2 * linux/drivers/block/ll_rw_blk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
6 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
7 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
8 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
9 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
13 * This handles all read/write requests to block devices
15 #include <linux/config.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/backing-dev.h>
19 #include <linux/bio.h>
20 #include <linux/blkdev.h>
21 #include <linux/highmem.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/string.h>
25 #include <linux/init.h>
26 #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
35 #include <scsi/scsi_cmnd.h>
37 static void blk_unplug_work(void *data);
38 static void blk_unplug_timeout(unsigned long data);
41 * For the allocated request tables
43 static kmem_cache_t *request_cachep;
45 static wait_queue_head_t congestion_wqh[2] = {
46 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
47 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
51 * Controlling structure to kblockd
53 static struct workqueue_struct *kblockd_workqueue;
55 unsigned long blk_max_low_pfn, blk_max_pfn;
57 EXPORT_SYMBOL(blk_max_low_pfn);
58 EXPORT_SYMBOL(blk_max_pfn);
60 /* Amount of time in which a process may batch requests */
61 #define BLK_BATCH_TIME (HZ/50UL)
63 /* Number of requests a "batching" process may submit */
64 #define BLK_BATCH_REQ 32
67 * Return the threshold (number of used requests) at which the queue is
68 * considered to be congested. It include a little hysteresis to keep the
69 * context switch rate down.
71 static inline int queue_congestion_on_threshold(struct request_queue *q)
75 ret = q->nr_requests - (q->nr_requests / 8) + 1;
77 if (ret > q->nr_requests)
84 * The threshold at which a queue is considered to be uncongested
86 static inline int queue_congestion_off_threshold(struct request_queue *q)
90 ret = q->nr_requests - (q->nr_requests / 8) - 1;
99 * A queue has just exitted congestion. Note this in the global counter of
100 * congested queues, and wake up anyone who was waiting for requests to be
103 static void clear_queue_congested(request_queue_t *q, int rw)
106 wait_queue_head_t *wqh = &congestion_wqh[rw];
108 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
109 clear_bit(bit, &q->backing_dev_info.state);
110 smp_mb__after_clear_bit();
111 if (waitqueue_active(wqh))
116 * A queue has just entered congestion. Flag that in the queue's VM-visible
117 * state flags and increment the global gounter of congested queues.
119 static void set_queue_congested(request_queue_t *q, int rw)
123 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
124 set_bit(bit, &q->backing_dev_info.state);
128 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
131 * Locates the passed device's request queue and returns the address of its
134 * Will return NULL if the request queue cannot be located.
136 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
138 struct backing_dev_info *ret = NULL;
139 request_queue_t *q = bdev_get_queue(bdev);
142 ret = &q->backing_dev_info;
146 void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
149 q->activity_data = data;
152 EXPORT_SYMBOL(blk_queue_activity_fn);
155 * blk_queue_prep_rq - set a prepare_request function for queue
157 * @pfn: prepare_request function
159 * It's possible for a queue to register a prepare_request callback which
160 * is invoked before the request is handed to the request_fn. The goal of
161 * the function is to prepare a request for I/O, it can be used to build a
162 * cdb from the request data for instance.
165 void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
170 EXPORT_SYMBOL(blk_queue_prep_rq);
173 * blk_queue_merge_bvec - set a merge_bvec function for queue
175 * @mbfn: merge_bvec_fn
177 * Usually queues have static limitations on the max sectors or segments that
178 * we can put in a request. Stacking drivers may have some settings that
179 * are dynamic, and thus we have to query the queue whether it is ok to
180 * add a new bio_vec to a bio at a given offset or not. If the block device
181 * has such limitations, it needs to register a merge_bvec_fn to control
182 * the size of bio's sent to it. Note that a block device *must* allow a
183 * single page to be added to an empty bio. The block device driver may want
184 * to use the bio_split() function to deal with these bio's. By default
185 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
188 void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
190 q->merge_bvec_fn = mbfn;
193 EXPORT_SYMBOL(blk_queue_merge_bvec);
196 * blk_queue_make_request - define an alternate make_request function for a device
197 * @q: the request queue for the device to be affected
198 * @mfn: the alternate make_request function
201 * The normal way for &struct bios to be passed to a device
202 * driver is for them to be collected into requests on a request
203 * queue, and then to allow the device driver to select requests
204 * off that queue when it is ready. This works well for many block
205 * devices. However some block devices (typically virtual devices
206 * such as md or lvm) do not benefit from the processing on the
207 * request queue, and are served best by having the requests passed
208 * directly to them. This can be achieved by providing a function
209 * to blk_queue_make_request().
212 * The driver that does this *must* be able to deal appropriately
213 * with buffers in "highmemory". This can be accomplished by either calling
214 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
215 * blk_queue_bounce() to create a buffer in normal memory.
217 void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
222 q->nr_requests = BLKDEV_MAX_RQ;
223 q->max_phys_segments = MAX_PHYS_SEGMENTS;
224 q->max_hw_segments = MAX_HW_SEGMENTS;
225 q->make_request_fn = mfn;
226 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
227 q->backing_dev_info.state = 0;
228 q->backing_dev_info.memory_backed = 0;
229 blk_queue_max_sectors(q, MAX_SECTORS);
230 blk_queue_hardsect_size(q, 512);
231 blk_queue_dma_alignment(q, 511);
233 q->unplug_thresh = 4; /* hmm */
234 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
235 if (q->unplug_delay == 0)
238 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
240 q->unplug_timer.function = blk_unplug_timeout;
241 q->unplug_timer.data = (unsigned long)q;
244 * by default assume old behaviour and bounce for any highmem page
246 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
248 blk_queue_activity_fn(q, NULL, NULL);
251 EXPORT_SYMBOL(blk_queue_make_request);
254 * blk_queue_bounce_limit - set bounce buffer limit for queue
255 * @q: the request queue for the device
256 * @dma_addr: bus address limit
259 * Different hardware can have different requirements as to what pages
260 * it can do I/O directly to. A low level driver can call
261 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
262 * buffers for doing I/O to pages residing above @page. By default
263 * the block layer sets this to the highest numbered "low" memory page.
265 void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
267 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
268 unsigned long mb = dma_addr >> 20;
269 static request_queue_t *last_q;
272 * set appropriate bounce gfp mask -- unfortunately we don't have a
273 * full 4GB zone, so we have to resort to low memory for any bounces.
274 * ISA has its own < 16MB zone.
276 if (bounce_pfn < blk_max_low_pfn) {
277 BUG_ON(dma_addr < BLK_BOUNCE_ISA);
278 init_emergency_isa_pool();
279 q->bounce_gfp = GFP_NOIO | GFP_DMA;
281 q->bounce_gfp = GFP_NOIO;
284 * keep this for debugging for now...
286 if (dma_addr != BLK_BOUNCE_HIGH && q != last_q) {
287 printk("blk: queue %p, ", q);
288 if (dma_addr == BLK_BOUNCE_ANY)
289 printk("no I/O memory limit\n");
291 printk("I/O limit %luMb (mask 0x%Lx)\n", mb, (long long) dma_addr);
294 q->bounce_pfn = bounce_pfn;
298 EXPORT_SYMBOL(blk_queue_bounce_limit);
301 * blk_queue_max_sectors - set max sectors for a request for this queue
302 * @q: the request queue for the device
303 * @max_sectors: max sectors in the usual 512b unit
306 * Enables a low level driver to set an upper limit on the size of
309 void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors)
311 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
312 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
313 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
316 q->max_sectors = max_sectors;
319 EXPORT_SYMBOL(blk_queue_max_sectors);
322 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
323 * @q: the request queue for the device
324 * @max_segments: max number of segments
327 * Enables a low level driver to set an upper limit on the number of
328 * physical data segments in a request. This would be the largest sized
329 * scatter list the driver could handle.
331 void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
335 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
338 q->max_phys_segments = max_segments;
341 EXPORT_SYMBOL(blk_queue_max_phys_segments);
344 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
345 * @q: the request queue for the device
346 * @max_segments: max number of segments
349 * Enables a low level driver to set an upper limit on the number of
350 * hw data segments in a request. This would be the largest number of
351 * address/length pairs the host adapter can actually give as once
354 void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
358 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
361 q->max_hw_segments = max_segments;
364 EXPORT_SYMBOL(blk_queue_max_hw_segments);
367 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
368 * @q: the request queue for the device
369 * @max_size: max size of segment in bytes
372 * Enables a low level driver to set an upper limit on the size of a
375 void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
377 if (max_size < PAGE_CACHE_SIZE) {
378 max_size = PAGE_CACHE_SIZE;
379 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
382 q->max_segment_size = max_size;
385 EXPORT_SYMBOL(blk_queue_max_segment_size);
388 * blk_queue_hardsect_size - set hardware sector size for the queue
389 * @q: the request queue for the device
390 * @size: the hardware sector size, in bytes
393 * This should typically be set to the lowest possible sector size
394 * that the hardware can operate on (possible without reverting to
395 * even internal read-modify-write operations). Usually the default
396 * of 512 covers most hardware.
