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;
46 * For queue allocation
48 static kmem_cache_t *requestq_cachep;
51 * For io context allocations
53 static kmem_cache_t *iocontext_cachep;
55 static wait_queue_head_t congestion_wqh[2] = {
56 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
57 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
61 * Controlling structure to kblockd
63 static struct workqueue_struct *kblockd_workqueue;
65 unsigned long blk_max_low_pfn, blk_max_pfn;
67 EXPORT_SYMBOL(blk_max_low_pfn);
68 EXPORT_SYMBOL(blk_max_pfn);
70 /* Amount of time in which a process may batch requests */
71 #define BLK_BATCH_TIME (HZ/50UL)
73 /* Number of requests a "batching" process may submit */
74 #define BLK_BATCH_REQ 32
77 * Return the threshold (number of used requests) at which the queue is
78 * considered to be congested. It include a little hysteresis to keep the
79 * context switch rate down.
81 static inline int queue_congestion_on_threshold(struct request_queue *q)
83 return q->nr_congestion_on;
87 * The threshold at which a queue is considered to be uncongested
89 static inline int queue_congestion_off_threshold(struct request_queue *q)
91 return q->nr_congestion_off;
94 static void blk_queue_congestion_threshold(struct request_queue *q)
98 nr = q->nr_requests - (q->nr_requests / 8) + 1;
99 if (nr > q->nr_requests)
101 q->nr_congestion_on = nr;
103 nr = q->nr_requests - (q->nr_requests / 8) - 1;
106 q->nr_congestion_off = nr;
110 * A queue has just exitted congestion. Note this in the global counter of
111 * congested queues, and wake up anyone who was waiting for requests to be
114 static void clear_queue_congested(request_queue_t *q, int rw)
117 wait_queue_head_t *wqh = &congestion_wqh[rw];
119 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
120 clear_bit(bit, &q->backing_dev_info.state);
121 smp_mb__after_clear_bit();
122 if (waitqueue_active(wqh))
127 * A queue has just entered congestion. Flag that in the queue's VM-visible
128 * state flags and increment the global gounter of congested queues.
130 static void set_queue_congested(request_queue_t *q, int rw)
134 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
135 set_bit(bit, &q->backing_dev_info.state);
139 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
142 * Locates the passed device's request queue and returns the address of its
145 * Will return NULL if the request queue cannot be located.
147 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
149 struct backing_dev_info *ret = NULL;
150 request_queue_t *q = bdev_get_queue(bdev);
153 ret = &q->backing_dev_info;
157 void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
160 q->activity_data = data;
163 EXPORT_SYMBOL(blk_queue_activity_fn);
166 * blk_queue_prep_rq - set a prepare_request function for queue
168 * @pfn: prepare_request function
170 * It's possible for a queue to register a prepare_request callback which
171 * is invoked before the request is handed to the request_fn. The goal of
172 * the function is to prepare a request for I/O, it can be used to build a
173 * cdb from the request data for instance.
176 void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
181 EXPORT_SYMBOL(blk_queue_prep_rq);
184 * blk_queue_merge_bvec - set a merge_bvec function for queue
186 * @mbfn: merge_bvec_fn
188 * Usually queues have static limitations on the max sectors or segments that
189 * we can put in a request. Stacking drivers may have some settings that
190 * are dynamic, and thus we have to query the queue whether it is ok to
191 * add a new bio_vec to a bio at a given offset or not. If the block device
192 * has such limitations, it needs to register a merge_bvec_fn to control
193 * the size of bio's sent to it. Note that a block device *must* allow a
194 * single page to be added to an empty bio. The block device driver may want
195 * to use the bio_split() function to deal with these bio's. By default
196 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
199 void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
201 q->merge_bvec_fn = mbfn;
204 EXPORT_SYMBOL(blk_queue_merge_bvec);
207 * blk_queue_make_request - define an alternate make_request function for a device
208 * @q: the request queue for the device to be affected
209 * @mfn: the alternate make_request function
212 * The normal way for &struct bios to be passed to a device
213 * driver is for them to be collected into requests on a request
214 * queue, and then to allow the device driver to select requests
215 * off that queue when it is ready. This works well for many block
216 * devices. However some block devices (typically virtual devices
217 * such as md or lvm) do not benefit from the processing on the
218 * request queue, and are served best by having the requests passed
219 * directly to them. This can be achieved by providing a function
220 * to blk_queue_make_request().
223 * The driver that does this *must* be able to deal appropriately
224 * with buffers in "highmemory". This can be accomplished by either calling
225 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
226 * blk_queue_bounce() to create a buffer in normal memory.
228 void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
233 q->nr_requests = BLKDEV_MAX_RQ;
234 q->max_phys_segments = MAX_PHYS_SEGMENTS;
235 q->max_hw_segments = MAX_HW_SEGMENTS;
236 q->make_request_fn = mfn;
237 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
238 q->backing_dev_info.state = 0;
239 q->backing_dev_info.memory_backed = 0;
240 blk_queue_max_sectors(q, MAX_SECTORS);
241 blk_queue_hardsect_size(q, 512);
242 blk_queue_dma_alignment(q, 511);
243 blk_queue_congestion_threshold(q);
245 q->unplug_thresh = 4; /* hmm */
246 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
247 if (q->unplug_delay == 0)
250 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
252 q->unplug_timer.function = blk_unplug_timeout;
253 q->unplug_timer.data = (unsigned long)q;
256 * by default assume old behaviour and bounce for any highmem page
258 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
260 blk_queue_activity_fn(q, NULL, NULL);
263 EXPORT_SYMBOL(blk_queue_make_request);
266 * blk_queue_bounce_limit - set bounce buffer limit for queue
267 * @q: the request queue for the device
268 * @dma_addr: bus address limit
271 * Different hardware can have different requirements as to what pages
272 * it can do I/O directly to. A low level driver can call
273 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
274 * buffers for doing I/O to pages residing above @page. By default
275 * the block layer sets this to the highest numbered "low" memory page.
277 void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
279 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
282 * set appropriate bounce gfp mask -- unfortunately we don't have a
283 * full 4GB zone, so we have to resort to low memory for any bounces.
284 * ISA has its own < 16MB zone.
286 if (bounce_pfn < blk_max_low_pfn) {
287 BUG_ON(dma_addr < BLK_BOUNCE_ISA);
288 init_emergency_isa_pool();
289 q->bounce_gfp = GFP_NOIO | GFP_DMA;
291 q->bounce_gfp = GFP_NOIO;
293 q->bounce_pfn = bounce_pfn;
296 EXPORT_SYMBOL(blk_queue_bounce_limit);
299 * blk_queue_max_sectors - set max sectors for a request for this queue
300 * @q: the request queue for the device
301 * @max_sectors: max sectors in the usual 512b unit
304 * Enables a low level driver to set an upper limit on the size of
307 void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors)
309 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
310 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
311 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
314 q->max_sectors = max_sectors;
317 EXPORT_SYMBOL(blk_queue_max_sectors);
320 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
321 * @q: the request queue for the device
322 * @max_segments: max number of segments
325 * Enables a low level driver to set an upper limit on the number of
326 * physical data segments in a request. This would be the largest sized
327 * scatter list the driver could handle.
329 void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
333 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
336 q->max_phys_segments = max_segments;
339 EXPORT_SYMBOL(blk_queue_max_phys_segments);
342 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
343 * @q: the request queue for the device
344 * @max_segments: max number of segments
347 * Enables a low level driver to set an upper limit on the number of
348 * hw data segments in a request. This would be the largest number of
349 * address/length pairs the host adapter can actually give as once
352 void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
356 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
359 q->max_hw_segments = max_segments;
362 EXPORT_SYMBOL(blk_queue_max_hw_segments);
365 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
366 * @q: the request queue for the device
367 * @max_size: max size of segment in bytes
370 * Enables a low level driver to set an upper limit on the size of a
373 void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
375 if (max_size < PAGE_CACHE_SIZE) {
376 max_size = PAGE_CACHE_SIZE;
377 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
380 q->max_segment_size = max_size;
383 EXPORT_SYMBOL(blk_queue_max_segment_size);
386 * blk_queue_hardsect_size - set hardware sector size for the queue
387 * @q: the request queue for the device
388 * @size: the hardware sector size, in bytes
391 * This should typically be set to the lowest possible sector size
392 * that the hardware can operate on (possible without reverting to
393 * even internal read-modify-write operations). Usually the default
394 * of 512 covers most hardware.
396 void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
398 q->hardsect_size = size;
401 EXPORT_SYMBOL(blk_queue_hardsect_size);
404 * Returns the minimum that is _not_ zero, unless both are zero.
