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> |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 29 | #include <linux/blktrace_api.h> |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 30 | #include <linux/fault-inject.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 32 | #include "blk.h" |
| 33 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 34 | static int __make_request(struct request_queue *q, struct bio *bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | |
| 36 | /* |
| 37 | * For the allocated request tables |
| 38 | */ |
Adrian Bunk | 5ece6c5 | 2008-02-18 13:45:51 +0100 | [diff] [blame] | 39 | static struct kmem_cache *request_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
| 41 | /* |
| 42 | * For queue allocation |
| 43 | */ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 44 | struct kmem_cache *blk_requestq_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | |
| 46 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | * Controlling structure to kblockd |
| 48 | */ |
Jens Axboe | ff856ba | 2006-01-09 16:02:34 +0100 | [diff] [blame] | 49 | static struct workqueue_struct *kblockd_workqueue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 51 | static void drive_stat_acct(struct request *rq, int new_io) |
| 52 | { |
Jens Axboe | 28f1370 | 2008-05-07 10:15:46 +0200 | [diff] [blame] | 53 | struct hd_struct *part; |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 54 | int rw = rq_data_dir(rq); |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 55 | int cpu; |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 56 | |
| 57 | if (!blk_fs_request(rq) || !rq->rq_disk) |
| 58 | return; |
| 59 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 60 | cpu = part_stat_lock(); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 61 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 62 | |
Jens Axboe | 28f1370 | 2008-05-07 10:15:46 +0200 | [diff] [blame] | 63 | if (!new_io) |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 64 | part_stat_inc(cpu, part, merges[rw]); |
Jens Axboe | 28f1370 | 2008-05-07 10:15:46 +0200 | [diff] [blame] | 65 | else { |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 66 | part_round_stats(cpu, part); |
| 67 | part_inc_in_flight(part); |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 68 | } |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 69 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 70 | part_stat_unlock(); |
Jens Axboe | 26b8256 | 2008-01-29 13:54:41 +0100 | [diff] [blame] | 71 | } |
| 72 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 73 | void blk_queue_congestion_threshold(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | { |
| 75 | int nr; |
| 76 | |
| 77 | nr = q->nr_requests - (q->nr_requests / 8) + 1; |
| 78 | if (nr > q->nr_requests) |
| 79 | nr = q->nr_requests; |
| 80 | q->nr_congestion_on = nr; |
| 81 | |
| 82 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; |
| 83 | if (nr < 1) |
| 84 | nr = 1; |
| 85 | q->nr_congestion_off = nr; |
| 86 | } |
| 87 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | /** |
| 89 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info |
| 90 | * @bdev: device |
| 91 | * |
| 92 | * Locates the passed device's request queue and returns the address of its |
| 93 | * backing_dev_info |
| 94 | * |
| 95 | * Will return NULL if the request queue cannot be located. |
| 96 | */ |
| 97 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) |
| 98 | { |
| 99 | struct backing_dev_info *ret = NULL; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 100 | struct request_queue *q = bdev_get_queue(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | |
| 102 | if (q) |
| 103 | ret = &q->backing_dev_info; |
| 104 | return ret; |
| 105 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
| 107 | |
FUJITA Tomonori | 2a4aa30 | 2008-04-29 09:54:36 +0200 | [diff] [blame] | 108 | void blk_rq_init(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | { |
FUJITA Tomonori | 1afb20f | 2008-04-25 12:26:28 +0200 | [diff] [blame] | 110 | memset(rq, 0, sizeof(*rq)); |
| 111 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | INIT_LIST_HEAD(&rq->queuelist); |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 113 | INIT_LIST_HEAD(&rq->timeout_list); |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 114 | rq->cpu = -1; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 115 | rq->q = q; |
| 116 | rq->sector = rq->hard_sector = (sector_t) -1; |
Jens Axboe | 2e662b6 | 2006-07-13 11:55:04 +0200 | [diff] [blame] | 117 | INIT_HLIST_NODE(&rq->hash); |
| 118 | RB_CLEAR_NODE(&rq->rb_node); |
FUJITA Tomonori | d7e3c32 | 2008-04-29 09:54:39 +0200 | [diff] [blame] | 119 | rq->cmd = rq->__cmd; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 120 | rq->tag = -1; |
Jens Axboe | 63a7138 | 2008-02-08 12:41:03 +0100 | [diff] [blame] | 121 | rq->ref_count = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | } |
