry queue_max_sectors_entry
= {
3940 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
3941 .show
= queue_max_sectors_show
,
3942 .store
= queue_max_sectors_store
,
3945 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
3946 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
3947 .show
= queue_max_hw_sectors_show
,
3950 static struct queue_sysfs_entry queue_iosched_entry
= {
3951 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
3952 .show
= elv_iosched_show
,
3953 .store
= elv_iosched_store
,
3956 static struct attribute
*default_attrs
[] = {
3957 &queue_requests_entry
.attr
,
3958 &queue_ra_entry
.attr
,
3959 &queue_max_hw_sectors_entry
.attr
,
3960 &queue_max_sectors_entry
.attr
,
3961 &queue_iosched_entry
.attr
,
3965 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3968 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3970 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3971 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
3976 mutex_lock(&q
->sysfs_lock
);
3977 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
3978 mutex_unlock(&q
->sysfs_lock
);
3981 res
= entry
->show(q
, page
);
3982 mutex_unlock(&q
->sysfs_lock
);
3987 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3988 const char *page
, size_t length
)
3990 struct queue_sysfs_entry
*entry
= to_queue(attr
);
3991 request_queue_t
*q
= container_of(kobj
, struct request_queue
, kobj
);
3997 mutex_lock(&q
->sysfs_lock
);
3998 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
3999 mutex_unlock(&q
->sysfs_lock
);
4002 res
= entry
->store(q
, page
, length
);
4003 mutex_unlock(&q
->sysfs_lock
);
4007 static struct sysfs_ops queue_sysfs_ops
= {
4008 .show
= queue_attr_show
,
4009 .store
= queue_attr_store
,
4012 static struct kobj_type queue_ktype
= {
4013 .sysfs_ops
= &queue_sysfs_ops
,
4014 .default_attrs
= default_attrs
,
4015 .release
= blk_release_queue
,
4018 int blk_register_queue(struct gendisk
*disk
)
4022 request_queue_t
*q
= disk
->queue
;
4024 if (!q
|| !q
->request_fn
)
4027 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4029 ret
= kobject_add(&q
->kobj
);
4033 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4035 ret
= elv_register_queue(q
);
4037 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4038 kobject_del(&q
->kobj
);
4045 void blk_unregister_queue(struct gendisk
*disk
)
4047 request_queue_t
*q
= disk
->queue
;
4049 if (q
&& q
->request_fn
) {
4050 elv_unregister_queue(q
);
4052 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4053 kobject_del(&q
->kobj
);
4054 kobject_put(&disk
->kobj
);