406 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
409 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
410 * @t: the stacking driver (top)
411 * @b: the underlying device (bottom)
413 void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
415 /* zero is "infinity" */
416 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
418 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
419 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
420 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
421 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
424 EXPORT_SYMBOL(blk_queue_stack_limits);
427 * blk_queue_segment_boundary - set boundary rules for segment merging
428 * @q: the request queue for the device
429 * @mask: the memory boundary mask
431 void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
433 if (mask < PAGE_CACHE_SIZE - 1) {
434 mask = PAGE_CACHE_SIZE - 1;
435 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
438 q->seg_boundary_mask = mask;
441 EXPORT_SYMBOL(blk_queue_segment_boundary);
444 * blk_queue_dma_alignment - set dma length and memory alignment
445 * @q: the request queue for the device
446 * @mask: alignment mask
449 * set required memory and length aligment for direct dma transactions.
450 * this is used when buiding direct io requests for the queue.
453 void blk_queue_dma_alignment(request_queue_t *q, int mask)
455 q->dma_alignment = mask;
458 EXPORT_SYMBOL(blk_queue_dma_alignment);
461 * blk_queue_find_tag - find a request by its tag and queue
463 * @q: The request queue for the device
464 * @tag: The tag of the request
467 * Should be used when a device returns a tag and you want to match
470 * no locks need be held.
472 struct request *blk_queue_find_tag(request_queue_t *q, int tag)
474 struct blk_queue_tag *bqt = q->queue_tags;
476 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
479 return bqt->tag_index[tag];
482 EXPORT_SYMBOL(blk_queue_find_tag);
485 * blk_queue_free_tags - release tag maintenance info
486 * @q: the request queue for the device
489 * blk_cleanup_queue() will take care of calling this function, if tagging
490 * has been used. So there's usually no need to call this directly, unless
491 * tagging is just being disabled but the queue remains in function.
493 void blk_queue_free_tags(request_queue_t *q)
495 struct blk_queue_tag *bqt = q->queue_tags;
500 if (atomic_dec_and_test(&bqt->refcnt)) {
502 BUG_ON(!list_empty(&bqt->busy_list));
504 kfree(bqt->tag_index);
505 bqt->tag_index = NULL;
513 q->queue_tags = NULL;
514 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
517 EXPORT_SYMBOL(blk_queue_free_tags);
520 init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
524 if (depth > q->nr_requests * 2) {
525 depth = q->nr_requests * 2;
526 printk(KERN_ERR "%s: adjusted depth to %d\n",
527 __FUNCTION__, depth);
530 tags->tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
531 if (!tags->tag_index)
534 bits = (depth / BLK_TAGS_PER_LONG) + 1;
535 tags->tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC);
539 memset(tags->tag_index, 0, depth * sizeof(struct request *));
540 memset(tags->tag_map, 0, bits * sizeof(unsigned long));
541 tags->max_depth = depth;
542 tags->real_max_depth = bits * BITS_PER_LONG;
545 * set the upper bits if the depth isn't a multiple of the word size
547 for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++)
548 __set_bit(i, tags->tag_map);
550 INIT_LIST_HEAD(&tags->busy_list);
552 atomic_set(&tags->refcnt, 1);
555 kfree(tags->tag_index);
560 * blk_queue_init_tags - initialize the queue tag info
561 * @q: the request queue for the device
562 * @depth: the maximum queue depth supported
564 int blk_queue_init_tags(request_queue_t *q, int depth,
565 struct blk_queue_tag *tags)
568 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
572 if (init_tag_map(q, tags, depth))
575 atomic_inc(&tags->refcnt);
578 * assign it, all done
580 q->queue_tags = tags;
581 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
588 EXPORT_SYMBOL(blk_queue_init_tags);
591 * blk_queue_resize_tags - change the queueing depth
592 * @q: the request queue for the device
593 * @new_depth: the new max command queueing depth
596 * Must be called with the queue lock held.
598 int blk_queue_resize_tags(request_queue_t *q, int new_depth)
600 struct blk_queue_tag *bqt = q->queue_tags;
601 struct request **tag_index;
602 unsigned long *tag_map;
609 * don't bother sizing down
611 if (new_depth <= bqt->real_max_depth) {
612 bqt->max_depth = new_depth;
617 * save the old state info, so we can copy it back
619 tag_index = bqt->tag_index;
620 tag_map = bqt->tag_map;
621 max_depth = bqt->real_max_depth;
623 if (init_tag_map(q, bqt, new_depth))
626 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
627 bits = max_depth / BLK_TAGS_PER_LONG;
628 memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long));
635 EXPORT_SYMBOL(blk_queue_resize_tags);
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, hw_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 hw_seg_size = 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))
846 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
849 seg_size += bv->bv_len;
850 hw_seg_size += bv->bv_len;
855 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
856 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
857 hw_seg_size += bv->bv_len;
860 if (hw_seg_size > bio->bi_hw_front_size)
861 bio->bi_hw_front_size = hw_seg_size;
862 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
868 seg_size = bv->bv_len;
871 if (hw_seg_size > bio->bi_hw_back_size)
872 bio->bi_hw_back_size = hw_seg_size;
873 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
874 bio->bi_hw_front_size = hw_seg_size;
875 bio->bi_phys_segments = nr_phys_segs;
876 bio->bi_hw_segments = nr_hw_segs;
877 bio->bi_flags |= (1 << BIO_SEG_VALID);
881 int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
884 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
887 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
889 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
893 * bio and nxt are contigous in memory, check if the queue allows
894 * these two to be merged into one
896 if (BIO_SEG_BOUNDARY(q, bio, nxt))
902 EXPORT_SYMBOL(blk_phys_contig_segment);
904 int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
907 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
908 blk_recount_segments(q, bio);
909 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
910 blk_recount_segments(q, nxt);
911 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
912 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
914 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
920 EXPORT_SYMBOL(blk_hw_contig_segment);
923 * map a request to scatterlist, return number of sg entries setup. Caller
924 * must make sure sg can hold rq->nr_phys_segments entries
926 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
928 struct bio_vec *bvec, *bvprv;
930 int nsegs, i, cluster;
933 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
939 rq_for_each_bio(bio, rq) {
941 * for each segment in bio
943 bio_for_each_segment(bvec, bio, i) {
944 int nbytes = bvec->bv_len;
946 if (bvprv && cluster) {
947 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
950 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
952 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
955 sg[nsegs - 1].length += nbytes;
958 memset(&sg[nsegs],0,sizeof(struct scatterlist));
959 sg[nsegs].page = bvec->bv_page;
960 sg[nsegs].length = nbytes;
961 sg[nsegs].offset = bvec->bv_offset;
966 } /* segments in bio */
972 EXPORT_SYMBOL(blk_rq_map_sg);
975 * the standard queue merge functions, can be overridden with device
976 * specific ones if so desired
979 static inline int ll_new_mergeable(request_queue_t *q,
983 int nr_phys_segs = bio_phys_segments(q, bio);
985 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
986 req->flags |= REQ_NOMERGE;
987 if (req == q->last_merge)
988 q->last_merge = NULL;
993 * A hw segment is just getting larger, bump just the phys
996 req->nr_phys_segments += nr_phys_segs;
1000 static inline int ll_new_hw_segment(request_queue_t *q,
1001 struct request *req,
1004 int nr_hw_segs = bio_hw_segments(q, bio);
1005 int nr_phys_segs = bio_phys_segments(q, bio);
1007 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1008 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1009 req->flags |= REQ_NOMERGE;
1010 if (req == q->last_merge)
1011 q->last_merge = NULL;
1016 * This will form the start of a new hw segment. Bump both
1019 req->nr_hw_segments += nr_hw_segs;
1020 req->nr_phys_segments += nr_phys_segs;
1024 static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1029 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1030 req->flags |= REQ_NOMERGE;
1031 if (req == q->last_merge)
1032 q->last_merge = NULL;
1035 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1036 blk_recount_segments(q, req->biotail);
1037 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1038 blk_recount_segments(q, bio);
1039 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1040 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1041 !BIOVEC_VIRT_OVERSIZE(len)) {
1042 int mergeable = ll_new_mergeable(q, req, bio);
1045 if (req->nr_hw_segments == 1)
1046 req->bio->bi_hw_front_size = len;
1047 if (bio->bi_hw_segments == 1)
1048 bio->bi_hw_back_size = len;
1053 return ll_new_hw_segment(q, req, bio);
1056 static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1061 if (req->nr_sectors + bio_sectors(bio) > q->max_sectors) {
1062 req->flags |= REQ_NOMERGE;
1063 if (req == q->last_merge)
1064 q->last_merge = NULL;
1067 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1068 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1069 blk_recount_segments(q, bio);
1070 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1071 blk_recount_segments(q, req->bio);
1072 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1073 !BIOVEC_VIRT_OVERSIZE(len)) {
1074 int mergeable = ll_new_mergeable(q, req, bio);
1077 if (bio->bi_hw_segments == 1)
1078 bio->bi_hw_front_size = len;
1079 if (req->nr_hw_segments == 1)
1080 req->biotail->bi_hw_back_size = len;
1085 return ll_new_hw_segment(q, req, bio);
1088 static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1089 struct request *next)
1091 int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments;
1092 int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1095 * First check if the either of the requests are re-queued
1096 * requests. Can't merge them if they are.