FUJITA Tomonori | 2a4aa30 | 2008-04-29 09:54:36 +0200 | [diff] [blame] | 123 | EXPORT_SYMBOL(blk_rq_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 125 | static void req_bio_endio(struct request *rq, struct bio *bio, |
| 126 | unsigned int nbytes, int error) |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 127 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 128 | struct request_queue *q = rq->q; |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 129 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 130 | if (&q->bar_rq != rq) { |
| 131 | if (error) |
| 132 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
| 133 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
| 134 | error = -EIO; |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 135 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 136 | if (unlikely(nbytes > bio->bi_size)) { |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 137 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 138 | __func__, nbytes, bio->bi_size); |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 139 | nbytes = bio->bi_size; |
| 140 | } |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 141 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 142 | bio->bi_size -= nbytes; |
| 143 | bio->bi_sector += (nbytes >> 9); |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 144 | |
| 145 | if (bio_integrity(bio)) |
| 146 | bio_integrity_advance(bio, nbytes); |
| 147 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 148 | if (bio->bi_size == 0) |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 149 | bio_endio(bio, error); |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 150 | } else { |
| 151 | |
| 152 | /* |
| 153 | * Okay, this is the barrier request in progress, just |
| 154 | * record the error; |
| 155 | */ |
| 156 | if (error && !q->orderr) |
| 157 | q->orderr = error; |
| 158 | } |
Tejun Heo | 797e7db | 2006-01-06 09:51:03 +0100 | [diff] [blame] | 159 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | void blk_dump_rq_flags(struct request *rq, char *msg) |
| 162 | { |
| 163 | int bit; |
| 164 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 165 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 166 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
| 167 | rq->cmd_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 169 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
| 170 | (unsigned long long)rq->sector, |
| 171 | rq->nr_sectors, |
| 172 | rq->current_nr_sectors); |
| 173 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", |
| 174 | rq->bio, rq->biotail, |
| 175 | rq->buffer, rq->data, |
| 176 | rq->data_len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 178 | if (blk_pc_request(rq)) { |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 179 | printk(KERN_INFO " cdb: "); |
FUJITA Tomonori | d34c87e | 2008-04-29 14:37:52 +0200 | [diff] [blame] | 180 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | printk("%02x ", rq->cmd[bit]); |
| 182 | printk("\n"); |
| 183 | } |
| 184 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | EXPORT_SYMBOL(blk_dump_rq_flags); |
| 186 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | /* |
| 188 | * "plug" the device if there are no outstanding requests: this will |
| 189 | * force the transfer to start only after we have put all the requests |
| 190 | * on the list. |
| 191 | * |
| 192 | * This is called with interrupts off and no requests on the queue and |
| 193 | * with the queue lock held. |
| 194 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 195 | void blk_plug_device(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | { |
| 197 | WARN_ON(!irqs_disabled()); |
| 198 | |
| 199 | /* |
| 200 | * don't plug a stopped queue, it must be paired with blk_start_queue() |
| 201 | * which will restart the queueing |
| 202 | */ |
Coywolf Qi Hunt | 7daac49 | 2006-04-19 10:14:49 +0200 | [diff] [blame] | 203 | if (blk_queue_stopped(q)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | return; |
| 205 | |
Jens Axboe | e48ec69 | 2008-07-03 13:18:54 +0200 | [diff] [blame] | 206 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 208 | blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); |
| 209 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | EXPORT_SYMBOL(blk_plug_device); |
| 212 | |
Jens Axboe | 6c5e0c4 | 2008-08-01 20:31:32 +0200 | [diff] [blame] | 213 | /** |
| 214 | * blk_plug_device_unlocked - plug a device without queue lock held |
| 215 | * @q: The &struct request_queue to plug |
| 216 | * |
| 217 | * Description: |
| 218 | * Like @blk_plug_device(), but grabs the queue lock and disables |
| 219 | * interrupts. |
| 220 | **/ |
| 221 | void blk_plug_device_unlocked(struct request_queue *q) |
| 222 | { |
| 223 | unsigned long flags; |
| 224 | |
| 225 | spin_lock_irqsave(q->queue_lock, flags); |
| 226 | blk_plug_device(q); |
