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/task_io_accounting_ops.h>
29 #include <linux/interrupt.h>
30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h>
32 #include <linux/fault-inject.h>
37 #include <scsi/scsi_cmnd.h>
39 static void blk_unplug_work(struct work_struct
*work
);
40 static void blk_unplug_timeout(unsigned long data
);
41 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
);
42 static void init_request_from_bio(struct request
*req
, struct bio
*bio
);
43 static int __make_request(struct request_queue
*q
, struct bio
*bio
);
44 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
);
45 static void blk_recalc_rq_segments(struct request
*rq
);
46 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
50 * For the allocated request tables
52 static struct kmem_cache
*request_cachep
;
55 * For queue allocation
57 static struct kmem_cache
*requestq_cachep
;
60 * For io context allocations
62 static struct kmem_cache
*iocontext_cachep
;
65 * Controlling structure to kblockd
67 static struct workqueue_struct
*kblockd_workqueue
;
69 unsigned long blk_max_low_pfn
, blk_max_pfn
;
71 EXPORT_SYMBOL(blk_max_low_pfn
);
72 EXPORT_SYMBOL(blk_max_pfn
);
74 static DEFINE_PER_CPU(struct list_head
, blk_cpu_done
);
76 /* Amount of time in which a process may batch requests */
77 #define BLK_BATCH_TIME (HZ/50UL)
79 /* Number of requests a "batching" process may submit */
80 #define BLK_BATCH_REQ 32
83 * Return the threshold (number of used requests) at which the queue is
84 * considered to be congested. It include a little hysteresis to keep the
85 * context switch rate down.
87 static inline int queue_congestion_on_threshold(struct request_queue
*q
)
89 return q
->nr_congestion_on
;
93 * The threshold at which a queue is considered to be uncongested
95 static inline int queue_congestion_off_threshold(struct request_queue
*q
)
97 return q
->nr_congestion_off
;
100 static void blk_queue_congestion_threshold(struct request_queue
*q
)
104 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
105 if (nr
> q
->nr_requests
)
107 q
->nr_congestion_on
= nr
;
109 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
112 q
->nr_congestion_off
= nr
;
116 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
119 * Locates the passed device's request queue and returns the address of its
122 * Will return NULL if the request queue cannot be located.
124 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
126 struct backing_dev_info
*ret
= NULL
;
127 struct request_queue
*q
= bdev_get_queue(bdev
);
130 ret
= &q
->backing_dev_info
;
133 EXPORT_SYMBOL(blk_get_backing_dev_info
);
136 * blk_queue_prep_rq - set a prepare_request function for queue
138 * @pfn: prepare_request function
140 * It's possible for a queue to register a prepare_request callback which
141 * is invoked before the request is handed to the request_fn. The goal of
142 * the function is to prepare a request for I/O, it can be used to build a
143 * cdb from the request data for instance.
146 void blk_queue_prep_rq(struct request_queue
*q
, prep_rq_fn
*pfn
)
151 EXPORT_SYMBOL(blk_queue_prep_rq
);
154 * blk_queue_merge_bvec - set a merge_bvec function for queue
156 * @mbfn: merge_bvec_fn
158 * Usually queues have static limitations on the max sectors or segments that
159 * we can put in a request. Stacking drivers may have some settings that
160 * are dynamic, and thus we have to query the queue whether it is ok to
161 * add a new bio_vec to a bio at a given offset or not. If the block device
162 * has such limitations, it needs to register a merge_bvec_fn to control
163 * the size of bio's sent to it. Note that a block device *must* allow a
164 * single page to be added to an empty bio. The block device driver may want
165 * to use the bio_split() function to deal with these bio's. By default
166 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
169 void blk_queue_merge_bvec(struct request_queue
*q
, merge_bvec_fn
*mbfn
)
171 q
->merge_bvec_fn
= mbfn
;
174 EXPORT_SYMBOL(blk_queue_merge_bvec
);
176 void blk_queue_softirq_done(struct request_queue
*q
, softirq_done_fn
*fn
)
178 q
->softirq_done_fn
= fn
;
181 EXPORT_SYMBOL(blk_queue_softirq_done
);
184 * blk_queue_make_request - define an alternate make_request function for a device
185 * @q: the request queue for the device to be affected
186 * @mfn: the alternate make_request function
189 * The normal way for &struct bios to be passed to a device
190 * driver is for them to be collected into requests on a request
191 * queue, and then to allow the device driver to select requests
192 * off that queue when it is ready. This works well for many block
193 * devices. However some block devices (typically virtual devices
194 * such as md or lvm) do not benefit from the processing on the
195 * request queue, and are served best by having the requests passed
196 * directly to them. This can be achieved by providing a function
197 * to blk_queue_make_request().
200 * The driver that does this *must* be able to deal appropriately
201 * with buffers in "highmemory". This can be accomplished by either calling
202 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
203 * blk_queue_bounce() to create a buffer in normal memory.
205 void blk_queue_make_request(struct request_queue
* q
, make_request_fn
* mfn
)
210 q
->nr_requests
= BLKDEV_MAX_RQ
;
211 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
212 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
213 q
->make_request_fn
= mfn
;
214 q
->backing_dev_info
.ra_pages
= (VM_MAX_READAHEAD
* 1024) / PAGE_CACHE_SIZE
;
215 q
->backing_dev_info
.state
= 0;
216 q
->backing_dev_info
.capabilities
= BDI_CAP_MAP_COPY
;
217 blk_queue_max_sectors(q
, SAFE_MAX_SECTORS
);
218 blk_queue_hardsect_size(q
, 512);
219 blk_queue_dma_alignment(q
, 511);
220 blk_queue_congestion_threshold(q
);
221 q
->nr_batching
= BLK_BATCH_REQ
;
223 q
->unplug_thresh
= 4; /* hmm */
224 q
->unplug_delay
= (3 * HZ
) / 1000; /* 3 milliseconds */
225 if (q
->unplug_delay
== 0)
228 INIT_WORK(&q
->unplug_work
, blk_unplug_work
);
230 q
->unplug_timer
.function
= blk_unplug_timeout
;
231 q
->unplug_timer
.data
= (unsigned long)q
;
234 * by default assume old behaviour and bounce for any highmem page
236 blk_queue_bounce_limit(q
, BLK_BOUNCE_HIGH
);
239 EXPORT_SYMBOL(blk_queue_make_request
);
241 static void rq_init(struct request_queue
*q
, struct request
*rq
)
243 INIT_LIST_HEAD(&rq
->queuelist
);
244 INIT_LIST_HEAD(&rq
->donelist
);
247 rq
->bio
= rq
->biotail
= NULL
;
248 INIT_HLIST_NODE(&rq
->hash
);
249 RB_CLEAR_NODE(&rq
->rb_node
);
257 rq
->nr_phys_segments
= 0;
260 rq
->end_io_data
= NULL
;
261 rq
->completion_data
= NULL
;
266 * blk_queue_ordered - does this queue support ordered writes
267 * @q: the request queue
268 * @ordered: one of QUEUE_ORDERED_*
269 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
272 * For journalled file systems, doing ordered writes on a commit
273 * block instead of explicitly doing wait_on_buffer (which is bad
274 * for performance) can be a big win. Block drivers supporting this
275 * feature should call this function and indicate so.
278 int blk_queue_ordered(struct request_queue
*q
, unsigned ordered
,
279 prepare_flush_fn
*prepare_flush_fn
)
281 if (ordered
& (QUEUE_ORDERED_PREFLUSH
| QUEUE_ORDERED_POSTFLUSH
) &&
282 prepare_flush_fn
== NULL
) {
283 printk(KERN_ERR
"blk_queue_ordered: prepare_flush_fn required\n");
287 if (ordered
!= QUEUE_ORDERED_NONE
&&
288 ordered
!= QUEUE_ORDERED_DRAIN
&&
289 ordered
!= QUEUE_ORDERED_DRAIN_FLUSH
&&
290 ordered
!= QUEUE_ORDERED_DRAIN_FUA
&&
291 ordered
!= QUEUE_ORDERED_TAG
&&
292 ordered
!= QUEUE_ORDERED_TAG_FLUSH
&&
293 ordered
!= QUEUE_ORDERED_TAG_FUA
) {
294 printk(KERN_ERR
"blk_queue_ordered: bad value %d\n", ordered
);
298 q
->ordered
= ordered
;
299 q
->next_ordered
= ordered
;
300 q
->prepare_flush_fn
= prepare_flush_fn
;
305 EXPORT_SYMBOL(blk_queue_ordered
);
308 * Cache flushing for ordered writes handling
310 inline unsigned blk_ordered_cur_seq(struct request_queue
*q
)
314 return 1 << ffz(q
->ordseq
);
317 unsigned blk_ordered_req_seq(struct request
*rq
)
319 struct request_queue
*q
= rq
->q
;
321 BUG_ON(q
->ordseq
== 0);
323 if (rq
== &q
->pre_flush_rq
)
324 return QUEUE_ORDSEQ_PREFLUSH
;
325 if (rq
== &q
->bar_rq
)
326 return QUEUE_ORDSEQ_BAR
;
327 if (rq
== &q
->post_flush_rq
)
328 return QUEUE_ORDSEQ_POSTFLUSH
;
331 * !fs requests don't need to follow barrier ordering. Always
332 * put them at the front. This fixes the following deadlock.
334 * http://thread.gmane.org/gmane.linux.kernel/537473
336 if (!blk_fs_request(rq
))
337 return QUEUE_ORDSEQ_DRAIN
;
339 if ((rq
->cmd_flags
& REQ_ORDERED_COLOR
) ==
340 (q
->orig_bar_rq
->cmd_flags
& REQ_ORDERED_COLOR
))
341 return QUEUE_ORDSEQ_DRAIN
;
343 return QUEUE_ORDSEQ_DONE
;
346 void blk_ordered_complete_seq(struct request_queue
*q
, unsigned seq
, int error
)
351 if (error
&& !q
->orderr
)
354 BUG_ON(q
->ordseq
& seq
);
357 if (blk_ordered_cur_seq(q
) != QUEUE_ORDSEQ_DONE
)
361 * Okay, sequence complete.
365 uptodate
= q
->orderr
;
370 end_that_request_first(rq
, uptodate
, rq
->hard_nr_sectors
);
371 end_that_request_last(rq
, uptodate
);
374 static void pre_flush_end_io(struct request
*rq
, int error
)
376 elv_completed_request(rq
->q
, rq
);
377 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_PREFLUSH
, error
);
380 static void bar_end_io(struct request
*rq
, int error
)
382 elv_completed_request(rq
->q
, rq
);
383 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_BAR
, error
);
386 static void post_flush_end_io(struct request
*rq
, int error
)
388 elv_completed_request(rq
->q
, rq
);
389 blk_ordered_complete_seq(rq
->q
, QUEUE_ORDSEQ_POSTFLUSH
, error
);
392 static void queue_flush(struct request_queue
*q
, unsigned which
)
395 rq_end_io_fn
*end_io
;
397 if (which
== QUEUE_ORDERED_PREFLUSH
) {
398 rq
= &q
->pre_flush_rq
;
399 end_io
= pre_flush_end_io
;
401 rq
= &q
->post_flush_rq
;
402 end_io
= post_flush_end_io
;
405 rq
->cmd_flags
= REQ_HARDBARRIER
;
407 rq
->elevator_private
= NULL
;
408 rq
->elevator_private2
= NULL
;
409 rq
->rq_disk
= q
->bar_rq
.rq_disk
;
411 q
->prepare_flush_fn(q
, rq
);
413 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
416 static inline struct request
*start_ordered(struct request_queue
*q
,
420 q
->ordered
= q
->next_ordered
;
421 q
->ordseq
|= QUEUE_ORDSEQ_STARTED
;
424 * Prep proxy barrier request.
426 blkdev_dequeue_request(rq
);
431 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
432 rq
->cmd_flags
|= REQ_RW
;
433 if (q
->ordered
& QUEUE_ORDERED_FUA
)
434 rq
->cmd_flags
|= REQ_FUA
;
435 rq
->elevator_private
= NULL
;
436 rq
->elevator_private2
= NULL
;
437 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
438 rq
->end_io
= bar_end_io
;
441 * Queue ordered sequence. As we stack them at the head, we
442 * need to queue in reverse order. Note that we rely on that
443 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
444 * request gets inbetween ordered sequence. If this request is
445 * an empty barrier, we don't need to do a postflush ever since
446 * there will be no data written between the pre and post flush.
447 * Hence a single flush will suffice.
449 if ((q
->ordered
& QUEUE_ORDERED_POSTFLUSH
) && !blk_empty_barrier(rq
))
450 queue_flush(q
, QUEUE_ORDERED_POSTFLUSH
);
452 q
->ordseq
|= QUEUE_ORDSEQ_POSTFLUSH
;
454 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
456 if (q
->ordered
& QUEUE_ORDERED_PREFLUSH
) {
457 queue_flush(q
, QUEUE_ORDERED_PREFLUSH
);
458 rq
= &q
->pre_flush_rq
;
460 q
->ordseq
|= QUEUE_ORDSEQ_PREFLUSH
;
462 if ((q
->ordered
& QUEUE_ORDERED_TAG
) || q
->in_flight
== 0)
463 q
->ordseq
|= QUEUE_ORDSEQ_DRAIN
;
470 int blk_do_ordered(struct request_queue
*q
, struct request
**rqp
)
472 struct request
*rq
= *rqp
;
473 const int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
479 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
) {
480 *rqp
= start_ordered(q
, rq
);
484 * This can happen when the queue switches to
485 * ORDERED_NONE while this request is on it.