1098 if (req->special || next->special)
1102 * Will it become to large?
1104 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1107 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1108 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1109 total_phys_segments--;
1111 if (total_phys_segments > q->max_phys_segments)
1114 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1115 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1116 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1118 * propagate the combined length to the end of the requests
1120 if (req->nr_hw_segments == 1)
1121 req->bio->bi_hw_front_size = len;
1122 if (next->nr_hw_segments == 1)
1123 next->biotail->bi_hw_back_size = len;
1124 total_hw_segments--;
1127 if (total_hw_segments > q->max_hw_segments)
1130 /* Merge is OK... */
1131 req->nr_phys_segments = total_phys_segments;
1132 req->nr_hw_segments = total_hw_segments;
1137 * "plug" the device if there are no outstanding requests: this will
1138 * force the transfer to start only after we have put all the requests
1141 * This is called with interrupts off and no requests on the queue and
1142 * with the queue lock held.
1144 void blk_plug_device(request_queue_t *q)
1146 WARN_ON(!irqs_disabled());
1149 * don't plug a stopped queue, it must be paired with blk_start_queue()
1150 * which will restart the queueing
1152 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1155 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1156 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1159 EXPORT_SYMBOL(blk_plug_device);
1162 * remove the queue from the plugged list, if present. called with
1163 * queue lock held and interrupts disabled.
1165 int blk_remove_plug(request_queue_t *q)
1167 WARN_ON(!irqs_disabled());
1169 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1172 del_timer(&q->unplug_timer);
1176 EXPORT_SYMBOL(blk_remove_plug);
1179 * remove the plug and let it rip..
1181 void __generic_unplug_device(request_queue_t *q)
1183 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1186 if (!blk_remove_plug(q))
1190 * was plugged, fire request_fn if queue has stuff to do
1192 if (elv_next_request(q))
1195 EXPORT_SYMBOL(__generic_unplug_device);
1198 * generic_unplug_device - fire a request queue
1199 * @q: The &request_queue_t in question
1202 * Linux uses plugging to build bigger requests queues before letting
1203 * the device have at them. If a queue is plugged, the I/O scheduler
1204 * is still adding and merging requests on the queue. Once the queue
1205 * gets unplugged, the request_fn defined for the queue is invoked and
1206 * transfers started.
1208 void generic_unplug_device(request_queue_t *q)
1210 spin_lock_irq(q->queue_lock);
1211 __generic_unplug_device(q);
1212 spin_unlock_irq(q->queue_lock);
1214 EXPORT_SYMBOL(generic_unplug_device);
1216 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1219 request_queue_t *q = bdi->unplug_io_data;
1222 * devices don't necessarily have an ->unplug_fn defined
1228 static void blk_unplug_work(void *data)
1230 request_queue_t *q = data;
1235 static void blk_unplug_timeout(unsigned long data)
1237 request_queue_t *q = (request_queue_t *)data;
1239 kblockd_schedule_work(&q->unplug_work);
1243 * blk_start_queue - restart a previously stopped queue
1244 * @q: The &request_queue_t in question
1247 * blk_start_queue() will clear the stop flag on the queue, and call
1248 * the request_fn for the queue if it was in a stopped state when
1249 * entered. Also see blk_stop_queue(). Queue lock must be held.
1251 void blk_start_queue(request_queue_t *q)
1253 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1256 * one level of recursion is ok and is much faster than kicking
1257 * the unplug handling
1259 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1261 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1264 kblockd_schedule_work(&q->unplug_work);
1268 EXPORT_SYMBOL(blk_start_queue);
1271 * blk_stop_queue - stop a queue
1272 * @q: The &request_queue_t in question
1275 * The Linux block layer assumes that a block driver will consume all
1276 * entries on the request queue when the request_fn strategy is called.
1277 * Often this will not happen, because of hardware limitations (queue
1278 * depth settings). If a device driver gets a 'queue full' response,
1279 * or if it simply chooses not to queue more I/O at one point, it can
1280 * call this function to prevent the request_fn from being called until
1281 * the driver has signalled it's ready to go again. This happens by calling
1282 * blk_start_queue() to restart queue operations. Queue lock must be held.
1284 void blk_stop_queue(request_queue_t *q)
1287 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1290 EXPORT_SYMBOL(blk_stop_queue);
1293 * blk_run_queue - run a single device queue
1294 * @q: The queue to run
1296 void blk_run_queue(struct request_queue *q)
1298 unsigned long flags;
1300 spin_lock_irqsave(q->queue_lock, flags);
1303 spin_unlock_irqrestore(q->queue_lock, flags);
1306 EXPORT_SYMBOL(blk_run_queue);
1309 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1310 * @q: the request queue to be released
1313 * blk_cleanup_queue is the pair to blk_init_queue() or
1314 * blk_queue_make_request(). It should be called when a request queue is
1315 * being released; typically when a block device is being de-registered.
1316 * Currently, its primary task it to free all the &struct request
1317 * structures that were allocated to the queue and the queue itself.
1320 * Hopefully the low level driver will have finished any
1321 * outstanding requests first...
1323 void blk_cleanup_queue(request_queue_t * q)
1325 struct request_list *rl = &q->rq;
1327 if (!atomic_dec_and_test(&q->refcnt))
1332 del_timer_sync(&q->unplug_timer);
1336 mempool_destroy(rl->rq_pool);
1338 if (blk_queue_tagged(q))
1339 blk_queue_free_tags(q);
1341 kmem_cache_free(requestq_cachep, q);
1344 EXPORT_SYMBOL(blk_cleanup_queue);
1346 static int blk_init_free_list(request_queue_t *q)
1348 struct request_list *rl = &q->rq;
1350 rl->count[READ] = rl->count[WRITE] = 0;
1351 init_waitqueue_head(&rl->wait[READ]);
1352 init_waitqueue_head(&rl->wait[WRITE]);
1354 rl->rq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep);
1362 static int __make_request(request_queue_t *, struct bio *);
1364 static elevator_t *chosen_elevator =
1365 #if defined(CONFIG_IOSCHED_CFQ)
1367 #elif defined(CONFIG_IOSCHED_AS)
1369 #elif defined(CONFIG_IOSCHED_DEADLINE)
1371 #elif defined(CONFIG_IOSCHED_NOOP)
1375 #error "You must have at least 1 I/O scheduler selected"
1378 #if defined(CONFIG_IOSCHED_AS) || defined(CONFIG_IOSCHED_DEADLINE) || defined (CONFIG_IOSCHED_NOOP)
1379 static int __init elevator_setup(char *str)
1381 #ifdef CONFIG_IOSCHED_DEADLINE
1382 if (!strcmp(str, "deadline"))
1383 chosen_elevator = &iosched_deadline;
1385 #ifdef CONFIG_IOSCHED_AS
1386 if (!strcmp(str, "as"))
1387 chosen_elevator = &iosched_as;
1389 #ifdef CONFIG_IOSCHED_CFQ
1390 if (!strcmp(str, "cfq"))
1391 chosen_elevator = &iosched_cfq;
1393 #ifdef CONFIG_IOSCHED_NOOP
1394 if (!strcmp(str, "noop"))
1395 chosen_elevator = &elevator_noop;
1400 __setup("elevator=", elevator_setup);
1401 #endif /* CONFIG_IOSCHED_AS || CONFIG_IOSCHED_DEADLINE || CONFIG_IOSCHED_NOOP */
1403 request_queue_t *blk_alloc_queue(int gfp_mask)
1405 request_queue_t *q = kmem_cache_alloc(requestq_cachep, gfp_mask);
1410 memset(q, 0, sizeof(*q));
1411 init_timer(&q->unplug_timer);
1412 atomic_set(&q->refcnt, 1);
1414 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1415 q->backing_dev_info.unplug_io_data = q;
1420 EXPORT_SYMBOL(blk_alloc_queue);
1423 * blk_init_queue - prepare a request queue for use with a block device
1424 * @rfn: The function to be called to process requests that have been
1425 * placed on the queue.
1426 * @lock: Request queue spin lock
1429 * If a block device wishes to use the standard request handling procedures,
1430 * which sorts requests and coalesces adjacent requests, then it must
1431 * call blk_init_queue(). The function @rfn will be called when there
1432 * are requests on the queue that need to be processed. If the device
1433 * supports plugging, then @rfn may not be called immediately when requests
1434 * are available on the queue, but may be called at some time later instead.