| 227 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 228 | } |
| 229 | EXPORT_SYMBOL(blk_plug_device_unlocked); |
| 230 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | /* |
| 232 | * remove the queue from the plugged list, if present. called with |
| 233 | * queue lock held and interrupts disabled. |
| 234 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 235 | int blk_remove_plug(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | { |
| 237 | WARN_ON(!irqs_disabled()); |
| 238 | |
Jens Axboe | e48ec69 | 2008-07-03 13:18:54 +0200 | [diff] [blame] | 239 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | return 0; |
| 241 | |
| 242 | del_timer(&q->unplug_timer); |
| 243 | return 1; |
| 244 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | EXPORT_SYMBOL(blk_remove_plug); |
| 246 | |
| 247 | /* |
| 248 | * remove the plug and let it rip.. |
| 249 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 250 | void __generic_unplug_device(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | { |
Coywolf Qi Hunt | 7daac49 | 2006-04-19 10:14:49 +0200 | [diff] [blame] | 252 | if (unlikely(blk_queue_stopped(q))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | return; |
| 254 | |
| 255 | if (!blk_remove_plug(q)) |
| 256 | return; |
| 257 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 258 | q->request_fn(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | } |
| 260 | EXPORT_SYMBOL(__generic_unplug_device); |
| 261 | |
| 262 | /** |
| 263 | * generic_unplug_device - fire a request queue |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 264 | * @q: The &struct request_queue in question |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | * |
| 266 | * Description: |
| 267 | * Linux uses plugging to build bigger requests queues before letting |
| 268 | * the device have at them. If a queue is plugged, the I/O scheduler |
| 269 | * is still adding and merging requests on the queue. Once the queue |
| 270 | * gets unplugged, the request_fn defined for the queue is invoked and |
| 271 | * transfers started. |
| 272 | **/ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 273 | void generic_unplug_device(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | { |
Jens Axboe | dbaf2c0 | 2008-05-07 09:48:17 +0200 | [diff] [blame] | 275 | if (blk_queue_plugged(q)) { |
| 276 | spin_lock_irq(q->queue_lock); |
| 277 | __generic_unplug_device(q); |
| 278 | spin_unlock_irq(q->queue_lock); |
| 279 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | } |
| 281 | EXPORT_SYMBOL(generic_unplug_device); |
| 282 | |
| 283 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, |
| 284 | struct page *page) |
| 285 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 286 | struct request_queue *q = bdi->unplug_io_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | |
Alan D. Brunelle | 2ad8b1e | 2007-11-07 14:26:56 -0500 | [diff] [blame] | 288 | blk_unplug(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | } |
| 290 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 291 | void blk_unplug_work(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 293 | struct request_queue *q = |
| 294 | container_of(work, struct request_queue, unplug_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 296 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
| 297 | q->rq.count[READ] + q->rq.count[WRITE]); |
| 298 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | q->unplug_fn(q); |
| 300 | } |
| 301 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 302 | void blk_unplug_timeout(unsigned long data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 304 | struct request_queue *q = (struct request_queue *)data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 306 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, |
| 307 | q->rq.count[READ] + q->rq.count[WRITE]); |
| 308 | |
Jens Axboe | 18887ad | 2008-07-28 13:08:45 +0200 | [diff] [blame] | 309 | kblockd_schedule_work(q, &q->unplug_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | } |
| 311 | |
Alan D. Brunelle | 2ad8b1e | 2007-11-07 14:26:56 -0500 | [diff] [blame] | 312 | void blk_unplug(struct request_queue *q) |
| 313 | { |
| 314 | /* |
| 315 | * devices don't necessarily have an ->unplug_fn defined |
| 316 | */ |
| 317 | if (q->unplug_fn) { |
| 318 | blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, |
| 319 | q->rq.count[READ] + q->rq.count[WRITE]); |
| 320 | |
| 321 | q->unplug_fn(q); |
| 322 | } |
| 323 | } |
| 324 | EXPORT_SYMBOL(blk_unplug); |
| 325 | |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 326 | static void blk_invoke_request_fn(struct request_queue *q) |
| 327 | { |
| 328 | /* |
| 329 | * one level of recursion is ok and is much faster than kicking |
| 330 | * the unplug handling |
| 331 | */ |
| 332 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { |
| 333 | q->request_fn(q); |
| 334 | queue_flag_clear(QUEUE_FLAG_REENTER, q); |
| 335 | } else { |