487 blkdev_dequeue_request(rq
);
488 end_that_request_first(rq
, -EOPNOTSUPP
,
489 rq
->hard_nr_sectors
);
490 end_that_request_last(rq
, -EOPNOTSUPP
);
497 * Ordered sequence in progress
500 /* Special requests are not subject to ordering rules. */
501 if (!blk_fs_request(rq
) &&
502 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
505 if (q
->ordered
& QUEUE_ORDERED_TAG
) {
506 /* Ordered by tag. Blocking the next barrier is enough. */
507 if (is_barrier
&& rq
!= &q
->bar_rq
)
510 /* Ordered by draining. Wait for turn. */
511 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
512 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
519 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
520 unsigned int nbytes
, int error
)
522 struct request_queue
*q
= rq
->q
;
524 if (&q
->bar_rq
!= rq
) {
526 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
527 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
530 if (unlikely(nbytes
> bio
->bi_size
)) {
531 printk("%s: want %u bytes done, only %u left\n",
532 __FUNCTION__
, nbytes
, bio
->bi_size
);
533 nbytes
= bio
->bi_size
;
536 bio
->bi_size
-= nbytes
;
537 bio
->bi_sector
+= (nbytes
>> 9);
538 if (bio
->bi_size
== 0)
539 bio_endio(bio
, error
);
543 * Okay, this is the barrier request in progress, just
546 if (error
&& !q
->orderr
)
552 * blk_queue_bounce_limit - set bounce buffer limit for queue
553 * @q: the request queue for the device
554 * @dma_addr: bus address limit
557 * Different hardware can have different requirements as to what pages
558 * it can do I/O directly to. A low level driver can call
559 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
560 * buffers for doing I/O to pages residing above @page.
562 void blk_queue_bounce_limit(struct request_queue
*q
, u64 dma_addr
)
564 unsigned long bounce_pfn
= dma_addr
>> PAGE_SHIFT
;
567 q
->bounce_gfp
= GFP_NOIO
;
568 #if BITS_PER_LONG == 64
569 /* Assume anything <= 4GB can be handled by IOMMU.
570 Actually some IOMMUs can handle everything, but I don't
571 know of a way to test this here. */
572 if (bounce_pfn
< (min_t(u64
,0xffffffff,BLK_BOUNCE_HIGH
) >> PAGE_SHIFT
))
574 q
->bounce_pfn
= max_low_pfn
;
576 if (bounce_pfn
< blk_max_low_pfn
)
578 q
->bounce_pfn
= bounce_pfn
;
581 init_emergency_isa_pool();
582 q
->bounce_gfp
= GFP_NOIO
| GFP_DMA
;
583 q
->bounce_pfn
= bounce_pfn
;
587 EXPORT_SYMBOL(blk_queue_bounce_limit
);
590 * blk_queue_max_sectors - set max sectors for a request for this queue
591 * @q: the request queue for the device
592 * @max_sectors: max sectors in the usual 512b unit
595 * Enables a low level driver to set an upper limit on the size of
598 void blk_queue_max_sectors(struct request_queue
*q
, unsigned int max_sectors
)
600 if ((max_sectors
<< 9) < PAGE_CACHE_SIZE
) {
601 max_sectors
= 1 << (PAGE_CACHE_SHIFT
- 9);
602 printk("%s: set to minimum %d\n", __FUNCTION__
, max_sectors
);
605 if (BLK_DEF_MAX_SECTORS
> max_sectors
)
606 q
->max_hw_sectors
= q
->max_sectors
= max_sectors
;
608 q
->max_sectors
= BLK_DEF_MAX_SECTORS
;
609 q
->max_hw_sectors
= max_sectors
;
613 EXPORT_SYMBOL(blk_queue_max_sectors
);
616 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
617 * @q: the request queue for the device
618 * @max_segments: max number of segments
621 * Enables a low level driver to set an upper limit on the number of
622 * physical data segments in a request. This would be the largest sized
623 * scatter list the driver could handle.
625 void blk_queue_max_phys_segments(struct request_queue
*q
,
626 unsigned short max_segments
)
630 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
633 q
->max_phys_segments
= max_segments
;
636 EXPORT_SYMBOL(blk_queue_max_phys_segments
);
639 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
640 * @q: the request queue for the device
641 * @max_segments: max number of segments
644 * Enables a low level driver to set an upper limit on the number of
645 * hw data segments in a request. This would be the largest number of
646 * address/length pairs the host adapter can actually give as once
649 void blk_queue_max_hw_segments(struct request_queue
*q
,
650 unsigned short max_segments
)
654 printk("%s: set to minimum %d\n", __FUNCTION__
, max_segments
);
657 q
->max_hw_segments
= max_segments
;
660 EXPORT_SYMBOL(blk_queue_max_hw_segments
);
663 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
664 * @q: the request queue for the device
665 * @max_size: max size of segment in bytes
668 * Enables a low level driver to set an upper limit on the size of a
671 void blk_queue_max_segment_size(struct request_queue
*q
, unsigned int max_size
)
673 if (max_size
< PAGE_CACHE_SIZE
) {
674 max_size
= PAGE_CACHE_SIZE
;
675 printk("%s: set to minimum %d\n", __FUNCTION__
, max_size
);
678 q
->max_segment_size
= max_size
;
681 EXPORT_SYMBOL(blk_queue_max_segment_size
);
684 * blk_queue_hardsect_size - set hardware sector size for the queue
685 * @q: the request queue for the device
686 * @size: the hardware sector size, in bytes
689 * This should typically be set to the lowest possible sector size
690 * that the hardware can operate on (possible without reverting to
691 * even internal read-modify-write operations). Usually the default
692 * of 512 covers most hardware.
694 void blk_queue_hardsect_size(struct request_queue
*q
, unsigned short size
)
696 q
->hardsect_size
= size
;
699 EXPORT_SYMBOL(blk_queue_hardsect_size
);
702 * Returns the minimum that is _not_ zero, unless both are zero.
704 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
707 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
708 * @t: the stacking driver (top)
709 * @b: the underlying device (bottom)
711 void blk_queue_stack_limits(struct request_queue
*t
, struct request_queue
*b
)
713 /* zero is "infinity" */
714 t
->max_sectors
= min_not_zero(t
->max_sectors
,b
->max_sectors
);
715 t
->max_hw_sectors
= min_not_zero(t
->max_hw_sectors
,b
->max_hw_sectors
);
717 t
->max_phys_segments
= min(t
->max_phys_segments
,b
->max_phys_segments
);
718 t
->max_hw_segments
= min(t
->max_hw_segments
,b
->max_hw_segments
);
719 t
->max_segment_size
= min(t
->max_segment_size
,b
->max_segment_size
);
720 t
->hardsect_size
= max(t
->hardsect_size
,b
->hardsect_size
);
721 if (!test_bit(QUEUE_FLAG_CLUSTER
, &b
->queue_flags
))
722 clear_bit(QUEUE_FLAG_CLUSTER
, &t
->queue_flags
);
725 EXPORT_SYMBOL(blk_queue_stack_limits
);
728 * blk_queue_segment_boundary - set boundary rules for segment merging
729 * @q: the request queue for the device
730 * @mask: the memory boundary mask
732 void blk_queue_segment_boundary(struct request_queue
*q
, unsigned long mask
)
734 if (mask
< PAGE_CACHE_SIZE
- 1) {
735 mask
= PAGE_CACHE_SIZE
- 1;
736 printk("%s: set to minimum %lx\n", __FUNCTION__
, mask
);
739 q
->seg_boundary_mask
= mask
;
742 EXPORT_SYMBOL(blk_queue_segment_boundary
);
745 * blk_queue_dma_alignment - set dma length and memory alignment
746 * @q: the request queue for the device
747 * @mask: alignment mask
750 * set required memory and length aligment for direct dma transactions.
751 * this is used when buiding direct io requests for the queue.
754 void blk_queue_dma_alignment(struct request_queue
*q
, int mask
)
756 q
->dma_alignment
= mask
;
759 EXPORT_SYMBOL(blk_queue_dma_alignment
);
762 * blk_queue_find_tag - find a request by its tag and queue
763 * @q: The request queue for the device
764 * @tag: The tag of the request
767 * Should be used when a device returns a tag and you want to match
770 * no locks need be held.
772 struct request
*blk_queue_find_tag(struct request_queue
*q
, int tag
)
774 return blk_map_queue_find_tag(q
->queue_tags
, tag
);
777 EXPORT_SYMBOL(blk_queue_find_tag
);
780 * __blk_free_tags - release a given set of tag maintenance info
781 * @bqt: the tag map to free
783 * Tries to free the specified @bqt@. Returns true if it was
784 * actually freed and false if there are still references using it
786 static int __blk_free_tags(struct blk_queue_tag
*bqt
)
790 retval
= atomic_dec_and_test(&bqt
->refcnt
);
793 BUG_ON(!list_empty(&bqt
->busy_list
));
795 kfree(bqt
->tag_index
);
796 bqt
->tag_index
= NULL
;
809 * __blk_queue_free_tags - release tag maintenance info
810 * @q: the request queue for the device
813 * blk_cleanup_queue() will take care of calling this function, if tagging
814 * has been used. So there's no need to call this directly.
816 static void __blk_queue_free_tags(struct request_queue
*q
)
818 struct blk_queue_tag
*bqt
= q
->queue_tags
;
823 __blk_free_tags(bqt
);
825 q
->queue_tags
= NULL
;
826 q
->queue_flags
&= ~(1 << QUEUE_FLAG_QUEUED
);
831 * blk_free_tags - release a given set of tag maintenance info
832 * @bqt: the tag map to free
834 * For externally managed @bqt@ frees the map. Callers of this
835 * function must guarantee to have released all the queues that
836 * might have been using this tag map.
838 void blk_free_tags(struct blk_queue_tag
*bqt
)
840 if (unlikely(!__blk_free_tags(bqt
)))
843 EXPORT_SYMBOL(blk_free_tags
);
846 * blk_queue_free_tags - release tag maintenance info
847 * @q: the request queue for the device
850 * This is used to disabled tagged queuing to a device, yet leave
853 void blk_queue_free_tags(struct request_queue
*q
)
855 clear_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
858 EXPORT_SYMBOL(blk_queue_free_tags
);
861 init_tag_map(struct request_queue
*q
, struct blk_queue_tag
*tags
, int depth
)
863 struct request
**tag_index
;
864 unsigned long *tag_map
;
867 if (q
&& depth
> q
->nr_requests
* 2) {
868 depth
= q
->nr_requests
* 2;
869 printk(KERN_ERR
"%s: adjusted depth to %d\n",
870 __FUNCTION__
, depth
);
873 tag_index
= kzalloc(depth
* sizeof(struct request
*), GFP_ATOMIC
);
877 nr_ulongs
= ALIGN(depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
878 tag_map
= kzalloc(nr_ulongs
* sizeof(unsigned long), GFP_ATOMIC
);
882 tags
->real_max_depth
= depth
;
883 tags
->max_depth
= depth
;
884 tags
->tag_index
= tag_index
;
885 tags
->tag_map
= tag_map
;
893 static struct blk_queue_tag
*__blk_queue_init_tags(struct request_queue
*q
,
896 struct blk_queue_tag
*tags
;
898 tags
= kmalloc(sizeof(struct blk_queue_tag
), GFP_ATOMIC
);
902 if (init_tag_map(q
, tags
, depth
))
905 INIT_LIST_HEAD(&tags
->busy_list
);
907 atomic_set(&tags
->refcnt
, 1);
915 * blk_init_tags - initialize the tag info for an external tag map
916 * @depth: the maximum queue depth supported
917 * @tags: the tag to use
919 struct blk_queue_tag
*blk_init_tags(int depth
)
921 return __blk_queue_init_tags(NULL
, depth
);
923 EXPORT_SYMBOL(blk_init_tags
);
926 * blk_queue_init_tags - initialize the queue tag info
927 * @q: the request queue for the device
928 * @depth: the maximum queue depth supported
929 * @tags: the tag to use
931 int blk_queue_init_tags(struct request_queue
*q
, int depth
,
932 struct blk_queue_tag
*tags
)
936 BUG_ON(tags
&& q
->queue_tags
&& tags
!= q
->queue_tags
);
938 if (!tags
&& !q
->queue_tags
) {
939 tags
= __blk_queue_init_tags(q
, depth
);
943 } else if (q
->queue_tags
) {
944 if ((rc
= blk_queue_resize_tags(q
, depth
)))
946 set_bit(QUEUE_FLAG_QUEUED
, &q
->queue_flags
);
949 atomic_inc(&tags
->refcnt
);
952 * assign it, all done
954 q
->queue_tags
= tags
;
955 q
->queue_flags
|= (1 << QUEUE_FLAG_QUEUED
);
962 EXPORT_SYMBOL(blk_queue_init_tags
);
965 * blk_queue_resize_tags - change the queueing depth
966 * @q: the request queue for the device
967 * @new_depth: the new max command queueing depth
970 * Must be called with the queue lock held.
972 int blk_queue_resize_tags(struct request_queue
*q
, int new_depth
)
974 struct blk_queue_tag
*bqt
= q
->queue_tags
;
975 struct request
**tag_index
;
976 unsigned long *tag_map
;
977 int max_depth
, nr_ulongs
;
983 * if we already have large enough real_max_depth. just
984 * adjust max_depth. *NOTE* as requests with tag value
985 * between new_depth and real_max_depth can be in-flight, tag
986 * map can not be shrunk blindly here.
988 if (new_depth
<= bqt
->real_max_depth
) {
989 bqt
->max_depth
= new_depth
;
994 * Currently cannot replace a shared tag map with a new
995 * one, so error out if this is the case
997 if (atomic_read(&bqt
->refcnt
) != 1)
1001 * save the old state info, so we can copy it back
1003 tag_index
= bqt
->tag_index
;
1004 tag_map
= bqt
->tag_map
;
1005 max_depth
= bqt
->real_max_depth
;
1007 if (init_tag_map(q
, bqt
, new_depth
))
1010 memcpy(bqt
->tag_index
, tag_index
, max_depth
* sizeof(struct request
*));
1011 nr_ulongs
= ALIGN(max_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1012 memcpy(bqt
->tag_map
, tag_map
, nr_ulongs
* sizeof(unsigned long));
1019 EXPORT_SYMBOL(blk_queue_resize_tags
);
1022 * blk_queue_end_tag - end tag operations for a request
1023 * @q: the request queue for the device
1024 * @rq: the request that has completed
1027 * Typically called when end_that_request_first() returns 0, meaning
1028 * all transfers have been done for a request. It's important to call
1029 * this function before end_that_request_last(), as that will put the
1030 * request back on the free list thus corrupting the internal tag list.
1033 * queue lock must be held.
1035 void blk_queue_end_tag(struct request_queue
*q
, struct request
*rq
)
1037 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1042 if (unlikely(tag
>= bqt
->real_max_depth
))
1044 * This can happen after tag depth has been reduced.
1045 * FIXME: how about a warning or info message here?