1435 * Plugged queues are generally unplugged when a buffer belonging to one
1436 * of the requests on the queue is needed, or due to memory pressure.
1438 * @rfn is not required, or even expected, to remove all requests off the
1439 * queue, but only as many as it can handle at a time. If it does leave
1440 * requests on the queue, it is responsible for arranging that the requests
1441 * get dealt with eventually.
1443 * The queue spin lock must be held while manipulating the requests on the
1446 * Function returns a pointer to the initialized request queue, or NULL if
1447 * it didn't succeed.
1450 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1451 * when the block device is deactivated (such as at module unload).
1453 request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1458 q = blk_alloc_queue(GFP_KERNEL);
1462 if (blk_init_free_list(q))
1467 printk("Using %s io scheduler\n", chosen_elevator->elevator_name);
1470 q->request_fn = rfn;
1471 q->back_merge_fn = ll_back_merge_fn;
1472 q->front_merge_fn = ll_front_merge_fn;
1473 q->merge_requests_fn = ll_merge_requests_fn;
1474 q->prep_rq_fn = NULL;
1475 q->unplug_fn = generic_unplug_device;
1476 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1477 q->queue_lock = lock;
1479 blk_queue_segment_boundary(q, 0xffffffff);
1481 blk_queue_make_request(q, __make_request);
1482 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1484 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1485 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1490 if (!elevator_init(q, chosen_elevator))
1493 blk_cleanup_queue(q);
1495 kmem_cache_free(requestq_cachep, q);
1499 EXPORT_SYMBOL(blk_init_queue);
1501 int blk_get_queue(request_queue_t *q)
1503 if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
1504 atomic_inc(&q->refcnt);
1511 EXPORT_SYMBOL(blk_get_queue);
1513 static inline void blk_free_request(request_queue_t *q, struct request *rq)
1515 elv_put_request(q, rq);
1516 mempool_free(rq, q->rq.rq_pool);
1519 static inline struct request *blk_alloc_request(request_queue_t *q,int gfp_mask)
1521 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1526 if (!elv_set_request(q, rq, gfp_mask))
1529 mempool_free(rq, q->rq.rq_pool);
1534 * ioc_batching returns true if the ioc is a valid batching request and
1535 * should be given priority access to a request.
1537 static inline int ioc_batching(struct io_context *ioc)
1543 * Make sure the process is able to allocate at least 1 request
1544 * even if the batch times out, otherwise we could theoretically
1547 return ioc->nr_batch_requests == BLK_BATCH_REQ ||
1548 (ioc->nr_batch_requests > 0
1549 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1553 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1554 * will cause the process to be a "batcher" on all queues in the system. This
1555 * is the behaviour we want though - once it gets a wakeup it should be given
1558 void ioc_set_batching(struct io_context *ioc)
1560 if (!ioc || ioc_batching(ioc))
1563 ioc->nr_batch_requests = BLK_BATCH_REQ;
1564 ioc->last_waited = jiffies;
1568 * A request has just been released. Account for it, update the full and
1569 * congestion status, wake up any waiters. Called under q->queue_lock.
1571 static void freed_request(request_queue_t *q, int rw)
1573 struct request_list *rl = &q->rq;
1576 if (rl->count[rw] < queue_congestion_off_threshold(q))
1577 clear_queue_congested(q, rw);
1578 if (rl->count[rw]+1 <= q->nr_requests) {
1579 if (waitqueue_active(&rl->wait[rw]))
1580 wake_up(&rl->wait[rw]);
1581 if (!waitqueue_active(&rl->wait[rw]))
1582 blk_clear_queue_full(q, rw);
1586 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1588 * Get a free request, queue_lock must not be held
1590 static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1592 struct request *rq = NULL;
1593 struct request_list *rl = &q->rq;
1594 struct io_context *ioc = get_io_context(gfp_mask);
1596 spin_lock_irq(q->queue_lock);
1598 if (!elv_may_queue(q, rw))
1601 if (rl->count[rw]+1 >= q->nr_requests) {
1603 * The queue will fill after this allocation, so set it as
1604 * full, and mark this process as "batching". This process
1605 * will be allowed to complete a batch of requests, others
1608 if (!blk_queue_full(q, rw)) {
1609 ioc_set_batching(ioc);
1610 blk_set_queue_full(q, rw);
1615 * The queue is full and the allocating process is not a
1616 * "batcher", and not exempted by the IO scheduler
1618 if (blk_queue_full(q, rw) && !ioc_batching(ioc))
1622 if (rl->count[rw] >= queue_congestion_on_threshold(q))
1623 set_queue_congested(q, rw);
1624 spin_unlock_irq(q->queue_lock);
1626 rq = blk_alloc_request(q, gfp_mask);
1629 * Allocation failed presumably due to memory. Undo anything
1630 * we might have messed up.
1632 * Allocating task should really be put onto the front of the
1633 * wait queue, but this is pretty rare.
1635 spin_lock_irq(q->queue_lock);
1636 freed_request(q, rw);
1640 if (ioc_batching(ioc))
1641 ioc->nr_batch_requests--;
1643 INIT_LIST_HEAD(&rq->queuelist);
1646 * first three bits are identical in rq->flags and bio->bi_rw,
1647 * see bio.h and blkdev.h
1652 rq->rq_status = RQ_ACTIVE;
1653 rq->bio = rq->biotail = NULL;
1665 put_io_context(ioc);
1669 elv_set_congested(q);
1670 spin_unlock_irq(q->queue_lock);
1675 * No available requests for this queue, unplug the device and wait for some
1676 * requests to become available.
1678 static struct request *get_request_wait(request_queue_t *q, int rw)
1683 generic_unplug_device(q);
1685 struct request_list *rl = &q->rq;
1687 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1688 TASK_UNINTERRUPTIBLE);
1690 rq = get_request(q, rw, GFP_NOIO);
1693 struct io_context *ioc;
1698 * After sleeping, we become a "batching" process and
1699 * will be able to allocate at least one request, and
1700 * up to a big batch of them for a small period time.
1701 * See ioc_batching, ioc_set_batching
1703 ioc = get_io_context(GFP_NOIO);
1704 ioc_set_batching(ioc);
1705 put_io_context(ioc);
1707 finish_wait(&rl->wait[rw], &wait);
1713 struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask)
1717 BUG_ON(rw != READ && rw != WRITE);
1719 if (gfp_mask & __GFP_WAIT)
1720 rq = get_request_wait(q, rw);
1722 rq = get_request(q, rw, gfp_mask);
1727 EXPORT_SYMBOL(blk_get_request);
1730 * blk_requeue_request - put a request back on queue
1731 * @q: request queue where request should be inserted
1732 * @rq: request to be inserted
1735 * Drivers often keep queueing requests until the hardware cannot accept
1736 * more, when that condition happens we need to put the request back
1737 * on the queue. Must be called with queue lock held.
1739 void blk_requeue_request(request_queue_t *q, struct request *rq)
1741 if (blk_rq_tagged(rq))
1742 blk_queue_end_tag(q, rq);
1744 elv_requeue_request(q, rq);
1747 EXPORT_SYMBOL(blk_requeue_request);
1750 * blk_insert_request - insert a special request in to a request queue
1751 * @q: request queue where request should be inserted
1752 * @rq: request to be inserted
1753 * @at_head: insert request at head or tail of queue
1754 * @data: private data
1755 * @reinsert: true if request it a reinsertion of previously processed one
1758 * Many block devices need to execute commands asynchronously, so they don't
1759 * block the whole kernel from preemption during request execution. This is
1760 * accomplished normally by inserting aritficial requests tagged as
1761 * REQ_SPECIAL in to the corresponding request queue, and letting them be
1762 * scheduled for actual execution by the request queue.
1764 * We have the option of inserting the head or the tail of the queue.
1765 * Typically we use the tail for new ioctls and so forth. We use the head
1766 * of the queue for things like a QUEUE_FULL message from a device, or a
1767 * host that is unable to accept a particular command.
1769 void blk_insert_request(request_queue_t *q, struct request *rq,
1770 int at_head, void *data, int reinsert)
1772 unsigned long flags;
1775 * tell I/O scheduler that this isn't a regular read/write (ie it
1776 * must not attempt merges on this) and that it acts as a soft
1779 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
1783 spin_lock_irqsave(q->queue_lock, flags);
1786 * If command is tagged, release the tag
1789 blk_requeue_request(q, rq);
1791 int where = ELEVATOR_INSERT_BACK;
1794 where = ELEVATOR_INSERT_FRONT;
1796 if (blk_rq_tagged(rq))
1797 blk_queue_end_tag(q, rq);
1799 drive_stat_acct(rq, rq->nr_sectors, 1);
1800 __elv_add_request(q, rq, where, 0);
1802 if (blk_queue_plugged(q))
1803 __generic_unplug_device(q);
1806 spin_unlock_irqrestore(q->queue_lock, flags);
1809 EXPORT_SYMBOL(blk_insert_request);
1812 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
1813 * @q: request queue where request should be inserted
1814 * @rw: READ or WRITE data
1815 * @ubuf: the user buffer
1816 * @len: length of user data
1819 * Data will be mapped directly for zero copy io, if possible. Otherwise
1820 * a kernel bounce buffer is used.