398 void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
400 q->hardsect_size = size;
403 EXPORT_SYMBOL(blk_queue_hardsect_size);
406 * Returns the minimum that is _not_ zero, unless both are zero.
408 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
411 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
412 * @t: the stacking driver (top)
413 * @b: the underlying device (bottom)
415 void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
417 /* zero is "infinity" */
418 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
420 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
421 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
422 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
423 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
426 EXPORT_SYMBOL(blk_queue_stack_limits);
429 * blk_queue_segment_boundary - set boundary rules for segment merging
430 * @q: the request queue for the device
431 * @mask: the memory boundary mask
433 void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
435 if (mask < PAGE_CACHE_SIZE - 1) {
436 mask = PAGE_CACHE_SIZE - 1;
437 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
440 q->seg_boundary_mask = mask;
443 EXPORT_SYMBOL(blk_queue_segment_boundary);
446 * blk_queue_dma_alignment - set dma length and memory alignment
447 * @q: the request queue for the device
448 * @mask: alignment mask
451 * set required memory and length aligment for direct dma transactions.
452 * this is used when buiding direct io requests for the queue.
455 void blk_queue_dma_alignment(request_queue_t *q, int mask)
457 q->dma_alignment = mask;
460 EXPORT_SYMBOL(blk_queue_dma_alignment);
463 * blk_queue_find_tag - find a request by its tag and queue
465 * @q: The request queue for the device
466 * @tag: The tag of the request
469 * Should be used when a device returns a tag and you want to match
472 * no locks need be held.
474 struct request *blk_queue_find_tag(request_queue_t *q, int tag)
476 struct blk_queue_tag *bqt = q->queue_tags;
478 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
481 return bqt->tag_index[tag];
484 EXPORT_SYMBOL(blk_queue_find_tag);
487 * blk_queue_free_tags - release tag maintenance info
488 * @q: the request queue for the device
491 * blk_cleanup_queue() will take care of calling this function, if tagging
492 * has been used. So there's usually no need to call this directly, unless
493 * tagging is just being disabled but the queue remains in function.
495 void blk_queue_free_tags(request_queue_t *q)
497 struct blk_queue_tag *bqt = q->queue_tags;
502 if (atomic_dec_and_test(&bqt->refcnt)) {
504 BUG_ON(!list_empty(&bqt->busy_list));
506 kfree(bqt->tag_index);
507 bqt->tag_index = NULL;
515 q->queue_tags = NULL;
516 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
519 EXPORT_SYMBOL(blk_queue_free_tags);
522 init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
526 if (depth > q->nr_requests * 2) {
527 depth = q->nr_requests * 2;
528 printk(KERN_ERR "%s: adjusted depth to %d\n",
529 __FUNCTION__, depth);
532 tags->tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
533 if (!tags->tag_index)
536 bits = (depth / BLK_TAGS_PER_LONG) + 1;
537 tags->tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC);
541 memset(tags->tag_index, 0, depth * sizeof(struct request *));
542 memset(tags->tag_map, 0, bits * sizeof(unsigned long));
543 tags->max_depth = depth;
544 tags->real_max_depth = bits * BITS_PER_LONG;
547 * set the upper bits if the depth isn't a multiple of the word size
549 for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++)
550 __set_bit(i, tags->tag_map);
552 INIT_LIST_HEAD(&tags->busy_list);
554 atomic_set(&tags->refcnt, 1);
557 kfree(tags->tag_index);
562 * blk_queue_init_tags - initialize the queue tag info
563 * @q: the request queue for the device
564 * @depth: the maximum queue depth supported
566 int blk_queue_init_tags(request_queue_t *q, int depth,
567 struct blk_queue_tag *tags)
570 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
574 if (init_tag_map(q, tags, depth))
577 atomic_inc(&tags->refcnt);
580 * assign it, all done
582 q->queue_tags = tags;
583 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
590 EXPORT_SYMBOL(blk_queue_init_tags);
593 * blk_queue_resize_tags - change the queueing depth
594 * @q: the request queue for the device
595 * @new_depth: the new max command queueing depth
598 * Must be called with the queue lock held.
600 int blk_queue_resize_tags(request_queue_t *q, int new_depth)
602 struct blk_queue_tag *bqt = q->queue_tags;
603 struct request **tag_index;
604 unsigned long *tag_map;
611 * don't bother sizing down
613 if (new_depth <= bqt->real_max_depth) {
614 bqt->max_depth = new_depth;
619 * save the old state info, so we can copy it back
621 tag_index = bqt->tag_index;
622 tag_map = bqt->tag_map;
623 max_depth = bqt->real_max_depth;
625 if (init_tag_map(q, bqt, new_depth))
628 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
629 bits = max_depth / BLK_TAGS_PER_LONG;
630 memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long));
638 * blk_queue_end_tag - end tag operations for a request
639 * @q: the request queue for the device
640 * @rq: the request that has completed
643 * Typically called when end_that_request_first() returns 0, meaning
644 * all transfers have been done for a request. It's important to call
645 * this function before end_that_request_last(), as that will put the
646 * request back on the free list thus corrupting the internal tag list.
649 * queue lock must be held.
651 void blk_queue_end_tag(request_queue_t *q, struct request *rq)
653 struct blk_queue_tag *bqt = q->queue_tags;
658 if (unlikely(tag >= bqt->real_max_depth))
661 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
662 printk("attempt to clear non-busy tag (%d)\n", tag);
666 list_del_init(&rq->queuelist);
667 rq->flags &= ~REQ_QUEUED;
670 if (unlikely(bqt->tag_index[tag] == NULL))
671 printk("tag %d is missing\n", tag);
673 bqt->tag_index[tag] = NULL;
677 EXPORT_SYMBOL(blk_queue_end_tag);
680 * blk_queue_start_tag - find a free tag and assign it
681 * @q: the request queue for the device
682 * @rq: the block request that needs tagging
685 * This can either be used as a stand-alone helper, or possibly be
686 * assigned as the queue &prep_rq_fn (in which case &struct request
687 * automagically gets a tag assigned). Note that this function
688 * assumes that any type of request can be queued! if this is not
689 * true for your device, you must check the request type before
690 * calling this function. The request will also be removed from
691 * the request queue, so it's the drivers responsibility to readd
692 * it if it should need to be restarted for some reason.
695 * queue lock must be held.
697 int blk_queue_start_tag(request_queue_t *q, struct request *rq)
699 struct blk_queue_tag *bqt = q->queue_tags;
700 unsigned long *map = bqt->tag_map;
703 if (unlikely((rq->flags & REQ_QUEUED))) {
705 "request %p for device [%s] already tagged %d",
706 rq, rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
710 for (map = bqt->tag_map; *map == -1UL; map++) {
711 tag += BLK_TAGS_PER_LONG;
713 if (tag >= bqt->max_depth)
718 __set_bit(tag, bqt->tag_map);
720 rq->flags |= REQ_QUEUED;
722 bqt->tag_index[tag] = rq;
723 blkdev_dequeue_request(rq);
724 list_add(&rq->queuelist, &bqt->busy_list);
729 EXPORT_SYMBOL(blk_queue_start_tag);
732 * blk_queue_invalidate_tags - invalidate all pending tags
733 * @q: the request queue for the device
736 * Hardware conditions may dictate a need to stop all pending requests.
737 * In this case, we will safely clear the block side of the tag queue and
738 * readd all requests to the request queue in the right order.
741 * queue lock must be held.
743 void blk_queue_invalidate_tags(request_queue_t *q)
745 struct blk_queue_tag *bqt = q->queue_tags;
746 struct list_head *tmp, *n;
749 list_for_each_safe(tmp, n, &bqt->busy_list) {
750 rq = list_entry_rq(tmp);
753 printk("bad tag found on list\n");
754 list_del_init(&rq->queuelist);
755 rq->flags &= ~REQ_QUEUED;
757 blk_queue_end_tag(q, rq);
759 rq->flags &= ~REQ_STARTED;
760 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
764 EXPORT_SYMBOL(blk_queue_invalidate_tags);
766 static char *rq_flags[] = {
784 "REQ_DRIVE_TASKFILE",
791 void blk_dump_rq_flags(struct request *rq, char *msg)
795 printk("%s: dev %s: flags = ", msg,
796 rq->rq_disk ? rq->rq_disk->disk_name : "?");
799 if (rq->flags & (1 << bit))
800 printk("%s ", rq_flags[bit]);
802 } while (bit < __REQ_NR_BITS);
804 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
806 rq->current_nr_sectors);
807 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
809 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
811 for (bit = 0; bit < sizeof(rq->cmd); bit++)
812 printk("%02x ", rq->cmd[bit]);
817 EXPORT_SYMBOL(blk_dump_rq_flags);
819 void blk_recount_segments(request_queue_t *q, struct bio *bio)
821 struct bio_vec *bv, *bvprv = NULL;
822 int i, nr_phys_segs, nr_hw_segs, seg_size, cluster;
823 int high, highprv = 1;
825 if (unlikely(!bio->bi_io_vec))
828 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
829 seg_size = nr_phys_segs = nr_hw_segs = 0;
830 bio_for_each_segment(bv, bio, i) {
832 * the trick here is making sure that a high page is never
833 * considered part of another segment, since that might
834 * change with the bounce page.