| 336 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); |
| 337 | kblockd_schedule_work(q, &q->unplug_work); |
| 338 | } |
| 339 | } |
| 340 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | /** |
| 342 | * blk_start_queue - restart a previously stopped queue |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 343 | * @q: The &struct request_queue in question |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | * |
| 345 | * Description: |
| 346 | * blk_start_queue() will clear the stop flag on the queue, and call |
| 347 | * the request_fn for the queue if it was in a stopped state when |
| 348 | * entered. Also see blk_stop_queue(). Queue lock must be held. |
| 349 | **/ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 350 | void blk_start_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | { |
Paolo 'Blaisorblade' Giarrusso | a038e25 | 2006-06-05 12:09:01 +0200 | [diff] [blame] | 352 | WARN_ON(!irqs_disabled()); |
| 353 | |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 354 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 355 | blk_invoke_request_fn(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | EXPORT_SYMBOL(blk_start_queue); |
| 358 | |
| 359 | /** |
| 360 | * blk_stop_queue - stop a queue |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 361 | * @q: The &struct request_queue in question |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | * |
| 363 | * Description: |
| 364 | * The Linux block layer assumes that a block driver will consume all |
| 365 | * entries on the request queue when the request_fn strategy is called. |
| 366 | * Often this will not happen, because of hardware limitations (queue |
| 367 | * depth settings). If a device driver gets a 'queue full' response, |
| 368 | * or if it simply chooses not to queue more I/O at one point, it can |
| 369 | * call this function to prevent the request_fn from being called until |
| 370 | * the driver has signalled it's ready to go again. This happens by calling |
| 371 | * blk_start_queue() to restart queue operations. Queue lock must be held. |
| 372 | **/ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 373 | void blk_stop_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | { |
| 375 | blk_remove_plug(q); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 376 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | } |
| 378 | EXPORT_SYMBOL(blk_stop_queue); |
| 379 | |
| 380 | /** |
| 381 | * blk_sync_queue - cancel any pending callbacks on a queue |
| 382 | * @q: the queue |
| 383 | * |
| 384 | * Description: |
| 385 | * The block layer may perform asynchronous callback activity |
| 386 | * on a queue, such as calling the unplug function after a timeout. |
| 387 | * A block device may call blk_sync_queue to ensure that any |
| 388 | * such activity is cancelled, thus allowing it to release resources |
Michael Opdenacker | 59c5159 | 2007-05-09 08:57:56 +0200 | [diff] [blame] | 389 | * that the callbacks might use. The caller must already have made sure |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | * that its ->make_request_fn will not re-add plugging prior to calling |
| 391 | * this function. |
| 392 | * |
| 393 | */ |
| 394 | void blk_sync_queue(struct request_queue *q) |
| 395 | { |
| 396 | del_timer_sync(&q->unplug_timer); |
Oleg Nesterov | abbeb88 | 2007-10-23 15:08:19 +0200 | [diff] [blame] | 397 | kblockd_flush_work(&q->unplug_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | } |
| 399 | EXPORT_SYMBOL(blk_sync_queue); |
| 400 | |
| 401 | /** |
| 402 | * blk_run_queue - run a single device queue |
| 403 | * @q: The queue to run |
| 404 | */ |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 405 | void __blk_run_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | blk_remove_plug(q); |
Jens Axboe | dac07ec | 2006-05-11 08:20:16 +0200 | [diff] [blame] | 408 | |
| 409 | /* |
| 410 | * Only recurse once to avoid overrunning the stack, let the unplug |
| 411 | * handling reinvoke the handler shortly if we already got there. |
| 412 | */ |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 413 | if (!elv_queue_empty(q)) |
| 414 | blk_invoke_request_fn(q); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 415 | } |
| 416 | EXPORT_SYMBOL(__blk_run_queue); |
Jens Axboe | dac07ec | 2006-05-11 08:20:16 +0200 | [diff] [blame] | 417 | |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 418 | /** |
| 419 | * blk_run_queue - run a single device queue |
| 420 | * @q: The queue to run |
| 421 | */ |
| 422 | void blk_run_queue(struct request_queue *q) |
| 423 | { |
| 424 | unsigned long flags; |
| 425 | |
| 426 | spin_lock_irqsave(q->queue_lock, flags); |
| 427 | __blk_run_queue(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 429 | } |
| 430 | EXPORT_SYMBOL(blk_run_queue); |
| 431 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 432 | void blk_put_queue(struct request_queue *q) |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 433 | { |