1822 * A matching blk_rq_unmap_user() must be issued at the end of io, while
1823 * still in process context.
1825 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
1826 * before being submitted to the device, as pages mapped may be out of
1827 * reach. It's the callers responsibility to make sure this happens. The
1828 * original bio must be passed back in to blk_rq_unmap_user() for proper
1831 struct request *blk_rq_map_user(request_queue_t *q, int rw, void __user *ubuf,
1834 unsigned long uaddr;
1838 if (len > (q->max_sectors << 9))
1839 return ERR_PTR(-EINVAL);
1840 if ((!len && ubuf) || (len && !ubuf))
1841 return ERR_PTR(-EINVAL);
1843 rq = blk_get_request(q, rw, __GFP_WAIT);
1845 return ERR_PTR(-ENOMEM);
1848 * if alignment requirement is satisfied, map in user pages for
1849 * direct dma. else, set up kernel bounce buffers
1851 uaddr = (unsigned long) ubuf;
1852 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
1853 bio = bio_map_user(q, NULL, uaddr, len, rw == READ);
1855 bio = bio_copy_user(q, uaddr, len, rw == READ);
1858 rq->bio = rq->biotail = bio;
1859 blk_rq_bio_prep(q, rq, bio);
1861 rq->buffer = rq->data = NULL;
1867 * bio is the err-ptr
1869 blk_put_request(rq);
1870 return (struct request *) bio;
1873 EXPORT_SYMBOL(blk_rq_map_user);
1876 * blk_rq_unmap_user - unmap a request with user data
1877 * @rq: request to be unmapped
1878 * @ubuf: user buffer
1879 * @ulen: length of user buffer
1882 * Unmap a request previously mapped by blk_rq_map_user().
1884 int blk_rq_unmap_user(struct request *rq, struct bio *bio, unsigned int ulen)
1889 if (bio_flagged(bio, BIO_USER_MAPPED))
1890 bio_unmap_user(bio);
1892 ret = bio_uncopy_user(bio);
1895 blk_put_request(rq);
1899 EXPORT_SYMBOL(blk_rq_unmap_user);
1902 * blk_execute_rq - insert a request into queue for execution
1903 * @q: queue to insert the request in
1904 * @bd_disk: matching gendisk
1905 * @rq: request to insert
1908 * Insert a fully prepared request at the back of the io scheduler queue
1911 int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
1914 DECLARE_COMPLETION(wait);
1915 char sense[SCSI_SENSE_BUFFERSIZE];
1918 rq->rq_disk = bd_disk;
1921 * we need an extra reference to the request, so we can look at
1922 * it after io completion
1927 memset(sense, 0, sizeof(sense));
1932 rq->flags |= REQ_NOMERGE;
1933 rq->waiting = &wait;
1934 elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1);
1935 generic_unplug_device(q);
1936 wait_for_completion(&wait);
1945 EXPORT_SYMBOL(blk_execute_rq);
1947 void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
1949 int rw = rq_data_dir(rq);
1951 if (!blk_fs_request(rq) || !rq->rq_disk)
1955 disk_stat_add(rq->rq_disk, read_sectors, nr_sectors);
1957 disk_stat_inc(rq->rq_disk, read_merges);
1958 } else if (rw == WRITE) {
1959 disk_stat_add(rq->rq_disk, write_sectors, nr_sectors);
1961 disk_stat_inc(rq->rq_disk, write_merges);
1964 disk_round_stats(rq->rq_disk);
1965 rq->rq_disk->in_flight++;
1970 * add-request adds a request to the linked list.
1971 * queue lock is held and interrupts disabled, as we muck with the
1972 * request queue list.
1974 static inline void add_request(request_queue_t * q, struct request * req)
1976 drive_stat_acct(req, req->nr_sectors, 1);
1979 q->activity_fn(q->activity_data, rq_data_dir(req));
1982 * elevator indicated where it wants this request to be
1983 * inserted at elevator_merge time
1985 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
1989 * disk_round_stats() - Round off the performance stats on a struct
1992 * The average IO queue length and utilisation statistics are maintained
1993 * by observing the current state of the queue length and the amount of
1994 * time it has been in this state for.
1996 * Normally, that accounting is done on IO completion, but that can result
1997 * in more than a second's worth of IO being accounted for within any one
1998 * second, leading to >100% utilisation. To deal with that, we call this
1999 * function to do a round-off before returning the results when reading
2000 * /proc/diskstats. This accounts immediately for all queue usage up to
2001 * the current jiffies and restarts the counters again.
2003 void disk_round_stats(struct gendisk *disk)
2005 unsigned long now = jiffies;
2007 disk_stat_add(disk, time_in_queue,
2008 disk->in_flight * (now - disk->stamp));
2011 if (disk->in_flight)
2012 disk_stat_add(disk, io_ticks, (now - disk->stamp_idle));
2013 disk->stamp_idle = now;
2017 * queue lock must be held
2019 void __blk_put_request(request_queue_t *q, struct request *req)
2021 struct request_list *rl = req->rl;
2025 if (unlikely(--req->ref_count))
2028 req->rq_status = RQ_INACTIVE;
2033 * Request may not have originated from ll_rw_blk. if not,
2034 * it didn't come out of our reserved rq pools
2037 int rw = rq_data_dir(req);
2039 elv_completed_request(q, req);
2041 BUG_ON(!list_empty(&req->queuelist));
2043 blk_free_request(q, req);
2044 freed_request(q, rw);
2048 void blk_put_request(struct request *req)
2051 * if req->rl isn't set, this request didnt originate from the
2052 * block layer, so it's safe to just disregard it
2055 unsigned long flags;
2056 request_queue_t *q = req->q;
2058 spin_lock_irqsave(q->queue_lock, flags);
2059 __blk_put_request(q, req);
2060 spin_unlock_irqrestore(q->queue_lock, flags);
2064 EXPORT_SYMBOL(blk_put_request);
2067 * blk_congestion_wait - wait for a queue to become uncongested
2068 * @rw: READ or WRITE
2069 * @timeout: timeout in jiffies
2071 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2072 * If no queues are congested then just wait for the next request to be
2075 long blk_congestion_wait(int rw, long timeout)
2079 wait_queue_head_t *wqh = &congestion_wqh[rw];
2081 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2082 ret = io_schedule_timeout(timeout);
2083 finish_wait(wqh, &wait);
2087 EXPORT_SYMBOL(blk_congestion_wait);
2090 * Has to be called with the request spinlock acquired
2092 static int attempt_merge(request_queue_t *q, struct request *req,
2093 struct request *next)
2095 if (!rq_mergeable(req) || !rq_mergeable(next))
2101 if (req->sector + req->nr_sectors != next->sector)
2104 if (rq_data_dir(req) != rq_data_dir(next)
2105 || req->rq_disk != next->rq_disk
2106 || next->waiting || next->special)
2110 * If we are allowed to merge, then append bio list
2111 * from next to rq and release next. merge_requests_fn
2112 * will have updated segment counts, update sector
2115 if (!q->merge_requests_fn(q, req, next))
2119 * At this point we have either done a back merge
2120 * or front merge. We need the smaller start_time of
2121 * the merged requests to be the current request
2122 * for accounting purposes.
2124 if (time_after(req->start_time, next->start_time))
2125 req->start_time = next->start_time;
2127 req->biotail->bi_next = next->bio;
2128 req->biotail = next->biotail;
2130 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2132 elv_merge_requests(q, req, next);
2135 disk_round_stats(req->rq_disk);
2136 req->rq_disk->in_flight--;
2139 __blk_put_request(q, next);
2143 static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2145 struct request *next = elv_latter_request(q, rq);
2148 return attempt_merge(q, rq, next);
2153 static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2155 struct request *prev = elv_former_request(q, rq);
2158 return attempt_merge(q, prev, rq);
2164 * blk_attempt_remerge - attempt to remerge active head with next request
2165 * @q: The &request_queue_t belonging to the device
2166 * @rq: The head request (usually)
2169 * For head-active devices, the queue can easily be unplugged so quickly
2170 * that proper merging is not done on the front request. This may hurt
2171 * performance greatly for some devices. The block layer cannot safely
2172 * do merging on that first request for these queues, but the driver can
2173 * call this function and make it happen any way. Only the driver knows
2174 * when it is safe to do so.