836 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
840 if (seg_size + bv->bv_len > q->max_segment_size)
842 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
844 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
847 seg_size += bv->bv_len;
852 if (!BIOVEC_VIRT_MERGEABLE(bvprv, bv))
858 seg_size = bv->bv_len;
862 bio->bi_phys_segments = nr_phys_segs;
863 bio->bi_hw_segments = nr_hw_segs;
864 bio->bi_flags |= (1 << BIO_SEG_VALID);
868 int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
871 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
874 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
876 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
880 * bio and nxt are contigous in memory, check if the queue allows
881 * these two to be merged into one
883 if (BIO_SEG_BOUNDARY(q, bio, nxt))
889 EXPORT_SYMBOL(blk_phys_contig_segment);
891 int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
894 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
897 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
899 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
903 * bio and nxt are contigous in memory, check if the queue allows
904 * these two to be merged into one
906 if (BIO_SEG_BOUNDARY(q, bio, nxt))
912 EXPORT_SYMBOL(blk_hw_contig_segment);
915 * map a request to scatterlist, return number of sg entries setup. Caller
916 * must make sure sg can hold rq->nr_phys_segments entries
918 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
920 struct bio_vec *bvec, *bvprv;
922 int nsegs, i, cluster;
925 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
931 rq_for_each_bio(bio, rq) {
933 * for each segment in bio
935 bio_for_each_segment(bvec, bio, i) {
936 int nbytes = bvec->bv_len;
938 if (bvprv && cluster) {
939 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
942 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
944 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
947 sg[nsegs - 1].length += nbytes;
950 memset(&sg[nsegs],0,sizeof(struct scatterlist));
951 sg[nsegs].page = bvec->bv_page;
952 sg[nsegs].length = nbytes;
953 sg[nsegs].offset = bvec->bv_offset;
958 } /* segments in bio */
964 EXPORT_SYMBOL(blk_rq_map_sg);
967 * the standard queue merge functions, can be overridden with device
968 * specific ones if so desired
971 static inline int ll_new_mergeable(request_queue_t *q,
975 int nr_phys_segs = bio_phys_segments(q, bio);
977 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
978 req->flags |= REQ_NOMERGE;
979 q->last_merge = NULL;
984 * A hw segment is just getting larger, bump just the phys
987 req->nr_phys_segments += nr_phys_segs;
991 static inline int ll_new_hw_segment(request_queue_t *q,
995 int nr_hw_segs = bio_hw_segments(q, bio);
996 int nr_phys_segs = bio_phys_segments(q, bio);
998 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
999 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1000 req->flags |= REQ_NOMERGE;
1001 q->last_merge = NULL;
1006 * This will form the start of a new hw segment. Bump both
1009 req->nr_hw_segments += nr_hw_segs;
1010 req->nr_phys_segments += nr_phys_segs;
1014 static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1017 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1018 req->flags |= REQ_NOMERGE;
1019 q->last_merge = NULL;
1023 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)))
1024 return ll_new_mergeable(q, req, bio);
1026 return ll_new_hw_segment(q, req, bio);
1029 static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1032 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1033 req->flags |= REQ_NOMERGE;
1034 q->last_merge = NULL;
1038 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)))
1039 return ll_new_mergeable(q, req, bio);
1041 return ll_new_hw_segment(q, req, bio);
1044 static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1045 struct request *next)
1047 int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments;
1048 int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1051 * First check if the either of the requests are re-queued
1052 * requests. Can't merge them if they are.
1054 if (req->special || next->special)
1058 * Will it become to large?
1060 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1063 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1064 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1065 total_phys_segments--;
1067 if (total_phys_segments > q->max_phys_segments)
1070 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1071 if (blk_hw_contig_segment(q, req->biotail, next->bio))
1072 total_hw_segments--;
1074 if (total_hw_segments > q->max_hw_segments)
1077 /* Merge is OK... */
1078 req->nr_phys_segments = total_phys_segments;
1079 req->nr_hw_segments = total_hw_segments;
1084 * "plug" the device if there are no outstanding requests: this will
1085 * force the transfer to start only after we have put all the requests
1088 * This is called with interrupts off and no requests on the queue and
1089 * with the queue lock held.
1091 void blk_plug_device(request_queue_t *q)
1093 WARN_ON(!irqs_disabled());
1096 * don't plug a stopped queue, it must be paired with blk_start_queue()
1097 * which will restart the queueing
1099 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1102 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1103 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1106 EXPORT_SYMBOL(blk_plug_device);
1109 * remove the queue from the plugged list, if present. called with
1110 * queue lock held and interrupts disabled.
1112 int blk_remove_plug(request_queue_t *q)
1114 WARN_ON(!irqs_disabled());
1116 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1119 del_timer(&q->unplug_timer);
1123 EXPORT_SYMBOL(blk_remove_plug);
1126 * remove the plug and let it rip..
1128 static inline void __generic_unplug_device(request_queue_t *q)
1130 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1133 if (!blk_remove_plug(q))
1137 * was plugged, fire request_fn if queue has stuff to do
1139 if (elv_next_request(q))
1144 * generic_unplug_device - fire a request queue
1145 * @data: The &request_queue_t in question
1148 * Linux uses plugging to build bigger requests queues before letting
1149 * the device have at them. If a queue is plugged, the I/O scheduler
1150 * is still adding and merging requests on the queue. Once the queue
1151 * gets unplugged, the request_fn defined for the queue is invoked and
1152 * transfers started.
1154 void generic_unplug_device(request_queue_t *q)
1156 spin_lock_irq(q->queue_lock);
1157 __generic_unplug_device(q);
1158 spin_unlock_irq(q->queue_lock);
1160 EXPORT_SYMBOL(generic_unplug_device);
1162 static void blk_backing_dev_unplug(struct backing_dev_info *bdi)
1164 request_queue_t *q = bdi->unplug_io_data;
1167 * devices don't necessarily have an ->unplug_fn defined
1173 static void blk_unplug_work(void *data)
1175 request_queue_t *q = data;
1180 static void blk_unplug_timeout(unsigned long data)
1182 request_queue_t *q = (request_queue_t *)data;
1184 kblockd_schedule_work(&q->unplug_work);
1188 * blk_start_queue - restart a previously stopped queue
1189 * @q: The &request_queue_t in question
1192 * blk_start_queue() will clear the stop flag on the queue, and call
1193 * the request_fn for the queue if it was in a stopped state when
1194 * entered. Also see blk_stop_queue(). Queue lock must be held.
1196 void blk_start_queue(request_queue_t *q)
1198 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1201 * one level of recursion is ok and is much faster than kicking
1202 * the unplug handling
1204 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1206 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1209 schedule_work(&q->unplug_work);
1213 EXPORT_SYMBOL(blk_start_queue);
1216 * blk_stop_queue - stop a queue
1217 * @q: The &request_queue_t in question
1220 * The Linux block layer assumes that a block driver will consume all
1221 * entries on the request queue when the request_fn strategy is called.
1222 * Often this will not happen, because of hardware limitations (queue
1223 * depth settings). If a device driver gets a 'queue full' response,
1224 * or if it simply chooses not to queue more I/O at one point, it can
1225 * call this function to prevent the request_fn from being called until
1226 * the driver has signalled it's ready to go again. This happens by calling
1227 * blk_start_queue() to restart queue operations. Queue lock must be held.
1229 void blk_stop_queue(request_queue_t *q)
1232 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1235 EXPORT_SYMBOL(blk_stop_queue);
1238 * blk_run_queue - run a single device queue
1239 * @q: The queue to run
1241 void blk_run_queue(struct request_queue *q)
1243 unsigned long flags;
1245 spin_lock_irqsave(q->queue_lock, flags);
1248 spin_unlock_irqrestore(q->queue_lock, flags);
1251 EXPORT_SYMBOL(blk_run_queue);
1254 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1255 * @q: the request queue to be released
1258 * blk_cleanup_queue is the pair to blk_init_queue() or
1259 * blk_queue_make_request(). It should be called when a request queue is
1260 * being released; typically when a block device is being de-registered.
1261 * Currently, its primary task it to free all the &struct request
1262 * structures that were allocated to the queue and the queue itself.
1265 * Hopefully the low level driver will have finished any
1266 * outstanding requests first...
