Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics |
| 4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE |
| 5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
| 7 | * - July2000 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
| 9 | */ |
| 10 | |
| 11 | /* |
| 12 | * This handles all read/write requests to block devices |
| 13 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/kernel.h> |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/backing-dev.h> |
| 17 | #include <linux/bio.h> |
| 18 | #include <linux/blkdev.h> |
| 19 | #include <linux/highmem.h> |
| 20 | #include <linux/mm.h> |
| 21 | #include <linux/kernel_stat.h> |
| 22 | #include <linux/string.h> |
| 23 | #include <linux/init.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/completion.h> |
| 25 | #include <linux/slab.h> |
| 26 | #include <linux/swap.h> |
| 27 | #include <linux/writeback.h> |
Andrew Morton | faccbd4b | 2006-12-10 02:19:35 -0800 | [diff] [blame] | 28 | #include <linux/task_io_accounting_ops.h> |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 29 | #include <linux/fault-inject.h> |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 30 | #include <linux/list_sort.h> |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 31 | #include <linux/delay.h> |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 32 | #include <linux/ratelimit.h> |
Lin Ming | 6c95466 | 2013-03-23 11:42:26 +0800 | [diff] [blame] | 33 | #include <linux/pm_runtime.h> |
Li Zefan | 5578213 | 2009-06-09 13:43:05 +0800 | [diff] [blame] | 34 | |
| 35 | #define CREATE_TRACE_POINTS |
| 36 | #include <trace/events/block.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 38 | #include "blk.h" |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 39 | #include "blk-cgroup.h" |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 40 | |
Mike Snitzer | d07335e | 2010-11-16 12:52:38 +0100 | [diff] [blame] | 41 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
Jun'ichi Nomura | b0da3f0 | 2009-10-01 21:16:13 +0200 | [diff] [blame] | 42 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
Linus Torvalds | 0a82a8d | 2013-04-18 09:00:26 -0700 | [diff] [blame] | 43 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
NeilBrown | cbae8d4 | 2012-12-14 20:49:27 +0100 | [diff] [blame] | 44 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
Ingo Molnar | 0bfc245 | 2008-11-26 11:59:56 +0100 | [diff] [blame] | 45 | |
Tejun Heo | a73f730 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 46 | DEFINE_IDA(blk_queue_ida); |
| 47 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | /* |
| 49 | * For the allocated request tables |
| 50 | */ |
Adrian Bunk | 5ece6c5 | 2008-02-18 13:45:51 +0100 | [diff] [blame] | 51 | static struct kmem_cache *request_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | |
| 53 | /* |
| 54 | * For queue allocation |
| 55 | */ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 56 | struct kmem_cache *blk_requestq_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
| 58 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | * Controlling structure to kblockd |
| 60 | */ |
Jens Axboe | ff856ba | 2006-01-09 16:02:34 +0100 | [diff] [blame] | 61 | static struct workqueue_struct *kblockd_workqueue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 63 | static void drive_stat_acct(struct request *rq, int new_io) |
| 64 | { |
Jens Axboe | 28f1370 | 2008-05-07 10:15:46 +0200 | [diff] [blame] | 65 | struct hd_struct *part; |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 66 | int rw = rq_data_dir(rq); |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 67 | int cpu; |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 68 | |
Jens Axboe | c2553b5 | 2009-04-24 08:10:11 +0200 | [diff] [blame] | 69 | if (!blk_do_io_stat(rq)) |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 70 | return; |
| 71 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 72 | cpu = part_stat_lock(); |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 73 | |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 74 | if (!new_io) { |
| 75 | part = rq->part; |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 76 | part_stat_inc(cpu, part, merges[rw]); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 77 | } else { |
| 78 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); |
Jens Axboe | 6c23a96 | 2011-01-07 08:43:37 +0100 | [diff] [blame] | 79 | if (!hd_struct_try_get(part)) { |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 80 | /* |
| 81 | * The partition is already being removed, |
| 82 | * the request will be accounted on the disk only |
| 83 | * |
| 84 | * We take a reference on disk->part0 although that |
| 85 | * partition will never be deleted, so we can treat |
| 86 | * it as any other partition. |
| 87 | */ |
| 88 | part = &rq->rq_disk->part0; |
Jens Axboe | 6c23a96 | 2011-01-07 08:43:37 +0100 | [diff] [blame] | 89 | hd_struct_get(part); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 90 | } |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 91 | part_round_stats(cpu, part); |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 92 | part_inc_in_flight(part, rw); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 93 | rq->part = part; |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 94 | } |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 95 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 96 | part_stat_unlock(); |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 97 | } |
| 98 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 99 | void blk_queue_congestion_threshold(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | { |
| 101 | int nr; |
| 102 | |
| 103 | nr = q->nr_requests - (q->nr_requests / 8) + 1; |
| 104 | if (nr > q->nr_requests) |
| 105 | nr = q->nr_requests; |
| 106 | q->nr_congestion_on = nr; |
| 107 | |
| 108 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; |
| 109 | if (nr < 1) |
| 110 | nr = 1; |
| 111 | q->nr_congestion_off = nr; |
| 112 | } |
| 113 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | /** |
| 115 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info |
| 116 | * @bdev: device |
| 117 | * |
| 118 | * Locates the passed device's request queue and returns the address of its |
| 119 | * backing_dev_info |
| 120 | * |
| 121 | * Will return NULL if the request queue cannot be located. |
| 122 | */ |
| 123 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) |
| 124 | { |
| 125 | struct backing_dev_info *ret = NULL; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 126 | struct request_queue *q = bdev_get_queue(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | |
| 128 | if (q) |
| 129 | ret = &q->backing_dev_info; |
| 130 | return ret; |
| 131 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
| 133 | |
FUJITA Tomonori | 2a4aa30 | 2008-04-29 09:54:36 +0200 | [diff] [blame] | 134 | void blk_rq_init(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | { |
FUJITA Tomonori | 1afb20f | 2008-04-25 12:26:28 +0200 | [diff] [blame] | 136 | memset(rq, 0, sizeof(*rq)); |
| 137 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | INIT_LIST_HEAD(&rq->queuelist); |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 139 | INIT_LIST_HEAD(&rq->timeout_list); |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 140 | rq->cpu = -1; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 141 | rq->q = q; |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 142 | rq->__sector = (sector_t) -1; |
Jens Axboe | 2e662b6 | 2006-07-13 11:55:04 +0200 | [diff] [blame] | 143 | INIT_HLIST_NODE(&rq->hash); |
| 144 | RB_CLEAR_NODE(&rq->rb_node); |
FUJITA Tomonori | d7e3c32 | 2008-04-29 09:54:39 +0200 | [diff] [blame] | 145 | rq->cmd = rq->__cmd; |
Li Zefan | e2494e1 | 2009-04-02 13:43:26 +0800 | [diff] [blame] | 146 | rq->cmd_len = BLK_MAX_CDB; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 147 | rq->tag = -1; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 148 | rq->ref_count = 1; |
Tejun Heo | b243ddc | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 149 | rq->start_time = jiffies; |
Divyesh Shah | 9195291 | 2010-04-01 15:01:41 -0700 | [diff] [blame] | 150 | set_start_time_ns(rq); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 151 | rq->part = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | } |
FUJITA Tomonori | 2a4aa30 | 2008-04-29 09:54:36 +0200 | [diff] [blame] | 153 | EXPORT_SYMBOL(blk_rq_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 155 | static void req_bio_endio(struct request *rq, struct bio *bio, |
| 156 | unsigned int nbytes, int error) |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 157 | { |
Tejun Heo | 143a87f | 2011-01-25 12:43:52 +0100 | [diff] [blame] | 158 | if (error) |
| 159 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
| 160 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
| 161 | error = -EIO; |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 162 | |
Tejun Heo | 143a87f | 2011-01-25 12:43:52 +0100 | [diff] [blame] | 163 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
| 164 | set_bit(BIO_QUIET, &bio->bi_flags); |
| 165 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 166 | bio_advance(bio, nbytes); |
Tejun Heo | 143a87f | 2011-01-25 12:43:52 +0100 | [diff] [blame] | 167 | |
| 168 | /* don't actually finish bio if it's part of flush sequence */ |
| 169 | if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) |
| 170 | bio_endio(bio, error); |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 171 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | void blk_dump_rq_flags(struct request *rq, char *msg) |
| 174 | { |
| 175 | int bit; |
| 176 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 177 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 178 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
| 179 | rq->cmd_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 181 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
| 182 | (unsigned long long)blk_rq_pos(rq), |
| 183 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); |
Tejun Heo | 731ec49 | 2009-04-23 11:05:20 +0900 | [diff] [blame] | 184 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 185 | rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 187 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 188 | printk(KERN_INFO " cdb: "); |
FUJITA Tomonori | d34c87e | 2008-04-29 14:37:52 +0200 | [diff] [blame] | 189 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | printk("%02x ", rq->cmd[bit]); |
| 191 | printk("\n"); |
| 192 | } |
| 193 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | EXPORT_SYMBOL(blk_dump_rq_flags); |
| 195 | |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 196 | static void blk_delay_work(struct work_struct *work) |
Jens Axboe | 6c5e0c4 | 2008-08-01 20:31:32 +0200 | [diff] [blame] | 197 | { |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 198 | struct request_queue *q; |
Jens Axboe | 6c5e0c4 | 2008-08-01 20:31:32 +0200 | [diff] [blame] | 199 | |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 200 | q = container_of(work, struct request_queue, delay_work.work); |
| 201 | spin_lock_irq(q->queue_lock); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 202 | __blk_run_queue(q); |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 203 | spin_unlock_irq(q->queue_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | |
| 206 | /** |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 207 | * blk_delay_queue - restart queueing after defined interval |
| 208 | * @q: The &struct request_queue in question |
| 209 | * @msecs: Delay in msecs |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | * |
| 211 | * Description: |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 212 | * Sometimes queueing needs to be postponed for a little while, to allow |
| 213 | * resources to come back. This function will make sure that queueing is |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 214 | * restarted around the specified time. Queue lock must be held. |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 215 | */ |
| 216 | void blk_delay_queue(struct request_queue *q, unsigned long msecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | { |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 218 | if (likely(!blk_queue_dead(q))) |
| 219 | queue_delayed_work(kblockd_workqueue, &q->delay_work, |
| 220 | msecs_to_jiffies(msecs)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | } |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 222 | EXPORT_SYMBOL(blk_delay_queue); |
Alan D. Brunelle | 2ad8b1e | 2007-11-07 14:26:56 -0500 | [diff] [blame] | 223 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | /** |
| 225 | * blk_start_queue - restart a previously stopped queue |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 226 | * @q: The &struct request_queue in question |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | * |
| 228 | * Description: |
| 229 | * blk_start_queue() will clear the stop flag on the queue, and call |
| 230 | * the request_fn for the queue if it was in a stopped state when |
| 231 | * entered. Also see blk_stop_queue(). Queue lock must be held. |
| 232 | **/ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 233 | void blk_start_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | { |
Paolo 'Blaisorblade' Giarrusso | a038e25 | 2006-06-05 12:09:01 +0200 | [diff] [blame] | 235 | WARN_ON(!irqs_disabled()); |
| 236 | |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 237 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 238 | __blk_run_queue(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | EXPORT_SYMBOL(blk_start_queue); |
| 241 | |
| 242 | /** |
| 243 | * blk_stop_queue - stop a queue |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 244 | * @q: The &struct request_queue in question |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | * |
| 246 | * Description: |
| 247 | * The Linux block layer assumes that a block driver will consume all |
| 248 | * entries on the request queue when the request_fn strategy is called. |
| 249 | * Often this will not happen, because of hardware limitations (queue |
| 250 | * depth settings). If a device driver gets a 'queue full' response, |
| 251 | * or if it simply chooses not to queue more I/O at one point, it can |
| 252 | * call this function to prevent the request_fn from being called until |
| 253 | * the driver has signalled it's ready to go again. This happens by calling |
| 254 | * blk_start_queue() to restart queue operations. Queue lock must be held. |
| 255 | **/ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 256 | void blk_stop_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 257 | { |
Tejun Heo | 136b572 | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 258 | cancel_delayed_work(&q->delay_work); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 259 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | } |
| 261 | EXPORT_SYMBOL(blk_stop_queue); |
| 262 | |
| 263 | /** |
| 264 | * blk_sync_queue - cancel any pending callbacks on a queue |
| 265 | * @q: the queue |
| 266 | * |
| 267 | * Description: |
| 268 | * The block layer may perform asynchronous callback activity |
| 269 | * on a queue, such as calling the unplug function after a timeout. |
| 270 | * A block device may call blk_sync_queue to ensure that any |
| 271 | * such activity is cancelled, thus allowing it to release resources |
Michael Opdenacker | 59c5159 | 2007-05-09 08:57:56 +0200 | [diff] [blame] | 272 | * that the callbacks might use. The caller must already have made sure |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | * that its ->make_request_fn will not re-add plugging prior to calling |
| 274 | * this function. |
| 275 | * |
Vivek Goyal | da52777 | 2011-03-02 19:05:33 -0500 | [diff] [blame] | 276 | * This function does not cancel any asynchronous activity arising |
| 277 | * out of elevator or throttling code. That would require elevaotor_exit() |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 278 | * and blkcg_exit_queue() to be called with queue lock initialized. |
Vivek Goyal | da52777 | 2011-03-02 19:05:33 -0500 | [diff] [blame] | 279 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | */ |
| 281 | void blk_sync_queue(struct request_queue *q) |
| 282 | { |
Jens Axboe | 70ed28b | 2008-11-19 14:38:39 +0100 | [diff] [blame] | 283 | del_timer_sync(&q->timeout); |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 284 | cancel_delayed_work_sync(&q->delay_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | } |
| 286 | EXPORT_SYMBOL(blk_sync_queue); |
| 287 | |
| 288 | /** |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 289 | * __blk_run_queue_uncond - run a queue whether or not it has been stopped |
| 290 | * @q: The queue to run |
| 291 | * |
| 292 | * Description: |
| 293 | * Invoke request handling on a queue if there are any pending requests. |
| 294 | * May be used to restart request handling after a request has completed. |
| 295 | * This variant runs the queue whether or not the queue has been |
| 296 | * stopped. Must be called with the queue lock held and interrupts |
| 297 | * disabled. See also @blk_run_queue. |
| 298 | */ |
| 299 | inline void __blk_run_queue_uncond(struct request_queue *q) |
| 300 | { |
| 301 | if (unlikely(blk_queue_dead(q))) |
| 302 | return; |
| 303 | |
Bart Van Assche | 24faf6f | 2012-11-28 13:46:45 +0100 | [diff] [blame] | 304 | /* |
| 305 | * Some request_fn implementations, e.g. scsi_request_fn(), unlock |
| 306 | * the queue lock internally. As a result multiple threads may be |
| 307 | * running such a request function concurrently. Keep track of the |
| 308 | * number of active request_fn invocations such that blk_drain_queue() |
| 309 | * can wait until all these request_fn calls have finished. |
| 310 | */ |
| 311 | q->request_fn_active++; |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 312 | q->request_fn(q); |
Bart Van Assche | 24faf6f | 2012-11-28 13:46:45 +0100 | [diff] [blame] | 313 | q->request_fn_active--; |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 314 | } |
| 315 | |
| 316 | /** |
Jens Axboe | 80a4b58 | 2008-10-14 09:51:06 +0200 | [diff] [blame] | 317 | * __blk_run_queue - run a single device queue |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | * @q: The queue to run |
Jens Axboe | 80a4b58 | 2008-10-14 09:51:06 +0200 | [diff] [blame] | 319 | * |
| 320 | * Description: |
| 321 | * See @blk_run_queue. This variant must be called with the queue lock |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 322 | * held and interrupts disabled. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | */ |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 324 | void __blk_run_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | { |
Tejun Heo | a538cd0 | 2009-04-23 11:05:17 +0900 | [diff] [blame] | 326 | if (unlikely(blk_queue_stopped(q))) |
| 327 | return; |
| 328 | |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 329 | __blk_run_queue_uncond(q); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 330 | } |
| 331 | EXPORT_SYMBOL(__blk_run_queue); |
Jens Axboe | dac07ec | 2006-05-11 08:20:16 +0200 | [diff] [blame] | 332 | |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 333 | /** |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 334 | * blk_run_queue_async - run a single device queue in workqueue context |
| 335 | * @q: The queue to run |
| 336 | * |
| 337 | * Description: |
| 338 | * Tells kblockd to perform the equivalent of @blk_run_queue on behalf |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 339 | * of us. The caller must hold the queue lock. |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 340 | */ |
| 341 | void blk_run_queue_async(struct request_queue *q) |
| 342 | { |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 343 | if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) |
Tejun Heo | e7c2f96 | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 344 | mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 345 | } |
Jens Axboe | c21e6be | 2011-04-19 13:32:46 +0200 | [diff] [blame] | 346 | EXPORT_SYMBOL(blk_run_queue_async); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 347 | |
| 348 | /** |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 349 | * blk_run_queue - run a single device queue |
| 350 | * @q: The queue to run |
Jens Axboe | 80a4b58 | 2008-10-14 09:51:06 +0200 | [diff] [blame] | 351 | * |
| 352 | * Description: |
| 353 | * Invoke request handling on this queue, if it has pending work to do. |
Tejun Heo | a7f5579 | 2009-04-23 11:05:17 +0900 | [diff] [blame] | 354 | * May be used to restart queueing when a request has completed. |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 355 | */ |
| 356 | void blk_run_queue(struct request_queue *q) |
| 357 | { |
| 358 | unsigned long flags; |
| 359 | |
| 360 | spin_lock_irqsave(q->queue_lock, flags); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 361 | __blk_run_queue(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 363 | } |
| 364 | EXPORT_SYMBOL(blk_run_queue); |
| 365 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 366 | void blk_put_queue(struct request_queue *q) |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 367 | { |
| 368 | kobject_put(&q->kobj); |
| 369 | } |