2176 void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2178 unsigned long flags;
2180 spin_lock_irqsave(q->queue_lock, flags);
2181 attempt_back_merge(q, rq);
2182 spin_unlock_irqrestore(q->queue_lock, flags);
2185 EXPORT_SYMBOL(blk_attempt_remerge);
2188 * Non-locking blk_attempt_remerge variant.
2190 void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
2192 attempt_back_merge(q, rq);
2195 EXPORT_SYMBOL(__blk_attempt_remerge);
2197 static int __make_request(request_queue_t *q, struct bio *bio)
2199 struct request *req, *freereq = NULL;
2200 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, ra;
2203 sector = bio->bi_sector;
2204 nr_sectors = bio_sectors(bio);
2205 cur_nr_sectors = bio_cur_sectors(bio);
2207 rw = bio_data_dir(bio);
2210 * low level driver can indicate that it wants pages above a
2211 * certain limit bounced to low memory (ie for highmem, or even
2212 * ISA dma in theory)
2214 blk_queue_bounce(q, &bio);
2216 spin_lock_prefetch(q->queue_lock);
2218 barrier = test_bit(BIO_RW_BARRIER, &bio->bi_rw);
2220 ra = bio->bi_rw & (1 << BIO_RW_AHEAD);
2223 spin_lock_irq(q->queue_lock);
2225 if (elv_queue_empty(q)) {
2232 el_ret = elv_merge(q, &req, bio);
2234 case ELEVATOR_BACK_MERGE:
2235 BUG_ON(!rq_mergeable(req));
2237 if (!q->back_merge_fn(q, req, bio))
2240 req->biotail->bi_next = bio;
2242 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2243 drive_stat_acct(req, nr_sectors, 0);
2244 if (!attempt_back_merge(q, req))
2245 elv_merged_request(q, req);
2248 case ELEVATOR_FRONT_MERGE:
2249 BUG_ON(!rq_mergeable(req));
2251 if (!q->front_merge_fn(q, req, bio))
2254 bio->bi_next = req->bio;
2255 req->cbio = req->bio = bio;
2256 req->nr_cbio_segments = bio_segments(bio);
2257 req->nr_cbio_sectors = bio_sectors(bio);
2260 * may not be valid. if the low level driver said
2261 * it didn't need a bounce buffer then it better
2262 * not touch req->buffer either...
2264 req->buffer = bio_data(bio);
2265 req->current_nr_sectors = cur_nr_sectors;
2266 req->hard_cur_sectors = cur_nr_sectors;
2267 req->sector = req->hard_sector = sector;
2268 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2269 drive_stat_acct(req, nr_sectors, 0);
2270 if (!attempt_front_merge(q, req))
2271 elv_merged_request(q, req);
2275 * elevator says don't/can't merge. get new request
2277 case ELEVATOR_NO_MERGE:
2281 printk("elevator returned crap (%d)\n", el_ret);
2286 * Grab a free request from the freelist - if that is empty, check
2287 * if we are doing read ahead and abort instead of blocking for
2295 spin_unlock_irq(q->queue_lock);
2296 if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) {
2303 freereq = get_request_wait(q, rw);
2308 req->flags |= REQ_CMD;
2311 * inherit FAILFAST from bio and don't stack up
2312 * retries for read ahead
2314 if (ra || test_bit(BIO_RW_FAILFAST, &bio->bi_rw))
2315 req->flags |= REQ_FAILFAST;
2318 * REQ_BARRIER implies no merging, but lets make it explicit
2321 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2324 req->hard_sector = req->sector = sector;
2325 req->hard_nr_sectors = req->nr_sectors = nr_sectors;
2326 req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
2327 req->nr_phys_segments = bio_phys_segments(q, bio);
2328 req->nr_hw_segments = bio_hw_segments(q, bio);
2329 req->nr_cbio_segments = bio_segments(bio);
2330 req->nr_cbio_sectors = bio_sectors(bio);
2331 req->buffer = bio_data(bio); /* see ->buffer comment above */
2332 req->waiting = NULL;
2333 req->cbio = req->bio = req->biotail = bio;
2334 req->rq_disk = bio->bi_bdev->bd_disk;
2335 req->start_time = jiffies;
2337 add_request(q, req);
2340 __blk_put_request(q, freereq);
2342 __generic_unplug_device(q);
2344 spin_unlock_irq(q->queue_lock);
2348 bio_endio(bio, nr_sectors << 9, -EWOULDBLOCK);
2353 * If bio->bi_dev is a partition, remap the location
2355 static inline void blk_partition_remap(struct bio *bio)
2357 struct block_device *bdev = bio->bi_bdev;
2359 if (bdev != bdev->bd_contains) {
2360 struct hd_struct *p = bdev->bd_part;
2362 switch (bio->bi_rw) {
2364 p->read_sectors += bio_sectors(bio);
2368 p->write_sectors += bio_sectors(bio);
2372 bio->bi_sector += p->start_sect;
2373 bio->bi_bdev = bdev->bd_contains;
2378 * generic_make_request: hand a buffer to its device driver for I/O
2379 * @bio: The bio describing the location in memory and on the device.
2381 * generic_make_request() is used to make I/O requests of block
2382 * devices. It is passed a &struct bio, which describes the I/O that needs
2385 * generic_make_request() does not return any status. The
2386 * success/failure status of the request, along with notification of
2387 * completion, is delivered asynchronously through the bio->bi_end_io
2388 * function described (one day) else where.
2390 * The caller of generic_make_request must make sure that bi_io_vec
2391 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2392 * set to describe the device address, and the
2393 * bi_end_io and optionally bi_private are set to describe how
2394 * completion notification should be signaled.
2396 * generic_make_request and the drivers it calls may use bi_next if this
2397 * bio happens to be merged with someone else, and may change bi_dev and
2398 * bi_sector for remaps as it sees fit. So the values of these fields
2399 * should NOT be depended on after the call to generic_make_request.
2401 void generic_make_request(struct bio *bio)
2405 int ret, nr_sectors = bio_sectors(bio);
2408 /* Test device or partition size, when known. */
2409 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2411 sector_t sector = bio->bi_sector;
2413 if (maxsector < nr_sectors ||
2414 maxsector - nr_sectors < sector) {
2415 char b[BDEVNAME_SIZE];
2416 /* This may well happen - the kernel calls
2417 * bread() without checking the size of the
2418 * device, e.g., when mounting a device. */
2420 "attempt to access beyond end of device\n");
2421 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2422 bdevname(bio->bi_bdev, b),
2424 (unsigned long long) sector + nr_sectors,
2425 (long long) maxsector);
2427 set_bit(BIO_EOF, &bio->bi_flags);
2433 * Resolve the mapping until finished. (drivers are
2434 * still free to implement/resolve their own stacking
2435 * by explicitly returning 0)
2437 * NOTE: we don't repeat the blk_size check for each new device.
2438 * Stacking drivers are expected to know what they are doing.
2441 char b[BDEVNAME_SIZE];
2443 q = bdev_get_queue(bio->bi_bdev);
2446 "generic_make_request: Trying to access "
2447 "nonexistent block-device %s (%Lu)\n",
2448 bdevname(bio->bi_bdev, b),
2449 (long long) bio->bi_sector);
2451 bio_endio(bio, bio->bi_size, -EIO);
2455 if (unlikely(bio_sectors(bio) > q->max_sectors)) {
2456 printk("bio too big device %s (%u > %u)\n",
2457 bdevname(bio->bi_bdev, b),
2463 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))
2467 * If this device has partitions, remap block n
2468 * of partition p to block n+start(p) of the disk.
2470 blk_partition_remap(bio);
2472 ret = q->make_request_fn(q, bio);
2476 EXPORT_SYMBOL(generic_make_request);
2479 * submit_bio: submit a bio to the block device layer for I/O
2480 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
2481 * @bio: The &struct bio which describes the I/O
2483 * submit_bio() is very similar in purpose to generic_make_request(), and
2484 * uses that function to do most of the work. Both are fairly rough
2485 * interfaces, @bio must be presetup and ready for I/O.
2488 void submit_bio(int rw, struct bio *bio)
2490 int count = bio_sectors(bio);
2492 BIO_BUG_ON(!bio->bi_size);
2493 BIO_BUG_ON(!bio->bi_io_vec);
2496 mod_page_state(pgpgout, count);
2498 mod_page_state(pgpgin, count);
2500 if (unlikely(block_dump)) {
2501 char b[BDEVNAME_SIZE];
2502 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
2503 current->comm, current->pid,
2504 (rw & WRITE) ? "WRITE" : "READ",
2505 (unsigned long long)bio->bi_sector,
2506 bdevname(bio->bi_bdev,b));
2509 generic_make_request(bio);
2512 EXPORT_SYMBOL(submit_bio);
2515 * blk_rq_next_segment
2516 * @rq: the request being processed
2519 * Points to the next segment in the request if the current segment
2520 * is complete. Leaves things unchanged if this segment is not over
2521 * or if no more segments are left in this request.