1268 void blk_cleanup_queue(request_queue_t * q)
1270 struct request_list *rl = &q->rq;
1272 if (!atomic_dec_and_test(&q->refcnt))
1277 del_timer_sync(&q->unplug_timer);
1281 mempool_destroy(rl->rq_pool);
1283 if (blk_queue_tagged(q))
1284 blk_queue_free_tags(q);
1289 EXPORT_SYMBOL(blk_cleanup_queue);
1291 static int blk_init_free_list(request_queue_t *q)
1293 struct request_list *rl = &q->rq;
1295 rl->count[READ] = rl->count[WRITE] = 0;
1296 init_waitqueue_head(&rl->wait[READ]);
1297 init_waitqueue_head(&rl->wait[WRITE]);
1299 rl->rq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep);
1307 static int __make_request(request_queue_t *, struct bio *);
1309 static elevator_t *chosen_elevator =
1310 #if defined(CONFIG_IOSCHED_AS)
1312 #elif defined(CONFIG_IOSCHED_DEADLINE)
1314 #elif defined(CONFIG_IOSCHED_CFQ)
1316 #elif defined(CONFIG_IOSCHED_NOOP)
1320 #error "You must have at least 1 I/O scheduler selected"
1323 #if defined(CONFIG_IOSCHED_AS) || defined(CONFIG_IOSCHED_DEADLINE) || defined (CONFIG_IOSCHED_NOOP)
1324 static int __init elevator_setup(char *str)
1326 #ifdef CONFIG_IOSCHED_DEADLINE
1327 if (!strcmp(str, "deadline"))
1328 chosen_elevator = &iosched_deadline;
1330 #ifdef CONFIG_IOSCHED_AS
1331 if (!strcmp(str, "as"))
1332 chosen_elevator = &iosched_as;
1334 #ifdef CONFIG_IOSCHED_CFQ
1335 if (!strcmp(str, "cfq"))
1336 chosen_elevator = &iosched_cfq;
1338 #ifdef CONFIG_IOSCHED_NOOP
1339 if (!strcmp(str, "noop"))
1340 chosen_elevator = &elevator_noop;
1345 __setup("elevator=", elevator_setup);
1346 #endif /* CONFIG_IOSCHED_AS || CONFIG_IOSCHED_DEADLINE || CONFIG_IOSCHED_NOOP */
1348 request_queue_t *blk_alloc_queue(int gfp_mask)
1350 request_queue_t *q = kmalloc(sizeof(*q), gfp_mask);
1355 memset(q, 0, sizeof(*q));
1356 init_timer(&q->unplug_timer);
1357 atomic_set(&q->refcnt, 1);
1359 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1360 q->backing_dev_info.unplug_io_data = q;
1365 EXPORT_SYMBOL(blk_alloc_queue);
1368 * blk_init_queue - prepare a request queue for use with a block device
1369 * @rfn: The function to be called to process requests that have been
1370 * placed on the queue.
1371 * @lock: Request queue spin lock
1374 * If a block device wishes to use the standard request handling procedures,
1375 * which sorts requests and coalesces adjacent requests, then it must
1376 * call blk_init_queue(). The function @rfn will be called when there
1377 * are requests on the queue that need to be processed. If the device
1378 * supports plugging, then @rfn may not be called immediately when requests
1379 * are available on the queue, but may be called at some time later instead.
1380 * Plugged queues are generally unplugged when a buffer belonging to one
1381 * of the requests on the queue is needed, or due to memory pressure.
1383 * @rfn is not required, or even expected, to remove all requests off the
1384 * queue, but only as many as it can handle at a time. If it does leave
1385 * requests on the queue, it is responsible for arranging that the requests
1386 * get dealt with eventually.
1388 * The queue spin lock must be held while manipulating the requests on the
1391 * Function returns a pointer to the initialized request queue, or NULL if
1392 * it didn't succeed.
1395 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1396 * when the block device is deactivated (such as at module unload).
1398 request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1403 q = blk_alloc_queue(GFP_KERNEL);
1407 if (blk_init_free_list(q))
1412 printk("Using %s io scheduler\n", chosen_elevator->elevator_name);
1415 if (elevator_init(q, chosen_elevator))
1418 q->request_fn = rfn;
1419 q->back_merge_fn = ll_back_merge_fn;
1420 q->front_merge_fn = ll_front_merge_fn;
1421 q->merge_requests_fn = ll_merge_requests_fn;
1422 q->prep_rq_fn = NULL;
1423 q->unplug_fn = generic_unplug_device;
1424 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1425 q->queue_lock = lock;
1427 blk_queue_segment_boundary(q, 0xffffffff);
1429 blk_queue_make_request(q, __make_request);
1430 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1432 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1433 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1437 blk_cleanup_queue(q);
1443 EXPORT_SYMBOL(blk_init_queue);
1445 int blk_get_queue(request_queue_t *q)
1447 if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
1448 atomic_inc(&q->refcnt);
1455 EXPORT_SYMBOL(blk_get_queue);
1457 static inline void blk_free_request(request_queue_t *q, struct request *rq)
1459 elv_put_request(q, rq);
1460 mempool_free(rq, q->rq.rq_pool);
1463 static inline struct request *blk_alloc_request(request_queue_t *q,int gfp_mask)
1465 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1470 if (!elv_set_request(q, rq, gfp_mask))
1473 mempool_free(rq, q->rq.rq_pool);
1478 * ioc_batching returns true if the ioc is a valid batching request and
1479 * should be given priority access to a request.
1481 static inline int ioc_batching(struct io_context *ioc)
1487 * Make sure the process is able to allocate at least 1 request
1488 * even if the batch times out, otherwise we could theoretically
1491 return ioc->nr_batch_requests == BLK_BATCH_REQ ||
1492 (ioc->nr_batch_requests > 0
1493 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1497 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1498 * will cause the process to be a "batcher" on all queues in the system. This
1499 * is the behaviour we want though - once it gets a wakeup it should be given
1502 void ioc_set_batching(struct io_context *ioc)
1504 if (!ioc || ioc_batching(ioc))
1507 ioc->nr_batch_requests = BLK_BATCH_REQ;
1508 ioc->last_waited = jiffies;
1512 * A request has just been released. Account for it, update the full and
1513 * congestion status, wake up any waiters. Called under q->queue_lock.
1515 static void freed_request(request_queue_t *q, int rw)
1517 struct request_list *rl = &q->rq;
1520 if (rl->count[rw] < queue_congestion_off_threshold(q))
1521 clear_queue_congested(q, rw);
1522 if (rl->count[rw]+1 <= q->nr_requests) {
1523 if (waitqueue_active(&rl->wait[rw]))
1524 wake_up(&rl->wait[rw]);
1525 if (!waitqueue_active(&rl->wait[rw]))
1526 blk_clear_queue_full(q, rw);
1530 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1532 * Get a free request, queue_lock must not be held
1534 static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1536 struct request *rq = NULL;
1537 struct request_list *rl = &q->rq;
1538 struct io_context *ioc = get_io_context(gfp_mask);
1540 spin_lock_irq(q->queue_lock);
1541 if (rl->count[rw]+1 >= q->nr_requests) {
1543 * The queue will fill after this allocation, so set it as
1544 * full, and mark this process as "batching". This process
1545 * will be allowed to complete a batch of requests, others
1548 if (!blk_queue_full(q, rw)) {
1549 ioc_set_batching(ioc);
1550 blk_set_queue_full(q, rw);
1554 if (blk_queue_full(q, rw)
1555 && !ioc_batching(ioc) && !elv_may_queue(q, rw)) {
1557 * The queue is full and the allocating process is not a
1558 * "batcher", and not exempted by the IO scheduler
1560 spin_unlock_irq(q->queue_lock);
1565 if (rl->count[rw] >= queue_congestion_on_threshold(q))
1566 set_queue_congested(q, rw);
1567 spin_unlock_irq(q->queue_lock);
1569 rq = blk_alloc_request(q, gfp_mask);
1572 * Allocation failed presumably due to memory. Undo anything
1573 * we might have messed up.
1575 * Allocating task should really be put onto the front of the
1576 * wait queue, but this is pretty rare.
1578 spin_lock_irq(q->queue_lock);
1579 freed_request(q, rw);
1580 spin_unlock_irq(q->queue_lock);
1584 if (ioc_batching(ioc))
1585 ioc->nr_batch_requests--;
1587 INIT_LIST_HEAD(&rq->queuelist);
1590 * first three bits are identical in rq->flags and bio->bi_rw,
1591 * see bio.h and blkdev.h
1596 rq->rq_status = RQ_ACTIVE;
1597 rq->bio = rq->biotail = NULL;
1609 put_io_context(ioc);
1614 * No available requests for this queue, unplug the device and wait for some
1615 * requests to become available.
1617 static struct request *get_request_wait(request_queue_t *q, int rw)
1622 generic_unplug_device(q);
1624 struct request_list *rl = &q->rq;
1626 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1627 TASK_UNINTERRUPTIBLE);
1629 rq = get_request(q, rw, GFP_NOIO);
1632 struct io_context *ioc;
1637 * After sleeping, we become a "batching" process and
1638 * will be able to allocate at least one request, and
1639 * up to a big batch of them for a small period time.