1049 list_del_init(&rq
->queuelist
);
1050 rq
->cmd_flags
&= ~REQ_QUEUED
;
1053 if (unlikely(bqt
->tag_index
[tag
] == NULL
))
1054 printk(KERN_ERR
"%s: tag %d is missing\n",
1057 bqt
->tag_index
[tag
] = NULL
;
1060 * We use test_and_clear_bit's memory ordering properties here.
1061 * The tag_map bit acts as a lock for tag_index[bit], so we need
1062 * a barrer before clearing the bit (precisely: release semantics).
1063 * Could use clear_bit_unlock when it is merged.
1065 if (unlikely(!test_and_clear_bit(tag
, bqt
->tag_map
))) {
1066 printk(KERN_ERR
"%s: attempt to clear non-busy tag (%d)\n",
1074 EXPORT_SYMBOL(blk_queue_end_tag
);
1077 * blk_queue_start_tag - find a free tag and assign it
1078 * @q: the request queue for the device
1079 * @rq: the block request that needs tagging
1082 * This can either be used as a stand-alone helper, or possibly be
1083 * assigned as the queue &prep_rq_fn (in which case &struct request
1084 * automagically gets a tag assigned). Note that this function
1085 * assumes that any type of request can be queued! if this is not
1086 * true for your device, you must check the request type before
1087 * calling this function. The request will also be removed from
1088 * the request queue, so it's the drivers responsibility to readd
1089 * it if it should need to be restarted for some reason.
1092 * queue lock must be held.
1094 int blk_queue_start_tag(struct request_queue
*q
, struct request
*rq
)
1096 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1099 if (unlikely((rq
->cmd_flags
& REQ_QUEUED
))) {
1101 "%s: request %p for device [%s] already tagged %d",
1103 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->tag
);
1108 * Protect against shared tag maps, as we may not have exclusive
1109 * access to the tag map.
1112 tag
= find_first_zero_bit(bqt
->tag_map
, bqt
->max_depth
);
1113 if (tag
>= bqt
->max_depth
)
1116 } while (test_and_set_bit(tag
, bqt
->tag_map
));
1118 * We rely on test_and_set_bit providing lock memory ordering semantics
1119 * (could use test_and_set_bit_lock when it is merged).
1122 rq
->cmd_flags
|= REQ_QUEUED
;
1124 bqt
->tag_index
[tag
] = rq
;
1125 blkdev_dequeue_request(rq
);
1126 list_add(&rq
->queuelist
, &bqt
->busy_list
);
1131 EXPORT_SYMBOL(blk_queue_start_tag
);
1134 * blk_queue_invalidate_tags - invalidate all pending tags
1135 * @q: the request queue for the device
1138 * Hardware conditions may dictate a need to stop all pending requests.
1139 * In this case, we will safely clear the block side of the tag queue and
1140 * readd all requests to the request queue in the right order.
1143 * queue lock must be held.
1145 void blk_queue_invalidate_tags(struct request_queue
*q
)
1147 struct blk_queue_tag
*bqt
= q
->queue_tags
;
1148 struct list_head
*tmp
, *n
;
1151 list_for_each_safe(tmp
, n
, &bqt
->busy_list
) {
1152 rq
= list_entry_rq(tmp
);
1154 if (rq
->tag
== -1) {
1156 "%s: bad tag found on list\n", __FUNCTION__
);
1157 list_del_init(&rq
->queuelist
);
1158 rq
->cmd_flags
&= ~REQ_QUEUED
;
1160 blk_queue_end_tag(q
, rq
);
1162 rq
->cmd_flags
&= ~REQ_STARTED
;
1163 __elv_add_request(q
, rq
, ELEVATOR_INSERT_BACK
, 0);
1167 EXPORT_SYMBOL(blk_queue_invalidate_tags
);
1169 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
1173 printk("%s: dev %s: type=%x, flags=%x\n", msg
,
1174 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
1177 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq
->sector
,
1179 rq
->current_nr_sectors
);
1180 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq
->bio
, rq
->biotail
, rq
->buffer
, rq
->data
, rq
->data_len
);
1182 if (blk_pc_request(rq
)) {
1184 for (bit
= 0; bit
< sizeof(rq
->cmd
); bit
++)
1185 printk("%02x ", rq
->cmd
[bit
]);
1190 EXPORT_SYMBOL(blk_dump_rq_flags
);
1192 void blk_recount_segments(struct request_queue
*q
, struct bio
*bio
)
1195 struct bio
*nxt
= bio
->bi_next
;
1197 rq
.bio
= rq
.biotail
= bio
;
1198 bio
->bi_next
= NULL
;
1199 blk_recalc_rq_segments(&rq
);
1201 bio
->bi_phys_segments
= rq
.nr_phys_segments
;
1202 bio
->bi_hw_segments
= rq
.nr_hw_segments
;
1203 bio
->bi_flags
|= (1 << BIO_SEG_VALID
);
1205 EXPORT_SYMBOL(blk_recount_segments
);
1207 static void blk_recalc_rq_segments(struct request
*rq
)
1211 unsigned int phys_size
;
1212 unsigned int hw_size
;
1213 struct bio_vec
*bv
, *bvprv
= NULL
;
1217 struct req_iterator iter
;
1218 int high
, highprv
= 1;
1219 struct request_queue
*q
= rq
->q
;
1224 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1225 hw_seg_size
= seg_size
= 0;
1226 phys_size
= hw_size
= nr_phys_segs
= nr_hw_segs
= 0;
1227 rq_for_each_segment(bv
, rq
, iter
) {
1229 * the trick here is making sure that a high page is never
1230 * considered part of another segment, since that might
1231 * change with the bounce page.
1233 high
= page_to_pfn(bv
->bv_page
) > q
->bounce_pfn
;
1234 if (high
|| highprv
)
1235 goto new_hw_segment
;
1237 if (seg_size
+ bv
->bv_len
> q
->max_segment_size
)
1239 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bv
))
1241 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bv
))
1243 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1244 goto new_hw_segment
;
1246 seg_size
+= bv
->bv_len
;
1247 hw_seg_size
+= bv
->bv_len
;
1252 if (BIOVEC_VIRT_MERGEABLE(bvprv
, bv
) &&
1253 !BIOVEC_VIRT_OVERSIZE(hw_seg_size
+ bv
->bv_len
))
1254 hw_seg_size
+= bv
->bv_len
;
1257 if (nr_hw_segs
== 1 &&
1258 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1259 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1260 hw_seg_size
= BIOVEC_VIRT_START_SIZE(bv
) + bv
->bv_len
;
1266 seg_size
= bv
->bv_len
;
1270 if (nr_hw_segs
== 1 &&
1271 hw_seg_size
> rq
->bio
->bi_hw_front_size
)
1272 rq
->bio
->bi_hw_front_size
= hw_seg_size
;
1273 if (hw_seg_size
> rq
->biotail
->bi_hw_back_size
)
1274 rq
->biotail
->bi_hw_back_size
= hw_seg_size
;
1275 rq
->nr_phys_segments
= nr_phys_segs
;
1276 rq
->nr_hw_segments
= nr_hw_segs
;
1279 static int blk_phys_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1282 if (!(q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
)))
1285 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)))
1287 if (bio
->bi_size
+ nxt
->bi_size
> q
->max_segment_size
)
1291 * bio and nxt are contigous in memory, check if the queue allows
1292 * these two to be merged into one
1294 if (BIO_SEG_BOUNDARY(q
, bio
, nxt
))
1300 static int blk_hw_contig_segment(struct request_queue
*q
, struct bio
*bio
,
1303 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1304 blk_recount_segments(q
, bio
);
1305 if (unlikely(!bio_flagged(nxt
, BIO_SEG_VALID
)))
1306 blk_recount_segments(q
, nxt
);
1307 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(nxt
)) ||
1308 BIOVEC_VIRT_OVERSIZE(bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
))
1310 if (bio
->bi_hw_back_size
+ nxt
->bi_hw_front_size
> q
->max_segment_size
)
1317 * map a request to scatterlist, return number of sg entries setup. Caller
1318 * must make sure sg can hold rq->nr_phys_segments entries
1320 int blk_rq_map_sg(struct request_queue
*q
, struct request
*rq
,
1321 struct scatterlist
*sg
)
1323 struct bio_vec
*bvec
, *bvprv
;
1324 struct req_iterator iter
;
1328 cluster
= q
->queue_flags
& (1 << QUEUE_FLAG_CLUSTER
);
1331 * for each bio in rq
1334 rq_for_each_segment(bvec
, rq
, iter
) {
1335 int nbytes
= bvec
->bv_len
;
1337 if (bvprv
&& cluster
) {
1338 if (sg
[nsegs
- 1].length
+ nbytes
> q
->max_segment_size
)
1341 if (!BIOVEC_PHYS_MERGEABLE(bvprv
, bvec
))
1343 if (!BIOVEC_SEG_BOUNDARY(q
, bvprv
, bvec
))
1346 sg
[nsegs
- 1].length
+= nbytes
;
1349 memset(&sg
[nsegs
],0,sizeof(struct scatterlist
));
1350 sg
[nsegs
].page
= bvec
->bv_page
;
1351 sg
[nsegs
].length
= nbytes
;
1352 sg
[nsegs
].offset
= bvec
->bv_offset
;
1357 } /* segments in rq */
1362 EXPORT_SYMBOL(blk_rq_map_sg
);
1365 * the standard queue merge functions, can be overridden with device
1366 * specific ones if so desired
1369 static inline int ll_new_mergeable(struct request_queue
*q
,
1370 struct request
*req
,
1373 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1375 if (req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1376 req
->cmd_flags
|= REQ_NOMERGE
;
1377 if (req
== q
->last_merge
)
1378 q
->last_merge
= NULL
;
1383 * A hw segment is just getting larger, bump just the phys
1386 req
->nr_phys_segments
+= nr_phys_segs
;
1390 static inline int ll_new_hw_segment(struct request_queue
*q
,
1391 struct request
*req
,
1394 int nr_hw_segs
= bio_hw_segments(q
, bio
);
1395 int nr_phys_segs
= bio_phys_segments(q
, bio
);
1397 if (req
->nr_hw_segments
+ nr_hw_segs
> q
->max_hw_segments
1398 || req
->nr_phys_segments
+ nr_phys_segs
> q
->max_phys_segments
) {
1399 req
->cmd_flags
|= REQ_NOMERGE
;
1400 if (req
== q
->last_merge
)
1401 q
->last_merge
= NULL
;
1406 * This will form the start of a new hw segment. Bump both
1409 req
->nr_hw_segments
+= nr_hw_segs
;
1410 req
->nr_phys_segments
+= nr_phys_segs
;
1414 static int ll_back_merge_fn(struct request_queue
*q
, struct request
*req
,
1417 unsigned short max_sectors
;
1420 if (unlikely(blk_pc_request(req
)))
1421 max_sectors
= q
->max_hw_sectors
;
1423 max_sectors
= q
->max_sectors
;
1425 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1426 req
->cmd_flags
|= REQ_NOMERGE
;
1427 if (req
== q
->last_merge
)
1428 q
->last_merge
= NULL
;
1431 if (unlikely(!bio_flagged(req
->biotail
, BIO_SEG_VALID
)))
1432 blk_recount_segments(q
, req
->biotail
);
1433 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1434 blk_recount_segments(q
, bio
);
1435 len
= req
->biotail
->bi_hw_back_size
+ bio
->bi_hw_front_size
;
1436 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req
->biotail
), __BVEC_START(bio
)) &&
1437 !BIOVEC_VIRT_OVERSIZE(len
)) {
1438 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1441 if (req
->nr_hw_segments
== 1)
1442 req
->bio
->bi_hw_front_size
= len
;
1443 if (bio
->bi_hw_segments
== 1)
1444 bio
->bi_hw_back_size
= len
;
1449 return ll_new_hw_segment(q
, req
, bio
);
1452 static int ll_front_merge_fn(struct request_queue
*q
, struct request
*req
,
1455 unsigned short max_sectors
;
1458 if (unlikely(blk_pc_request(req
)))
1459 max_sectors
= q
->max_hw_sectors
;
1461 max_sectors
= q
->max_sectors
;
1464 if (req
->nr_sectors
+ bio_sectors(bio
) > max_sectors
) {
1465 req
->cmd_flags
|= REQ_NOMERGE
;
1466 if (req
== q
->last_merge
)
1467 q
->last_merge
= NULL
;
1470 len
= bio
->bi_hw_back_size
+ req
->bio
->bi_hw_front_size
;
1471 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
1472 blk_recount_segments(q
, bio
);
1473 if (unlikely(!bio_flagged(req
->bio
, BIO_SEG_VALID
)))
1474 blk_recount_segments(q
, req
->bio
);
1475 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio
), __BVEC_START(req
->bio
)) &&
1476 !BIOVEC_VIRT_OVERSIZE(len
)) {
1477 int mergeable
= ll_new_mergeable(q
, req
, bio
);
1480 if (bio
->bi_hw_segments
== 1)
1481 bio
->bi_hw_front_size
= len
;
1482 if (req
->nr_hw_segments
== 1)
1483 req
->biotail
->bi_hw_back_size
= len
;
1488 return ll_new_hw_segment(q
, req
, bio
);
1491 static int ll_merge_requests_fn(struct request_queue
*q
, struct request
*req
,
1492 struct request
*next
)
1494 int total_phys_segments
;
1495 int total_hw_segments
;
1498 * First check if the either of the requests are re-queued
1499 * requests. Can't merge them if they are.
1501 if (req
->special
|| next
->special
)
1505 * Will it become too large?