Jens Axboe | d86e0e8 | 2011-05-27 07:44:43 +0200 | [diff] [blame] | 370 | EXPORT_SYMBOL(blk_put_queue); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 371 | |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 372 | /** |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 373 | * __blk_drain_queue - drain requests from request_queue |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 374 | * @q: queue to drain |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 375 | * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 376 | * |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 377 | * Drain requests from @q. If @drain_all is set, all requests are drained. |
| 378 | * If not, only ELVPRIV requests are drained. The caller is responsible |
| 379 | * for ensuring that no new requests which need to be drained are queued. |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 380 | */ |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 381 | static void __blk_drain_queue(struct request_queue *q, bool drain_all) |
| 382 | __releases(q->queue_lock) |
| 383 | __acquires(q->queue_lock) |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 384 | { |
Asias He | 458f27a | 2012-06-15 08:45:25 +0200 | [diff] [blame] | 385 | int i; |
| 386 | |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 387 | lockdep_assert_held(q->queue_lock); |
| 388 | |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 389 | while (true) { |
Tejun Heo | 481a7d6 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 390 | bool drain = false; |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 391 | |
Tejun Heo | b855b04 | 2012-03-06 21:24:55 +0100 | [diff] [blame] | 392 | /* |
| 393 | * The caller might be trying to drain @q before its |
| 394 | * elevator is initialized. |
| 395 | */ |
| 396 | if (q->elevator) |
| 397 | elv_drain_elevator(q); |
| 398 | |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 399 | blkcg_drain_queue(q); |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 400 | |
Tejun Heo | 4eabc94 | 2011-12-15 20:03:04 +0100 | [diff] [blame] | 401 | /* |
| 402 | * This function might be called on a queue which failed |
Tejun Heo | b855b04 | 2012-03-06 21:24:55 +0100 | [diff] [blame] | 403 | * driver init after queue creation or is not yet fully |
| 404 | * active yet. Some drivers (e.g. fd and loop) get unhappy |
| 405 | * in such cases. Kick queue iff dispatch queue has |
| 406 | * something on it and @q has request_fn set. |
Tejun Heo | 4eabc94 | 2011-12-15 20:03:04 +0100 | [diff] [blame] | 407 | */ |
Tejun Heo | b855b04 | 2012-03-06 21:24:55 +0100 | [diff] [blame] | 408 | if (!list_empty(&q->queue_head) && q->request_fn) |
Tejun Heo | 4eabc94 | 2011-12-15 20:03:04 +0100 | [diff] [blame] | 409 | __blk_run_queue(q); |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 410 | |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 411 | drain |= q->nr_rqs_elvpriv; |
Bart Van Assche | 24faf6f | 2012-11-28 13:46:45 +0100 | [diff] [blame] | 412 | drain |= q->request_fn_active; |
Tejun Heo | 481a7d6 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 413 | |
| 414 | /* |
| 415 | * Unfortunately, requests are queued at and tracked from |
| 416 | * multiple places and there's no single counter which can |
| 417 | * be drained. Check all the queues and counters. |
| 418 | */ |
| 419 | if (drain_all) { |
| 420 | drain |= !list_empty(&q->queue_head); |
| 421 | for (i = 0; i < 2; i++) { |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 422 | drain |= q->nr_rqs[i]; |
Tejun Heo | 481a7d6 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 423 | drain |= q->in_flight[i]; |
| 424 | drain |= !list_empty(&q->flush_queue[i]); |
| 425 | } |
| 426 | } |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 427 | |
Tejun Heo | 481a7d6 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 428 | if (!drain) |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 429 | break; |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 430 | |
| 431 | spin_unlock_irq(q->queue_lock); |
| 432 | |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 433 | msleep(10); |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 434 | |
| 435 | spin_lock_irq(q->queue_lock); |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 436 | } |
Asias He | 458f27a | 2012-06-15 08:45:25 +0200 | [diff] [blame] | 437 | |
| 438 | /* |
| 439 | * With queue marked dead, any woken up waiter will fail the |
| 440 | * allocation path, so the wakeup chaining is lost and we're |
| 441 | * left with hung waiters. We need to wake up those waiters. |
| 442 | */ |
| 443 | if (q->request_fn) { |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 444 | struct request_list *rl; |
| 445 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 446 | blk_queue_for_each_rl(rl, q) |
| 447 | for (i = 0; i < ARRAY_SIZE(rl->wait); i++) |
| 448 | wake_up_all(&rl->wait[i]); |
Asias He | 458f27a | 2012-06-15 08:45:25 +0200 | [diff] [blame] | 449 | } |
Tejun Heo | e3c78ca | 2011-10-19 14:32:38 +0200 | [diff] [blame] | 450 | } |
| 451 | |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 452 | /** |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 453 | * blk_queue_bypass_start - enter queue bypass mode |
| 454 | * @q: queue of interest |
| 455 | * |
| 456 | * In bypass mode, only the dispatch FIFO queue of @q is used. This |
| 457 | * function makes @q enter bypass mode and drains all requests which were |
Tejun Heo | 6ecf23a | 2012-03-05 13:14:59 -0800 | [diff] [blame] | 458 | * throttled or issued before. On return, it's guaranteed that no request |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 459 | * is being throttled or has ELVPRIV set and blk_queue_bypass() %true |
| 460 | * inside queue or RCU read lock. |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 461 | */ |
| 462 | void blk_queue_bypass_start(struct request_queue *q) |
| 463 | { |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 464 | bool drain; |
| 465 | |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 466 | spin_lock_irq(q->queue_lock); |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 467 | drain = !q->bypass_depth++; |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 468 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
| 469 | spin_unlock_irq(q->queue_lock); |
| 470 | |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 471 | if (drain) { |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 472 | spin_lock_irq(q->queue_lock); |
| 473 | __blk_drain_queue(q, false); |
| 474 | spin_unlock_irq(q->queue_lock); |
| 475 | |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 476 | /* ensure blk_queue_bypass() is %true inside RCU read lock */ |
| 477 | synchronize_rcu(); |
| 478 | } |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 479 | } |
| 480 | EXPORT_SYMBOL_GPL(blk_queue_bypass_start); |
| 481 | |
| 482 | /** |
| 483 | * blk_queue_bypass_end - leave queue bypass mode |
| 484 | * @q: queue of interest |
| 485 | * |
| 486 | * Leave bypass mode and restore the normal queueing behavior. |
| 487 | */ |
| 488 | void blk_queue_bypass_end(struct request_queue *q) |
| 489 | { |
| 490 | spin_lock_irq(q->queue_lock); |
| 491 | if (!--q->bypass_depth) |
| 492 | queue_flag_clear(QUEUE_FLAG_BYPASS, q); |
| 493 | WARN_ON_ONCE(q->bypass_depth < 0); |
| 494 | spin_unlock_irq(q->queue_lock); |
| 495 | } |
| 496 | EXPORT_SYMBOL_GPL(blk_queue_bypass_end); |
| 497 | |
| 498 | /** |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 499 | * blk_cleanup_queue - shutdown a request queue |
| 500 | * @q: request queue to shutdown |
| 501 | * |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 502 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
| 503 | * put it. All future requests will be failed immediately with -ENODEV. |
Vivek Goyal | c94a96a | 2011-03-02 19:04:42 -0500 | [diff] [blame] | 504 | */ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 505 | void blk_cleanup_queue(struct request_queue *q) |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 506 | { |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 507 | spinlock_t *lock = q->queue_lock; |
Jens Axboe | e3335de | 2008-09-18 09:22:54 -0700 | [diff] [blame] | 508 | |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 509 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 510 | mutex_lock(&q->sysfs_lock); |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 511 | queue_flag_set_unlocked(QUEUE_FLAG_DYING, q); |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 512 | spin_lock_irq(lock); |
Tejun Heo | 6ecf23a | 2012-03-05 13:14:59 -0800 | [diff] [blame] | 513 | |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 514 | /* |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 515 | * A dying queue is permanently in bypass mode till released. Note |
Tejun Heo | 80fd997 | 2012-04-13 14:50:53 -0700 | [diff] [blame] | 516 | * that, unlike blk_queue_bypass_start(), we aren't performing |
| 517 | * synchronize_rcu() after entering bypass mode to avoid the delay |
| 518 | * as some drivers create and destroy a lot of queues while |
| 519 | * probing. This is still safe because blk_release_queue() will be |
| 520 | * called only after the queue refcnt drops to zero and nothing, |
| 521 | * RCU or not, would be traversing the queue by then. |
| 522 | */ |
Tejun Heo | 6ecf23a | 2012-03-05 13:14:59 -0800 | [diff] [blame] | 523 | q->bypass_depth++; |
| 524 | queue_flag_set(QUEUE_FLAG_BYPASS, q); |
| 525 | |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 526 | queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
| 527 | queue_flag_set(QUEUE_FLAG_NOXMERGES, q); |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 528 | queue_flag_set(QUEUE_FLAG_DYING, q); |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 529 | spin_unlock_irq(lock); |
| 530 | mutex_unlock(&q->sysfs_lock); |
| 531 | |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 532 | /* |
| 533 | * Drain all requests queued before DYING marking. Set DEAD flag to |
| 534 | * prevent that q->request_fn() gets invoked after draining finished. |
| 535 | */ |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 536 | spin_lock_irq(lock); |
| 537 | __blk_drain_queue(q, true); |
Bart Van Assche | c246e80 | 2012-12-06 14:32:01 +0100 | [diff] [blame] | 538 | queue_flag_set(QUEUE_FLAG_DEAD, q); |
Bart Van Assche | 807592a | 2012-11-28 13:43:38 +0100 | [diff] [blame] | 539 | spin_unlock_irq(lock); |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 540 | |
| 541 | /* @q won't process any more request, flush async actions */ |
| 542 | del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); |
| 543 | blk_sync_queue(q); |
| 544 | |
Asias He | 5e5cfac | 2012-05-24 23:28:52 +0800 | [diff] [blame] | 545 | spin_lock_irq(lock); |
| 546 | if (q->queue_lock != &q->__queue_lock) |
| 547 | q->queue_lock = &q->__queue_lock; |
| 548 | spin_unlock_irq(lock); |
| 549 | |
Tejun Heo | c9a929d | 2011-10-19 14:42:16 +0200 | [diff] [blame] | 550 | /* @q is and will stay empty, shutdown and put */ |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 551 | blk_put_queue(q); |
| 552 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | EXPORT_SYMBOL(blk_cleanup_queue); |
| 554 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 555 | int blk_init_rl(struct request_list *rl, struct request_queue *q, |
| 556 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | { |
Mike Snitzer | 1abec4f | 2010-05-25 13:15:15 -0400 | [diff] [blame] | 558 | if (unlikely(rl->rq_pool)) |
| 559 | return 0; |
| 560 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 561 | rl->q = q; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 562 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
| 563 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 564 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
| 565 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 567 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
Tejun Heo | a91a5ac | 2012-06-04 20:40:53 -0700 | [diff] [blame] | 568 | mempool_free_slab, request_cachep, |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 569 | gfp_mask, q->node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | if (!rl->rq_pool) |
| 571 | return -ENOMEM; |
| 572 | |
| 573 | return 0; |
| 574 | } |
| 575 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 576 | void blk_exit_rl(struct request_list *rl) |
| 577 | { |
| 578 | if (rl->rq_pool) |
| 579 | mempool_destroy(rl->rq_pool); |
| 580 | } |
| 581 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 582 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 583 | { |
Ezequiel Garcia | c304a51 | 2012-11-10 10:39:44 +0100 | [diff] [blame] | 584 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 585 | } |
| 586 | EXPORT_SYMBOL(blk_alloc_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 588 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 589 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 590 | struct request_queue *q; |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 591 | int err; |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 592 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 593 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 594 | gfp_mask | __GFP_ZERO, node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | if (!q) |
| 596 | return NULL; |
| 597 | |
Dan Carpenter | 00380a4 | 2012-03-23 09:58:54 +0100 | [diff] [blame] | 598 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
Tejun Heo | a73f730 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 599 | if (q->id < 0) |
| 600 | goto fail_q; |
| 601 | |
Jens Axboe | 0989a02 | 2009-06-12 14:42:56 +0200 | [diff] [blame] | 602 | q->backing_dev_info.ra_pages = |
| 603 | (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; |
| 604 | q->backing_dev_info.state = 0; |
| 605 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; |
Jens Axboe | d993831 | 2009-06-12 14:45:52 +0200 | [diff] [blame] | 606 | q->backing_dev_info.name = "block"; |
Mike Snitzer | 5151412 | 2011-11-23 10:59:13 +0100 | [diff] [blame] | 607 | q->node = node_id; |
Jens Axboe | 0989a02 | 2009-06-12 14:42:56 +0200 | [diff] [blame] | 608 | |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 609 | err = bdi_init(&q->backing_dev_info); |
Tejun Heo | a73f730 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 610 | if (err) |
| 611 | goto fail_id; |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 612 | |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 613 | setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, |
| 614 | laptop_mode_timer_fn, (unsigned long) q); |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 615 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
Tejun Heo | b855b04 | 2012-03-06 21:24:55 +0100 | [diff] [blame] | 616 | INIT_LIST_HEAD(&q->queue_head); |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 617 | INIT_LIST_HEAD(&q->timeout_list); |
Tejun Heo | a612fdd | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 618 | INIT_LIST_HEAD(&q->icq_list); |
Tejun Heo | 4eef304 | 2012-03-05 13:15:18 -0800 | [diff] [blame] | 619 | #ifdef CONFIG_BLK_CGROUP |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 620 | INIT_LIST_HEAD(&q->blkg_list); |
Tejun Heo | 4eef304 | 2012-03-05 13:15:18 -0800 | [diff] [blame] | 621 | #endif |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 622 | INIT_LIST_HEAD(&q->flush_queue[0]); |
| 623 | INIT_LIST_HEAD(&q->flush_queue[1]); |
| 624 | INIT_LIST_HEAD(&q->flush_data_in_flight); |
Jens Axboe | 3cca6dc | 2011-03-02 11:08:00 -0500 | [diff] [blame] | 625 | INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 626 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 627 | kobject_init(&q->kobj, &blk_queue_ktype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 629 | mutex_init(&q->sysfs_lock); |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 630 | spin_lock_init(&q->__queue_lock); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 631 | |
Vivek Goyal | c94a96a | 2011-03-02 19:04:42 -0500 | [diff] [blame] | 632 | /* |
| 633 | * By default initialize queue_lock to internal lock and driver can |
| 634 | * override it later if need be. |
| 635 | */ |
| 636 | q->queue_lock = &q->__queue_lock; |
| 637 | |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 638 | /* |
| 639 | * A queue starts its life with bypass turned on to avoid |
| 640 | * unnecessary bypass on/off overhead and nasty surprises during |
Tejun Heo | 749fefe | 2012-09-20 14:08:52 -0700 | [diff] [blame] | 641 | * init. The initial bypass will be finished when the queue is |
| 642 | * registered by blk_register_queue(). |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 643 | */ |
| 644 | q->bypass_depth = 1; |
| 645 | __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); |
| 646 | |
Tejun Heo | 5efd611 | 2012-03-05 13:15:12 -0800 | [diff] [blame] | 647 | if (blkcg_init_queue(q)) |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 648 | goto fail_id; |
| 649 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | return q; |
Tejun Heo | a73f730 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 651 | |
| 652 | fail_id: |
| 653 | ida_simple_remove(&blk_queue_ida, q->id); |
| 654 | fail_q: |
| 655 | kmem_cache_free(blk_requestq_cachep, q); |
| 656 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | } |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 658 | EXPORT_SYMBOL(blk_alloc_queue_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 659 | |
| 660 | /** |
| 661 | * blk_init_queue - prepare a request queue for use with a block device |
| 662 | * @rfn: The function to be called to process requests that have been |
| 663 | * placed on the queue. |
| 664 | * @lock: Request queue spin lock |
| 665 | * |
| 666 | * Description: |
| 667 | * If a block device wishes to use the standard request handling procedures, |
| 668 | * which sorts requests and coalesces adjacent requests, then it must |
| 669 | * call blk_init_queue(). The function @rfn will be called when there |
| 670 | * are requests on the queue that need to be processed. If the device |
| 671 | * supports plugging, then @rfn may not be called immediately when requests |
| 672 | * are available on the queue, but may be called at some time later instead. |
| 673 | * Plugged queues are generally unplugged when a buffer belonging to one |
| 674 | * of the requests on the queue is needed, or due to memory pressure. |
| 675 | * |
| 676 | * @rfn is not required, or even expected, to remove all requests off the |
| 677 | * queue, but only as many as it can handle at a time. If it does leave |
| 678 | * requests on the queue, it is responsible for arranging that the requests |
| 679 | * get dealt with eventually. |
| 680 | * |
| 681 | * The queue spin lock must be held while manipulating the requests on the |
Paolo 'Blaisorblade' Giarrusso | a038e25 | 2006-06-05 12:09:01 +0200 | [diff] [blame] | 682 | * request queue; this lock will be taken also from interrupt context, so irq |
| 683 | * disabling is needed for it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | * |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 685 | * Function returns a pointer to the initialized request queue, or %NULL if |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | * it didn't succeed. |
| 687 | * |
| 688 | * Note: |
| 689 | * blk_init_queue() must be paired with a blk_cleanup_queue() call |
| 690 | * when the block device is deactivated (such as at module unload). |
| 691 | **/ |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 692 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 693 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | { |
Ezequiel Garcia | c304a51 | 2012-11-10 10:39:44 +0100 | [diff] [blame] | 695 | return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 696 | } |
| 697 | EXPORT_SYMBOL(blk_init_queue); |
| 698 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 699 | struct request_queue * |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 700 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
| 701 | { |
Mike Snitzer | c86d1b8 | 2010-06-03 11:34:52 -0600 | [diff] [blame] | 702 | struct request_queue *uninit_q, *q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | |
Mike Snitzer | c86d1b8 | 2010-06-03 11:34:52 -0600 | [diff] [blame] | 704 | uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
| 705 | if (!uninit_q) |
| 706 | return NULL; |
| 707 | |
Mike Snitzer | 5151412 | 2011-11-23 10:59:13 +0100 | [diff] [blame] | 708 | q = blk_init_allocated_queue(uninit_q, rfn, lock); |
Mike Snitzer | c86d1b8 | 2010-06-03 11:34:52 -0600 | [diff] [blame] | 709 | if (!q) |
| 710 | blk_cleanup_queue(uninit_q); |
| 711 | |
| 712 | return q; |
Mike Snitzer | 01effb0 | 2010-05-11 08:57:42 +0200 | [diff] [blame] | 713 | } |
| 714 | EXPORT_SYMBOL(blk_init_queue_node); |
| 715 | |
| 716 | struct request_queue * |
| 717 | blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, |
| 718 | spinlock_t *lock) |
| 719 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 720 | if (!q) |
| 721 | return NULL; |
| 722 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 723 | if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) |
Al Viro | 8669aaf | 2006-03-18 13:50:00 -0500 | [diff] [blame] | 724 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 725 | |
| 726 | q->request_fn = rfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | q->prep_rq_fn = NULL; |
James Bottomley | 28018c2 | 2010-07-01 19:49:17 +0900 | [diff] [blame] | 728 | q->unprep_rq_fn = NULL; |
Tejun Heo | 60ea822 | 2012-09-20 14:09:30 -0700 | [diff] [blame] | 729 | q->queue_flags |= QUEUE_FLAG_DEFAULT; |
Vivek Goyal | c94a96a | 2011-03-02 19:04:42 -0500 | [diff] [blame] | 730 | |
| 731 | /* Override internal queue lock with supplied lock pointer */ |
| 732 | if (lock) |
| 733 | q->queue_lock = lock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | |
Jens Axboe | f3b144a | 2009-03-06 08:48:33 +0100 | [diff] [blame] | 735 | /* |