1640 * See ioc_batching, ioc_set_batching
1642 ioc = get_io_context(GFP_NOIO);
1643 ioc_set_batching(ioc);
1644 put_io_context(ioc);
1646 finish_wait(&rl->wait[rw], &wait);
1652 struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask)
1656 BUG_ON(rw != READ && rw != WRITE);
1658 if (gfp_mask & __GFP_WAIT)
1659 rq = get_request_wait(q, rw);
1661 rq = get_request(q, rw, gfp_mask);
1666 EXPORT_SYMBOL(blk_get_request);
1669 * blk_requeue_request - put a request back on queue
1670 * @q: request queue where request should be inserted
1671 * @rq: request to be inserted
1674 * Drivers often keep queueing requests until the hardware cannot accept
1675 * more, when that condition happens we need to put the request back
1676 * on the queue. Must be called with queue lock held.
1678 void blk_requeue_request(request_queue_t *q, struct request *rq)
1680 if (blk_rq_tagged(rq))
1681 blk_queue_end_tag(q, rq);
1683 elv_requeue_request(q, rq);
1686 EXPORT_SYMBOL(blk_requeue_request);
1689 * blk_insert_request - insert a special request in to a request queue
1690 * @q: request queue where request should be inserted
1691 * @rq: request to be inserted
1692 * @at_head: insert request at head or tail of queue
1693 * @data: private data
1694 * @reinsert: true if request it a reinsertion of previously processed one
1697 * Many block devices need to execute commands asynchronously, so they don't
1698 * block the whole kernel from preemption during request execution. This is
1699 * accomplished normally by inserting aritficial requests tagged as
1700 * REQ_SPECIAL in to the corresponding request queue, and letting them be
1701 * scheduled for actual execution by the request queue.
1703 * We have the option of inserting the head or the tail of the queue.
1704 * Typically we use the tail for new ioctls and so forth. We use the head
1705 * of the queue for things like a QUEUE_FULL message from a device, or a
1706 * host that is unable to accept a particular command.
1708 void blk_insert_request(request_queue_t *q, struct request *rq,
1709 int at_head, void *data, int reinsert)
1711 unsigned long flags;
1714 * tell I/O scheduler that this isn't a regular read/write (ie it
1715 * must not attempt merges on this) and that it acts as a soft
1718 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
1722 spin_lock_irqsave(q->queue_lock, flags);
1725 * If command is tagged, release the tag
1728 blk_requeue_request(q, rq);
1730 int where = ELEVATOR_INSERT_BACK;
1733 where = ELEVATOR_INSERT_FRONT;
1735 if (blk_rq_tagged(rq))
1736 blk_queue_end_tag(q, rq);
1738 drive_stat_acct(rq, rq->nr_sectors, 1);
1739 __elv_add_request(q, rq, where, 0);
1741 if (blk_queue_plugged(q))
1742 __generic_unplug_device(q);
1745 spin_unlock_irqrestore(q->queue_lock, flags);
1748 EXPORT_SYMBOL(blk_insert_request);
1751 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
1752 * @q: request queue where request should be inserted
1753 * @rw: READ or WRITE data
1754 * @ubuf: the user buffer
1755 * @len: length of user data
1758 * Data will be mapped directly for zero copy io, if possible. Otherwise
1759 * a kernel bounce buffer is used.
1761 * A matching blk_rq_unmap_user() must be issued at the end of io, while
1762 * still in process context.
1764 struct request *blk_rq_map_user(request_queue_t *q, int rw, void __user *ubuf,
1767 struct request *rq = NULL;
1772 rq = blk_get_request(q, rw, __GFP_WAIT);
1774 return ERR_PTR(-ENOMEM);
1776 bio = bio_map_user(q, NULL, (unsigned long) ubuf, len, rw == READ);
1778 int bytes = (len + 511) & ~511;
1780 buf = kmalloc(bytes, q->bounce_gfp | GFP_USER);
1787 if (copy_from_user(buf, ubuf, len)) {
1792 memset(buf, 0, len);
1795 rq->bio = rq->biotail = bio;
1797 blk_rq_bio_prep(q, rq, bio);
1799 rq->buffer = rq->data = buf;
1806 bio_unmap_user(bio, 1);
1808 blk_put_request(rq);
1810 return ERR_PTR(ret);
1813 EXPORT_SYMBOL(blk_rq_map_user);
1816 * blk_rq_unmap_user - unmap a request with user data
1817 * @rq: request to be unmapped
1818 * @ubuf: user buffer
1819 * @ulen: length of user buffer
1822 * Unmap a request previously mapped by blk_rq_map_user().
1824 int blk_rq_unmap_user(struct request *rq, void __user *ubuf, struct bio *bio,
1827 const int read = rq_data_dir(rq) == READ;
1831 bio_unmap_user(bio, read);
1833 if (read && copy_to_user(ubuf, rq->buffer, ulen))
1838 blk_put_request(rq);
1842 EXPORT_SYMBOL(blk_rq_unmap_user);
1845 * blk_execute_rq - insert a request into queue for execution
1846 * @q: queue to insert the request in
1847 * @bd_disk: matching gendisk
1848 * @rq: request to insert
1851 * Insert a fully prepared request at the back of the io scheduler queue
1854 int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
1857 DECLARE_COMPLETION(wait);
1858 char sense[SCSI_SENSE_BUFFERSIZE];
1861 rq->rq_disk = bd_disk;
1864 * we need an extra reference to the request, so we can look at
1865 * it after io completion
1870 memset(sense, 0, sizeof(sense));
1875 rq->flags |= REQ_NOMERGE;
1876 rq->waiting = &wait;
1877 elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
1878 generic_unplug_device(q);
1879 wait_for_completion(&wait);
1887 EXPORT_SYMBOL(blk_execute_rq);
1889 void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
1891 int rw = rq_data_dir(rq);
1893 if (!blk_fs_request(rq) || !rq->rq_disk)
1897 disk_stat_add(rq->rq_disk, read_sectors, nr_sectors);
1899 disk_stat_inc(rq->rq_disk, read_merges);
1900 } else if (rw == WRITE) {
1901 disk_stat_add(rq->rq_disk, write_sectors, nr_sectors);
1903 disk_stat_inc(rq->rq_disk, write_merges);
1906 disk_round_stats(rq->rq_disk);
1907 rq->rq_disk->in_flight++;
1912 * add-request adds a request to the linked list.
1913 * queue lock is held and interrupts disabled, as we muck with the
1914 * request queue list.
1916 static inline void add_request(request_queue_t * q, struct request * req)
1918 drive_stat_acct(req, req->nr_sectors, 1);
1921 q->activity_fn(q->activity_data, rq_data_dir(req));
1924 * elevator indicated where it wants this request to be
1925 * inserted at elevator_merge time
1927 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1931 * disk_round_stats() - Round off the performance stats on a struct
1934 * The average IO queue length and utilisation statistics are maintained
1935 * by observing the current state of the queue length and the amount of
1936 * time it has been in this state for.
1938 * Normally, that accounting is done on IO completion, but that can result
1939 * in more than a second's worth of IO being accounted for within any one
1940 * second, leading to >100% utilisation. To deal with that, we call this
1941 * function to do a round-off before returning the results when reading
1942 * /proc/diskstats. This accounts immediately for all queue usage up to
1943 * the current jiffies and restarts the counters again.
1945 void disk_round_stats(struct gendisk *disk)
1947 unsigned long now = jiffies;
1949 disk_stat_add(disk, time_in_queue,
1950 disk->in_flight * (now - disk->stamp));
1953 if (disk->in_flight)
1954 disk_stat_add(disk, io_ticks, (now - disk->stamp_idle));
1955 disk->stamp_idle = now;
1959 * queue lock must be held
1961 void __blk_put_request(request_queue_t *q, struct request *req)
1963 struct request_list *rl = req->rl;
1967 if (unlikely(--req->ref_count))
1970 req->rq_status = RQ_INACTIVE;
1975 * Request may not have originated from ll_rw_blk. if not,
1976 * it didn't come out of our reserved rq pools
1979 int rw = rq_data_dir(req);
1981 elv_completed_request(q, req);
1983 BUG_ON(!list_empty(&req->queuelist));
1985 blk_free_request(q, req);
1986 freed_request(q, rw);
1990 void blk_put_request(struct request *req)
1993 * if req->rl isn't set, this request didnt originate from the
1994 * block layer, so it's safe to just disregard it
1997 unsigned long flags;
1998 request_queue_t *q = req->q;
2000 spin_lock_irqsave(q->queue_lock, flags);
2001 __blk_put_request(q, req);
2002 spin_unlock_irqrestore(q->queue_lock, flags);
2006 EXPORT_SYMBOL(blk_put_request);
2009 * blk_congestion_wait - wait for a queue to become uncongested
2010 * @rw: READ or WRITE
2011 * @timeout: timeout in jiffies
2013 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2014 * If no queues are congested then just wait for the next request to be
2017 long blk_congestion_wait(int rw, long timeout)
2021 wait_queue_head_t *wqh = &congestion_wqh[rw];
2023 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2024 ret = io_schedule_timeout(timeout);
2025 finish_wait(wqh, &wait);
2029 EXPORT_SYMBOL(blk_congestion_wait);
2032 * Has to be called with the request spinlock acquired
2034 static int attempt_merge(request_queue_t *q, struct request *req,
2035 struct request *next)
2037 if (!rq_mergeable(req) || !rq_mergeable(next))
2043 if (req->sector + req->nr_sectors != next->sector)
2046 if (rq_data_dir(req) != rq_data_dir(next)
2047 || req->rq_disk != next->rq_disk
2048 || next->waiting || next->special)
2052 * If we are allowed to merge, then append bio list
2053 * from next to rq and release next. merge_requests_fn
2054 * will have updated segment counts, update sector
2057 if (!q->merge_requests_fn(q, req, next))
2061 * At this point we have either done a back merge
2062 * or front merge. We need the smaller start_time of
2063 * the merged requests to be the current request
2064 * for accounting purposes.