1507 if ((req
->nr_sectors
+ next
->nr_sectors
) > q
->max_sectors
)
1510 total_phys_segments
= req
->nr_phys_segments
+ next
->nr_phys_segments
;
1511 if (blk_phys_contig_segment(q
, req
->biotail
, next
->bio
))
1512 total_phys_segments
--;
1514 if (total_phys_segments
> q
->max_phys_segments
)
1517 total_hw_segments
= req
->nr_hw_segments
+ next
->nr_hw_segments
;
1518 if (blk_hw_contig_segment(q
, req
->biotail
, next
->bio
)) {
1519 int len
= req
->biotail
->bi_hw_back_size
+ next
->bio
->bi_hw_front_size
;
1521 * propagate the combined length to the end of the requests
1523 if (req
->nr_hw_segments
== 1)
1524 req
->bio
->bi_hw_front_size
= len
;
1525 if (next
->nr_hw_segments
== 1)
1526 next
->biotail
->bi_hw_back_size
= len
;
1527 total_hw_segments
--;
1530 if (total_hw_segments
> q
->max_hw_segments
)
1533 /* Merge is OK... */
1534 req
->nr_phys_segments
= total_phys_segments
;
1535 req
->nr_hw_segments
= total_hw_segments
;
1540 * "plug" the device if there are no outstanding requests: this will
1541 * force the transfer to start only after we have put all the requests
1544 * This is called with interrupts off and no requests on the queue and
1545 * with the queue lock held.
1547 void blk_plug_device(struct request_queue
*q
)
1549 WARN_ON(!irqs_disabled());
1552 * don't plug a stopped queue, it must be paired with blk_start_queue()
1553 * which will restart the queueing
1555 if (blk_queue_stopped(q
))
1558 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
)) {
1559 mod_timer(&q
->unplug_timer
, jiffies
+ q
->unplug_delay
);
1560 blk_add_trace_generic(q
, NULL
, 0, BLK_TA_PLUG
);
1564 EXPORT_SYMBOL(blk_plug_device
);
1567 * remove the queue from the plugged list, if present. called with
1568 * queue lock held and interrupts disabled.
1570 int blk_remove_plug(struct request_queue
*q
)
1572 WARN_ON(!irqs_disabled());
1574 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED
, &q
->queue_flags
))
1577 del_timer(&q
->unplug_timer
);
1581 EXPORT_SYMBOL(blk_remove_plug
);
1584 * remove the plug and let it rip..
1586 void __generic_unplug_device(struct request_queue
*q
)
1588 if (unlikely(blk_queue_stopped(q
)))
1591 if (!blk_remove_plug(q
))
1596 EXPORT_SYMBOL(__generic_unplug_device
);
1599 * generic_unplug_device - fire a request queue
1600 * @q: The &struct request_queue in question
1603 * Linux uses plugging to build bigger requests queues before letting
1604 * the device have at them. If a queue is plugged, the I/O scheduler
1605 * is still adding and merging requests on the queue. Once the queue
1606 * gets unplugged, the request_fn defined for the queue is invoked and
1607 * transfers started.
1609 void generic_unplug_device(struct request_queue
*q
)
1611 spin_lock_irq(q
->queue_lock
);
1612 __generic_unplug_device(q
);
1613 spin_unlock_irq(q
->queue_lock
);
1615 EXPORT_SYMBOL(generic_unplug_device
);
1617 static void blk_backing_dev_unplug(struct backing_dev_info
*bdi
,
1620 struct request_queue
*q
= bdi
->unplug_io_data
;
1623 * devices don't necessarily have an ->unplug_fn defined
1626 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1627 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1633 static void blk_unplug_work(struct work_struct
*work
)
1635 struct request_queue
*q
=
1636 container_of(work
, struct request_queue
, unplug_work
);
1638 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_IO
, NULL
,
1639 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1644 static void blk_unplug_timeout(unsigned long data
)
1646 struct request_queue
*q
= (struct request_queue
*)data
;
1648 blk_add_trace_pdu_int(q
, BLK_TA_UNPLUG_TIMER
, NULL
,
1649 q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]);
1651 kblockd_schedule_work(&q
->unplug_work
);
1655 * blk_start_queue - restart a previously stopped queue
1656 * @q: The &struct request_queue in question
1659 * blk_start_queue() will clear the stop flag on the queue, and call
1660 * the request_fn for the queue if it was in a stopped state when
1661 * entered. Also see blk_stop_queue(). Queue lock must be held.
1663 void blk_start_queue(struct request_queue
*q
)
1665 WARN_ON(!irqs_disabled());
1667 clear_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1670 * one level of recursion is ok and is much faster than kicking
1671 * the unplug handling
1673 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1675 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1678 kblockd_schedule_work(&q
->unplug_work
);
1682 EXPORT_SYMBOL(blk_start_queue
);
1685 * blk_stop_queue - stop a queue
1686 * @q: The &struct request_queue in question
1689 * The Linux block layer assumes that a block driver will consume all
1690 * entries on the request queue when the request_fn strategy is called.
1691 * Often this will not happen, because of hardware limitations (queue
1692 * depth settings). If a device driver gets a 'queue full' response,
1693 * or if it simply chooses not to queue more I/O at one point, it can
1694 * call this function to prevent the request_fn from being called until
1695 * the driver has signalled it's ready to go again. This happens by calling
1696 * blk_start_queue() to restart queue operations. Queue lock must be held.
1698 void blk_stop_queue(struct request_queue
*q
)
1701 set_bit(QUEUE_FLAG_STOPPED
, &q
->queue_flags
);
1703 EXPORT_SYMBOL(blk_stop_queue
);
1706 * blk_sync_queue - cancel any pending callbacks on a queue
1710 * The block layer may perform asynchronous callback activity
1711 * on a queue, such as calling the unplug function after a timeout.
1712 * A block device may call blk_sync_queue to ensure that any
1713 * such activity is cancelled, thus allowing it to release resources
1714 * that the callbacks might use. The caller must already have made sure
1715 * that its ->make_request_fn will not re-add plugging prior to calling
1719 void blk_sync_queue(struct request_queue
*q
)
1721 del_timer_sync(&q
->unplug_timer
);
1723 EXPORT_SYMBOL(blk_sync_queue
);
1726 * blk_run_queue - run a single device queue
1727 * @q: The queue to run
1729 void blk_run_queue(struct request_queue
*q
)
1731 unsigned long flags
;
1733 spin_lock_irqsave(q
->queue_lock
, flags
);
1737 * Only recurse once to avoid overrunning the stack, let the unplug
1738 * handling reinvoke the handler shortly if we already got there.
1740 if (!elv_queue_empty(q
)) {
1741 if (!test_and_set_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
)) {
1743 clear_bit(QUEUE_FLAG_REENTER
, &q
->queue_flags
);
1746 kblockd_schedule_work(&q
->unplug_work
);
1750 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1752 EXPORT_SYMBOL(blk_run_queue
);
1755 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
1756 * @kobj: the kobj belonging of the request queue to be released
1759 * blk_cleanup_queue is the pair to blk_init_queue() or
1760 * blk_queue_make_request(). It should be called when a request queue is
1761 * being released; typically when a block device is being de-registered.
1762 * Currently, its primary task it to free all the &struct request
1763 * structures that were allocated to the queue and the queue itself.
1766 * Hopefully the low level driver will have finished any
1767 * outstanding requests first...
1769 static void blk_release_queue(struct kobject
*kobj
)
1771 struct request_queue
*q
=
1772 container_of(kobj
, struct request_queue
, kobj
);
1773 struct request_list
*rl
= &q
->rq
;
1778 mempool_destroy(rl
->rq_pool
);
1781 __blk_queue_free_tags(q
);
1783 blk_trace_shutdown(q
);
1785 kmem_cache_free(requestq_cachep
, q
);
1788 void blk_put_queue(struct request_queue
*q
)
1790 kobject_put(&q
->kobj
);
1792 EXPORT_SYMBOL(blk_put_queue
);
1794 void blk_cleanup_queue(struct request_queue
* q
)
1796 mutex_lock(&q
->sysfs_lock
);
1797 set_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
);
1798 mutex_unlock(&q
->sysfs_lock
);
1801 elevator_exit(q
->elevator
);
1806 EXPORT_SYMBOL(blk_cleanup_queue
);
1808 static int blk_init_free_list(struct request_queue
*q
)
1810 struct request_list
*rl
= &q
->rq
;
1812 rl
->count
[READ
] = rl
->count
[WRITE
] = 0;
1813 rl
->starved
[READ
] = rl
->starved
[WRITE
] = 0;
1815 init_waitqueue_head(&rl
->wait
[READ
]);
1816 init_waitqueue_head(&rl
->wait
[WRITE
]);
1818 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, mempool_alloc_slab
,
1819 mempool_free_slab
, request_cachep
, q
->node
);
1827 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
1829 return blk_alloc_queue_node(gfp_mask
, -1);
1831 EXPORT_SYMBOL(blk_alloc_queue
);
1833 static struct kobj_type queue_ktype
;
1835 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
1837 struct request_queue
*q
;
1839 q
= kmem_cache_alloc_node(requestq_cachep
,
1840 gfp_mask
| __GFP_ZERO
, node_id
);
1844 init_timer(&q
->unplug_timer
);
1846 kobject_set_name(&q
->kobj
, "%s", "queue");
1847 q
->kobj
.ktype
= &queue_ktype
;
1848 kobject_init(&q
->kobj
);
1850 q
->backing_dev_info
.unplug_io_fn
= blk_backing_dev_unplug
;
1851 q
->backing_dev_info
.unplug_io_data
= q
;
1853 mutex_init(&q
->sysfs_lock
);
1857 EXPORT_SYMBOL(blk_alloc_queue_node
);
1860 * blk_init_queue - prepare a request queue for use with a block device
1861 * @rfn: The function to be called to process requests that have been
1862 * placed on the queue.
1863 * @lock: Request queue spin lock
1866 * If a block device wishes to use the standard request handling procedures,
1867 * which sorts requests and coalesces adjacent requests, then it must
1868 * call blk_init_queue(). The function @rfn will be called when there
1869 * are requests on the queue that need to be processed. If the device
1870 * supports plugging, then @rfn may not be called immediately when requests
1871 * are available on the queue, but may be called at some time later instead.
1872 * Plugged queues are generally unplugged when a buffer belonging to one
1873 * of the requests on the queue is needed, or due to memory pressure.
1875 * @rfn is not required, or even expected, to remove all requests off the
1876 * queue, but only as many as it can handle at a time. If it does leave
1877 * requests on the queue, it is responsible for arranging that the requests
1878 * get dealt with eventually.
1880 * The queue spin lock must be held while manipulating the requests on the
1881 * request queue; this lock will be taken also from interrupt context, so irq
1882 * disabling is needed for it.
1884 * Function returns a pointer to the initialized request queue, or NULL if
1885 * it didn't succeed.
1888 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1889 * when the block device is deactivated (such as at module unload).
1892 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1894 return blk_init_queue_node(rfn
, lock
, -1);
1896 EXPORT_SYMBOL(blk_init_queue
);
1898 struct request_queue
*
1899 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1901 struct request_queue
*q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1907 if (blk_init_free_list(q
)) {
1908 kmem_cache_free(requestq_cachep
, q
);
1913 * if caller didn't supply a lock, they get per-queue locking with
1917 spin_lock_init(&q
->__queue_lock
);
1918 lock
= &q
->__queue_lock
;
1921 q
->request_fn
= rfn
;
1922 q
->prep_rq_fn
= NULL
;
1923 q
->unplug_fn
= generic_unplug_device
;
1924 q
->queue_flags
= (1 << QUEUE_FLAG_CLUSTER
);
1925 q
->queue_lock
= lock
;
1927 blk_queue_segment_boundary(q
, 0xffffffff);
1929 blk_queue_make_request(q
, __make_request
);
1930 blk_queue_max_segment_size(q
, MAX_SEGMENT_SIZE
);
1932 blk_queue_max_hw_segments(q
, MAX_HW_SEGMENTS
);
1933 blk_queue_max_phys_segments(q
, MAX_PHYS_SEGMENTS
);
1935 q
->sg_reserved_size
= INT_MAX
;
1940 if (!elevator_init(q
, NULL
)) {
1941 blk_queue_congestion_threshold(q
);
1948 EXPORT_SYMBOL(blk_init_queue_node
);
1950 int blk_get_queue(struct request_queue
*q
)
1952 if (likely(!test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
))) {
1953 kobject_get(&q
->kobj
);
1960 EXPORT_SYMBOL(blk_get_queue
);
1962 static inline void blk_free_request(struct request_queue
*q
, struct request
*rq
)
1964 if (rq
->cmd_flags
& REQ_ELVPRIV
)
1965 elv_put_request(q
, rq
);
1966 mempool_free(rq
, q
->rq
.rq_pool
);
1969 static struct request
*
1970 blk_alloc_request(struct request_queue
*q
, int rw
, int priv
, gfp_t gfp_mask
)
1972 struct request
*rq
= mempool_alloc(q
->rq
.rq_pool
, gfp_mask
);
1978 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
1979 * see bio.h and blkdev.h
1981 rq
->cmd_flags
= rw
| REQ_ALLOCED
;
1984 if (unlikely(elv_set_request(q
, rq
, gfp_mask
))) {
1985 mempool_free(rq
, q
->rq
.rq_pool
);
1988 rq
->cmd_flags
|= REQ_ELVPRIV
;
1995 * ioc_batching returns true if the ioc is a valid batching request and
1996 * should be given priority access to a request.
1998 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
2004 * Make sure the process is able to allocate at least 1 request
2005 * even if the batch times out, otherwise we could theoretically
2008 return ioc
->nr_batch_requests
== q
->nr_batching
||
2009 (ioc
->nr_batch_requests
> 0
2010 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
2014 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2015 * will cause the process to be a "batcher" on all queues in the system. This
2016 * is the behaviour we want though - once it gets a wakeup it should be given
2019 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
2021 if (!ioc
|| ioc_batching(q
, ioc
))
2024 ioc
->nr_batch_requests
= q
->nr_batching
;
2025 ioc
->last_waited
= jiffies
;
2028 static void __freed_request(struct request_queue
*q
, int rw
)
2030 struct request_list
*rl
= &q
->rq
;
2032 if (rl
->count
[rw
] < queue_congestion_off_threshold(q
))
2033 blk_clear_queue_congested(q
, rw
);
2035 if (rl
->count
[rw
] + 1 <= q
->nr_requests
) {
2036 if (waitqueue_active(&rl
->wait
[rw
]))
2037 wake_up(&rl
->wait
[rw
]);
2039 blk_clear_queue_full(q
, rw
);
2044 * A request has just been released. Account for it, update the full and
2045 * congestion status, wake up any waiters. Called under q->queue_lock.
2047 static void freed_request(struct request_queue
*q
, int rw
, int priv
)
2049 struct request_list
*rl
= &q
->rq
;
2055 __freed_request(q
, rw
);
2057 if (unlikely(rl
->starved
[rw
^ 1]))
2058 __freed_request(q
, rw
^ 1);
2061 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2063 * Get a free request, queue_lock must be held.