| 736 | * This also sets hw/phys segments, boundary and size |
| 737 | */ |
Jens Axboe | c20e8de | 2011-09-12 12:03:37 +0200 | [diff] [blame] | 738 | blk_queue_make_request(q, blk_queue_bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | |
Alan Stern | 44ec954 | 2007-02-20 11:01:57 -0500 | [diff] [blame] | 740 | q->sg_reserved_size = INT_MAX; |
| 741 | |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 742 | /* init elevator */ |
| 743 | if (elevator_init(q, NULL)) |
| 744 | return NULL; |
Tejun Heo | b82d4b1 | 2012-04-13 13:11:31 -0700 | [diff] [blame] | 745 | return q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | } |
Mike Snitzer | 5151412 | 2011-11-23 10:59:13 +0100 | [diff] [blame] | 747 | EXPORT_SYMBOL(blk_init_allocated_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | |
Tejun Heo | 09ac46c | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 749 | bool blk_get_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | { |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 751 | if (likely(!blk_queue_dying(q))) { |
Tejun Heo | 09ac46c | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 752 | __blk_get_queue(q); |
| 753 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | } |
| 755 | |
Tejun Heo | 09ac46c | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 756 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | } |
Jens Axboe | d86e0e8 | 2011-05-27 07:44:43 +0200 | [diff] [blame] | 758 | EXPORT_SYMBOL(blk_get_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 760 | static inline void blk_free_request(struct request_list *rl, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | { |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 762 | if (rq->cmd_flags & REQ_ELVPRIV) { |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 763 | elv_put_request(rl->q, rq); |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 764 | if (rq->elv.icq) |
Tejun Heo | 11a3122 | 2012-02-07 07:51:30 +0100 | [diff] [blame] | 765 | put_io_context(rq->elv.icq->ioc); |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 766 | } |
| 767 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 768 | mempool_free(rq, rl->rq_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 769 | } |
| 770 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | /* |
| 772 | * ioc_batching returns true if the ioc is a valid batching request and |
| 773 | * should be given priority access to a request. |
| 774 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 775 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | { |
| 777 | if (!ioc) |
| 778 | return 0; |
| 779 | |
| 780 | /* |
| 781 | * Make sure the process is able to allocate at least 1 request |
| 782 | * even if the batch times out, otherwise we could theoretically |
| 783 | * lose wakeups. |
| 784 | */ |
| 785 | return ioc->nr_batch_requests == q->nr_batching || |
| 786 | (ioc->nr_batch_requests > 0 |
| 787 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); |
| 788 | } |
| 789 | |
| 790 | /* |
| 791 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This |
| 792 | * will cause the process to be a "batcher" on all queues in the system. This |
| 793 | * is the behaviour we want though - once it gets a wakeup it should be given |
| 794 | * a nice run. |
| 795 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 796 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | { |
| 798 | if (!ioc || ioc_batching(q, ioc)) |
| 799 | return; |
| 800 | |
| 801 | ioc->nr_batch_requests = q->nr_batching; |
| 802 | ioc->last_waited = jiffies; |
| 803 | } |
| 804 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 805 | static void __freed_request(struct request_list *rl, int sync) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | { |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 807 | struct request_queue *q = rl->q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 809 | /* |
| 810 | * bdi isn't aware of blkcg yet. As all async IOs end up root |
| 811 | * blkcg anyway, just use root blkcg state. |
| 812 | */ |
| 813 | if (rl == &q->root_rl && |
| 814 | rl->count[sync] < queue_congestion_off_threshold(q)) |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 815 | blk_clear_queue_congested(q, sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 816 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 817 | if (rl->count[sync] + 1 <= q->nr_requests) { |
| 818 | if (waitqueue_active(&rl->wait[sync])) |
| 819 | wake_up(&rl->wait[sync]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 820 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 821 | blk_clear_rl_full(rl, sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 822 | } |
| 823 | } |
| 824 | |
| 825 | /* |
| 826 | * A request has just been released. Account for it, update the full and |
| 827 | * congestion status, wake up any waiters. Called under q->queue_lock. |
| 828 | */ |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 829 | static void freed_request(struct request_list *rl, unsigned int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | { |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 831 | struct request_queue *q = rl->q; |
Tejun Heo | 75eb6c3 | 2011-10-19 14:31:22 +0200 | [diff] [blame] | 832 | int sync = rw_is_sync(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 833 | |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 834 | q->nr_rqs[sync]--; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 835 | rl->count[sync]--; |
Tejun Heo | 75eb6c3 | 2011-10-19 14:31:22 +0200 | [diff] [blame] | 836 | if (flags & REQ_ELVPRIV) |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 837 | q->nr_rqs_elvpriv--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 839 | __freed_request(rl, sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 841 | if (unlikely(rl->starved[sync ^ 1])) |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 842 | __freed_request(rl, sync ^ 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 843 | } |
| 844 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 845 | /* |
Mike Snitzer | 9d5a4e9 | 2011-02-11 11:05:46 +0100 | [diff] [blame] | 846 | * Determine if elevator data should be initialized when allocating the |
| 847 | * request associated with @bio. |
| 848 | */ |
| 849 | static bool blk_rq_should_init_elevator(struct bio *bio) |
| 850 | { |
| 851 | if (!bio) |
| 852 | return true; |
| 853 | |
| 854 | /* |
| 855 | * Flush requests do not use the elevator so skip initialization. |
| 856 | * This allows a request to share the flush and elevator data. |
| 857 | */ |
| 858 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) |
| 859 | return false; |
| 860 | |
| 861 | return true; |
| 862 | } |
| 863 | |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 864 | /** |
Tejun Heo | 852c788 | 2012-03-05 13:15:27 -0800 | [diff] [blame] | 865 | * rq_ioc - determine io_context for request allocation |
| 866 | * @bio: request being allocated is for this bio (can be %NULL) |
| 867 | * |
| 868 | * Determine io_context to use for request allocation for @bio. May return |
| 869 | * %NULL if %current->io_context doesn't exist. |
| 870 | */ |
| 871 | static struct io_context *rq_ioc(struct bio *bio) |
| 872 | { |
| 873 | #ifdef CONFIG_BLK_CGROUP |
| 874 | if (bio && bio->bi_ioc) |
| 875 | return bio->bi_ioc; |
| 876 | #endif |
| 877 | return current->io_context; |
| 878 | } |
| 879 | |
| 880 | /** |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 881 | * __get_request - get a free request |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 882 | * @rl: request list to allocate from |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 883 | * @rw_flags: RW and SYNC flags |
| 884 | * @bio: bio to allocate request for (can be %NULL) |
| 885 | * @gfp_mask: allocation mask |
| 886 | * |
| 887 | * Get a free request from @q. This function may fail under memory |
| 888 | * pressure or if @q is dead. |
| 889 | * |
| 890 | * Must be callled with @q->queue_lock held and, |
| 891 | * Returns %NULL on failure, with @q->queue_lock held. |
| 892 | * Returns !%NULL on success, with @q->queue_lock *not held*. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | */ |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 894 | static struct request *__get_request(struct request_list *rl, int rw_flags, |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 895 | struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | { |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 897 | struct request_queue *q = rl->q; |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 898 | struct request *rq; |
Tejun Heo | 7f4b35d | 2012-06-04 20:40:56 -0700 | [diff] [blame] | 899 | struct elevator_type *et = q->elevator->type; |
| 900 | struct io_context *ioc = rq_ioc(bio); |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 901 | struct io_cq *icq = NULL; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 902 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
Tejun Heo | 75eb6c3 | 2011-10-19 14:31:22 +0200 | [diff] [blame] | 903 | int may_queue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 905 | if (unlikely(blk_queue_dying(q))) |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 906 | return NULL; |
| 907 | |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 908 | may_queue = elv_may_queue(q, rw_flags); |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 909 | if (may_queue == ELV_MQUEUE_NO) |
| 910 | goto rq_starved; |
| 911 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 912 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
| 913 | if (rl->count[is_sync]+1 >= q->nr_requests) { |
Tejun Heo | f2dbd76 | 2011-12-14 00:33:40 +0100 | [diff] [blame] | 914 | /* |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 915 | * The queue will fill after this allocation, so set |
| 916 | * it as full, and mark this process as "batching". |
| 917 | * This process will be allowed to complete a batch of |
| 918 | * requests, others will be blocked. |
| 919 | */ |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 920 | if (!blk_rl_full(rl, is_sync)) { |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 921 | ioc_set_batching(q, ioc); |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 922 | blk_set_rl_full(rl, is_sync); |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 923 | } else { |
| 924 | if (may_queue != ELV_MQUEUE_MUST |
| 925 | && !ioc_batching(q, ioc)) { |
| 926 | /* |
| 927 | * The queue is full and the allocating |
| 928 | * process is not a "batcher", and not |
| 929 | * exempted by the IO scheduler |
| 930 | */ |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 931 | return NULL; |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 932 | } |
| 933 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | } |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 935 | /* |
| 936 | * bdi isn't aware of blkcg yet. As all async IOs end up |
| 937 | * root blkcg anyway, just use root blkcg state. |
| 938 | */ |
| 939 | if (rl == &q->root_rl) |
| 940 | blk_set_queue_congested(q, is_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 941 | } |
| 942 | |
Jens Axboe | 082cf69 | 2005-06-28 16:35:11 +0200 | [diff] [blame] | 943 | /* |
| 944 | * Only allow batching queuers to allocate up to 50% over the defined |
| 945 | * limit of requests, otherwise we could have thousands of requests |
| 946 | * allocated with any setting of ->nr_requests |
| 947 | */ |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 948 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 949 | return NULL; |
Hugh Dickins | fd782a4 | 2005-06-29 15:15:40 +0100 | [diff] [blame] | 950 | |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 951 | q->nr_rqs[is_sync]++; |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 952 | rl->count[is_sync]++; |
| 953 | rl->starved[is_sync] = 0; |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 954 | |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 955 | /* |
| 956 | * Decide whether the new request will be managed by elevator. If |
| 957 | * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will |
| 958 | * prevent the current elevator from being destroyed until the new |
| 959 | * request is freed. This guarantees icq's won't be destroyed and |
| 960 | * makes creating new ones safe. |
| 961 | * |
| 962 | * Also, lookup icq while holding queue_lock. If it doesn't exist, |
| 963 | * it will be created after releasing queue_lock. |
| 964 | */ |
Tejun Heo | d732580 | 2012-03-05 13:14:58 -0800 | [diff] [blame] | 965 | if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { |
Tejun Heo | 75eb6c3 | 2011-10-19 14:31:22 +0200 | [diff] [blame] | 966 | rw_flags |= REQ_ELVPRIV; |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 967 | q->nr_rqs_elvpriv++; |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 968 | if (et->icq_cache && ioc) |
| 969 | icq = ioc_lookup_icq(ioc, q); |
Mike Snitzer | 9d5a4e9 | 2011-02-11 11:05:46 +0100 | [diff] [blame] | 970 | } |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 971 | |
Jens Axboe | f253b86 | 2010-10-24 22:06:02 +0200 | [diff] [blame] | 972 | if (blk_queue_io_stat(q)) |
| 973 | rw_flags |= REQ_IO_STAT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 974 | spin_unlock_irq(q->queue_lock); |
| 975 | |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 976 | /* allocate and init request */ |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 977 | rq = mempool_alloc(rl->rq_pool, gfp_mask); |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 978 | if (!rq) |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 979 | goto fail_alloc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 981 | blk_rq_init(q, rq); |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 982 | blk_rq_set_rl(rq, rl); |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 983 | rq->cmd_flags = rw_flags | REQ_ALLOCED; |
| 984 | |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 985 | /* init elvpriv */ |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 986 | if (rw_flags & REQ_ELVPRIV) { |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 987 | if (unlikely(et->icq_cache && !icq)) { |
Tejun Heo | 7f4b35d | 2012-06-04 20:40:56 -0700 | [diff] [blame] | 988 | if (ioc) |
| 989 | icq = ioc_create_icq(ioc, q, gfp_mask); |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 990 | if (!icq) |
| 991 | goto fail_elvpriv; |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 992 | } |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 993 | |
| 994 | rq->elv.icq = icq; |
| 995 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) |
| 996 | goto fail_elvpriv; |
| 997 | |
| 998 | /* @rq->elv.icq holds io_context until @rq is freed */ |
Tejun Heo | 29e2b09 | 2012-04-19 16:29:21 -0700 | [diff] [blame] | 999 | if (icq) |
| 1000 | get_io_context(icq->ioc); |
| 1001 | } |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 1002 | out: |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 1003 | /* |
| 1004 | * ioc may be NULL here, and ioc_batching will be false. That's |
| 1005 | * OK, if the queue is under the request limit then requests need |
| 1006 | * not count toward the nr_batch_requests limit. There will always |
| 1007 | * be some limit enforced by BLK_BATCH_TIME. |
| 1008 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | if (ioc_batching(q, ioc)) |
| 1010 | ioc->nr_batch_requests--; |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1011 | |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 1012 | trace_block_getrq(q, bio, rw_flags & 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1013 | return rq; |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 1014 | |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 1015 | fail_elvpriv: |
| 1016 | /* |
| 1017 | * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed |
| 1018 | * and may fail indefinitely under memory pressure and thus |
| 1019 | * shouldn't stall IO. Treat this request as !elvpriv. This will |
| 1020 | * disturb iosched and blkcg but weird is bettern than dead. |
| 1021 | */ |
| 1022 | printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n", |
| 1023 | dev_name(q->backing_dev_info.dev)); |
| 1024 | |
| 1025 | rq->cmd_flags &= ~REQ_ELVPRIV; |
| 1026 | rq->elv.icq = NULL; |
| 1027 | |
| 1028 | spin_lock_irq(q->queue_lock); |
Tejun Heo | 8a5ecdd | 2012-06-04 20:40:58 -0700 | [diff] [blame] | 1029 | q->nr_rqs_elvpriv--; |
Tejun Heo | aaf7c68 | 2012-04-19 16:29:22 -0700 | [diff] [blame] | 1030 | spin_unlock_irq(q->queue_lock); |
| 1031 | goto out; |
| 1032 | |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 1033 | fail_alloc: |
| 1034 | /* |
| 1035 | * Allocation failed presumably due to memory. Undo anything we |
| 1036 | * might have messed up. |
| 1037 | * |
| 1038 | * Allocating task should really be put onto the front of the wait |
| 1039 | * queue, but this is pretty rare. |
| 1040 | */ |
| 1041 | spin_lock_irq(q->queue_lock); |
Tejun Heo | 5b788ce | 2012-06-04 20:40:59 -0700 | [diff] [blame] | 1042 | freed_request(rl, rw_flags); |
Tejun Heo | b679281 | 2012-03-05 13:15:23 -0800 | [diff] [blame] | 1043 | |
| 1044 | /* |
| 1045 | * in the very unlikely event that allocation failed and no |
| 1046 | * requests for this direction was pending, mark us starved so that |
| 1047 | * freeing of a request in the other direction will notice |
| 1048 | * us. another possible fix would be to split the rq mempool into |
| 1049 | * READ and WRITE |
| 1050 | */ |
| 1051 | rq_starved: |
| 1052 | if (unlikely(rl->count[is_sync] == 0)) |
| 1053 | rl->starved[is_sync] = 1; |
| 1054 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | } |
| 1056 | |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1057 | /** |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1058 | * get_request - get a free request |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1059 | * @q: request_queue to allocate request from |
| 1060 | * @rw_flags: RW and SYNC flags |
| 1061 | * @bio: bio to allocate request for (can be %NULL) |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1062 | * @gfp_mask: allocation mask |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1063 | * |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1064 | * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this |
| 1065 | * function keeps retrying under memory pressure and fails iff @q is dead. |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1066 | * |
| 1067 | * Must be callled with @q->queue_lock held and, |
| 1068 | * Returns %NULL on failure, with @q->queue_lock held. |
| 1069 | * Returns !%NULL on success, with @q->queue_lock *not held*. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | */ |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1071 | static struct request *get_request(struct request_queue *q, int rw_flags, |
| 1072 | struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1073 | { |
Jens Axboe | 1faa16d | 2009-04-06 14:48:01 +0200 | [diff] [blame] | 1074 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1075 | DEFINE_WAIT(wait); |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1076 | struct request_list *rl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | struct request *rq; |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1078 | |
| 1079 | rl = blk_get_rl(q, bio); /* transferred to @rq on success */ |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1080 | retry: |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1081 | rq = __get_request(rl, rw_flags, bio, gfp_mask); |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1082 | if (rq) |
| 1083 | return rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1084 | |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 1085 | if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) { |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1086 | blk_put_rl(rl); |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1087 | return NULL; |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1088 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1089 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1090 | /* wait on @rl and retry */ |
| 1091 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
| 1092 | TASK_UNINTERRUPTIBLE); |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1093 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1094 | trace_block_sleeprq(q, bio, rw_flags & 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1095 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1096 | spin_unlock_irq(q->queue_lock); |
| 1097 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1098 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1099 | /* |
| 1100 | * After sleeping, we become a "batching" process and will be able |
| 1101 | * to allocate at least one request, and up to a big batch of them |
| 1102 | * for a small period time. See ioc_batching, ioc_set_batching |
| 1103 | */ |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1104 | ioc_set_batching(q, current->io_context); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1106 | spin_lock_irq(q->queue_lock); |
| 1107 | finish_wait(&rl->wait[is_sync], &wait); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1108 | |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1109 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1110 | } |
| 1111 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1112 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | { |
| 1114 | struct request *rq; |
| 1115 | |
| 1116 | BUG_ON(rw != READ && rw != WRITE); |
| 1117 | |
Tejun Heo | 7f4b35d | 2012-06-04 20:40:56 -0700 | [diff] [blame] | 1118 | /* create ioc upfront */ |
| 1119 | create_io_context(gfp_mask, q->node); |