2523 * Meant to be used for bio traversal during I/O submission
2524 * Does not affect any I/O completions or update completion state
2525 * in the request, and does not modify any bio fields.
2527 * Decrementing rq->nr_sectors, rq->current_nr_sectors and
2528 * rq->nr_cbio_sectors as data is transferred is the caller's
2529 * responsibility and should be done before calling this routine.
2531 void blk_rq_next_segment(struct request *rq)
2533 if (rq->current_nr_sectors > 0)
2536 if (rq->nr_cbio_sectors > 0) {
2537 --rq->nr_cbio_segments;
2538 rq->current_nr_sectors = blk_rq_vec(rq)->bv_len >> 9;
2540 if ((rq->cbio = rq->cbio->bi_next)) {
2541 rq->nr_cbio_segments = bio_segments(rq->cbio);
2542 rq->nr_cbio_sectors = bio_sectors(rq->cbio);
2543 rq->current_nr_sectors = bio_cur_sectors(rq->cbio);
2547 /* remember the size of this segment before we start I/O */
2548 rq->hard_cur_sectors = rq->current_nr_sectors;
2552 * process_that_request_first - process partial request submission
2553 * @req: the request being processed
2554 * @nr_sectors: number of sectors I/O has been submitted on
2557 * May be used for processing bio's while submitting I/O without
2558 * signalling completion. Fails if more data is requested than is
2559 * available in the request in which case it doesn't advance any
2562 * Assumes a request is correctly set up. No sanity checks.
2565 * 0 - no more data left to submit (not processed)
2566 * 1 - data available to submit for this request (processed)
2568 int process_that_request_first(struct request *req, unsigned int nr_sectors)
2572 if (req->nr_sectors < nr_sectors)
2575 req->nr_sectors -= nr_sectors;
2576 req->sector += nr_sectors;
2577 while (nr_sectors) {
2578 nsect = min_t(unsigned, req->current_nr_sectors, nr_sectors);
2579 req->current_nr_sectors -= nsect;
2580 nr_sectors -= nsect;
2582 req->nr_cbio_sectors -= nsect;
2583 blk_rq_next_segment(req);
2589 EXPORT_SYMBOL(process_that_request_first);
2591 void blk_recalc_rq_segments(struct request *rq)
2593 struct bio *bio, *prevbio = NULL;
2594 int nr_phys_segs, nr_hw_segs;
2599 nr_phys_segs = nr_hw_segs = 0;
2600 rq_for_each_bio(bio, rq) {
2601 /* Force bio hw/phys segs to be recalculated. */
2602 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
2604 nr_phys_segs += bio_phys_segments(rq->q, bio);
2605 nr_hw_segs += bio_hw_segments(rq->q, bio);
2607 if (blk_phys_contig_segment(rq->q, prevbio, bio))
2609 if (blk_hw_contig_segment(rq->q, prevbio, bio))
2615 rq->nr_phys_segments = nr_phys_segs;
2616 rq->nr_hw_segments = nr_hw_segs;
2619 void blk_recalc_rq_sectors(struct request *rq, int nsect)
2621 if (blk_fs_request(rq)) {
2622 rq->hard_sector += nsect;
2623 rq->hard_nr_sectors -= nsect;
2626 * Move the I/O submission pointers ahead if required,
2627 * i.e. for drivers not aware of rq->cbio.
2629 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
2630 (rq->sector <= rq->hard_sector)) {
2631 rq->sector = rq->hard_sector;
2632 rq->nr_sectors = rq->hard_nr_sectors;
2633 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
2634 rq->current_nr_sectors = rq->hard_cur_sectors;
2635 rq->nr_cbio_segments = bio_segments(rq->bio);
2636 rq->nr_cbio_sectors = bio_sectors(rq->bio);
2637 rq->buffer = bio_data(rq->bio);
2643 * if total number of sectors is less than the first segment
2644 * size, something has gone terribly wrong
2646 if (rq->nr_sectors < rq->current_nr_sectors) {
2647 printk("blk: request botched\n");
2648 rq->nr_sectors = rq->current_nr_sectors;
2653 static int __end_that_request_first(struct request *req, int uptodate,
2656 int total_bytes, bio_nbytes, error = 0, next_idx = 0;
2660 * for a REQ_BLOCK_PC request, we want to carry any eventual
2661 * sense key with us all the way through
2663 if (!blk_pc_request(req))
2668 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
2669 printk("end_request: I/O error, dev %s, sector %llu\n",
2670 req->rq_disk ? req->rq_disk->disk_name : "?",
2671 (unsigned long long)req->sector);
2674 total_bytes = bio_nbytes = 0;
2675 while ((bio = req->bio) != NULL) {
2678 if (nr_bytes >= bio->bi_size) {
2679 req->bio = bio->bi_next;
2680 nbytes = bio->bi_size;
2681 bio_endio(bio, nbytes, error);
2685 int idx = bio->bi_idx + next_idx;
2687 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
2688 blk_dump_rq_flags(req, "__end_that");
2689 printk("%s: bio idx %d >= vcnt %d\n",
2691 bio->bi_idx, bio->bi_vcnt);
2695 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2696 BIO_BUG_ON(nbytes > bio->bi_size);
2699 * not a complete bvec done
2701 if (unlikely(nbytes > nr_bytes)) {
2702 bio_nbytes += nr_bytes;
2703 total_bytes += nr_bytes;
2708 * advance to the next vector
2711 bio_nbytes += nbytes;
2714 total_bytes += nbytes;
2717 if ((bio = req->bio)) {
2719 * end more in this run, or just return 'not-done'
2721 if (unlikely(nr_bytes <= 0))
2733 * if the request wasn't completed, update state
2736 bio_endio(bio, bio_nbytes, error);
2737 bio->bi_idx += next_idx;
2738 bio_iovec(bio)->bv_offset += nr_bytes;
2739 bio_iovec(bio)->bv_len -= nr_bytes;
2742 blk_recalc_rq_sectors(req, total_bytes >> 9);
2743 blk_recalc_rq_segments(req);
2748 * end_that_request_first - end I/O on a request
2749 * @req: the request being processed
2750 * @uptodate: 0 for I/O error
2751 * @nr_sectors: number of sectors to end I/O on
2754 * Ends I/O on a number of sectors attached to @req, and sets it up
2755 * for the next range of segments (if any) in the cluster.
2758 * 0 - we are done with this request, call end_that_request_last()
2759 * 1 - still buffers pending for this request
2761 int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
2763 return __end_that_request_first(req, uptodate, nr_sectors << 9);
2766 EXPORT_SYMBOL(end_that_request_first);
2769 * end_that_request_chunk - end I/O on a request
2770 * @req: the request being processed
2771 * @uptodate: 0 for I/O error
2772 * @nr_bytes: number of bytes to complete
2775 * Ends I/O on a number of bytes attached to @req, and sets it up
2776 * for the next range of segments (if any). Like end_that_request_first(),
2777 * but deals with bytes instead of sectors.