2064 * Returns NULL on failure, with queue_lock held.
2065 * Returns !NULL on success, with queue_lock *not held*.
2067 static struct request
*get_request(struct request_queue
*q
, int rw_flags
,
2068 struct bio
*bio
, gfp_t gfp_mask
)
2070 struct request
*rq
= NULL
;
2071 struct request_list
*rl
= &q
->rq
;
2072 struct io_context
*ioc
= NULL
;
2073 const int rw
= rw_flags
& 0x01;
2074 int may_queue
, priv
;
2076 may_queue
= elv_may_queue(q
, rw_flags
);
2077 if (may_queue
== ELV_MQUEUE_NO
)
2080 if (rl
->count
[rw
]+1 >= queue_congestion_on_threshold(q
)) {
2081 if (rl
->count
[rw
]+1 >= q
->nr_requests
) {
2082 ioc
= current_io_context(GFP_ATOMIC
, q
->node
);
2084 * The queue will fill after this allocation, so set
2085 * it as full, and mark this process as "batching".
2086 * This process will be allowed to complete a batch of
2087 * requests, others will be blocked.
2089 if (!blk_queue_full(q
, rw
)) {
2090 ioc_set_batching(q
, ioc
);
2091 blk_set_queue_full(q
, rw
);
2093 if (may_queue
!= ELV_MQUEUE_MUST
2094 && !ioc_batching(q
, ioc
)) {
2096 * The queue is full and the allocating
2097 * process is not a "batcher", and not
2098 * exempted by the IO scheduler
2104 blk_set_queue_congested(q
, rw
);
2108 * Only allow batching queuers to allocate up to 50% over the defined
2109 * limit of requests, otherwise we could have thousands of requests
2110 * allocated with any setting of ->nr_requests
2112 if (rl
->count
[rw
] >= (3 * q
->nr_requests
/ 2))
2116 rl
->starved
[rw
] = 0;
2118 priv
= !test_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
2122 spin_unlock_irq(q
->queue_lock
);
2124 rq
= blk_alloc_request(q
, rw_flags
, priv
, gfp_mask
);
2125 if (unlikely(!rq
)) {
2127 * Allocation failed presumably due to memory. Undo anything
2128 * we might have messed up.
2130 * Allocating task should really be put onto the front of the
2131 * wait queue, but this is pretty rare.
2133 spin_lock_irq(q
->queue_lock
);
2134 freed_request(q
, rw
, priv
);
2137 * in the very unlikely event that allocation failed and no
2138 * requests for this direction was pending, mark us starved
2139 * so that freeing of a request in the other direction will
2140 * notice us. another possible fix would be to split the
2141 * rq mempool into READ and WRITE
2144 if (unlikely(rl
->count
[rw
] == 0))
2145 rl
->starved
[rw
] = 1;
2151 * ioc may be NULL here, and ioc_batching will be false. That's
2152 * OK, if the queue is under the request limit then requests need
2153 * not count toward the nr_batch_requests limit. There will always
2154 * be some limit enforced by BLK_BATCH_TIME.
2156 if (ioc_batching(q
, ioc
))
2157 ioc
->nr_batch_requests
--;
2161 blk_add_trace_generic(q
, bio
, rw
, BLK_TA_GETRQ
);
2167 * No available requests for this queue, unplug the device and wait for some
2168 * requests to become available.
2170 * Called with q->queue_lock held, and returns with it unlocked.
2172 static struct request
*get_request_wait(struct request_queue
*q
, int rw_flags
,
2175 const int rw
= rw_flags
& 0x01;
2178 rq
= get_request(q
, rw_flags
, 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_flags
, 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(struct request_queue
*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(struct request_queue
*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(struct request_queue
*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(struct request_queue
*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 int blk_rq_append_bio(struct request_queue
*q
, struct request
*rq
,
2344 blk_rq_bio_prep(q
, rq
, bio
);
2345 else if (!ll_back_merge_fn(q
, rq
, bio
))
2348 rq
->biotail
->bi_next
= bio
;
2351 rq
->data_len
+= bio
->bi_size
;
2355 EXPORT_SYMBOL(blk_rq_append_bio
);
2357 static int __blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2358 void __user
*ubuf
, unsigned int len
)
2360 unsigned long uaddr
;
2361 struct bio
*bio
, *orig_bio
;
2364 reading
= rq_data_dir(rq
) == READ
;
2367 * if alignment requirement is satisfied, map in user pages for
2368 * direct dma. else, set up kernel bounce buffers
2370 uaddr
= (unsigned long) ubuf
;
2371 if (!(uaddr
& queue_dma_alignment(q
)) && !(len
& queue_dma_alignment(q
)))
2372 bio
= bio_map_user(q
, NULL
, uaddr
, len
, reading
);
2374 bio
= bio_copy_user(q
, uaddr
, len
, reading
);
2377 return PTR_ERR(bio
);
2380 blk_queue_bounce(q
, &bio
);
2383 * We link the bounce buffer in and could have to traverse it
2384 * later so we have to get a ref to prevent it from being freed
2388 ret
= blk_rq_append_bio(q
, rq
, bio
);
2390 return bio
->bi_size
;
2392 /* if it was boucned we must call the end io function */
2394 __blk_rq_unmap_user(orig_bio
);
2400 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2401 * @q: request queue where request should be inserted
2402 * @rq: request structure to fill
2403 * @ubuf: the user buffer
2404 * @len: length of user data
2407 * Data will be mapped directly for zero copy io, if possible. Otherwise
2408 * a kernel bounce buffer is used.
2410 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2411 * still in process context.
2413 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2414 * before being submitted to the device, as pages mapped may be out of
2415 * reach. It's the callers responsibility to make sure this happens. The
2416 * original bio must be passed back in to blk_rq_unmap_user() for proper
2419 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
2420 void __user
*ubuf
, unsigned long len
)
2422 unsigned long bytes_read
= 0;
2423 struct bio
*bio
= NULL
;
2426 if (len
> (q
->max_hw_sectors
<< 9))
2431 while (bytes_read
!= len
) {
2432 unsigned long map_len
, end
, start
;
2434 map_len
= min_t(unsigned long, len
- bytes_read
, BIO_MAX_SIZE
);
2435 end
= ((unsigned long)ubuf
+ map_len
+ PAGE_SIZE
- 1)
2437 start
= (unsigned long)ubuf
>> PAGE_SHIFT
;
2440 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2441 * pages. If this happens we just lower the requested
2442 * mapping len by a page so that we can fit
2444 if (end
- start
> BIO_MAX_PAGES
)
2445 map_len
-= PAGE_SIZE
;
2447 ret
= __blk_rq_map_user(q
, rq
, ubuf
, map_len
);
2456 rq
->buffer
= rq
->data
= NULL
;
2459 blk_rq_unmap_user(bio
);
2463 EXPORT_SYMBOL(blk_rq_map_user
);
2466 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2467 * @q: request queue where request should be inserted
2468 * @rq: request to map data to
2469 * @iov: pointer to the iovec
2470 * @iov_count: number of elements in the iovec
2471 * @len: I/O byte count
2474 * Data will be mapped directly for zero copy io, if possible. Otherwise
2475 * a kernel bounce buffer is used.
2477 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2478 * still in process context.
2480 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2481 * before being submitted to the device, as pages mapped may be out of
2482 * reach. It's the callers responsibility to make sure this happens. The
2483 * original bio must be passed back in to blk_rq_unmap_user() for proper
2486 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
2487 struct sg_iovec
*iov
, int iov_count
, unsigned int len
)
2491 if (!iov
|| iov_count
<= 0)
2494 /* we don't allow misaligned data like bio_map_user() does. If the
2495 * user is using sg, they're expected to know the alignment constraints
2496 * and respect them accordingly */
2497 bio
= bio_map_user_iov(q
, NULL
, iov
, iov_count
, rq_data_dir(rq
)== READ
);
2499 return PTR_ERR(bio
);
2501 if (bio
->bi_size
!= len
) {
2503 bio_unmap_user(bio
);
2508 blk_rq_bio_prep(q
, rq
, bio
);
2509 rq
->buffer
= rq
->data
= NULL
;
2513 EXPORT_SYMBOL(blk_rq_map_user_iov
);
2516 * blk_rq_unmap_user - unmap a request with user data
2517 * @bio: start of bio list
2520 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2521 * supply the original rq->bio from the blk_rq_map_user() return, since
2522 * the io completion may have changed rq->bio.
2524 int blk_rq_unmap_user(struct bio
*bio
)
2526 struct bio
*mapped_bio
;
2531 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
2532 mapped_bio
= bio
->bi_private
;
2534 ret2
= __blk_rq_unmap_user(mapped_bio
);
2540 bio_put(mapped_bio
);
2546 EXPORT_SYMBOL(blk_rq_unmap_user
);
2549 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2550 * @q: request queue where request should be inserted
2551 * @rq: request to fill
2552 * @kbuf: the kernel buffer
2553 * @len: length of user data
2554 * @gfp_mask: memory allocation flags
2556 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
2557 unsigned int len
, gfp_t gfp_mask
)
2561 if (len
> (q
->max_hw_sectors
<< 9))
2566 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
2568 return PTR_ERR(bio
);
2570 if (rq_data_dir(rq
) == WRITE
)
2571 bio
->bi_rw
|= (1 << BIO_RW
);
2573 blk_rq_bio_prep(q
, rq
, bio
);
2574 blk_queue_bounce(q
, &rq
->bio
);
2575 rq
->buffer
= rq
->data
= NULL
;
2579 EXPORT_SYMBOL(blk_rq_map_kern
);
2582 * blk_execute_rq_nowait - insert a request into queue for execution
2583 * @q: queue to insert the request in
2584 * @bd_disk: matching gendisk
2585 * @rq: request to insert
2586 * @at_head: insert request at head or tail of queue
2587 * @done: I/O completion handler
2590 * Insert a fully prepared request at the back of the io scheduler queue
2591 * for execution. Don't wait for completion.
2593 void blk_execute_rq_nowait(struct request_queue
*q
, struct gendisk
*bd_disk
,
2594 struct request
*rq
, int at_head
,
2597 int where
= at_head
? ELEVATOR_INSERT_FRONT
: ELEVATOR_INSERT_BACK
;
2599 rq
->rq_disk
= bd_disk
;
2600 rq
->cmd_flags
|= REQ_NOMERGE
;
2602 WARN_ON(irqs_disabled());
2603 spin_lock_irq(q
->queue_lock
);
2604 __elv_add_request(q
, rq
, where
, 1);
2605 __generic_unplug_device(q
);
2606 spin_unlock_irq(q
->queue_lock
);
2608 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait
);
2611 * blk_execute_rq - insert a request into queue for execution
2612 * @q: queue to insert the request in
2613 * @bd_disk: matching gendisk
2614 * @rq: request to insert
2615 * @at_head: insert request at head or tail of queue
2618 * Insert a fully prepared request at the back of the io scheduler queue
2619 * for execution and wait for completion.
2621 int blk_execute_rq(struct request_queue
*q
, struct gendisk
*bd_disk
,
2622 struct request
*rq
, int at_head
)
2624 DECLARE_COMPLETION_ONSTACK(wait
);
2625 char sense
[SCSI_SENSE_BUFFERSIZE
];
2629 * we need an extra reference to the request, so we can look at
2630 * it after io completion
2635 memset(sense
, 0, sizeof(sense
));
2640 rq
->end_io_data
= &wait
;
2641 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, blk_end_sync_rq
);
2642 wait_for_completion(&wait
);
2650 EXPORT_SYMBOL(blk_execute_rq
);
2652 static void bio_end_empty_barrier(struct bio
*bio
, int err
)
2655 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2657 complete(bio
->bi_private
);
2661 * blkdev_issue_flush - queue a flush
2662 * @bdev: blockdev to issue flush for
2663 * @error_sector: error sector
2666 * Issue a flush for the block device in question. Caller can supply
2667 * room for storing the error offset in case of a flush error, if they
2668 * wish to. Caller must run wait_for_completion() on its own.
2670 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
2672 DECLARE_COMPLETION_ONSTACK(wait
);
2673 struct request_queue
*q
;
2677 if (bdev
->bd_disk
== NULL
)
2680 q
= bdev_get_queue(bdev
);
2684 bio
= bio_alloc(GFP_KERNEL
, 0);
2688 bio
->bi_end_io
= bio_end_empty_barrier
;
2689 bio
->bi_private
= &wait
;
2690 bio
->bi_bdev
= bdev
;
2691 submit_bio(1 << BIO_RW_BARRIER
, bio
);
2693 wait_for_completion(&wait
);
2696 * The driver must store the error location in ->bi_sector, if
2697 * it supports it. For non-stacked drivers, this should be copied
2701 *error_sector
= bio
->bi_sector
;
2704 if (!bio_flagged(bio
, BIO_UPTODATE
))
2711 EXPORT_SYMBOL(blkdev_issue_flush
);
2713 static void drive_stat_acct(struct request
*rq
, int nr_sectors
, int new_io
)
2715 int rw
= rq_data_dir(rq
);
2717 if (!blk_fs_request(rq
) || !rq
->rq_disk
)
2721 __disk_stat_inc(rq
->rq_disk
, merges
[rw
]);
2723 disk_round_stats(rq
->rq_disk
);
2724 rq
->rq_disk
->in_flight
++;
2729 * add-request adds a request to the linked list.
2730 * queue lock is held and interrupts disabled, as we muck with the
2731 * request queue list.
2733 static inline void add_request(struct request_queue
* q
, struct request
* req
)
2735 drive_stat_acct(req
, req
->nr_sectors
, 1);
2738 * elevator indicated where it wants this request to be
2739 * inserted at elevator_merge time
2741 __elv_add_request(q
, req
, ELEVATOR_INSERT_SORT
, 0);
2745 * disk_round_stats() - Round off the performance stats on a struct
2748 * The average IO queue length and utilisation statistics are maintained
2749 * by observing the current state of the queue length and the amount of
2750 * time it has been in this state for.