| 1120 | |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1121 | spin_lock_irq(q->queue_lock); |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1122 | rq = get_request(q, rw, NULL, gfp_mask); |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1123 | if (!rq) |
| 1124 | spin_unlock_irq(q->queue_lock); |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1125 | /* q->queue_lock is unlocked at this point */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | |
| 1127 | return rq; |
| 1128 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | EXPORT_SYMBOL(blk_get_request); |
| 1130 | |
| 1131 | /** |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1132 | * blk_make_request - given a bio, allocate a corresponding struct request. |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 1133 | * @q: target request queue |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1134 | * @bio: The bio describing the memory mappings that will be submitted for IO. |
| 1135 | * It may be a chained-bio properly constructed by block/bio layer. |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 1136 | * @gfp_mask: gfp flags to be used for memory allocation |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 1137 | * |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1138 | * blk_make_request is the parallel of generic_make_request for BLOCK_PC |
| 1139 | * type commands. Where the struct request needs to be farther initialized by |
| 1140 | * the caller. It is passed a &struct bio, which describes the memory info of |
| 1141 | * the I/O transfer. |
| 1142 | * |
| 1143 | * The caller of blk_make_request must make sure that bi_io_vec |
| 1144 | * are set to describe the memory buffers. That bio_data_dir() will return |
| 1145 | * the needed direction of the request. (And all bio's in the passed bio-chain |
| 1146 | * are properly set accordingly) |
| 1147 | * |
| 1148 | * If called under none-sleepable conditions, mapped bio buffers must not |
| 1149 | * need bouncing, by calling the appropriate masked or flagged allocator, |
| 1150 | * suitable for the target device. Otherwise the call to blk_queue_bounce will |
| 1151 | * BUG. |
Jens Axboe | 53674ac | 2009-05-19 19:52:35 +0200 | [diff] [blame] | 1152 | * |
| 1153 | * WARNING: When allocating/cloning a bio-chain, careful consideration should be |
| 1154 | * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for |
| 1155 | * anything but the first bio in the chain. Otherwise you risk waiting for IO |
| 1156 | * completion of a bio that hasn't been submitted yet, thus resulting in a |
| 1157 | * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead |
| 1158 | * of bio_alloc(), as that avoids the mempool deadlock. |
| 1159 | * If possible a big IO should be split into smaller parts when allocation |
| 1160 | * fails. Partial allocation should not be an error, or you risk a live-lock. |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 1161 | */ |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1162 | struct request *blk_make_request(struct request_queue *q, struct bio *bio, |
| 1163 | gfp_t gfp_mask) |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 1164 | { |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1165 | struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); |
| 1166 | |
| 1167 | if (unlikely(!rq)) |
| 1168 | return ERR_PTR(-ENOMEM); |
| 1169 | |
| 1170 | for_each_bio(bio) { |
| 1171 | struct bio *bounce_bio = bio; |
| 1172 | int ret; |
| 1173 | |
| 1174 | blk_queue_bounce(q, &bounce_bio); |
| 1175 | ret = blk_rq_append_bio(q, rq, bounce_bio); |
| 1176 | if (unlikely(ret)) { |
| 1177 | blk_put_request(rq); |
| 1178 | return ERR_PTR(ret); |
| 1179 | } |
| 1180 | } |
| 1181 | |
| 1182 | return rq; |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 1183 | } |
Boaz Harrosh | 79eb63e | 2009-05-17 18:57:15 +0300 | [diff] [blame] | 1184 | EXPORT_SYMBOL(blk_make_request); |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 1185 | |
| 1186 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1187 | * blk_requeue_request - put a request back on queue |
| 1188 | * @q: request queue where request should be inserted |
| 1189 | * @rq: request to be inserted |
| 1190 | * |
| 1191 | * Description: |
| 1192 | * Drivers often keep queueing requests until the hardware cannot accept |
| 1193 | * more, when that condition happens we need to put the request back |
| 1194 | * on the queue. Must be called with queue lock held. |
| 1195 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1196 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | { |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 1198 | blk_delete_timer(rq); |
| 1199 | blk_clear_rq_complete(rq); |
Arnaldo Carvalho de Melo | 5f3ea37 | 2008-10-30 08:34:33 +0100 | [diff] [blame] | 1200 | trace_block_rq_requeue(q, rq); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1201 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1202 | if (blk_rq_tagged(rq)) |
| 1203 | blk_queue_end_tag(q, rq); |
| 1204 | |
James Bottomley | ba396a6 | 2009-05-27 14:17:08 +0200 | [diff] [blame] | 1205 | BUG_ON(blk_queued_rq(rq)); |
| 1206 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | elv_requeue_request(q, rq); |
| 1208 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1209 | EXPORT_SYMBOL(blk_requeue_request); |
| 1210 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1211 | static void add_acct_request(struct request_queue *q, struct request *rq, |
| 1212 | int where) |
| 1213 | { |
| 1214 | drive_stat_acct(rq, 1); |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 1215 | __elv_add_request(q, rq, where); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1216 | } |
| 1217 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1218 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
| 1219 | unsigned long now) |
| 1220 | { |
| 1221 | if (now == part->stamp) |
| 1222 | return; |
| 1223 | |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 1224 | if (part_in_flight(part)) { |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1225 | __part_stat_add(cpu, part, time_in_queue, |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 1226 | part_in_flight(part) * (now - part->stamp)); |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1227 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
| 1228 | } |
| 1229 | part->stamp = now; |
| 1230 | } |
| 1231 | |
| 1232 | /** |
Randy Dunlap | 496aa8a | 2008-10-16 07:46:23 +0200 | [diff] [blame] | 1233 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
| 1234 | * @cpu: cpu number for stats access |
| 1235 | * @part: target partition |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | * |
| 1237 | * The average IO queue length and utilisation statistics are maintained |
| 1238 | * by observing the current state of the queue length and the amount of |
| 1239 | * time it has been in this state for. |
| 1240 | * |
| 1241 | * Normally, that accounting is done on IO completion, but that can result |
| 1242 | * in more than a second's worth of IO being accounted for within any one |
| 1243 | * second, leading to >100% utilisation. To deal with that, we call this |
| 1244 | * function to do a round-off before returning the results when reading |
| 1245 | * /proc/diskstats. This accounts immediately for all queue usage up to |
| 1246 | * the current jiffies and restarts the counters again. |
| 1247 | */ |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 1248 | void part_round_stats(int cpu, struct hd_struct *part) |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1249 | { |
| 1250 | unsigned long now = jiffies; |
| 1251 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1252 | if (part->partno) |
| 1253 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); |
| 1254 | part_round_stats_single(cpu, part, now); |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1255 | } |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1256 | EXPORT_SYMBOL_GPL(part_round_stats); |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1257 | |
Lin Ming | c815881 | 2013-03-23 11:42:27 +0800 | [diff] [blame] | 1258 | #ifdef CONFIG_PM_RUNTIME |
| 1259 | static void blk_pm_put_request(struct request *rq) |
| 1260 | { |
| 1261 | if (rq->q->dev && !(rq->cmd_flags & REQ_PM) && !--rq->q->nr_pending) |
| 1262 | pm_runtime_mark_last_busy(rq->q->dev); |
| 1263 | } |
| 1264 | #else |
| 1265 | static inline void blk_pm_put_request(struct request *rq) {} |
| 1266 | #endif |
| 1267 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | /* |
| 1269 | * queue lock must be held |
| 1270 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1271 | void __blk_put_request(struct request_queue *q, struct request *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1272 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | if (unlikely(!q)) |
| 1274 | return; |
| 1275 | if (unlikely(--req->ref_count)) |
| 1276 | return; |
| 1277 | |
Lin Ming | c815881 | 2013-03-23 11:42:27 +0800 | [diff] [blame] | 1278 | blk_pm_put_request(req); |
| 1279 | |
Tejun Heo | 8922e16 | 2005-10-20 16:23:44 +0200 | [diff] [blame] | 1280 | elv_completed_request(q, req); |
| 1281 | |
Boaz Harrosh | 1cd96c2 | 2009-03-24 12:35:07 +0100 | [diff] [blame] | 1282 | /* this is a bio leak */ |
| 1283 | WARN_ON(req->bio != NULL); |
| 1284 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | /* |
| 1286 | * Request may not have originated from ll_rw_blk. if not, |
| 1287 | * it didn't come out of our reserved rq pools |
| 1288 | */ |
Jens Axboe | 49171e5 | 2006-08-10 08:59:11 +0200 | [diff] [blame] | 1289 | if (req->cmd_flags & REQ_ALLOCED) { |
Tejun Heo | 75eb6c3 | 2011-10-19 14:31:22 +0200 | [diff] [blame] | 1290 | unsigned int flags = req->cmd_flags; |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1291 | struct request_list *rl = blk_rq_rl(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | BUG_ON(!list_empty(&req->queuelist)); |
Jens Axboe | 9817064 | 2006-07-28 09:23:08 +0200 | [diff] [blame] | 1294 | BUG_ON(!hlist_unhashed(&req->hash)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1295 | |
Tejun Heo | a051661 | 2012-06-26 15:05:44 -0700 | [diff] [blame] | 1296 | blk_free_request(rl, req); |
| 1297 | freed_request(rl, flags); |
| 1298 | blk_put_rl(rl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | } |
| 1300 | } |
Mike Christie | 6e39b69 | 2005-11-11 05:30:24 -0600 | [diff] [blame] | 1301 | EXPORT_SYMBOL_GPL(__blk_put_request); |
| 1302 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | void blk_put_request(struct request *req) |
| 1304 | { |
Tejun Heo | 8922e16 | 2005-10-20 16:23:44 +0200 | [diff] [blame] | 1305 | unsigned long flags; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1306 | struct request_queue *q = req->q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1307 | |
FUJITA Tomonori | 52a93ba | 2008-07-15 21:21:45 +0200 | [diff] [blame] | 1308 | spin_lock_irqsave(q->queue_lock, flags); |
| 1309 | __blk_put_request(q, req); |
| 1310 | spin_unlock_irqrestore(q->queue_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1311 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1312 | EXPORT_SYMBOL(blk_put_request); |
| 1313 | |
Christoph Hellwig | 66ac028 | 2010-06-18 16:59:42 +0200 | [diff] [blame] | 1314 | /** |
| 1315 | * blk_add_request_payload - add a payload to a request |
| 1316 | * @rq: request to update |
| 1317 | * @page: page backing the payload |
| 1318 | * @len: length of the payload. |
| 1319 | * |
| 1320 | * This allows to later add a payload to an already submitted request by |
| 1321 | * a block driver. The driver needs to take care of freeing the payload |
| 1322 | * itself. |
| 1323 | * |
| 1324 | * Note that this is a quite horrible hack and nothing but handling of |
| 1325 | * discard requests should ever use it. |
| 1326 | */ |
| 1327 | void blk_add_request_payload(struct request *rq, struct page *page, |
| 1328 | unsigned int len) |
| 1329 | { |
| 1330 | struct bio *bio = rq->bio; |
| 1331 | |
| 1332 | bio->bi_io_vec->bv_page = page; |
| 1333 | bio->bi_io_vec->bv_offset = 0; |
| 1334 | bio->bi_io_vec->bv_len = len; |
| 1335 | |
| 1336 | bio->bi_size = len; |
| 1337 | bio->bi_vcnt = 1; |
| 1338 | bio->bi_phys_segments = 1; |
| 1339 | |
| 1340 | rq->__data_len = rq->resid_len = len; |
| 1341 | rq->nr_phys_segments = 1; |
| 1342 | rq->buffer = bio_data(bio); |
| 1343 | } |
| 1344 | EXPORT_SYMBOL_GPL(blk_add_request_payload); |
| 1345 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1346 | static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, |
| 1347 | struct bio *bio) |
| 1348 | { |
| 1349 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; |
| 1350 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1351 | if (!ll_back_merge_fn(q, req, bio)) |
| 1352 | return false; |
| 1353 | |
Tejun Heo | 8c1cf6b | 2013-01-11 13:06:34 -0800 | [diff] [blame] | 1354 | trace_block_bio_backmerge(q, req, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1355 | |
| 1356 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
| 1357 | blk_rq_set_mixed_merge(req); |
| 1358 | |
| 1359 | req->biotail->bi_next = bio; |
| 1360 | req->biotail = bio; |
| 1361 | req->__data_len += bio->bi_size; |
| 1362 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
| 1363 | |
| 1364 | drive_stat_acct(req, 0); |
| 1365 | return true; |
| 1366 | } |
| 1367 | |
| 1368 | static bool bio_attempt_front_merge(struct request_queue *q, |
| 1369 | struct request *req, struct bio *bio) |
| 1370 | { |
| 1371 | const int ff = bio->bi_rw & REQ_FAILFAST_MASK; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1372 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1373 | if (!ll_front_merge_fn(q, req, bio)) |
| 1374 | return false; |
| 1375 | |
Tejun Heo | 8c1cf6b | 2013-01-11 13:06:34 -0800 | [diff] [blame] | 1376 | trace_block_bio_frontmerge(q, req, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1377 | |
| 1378 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
| 1379 | blk_rq_set_mixed_merge(req); |
| 1380 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1381 | bio->bi_next = req->bio; |
| 1382 | req->bio = bio; |
| 1383 | |
| 1384 | /* |
| 1385 | * may not be valid. if the low level driver said |
| 1386 | * it didn't need a bounce buffer then it better |
| 1387 | * not touch req->buffer either... |
| 1388 | */ |
| 1389 | req->buffer = bio_data(bio); |
| 1390 | req->__sector = bio->bi_sector; |
| 1391 | req->__data_len += bio->bi_size; |
| 1392 | req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); |
| 1393 | |
| 1394 | drive_stat_acct(req, 0); |
| 1395 | return true; |
| 1396 | } |
| 1397 | |
Tejun Heo | bd87b58 | 2011-10-19 14:33:08 +0200 | [diff] [blame] | 1398 | /** |
| 1399 | * attempt_plug_merge - try to merge with %current's plugged list |
| 1400 | * @q: request_queue new bio is being queued at |
| 1401 | * @bio: new bio being queued |
| 1402 | * @request_count: out parameter for number of traversed plugged requests |
| 1403 | * |
| 1404 | * Determine whether @bio being queued on @q can be merged with a request |
| 1405 | * on %current's plugged list. Returns %true if merge was successful, |
| 1406 | * otherwise %false. |
| 1407 | * |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 1408 | * Plugging coalesces IOs from the same issuer for the same purpose without |
| 1409 | * going through @q->queue_lock. As such it's more of an issuing mechanism |
| 1410 | * than scheduling, and the request, while may have elvpriv data, is not |
| 1411 | * added on the elevator at this point. In addition, we don't have |
| 1412 | * reliable access to the elevator outside queue lock. Only check basic |
| 1413 | * merging parameters without querying the elevator. |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1414 | */ |
Tejun Heo | bd87b58 | 2011-10-19 14:33:08 +0200 | [diff] [blame] | 1415 | static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, |
| 1416 | unsigned int *request_count) |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1417 | { |
| 1418 | struct blk_plug *plug; |
| 1419 | struct request *rq; |
| 1420 | bool ret = false; |
| 1421 | |
Tejun Heo | bd87b58 | 2011-10-19 14:33:08 +0200 | [diff] [blame] | 1422 | plug = current->plug; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1423 | if (!plug) |
| 1424 | goto out; |
Shaohua Li | 56ebdaf | 2011-08-24 16:04:34 +0200 | [diff] [blame] | 1425 | *request_count = 0; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1426 | |
| 1427 | list_for_each_entry_reverse(rq, &plug->list, queuelist) { |
| 1428 | int el_ret; |
| 1429 | |
Shaohua Li | 1b2e19f | 2012-04-06 11:37:47 -0600 | [diff] [blame] | 1430 | if (rq->q == q) |
| 1431 | (*request_count)++; |
Shaohua Li | 56ebdaf | 2011-08-24 16:04:34 +0200 | [diff] [blame] | 1432 | |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 1433 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1434 | continue; |
| 1435 | |
Tejun Heo | 050c8ea | 2012-02-08 09:19:38 +0100 | [diff] [blame] | 1436 | el_ret = blk_try_merge(rq, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1437 | if (el_ret == ELEVATOR_BACK_MERGE) { |
| 1438 | ret = bio_attempt_back_merge(q, rq, bio); |
| 1439 | if (ret) |
| 1440 | break; |
| 1441 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { |
| 1442 | ret = bio_attempt_front_merge(q, rq, bio); |
| 1443 | if (ret) |
| 1444 | break; |
| 1445 | } |
| 1446 | } |
| 1447 | out: |
| 1448 | return ret; |
| 1449 | } |
| 1450 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 1451 | void init_request_from_bio(struct request *req, struct bio *bio) |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1452 | { |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 1453 | req->cmd_type = REQ_TYPE_FS; |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1454 | |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 1455 | req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; |
| 1456 | if (bio->bi_rw & REQ_RAHEAD) |
Tejun Heo | a82afdf | 2009-07-03 17:48:16 +0900 | [diff] [blame] | 1457 | req->cmd_flags |= REQ_FAILFAST_MASK; |
Jens Axboe | b31dc66 | 2006-06-13 08:26:10 +0200 | [diff] [blame] | 1458 | |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1459 | req->errors = 0; |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 1460 | req->__sector = bio->bi_sector; |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1461 | req->ioprio = bio_prio(bio); |
NeilBrown | bc1c56f | 2007-08-16 13:31:30 +0200 | [diff] [blame] | 1462 | blk_rq_bio_prep(req->q, req, bio); |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1463 | } |
| 1464 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1465 | void blk_queue_bio(struct request_queue *q, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | { |
Jiri Slaby | 5e00d1b | 2010-08-12 14:31:06 +0200 | [diff] [blame] | 1467 | const bool sync = !!(bio->bi_rw & REQ_SYNC); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1468 | struct blk_plug *plug; |
| 1469 | int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; |
| 1470 | struct request *req; |
Shaohua Li | 56ebdaf | 2011-08-24 16:04:34 +0200 | [diff] [blame] | 1471 | unsigned int request_count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1472 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | /* |
| 1474 | * low level driver can indicate that it wants pages above a |
| 1475 | * certain limit bounced to low memory (ie for highmem, or even |
| 1476 | * ISA dma in theory) |
| 1477 | */ |
| 1478 | blk_queue_bounce(q, &bio); |
| 1479 | |
Darrick J. Wong | ffecfd1 | 2013-02-21 16:42:55 -0800 | [diff] [blame] | 1480 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) { |
| 1481 | bio_endio(bio, -EIO); |
| 1482 | return; |
| 1483 | } |
| 1484 | |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 1485 | if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1486 | spin_lock_irq(q->queue_lock); |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 1487 | where = ELEVATOR_INSERT_FLUSH; |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 1488 | goto get_rq; |
| 1489 | } |
| 1490 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1491 | /* |
| 1492 | * Check if we can merge with the plugged list before grabbing |
| 1493 | * any locks. |
| 1494 | */ |
Tejun Heo | bd87b58 | 2011-10-19 14:33:08 +0200 | [diff] [blame] | 1495 | if (attempt_plug_merge(q, bio, &request_count)) |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1496 | return; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1497 | |
| 1498 | spin_lock_irq(q->queue_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1499 | |
| 1500 | el_ret = elv_merge(q, &req, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1501 | if (el_ret == ELEVATOR_BACK_MERGE) { |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1502 | if (bio_attempt_back_merge(q, req, bio)) { |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 1503 | elv_bio_merged(q, req, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1504 | if (!attempt_back_merge(q, req)) |
| 1505 | elv_merged_request(q, req, el_ret); |
| 1506 | goto out_unlock; |
Tejun Heo | 80a761f | 2009-07-03 17:48:17 +0900 | [diff] [blame] | 1507 | } |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1508 | } else if (el_ret == ELEVATOR_FRONT_MERGE) { |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1509 | if (bio_attempt_front_merge(q, req, bio)) { |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 1510 | elv_bio_merged(q, req, bio); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1511 | if (!attempt_front_merge(q, req)) |
| 1512 | elv_merged_request(q, req, el_ret); |
| 1513 | goto out_unlock; |
| 1514 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1515 | } |