| 434 | kobject_put(&q->kobj); |
| 435 | } |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 436 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 437 | void blk_cleanup_queue(struct request_queue *q) |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 438 | { |
Jens Axboe | e3335de9 | 2008-09-18 09:22:54 -0700 | [diff] [blame] | 439 | /* |
| 440 | * We know we have process context here, so we can be a little |
| 441 | * cautious and ensure that pending block actions on this device |
| 442 | * are done before moving on. Going into this function, we should |
| 443 | * not have processes doing IO to this device. |
| 444 | */ |
| 445 | blk_sync_queue(q); |
| 446 | |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 447 | mutex_lock(&q->sysfs_lock); |
Nick Piggin | 75ad23b | 2008-04-29 14:48:33 +0200 | [diff] [blame] | 448 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 449 | mutex_unlock(&q->sysfs_lock); |
| 450 | |
| 451 | if (q->elevator) |
| 452 | elevator_exit(q->elevator); |
| 453 | |
| 454 | blk_put_queue(q); |
| 455 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | EXPORT_SYMBOL(blk_cleanup_queue); |
| 457 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 458 | static int blk_init_free_list(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | { |
| 460 | struct request_list *rl = &q->rq; |
| 461 | |
| 462 | rl->count[READ] = rl->count[WRITE] = 0; |
| 463 | rl->starved[READ] = rl->starved[WRITE] = 0; |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 464 | rl->elvpriv = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | init_waitqueue_head(&rl->wait[READ]); |
| 466 | init_waitqueue_head(&rl->wait[WRITE]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 468 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
| 469 | mempool_free_slab, request_cachep, q->node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | |
| 471 | if (!rl->rq_pool) |
| 472 | return -ENOMEM; |
| 473 | |
| 474 | return 0; |
| 475 | } |
| 476 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 477 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | { |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 479 | return blk_alloc_queue_node(gfp_mask, -1); |
| 480 | } |
| 481 | EXPORT_SYMBOL(blk_alloc_queue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 483 | 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] | 484 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 485 | struct request_queue *q; |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 486 | int err; |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 487 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 488 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 489 | gfp_mask | __GFP_ZERO, node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | if (!q) |
| 491 | return NULL; |
| 492 | |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 493 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
| 494 | q->backing_dev_info.unplug_io_data = q; |
| 495 | err = bdi_init(&q->backing_dev_info); |
| 496 | if (err) { |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 497 | kmem_cache_free(blk_requestq_cachep, q); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 498 | return NULL; |
| 499 | } |
| 500 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | init_timer(&q->unplug_timer); |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 502 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
| 503 | INIT_LIST_HEAD(&q->timeout_list); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 504 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 505 | kobject_init(&q->kobj, &blk_queue_ktype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 506 | |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 507 | mutex_init(&q->sysfs_lock); |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 508 | spin_lock_init(&q->__queue_lock); |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 509 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | return q; |
| 511 | } |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 512 | EXPORT_SYMBOL(blk_alloc_queue_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | |
| 514 | /** |
| 515 | * blk_init_queue - prepare a request queue for use with a block device |
| 516 | * @rfn: The function to be called to process requests that have been |
| 517 | * placed on the queue. |
| 518 | * @lock: Request queue spin lock |
| 519 | * |
| 520 | * Description: |
| 521 | * If a block device wishes to use the standard request handling procedures, |
| 522 | * which sorts requests and coalesces adjacent requests, then it must |
| 523 | * call blk_init_queue(). The function @rfn will be called when there |
| 524 | * are requests on the queue that need to be processed. If the device |
| 525 | * supports plugging, then @rfn may not be called immediately when requests |
| 526 | * are available on the queue, but may be called at some time later instead. |