2780 * 0 - we are done with this request, call end_that_request_last()
2781 * 1 - still buffers pending for this request
2783 int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
2785 return __end_that_request_first(req, uptodate, nr_bytes);
2788 EXPORT_SYMBOL(end_that_request_chunk);
2791 * queue lock must be held
2793 void end_that_request_last(struct request *req)
2795 struct gendisk *disk = req->rq_disk;
2796 struct completion *waiting = req->waiting;
2798 if (unlikely(laptop_mode) && blk_fs_request(req))
2799 laptop_io_completion();
2801 if (disk && blk_fs_request(req)) {
2802 unsigned long duration = jiffies - req->start_time;
2803 switch (rq_data_dir(req)) {
2805 disk_stat_inc(disk, writes);
2806 disk_stat_add(disk, write_ticks, duration);
2809 disk_stat_inc(disk, reads);
2810 disk_stat_add(disk, read_ticks, duration);
2813 disk_round_stats(disk);
2816 __blk_put_request(req->q, req);
2817 /* Do this LAST! The structure may be freed immediately afterwards */
2822 EXPORT_SYMBOL(end_that_request_last);
2824 void end_request(struct request *req, int uptodate)
2826 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
2827 add_disk_randomness(req->rq_disk);
2828 blkdev_dequeue_request(req);
2829 end_that_request_last(req);
2833 EXPORT_SYMBOL(end_request);
2835 void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
2837 /* first three bits are identical in rq->flags and bio->bi_rw */
2838 rq->flags |= (bio->bi_rw & 7);
2840 rq->nr_phys_segments = bio_phys_segments(q, bio);
2841 rq->nr_hw_segments = bio_hw_segments(q, bio);
2842 rq->current_nr_sectors = bio_cur_sectors(bio);
2843 rq->hard_cur_sectors = rq->current_nr_sectors;
2844 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
2845 rq->nr_cbio_segments = bio_segments(bio);
2846 rq->nr_cbio_sectors = bio_sectors(bio);
2847 rq->buffer = bio_data(bio);
2849 rq->cbio = rq->bio = rq->biotail = bio;
2852 EXPORT_SYMBOL(blk_rq_bio_prep);
2854 void blk_rq_prep_restart(struct request *rq)
2858 bio = rq->cbio = rq->bio;
2860 rq->nr_cbio_segments = bio_segments(bio);
2861 rq->nr_cbio_sectors = bio_sectors(bio);
2862 rq->hard_cur_sectors = bio_cur_sectors(bio);
2863 rq->buffer = bio_data(bio);
2865 rq->sector = rq->hard_sector;
2866 rq->nr_sectors = rq->hard_nr_sectors;
2867 rq->current_nr_sectors = rq->hard_cur_sectors;
2870 EXPORT_SYMBOL(blk_rq_prep_restart);
2872 int kblockd_schedule_work(struct work_struct *work)
2874 return queue_work(kblockd_workqueue, work);
2877 EXPORT_SYMBOL(kblockd_schedule_work);
2879 void kblockd_flush(void)
2881 flush_workqueue(kblockd_workqueue);
2884 int __init blk_dev_init(void)
2886 kblockd_workqueue = create_workqueue("kblockd");
2887 if (!kblockd_workqueue)
2888 panic("Failed to create kblockd\n");
2890 request_cachep = kmem_cache_create("blkdev_requests",
2891 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
2893 requestq_cachep = kmem_cache_create("blkdev_queue",
2894 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
2896 iocontext_cachep = kmem_cache_create("blkdev_ioc",
2897 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
2899 blk_max_low_pfn = max_low_pfn;
2900 blk_max_pfn = max_pfn;
2905 * IO Context helper functions
2907 void put_io_context(struct io_context *ioc)
2912 BUG_ON(atomic_read(&ioc->refcount) == 0);
2914 if (atomic_dec_and_test(&ioc->refcount)) {
2915 if (ioc->aic && ioc->aic->dtor)
2916 ioc->aic->dtor(ioc->aic);
2917 kmem_cache_free(iocontext_cachep, ioc);
2921 /* Called by the exitting task */
2922 void exit_io_context(void)
2924 unsigned long flags;
2925 struct io_context *ioc;
2927 local_irq_save(flags);
2928 ioc = current->io_context;
2930 if (ioc->aic && ioc->aic->exit)
2931 ioc->aic->exit(ioc->aic);
2932 put_io_context(ioc);
2933 current->io_context = NULL;
2936 local_irq_restore(flags);
2940 * If the current task has no IO context then create one and initialise it.
2941 * If it does have a context, take a ref on it.
2943 * This is always called in the context of the task which submitted the I/O.
2944 * But weird things happen, so we disable local interrupts to ensure exclusive
2945 * access to *current.
2947 struct io_context *get_io_context(int gfp_flags)
2949 struct task_struct *tsk = current;
2950 unsigned long flags;
2951 struct io_context *ret;
2953 local_irq_save(flags);
2954 ret = tsk->io_context;
2956 ret = kmem_cache_alloc(iocontext_cachep, GFP_ATOMIC);
2958 atomic_set(&ret->refcount, 1);
2959 ret->pid = tsk->pid;
2960 ret->last_waited = jiffies; /* doesn't matter... */
2961 ret->nr_batch_requests = 0; /* because this is 0 */
2963 tsk->io_context = ret;
2967 atomic_inc(&ret->refcount);
2968 local_irq_restore(flags);
2972 void copy_io_context(struct io_context **pdst, struct io_context **psrc)
2974 struct io_context *src = *psrc;
2975 struct io_context *dst = *pdst;
2978 BUG_ON(atomic_read(&src->refcount) == 0);
2979 atomic_inc(&src->refcount);
2980 put_io_context(dst);
2985 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
2987 struct io_context *temp;
2997 struct queue_sysfs_entry {
2998 struct attribute attr;
2999 ssize_t (*show)(struct request_queue *, char *);
3000 ssize_t (*store)(struct request_queue *, const char *, size_t);
3004 queue_var_show(unsigned int var, char *page)
3006 return sprintf(page, "%d\n", var);
3010 queue_var_store(unsigned long *var, const char *page, size_t count)
3012 char *p = (char *) page;
3014 *var = simple_strtoul(p, &p, 10);
3018 static ssize_t queue_requests_show(struct request_queue *q, char *page)
3020 return queue_var_show(q->nr_requests, (page));
3024 queue_requests_store(struct request_queue *q, const char *page, size_t count)
3026 struct request_list *rl = &q->rq;
3028 int ret = queue_var_store(&q->nr_requests, page, count);
3029 if (q->nr_requests < BLKDEV_MIN_RQ)
3030 q->nr_requests = BLKDEV_MIN_RQ;
3031 blk_queue_congestion_threshold(q);
3033 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3034 set_queue_congested(q, READ);
3035 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3036 clear_queue_congested(q, READ);
3038 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3039 set_queue_congested(q, WRITE);
3040 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3041 clear_queue_congested(q, WRITE);
3043 if (rl->count[READ] >= q->nr_requests) {
3044 blk_set_queue_full(q, READ);
3045 } else if (rl->count[READ]+1 <= q->nr_requests) {
3046 blk_clear_queue_full(q, READ);
3047 wake_up(&rl->wait[READ]);
3050 if (rl->count[WRITE] >= q->nr_requests) {
3051 blk_set_queue_full(q, WRITE);
3052 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3053 blk_clear_queue_full(q, WRITE);
3054 wake_up(&rl->wait[WRITE]);
3059 static ssize_t queue_ra_show(struct request_queue *q, char *page)
3061 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3063 return queue_var_show(ra_kb, (page));
3067 queue_ra_store(struct request_queue *q, const char *page, size_t count)
3069 unsigned long ra_kb;
3070 ssize_t ret = queue_var_store(&ra_kb, page, count);
3072 if (ra_kb > (q->max_sectors >> 1))
3073 ra_kb = (q->max_sectors >> 1);
3075 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3079 static struct queue_sysfs_entry queue_requests_entry = {
3080 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3081 .show = queue_requests_show,
3082 .store = queue_requests_store,
3085 static struct queue_sysfs_entry queue_ra_entry = {
3086 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3087 .show = queue_ra_show,
3088 .store = queue_ra_store,
3091 static struct attribute *default_attrs[] = {
3092 &queue_requests_entry.attr,
3093 &queue_ra_entry.attr,
3097 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3100 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3102 struct queue_sysfs_entry *entry = to_queue(attr);
3103 struct request_queue *q;
3105 q = container_of(kobj, struct request_queue, kobj);
3109 return entry->show(q, page);
3113 queue_attr_store(struct kobject *kobj, struct attribute *attr,
3114 const char *page, size_t length)
3116 struct queue_sysfs_entry *entry = to_queue(attr);
3117 struct request_queue *q;
3119 q = container_of(kobj, struct request_queue, kobj);
3123 return entry->store(q, page, length);
3126 static struct sysfs_ops queue_sysfs_ops = {
3127 .show = queue_attr_show,
3128 .store = queue_attr_store,
3131 struct kobj_type queue_ktype = {
3132 .sysfs_ops = &queue_sysfs_ops,
3133 .default_attrs = default_attrs,
3136 int blk_register_queue(struct gendisk *disk)
3140 request_queue_t *q = disk->queue;
3142 if (!q || !q->request_fn)
3145 q->kobj.parent = kobject_get(&disk->kobj);
3146 if (!q->kobj.parent)
3149 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3150 q->kobj.ktype = &queue_ktype;
3152 ret = kobject_register(&q->kobj);
3156 ret = elv_register_queue(q);
3158 kobject_unregister(&q->kobj);
3165 void blk_unregister_queue(struct gendisk *disk)
3167 request_queue_t *q = disk->queue;
3169 if (q && q->request_fn) {
3170 elv_unregister_queue(q);
3172 kobject_unregister(&q->kobj);
3173 kobject_put(&disk->kobj);
3177 asmlinkage int sys_ioprio_set(int ioprio)
3179 if (ioprio < IOPRIO_IDLE || ioprio > IOPRIO_RT)
3181 if (ioprio == IOPRIO_RT && !capable(CAP_SYS_ADMIN))
3184 printk("%s: set ioprio %d\n", current->comm, ioprio);
3185 current->ioprio = ioprio;
3189 asmlinkage int sys_ioprio_get(void)
3191 return current->ioprio;