| 1516 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | get_rq: |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1518 | /* |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 1519 | * This sync check and mask will be re-done in init_request_from_bio(), |
| 1520 | * but we need to set it earlier to expose the sync flag to the |
| 1521 | * rq allocator and io schedulers. |
| 1522 | */ |
| 1523 | rw_flags = bio_data_dir(bio); |
| 1524 | if (sync) |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 1525 | rw_flags |= REQ_SYNC; |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 1526 | |
| 1527 | /* |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1528 | * Grab a free request. This is might sleep but can not fail. |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1529 | * Returns with the queue unlocked. |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1530 | */ |
Tejun Heo | a06e05e | 2012-06-04 20:40:55 -0700 | [diff] [blame] | 1531 | req = get_request(q, rw_flags, bio, GFP_NOIO); |
Tejun Heo | da8303c | 2011-10-19 14:33:05 +0200 | [diff] [blame] | 1532 | if (unlikely(!req)) { |
| 1533 | bio_endio(bio, -ENODEV); /* @q is dead */ |
| 1534 | goto out_unlock; |
| 1535 | } |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1536 | |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1537 | /* |
| 1538 | * After dropping the lock and possibly sleeping here, our request |
| 1539 | * may now be mergeable after it had proven unmergeable (above). |
| 1540 | * We don't worry about that case for efficiency. It won't happen |
| 1541 | * often, and the elevators are able to handle it. |
| 1542 | */ |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1543 | init_request_from_bio(req, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | |
Tao Ma | 9562ad9 | 2011-10-24 16:11:30 +0200 | [diff] [blame] | 1545 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) |
Jens Axboe | 11ccf11 | 2011-07-26 15:01:15 +0200 | [diff] [blame] | 1546 | req->cpu = raw_smp_processor_id(); |
Tejun Heo | dd83100 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 1547 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1548 | plug = current->plug; |
Jens Axboe | 721a960 | 2011-03-09 11:56:30 +0100 | [diff] [blame] | 1549 | if (plug) { |
Jens Axboe | dc6d36c | 2011-04-12 10:28:28 +0200 | [diff] [blame] | 1550 | /* |
| 1551 | * If this is the first request added after a plug, fire |
| 1552 | * of a plug trace. If others have been added before, check |
| 1553 | * if we have multiple devices in this plug. If so, make a |
| 1554 | * note to sort the list before dispatch. |
| 1555 | */ |
| 1556 | if (list_empty(&plug->list)) |
| 1557 | trace_block_plug(q); |
Shaohua Li | 3540d5e | 2011-11-16 09:21:50 +0100 | [diff] [blame] | 1558 | else { |
Shaohua Li | 019ceb7 | 2011-11-16 09:21:50 +0100 | [diff] [blame] | 1559 | if (request_count >= BLK_MAX_REQUEST_COUNT) { |
Shaohua Li | 3540d5e | 2011-11-16 09:21:50 +0100 | [diff] [blame] | 1560 | blk_flush_plug_list(plug, false); |
Shaohua Li | 019ceb7 | 2011-11-16 09:21:50 +0100 | [diff] [blame] | 1561 | trace_block_plug(q); |
| 1562 | } |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1563 | } |
Shaohua Li | a632716 | 2011-08-24 16:04:32 +0200 | [diff] [blame] | 1564 | list_add_tail(&req->queuelist, &plug->list); |
| 1565 | drive_stat_acct(req, 1); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1566 | } else { |
| 1567 | spin_lock_irq(q->queue_lock); |
| 1568 | add_acct_request(q, req, where); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 1569 | __blk_run_queue(q); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 1570 | out_unlock: |
| 1571 | spin_unlock_irq(q->queue_lock); |
| 1572 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1573 | } |
Jens Axboe | c20e8de | 2011-09-12 12:03:37 +0200 | [diff] [blame] | 1574 | EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | |
| 1576 | /* |
| 1577 | * If bio->bi_dev is a partition, remap the location |
| 1578 | */ |
| 1579 | static inline void blk_partition_remap(struct bio *bio) |
| 1580 | { |
| 1581 | struct block_device *bdev = bio->bi_bdev; |
| 1582 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1583 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | struct hd_struct *p = bdev->bd_part; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1585 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | bio->bi_sector += p->start_sect; |
| 1587 | bio->bi_bdev = bdev->bd_contains; |
Alan D. Brunelle | c7149d6 | 2007-08-07 15:30:23 +0200 | [diff] [blame] | 1588 | |
Mike Snitzer | d07335e | 2010-11-16 12:52:38 +0100 | [diff] [blame] | 1589 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, |
| 1590 | bdev->bd_dev, |
| 1591 | bio->bi_sector - p->start_sect); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1592 | } |
| 1593 | } |
| 1594 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | static void handle_bad_sector(struct bio *bio) |
| 1596 | { |
| 1597 | char b[BDEVNAME_SIZE]; |
| 1598 | |
| 1599 | printk(KERN_INFO "attempt to access beyond end of device\n"); |
| 1600 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", |
| 1601 | bdevname(bio->bi_bdev, b), |
| 1602 | bio->bi_rw, |
Kent Overstreet | f73a1c7 | 2012-09-25 15:05:12 -0700 | [diff] [blame] | 1603 | (unsigned long long)bio_end_sector(bio), |
Mike Snitzer | 77304d2 | 2010-11-08 14:39:12 +0100 | [diff] [blame] | 1604 | (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1605 | |
| 1606 | set_bit(BIO_EOF, &bio->bi_flags); |
| 1607 | } |
| 1608 | |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1609 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
| 1610 | |
| 1611 | static DECLARE_FAULT_ATTR(fail_make_request); |
| 1612 | |
| 1613 | static int __init setup_fail_make_request(char *str) |
| 1614 | { |
| 1615 | return setup_fault_attr(&fail_make_request, str); |
| 1616 | } |
| 1617 | __setup("fail_make_request=", setup_fail_make_request); |
| 1618 | |
Akinobu Mita | b2c9cd3 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1619 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1620 | { |
Akinobu Mita | b2c9cd3 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1621 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1622 | } |
| 1623 | |
| 1624 | static int __init fail_make_request_debugfs(void) |
| 1625 | { |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 1626 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
| 1627 | NULL, &fail_make_request); |
| 1628 | |
| 1629 | return IS_ERR(dir) ? PTR_ERR(dir) : 0; |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1630 | } |
| 1631 | |
| 1632 | late_initcall(fail_make_request_debugfs); |
| 1633 | |
| 1634 | #else /* CONFIG_FAIL_MAKE_REQUEST */ |
| 1635 | |
Akinobu Mita | b2c9cd3 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1636 | static inline bool should_fail_request(struct hd_struct *part, |
| 1637 | unsigned int bytes) |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1638 | { |
Akinobu Mita | b2c9cd3 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1639 | return false; |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1640 | } |
| 1641 | |
| 1642 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
| 1643 | |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1644 | /* |
| 1645 | * Check whether this bio extends beyond the end of the device. |
| 1646 | */ |
| 1647 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) |
| 1648 | { |
| 1649 | sector_t maxsector; |
| 1650 | |
| 1651 | if (!nr_sectors) |
| 1652 | return 0; |
| 1653 | |
| 1654 | /* Test device or partition size, when known. */ |
Mike Snitzer | 77304d2 | 2010-11-08 14:39:12 +0100 | [diff] [blame] | 1655 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1656 | if (maxsector) { |
| 1657 | sector_t sector = bio->bi_sector; |
| 1658 | |
| 1659 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { |
| 1660 | /* |
| 1661 | * This may well happen - the kernel calls bread() |
| 1662 | * without checking the size of the device, e.g., when |
| 1663 | * mounting a device. |
| 1664 | */ |
| 1665 | handle_bad_sector(bio); |
| 1666 | return 1; |
| 1667 | } |
| 1668 | } |
| 1669 | |
| 1670 | return 0; |
| 1671 | } |
| 1672 | |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1673 | static noinline_for_stack bool |
| 1674 | generic_make_request_checks(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1675 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1676 | struct request_queue *q; |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1677 | int nr_sectors = bio_sectors(bio); |
Jens Axboe | 51fd77b | 2007-11-02 08:49:08 +0100 | [diff] [blame] | 1678 | int err = -EIO; |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1679 | char b[BDEVNAME_SIZE]; |
| 1680 | struct hd_struct *part; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | |
| 1682 | might_sleep(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1683 | |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1684 | if (bio_check_eod(bio, nr_sectors)) |
| 1685 | goto end_io; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1686 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1687 | q = bdev_get_queue(bio->bi_bdev); |
| 1688 | if (unlikely(!q)) { |
| 1689 | printk(KERN_ERR |
| 1690 | "generic_make_request: Trying to access " |
| 1691 | "nonexistent block-device %s (%Lu)\n", |
| 1692 | bdevname(bio->bi_bdev, b), |
| 1693 | (long long) bio->bi_sector); |
| 1694 | goto end_io; |
| 1695 | } |
| 1696 | |
Martin K. Petersen | e2a60da | 2012-09-18 12:19:25 -0400 | [diff] [blame] | 1697 | if (likely(bio_is_rw(bio) && |
| 1698 | nr_sectors > queue_max_hw_sectors(q))) { |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1699 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
| 1700 | bdevname(bio->bi_bdev, b), |
| 1701 | bio_sectors(bio), |
| 1702 | queue_max_hw_sectors(q)); |
| 1703 | goto end_io; |
| 1704 | } |
| 1705 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1706 | part = bio->bi_bdev->bd_part; |
| 1707 | if (should_fail_request(part, bio->bi_size) || |
| 1708 | should_fail_request(&part_to_disk(part)->part0, |
| 1709 | bio->bi_size)) |
| 1710 | goto end_io; |
| 1711 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1712 | /* |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1713 | * If this device has partitions, remap block n |
| 1714 | * of partition p to block n+start(p) of the disk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1715 | */ |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1716 | blk_partition_remap(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1717 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1718 | if (bio_check_eod(bio, nr_sectors)) |
| 1719 | goto end_io; |
| 1720 | |
| 1721 | /* |
| 1722 | * Filter flush bio's early so that make_request based |
| 1723 | * drivers without flush support don't have to worry |
| 1724 | * about them. |
| 1725 | */ |
| 1726 | if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { |
| 1727 | bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); |
| 1728 | if (!nr_sectors) { |
| 1729 | err = 0; |
Tejun Heo | a738467 | 2008-11-28 13:32:03 +0900 | [diff] [blame] | 1730 | goto end_io; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1731 | } |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1732 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1733 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1734 | if ((bio->bi_rw & REQ_DISCARD) && |
| 1735 | (!blk_queue_discard(q) || |
Martin K. Petersen | e2a60da | 2012-09-18 12:19:25 -0400 | [diff] [blame] | 1736 | ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) { |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1737 | err = -EOPNOTSUPP; |
| 1738 | goto end_io; |
| 1739 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1740 | |
Martin K. Petersen | 4363ac7 | 2012-09-18 12:19:27 -0400 | [diff] [blame] | 1741 | if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1742 | err = -EOPNOTSUPP; |
| 1743 | goto end_io; |
| 1744 | } |
| 1745 | |
Tejun Heo | 7f4b35d | 2012-06-04 20:40:56 -0700 | [diff] [blame] | 1746 | /* |
| 1747 | * Various block parts want %current->io_context and lazy ioc |
| 1748 | * allocation ends up trading a lot of pain for a small amount of |
| 1749 | * memory. Just allocate it upfront. This may fail and block |
| 1750 | * layer knows how to live with it. |
| 1751 | */ |
| 1752 | create_io_context(GFP_ATOMIC, q->node); |
| 1753 | |
Tejun Heo | bc16a4f | 2011-10-19 14:33:01 +0200 | [diff] [blame] | 1754 | if (blk_throtl_bio(q, bio)) |
| 1755 | return false; /* throttled, will be resubmitted later */ |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1756 | |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 1757 | trace_block_bio_queue(q, bio); |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1758 | return true; |
Tejun Heo | a738467 | 2008-11-28 13:32:03 +0900 | [diff] [blame] | 1759 | |
| 1760 | end_io: |
| 1761 | bio_endio(bio, err); |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1762 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1763 | } |
| 1764 | |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1765 | /** |
| 1766 | * generic_make_request - hand a buffer to its device driver for I/O |
| 1767 | * @bio: The bio describing the location in memory and on the device. |
| 1768 | * |
| 1769 | * generic_make_request() is used to make I/O requests of block |
| 1770 | * devices. It is passed a &struct bio, which describes the I/O that needs |
| 1771 | * to be done. |
| 1772 | * |
| 1773 | * generic_make_request() does not return any status. The |
| 1774 | * success/failure status of the request, along with notification of |
| 1775 | * completion, is delivered asynchronously through the bio->bi_end_io |
| 1776 | * function described (one day) else where. |
| 1777 | * |
| 1778 | * The caller of generic_make_request must make sure that bi_io_vec |
| 1779 | * are set to describe the memory buffer, and that bi_dev and bi_sector are |
| 1780 | * set to describe the device address, and the |
| 1781 | * bi_end_io and optionally bi_private are set to describe how |
| 1782 | * completion notification should be signaled. |
| 1783 | * |
| 1784 | * generic_make_request and the drivers it calls may use bi_next if this |
| 1785 | * bio happens to be merged with someone else, and may resubmit the bio to |
| 1786 | * a lower device by calling into generic_make_request recursively, which |
| 1787 | * means the bio should NOT be touched after the call to ->make_request_fn. |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1788 | */ |
| 1789 | void generic_make_request(struct bio *bio) |
| 1790 | { |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1791 | struct bio_list bio_list_on_stack; |
| 1792 | |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1793 | if (!generic_make_request_checks(bio)) |
| 1794 | return; |
| 1795 | |
| 1796 | /* |
| 1797 | * We only want one ->make_request_fn to be active at a time, else |
| 1798 | * stack usage with stacked devices could be a problem. So use |
| 1799 | * current->bio_list to keep a list of requests submited by a |
| 1800 | * make_request_fn function. current->bio_list is also used as a |
| 1801 | * flag to say if generic_make_request is currently active in this |
| 1802 | * task or not. If it is NULL, then no make_request is active. If |
| 1803 | * it is non-NULL, then a make_request is active, and new requests |
| 1804 | * should be added at the tail |
| 1805 | */ |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1806 | if (current->bio_list) { |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1807 | bio_list_add(current->bio_list, bio); |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1808 | return; |
| 1809 | } |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1810 | |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1811 | /* following loop may be a bit non-obvious, and so deserves some |
| 1812 | * explanation. |
| 1813 | * Before entering the loop, bio->bi_next is NULL (as all callers |
| 1814 | * ensure that) so we have a list with a single bio. |
| 1815 | * We pretend that we have just taken it off a longer list, so |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1816 | * we assign bio_list to a pointer to the bio_list_on_stack, |
| 1817 | * thus initialising the bio_list of new bios to be |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1818 | * added. ->make_request() may indeed add some more bios |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1819 | * through a recursive call to generic_make_request. If it |
| 1820 | * did, we find a non-NULL value in bio_list and re-enter the loop |
| 1821 | * from the top. In this case we really did just take the bio |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1822 | * of the top of the list (no pretending) and so remove it from |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1823 | * bio_list, and call into ->make_request() again. |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1824 | */ |
| 1825 | BUG_ON(bio->bi_next); |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1826 | bio_list_init(&bio_list_on_stack); |
| 1827 | current->bio_list = &bio_list_on_stack; |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1828 | do { |
Christoph Hellwig | 27a84d5 | 2011-09-15 14:01:40 +0200 | [diff] [blame] | 1829 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
| 1830 | |
| 1831 | q->make_request_fn(q, bio); |
| 1832 | |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1833 | bio = bio_list_pop(current->bio_list); |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1834 | } while (bio); |
Akinobu Mita | bddd87c | 2010-02-23 08:55:42 +0100 | [diff] [blame] | 1835 | current->bio_list = NULL; /* deactivate */ |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1836 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1837 | EXPORT_SYMBOL(generic_make_request); |
| 1838 | |
| 1839 | /** |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1840 | * submit_bio - submit a bio to the block device layer for I/O |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1841 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
| 1842 | * @bio: The &struct bio which describes the I/O |
| 1843 | * |
| 1844 | * submit_bio() is very similar in purpose to generic_make_request(), and |
| 1845 | * uses that function to do most of the work. Both are fairly rough |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1846 | * interfaces; @bio must be presetup and ready for I/O. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | * |
| 1848 | */ |
| 1849 | void submit_bio(int rw, struct bio *bio) |
| 1850 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1851 | bio->bi_rw |= rw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1852 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1853 | /* |
| 1854 | * If it's a regular read/write or a barrier with data attached, |
| 1855 | * go through the normal accounting stuff before submission. |
| 1856 | */ |
Martin K. Petersen | e2a60da | 2012-09-18 12:19:25 -0400 | [diff] [blame] | 1857 | if (bio_has_data(bio)) { |
Martin K. Petersen | 4363ac7 | 2012-09-18 12:19:27 -0400 | [diff] [blame] | 1858 | unsigned int count; |
| 1859 | |
| 1860 | if (unlikely(rw & REQ_WRITE_SAME)) |
| 1861 | count = bdev_logical_block_size(bio->bi_bdev) >> 9; |
| 1862 | else |
| 1863 | count = bio_sectors(bio); |
| 1864 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1865 | if (rw & WRITE) { |
| 1866 | count_vm_events(PGPGOUT, count); |
| 1867 | } else { |
| 1868 | task_io_account_read(bio->bi_size); |
| 1869 | count_vm_events(PGPGIN, count); |
| 1870 | } |
| 1871 | |
| 1872 | if (unlikely(block_dump)) { |
| 1873 | char b[BDEVNAME_SIZE]; |
San Mehat | 8dcbdc7 | 2010-09-14 08:48:01 +0200 | [diff] [blame] | 1874 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
Pavel Emelyanov | ba25f9d | 2007-10-18 23:40:40 -0700 | [diff] [blame] | 1875 | current->comm, task_pid_nr(current), |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1876 | (rw & WRITE) ? "WRITE" : "READ", |
| 1877 | (unsigned long long)bio->bi_sector, |
San Mehat | 8dcbdc7 | 2010-09-14 08:48:01 +0200 | [diff] [blame] | 1878 | bdevname(bio->bi_bdev, b), |
| 1879 | count); |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1880 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1881 | } |
| 1882 | |
| 1883 | generic_make_request(bio); |
| 1884 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1885 | EXPORT_SYMBOL(submit_bio); |
| 1886 | |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 1887 | /** |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1888 | * blk_rq_check_limits - Helper function to check a request for the queue limit |
| 1889 | * @q: the queue |
| 1890 | * @rq: the request being checked |
| 1891 | * |
| 1892 | * Description: |
| 1893 | * @rq may have been made based on weaker limitations of upper-level queues |
| 1894 | * in request stacking drivers, and it may violate the limitation of @q. |
| 1895 | * Since the block layer and the underlying device driver trust @rq |
| 1896 | * after it is inserted to @q, it should be checked against @q before |
| 1897 | * the insertion using this generic function. |
| 1898 | * |
| 1899 | * This function should also be useful for request stacking drivers |
Stefan Weil | eef35c2 | 2010-08-06 21:11:15 +0200 | [diff] [blame] | 1900 | * in some cases below, so export this function. |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1901 | * Request stacking drivers like request-based dm may change the queue |
| 1902 | * limits while requests are in the queue (e.g. dm's table swapping). |
| 1903 | * Such request stacking drivers should check those requests agaist |