| 527 | * Plugged queues are generally unplugged when a buffer belonging to one |
| 528 | * of the requests on the queue is needed, or due to memory pressure. |
| 529 | * |
| 530 | * @rfn is not required, or even expected, to remove all requests off the |
| 531 | * queue, but only as many as it can handle at a time. If it does leave |
| 532 | * requests on the queue, it is responsible for arranging that the requests |
| 533 | * get dealt with eventually. |
| 534 | * |
| 535 | * 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] | 536 | * request queue; this lock will be taken also from interrupt context, so irq |
| 537 | * disabling is needed for it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | * |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 539 | * Function returns a pointer to the initialized request queue, or %NULL if |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | * it didn't succeed. |
| 541 | * |
| 542 | * Note: |
| 543 | * blk_init_queue() must be paired with a blk_cleanup_queue() call |
| 544 | * when the block device is deactivated (such as at module unload). |
| 545 | **/ |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 546 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 547 | 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] | 548 | { |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 549 | return blk_init_queue_node(rfn, lock, -1); |
| 550 | } |
| 551 | EXPORT_SYMBOL(blk_init_queue); |
| 552 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 553 | struct request_queue * |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 554 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
| 555 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 556 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | |
| 558 | if (!q) |
| 559 | return NULL; |
| 560 | |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 561 | q->node = node_id; |
Al Viro | 8669aaf | 2006-03-18 13:50:00 -0500 | [diff] [blame] | 562 | if (blk_init_free_list(q)) { |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 563 | kmem_cache_free(blk_requestq_cachep, q); |
Al Viro | 8669aaf | 2006-03-18 13:50:00 -0500 | [diff] [blame] | 564 | return NULL; |
| 565 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | |
| 152587d | 2005-04-12 16:22:06 -0500 | [diff] [blame] | 567 | /* |
| 568 | * if caller didn't supply a lock, they get per-queue locking with |
| 569 | * our embedded lock |
| 570 | */ |
Neil Brown | e7e72bf | 2008-05-14 16:05:54 -0700 | [diff] [blame] | 571 | if (!lock) |
| 152587d | 2005-04-12 16:22:06 -0500 | [diff] [blame] | 572 | lock = &q->__queue_lock; |
| 152587d | 2005-04-12 16:22:06 -0500 | [diff] [blame] | 573 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | q->request_fn = rfn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | q->prep_rq_fn = NULL; |
| 576 | q->unplug_fn = generic_unplug_device; |
Kiyoshi Ueda | 4ee5eaf | 2008-09-18 10:46:13 -0400 | [diff] [blame] | 577 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER | |
| 578 | 1 << QUEUE_FLAG_STACKABLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | q->queue_lock = lock; |
| 580 | |
| 581 | blk_queue_segment_boundary(q, 0xffffffff); |
| 582 | |
| 583 | blk_queue_make_request(q, __make_request); |
| 584 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); |
| 585 | |
| 586 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); |
| 587 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
| 588 | |
Alan Stern | 44ec954 | 2007-02-20 11:01:57 -0500 | [diff] [blame] | 589 | q->sg_reserved_size = INT_MAX; |
| 590 | |
FUJITA Tomonori | abf5439 | 2008-08-16 14:10:05 +0900 | [diff] [blame] | 591 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
| 592 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | /* |
| 594 | * all done |
| 595 | */ |
| 596 | if (!elevator_init(q, NULL)) { |
| 597 | blk_queue_congestion_threshold(q); |
| 598 | return q; |
| 599 | } |
| 600 | |
Al Viro | 8669aaf | 2006-03-18 13:50:00 -0500 | [diff] [blame] | 601 | blk_put_queue(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 602 | return NULL; |
| 603 | } |
Christoph Lameter | 1946089 | 2005-06-23 00:08:19 -0700 | [diff] [blame] | 604 | EXPORT_SYMBOL(blk_init_queue_node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 606 | int blk_get_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | { |
Nick Piggin | fde6ad2 | 2005-06-23 00:08:53 -0700 | [diff] [blame] | 608 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
Al Viro | 483f4af | 2006-03-18 18:34:37 -0500 | [diff] [blame] | 609 | kobject_get(&q->kobj); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | return 0; |
| 611 | } |
| 612 | |
| 613 | return 1; |
| 614 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 616 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | { |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 618 | if (rq->cmd_flags & REQ_ELVPRIV) |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 619 | elv_put_request(q, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | mempool_free(rq, q->rq.rq_pool); |