2752 * Normally, that accounting is done on IO completion, but that can result
2753 * in more than a second's worth of IO being accounted for within any one
2754 * second, leading to >100% utilisation. To deal with that, we call this
2755 * function to do a round-off before returning the results when reading
2756 * /proc/diskstats. This accounts immediately for all queue usage up to
2757 * the current jiffies and restarts the counters again.
2759 void disk_round_stats(struct gendisk
*disk
)
2761 unsigned long now
= jiffies
;
2763 if (now
== disk
->stamp
)
2766 if (disk
->in_flight
) {
2767 __disk_stat_add(disk
, time_in_queue
,
2768 disk
->in_flight
* (now
- disk
->stamp
));
2769 __disk_stat_add(disk
, io_ticks
, (now
- disk
->stamp
));
2774 EXPORT_SYMBOL_GPL(disk_round_stats
);
2777 * queue lock must be held
2779 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
2783 if (unlikely(--req
->ref_count
))
2786 elv_completed_request(q
, req
);
2789 * Request may not have originated from ll_rw_blk. if not,
2790 * it didn't come out of our reserved rq pools
2792 if (req
->cmd_flags
& REQ_ALLOCED
) {
2793 int rw
= rq_data_dir(req
);
2794 int priv
= req
->cmd_flags
& REQ_ELVPRIV
;
2796 BUG_ON(!list_empty(&req
->queuelist
));
2797 BUG_ON(!hlist_unhashed(&req
->hash
));
2799 blk_free_request(q
, req
);
2800 freed_request(q
, rw
, priv
);
2804 EXPORT_SYMBOL_GPL(__blk_put_request
);
2806 void blk_put_request(struct request
*req
)
2808 unsigned long flags
;
2809 struct request_queue
*q
= req
->q
;
2812 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2813 * following if (q) test.
2816 spin_lock_irqsave(q
->queue_lock
, flags
);
2817 __blk_put_request(q
, req
);
2818 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2822 EXPORT_SYMBOL(blk_put_request
);
2825 * blk_end_sync_rq - executes a completion event on a request
2826 * @rq: request to complete
2827 * @error: end io status of the request
2829 void blk_end_sync_rq(struct request
*rq
, int error
)
2831 struct completion
*waiting
= rq
->end_io_data
;
2833 rq
->end_io_data
= NULL
;
2834 __blk_put_request(rq
->q
, rq
);
2837 * complete last, if this is a stack request the process (and thus
2838 * the rq pointer) could be invalid right after this complete()
2842 EXPORT_SYMBOL(blk_end_sync_rq
);
2845 * Has to be called with the request spinlock acquired
2847 static int attempt_merge(struct request_queue
*q
, struct request
*req
,
2848 struct request
*next
)
2850 if (!rq_mergeable(req
) || !rq_mergeable(next
))
2856 if (req
->sector
+ req
->nr_sectors
!= next
->sector
)
2859 if (rq_data_dir(req
) != rq_data_dir(next
)
2860 || req
->rq_disk
!= next
->rq_disk
2865 * If we are allowed to merge, then append bio list
2866 * from next to rq and release next. merge_requests_fn
2867 * will have updated segment counts, update sector
2870 if (!ll_merge_requests_fn(q
, req
, next
))
2874 * At this point we have either done a back merge
2875 * or front merge. We need the smaller start_time of
2876 * the merged requests to be the current request
2877 * for accounting purposes.
2879 if (time_after(req
->start_time
, next
->start_time
))
2880 req
->start_time
= next
->start_time
;
2882 req
->biotail
->bi_next
= next
->bio
;
2883 req
->biotail
= next
->biotail
;
2885 req
->nr_sectors
= req
->hard_nr_sectors
+= next
->hard_nr_sectors
;
2887 elv_merge_requests(q
, req
, next
);
2890 disk_round_stats(req
->rq_disk
);
2891 req
->rq_disk
->in_flight
--;
2894 req
->ioprio
= ioprio_best(req
->ioprio
, next
->ioprio
);
2896 __blk_put_request(q
, next
);
2900 static inline int attempt_back_merge(struct request_queue
*q
,
2903 struct request
*next
= elv_latter_request(q
, rq
);
2906 return attempt_merge(q
, rq
, next
);
2911 static inline int attempt_front_merge(struct request_queue
*q
,
2914 struct request
*prev
= elv_former_request(q
, rq
);
2917 return attempt_merge(q
, prev
, rq
);
2922 static void init_request_from_bio(struct request
*req
, struct bio
*bio
)
2924 req
->cmd_type
= REQ_TYPE_FS
;
2927 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2929 if (bio_rw_ahead(bio
) || bio_failfast(bio
))
2930 req
->cmd_flags
|= REQ_FAILFAST
;
2933 * REQ_BARRIER implies no merging, but lets make it explicit
2935 if (unlikely(bio_barrier(bio
)))
2936 req
->cmd_flags
|= (REQ_HARDBARRIER
| REQ_NOMERGE
);
2939 req
->cmd_flags
|= REQ_RW_SYNC
;
2940 if (bio_rw_meta(bio
))
2941 req
->cmd_flags
|= REQ_RW_META
;
2944 req
->hard_sector
= req
->sector
= bio
->bi_sector
;
2945 req
->ioprio
= bio_prio(bio
);
2946 req
->start_time
= jiffies
;
2947 blk_rq_bio_prep(req
->q
, req
, bio
);
2950 static int __make_request(struct request_queue
*q
, struct bio
*bio
)
2952 struct request
*req
;
2953 int el_ret
, nr_sectors
, barrier
, err
;
2954 const unsigned short prio
= bio_prio(bio
);
2955 const int sync
= bio_sync(bio
);
2958 nr_sectors
= bio_sectors(bio
);
2961 * low level driver can indicate that it wants pages above a
2962 * certain limit bounced to low memory (ie for highmem, or even
2963 * ISA dma in theory)
2965 blk_queue_bounce(q
, &bio
);
2967 barrier
= bio_barrier(bio
);
2968 if (unlikely(barrier
) && (q
->next_ordered
== QUEUE_ORDERED_NONE
)) {
2973 spin_lock_irq(q
->queue_lock
);
2975 if (unlikely(barrier
) || elv_queue_empty(q
))
2978 el_ret
= elv_merge(q
, &req
, bio
);
2980 case ELEVATOR_BACK_MERGE
:
2981 BUG_ON(!rq_mergeable(req
));
2983 if (!ll_back_merge_fn(q
, req
, bio
))
2986 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
);
2988 req
->biotail
->bi_next
= bio
;
2990 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
2991 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
2992 drive_stat_acct(req
, nr_sectors
, 0);
2993 if (!attempt_back_merge(q
, req
))
2994 elv_merged_request(q
, req
, el_ret
);
2997 case ELEVATOR_FRONT_MERGE
:
2998 BUG_ON(!rq_mergeable(req
));
3000 if (!ll_front_merge_fn(q
, req
, bio
))
3003 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
);
3005 bio
->bi_next
= req
->bio
;
3009 * may not be valid. if the low level driver said
3010 * it didn't need a bounce buffer then it better
3011 * not touch req->buffer either...
3013 req
->buffer
= bio_data(bio
);
3014 req
->current_nr_sectors
= bio_cur_sectors(bio
);
3015 req
->hard_cur_sectors
= req
->current_nr_sectors
;
3016 req
->sector
= req
->hard_sector
= bio
->bi_sector
;
3017 req
->nr_sectors
= req
->hard_nr_sectors
+= nr_sectors
;
3018 req
->ioprio
= ioprio_best(req
->ioprio
, prio
);
3019 drive_stat_acct(req
, nr_sectors
, 0);
3020 if (!attempt_front_merge(q
, req
))
3021 elv_merged_request(q
, req
, el_ret
);
3024 /* ELV_NO_MERGE: elevator says don't/can't merge. */
3031 * This sync check and mask will be re-done in init_request_from_bio(),
3032 * but we need to set it earlier to expose the sync flag to the
3033 * rq allocator and io schedulers.
3035 rw_flags
= bio_data_dir(bio
);
3037 rw_flags
|= REQ_RW_SYNC
;
3040 * Grab a free request. This is might sleep but can not fail.
3041 * Returns with the queue unlocked.
3043 req
= get_request_wait(q
, rw_flags
, bio
);
3046 * After dropping the lock and possibly sleeping here, our request
3047 * may now be mergeable after it had proven unmergeable (above).
3048 * We don't worry about that case for efficiency. It won't happen
3049 * often, and the elevators are able to handle it.
3051 init_request_from_bio(req
, bio
);
3053 spin_lock_irq(q
->queue_lock
);
3054 if (elv_queue_empty(q
))
3056 add_request(q
, req
);
3059 __generic_unplug_device(q
);
3061 spin_unlock_irq(q
->queue_lock
);
3065 bio_endio(bio
, err
);
3070 * If bio->bi_dev is a partition, remap the location
3072 static inline void blk_partition_remap(struct bio
*bio
)
3074 struct block_device
*bdev
= bio
->bi_bdev
;
3076 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
3077 struct hd_struct
*p
= bdev
->bd_part
;
3078 const int rw
= bio_data_dir(bio
);
3080 p
->sectors
[rw
] += bio_sectors(bio
);
3083 bio
->bi_sector
+= p
->start_sect
;
3084 bio
->bi_bdev
= bdev
->bd_contains
;
3086 blk_add_trace_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
3087 bdev
->bd_dev
, bio
->bi_sector
,
3088 bio
->bi_sector
- p
->start_sect
);
3092 static void handle_bad_sector(struct bio
*bio
)
3094 char b
[BDEVNAME_SIZE
];
3096 printk(KERN_INFO
"attempt to access beyond end of device\n");
3097 printk(KERN_INFO
"%s: rw=%ld, want=%Lu, limit=%Lu\n",
3098 bdevname(bio
->bi_bdev
, b
),
3100 (unsigned long long)bio
->bi_sector
+ bio_sectors(bio
),
3101 (long long)(bio
->bi_bdev
->bd_inode
->i_size
>> 9));
3103 set_bit(BIO_EOF
, &bio
->bi_flags
);
3106 #ifdef CONFIG_FAIL_MAKE_REQUEST
3108 static DECLARE_FAULT_ATTR(fail_make_request
);
3110 static int __init
setup_fail_make_request(char *str
)
3112 return setup_fault_attr(&fail_make_request
, str
);
3114 __setup("fail_make_request=", setup_fail_make_request
);
3116 static int should_fail_request(struct bio
*bio
)
3118 if ((bio
->bi_bdev
->bd_disk
->flags
& GENHD_FL_FAIL
) ||
3119 (bio
->bi_bdev
->bd_part
&& bio
->bi_bdev
->bd_part
->make_it_fail
))
3120 return should_fail(&fail_make_request
, bio
->bi_size
);
3125 static int __init
fail_make_request_debugfs(void)
3127 return init_fault_attr_dentries(&fail_make_request
,
3128 "fail_make_request");
3131 late_initcall(fail_make_request_debugfs
);
3133 #else /* CONFIG_FAIL_MAKE_REQUEST */
3135 static inline int should_fail_request(struct bio
*bio
)
3140 #endif /* CONFIG_FAIL_MAKE_REQUEST */
3143 * Check whether this bio extends beyond the end of the device.
3145 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
3152 /* Test device or partition size, when known. */
3153 maxsector
= bio
->bi_bdev
->bd_inode
->i_size
>> 9;
3155 sector_t sector
= bio
->bi_sector
;
3157 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
3159 * This may well happen - the kernel calls bread()
3160 * without checking the size of the device, e.g., when
3161 * mounting a device.
3163 handle_bad_sector(bio
);
3172 * generic_make_request: hand a buffer to its device driver for I/O
3173 * @bio: The bio describing the location in memory and on the device.
3175 * generic_make_request() is used to make I/O requests of block
3176 * devices. It is passed a &struct bio, which describes the I/O that needs
3179 * generic_make_request() does not return any status. The
3180 * success/failure status of the request, along with notification of
3181 * completion, is delivered asynchronously through the bio->bi_end_io
3182 * function described (one day) else where.
3184 * The caller of generic_make_request must make sure that bi_io_vec
3185 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3186 * set to describe the device address, and the
3187 * bi_end_io and optionally bi_private are set to describe how
3188 * completion notification should be signaled.
3190 * generic_make_request and the drivers it calls may use bi_next if this
3191 * bio happens to be merged with someone else, and may change bi_dev and
3192 * bi_sector for remaps as it sees fit. So the values of these fields
3193 * should NOT be depended on after the call to generic_make_request.
3195 static inline void __generic_make_request(struct bio
*bio
)
3197 struct request_queue
*q
;
3198 sector_t old_sector
;
3199 int ret
, nr_sectors
= bio_sectors(bio
);
3204 if (bio_check_eod(bio
, nr_sectors
))
3208 * Resolve the mapping until finished. (drivers are
3209 * still free to implement/resolve their own stacking
3210 * by explicitly returning 0)
3212 * NOTE: we don't repeat the blk_size check for each new device.
3213 * Stacking drivers are expected to know what they are doing.
3218 char b
[BDEVNAME_SIZE
];
3220 q
= bdev_get_queue(bio
->bi_bdev
);
3223 "generic_make_request: Trying to access "
3224 "nonexistent block-device %s (%Lu)\n",
3225 bdevname(bio
->bi_bdev
, b
),
3226 (long long) bio
->bi_sector
);
3228 bio_endio(bio
, -EIO
);
3232 if (unlikely(nr_sectors
> q
->max_hw_sectors
)) {
3233 printk("bio too big device %s (%u > %u)\n",
3234 bdevname(bio
->bi_bdev
, b
),
3240 if (unlikely(test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)))
3243 if (should_fail_request(bio
))
3247 * If this device has partitions, remap block n
3248 * of partition p to block n+start(p) of the disk.
3250 blk_partition_remap(bio
);
3252 if (old_sector
!= -1)
3253 blk_add_trace_remap(q
, bio
, old_dev
, bio
->bi_sector
,
3256 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
);
3258 old_sector
= bio
->bi_sector
;
3259 old_dev
= bio
->bi_bdev
->bd_dev
;
3261 if (bio_check_eod(bio
, nr_sectors
))
3264 ret
= q
->make_request_fn(q
, bio
);
3269 * We only want one ->make_request_fn to be active at a time,
3270 * else stack usage with stacked devices could be a problem.
3271 * So use current->bio_{list,tail} to keep a list of requests
3272 * submited by a make_request_fn function.