| 1904 | * the new queue limits again when they dispatch those requests, |
| 1905 | * although such checkings are also done against the old queue limits |
| 1906 | * when submitting requests. |
| 1907 | */ |
| 1908 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) |
| 1909 | { |
Martin K. Petersen | e2a60da | 2012-09-18 12:19:25 -0400 | [diff] [blame] | 1910 | if (!rq_mergeable(rq)) |
ike Snitzer | 3383977 | 2010-08-08 12:11:33 -0400 | [diff] [blame] | 1911 | return 0; |
| 1912 | |
Martin K. Petersen | f31dc1c | 2012-09-18 12:19:26 -0400 | [diff] [blame] | 1913 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) { |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1914 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
| 1915 | return -EIO; |
| 1916 | } |
| 1917 | |
| 1918 | /* |
| 1919 | * queue's settings related to segment counting like q->bounce_pfn |
| 1920 | * may differ from that of other stacking queues. |
| 1921 | * Recalculate it to check the request correctly on this queue's |
| 1922 | * limitation. |
| 1923 | */ |
| 1924 | blk_recalc_rq_segments(rq); |
Martin K. Petersen | 8a78362 | 2010-02-26 00:20:39 -0500 | [diff] [blame] | 1925 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1926 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
| 1927 | return -EIO; |
| 1928 | } |
| 1929 | |
| 1930 | return 0; |
| 1931 | } |
| 1932 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); |
| 1933 | |
| 1934 | /** |
| 1935 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request |
| 1936 | * @q: the queue to submit the request |
| 1937 | * @rq: the request being queued |
| 1938 | */ |
| 1939 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
| 1940 | { |
| 1941 | unsigned long flags; |
Jeff Moyer | 4853aba | 2011-08-15 21:37:25 +0200 | [diff] [blame] | 1942 | int where = ELEVATOR_INSERT_BACK; |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1943 | |
| 1944 | if (blk_rq_check_limits(q, rq)) |
| 1945 | return -EIO; |
| 1946 | |
Akinobu Mita | b2c9cd3 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1947 | if (rq->rq_disk && |
| 1948 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1949 | return -EIO; |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1950 | |
| 1951 | spin_lock_irqsave(q->queue_lock, flags); |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 1952 | if (unlikely(blk_queue_dying(q))) { |
Tejun Heo | 8ba6143 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 1953 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1954 | return -ENODEV; |
| 1955 | } |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1956 | |
| 1957 | /* |
| 1958 | * Submitting request must be dequeued before calling this function |
| 1959 | * because it will be linked to another request_queue |
| 1960 | */ |
| 1961 | BUG_ON(blk_queued_rq(rq)); |
| 1962 | |
Jeff Moyer | 4853aba | 2011-08-15 21:37:25 +0200 | [diff] [blame] | 1963 | if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) |
| 1964 | where = ELEVATOR_INSERT_FLUSH; |
| 1965 | |
| 1966 | add_acct_request(q, rq, where); |
Jeff Moyer | e67b77c | 2011-10-17 12:57:23 +0200 | [diff] [blame] | 1967 | if (where == ELEVATOR_INSERT_FLUSH) |
| 1968 | __blk_run_queue(q); |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1969 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1970 | |
| 1971 | return 0; |
| 1972 | } |
| 1973 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); |
| 1974 | |
Tejun Heo | 80a761f | 2009-07-03 17:48:17 +0900 | [diff] [blame] | 1975 | /** |
| 1976 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary |
| 1977 | * @rq: request to examine |
| 1978 | * |
| 1979 | * Description: |
| 1980 | * A request could be merge of IOs which require different failure |
| 1981 | * handling. This function determines the number of bytes which |
| 1982 | * can be failed from the beginning of the request without |
| 1983 | * crossing into area which need to be retried further. |
| 1984 | * |
| 1985 | * Return: |
| 1986 | * The number of bytes to fail. |
| 1987 | * |
| 1988 | * Context: |
| 1989 | * queue_lock must be held. |
| 1990 | */ |
| 1991 | unsigned int blk_rq_err_bytes(const struct request *rq) |
| 1992 | { |
| 1993 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
| 1994 | unsigned int bytes = 0; |
| 1995 | struct bio *bio; |
| 1996 | |
| 1997 | if (!(rq->cmd_flags & REQ_MIXED_MERGE)) |
| 1998 | return blk_rq_bytes(rq); |
| 1999 | |
| 2000 | /* |
| 2001 | * Currently the only 'mixing' which can happen is between |
| 2002 | * different fastfail types. We can safely fail portions |
| 2003 | * which have all the failfast bits that the first one has - |
| 2004 | * the ones which are at least as eager to fail as the first |
| 2005 | * one. |
| 2006 | */ |
| 2007 | for (bio = rq->bio; bio; bio = bio->bi_next) { |
| 2008 | if ((bio->bi_rw & ff) != ff) |
| 2009 | break; |
| 2010 | bytes += bio->bi_size; |
| 2011 | } |
| 2012 | |
| 2013 | /* this could lead to infinite loop */ |
| 2014 | BUG_ON(blk_rq_bytes(rq) && !bytes); |
| 2015 | return bytes; |
| 2016 | } |
| 2017 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); |
| 2018 | |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2019 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
| 2020 | { |
Jens Axboe | c2553b5 | 2009-04-24 08:10:11 +0200 | [diff] [blame] | 2021 | if (blk_do_io_stat(req)) { |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2022 | const int rw = rq_data_dir(req); |
| 2023 | struct hd_struct *part; |
| 2024 | int cpu; |
| 2025 | |
| 2026 | cpu = part_stat_lock(); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 2027 | part = req->part; |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2028 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); |
| 2029 | part_stat_unlock(); |
| 2030 | } |
| 2031 | } |
| 2032 | |
| 2033 | static void blk_account_io_done(struct request *req) |
| 2034 | { |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2035 | /* |
Tejun Heo | dd4c133 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 2036 | * Account IO completion. flush_rq isn't accounted as a |
| 2037 | * normal IO on queueing nor completion. Accounting the |
| 2038 | * containing request is enough. |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2039 | */ |
Tejun Heo | 414b4ff | 2011-01-25 12:43:49 +0100 | [diff] [blame] | 2040 | if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2041 | unsigned long duration = jiffies - req->start_time; |
| 2042 | const int rw = rq_data_dir(req); |
| 2043 | struct hd_struct *part; |
| 2044 | int cpu; |
| 2045 | |
| 2046 | cpu = part_stat_lock(); |
Jerome Marchand | 09e099d | 2011-01-05 16:57:38 +0100 | [diff] [blame] | 2047 | part = req->part; |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2048 | |
| 2049 | part_stat_inc(cpu, part, ios[rw]); |
| 2050 | part_stat_add(cpu, part, ticks[rw], duration); |
| 2051 | part_round_stats(cpu, part); |
Nikanth Karthikesan | 316d315 | 2009-10-06 20:16:55 +0200 | [diff] [blame] | 2052 | part_dec_in_flight(part, rw); |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2053 | |
Jens Axboe | 6c23a96 | 2011-01-07 08:43:37 +0100 | [diff] [blame] | 2054 | hd_struct_put(part); |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2055 | part_stat_unlock(); |
| 2056 | } |
| 2057 | } |
| 2058 | |
Lin Ming | c815881 | 2013-03-23 11:42:27 +0800 | [diff] [blame] | 2059 | #ifdef CONFIG_PM_RUNTIME |
| 2060 | /* |
| 2061 | * Don't process normal requests when queue is suspended |
| 2062 | * or in the process of suspending/resuming |
| 2063 | */ |
| 2064 | static struct request *blk_pm_peek_request(struct request_queue *q, |
| 2065 | struct request *rq) |
| 2066 | { |
| 2067 | if (q->dev && (q->rpm_status == RPM_SUSPENDED || |
| 2068 | (q->rpm_status != RPM_ACTIVE && !(rq->cmd_flags & REQ_PM)))) |
| 2069 | return NULL; |
| 2070 | else |
| 2071 | return rq; |
| 2072 | } |
| 2073 | #else |
| 2074 | static inline struct request *blk_pm_peek_request(struct request_queue *q, |
| 2075 | struct request *rq) |
| 2076 | { |
| 2077 | return rq; |
| 2078 | } |
| 2079 | #endif |
| 2080 | |
Tejun Heo | 53a0880 | 2008-12-03 12:41:26 +0100 | [diff] [blame] | 2081 | /** |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2082 | * blk_peek_request - peek at the top of a request queue |
| 2083 | * @q: request queue to peek at |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2084 | * |
| 2085 | * Description: |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2086 | * Return the request at the top of @q. The returned request |
| 2087 | * should be started using blk_start_request() before LLD starts |
| 2088 | * processing it. |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2089 | * |
| 2090 | * Return: |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2091 | * Pointer to the request at the top of @q if available. Null |
| 2092 | * otherwise. |
| 2093 | * |
| 2094 | * Context: |
| 2095 | * queue_lock must be held. |
| 2096 | */ |
| 2097 | struct request *blk_peek_request(struct request_queue *q) |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2098 | { |
| 2099 | struct request *rq; |
| 2100 | int ret; |
| 2101 | |
| 2102 | while ((rq = __elv_next_request(q)) != NULL) { |
Lin Ming | c815881 | 2013-03-23 11:42:27 +0800 | [diff] [blame] | 2103 | |
| 2104 | rq = blk_pm_peek_request(q, rq); |
| 2105 | if (!rq) |
| 2106 | break; |
| 2107 | |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2108 | if (!(rq->cmd_flags & REQ_STARTED)) { |
| 2109 | /* |
| 2110 | * This is the first time the device driver |
| 2111 | * sees this request (possibly after |
| 2112 | * requeueing). Notify IO scheduler. |
| 2113 | */ |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 2114 | if (rq->cmd_flags & REQ_SORTED) |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2115 | elv_activate_rq(q, rq); |
| 2116 | |
| 2117 | /* |
| 2118 | * just mark as started even if we don't start |
| 2119 | * it, a request that has been delayed should |
| 2120 | * not be passed by new incoming requests |
| 2121 | */ |
| 2122 | rq->cmd_flags |= REQ_STARTED; |
| 2123 | trace_block_rq_issue(q, rq); |
| 2124 | } |
| 2125 | |
| 2126 | if (!q->boundary_rq || q->boundary_rq == rq) { |
| 2127 | q->end_sector = rq_end_sector(rq); |
| 2128 | q->boundary_rq = NULL; |
| 2129 | } |
| 2130 | |
| 2131 | if (rq->cmd_flags & REQ_DONTPREP) |
| 2132 | break; |
| 2133 | |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2134 | if (q->dma_drain_size && blk_rq_bytes(rq)) { |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2135 | /* |
| 2136 | * make sure space for the drain appears we |
| 2137 | * know we can do this because max_hw_segments |
| 2138 | * has been adjusted to be one fewer than the |
| 2139 | * device can handle |
| 2140 | */ |
| 2141 | rq->nr_phys_segments++; |
| 2142 | } |
| 2143 | |
| 2144 | if (!q->prep_rq_fn) |
| 2145 | break; |
| 2146 | |
| 2147 | ret = q->prep_rq_fn(q, rq); |
| 2148 | if (ret == BLKPREP_OK) { |
| 2149 | break; |
| 2150 | } else if (ret == BLKPREP_DEFER) { |
| 2151 | /* |
| 2152 | * the request may have been (partially) prepped. |
| 2153 | * we need to keep this request in the front to |
| 2154 | * avoid resource deadlock. REQ_STARTED will |
| 2155 | * prevent other fs requests from passing this one. |
| 2156 | */ |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2157 | if (q->dma_drain_size && blk_rq_bytes(rq) && |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2158 | !(rq->cmd_flags & REQ_DONTPREP)) { |
| 2159 | /* |
| 2160 | * remove the space for the drain we added |
| 2161 | * so that we don't add it again |
| 2162 | */ |
| 2163 | --rq->nr_phys_segments; |
| 2164 | } |
| 2165 | |
| 2166 | rq = NULL; |
| 2167 | break; |
| 2168 | } else if (ret == BLKPREP_KILL) { |
| 2169 | rq->cmd_flags |= REQ_QUIET; |
James Bottomley | c143dc9 | 2009-05-30 06:43:49 +0200 | [diff] [blame] | 2170 | /* |
| 2171 | * Mark this request as started so we don't trigger |
| 2172 | * any debug logic in the end I/O path. |
| 2173 | */ |
| 2174 | blk_start_request(rq); |
Tejun Heo | 40cbbb7 | 2009-04-23 11:05:19 +0900 | [diff] [blame] | 2175 | __blk_end_request_all(rq, -EIO); |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2176 | } else { |
| 2177 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); |
| 2178 | break; |
| 2179 | } |
| 2180 | } |
| 2181 | |
| 2182 | return rq; |
| 2183 | } |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2184 | EXPORT_SYMBOL(blk_peek_request); |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2185 | |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2186 | void blk_dequeue_request(struct request *rq) |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2187 | { |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2188 | struct request_queue *q = rq->q; |
| 2189 | |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2190 | BUG_ON(list_empty(&rq->queuelist)); |
| 2191 | BUG_ON(ELV_ON_HASH(rq)); |
| 2192 | |
| 2193 | list_del_init(&rq->queuelist); |
| 2194 | |
| 2195 | /* |
| 2196 | * the time frame between a request being removed from the lists |
| 2197 | * and to it is freed is accounted as io that is in progress at |
| 2198 | * the driver side. |
| 2199 | */ |
Divyesh Shah | 9195291 | 2010-04-01 15:01:41 -0700 | [diff] [blame] | 2200 | if (blk_account_rq(rq)) { |
Jens Axboe | 0a7ae2f | 2009-05-20 08:54:31 +0200 | [diff] [blame] | 2201 | q->in_flight[rq_is_sync(rq)]++; |
Divyesh Shah | 9195291 | 2010-04-01 15:01:41 -0700 | [diff] [blame] | 2202 | set_io_start_time_ns(rq); |
| 2203 | } |
Tejun Heo | 158dbda | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2204 | } |
| 2205 | |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2206 | /** |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2207 | * blk_start_request - start request processing on the driver |
| 2208 | * @req: request to dequeue |
| 2209 | * |
| 2210 | * Description: |
| 2211 | * Dequeue @req and start timeout timer on it. This hands off the |
| 2212 | * request to the driver. |
| 2213 | * |
| 2214 | * Block internal functions which don't want to start timer should |
| 2215 | * call blk_dequeue_request(). |
| 2216 | * |
| 2217 | * Context: |
| 2218 | * queue_lock must be held. |
| 2219 | */ |
| 2220 | void blk_start_request(struct request *req) |
| 2221 | { |
| 2222 | blk_dequeue_request(req); |
| 2223 | |
| 2224 | /* |
Tejun Heo | 5f49f63 | 2009-05-19 18:33:05 +0900 | [diff] [blame] | 2225 | * We are now handing the request to the hardware, initialize |
| 2226 | * resid_len to full count and add the timeout handler. |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2227 | */ |
Tejun Heo | 5f49f63 | 2009-05-19 18:33:05 +0900 | [diff] [blame] | 2228 | req->resid_len = blk_rq_bytes(req); |
FUJITA Tomonori | dbb66c4 | 2009-06-09 05:47:10 +0200 | [diff] [blame] | 2229 | if (unlikely(blk_bidi_rq(req))) |
| 2230 | req->next_rq->resid_len = blk_rq_bytes(req->next_rq); |
| 2231 | |
Tejun Heo | 9934c8c | 2009-05-08 11:54:16 +0900 | [diff] [blame] | 2232 | blk_add_timer(req); |
| 2233 | } |
| 2234 | EXPORT_SYMBOL(blk_start_request); |
| 2235 | |
| 2236 | /** |
| 2237 | * blk_fetch_request - fetch a request from a request queue |
| 2238 | * @q: request queue to fetch a request from |
| 2239 | * |
| 2240 | * Description: |
| 2241 | * Return the request at the top of @q. The request is started on |
| 2242 | * return and LLD can start processing it immediately. |
| 2243 | * |
| 2244 | * Return: |
| 2245 | * Pointer to the request at the top of @q if available. Null |
| 2246 | * otherwise. |
| 2247 | * |
| 2248 | * Context: |
| 2249 | * queue_lock must be held. |
| 2250 | */ |
| 2251 | struct request *blk_fetch_request(struct request_queue *q) |
| 2252 | { |
| 2253 | struct request *rq; |
| 2254 | |
| 2255 | rq = blk_peek_request(q); |
| 2256 | if (rq) |
| 2257 | blk_start_request(rq); |
| 2258 | return rq; |
| 2259 | } |
| 2260 | EXPORT_SYMBOL(blk_fetch_request); |
| 2261 | |
| 2262 | /** |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2263 | * blk_update_request - Special helper function for request stacking drivers |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2264 | * @req: the request being processed |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2265 | * @error: %0 for success, < %0 for error |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2266 | * @nr_bytes: number of bytes to complete @req |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2267 | * |
| 2268 | * Description: |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2269 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
| 2270 | * the request structure even if @req doesn't have leftover. |
| 2271 | * If @req has leftover, sets it up for the next range of segments. |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2272 | * |
| 2273 | * This special helper function is only for request stacking drivers |
| 2274 | * (e.g. request-based dm) so that they can handle partial completion. |
| 2275 | * Actual device drivers should use blk_end_request instead. |
| 2276 | * |
| 2277 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees |
| 2278 | * %false return from this function. |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2279 | * |
| 2280 | * Return: |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2281 | * %false - this request doesn't have any more data |
| 2282 | * %true - this request has more data |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 2283 | **/ |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2284 | bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2285 | { |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2286 | int total_bytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2287 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2288 | if (!req->bio) |
| 2289 | return false; |
| 2290 | |
Arnaldo Carvalho de Melo | 5f3ea37 | 2008-10-30 08:34:33 +0100 | [diff] [blame] | 2291 | trace_block_rq_complete(req->q, req); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 2292 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2293 | /* |
Tejun Heo | 6f41469 | 2009-04-19 07:00:41 +0900 | [diff] [blame] | 2294 | * For fs requests, rq is just carrier of independent bio's |
| 2295 | * and each partial completion should be handled separately. |
| 2296 | * Reset per-request error on each partial completion. |
| 2297 | * |
| 2298 | * TODO: tj: This is too subtle. It would be better to let |
| 2299 | * low level drivers do what they see fit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | */ |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 2301 | if (req->cmd_type == REQ_TYPE_FS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | req->errors = 0; |
| 2303 | |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 2304 | if (error && req->cmd_type == REQ_TYPE_FS && |
| 2305 | !(req->cmd_flags & REQ_QUIET)) { |
Hannes Reinecke | 7977556 | 2011-01-18 10:13:13 +0100 | [diff] [blame] | 2306 | char *error_type; |
| 2307 | |
| 2308 | switch (error) { |
| 2309 | case -ENOLINK: |
| 2310 | error_type = "recoverable transport"; |
| 2311 | break; |
| 2312 | case -EREMOTEIO: |
| 2313 | error_type = "critical target"; |
| 2314 | break; |
| 2315 | case -EBADE: |
| 2316 | error_type = "critical nexus"; |
| 2317 | break; |
Hannes Reinecke | d1ffc1f | 2013-01-30 09:26:16 +0000 | [diff] [blame] | 2318 | case -ETIMEDOUT: |
| 2319 | error_type = "timeout"; |
| 2320 | break; |
Hannes Reinecke | a9d6ceb8 | 2013-07-01 15:16:25 +0200 | [diff] [blame] | 2321 | case -ENOSPC: |
| 2322 | error_type = "critical space allocation"; |
| 2323 | break; |
Hannes Reinecke | 7e782af | 2013-07-01 15:16:26 +0200 | [diff] [blame] | 2324 | case -ENODATA: |
| 2325 | error_type = "critical medium"; |
| 2326 | break; |
Hannes Reinecke | 7977556 | 2011-01-18 10:13:13 +0100 | [diff] [blame] | 2327 | case -EIO: |
| 2328 | default: |
| 2329 | error_type = "I/O"; |
| 2330 | break; |
| 2331 | } |
Yi Zou | 37d7b34 | 2012-08-30 16:26:25 -0700 | [diff] [blame] | 2332 | printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", |
| 2333 | error_type, req->rq_disk ? |
| 2334 | req->rq_disk->disk_name : "?", |
| 2335 | (unsigned long long)blk_rq_pos(req)); |
| 2336 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | } |
| 2338 | |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2339 | blk_account_io_completion(req, nr_bytes); |
Jens Axboe | d72d904 | 2005-11-01 08:35:42 +0100 | [diff] [blame] | 2340 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2341 | total_bytes = 0; |
| 2342 | while (req->bio) { |
| 2343 | struct bio *bio = req->bio; |
| 2344 | unsigned bio_bytes = min(bio->bi_size, nr_bytes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2345 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2346 | if (bio_bytes == bio->bi_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2347 | req->bio = bio->bi_next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2349 | req_bio_endio(req, bio, bio_bytes, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2351 | total_bytes += bio_bytes; |
| 2352 | nr_bytes -= bio_bytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2353 | |
Kent Overstreet | f79ea41 | 2012-09-20 16:38:30 -0700 | [diff] [blame] | 2354 | if (!nr_bytes) |
| 2355 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2356 | } |
| 2357 | |
| 2358 | /* |
| 2359 | * completely done |
| 2360 | */ |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2361 | if (!req->bio) { |
| 2362 | /* |
| 2363 | * Reset counters so that the request stacking driver |
| 2364 | * can find how many bytes remain in the request |
| 2365 | * later. |