2066 if (time_after(req->start_time, next->start_time))
2067 req->start_time = next->start_time;
2069 req->biotail->bi_next = next->bio;
2070 req->biotail = next->biotail;
2072 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2074 elv_merge_requests(q, req, next);
2077 disk_round_stats(req->rq_disk);
2078 req->rq_disk->in_flight--;
2081 __blk_put_request(q, next);
2085 static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2087 struct request *next = elv_latter_request(q, rq);
2090 return attempt_merge(q, rq, next);
2095 static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2097 struct request *prev = elv_former_request(q, rq);
2100 return attempt_merge(q, prev, rq);
2106 * blk_attempt_remerge - attempt to remerge active head with next request
2107 * @q: The &request_queue_t belonging to the device
2108 * @rq: The head request (usually)
2111 * For head-active devices, the queue can easily be unplugged so quickly
2112 * that proper merging is not done on the front request. This may hurt
2113 * performance greatly for some devices. The block layer cannot safely
2114 * do merging on that first request for these queues, but the driver can
2115 * call this function and make it happen any way. Only the driver knows
2116 * when it is safe to do so.
2118 void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2120 unsigned long flags;
2122 spin_lock_irqsave(q->queue_lock, flags);
2123 attempt_back_merge(q, rq);
2124 spin_unlock_irqrestore(q->queue_lock, flags);
2127 EXPORT_SYMBOL(blk_attempt_remerge);
2130 * Non-locking blk_attempt_remerge variant.
2132 void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
2134 attempt_back_merge(q, rq);
2137 EXPORT_SYMBOL(__blk_attempt_remerge);
2139 static int __make_request(request_queue_t *q, struct bio *bio)
2141 struct request *req, *freereq = NULL;
2142 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, ra;
2145 sector = bio->bi_sector;
2146 nr_sectors = bio_sectors(bio);
2147 cur_nr_sectors = bio_cur_sectors(bio);
2149 rw = bio_data_dir(bio);
2152 * low level driver can indicate that it wants pages above a
2153 * certain limit bounced to low memory (ie for highmem, or even
2154 * ISA dma in theory)
2156 blk_queue_bounce(q, &bio);
2158 spin_lock_prefetch(q->queue_lock);
2160 barrier = test_bit(BIO_RW_BARRIER, &bio->bi_rw);
2162 ra = bio->bi_rw & (1 << BIO_RW_AHEAD);
2165 spin_lock_irq(q->queue_lock);
2167 if (elv_queue_empty(q)) {
2174 el_ret = elv_merge(q, &req, bio);
2176 case ELEVATOR_BACK_MERGE:
2177 BUG_ON(!rq_mergeable(req));
2179 if (!q->back_merge_fn(q, req, bio))
2182 req->biotail->bi_next = bio;
2184 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2185 drive_stat_acct(req, nr_sectors, 0);
2186 if (!attempt_back_merge(q, req))
2187 elv_merged_request(q, req);
2190 case ELEVATOR_FRONT_MERGE:
2191 BUG_ON(!rq_mergeable(req));
2193 if (!q->front_merge_fn(q, req, bio))
2196 bio->bi_next = req->bio;
2197 req->cbio = req->bio = bio;
2198 req->nr_cbio_segments = bio_segments(bio);
2199 req->nr_cbio_sectors = bio_sectors(bio);
2202 * may not be valid. if the low level driver said
2203 * it didn't need a bounce buffer then it better
2204 * not touch req->buffer either...
2206 req->buffer = bio_data(bio);
2207 req->current_nr_sectors = cur_nr_sectors;
2208 req->hard_cur_sectors = cur_nr_sectors;
2209 req->sector = req->hard_sector = sector;
2210 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2211 drive_stat_acct(req, nr_sectors, 0);
2212 if (!attempt_front_merge(q, req))
2213 elv_merged_request(q, req);
2217 * elevator says don't/can't merge. get new request
2219 case ELEVATOR_NO_MERGE:
2223 printk("elevator returned crap (%d)\n", el_ret);
2228 * Grab a free request from the freelist - if that is empty, check
2229 * if we are doing read ahead and abort instead of blocking for
2237 spin_unlock_irq(q->queue_lock);
2238 if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) {
2245 freereq = get_request_wait(q, rw);
2250 req->flags |= REQ_CMD;
2253 * inherit FAILFAST from bio and don't stack up
2254 * retries for read ahead
2256 if (ra || test_bit(BIO_RW_FAILFAST, &bio->bi_rw))
2257 req->flags |= REQ_FAILFAST;
2260 * REQ_BARRIER implies no merging, but lets make it explicit
2263 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2266 req->hard_sector = req->sector = sector;
2267 req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2268 req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2269 req->nr_phys_segments = bio_phys_segments(q, bio);
2270 req->nr_hw_segments = bio_hw_segments(q, bio);
2271 req->nr_cbio_segments = bio_segments(bio);
2272 req->nr_cbio_sectors = bio_sectors(bio);
2273 req->buffer = bio_data(bio); /* see ->buffer comment above */
2274 req->waiting = NULL;
2275 req->cbio = req->bio = req->biotail = bio;
2276 req->rq_disk = bio->bi_bdev->bd_disk;
2277 req->start_time = jiffies;
2279 add_request(q, req);
2282 __blk_put_request(q, freereq);
2284 if (blk_queue_plugged(q)) {
2285 int nrq = q->rq.count[READ] + q->rq.count[WRITE] - q->in_flight;
2287 if (nrq == q->unplug_thresh || bio_sync(bio))
2288 __generic_unplug_device(q);
2290 spin_unlock_irq(q->queue_lock);
2294 bio_endio(bio, nr_sectors << 9, -EWOULDBLOCK);
2299 * If bio->bi_dev is a partition, remap the location
2301 static inline void blk_partition_remap(struct bio *bio)
2303 struct block_device *bdev = bio->bi_bdev;
2305 if (bdev != bdev->bd_contains) {
2306 struct hd_struct *p = bdev->bd_part;
2308 switch (bio->bi_rw) {
2310 p->read_sectors += bio_sectors(bio);
2314 p->write_sectors += bio_sectors(bio);
2318 bio->bi_sector += p->start_sect;
2319 bio->bi_bdev = bdev->bd_contains;
2324 * generic_make_request: hand a buffer to its device driver for I/O
2325 * @bio: The bio describing the location in memory and on the device.
2327 * generic_make_request() is used to make I/O requests of block
2328 * devices. It is passed a &struct bio, which describes the I/O that needs
2331 * generic_make_request() does not return any status. The
2332 * success/failure status of the request, along with notification of
2333 * completion, is delivered asynchronously through the bio->bi_end_io
2334 * function described (one day) else where.
2336 * The caller of generic_make_request must make sure that bi_io_vec
2337 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2338 * set to describe the device address, and the
2339 * bi_end_io and optionally bi_private are set to describe how
2340 * completion notification should be signaled.
2342 * generic_make_request and the drivers it calls may use bi_next if this
2343 * bio happens to be merged with someone else, and may change bi_dev and
2344 * bi_sector for remaps as it sees fit. So the values of these fields
2345 * should NOT be depended on after the call to generic_make_request.
2347 void generic_make_request(struct bio *bio)
2351 int ret, nr_sectors = bio_sectors(bio);
2353 /* Test device or partition size, when known. */
2354 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2356 sector_t sector = bio->bi_sector;
2358 if (maxsector < nr_sectors ||
2359 maxsector - nr_sectors < sector) {
2360 char b[BDEVNAME_SIZE];
2361 /* This may well happen - the kernel calls
2362 * bread() without checking the size of the
2363 * device, e.g., when mounting a device. */
2365 "attempt to access beyond end of device\n");
2366 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2367 bdevname(bio->bi_bdev, b),
2369 (unsigned long long) sector + nr_sectors,
2370 (long long) maxsector);
2372 set_bit(BIO_EOF, &bio->bi_flags);
2378 * Resolve the mapping until finished. (drivers are
2379 * still free to implement/resolve their own stacking
2380 * by explicitly returning 0)
2382 * NOTE: we don't repeat the blk_size check for each new device.
2383 * Stacking drivers are expected to know what they are doing.