3273 * current->bio_tail is also used as a flag to say if
3274 * generic_make_request is currently active in this task or not.
3275 * If it is NULL, then no make_request is active. If it is non-NULL,
3276 * then a make_request is active, and new requests should be added
3279 void generic_make_request(struct bio
*bio
)
3281 if (current
->bio_tail
) {
3282 /* make_request is active */
3283 *(current
->bio_tail
) = bio
;
3284 bio
->bi_next
= NULL
;
3285 current
->bio_tail
= &bio
->bi_next
;
3288 /* following loop may be a bit non-obvious, and so deserves some
3290 * Before entering the loop, bio->bi_next is NULL (as all callers
3291 * ensure that) so we have a list with a single bio.
3292 * We pretend that we have just taken it off a longer list, so
3293 * we assign bio_list to the next (which is NULL) and bio_tail
3294 * to &bio_list, thus initialising the bio_list of new bios to be
3295 * added. __generic_make_request may indeed add some more bios
3296 * through a recursive call to generic_make_request. If it
3297 * did, we find a non-NULL value in bio_list and re-enter the loop
3298 * from the top. In this case we really did just take the bio
3299 * of the top of the list (no pretending) and so fixup bio_list and
3300 * bio_tail or bi_next, and call into __generic_make_request again.
3302 * The loop was structured like this to make only one call to
3303 * __generic_make_request (which is important as it is large and
3304 * inlined) and to keep the structure simple.
3306 BUG_ON(bio
->bi_next
);
3308 current
->bio_list
= bio
->bi_next
;
3309 if (bio
->bi_next
== NULL
)
3310 current
->bio_tail
= ¤t
->bio_list
;
3312 bio
->bi_next
= NULL
;
3313 __generic_make_request(bio
);
3314 bio
= current
->bio_list
;
3316 current
->bio_tail
= NULL
; /* deactivate */
3319 EXPORT_SYMBOL(generic_make_request
);
3322 * submit_bio: submit a bio to the block device layer for I/O
3323 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3324 * @bio: The &struct bio which describes the I/O
3326 * submit_bio() is very similar in purpose to generic_make_request(), and
3327 * uses that function to do most of the work. Both are fairly rough
3328 * interfaces, @bio must be presetup and ready for I/O.
3331 void submit_bio(int rw
, struct bio
*bio
)
3333 int count
= bio_sectors(bio
);
3338 * If it's a regular read/write or a barrier with data attached,
3339 * go through the normal accounting stuff before submission.
3341 if (!bio_empty_barrier(bio
)) {
3343 BIO_BUG_ON(!bio
->bi_size
);
3344 BIO_BUG_ON(!bio
->bi_io_vec
);
3347 count_vm_events(PGPGOUT
, count
);
3349 task_io_account_read(bio
->bi_size
);
3350 count_vm_events(PGPGIN
, count
);
3353 if (unlikely(block_dump
)) {
3354 char b
[BDEVNAME_SIZE
];
3355 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s\n",
3356 current
->comm
, current
->pid
,
3357 (rw
& WRITE
) ? "WRITE" : "READ",
3358 (unsigned long long)bio
->bi_sector
,
3359 bdevname(bio
->bi_bdev
,b
));
3363 generic_make_request(bio
);
3366 EXPORT_SYMBOL(submit_bio
);
3368 static void blk_recalc_rq_sectors(struct request
*rq
, int nsect
)
3370 if (blk_fs_request(rq
)) {
3371 rq
->hard_sector
+= nsect
;
3372 rq
->hard_nr_sectors
-= nsect
;
3375 * Move the I/O submission pointers ahead if required.
3377 if ((rq
->nr_sectors
>= rq
->hard_nr_sectors
) &&
3378 (rq
->sector
<= rq
->hard_sector
)) {
3379 rq
->sector
= rq
->hard_sector
;
3380 rq
->nr_sectors
= rq
->hard_nr_sectors
;
3381 rq
->hard_cur_sectors
= bio_cur_sectors(rq
->bio
);
3382 rq
->current_nr_sectors
= rq
->hard_cur_sectors
;
3383 rq
->buffer
= bio_data(rq
->bio
);
3387 * if total number of sectors is less than the first segment
3388 * size, something has gone terribly wrong
3390 if (rq
->nr_sectors
< rq
->current_nr_sectors
) {
3391 printk("blk: request botched\n");
3392 rq
->nr_sectors
= rq
->current_nr_sectors
;
3397 static int __end_that_request_first(struct request
*req
, int uptodate
,
3400 int total_bytes
, bio_nbytes
, error
, next_idx
= 0;
3403 blk_add_trace_rq(req
->q
, req
, BLK_TA_COMPLETE
);
3406 * extend uptodate bool to allow < 0 value to be direct io error
3409 if (end_io_error(uptodate
))
3410 error
= !uptodate
? -EIO
: uptodate
;
3413 * for a REQ_BLOCK_PC request, we want to carry any eventual
3414 * sense key with us all the way through
3416 if (!blk_pc_request(req
))
3420 if (blk_fs_request(req
) && !(req
->cmd_flags
& REQ_QUIET
))
3421 printk("end_request: I/O error, dev %s, sector %llu\n",
3422 req
->rq_disk
? req
->rq_disk
->disk_name
: "?",
3423 (unsigned long long)req
->sector
);
3426 if (blk_fs_request(req
) && req
->rq_disk
) {
3427 const int rw
= rq_data_dir(req
);
3429 disk_stat_add(req
->rq_disk
, sectors
[rw
], nr_bytes
>> 9);
3432 total_bytes
= bio_nbytes
= 0;
3433 while ((bio
= req
->bio
) != NULL
) {
3437 * For an empty barrier request, the low level driver must
3438 * store a potential error location in ->sector. We pass
3439 * that back up in ->bi_sector.
3441 if (blk_empty_barrier(req
))
3442 bio
->bi_sector
= req
->sector
;
3444 if (nr_bytes
>= bio
->bi_size
) {
3445 req
->bio
= bio
->bi_next
;
3446 nbytes
= bio
->bi_size
;
3447 req_bio_endio(req
, bio
, nbytes
, error
);
3451 int idx
= bio
->bi_idx
+ next_idx
;
3453 if (unlikely(bio
->bi_idx
>= bio
->bi_vcnt
)) {
3454 blk_dump_rq_flags(req
, "__end_that");
3455 printk("%s: bio idx %d >= vcnt %d\n",
3457 bio
->bi_idx
, bio
->bi_vcnt
);
3461 nbytes
= bio_iovec_idx(bio
, idx
)->bv_len
;
3462 BIO_BUG_ON(nbytes
> bio
->bi_size
);
3465 * not a complete bvec done
3467 if (unlikely(nbytes
> nr_bytes
)) {
3468 bio_nbytes
+= nr_bytes
;
3469 total_bytes
+= nr_bytes
;
3474 * advance to the next vector
3477 bio_nbytes
+= nbytes
;
3480 total_bytes
+= nbytes
;
3483 if ((bio
= req
->bio
)) {
3485 * end more in this run, or just return 'not-done'
3487 if (unlikely(nr_bytes
<= 0))
3499 * if the request wasn't completed, update state
3502 req_bio_endio(req
, bio
, bio_nbytes
, error
);
3503 bio
->bi_idx
+= next_idx
;
3504 bio_iovec(bio
)->bv_offset
+= nr_bytes
;
3505 bio_iovec(bio
)->bv_len
-= nr_bytes
;
3508 blk_recalc_rq_sectors(req
, total_bytes
>> 9);
3509 blk_recalc_rq_segments(req
);
3514 * end_that_request_first - end I/O on a request
3515 * @req: the request being processed
3516 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3517 * @nr_sectors: number of sectors to end I/O on
3520 * Ends I/O on a number of sectors attached to @req, and sets it up
3521 * for the next range of segments (if any) in the cluster.
3524 * 0 - we are done with this request, call end_that_request_last()
3525 * 1 - still buffers pending for this request
3527 int end_that_request_first(struct request
*req
, int uptodate
, int nr_sectors
)
3529 return __end_that_request_first(req
, uptodate
, nr_sectors
<< 9);
3532 EXPORT_SYMBOL(end_that_request_first
);
3535 * end_that_request_chunk - end I/O on a request
3536 * @req: the request being processed
3537 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3538 * @nr_bytes: number of bytes to complete
3541 * Ends I/O on a number of bytes attached to @req, and sets it up
3542 * for the next range of segments (if any). Like end_that_request_first(),
3543 * but deals with bytes instead of sectors.
3546 * 0 - we are done with this request, call end_that_request_last()
3547 * 1 - still buffers pending for this request
3549 int end_that_request_chunk(struct request
*req
, int uptodate
, int nr_bytes
)
3551 return __end_that_request_first(req
, uptodate
, nr_bytes
);
3554 EXPORT_SYMBOL(end_that_request_chunk
);
3557 * splice the completion data to a local structure and hand off to
3558 * process_completion_queue() to complete the requests
3560 static void blk_done_softirq(struct softirq_action
*h
)
3562 struct list_head
*cpu_list
, local_list
;
3564 local_irq_disable();
3565 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3566 list_replace_init(cpu_list
, &local_list
);
3569 while (!list_empty(&local_list
)) {
3570 struct request
*rq
= list_entry(local_list
.next
, struct request
, donelist
);
3572 list_del_init(&rq
->donelist
);
3573 rq
->q
->softirq_done_fn(rq
);
3577 static int __cpuinit
blk_cpu_notify(struct notifier_block
*self
, unsigned long action
,
3581 * If a CPU goes away, splice its entries to the current CPU
3582 * and trigger a run of the softirq
3584 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3585 int cpu
= (unsigned long) hcpu
;
3587 local_irq_disable();
3588 list_splice_init(&per_cpu(blk_cpu_done
, cpu
),
3589 &__get_cpu_var(blk_cpu_done
));
3590 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3598 static struct notifier_block blk_cpu_notifier __cpuinitdata
= {
3599 .notifier_call
= blk_cpu_notify
,
3603 * blk_complete_request - end I/O on a request
3604 * @req: the request being processed
3607 * Ends all I/O on a request. It does not handle partial completions,
3608 * unless the driver actually implements this in its completion callback
3609 * through requeueing. The actual completion happens out-of-order,
3610 * through a softirq handler. The user must have registered a completion
3611 * callback through blk_queue_softirq_done().
3614 void blk_complete_request(struct request
*req
)
3616 struct list_head
*cpu_list
;
3617 unsigned long flags
;
3619 BUG_ON(!req
->q
->softirq_done_fn
);
3621 local_irq_save(flags
);
3623 cpu_list
= &__get_cpu_var(blk_cpu_done
);
3624 list_add_tail(&req
->donelist
, cpu_list
);
3625 raise_softirq_irqoff(BLOCK_SOFTIRQ
);
3627 local_irq_restore(flags
);
3630 EXPORT_SYMBOL(blk_complete_request
);
3633 * queue lock must be held
3635 void end_that_request_last(struct request
*req
, int uptodate
)
3637 struct gendisk
*disk
= req
->rq_disk
;
3641 * extend uptodate bool to allow < 0 value to be direct io error
3644 if (end_io_error(uptodate
))
3645 error
= !uptodate
? -EIO
: uptodate
;
3647 if (unlikely(laptop_mode
) && blk_fs_request(req
))
3648 laptop_io_completion();
3651 * Account IO completion. bar_rq isn't accounted as a normal
3652 * IO on queueing nor completion. Accounting the containing
3653 * request is enough.
3655 if (disk
&& blk_fs_request(req
) && req
!= &req
->q
->bar_rq
) {
3656 unsigned long duration
= jiffies
- req
->start_time
;
3657 const int rw
= rq_data_dir(req
);
3659 __disk_stat_inc(disk
, ios
[rw
]);
3660 __disk_stat_add(disk
, ticks
[rw
], duration
);
3661 disk_round_stats(disk
);
3665 req
->end_io(req
, error
);
3667 __blk_put_request(req
->q
, req
);
3670 EXPORT_SYMBOL(end_that_request_last
);
3672 static inline void __end_request(struct request
*rq
, int uptodate
,
3673 unsigned int nr_bytes
, int dequeue
)
3675 if (!end_that_request_chunk(rq
, uptodate
, nr_bytes
)) {
3677 blkdev_dequeue_request(rq
);
3678 add_disk_randomness(rq
->rq_disk
);
3679 end_that_request_last(rq
, uptodate
);
3683 static unsigned int rq_byte_size(struct request
*rq
)
3685 if (blk_fs_request(rq
))
3686 return rq
->hard_nr_sectors
<< 9;
3688 return rq
->data_len
;
3692 * end_queued_request - end all I/O on a queued request
3693 * @rq: the request being processed
3694 * @uptodate: error value or 0/1 uptodate flag
3697 * Ends all I/O on a request, and removes it from the block layer queues.
3698 * Not suitable for normal IO completion, unless the driver still has
3699 * the request attached to the block layer.
3702 void end_queued_request(struct request
*rq
, int uptodate
)
3704 __end_request(rq
, uptodate
, rq_byte_size(rq
), 1);
3706 EXPORT_SYMBOL(end_queued_request
);
3709 * end_dequeued_request - end all I/O on a dequeued request
3710 * @rq: the request being processed
3711 * @uptodate: error value or 0/1 uptodate flag
3714 * Ends all I/O on a request. The request must already have been
3715 * dequeued using blkdev_dequeue_request(), as is normally the case
3719 void end_dequeued_request(struct request
*rq
, int uptodate
)
3721 __end_request(rq
, uptodate
, rq_byte_size(rq
), 0);
3723 EXPORT_SYMBOL(end_dequeued_request
);
3727 * end_request - end I/O on the current segment of the request
3728 * @rq: the request being processed
3729 * @uptodate: error value or 0/1 uptodate flag
3732 * Ends I/O on the current segment of a request. If that is the only
3733 * remaining segment, the request is also completed and freed.
3735 * This is a remnant of how older block drivers handled IO completions.
3736 * Modern drivers typically end IO on the full request in one go, unless
3737 * they have a residual value to account for. For that case this function
3738 * isn't really useful, unless the residual just happens to be the
3739 * full current segment. In other words, don't use this function in new
3740 * code. Either use end_request_completely(), or the
3741 * end_that_request_chunk() (along with end_that_request_last()) for
3742 * partial completions.