| 2366 | */ |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 2367 | req->__data_len = 0; |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2368 | return false; |
| 2369 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2370 | |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 2371 | req->__data_len -= total_bytes; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2372 | req->buffer = bio_data(req->bio); |
| 2373 | |
| 2374 | /* update sector only for requests with clear definition of sector */ |
Martin K. Petersen | e2a60da | 2012-09-18 12:19:25 -0400 | [diff] [blame] | 2375 | if (req->cmd_type == REQ_TYPE_FS) |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 2376 | req->__sector += total_bytes >> 9; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2377 | |
Tejun Heo | 80a761f | 2009-07-03 17:48:17 +0900 | [diff] [blame] | 2378 | /* mixed attributes always follow the first bio */ |
| 2379 | if (req->cmd_flags & REQ_MIXED_MERGE) { |
| 2380 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
| 2381 | req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; |
| 2382 | } |
| 2383 | |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2384 | /* |
| 2385 | * If total number of sectors is less than the first segment |
| 2386 | * size, something has gone terribly wrong. |
| 2387 | */ |
| 2388 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { |
Jens Axboe | 8182924 | 2011-03-30 09:51:33 +0200 | [diff] [blame] | 2389 | blk_dump_rq_flags(req, "request botched"); |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 2390 | req->__data_len = blk_rq_cur_bytes(req); |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2391 | } |
| 2392 | |
| 2393 | /* recalculate the number of segments */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2394 | blk_recalc_rq_segments(req); |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2395 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2396 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2397 | } |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2398 | EXPORT_SYMBOL_GPL(blk_update_request); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2399 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2400 | static bool blk_update_bidi_request(struct request *rq, int error, |
| 2401 | unsigned int nr_bytes, |
| 2402 | unsigned int bidi_bytes) |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2403 | { |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2404 | if (blk_update_request(rq, error, nr_bytes)) |
| 2405 | return true; |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2406 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2407 | /* Bidi request must be completed as a whole */ |
| 2408 | if (unlikely(blk_bidi_rq(rq)) && |
| 2409 | blk_update_request(rq->next_rq, error, bidi_bytes)) |
| 2410 | return true; |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2411 | |
Jens Axboe | e2e1a14 | 2010-06-09 10:42:09 +0200 | [diff] [blame] | 2412 | if (blk_queue_add_random(rq->q)) |
| 2413 | add_disk_randomness(rq->rq_disk); |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2414 | |
| 2415 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2416 | } |
| 2417 | |
James Bottomley | 28018c2 | 2010-07-01 19:49:17 +0900 | [diff] [blame] | 2418 | /** |
| 2419 | * blk_unprep_request - unprepare a request |
| 2420 | * @req: the request |
| 2421 | * |
| 2422 | * This function makes a request ready for complete resubmission (or |
| 2423 | * completion). It happens only after all error handling is complete, |
| 2424 | * so represents the appropriate moment to deallocate any resources |
| 2425 | * that were allocated to the request in the prep_rq_fn. The queue |
| 2426 | * lock is held when calling this. |
| 2427 | */ |
| 2428 | void blk_unprep_request(struct request *req) |
| 2429 | { |
| 2430 | struct request_queue *q = req->q; |
| 2431 | |
| 2432 | req->cmd_flags &= ~REQ_DONTPREP; |
| 2433 | if (q->unprep_rq_fn) |
| 2434 | q->unprep_rq_fn(q, req); |
| 2435 | } |
| 2436 | EXPORT_SYMBOL_GPL(blk_unprep_request); |
| 2437 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2438 | /* |
| 2439 | * queue lock must be held |
| 2440 | */ |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2441 | static void blk_finish_request(struct request *req, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2442 | { |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 2443 | if (blk_rq_tagged(req)) |
| 2444 | blk_queue_end_tag(req->q, req); |
| 2445 | |
James Bottomley | ba396a6 | 2009-05-27 14:17:08 +0200 | [diff] [blame] | 2446 | BUG_ON(blk_queued_rq(req)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2447 | |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 2448 | if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) |
Matthew Garrett | 31373d0 | 2010-04-06 14:25:14 +0200 | [diff] [blame] | 2449 | laptop_io_completion(&req->q->backing_dev_info); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | |
Mike Anderson | e78042e | 2008-10-30 02:16:20 -0700 | [diff] [blame] | 2451 | blk_delete_timer(req); |
| 2452 | |
James Bottomley | 28018c2 | 2010-07-01 19:49:17 +0900 | [diff] [blame] | 2453 | if (req->cmd_flags & REQ_DONTPREP) |
| 2454 | blk_unprep_request(req); |
| 2455 | |
| 2456 | |
Jens Axboe | bc58ba9 | 2009-01-23 10:54:44 +0100 | [diff] [blame] | 2457 | blk_account_io_done(req); |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 2458 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | if (req->end_io) |
Tejun Heo | 8ffdc65 | 2006-01-06 09:49:03 +0100 | [diff] [blame] | 2460 | req->end_io(req, error); |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 2461 | else { |
| 2462 | if (blk_bidi_rq(req)) |
| 2463 | __blk_put_request(req->next_rq->q, req->next_rq); |
| 2464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2465 | __blk_put_request(req->q, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2466 | } |
| 2467 | } |
| 2468 | |
Kiyoshi Ueda | 3b11313 | 2007-12-11 17:41:17 -0500 | [diff] [blame] | 2469 | /** |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2470 | * blk_end_bidi_request - Complete a bidi request |
| 2471 | * @rq: the request to complete |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 2472 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 2473 | * @nr_bytes: number of bytes to complete @rq |
| 2474 | * @bidi_bytes: number of bytes to complete @rq->next_rq |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2475 | * |
| 2476 | * Description: |
| 2477 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2478 | * Drivers that supports bidi can safely call this member for any |
| 2479 | * type of request, bidi or uni. In the later case @bidi_bytes is |
| 2480 | * just ignored. |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2481 | * |
| 2482 | * Return: |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2483 | * %false - we are done with this request |
| 2484 | * %true - still buffers pending for this request |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2485 | **/ |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2486 | static bool blk_end_bidi_request(struct request *rq, int error, |
| 2487 | unsigned int nr_bytes, unsigned int bidi_bytes) |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2488 | { |
| 2489 | struct request_queue *q = rq->q; |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2490 | unsigned long flags; |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2491 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2492 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
| 2493 | return true; |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2494 | |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2495 | spin_lock_irqsave(q->queue_lock, flags); |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2496 | blk_finish_request(rq, error); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2497 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 2498 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2499 | return false; |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 2500 | } |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 2501 | |
| 2502 | /** |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2503 | * __blk_end_bidi_request - Complete a bidi request with queue lock held |
| 2504 | * @rq: the request to complete |
| 2505 | * @error: %0 for success, < %0 for error |
| 2506 | * @nr_bytes: number of bytes to complete @rq |
| 2507 | * @bidi_bytes: number of bytes to complete @rq->next_rq |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2508 | * |
| 2509 | * Description: |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2510 | * Identical to blk_end_bidi_request() except that queue lock is |
| 2511 | * assumed to be locked on entry and remains so on return. |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2512 | * |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2513 | * Return: |
| 2514 | * %false - we are done with this request |
| 2515 | * %true - still buffers pending for this request |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2516 | **/ |
Jeff Moyer | 4853aba | 2011-08-15 21:37:25 +0200 | [diff] [blame] | 2517 | bool __blk_end_bidi_request(struct request *rq, int error, |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2518 | unsigned int nr_bytes, unsigned int bidi_bytes) |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2519 | { |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2520 | if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) |
| 2521 | return true; |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2522 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2523 | blk_finish_request(rq, error); |
Tejun Heo | 5efccd1 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2524 | |
Tejun Heo | 2e60e02 | 2009-04-23 11:05:18 +0900 | [diff] [blame] | 2525 | return false; |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2526 | } |
| 2527 | |
| 2528 | /** |
| 2529 | * blk_end_request - Helper function for drivers to complete the request. |
| 2530 | * @rq: the request being processed |
| 2531 | * @error: %0 for success, < %0 for error |
| 2532 | * @nr_bytes: number of bytes to complete |
| 2533 | * |
| 2534 | * Description: |
| 2535 | * Ends I/O on a number of bytes attached to @rq. |
| 2536 | * If @rq has leftover, sets it up for the next range of segments. |
| 2537 | * |
| 2538 | * Return: |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2539 | * %false - we are done with this request |
| 2540 | * %true - still buffers pending for this request |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2541 | **/ |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2542 | bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2543 | { |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2544 | return blk_end_bidi_request(rq, error, nr_bytes, 0); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2545 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2546 | EXPORT_SYMBOL(blk_end_request); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2547 | |
| 2548 | /** |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2549 | * blk_end_request_all - Helper function for drives to finish the request. |
| 2550 | * @rq: the request to finish |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2551 | * @error: %0 for success, < %0 for error |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2552 | * |
| 2553 | * Description: |
| 2554 | * Completely finish @rq. |
| 2555 | */ |
| 2556 | void blk_end_request_all(struct request *rq, int error) |
| 2557 | { |
| 2558 | bool pending; |
| 2559 | unsigned int bidi_bytes = 0; |
| 2560 | |
| 2561 | if (unlikely(blk_bidi_rq(rq))) |
| 2562 | bidi_bytes = blk_rq_bytes(rq->next_rq); |
| 2563 | |
| 2564 | pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
| 2565 | BUG_ON(pending); |
| 2566 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2567 | EXPORT_SYMBOL(blk_end_request_all); |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2568 | |
| 2569 | /** |
| 2570 | * blk_end_request_cur - Helper function to finish the current request chunk. |
| 2571 | * @rq: the request to finish the current chunk for |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2572 | * @error: %0 for success, < %0 for error |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2573 | * |
| 2574 | * Description: |
| 2575 | * Complete the current consecutively mapped chunk from @rq. |
| 2576 | * |
| 2577 | * Return: |
| 2578 | * %false - we are done with this request |
| 2579 | * %true - still buffers pending for this request |
| 2580 | */ |
| 2581 | bool blk_end_request_cur(struct request *rq, int error) |
| 2582 | { |
| 2583 | return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
| 2584 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2585 | EXPORT_SYMBOL(blk_end_request_cur); |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2586 | |
| 2587 | /** |
Tejun Heo | 80a761f | 2009-07-03 17:48:17 +0900 | [diff] [blame] | 2588 | * blk_end_request_err - Finish a request till the next failure boundary. |
| 2589 | * @rq: the request to finish till the next failure boundary for |
| 2590 | * @error: must be negative errno |
| 2591 | * |
| 2592 | * Description: |
| 2593 | * Complete @rq till the next failure boundary. |
| 2594 | * |
| 2595 | * Return: |
| 2596 | * %false - we are done with this request |
| 2597 | * %true - still buffers pending for this request |
| 2598 | */ |
| 2599 | bool blk_end_request_err(struct request *rq, int error) |
| 2600 | { |
| 2601 | WARN_ON(error >= 0); |
| 2602 | return blk_end_request(rq, error, blk_rq_err_bytes(rq)); |
| 2603 | } |
| 2604 | EXPORT_SYMBOL_GPL(blk_end_request_err); |
| 2605 | |
| 2606 | /** |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2607 | * __blk_end_request - Helper function for drivers to complete the request. |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2608 | * @rq: the request being processed |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2609 | * @error: %0 for success, < %0 for error |
| 2610 | * @nr_bytes: number of bytes to complete |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2611 | * |
| 2612 | * Description: |
| 2613 | * Must be called with queue lock held unlike blk_end_request(). |
| 2614 | * |
| 2615 | * Return: |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2616 | * %false - we are done with this request |
| 2617 | * %true - still buffers pending for this request |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2618 | **/ |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2619 | bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2620 | { |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2621 | return __blk_end_bidi_request(rq, error, nr_bytes, 0); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2622 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2623 | EXPORT_SYMBOL(__blk_end_request); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2624 | |
| 2625 | /** |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2626 | * __blk_end_request_all - Helper function for drives to finish the request. |
| 2627 | * @rq: the request to finish |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2628 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 2629 | * |
| 2630 | * Description: |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2631 | * Completely finish @rq. Must be called with queue lock held. |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 2632 | */ |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2633 | void __blk_end_request_all(struct request *rq, int error) |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 2634 | { |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2635 | bool pending; |
| 2636 | unsigned int bidi_bytes = 0; |
| 2637 | |
| 2638 | if (unlikely(blk_bidi_rq(rq))) |
| 2639 | bidi_bytes = blk_rq_bytes(rq->next_rq); |
| 2640 | |
| 2641 | pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); |
| 2642 | BUG_ON(pending); |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 2643 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2644 | EXPORT_SYMBOL(__blk_end_request_all); |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 2645 | |
| 2646 | /** |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2647 | * __blk_end_request_cur - Helper function to finish the current request chunk. |
| 2648 | * @rq: the request to finish the current chunk for |
Randy Dunlap | 8ebf975 | 2009-06-11 20:00:41 -0700 | [diff] [blame] | 2649 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2650 | * |
| 2651 | * Description: |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2652 | * Complete the current consecutively mapped chunk from @rq. Must |
| 2653 | * be called with queue lock held. |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2654 | * |
| 2655 | * Return: |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2656 | * %false - we are done with this request |
| 2657 | * %true - still buffers pending for this request |
| 2658 | */ |
| 2659 | bool __blk_end_request_cur(struct request *rq, int error) |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2660 | { |
FUJITA Tomonori | b1f7449 | 2009-05-11 17:56:09 +0900 | [diff] [blame] | 2661 | return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2662 | } |
Jens Axboe | 56ad174 | 2009-07-28 22:11:24 +0200 | [diff] [blame] | 2663 | EXPORT_SYMBOL(__blk_end_request_cur); |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2664 | |
Tejun Heo | 80a761f | 2009-07-03 17:48:17 +0900 | [diff] [blame] | 2665 | /** |
| 2666 | * __blk_end_request_err - Finish a request till the next failure boundary. |
| 2667 | * @rq: the request to finish till the next failure boundary for |
| 2668 | * @error: must be negative errno |
| 2669 | * |
| 2670 | * Description: |
| 2671 | * Complete @rq till the next failure boundary. Must be called |
| 2672 | * with queue lock held. |
| 2673 | * |
| 2674 | * Return: |
| 2675 | * %false - we are done with this request |
| 2676 | * %true - still buffers pending for this request |
| 2677 | */ |
| 2678 | bool __blk_end_request_err(struct request *rq, int error) |
| 2679 | { |
| 2680 | WARN_ON(error >= 0); |
| 2681 | return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); |
| 2682 | } |
| 2683 | EXPORT_SYMBOL_GPL(__blk_end_request_err); |
| 2684 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 2685 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
| 2686 | struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2687 | { |
Tejun Heo | a82afdf | 2009-07-03 17:48:16 +0900 | [diff] [blame] | 2688 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 2689 | rq->cmd_flags |= bio->bi_rw & REQ_WRITE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2690 | |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 2691 | if (bio_has_data(bio)) { |
| 2692 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 2693 | rq->buffer = bio_data(bio); |
| 2694 | } |
Tejun Heo | a2dec7b | 2009-05-07 22:24:44 +0900 | [diff] [blame] | 2695 | rq->__data_len = bio->bi_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2696 | rq->bio = rq->biotail = bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2697 | |
NeilBrown | 6684657 | 2007-08-16 13:31:28 +0200 | [diff] [blame] | 2698 | if (bio->bi_bdev) |
| 2699 | rq->rq_disk = bio->bi_bdev->bd_disk; |
| 2700 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2701 | |
Ilya Loginov | 2d4dc89 | 2009-11-26 09:16:19 +0100 | [diff] [blame] | 2702 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
| 2703 | /** |
| 2704 | * rq_flush_dcache_pages - Helper function to flush all pages in a request |
| 2705 | * @rq: the request to be flushed |
| 2706 | * |
| 2707 | * Description: |
| 2708 | * Flush all pages in @rq. |
| 2709 | */ |
| 2710 | void rq_flush_dcache_pages(struct request *rq) |
| 2711 | { |
| 2712 | struct req_iterator iter; |
| 2713 | struct bio_vec *bvec; |
| 2714 | |
| 2715 | rq_for_each_segment(bvec, rq, iter) |
| 2716 | flush_dcache_page(bvec->bv_page); |
| 2717 | } |
| 2718 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); |
| 2719 | #endif |
| 2720 | |
Kiyoshi Ueda | ef9e3fa | 2008-10-01 16:12:15 +0200 | [diff] [blame] | 2721 | /** |
| 2722 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy |
| 2723 | * @q : the queue of the device being checked |
| 2724 | * |
| 2725 | * Description: |
| 2726 | * Check if underlying low-level drivers of a device are busy. |
| 2727 | * If the drivers want to export their busy state, they must set own |
| 2728 | * exporting function using blk_queue_lld_busy() first. |
| 2729 | * |
| 2730 | * Basically, this function is used only by request stacking drivers |
| 2731 | * to stop dispatching requests to underlying devices when underlying |
| 2732 | * devices are busy. This behavior helps more I/O merging on the queue |
| 2733 | * of the request stacking driver and prevents I/O throughput regression |
| 2734 | * on burst I/O load. |
| 2735 | * |
| 2736 | * Return: |
| 2737 | * 0 - Not busy (The request stacking driver should dispatch request) |
| 2738 | * 1 - Busy (The request stacking driver should stop dispatching request) |
| 2739 | */ |
| 2740 | int blk_lld_busy(struct request_queue *q) |
| 2741 | { |
| 2742 | if (q->lld_busy_fn) |
| 2743 | return q->lld_busy_fn(q); |
| 2744 | |
| 2745 | return 0; |
| 2746 | } |
| 2747 | EXPORT_SYMBOL_GPL(blk_lld_busy); |
| 2748 | |
Kiyoshi Ueda | b0fd271 | 2009-06-11 13:10:16 +0200 | [diff] [blame] | 2749 | /** |
| 2750 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request |
| 2751 | * @rq: the clone request to be cleaned up |
| 2752 | * |
| 2753 | * Description: |
| 2754 | * Free all bios in @rq for a cloned request. |
| 2755 | */ |