2386 char b[BDEVNAME_SIZE];
2388 q = bdev_get_queue(bio->bi_bdev);
2391 "generic_make_request: Trying to access "
2392 "nonexistent block-device %s (%Lu)\n",
2393 bdevname(bio->bi_bdev, b),
2394 (long long) bio->bi_sector);
2396 bio_endio(bio, bio->bi_size, -EIO);
2400 if (unlikely(bio_sectors(bio) > q->max_sectors)) {
2401 printk("bio too big device %s (%u > %u)\n",
2402 bdevname(bio->bi_bdev, b),
2408 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))
2412 * If this device has partitions, remap block n
2413 * of partition p to block n+start(p) of the disk.
2415 blk_partition_remap(bio);
2417 ret = q->make_request_fn(q, bio);
2421 EXPORT_SYMBOL(generic_make_request);
2424 * submit_bio: submit a bio to the block device layer for I/O
2425 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2426 * @bio: The &struct bio which describes the I/O
2428 * submit_bio() is very similar in purpose to generic_make_request(), and
2429 * uses that function to do most of the work. Both are fairly rough
2430 * interfaces, @bio must be presetup and ready for I/O.
2433 void submit_bio(int rw, struct bio *bio)
2435 int count = bio_sectors(bio);
2437 BIO_BUG_ON(!bio->bi_size);
2438 BIO_BUG_ON(!bio->bi_io_vec);
2441 mod_page_state(pgpgout, count);
2443 mod_page_state(pgpgin, count);
2445 if (unlikely(block_dump)) {
2446 char b[BDEVNAME_SIZE];
2447 printk("%s(%d): %s block %Lu on %s\n",
2448 current->comm, current->pid,
2449 (rw & WRITE) ? "WRITE" : "READ",
2450 (unsigned long long)bio->bi_sector,
2451 bdevname(bio->bi_bdev,b));
2454 generic_make_request(bio);
2457 EXPORT_SYMBOL(submit_bio);
2460 * blk_rq_next_segment
2461 * @rq: the request being processed
2464 * Points to the next segment in the request if the current segment
2465 * is complete. Leaves things unchanged if this segment is not over
2466 * or if no more segments are left in this request.
2468 * Meant to be used for bio traversal during I/O submission
2469 * Does not affect any I/O completions or update completion state
2470 * in the request, and does not modify any bio fields.
2472 * Decrementing rq->nr_sectors, rq->current_nr_sectors and
2473 * rq->nr_cbio_sectors as data is transferred is the caller's
2474 * responsibility and should be done before calling this routine.
2476 void blk_rq_next_segment(struct request *rq)
2478 if (rq->current_nr_sectors > 0)
2481 if (rq->nr_cbio_sectors > 0) {
2482 --rq->nr_cbio_segments;
2483 rq->current_nr_sectors = blk_rq_vec(rq)->bv_len >> 9;
2485 if ((rq->cbio = rq->cbio->bi_next)) {
2486 rq->nr_cbio_segments = bio_segments(rq->cbio);
2487 rq->nr_cbio_sectors = bio_sectors(rq->cbio);
2488 rq->current_nr_sectors = bio_cur_sectors(rq->cbio);
2492 /* remember the size of this segment before we start I/O */
2493 rq->hard_cur_sectors = rq->current_nr_sectors;
2497 * process_that_request_first - process partial request submission
2498 * @req: the request being processed
2499 * @nr_sectors: number of sectors I/O has been submitted on
2502 * May be used for processing bio's while submitting I/O without
2503 * signalling completion. Fails if more data is requested than is
2504 * available in the request in which case it doesn't advance any
2507 * Assumes a request is correctly set up. No sanity checks.
2510 * 0 - no more data left to submit (not processed)
2511 * 1 - data available to submit for this request (processed)
2513 int process_that_request_first(struct request *req, unsigned int nr_sectors)
2517 if (req->nr_sectors < nr_sectors)
2520 req->nr_sectors -= nr_sectors;
2521 req->sector += nr_sectors;
2522 while (nr_sectors) {
2523 nsect = min_t(unsigned, req->current_nr_sectors, nr_sectors);
2524 req->current_nr_sectors -= nsect;
2525 nr_sectors -= nsect;
2527 req->nr_cbio_sectors -= nsect;
2528 blk_rq_next_segment(req);
2534 EXPORT_SYMBOL(process_that_request_first);
2536 void blk_recalc_rq_segments(struct request *rq)
2539 int nr_phys_segs, nr_hw_segs;
2544 nr_phys_segs = nr_hw_segs = 0;
2545 rq_for_each_bio(bio, rq) {
2546 /* Force bio hw/phys segs to be recalculated. */
2547 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
2549 nr_phys_segs += bio_phys_segments(rq->q, bio);
2550 nr_hw_segs += bio_hw_segments(rq->q, bio);
2553 rq->nr_phys_segments = nr_phys_segs;
2554 rq->nr_hw_segments = nr_hw_segs;
2557 void blk_recalc_rq_sectors(struct request *rq, int nsect)
2559 if (blk_fs_request(rq)) {
2560 rq->hard_sector += nsect;
2561 rq->hard_nr_sectors -= nsect;
2564 * Move the I/O submission pointers ahead if required,
2565 * i.e. for drivers not aware of rq->cbio.
2567 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
2568 (rq->sector <= rq->hard_sector)) {
2569 rq->sector = rq->hard_sector;
2570 rq->nr_sectors = rq->hard_nr_sectors;
2571 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
2572 rq->current_nr_sectors = rq->hard_cur_sectors;
2573 rq->nr_cbio_segments = bio_segments(rq->bio);
2574 rq->nr_cbio_sectors = bio_sectors(rq->bio);
2575 rq->buffer = bio_data(rq->bio);
2581 * if total number of sectors is less than the first segment
2582 * size, something has gone terribly wrong
2584 if (rq->nr_sectors < rq->current_nr_sectors) {
2585 printk("blk: request botched\n");
2586 rq->nr_sectors = rq->current_nr_sectors;
2591 static int __end_that_request_first(struct request *req, int uptodate,
2594 int total_bytes, bio_nbytes, error = 0, next_idx = 0;
2598 * for a REQ_BLOCK_PC request, we want to carry any eventual
2599 * sense key with us all the way through
2601 if (!blk_pc_request(req))
2606 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
2607 printk("end_request: I/O error, dev %s, sector %llu\n",
2608 req->rq_disk ? req->rq_disk->disk_name : "?",
2609 (unsigned long long)req->sector);
2612 total_bytes = bio_nbytes = 0;
2613 while ((bio = req->bio)) {
2616 if (nr_bytes >= bio->bi_size) {
2617 req->bio = bio->bi_next;
2618 nbytes = bio->bi_size;
2619 bio_endio(bio, nbytes, error);
2623 int idx = bio->bi_idx + next_idx;
2625 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
2626 blk_dump_rq_flags(req, "__end_that");
2627 printk("%s: bio idx %d >= vcnt %d\n",
2629 bio->bi_idx, bio->bi_vcnt);
2633 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2634 BIO_BUG_ON(nbytes > bio->bi_size);
2637 * not a complete bvec done
2639 if (unlikely(nbytes > nr_bytes)) {
2640 bio_nbytes += nr_bytes;
2641 total_bytes += nr_bytes;
2646 * advance to the next vector
2649 bio_nbytes += nbytes;
2652 total_bytes += nbytes;
2655 if ((bio = req->bio)) {
2657 * end more in this run, or just return 'not-done'
2659 if (unlikely(nr_bytes <= 0))
2671 * if the request wasn't completed, update state
2674 bio_endio(bio, bio_nbytes, error);
2675 bio->bi_idx += next_idx;
2676 bio_iovec(bio)->bv_offset += nr_bytes;
2677 bio_iovec(bio)->bv_len -= nr_bytes;
2680 blk_recalc_rq_sectors(req, total_bytes >> 9);
2681 blk_recalc_rq_segments(req);
2686 * end_that_request_first - end I/O on a request
2687 * @req: the request being processed
2688 * @uptodate: 0 for I/O error
2689 * @nr_sectors: number of sectors to end I/O on
2692 * Ends I/O on a number of sectors attached to @req, and sets it up
2693 * for the next range of segments (if any) in the cluster.
2696 * 0 - we are done with this request, call end_that_request_last()
2697 * 1 - still buffers pending for this request
2699 int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
2701 return __end_that_request_first(req, uptodate, nr_sectors << 9);
2704 EXPORT_SYMBOL(end_that_request_first);
2707 * end_that_request_chunk - end I/O on a request
2708 * @req: the request being processed
2709 * @uptodate: 0 for I/O error
2710 * @nr_bytes: number of bytes to complete
2713 * Ends I/O on a number of bytes attached to @req, and sets it up
2714 * for the next range of segments (if any). Like end_that_request_first(),
2715 * but deals with bytes instead of sectors.