3745 void end_request(struct request
*req
, int uptodate
)
3747 __end_request(req
, uptodate
, req
->hard_cur_sectors
<< 9, 1);
3749 EXPORT_SYMBOL(end_request
);
3751 static void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3754 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3755 rq
->cmd_flags
|= (bio
->bi_rw
& 3);
3757 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3758 rq
->nr_hw_segments
= bio_hw_segments(q
, bio
);
3759 rq
->current_nr_sectors
= bio_cur_sectors(bio
);
3760 rq
->hard_cur_sectors
= rq
->current_nr_sectors
;
3761 rq
->hard_nr_sectors
= rq
->nr_sectors
= bio_sectors(bio
);
3762 rq
->buffer
= bio_data(bio
);
3763 rq
->data_len
= bio
->bi_size
;
3765 rq
->bio
= rq
->biotail
= bio
;
3768 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
3771 int kblockd_schedule_work(struct work_struct
*work
)
3773 return queue_work(kblockd_workqueue
, work
);
3776 EXPORT_SYMBOL(kblockd_schedule_work
);
3778 void kblockd_flush_work(struct work_struct
*work
)
3780 cancel_work_sync(work
);
3782 EXPORT_SYMBOL(kblockd_flush_work
);
3784 int __init
blk_dev_init(void)
3788 kblockd_workqueue
= create_workqueue("kblockd");
3789 if (!kblockd_workqueue
)
3790 panic("Failed to create kblockd\n");
3792 request_cachep
= kmem_cache_create("blkdev_requests",
3793 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3795 requestq_cachep
= kmem_cache_create("blkdev_queue",
3796 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3798 iocontext_cachep
= kmem_cache_create("blkdev_ioc",
3799 sizeof(struct io_context
), 0, SLAB_PANIC
, NULL
);
3801 for_each_possible_cpu(i
)
3802 INIT_LIST_HEAD(&per_cpu(blk_cpu_done
, i
));
3804 open_softirq(BLOCK_SOFTIRQ
, blk_done_softirq
, NULL
);
3805 register_hotcpu_notifier(&blk_cpu_notifier
);
3807 blk_max_low_pfn
= max_low_pfn
- 1;
3808 blk_max_pfn
= max_pfn
- 1;
3814 * IO Context helper functions
3816 void put_io_context(struct io_context
*ioc
)
3821 BUG_ON(atomic_read(&ioc
->refcount
) == 0);
3823 if (atomic_dec_and_test(&ioc
->refcount
)) {
3824 struct cfq_io_context
*cic
;
3827 if (ioc
->aic
&& ioc
->aic
->dtor
)
3828 ioc
->aic
->dtor(ioc
->aic
);
3829 if (ioc
->cic_root
.rb_node
!= NULL
) {
3830 struct rb_node
*n
= rb_first(&ioc
->cic_root
);
3832 cic
= rb_entry(n
, struct cfq_io_context
, rb_node
);
3837 kmem_cache_free(iocontext_cachep
, ioc
);
3840 EXPORT_SYMBOL(put_io_context
);
3842 /* Called by the exitting task */
3843 void exit_io_context(void)
3845 struct io_context
*ioc
;
3846 struct cfq_io_context
*cic
;
3849 ioc
= current
->io_context
;
3850 current
->io_context
= NULL
;
3851 task_unlock(current
);
3854 if (ioc
->aic
&& ioc
->aic
->exit
)
3855 ioc
->aic
->exit(ioc
->aic
);
3856 if (ioc
->cic_root
.rb_node
!= NULL
) {
3857 cic
= rb_entry(rb_first(&ioc
->cic_root
), struct cfq_io_context
, rb_node
);
3861 put_io_context(ioc
);
3865 * If the current task has no IO context then create one and initialise it.
3866 * Otherwise, return its existing IO context.
3868 * This returned IO context doesn't have a specifically elevated refcount,
3869 * but since the current task itself holds a reference, the context can be
3870 * used in general code, so long as it stays within `current` context.
3872 static struct io_context
*current_io_context(gfp_t gfp_flags
, int node
)
3874 struct task_struct
*tsk
= current
;
3875 struct io_context
*ret
;
3877 ret
= tsk
->io_context
;
3881 ret
= kmem_cache_alloc_node(iocontext_cachep
, gfp_flags
, node
);
3883 atomic_set(&ret
->refcount
, 1);
3884 ret
->task
= current
;
3885 ret
->ioprio_changed
= 0;
3886 ret
->last_waited
= jiffies
; /* doesn't matter... */
3887 ret
->nr_batch_requests
= 0; /* because this is 0 */
3889 ret
->cic_root
.rb_node
= NULL
;
3890 ret
->ioc_data
= NULL
;
3891 /* make sure set_task_ioprio() sees the settings above */
3893 tsk
->io_context
= ret
;
3900 * If the current task has no IO context then create one and initialise it.
3901 * If it does have a context, take a ref on it.
3903 * This is always called in the context of the task which submitted the I/O.
3905 struct io_context
*get_io_context(gfp_t gfp_flags
, int node
)
3907 struct io_context
*ret
;
3908 ret
= current_io_context(gfp_flags
, node
);
3910 atomic_inc(&ret
->refcount
);
3913 EXPORT_SYMBOL(get_io_context
);
3915 void copy_io_context(struct io_context
**pdst
, struct io_context
**psrc
)
3917 struct io_context
*src
= *psrc
;
3918 struct io_context
*dst
= *pdst
;
3921 BUG_ON(atomic_read(&src
->refcount
) == 0);
3922 atomic_inc(&src
->refcount
);
3923 put_io_context(dst
);
3927 EXPORT_SYMBOL(copy_io_context
);
3929 void swap_io_context(struct io_context
**ioc1
, struct io_context
**ioc2
)
3931 struct io_context
*temp
;
3936 EXPORT_SYMBOL(swap_io_context
);
3941 struct queue_sysfs_entry
{
3942 struct attribute attr
;
3943 ssize_t (*show
)(struct request_queue
*, char *);
3944 ssize_t (*store
)(struct request_queue
*, const char *, size_t);
3948 queue_var_show(unsigned int var
, char *page
)
3950 return sprintf(page
, "%d\n", var
);
3954 queue_var_store(unsigned long *var
, const char *page
, size_t count
)
3956 char *p
= (char *) page
;
3958 *var
= simple_strtoul(p
, &p
, 10);
3962 static ssize_t
queue_requests_show(struct request_queue
*q
, char *page
)
3964 return queue_var_show(q
->nr_requests
, (page
));
3968 queue_requests_store(struct request_queue
*q
, const char *page
, size_t count
)
3970 struct request_list
*rl
= &q
->rq
;
3972 int ret
= queue_var_store(&nr
, page
, count
);
3973 if (nr
< BLKDEV_MIN_RQ
)
3976 spin_lock_irq(q
->queue_lock
);
3977 q
->nr_requests
= nr
;
3978 blk_queue_congestion_threshold(q
);
3980 if (rl
->count
[READ
] >= queue_congestion_on_threshold(q
))
3981 blk_set_queue_congested(q
, READ
);
3982 else if (rl
->count
[READ
] < queue_congestion_off_threshold(q
))
3983 blk_clear_queue_congested(q
, READ
);
3985 if (rl
->count
[WRITE
] >= queue_congestion_on_threshold(q
))
3986 blk_set_queue_congested(q
, WRITE
);
3987 else if (rl
->count
[WRITE
] < queue_congestion_off_threshold(q
))
3988 blk_clear_queue_congested(q
, WRITE
);
3990 if (rl
->count
[READ
] >= q
->nr_requests
) {
3991 blk_set_queue_full(q
, READ
);
3992 } else if (rl
->count
[READ
]+1 <= q
->nr_requests
) {
3993 blk_clear_queue_full(q
, READ
);
3994 wake_up(&rl
->wait
[READ
]);
3997 if (rl
->count
[WRITE
] >= q
->nr_requests
) {
3998 blk_set_queue_full(q
, WRITE
);
3999 } else if (rl
->count
[WRITE
]+1 <= q
->nr_requests
) {
4000 blk_clear_queue_full(q
, WRITE
);
4001 wake_up(&rl
->wait
[WRITE
]);
4003 spin_unlock_irq(q
->queue_lock
);
4007 static ssize_t
queue_ra_show(struct request_queue
*q
, char *page
)
4009 int ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4011 return queue_var_show(ra_kb
, (page
));
4015 queue_ra_store(struct request_queue
*q
, const char *page
, size_t count
)
4017 unsigned long ra_kb
;
4018 ssize_t ret
= queue_var_store(&ra_kb
, page
, count
);
4020 spin_lock_irq(q
->queue_lock
);
4021 q
->backing_dev_info
.ra_pages
= ra_kb
>> (PAGE_CACHE_SHIFT
- 10);
4022 spin_unlock_irq(q
->queue_lock
);
4027 static ssize_t
queue_max_sectors_show(struct request_queue
*q
, char *page
)
4029 int max_sectors_kb
= q
->max_sectors
>> 1;
4031 return queue_var_show(max_sectors_kb
, (page
));
4035 queue_max_sectors_store(struct request_queue
*q
, const char *page
, size_t count
)
4037 unsigned long max_sectors_kb
,
4038 max_hw_sectors_kb
= q
->max_hw_sectors
>> 1,
4039 page_kb
= 1 << (PAGE_CACHE_SHIFT
- 10);
4040 ssize_t ret
= queue_var_store(&max_sectors_kb
, page
, count
);
4043 if (max_sectors_kb
> max_hw_sectors_kb
|| max_sectors_kb
< page_kb
)
4046 * Take the queue lock to update the readahead and max_sectors
4047 * values synchronously:
4049 spin_lock_irq(q
->queue_lock
);
4051 * Trim readahead window as well, if necessary:
4053 ra_kb
= q
->backing_dev_info
.ra_pages
<< (PAGE_CACHE_SHIFT
- 10);
4054 if (ra_kb
> max_sectors_kb
)
4055 q
->backing_dev_info
.ra_pages
=
4056 max_sectors_kb
>> (PAGE_CACHE_SHIFT
- 10);
4058 q
->max_sectors
= max_sectors_kb
<< 1;
4059 spin_unlock_irq(q
->queue_lock
);
4064 static ssize_t
queue_max_hw_sectors_show(struct request_queue
*q
, char *page
)
4066 int max_hw_sectors_kb
= q
->max_hw_sectors
>> 1;
4068 return queue_var_show(max_hw_sectors_kb
, (page
));
4072 static struct queue_sysfs_entry queue_requests_entry
= {
4073 .attr
= {.name
= "nr_requests", .mode
= S_IRUGO
| S_IWUSR
},
4074 .show
= queue_requests_show
,
4075 .store
= queue_requests_store
,
4078 static struct queue_sysfs_entry queue_ra_entry
= {
4079 .attr
= {.name
= "read_ahead_kb", .mode
= S_IRUGO
| S_IWUSR
},
4080 .show
= queue_ra_show
,
4081 .store
= queue_ra_store
,
4084 static struct queue_sysfs_entry queue_max_sectors_entry
= {
4085 .attr
= {.name
= "max_sectors_kb", .mode
= S_IRUGO
| S_IWUSR
},
4086 .show
= queue_max_sectors_show
,
4087 .store
= queue_max_sectors_store
,
4090 static struct queue_sysfs_entry queue_max_hw_sectors_entry
= {
4091 .attr
= {.name
= "max_hw_sectors_kb", .mode
= S_IRUGO
},
4092 .show
= queue_max_hw_sectors_show
,
4095 static struct queue_sysfs_entry queue_iosched_entry
= {
4096 .attr
= {.name
= "scheduler", .mode
= S_IRUGO
| S_IWUSR
},
4097 .show
= elv_iosched_show
,
4098 .store
= elv_iosched_store
,
4101 static struct attribute
*default_attrs
[] = {
4102 &queue_requests_entry
.attr
,
4103 &queue_ra_entry
.attr
,
4104 &queue_max_hw_sectors_entry
.attr
,
4105 &queue_max_sectors_entry
.attr
,
4106 &queue_iosched_entry
.attr
,
4110 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4113 queue_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4115 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4116 struct request_queue
*q
=
4117 container_of(kobj
, struct request_queue
, kobj
);
4122 mutex_lock(&q
->sysfs_lock
);
4123 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4124 mutex_unlock(&q
->sysfs_lock
);
4127 res
= entry
->show(q
, page
);
4128 mutex_unlock(&q
->sysfs_lock
);
4133 queue_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4134 const char *page
, size_t length
)
4136 struct queue_sysfs_entry
*entry
= to_queue(attr
);
4137 struct request_queue
*q
= container_of(kobj
, struct request_queue
, kobj
);
4143 mutex_lock(&q
->sysfs_lock
);
4144 if (test_bit(QUEUE_FLAG_DEAD
, &q
->queue_flags
)) {
4145 mutex_unlock(&q
->sysfs_lock
);
4148 res
= entry
->store(q
, page
, length
);
4149 mutex_unlock(&q
->sysfs_lock
);
4153 static struct sysfs_ops queue_sysfs_ops
= {
4154 .show
= queue_attr_show
,
4155 .store
= queue_attr_store
,
4158 static struct kobj_type queue_ktype
= {
4159 .sysfs_ops
= &queue_sysfs_ops
,
4160 .default_attrs
= default_attrs
,
4161 .release
= blk_release_queue
,
4164 int blk_register_queue(struct gendisk
*disk
)
4168 struct request_queue
*q
= disk
->queue
;
4170 if (!q
|| !q
->request_fn
)
4173 q
->kobj
.parent
= kobject_get(&disk
->kobj
);
4175 ret
= kobject_add(&q
->kobj
);
4179 kobject_uevent(&q
->kobj
, KOBJ_ADD
);
4181 ret
= elv_register_queue(q
);
4183 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4184 kobject_del(&q
->kobj
);
4191 void blk_unregister_queue(struct gendisk
*disk
)
4193 struct request_queue
*q
= disk
->queue
;
4195 if (q
&& q
->request_fn
) {
4196 elv_unregister_queue(q
);
4198 kobject_uevent(&q
->kobj
, KOBJ_REMOVE
);
4199 kobject_del(&q
->kobj
);
4200 kobject_put(&disk
->kobj
);