| 2756 | void blk_rq_unprep_clone(struct request *rq) |
| 2757 | { |
| 2758 | struct bio *bio; |
| 2759 | |
| 2760 | while ((bio = rq->bio) != NULL) { |
| 2761 | rq->bio = bio->bi_next; |
| 2762 | |
| 2763 | bio_put(bio); |
| 2764 | } |
| 2765 | } |
| 2766 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); |
| 2767 | |
| 2768 | /* |
| 2769 | * Copy attributes of the original request to the clone request. |
| 2770 | * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. |
| 2771 | */ |
| 2772 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) |
| 2773 | { |
| 2774 | dst->cpu = src->cpu; |
Tejun Heo | 3a2edd0 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 2775 | dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; |
Kiyoshi Ueda | b0fd271 | 2009-06-11 13:10:16 +0200 | [diff] [blame] | 2776 | dst->cmd_type = src->cmd_type; |
| 2777 | dst->__sector = blk_rq_pos(src); |
| 2778 | dst->__data_len = blk_rq_bytes(src); |
| 2779 | dst->nr_phys_segments = src->nr_phys_segments; |
| 2780 | dst->ioprio = src->ioprio; |
| 2781 | dst->extra_len = src->extra_len; |
| 2782 | } |
| 2783 | |
| 2784 | /** |
| 2785 | * blk_rq_prep_clone - Helper function to setup clone request |
| 2786 | * @rq: the request to be setup |
| 2787 | * @rq_src: original request to be cloned |
| 2788 | * @bs: bio_set that bios for clone are allocated from |
| 2789 | * @gfp_mask: memory allocation mask for bio |
| 2790 | * @bio_ctr: setup function to be called for each clone bio. |
| 2791 | * Returns %0 for success, non %0 for failure. |
| 2792 | * @data: private data to be passed to @bio_ctr |
| 2793 | * |
| 2794 | * Description: |
| 2795 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. |
| 2796 | * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) |
| 2797 | * are not copied, and copying such parts is the caller's responsibility. |
| 2798 | * Also, pages which the original bios are pointing to are not copied |
| 2799 | * and the cloned bios just point same pages. |
| 2800 | * So cloned bios must be completed before original bios, which means |
| 2801 | * the caller must complete @rq before @rq_src. |
| 2802 | */ |
| 2803 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, |
| 2804 | struct bio_set *bs, gfp_t gfp_mask, |
| 2805 | int (*bio_ctr)(struct bio *, struct bio *, void *), |
| 2806 | void *data) |
| 2807 | { |
| 2808 | struct bio *bio, *bio_src; |
| 2809 | |
| 2810 | if (!bs) |
| 2811 | bs = fs_bio_set; |
| 2812 | |
| 2813 | blk_rq_init(NULL, rq); |
| 2814 | |
| 2815 | __rq_for_each_bio(bio_src, rq_src) { |
Kent Overstreet | bf800ef | 2012-09-06 15:35:02 -0700 | [diff] [blame] | 2816 | bio = bio_clone_bioset(bio_src, gfp_mask, bs); |
Kiyoshi Ueda | b0fd271 | 2009-06-11 13:10:16 +0200 | [diff] [blame] | 2817 | if (!bio) |
| 2818 | goto free_and_out; |
| 2819 | |
Kiyoshi Ueda | b0fd271 | 2009-06-11 13:10:16 +0200 | [diff] [blame] | 2820 | if (bio_ctr && bio_ctr(bio, bio_src, data)) |
| 2821 | goto free_and_out; |
| 2822 | |
| 2823 | if (rq->bio) { |
| 2824 | rq->biotail->bi_next = bio; |
| 2825 | rq->biotail = bio; |
| 2826 | } else |
| 2827 | rq->bio = rq->biotail = bio; |
| 2828 | } |
| 2829 | |
| 2830 | __blk_rq_prep_clone(rq, rq_src); |
| 2831 | |
| 2832 | return 0; |
| 2833 | |
| 2834 | free_and_out: |
| 2835 | if (bio) |
Kent Overstreet | 4254bba | 2012-09-06 15:35:00 -0700 | [diff] [blame] | 2836 | bio_put(bio); |
Kiyoshi Ueda | b0fd271 | 2009-06-11 13:10:16 +0200 | [diff] [blame] | 2837 | blk_rq_unprep_clone(rq); |
| 2838 | |
| 2839 | return -ENOMEM; |
| 2840 | } |
| 2841 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); |
| 2842 | |
Jens Axboe | 18887ad | 2008-07-28 13:08:45 +0200 | [diff] [blame] | 2843 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2844 | { |
| 2845 | return queue_work(kblockd_workqueue, work); |
| 2846 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2847 | EXPORT_SYMBOL(kblockd_schedule_work); |
| 2848 | |
Vivek Goyal | e43473b | 2010-09-15 17:06:35 -0400 | [diff] [blame] | 2849 | int kblockd_schedule_delayed_work(struct request_queue *q, |
| 2850 | struct delayed_work *dwork, unsigned long delay) |
| 2851 | { |
| 2852 | return queue_delayed_work(kblockd_workqueue, dwork, delay); |
| 2853 | } |
| 2854 | EXPORT_SYMBOL(kblockd_schedule_delayed_work); |
| 2855 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2856 | #define PLUG_MAGIC 0x91827364 |
| 2857 | |
Suresh Jayaraman | 75df713 | 2011-09-21 10:00:16 +0200 | [diff] [blame] | 2858 | /** |
| 2859 | * blk_start_plug - initialize blk_plug and track it inside the task_struct |
| 2860 | * @plug: The &struct blk_plug that needs to be initialized |
| 2861 | * |
| 2862 | * Description: |
| 2863 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
| 2864 | * pending I/O should the task end up blocking between blk_start_plug() and |
| 2865 | * blk_finish_plug(). This is important from a performance perspective, but |
| 2866 | * also ensures that we don't deadlock. For instance, if the task is blocking |
| 2867 | * for a memory allocation, memory reclaim could end up wanting to free a |
| 2868 | * page belonging to that request that is currently residing in our private |
| 2869 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid |
| 2870 | * this kind of deadlock. |
| 2871 | */ |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2872 | void blk_start_plug(struct blk_plug *plug) |
| 2873 | { |
| 2874 | struct task_struct *tsk = current; |
| 2875 | |
| 2876 | plug->magic = PLUG_MAGIC; |
| 2877 | INIT_LIST_HEAD(&plug->list); |
NeilBrown | 048c937 | 2011-04-18 09:52:22 +0200 | [diff] [blame] | 2878 | INIT_LIST_HEAD(&plug->cb_list); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2879 | |
| 2880 | /* |
| 2881 | * If this is a nested plug, don't actually assign it. It will be |
| 2882 | * flushed on its own. |
| 2883 | */ |
| 2884 | if (!tsk->plug) { |
| 2885 | /* |
| 2886 | * Store ordering should not be needed here, since a potential |
| 2887 | * preempt will imply a full memory barrier |
| 2888 | */ |
| 2889 | tsk->plug = plug; |
| 2890 | } |
| 2891 | } |
| 2892 | EXPORT_SYMBOL(blk_start_plug); |
| 2893 | |
| 2894 | static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) |
| 2895 | { |
| 2896 | struct request *rqa = container_of(a, struct request, queuelist); |
| 2897 | struct request *rqb = container_of(b, struct request, queuelist); |
| 2898 | |
Jianpeng Ma | 975927b | 2012-10-25 21:58:17 +0200 | [diff] [blame] | 2899 | return !(rqa->q < rqb->q || |
| 2900 | (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2901 | } |
| 2902 | |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 2903 | /* |
| 2904 | * If 'from_schedule' is true, then postpone the dispatch of requests |
| 2905 | * until a safe kblockd context. We due this to avoid accidental big |
| 2906 | * additional stack usage in driver dispatch, in places where the originally |
| 2907 | * plugger did not intend it. |
| 2908 | */ |
Jens Axboe | f660378 | 2011-04-15 15:49:07 +0200 | [diff] [blame] | 2909 | static void queue_unplugged(struct request_queue *q, unsigned int depth, |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 2910 | bool from_schedule) |
Jens Axboe | 99e2259 | 2011-04-18 09:59:55 +0200 | [diff] [blame] | 2911 | __releases(q->queue_lock) |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 2912 | { |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 2913 | trace_block_unplug(q, depth, !from_schedule); |
Jens Axboe | 99e2259 | 2011-04-18 09:59:55 +0200 | [diff] [blame] | 2914 | |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 2915 | if (from_schedule) |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 2916 | blk_run_queue_async(q); |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 2917 | else |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 2918 | __blk_run_queue(q); |
Bart Van Assche | 7046057 | 2012-11-28 13:45:56 +0100 | [diff] [blame] | 2919 | spin_unlock(q->queue_lock); |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 2920 | } |
| 2921 | |
NeilBrown | 74018dc | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2922 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
NeilBrown | 048c937 | 2011-04-18 09:52:22 +0200 | [diff] [blame] | 2923 | { |
| 2924 | LIST_HEAD(callbacks); |
| 2925 | |
Shaohua Li | 2a7d555 | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2926 | while (!list_empty(&plug->cb_list)) { |
| 2927 | list_splice_init(&plug->cb_list, &callbacks); |
NeilBrown | 048c937 | 2011-04-18 09:52:22 +0200 | [diff] [blame] | 2928 | |
Shaohua Li | 2a7d555 | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2929 | while (!list_empty(&callbacks)) { |
| 2930 | struct blk_plug_cb *cb = list_first_entry(&callbacks, |
NeilBrown | 048c937 | 2011-04-18 09:52:22 +0200 | [diff] [blame] | 2931 | struct blk_plug_cb, |
| 2932 | list); |
Shaohua Li | 2a7d555 | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2933 | list_del(&cb->list); |
NeilBrown | 74018dc | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2934 | cb->callback(cb, from_schedule); |
Shaohua Li | 2a7d555 | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2935 | } |
NeilBrown | 048c937 | 2011-04-18 09:52:22 +0200 | [diff] [blame] | 2936 | } |
| 2937 | } |
| 2938 | |
NeilBrown | 9cbb175 | 2012-07-31 09:08:14 +0200 | [diff] [blame] | 2939 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
| 2940 | int size) |
| 2941 | { |
| 2942 | struct blk_plug *plug = current->plug; |
| 2943 | struct blk_plug_cb *cb; |
| 2944 | |
| 2945 | if (!plug) |
| 2946 | return NULL; |
| 2947 | |
| 2948 | list_for_each_entry(cb, &plug->cb_list, list) |
| 2949 | if (cb->callback == unplug && cb->data == data) |
| 2950 | return cb; |
| 2951 | |
| 2952 | /* Not currently on the callback list */ |
| 2953 | BUG_ON(size < sizeof(*cb)); |
| 2954 | cb = kzalloc(size, GFP_ATOMIC); |
| 2955 | if (cb) { |
| 2956 | cb->data = data; |
| 2957 | cb->callback = unplug; |
| 2958 | list_add(&cb->list, &plug->cb_list); |
| 2959 | } |
| 2960 | return cb; |
| 2961 | } |
| 2962 | EXPORT_SYMBOL(blk_check_plugged); |
| 2963 | |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 2964 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2965 | { |
| 2966 | struct request_queue *q; |
| 2967 | unsigned long flags; |
| 2968 | struct request *rq; |
NeilBrown | 109b812 | 2011-04-11 14:13:10 +0200 | [diff] [blame] | 2969 | LIST_HEAD(list); |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 2970 | unsigned int depth; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2971 | |
| 2972 | BUG_ON(plug->magic != PLUG_MAGIC); |
| 2973 | |
NeilBrown | 74018dc | 2012-07-31 09:08:15 +0200 | [diff] [blame] | 2974 | flush_plug_callbacks(plug, from_schedule); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2975 | if (list_empty(&plug->list)) |
| 2976 | return; |
| 2977 | |
NeilBrown | 109b812 | 2011-04-11 14:13:10 +0200 | [diff] [blame] | 2978 | list_splice_init(&plug->list, &list); |
| 2979 | |
Jianpeng Ma | 422765c | 2013-01-11 14:46:09 +0100 | [diff] [blame] | 2980 | list_sort(NULL, &list, plug_rq_cmp); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2981 | |
| 2982 | q = NULL; |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 2983 | depth = 0; |
Jens Axboe | 1881127 | 2011-04-12 10:11:24 +0200 | [diff] [blame] | 2984 | |
| 2985 | /* |
| 2986 | * Save and disable interrupts here, to avoid doing it for every |
| 2987 | * queue lock we have to take. |
| 2988 | */ |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2989 | local_irq_save(flags); |
NeilBrown | 109b812 | 2011-04-11 14:13:10 +0200 | [diff] [blame] | 2990 | while (!list_empty(&list)) { |
| 2991 | rq = list_entry_rq(list.next); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2992 | list_del_init(&rq->queuelist); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 2993 | BUG_ON(!rq->q); |
| 2994 | if (rq->q != q) { |
Jens Axboe | 99e2259 | 2011-04-18 09:59:55 +0200 | [diff] [blame] | 2995 | /* |
| 2996 | * This drops the queue lock |
| 2997 | */ |
| 2998 | if (q) |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 2999 | queue_unplugged(q, depth, from_schedule); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3000 | q = rq->q; |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 3001 | depth = 0; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3002 | spin_lock(q->queue_lock); |
| 3003 | } |
Tejun Heo | 8ba6143 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 3004 | |
| 3005 | /* |
| 3006 | * Short-circuit if @q is dead |
| 3007 | */ |
Bart Van Assche | 3f3299d | 2012-11-28 13:42:38 +0100 | [diff] [blame] | 3008 | if (unlikely(blk_queue_dying(q))) { |
Tejun Heo | 8ba6143 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 3009 | __blk_end_request_all(rq, -ENODEV); |
| 3010 | continue; |
| 3011 | } |
| 3012 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3013 | /* |
| 3014 | * rq is already accounted, so use raw insert |
| 3015 | */ |
Jens Axboe | 401a18e | 2011-03-25 16:57:52 +0100 | [diff] [blame] | 3016 | if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) |
| 3017 | __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); |
| 3018 | else |
| 3019 | __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); |
Jens Axboe | 94b5eb2 | 2011-04-12 10:12:19 +0200 | [diff] [blame] | 3020 | |
| 3021 | depth++; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3022 | } |
| 3023 | |
Jens Axboe | 99e2259 | 2011-04-18 09:59:55 +0200 | [diff] [blame] | 3024 | /* |
| 3025 | * This drops the queue lock |
| 3026 | */ |
| 3027 | if (q) |
Jens Axboe | 49cac01 | 2011-04-16 13:51:05 +0200 | [diff] [blame] | 3028 | queue_unplugged(q, depth, from_schedule); |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3029 | |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3030 | local_irq_restore(flags); |
| 3031 | } |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3032 | |
| 3033 | void blk_finish_plug(struct blk_plug *plug) |
| 3034 | { |
Jens Axboe | f660378 | 2011-04-15 15:49:07 +0200 | [diff] [blame] | 3035 | blk_flush_plug_list(plug, false); |
Christoph Hellwig | 88b996c | 2011-04-15 15:20:10 +0200 | [diff] [blame] | 3036 | |
| 3037 | if (plug == current->plug) |
| 3038 | current->plug = NULL; |
Jens Axboe | 73c1010 | 2011-03-08 13:19:51 +0100 | [diff] [blame] | 3039 | } |
| 3040 | EXPORT_SYMBOL(blk_finish_plug); |
| 3041 | |
Lin Ming | 6c95466 | 2013-03-23 11:42:26 +0800 | [diff] [blame] | 3042 | #ifdef CONFIG_PM_RUNTIME |
| 3043 | /** |
| 3044 | * blk_pm_runtime_init - Block layer runtime PM initialization routine |
| 3045 | * @q: the queue of the device |
| 3046 | * @dev: the device the queue belongs to |
| 3047 | * |
| 3048 | * Description: |
| 3049 | * Initialize runtime-PM-related fields for @q and start auto suspend for |
| 3050 | * @dev. Drivers that want to take advantage of request-based runtime PM |
| 3051 | * should call this function after @dev has been initialized, and its |
| 3052 | * request queue @q has been allocated, and runtime PM for it can not happen |
| 3053 | * yet(either due to disabled/forbidden or its usage_count > 0). In most |
| 3054 | * cases, driver should call this function before any I/O has taken place. |
| 3055 | * |
| 3056 | * This function takes care of setting up using auto suspend for the device, |
| 3057 | * the autosuspend delay is set to -1 to make runtime suspend impossible |
| 3058 | * until an updated value is either set by user or by driver. Drivers do |
| 3059 | * not need to touch other autosuspend settings. |
| 3060 | * |
| 3061 | * The block layer runtime PM is request based, so only works for drivers |
| 3062 | * that use request as their IO unit instead of those directly use bio's. |
| 3063 | */ |
| 3064 | void blk_pm_runtime_init(struct request_queue *q, struct device *dev) |
| 3065 | { |
| 3066 | q->dev = dev; |
| 3067 | q->rpm_status = RPM_ACTIVE; |
| 3068 | pm_runtime_set_autosuspend_delay(q->dev, -1); |
| 3069 | pm_runtime_use_autosuspend(q->dev); |
| 3070 | } |
| 3071 | EXPORT_SYMBOL(blk_pm_runtime_init); |
| 3072 | |
| 3073 | /** |
| 3074 | * blk_pre_runtime_suspend - Pre runtime suspend check |
| 3075 | * @q: the queue of the device |
| 3076 | * |
| 3077 | * Description: |
| 3078 | * This function will check if runtime suspend is allowed for the device |
| 3079 | * by examining if there are any requests pending in the queue. If there |
| 3080 | * are requests pending, the device can not be runtime suspended; otherwise, |
| 3081 | * the queue's status will be updated to SUSPENDING and the driver can |
| 3082 | * proceed to suspend the device. |
| 3083 | * |
| 3084 | * For the not allowed case, we mark last busy for the device so that |
| 3085 | * runtime PM core will try to autosuspend it some time later. |
| 3086 | * |
| 3087 | * This function should be called near the start of the device's |
| 3088 | * runtime_suspend callback. |
| 3089 | * |
| 3090 | * Return: |
| 3091 | * 0 - OK to runtime suspend the device |
| 3092 | * -EBUSY - Device should not be runtime suspended |
| 3093 | */ |
| 3094 | int blk_pre_runtime_suspend(struct request_queue *q) |
| 3095 | { |
| 3096 | int ret = 0; |
| 3097 | |
| 3098 | spin_lock_irq(q->queue_lock); |
| 3099 | if (q->nr_pending) { |
| 3100 | ret = -EBUSY; |
| 3101 | pm_runtime_mark_last_busy(q->dev); |
| 3102 | } else { |
| 3103 | q->rpm_status = RPM_SUSPENDING; |
| 3104 | } |
| 3105 | spin_unlock_irq(q->queue_lock); |
| 3106 | return ret; |
| 3107 | } |
| 3108 | EXPORT_SYMBOL(blk_pre_runtime_suspend); |
| 3109 | |
| 3110 | /** |
| 3111 | * blk_post_runtime_suspend - Post runtime suspend processing |
| 3112 | * @q: the queue of the device |
| 3113 | * @err: return value of the device's runtime_suspend function |
| 3114 | * |
| 3115 | * Description: |
| 3116 | * Update the queue's runtime status according to the return value of the |
| 3117 | * device's runtime suspend function and mark last busy for the device so |
| 3118 | * that PM core will try to auto suspend the device at a later time. |
| 3119 | * |
| 3120 | * This function should be called near the end of the device's |
| 3121 | * runtime_suspend callback. |
| 3122 | */ |
| 3123 | void blk_post_runtime_suspend(struct request_queue *q, int err) |
| 3124 | { |
| 3125 | spin_lock_irq(q->queue_lock); |
| 3126 | if (!err) { |
| 3127 | q->rpm_status = RPM_SUSPENDED; |
| 3128 | } else { |
| 3129 | q->rpm_status = RPM_ACTIVE; |
| 3130 | pm_runtime_mark_last_busy(q->dev); |
| 3131 | } |
| 3132 | spin_unlock_irq(q->queue_lock); |
| 3133 | } |
| 3134 | EXPORT_SYMBOL(blk_post_runtime_suspend); |
| 3135 | |
| 3136 | /** |
| 3137 | * blk_pre_runtime_resume - Pre runtime resume processing |
| 3138 | * @q: the queue of the device |
| 3139 | * |
| 3140 | * Description: |
| 3141 | * Update the queue's runtime status to RESUMING in preparation for the |
| 3142 | * runtime resume of the device. |
| 3143 | * |
| 3144 | * This function should be called near the start of the device's |
| 3145 | * runtime_resume callback. |
| 3146 | */ |
| 3147 | void blk_pre_runtime_resume(struct request_queue *q) |
| 3148 | { |
| 3149 | spin_lock_irq(q->queue_lock); |
| 3150 | q->rpm_status = RPM_RESUMING; |
| 3151 | spin_unlock_irq(q->queue_lock); |
| 3152 | } |
| 3153 | EXPORT_SYMBOL(blk_pre_runtime_resume); |
| 3154 | |
| 3155 | /** |
| 3156 | * blk_post_runtime_resume - Post runtime resume processing |
| 3157 | * @q: the queue of the device |
| 3158 | * @err: return value of the device's runtime_resume function |
| 3159 | * |
| 3160 | * Description: |
| 3161 | * Update the queue's runtime status according to the return value of the |
| 3162 | * device's runtime_resume function. If it is successfully resumed, process |
| 3163 | * the requests that are queued into the device's queue when it is resuming |
| 3164 | * and then mark last busy and initiate autosuspend for it. |
| 3165 | * |
| 3166 | * This function should be called near the end of the device's |
| 3167 | * runtime_resume callback. |
| 3168 | */ |
| 3169 | void blk_post_runtime_resume(struct request_queue *q, int err) |
| 3170 | { |
| 3171 | spin_lock_irq(q->queue_lock); |
| 3172 | if (!err) { |
| 3173 | q->rpm_status = RPM_ACTIVE; |
| 3174 | __blk_run_queue(q); |
| 3175 | pm_runtime_mark_last_busy(q->dev); |
Aaron Lu | c60855c | 2013-05-17 15:47:20 +0800 | [diff] [blame] | 3176 | pm_request_autosuspend(q->dev); |
Lin Ming | 6c95466 | 2013-03-23 11:42:26 +0800 | [diff] [blame] | 3177 | } else { |
| 3178 | q->rpm_status = RPM_SUSPENDED; |
| 3179 | } |
| 3180 | spin_unlock_irq(q->queue_lock); |
| 3181 | } |
| 3182 | EXPORT_SYMBOL(blk_post_runtime_resume); |
| 3183 | #endif |
| 3184 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3185 | int __init blk_dev_init(void) |
| 3186 | { |
Nikanth Karthikesan | 9eb55b0 | 2009-04-27 14:53:54 +0200 | [diff] [blame] | 3187 | BUILD_BUG_ON(__REQ_NR_BITS > 8 * |
| 3188 | sizeof(((struct request *)0)->cmd_flags)); |
| 3189 | |
Tejun Heo | 89b90be | 2011-01-03 15:01:47 +0100 | [diff] [blame] | 3190 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
| 3191 | kblockd_workqueue = alloc_workqueue("kblockd", |
Viresh Kumar | 695588f | 2013-04-24 17:12:56 +0530 | [diff] [blame] | 3192 | WQ_MEM_RECLAIM | WQ_HIGHPRI | |
| 3193 | WQ_POWER_EFFICIENT, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3194 | if (!kblockd_workqueue) |
| 3195 | panic("Failed to create kblockd\n"); |
| 3196 | |
| 3197 | request_cachep = kmem_cache_create("blkdev_requests", |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 3198 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3199 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 3200 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3201 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3202 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | return 0; |
| 3204 | } |