2718 * 0 - we are done with this request, call end_that_request_last()
2719 * 1 - still buffers pending for this request
2721 int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
2723 return __end_that_request_first(req, uptodate, nr_bytes);
2726 EXPORT_SYMBOL(end_that_request_chunk);
2729 * queue lock must be held
2731 void end_that_request_last(struct request *req)
2733 struct gendisk *disk = req->rq_disk;
2734 struct completion *waiting = req->waiting;
2736 if (unlikely(laptop_mode) && blk_fs_request(req))
2737 laptop_io_completion();
2739 if (disk && blk_fs_request(req)) {
2740 unsigned long duration = jiffies - req->start_time;
2741 switch (rq_data_dir(req)) {
2743 disk_stat_inc(disk, writes);
2744 disk_stat_add(disk, write_ticks, duration);
2747 disk_stat_inc(disk, reads);
2748 disk_stat_add(disk, read_ticks, duration);
2751 disk_round_stats(disk);
2754 __blk_put_request(req->q, req);
2755 /* Do this LAST! The structure may be freed immediately afterwards */
2760 EXPORT_SYMBOL(end_that_request_last);
2762 void end_request(struct request *req, int uptodate)
2764 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
2765 add_disk_randomness(req->rq_disk);
2766 blkdev_dequeue_request(req);
2767 end_that_request_last(req);
2771 EXPORT_SYMBOL(end_request);
2773 void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
2775 /* first three bits are identical in rq->flags and bio->bi_rw */
2776 rq->flags |= (bio->bi_rw & 7);
2778 rq->nr_phys_segments = bio_phys_segments(q, bio);
2779 rq->nr_hw_segments = bio_hw_segments(q, bio);
2780 rq->current_nr_sectors = bio_cur_sectors(bio);
2781 rq->hard_cur_sectors = rq->current_nr_sectors;
2782 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2783 rq->nr_cbio_segments = bio_segments(bio);
2784 rq->nr_cbio_sectors = bio_sectors(bio);
2785 rq->buffer = bio_data(bio);
2787 rq->cbio = rq->bio = rq->biotail = bio;
2790 EXPORT_SYMBOL(blk_rq_bio_prep);
2792 void blk_rq_prep_restart(struct request *rq)
2796 bio = rq->cbio = rq->bio;
2798 rq->nr_cbio_segments = bio_segments(bio);
2799 rq->nr_cbio_sectors = bio_sectors(bio);
2800 rq->hard_cur_sectors = bio_cur_sectors(bio);
2801 rq->buffer = bio_data(bio);
2803 rq->sector = rq->hard_sector;
2804 rq->nr_sectors = rq->hard_nr_sectors;
2805 rq->current_nr_sectors = rq->hard_cur_sectors;
2808 EXPORT_SYMBOL(blk_rq_prep_restart);
2810 int kblockd_schedule_work(struct work_struct *work)
2812 return queue_work(kblockd_workqueue, work);
2815 void kblockd_flush(void)
2817 flush_workqueue(kblockd_workqueue);
2820 int __init blk_dev_init(void)
2822 kblockd_workqueue = create_workqueue("kblockd");
2823 if (!kblockd_workqueue)
2824 panic("Failed to create kblockd\n");
2826 request_cachep = kmem_cache_create("blkdev_requests",
2827 sizeof(struct request), 0, 0, NULL, NULL);
2828 if (!request_cachep)
2829 panic("Can't create request pool slab cache\n");
2831 blk_max_low_pfn = max_low_pfn;
2832 blk_max_pfn = max_pfn;
2837 * IO Context helper functions
2839 void put_io_context(struct io_context *ioc)
2844 BUG_ON(atomic_read(&ioc->refcount) == 0);
2846 if (atomic_dec_and_test(&ioc->refcount)) {
2847 if (ioc->aic && ioc->aic->dtor)
2848 ioc->aic->dtor(ioc->aic);
2853 /* Called by the exitting task */
2854 void exit_io_context(void)
2856 unsigned long flags;
2857 struct io_context *ioc;
2859 local_irq_save(flags);
2860 ioc = current->io_context;
2862 if (ioc->aic && ioc->aic->exit)
2863 ioc->aic->exit(ioc->aic);
2864 put_io_context(ioc);
2865 current->io_context = NULL;
2868 local_irq_restore(flags);
2872 * If the current task has no IO context then create one and initialise it.
2873 * If it does have a context, take a ref on it.
2875 * This is always called in the context of the task which submitted the I/O.
2876 * But weird things happen, so we disable local interrupts to ensure exclusive
2877 * access to *current.
2879 struct io_context *get_io_context(int gfp_flags)
2881 struct task_struct *tsk = current;
2882 unsigned long flags;
2883 struct io_context *ret;
2885 local_irq_save(flags);
2886 ret = tsk->io_context;
2888 ret = kmalloc(sizeof(*ret), GFP_ATOMIC);
2890 atomic_set(&ret->refcount, 1);
2891 ret->pid = tsk->pid;
2892 ret->last_waited = jiffies; /* doesn't matter... */
2893 ret->nr_batch_requests = 0; /* because this is 0 */
2895 tsk->io_context = ret;
2899 atomic_inc(&ret->refcount);
2900 local_irq_restore(flags);
2904 void copy_io_context(struct io_context **pdst, struct io_context **psrc)
2906 struct io_context *src = *psrc;
2907 struct io_context *dst = *pdst;
2910 BUG_ON(atomic_read(&src->refcount) == 0);
2911 atomic_inc(&src->refcount);
2912 put_io_context(dst);
2917 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
2919 struct io_context *temp;
2929 struct queue_sysfs_entry {
2930 struct attribute attr;
2931 ssize_t (*show)(struct request_queue *, char *);
2932 ssize_t (*store)(struct request_queue *, const char *, size_t);
2936 queue_var_show(unsigned int var, char *page)
2938 return sprintf(page, "%d\n", var);
2942 queue_var_store(unsigned long *var, const char *page, size_t count)
2944 char *p = (char *) page;
2946 *var = simple_strtoul(p, &p, 10);
2950 static ssize_t queue_requests_show(struct request_queue *q, char *page)
2952 return queue_var_show(q->nr_requests, (page));
2956 queue_requests_store(struct request_queue *q, const char *page, size_t count)
2958 struct request_list *rl = &q->rq;
2960 int ret = queue_var_store(&q->nr_requests, page, count);
2961 if (q->nr_requests < BLKDEV_MIN_RQ)
2962 q->nr_requests = BLKDEV_MIN_RQ;
2964 if (rl->count[READ] >= queue_congestion_on_threshold(q))
2965 set_queue_congested(q, READ);
2966 else if (rl->count[READ] < queue_congestion_off_threshold(q))
2967 clear_queue_congested(q, READ);
2969 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
2970 set_queue_congested(q, WRITE);
2971 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
2972 clear_queue_congested(q, WRITE);
2974 if (rl->count[READ] >= q->nr_requests) {
2975 blk_set_queue_full(q, READ);
2976 } else if (rl->count[READ]+1 <= q->nr_requests) {
2977 blk_clear_queue_full(q, READ);
2978 wake_up(&rl->wait[READ]);
2981 if (rl->count[WRITE] >= q->nr_requests) {
2982 blk_set_queue_full(q, WRITE);
2983 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
2984 blk_clear_queue_full(q, WRITE);
2985 wake_up(&rl->wait[WRITE]);
2990 static struct queue_sysfs_entry queue_requests_entry = {
2991 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
2992 .show = queue_requests_show,
2993 .store = queue_requests_store,
2996 static struct attribute *default_attrs[] = {
2997 &queue_requests_entry.attr,
3001 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3004 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3006 struct queue_sysfs_entry *entry = to_queue(attr);
3007 struct request_queue *q;
3009 q = container_of(kobj, struct request_queue, kobj);
3013 return entry->show(q, page);
3017 queue_attr_store(struct kobject *kobj, struct attribute *attr,
3018 const char *page, size_t length)
3020 struct queue_sysfs_entry *entry = to_queue(attr);
3021 struct request_queue *q;
3023 q = container_of(kobj, struct request_queue, kobj);
3027 return entry->store(q, page, length);
3030 static struct sysfs_ops queue_sysfs_ops = {
3031 .show = queue_attr_show,
3032 .store = queue_attr_store,
3035 struct kobj_type queue_ktype = {
3036 .sysfs_ops = &queue_sysfs_ops,
3037 .default_attrs = default_attrs,
3040 int blk_register_queue(struct gendisk *disk)
3044 request_queue_t *q = disk->queue;
3046 if (!q || !q->request_fn)
3049 q->kobj.parent = kobject_get(&disk->kobj);
3050 if (!q->kobj.parent)
3053 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3054 q->kobj.ktype = &queue_ktype;
3056 ret = kobject_register(&q->kobj);
3060 ret = elv_register_queue(q);
3062 kobject_unregister(&q->kobj);
3069 void blk_unregister_queue(struct gendisk *disk)
3071 request_queue_t *q = disk->queue;
3073 if (q && q->request_fn) {
3074 elv_unregister_queue(q);
3076 kobject_unregister(&q->kobj);
3077 kobject_put(&disk->kobj);