| 621 | } |
| 622 | |
Jens Axboe | 1ea25ecb | 2006-07-18 22:24:11 +0200 | [diff] [blame] | 623 | static struct request * |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 624 | blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | { |
| 626 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); |
| 627 | |
| 628 | if (!rq) |
| 629 | return NULL; |
| 630 | |
FUJITA Tomonori | 2a4aa30 | 2008-04-29 09:54:36 +0200 | [diff] [blame] | 631 | blk_rq_init(q, rq); |
FUJITA Tomonori | 1afb20f | 2008-04-25 12:26:28 +0200 | [diff] [blame] | 632 | |
Jens Axboe | 49171e5 | 2006-08-10 08:59:11 +0200 | [diff] [blame] | 633 | rq->cmd_flags = rw | REQ_ALLOCED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 635 | if (priv) { |
Jens Axboe | cb78b28 | 2006-07-28 09:32:57 +0200 | [diff] [blame] | 636 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 637 | mempool_free(rq, q->rq.rq_pool); |
| 638 | return NULL; |
| 639 | } |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 640 | rq->cmd_flags |= REQ_ELVPRIV; |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 641 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 643 | return rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | } |
| 645 | |
| 646 | /* |
| 647 | * ioc_batching returns true if the ioc is a valid batching request and |
| 648 | * should be given priority access to a request. |
| 649 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 650 | 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] | 651 | { |
| 652 | if (!ioc) |
| 653 | return 0; |
| 654 | |
| 655 | /* |
| 656 | * Make sure the process is able to allocate at least 1 request |
| 657 | * even if the batch times out, otherwise we could theoretically |
| 658 | * lose wakeups. |
| 659 | */ |
| 660 | return ioc->nr_batch_requests == q->nr_batching || |
| 661 | (ioc->nr_batch_requests > 0 |
| 662 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); |
| 663 | } |
| 664 | |
| 665 | /* |
| 666 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This |
| 667 | * will cause the process to be a "batcher" on all queues in the system. This |
| 668 | * is the behaviour we want though - once it gets a wakeup it should be given |
| 669 | * a nice run. |
| 670 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 671 | 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] | 672 | { |
| 673 | if (!ioc || ioc_batching(q, ioc)) |
| 674 | return; |
| 675 | |
| 676 | ioc->nr_batch_requests = q->nr_batching; |
| 677 | ioc->last_waited = jiffies; |
| 678 | } |
| 679 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 680 | static void __freed_request(struct request_queue *q, int rw) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | { |
| 682 | struct request_list *rl = &q->rq; |
| 683 | |
| 684 | if (rl->count[rw] < queue_congestion_off_threshold(q)) |
Thomas Maier | 79e2de4 | 2006-10-19 23:28:15 -0700 | [diff] [blame] | 685 | blk_clear_queue_congested(q, rw); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | |
| 687 | if (rl->count[rw] + 1 <= q->nr_requests) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | if (waitqueue_active(&rl->wait[rw])) |
| 689 | wake_up(&rl->wait[rw]); |
| 690 | |
| 691 | blk_clear_queue_full(q, rw); |
| 692 | } |
| 693 | } |
| 694 | |
| 695 | /* |
| 696 | * A request has just been released. Account for it, update the full and |
| 697 | * congestion status, wake up any waiters. Called under q->queue_lock. |
| 698 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 699 | static void freed_request(struct request_queue *q, int rw, int priv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 700 | { |
| 701 | struct request_list *rl = &q->rq; |
| 702 | |
| 703 | rl->count[rw]--; |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 704 | if (priv) |
| 705 | rl->elvpriv--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | |
| 707 | __freed_request(q, rw); |
| 708 | |
| 709 | if (unlikely(rl->starved[rw ^ 1])) |
| 710 | __freed_request(q, rw ^ 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | } |
| 712 | |
| 713 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) |
| 714 | /* |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 715 | * Get a free request, queue_lock must be held. |
| 716 | * Returns NULL on failure, with queue_lock held. |
| 717 | * Returns !NULL on success, with queue_lock *not held*. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 719 | static struct request *get_request(struct request_queue *q, int rw_flags, |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 720 | struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | { |
| 722 | struct request *rq = NULL; |
| 723 | struct request_list *rl = &q->rq; |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 724 | struct io_context *ioc = NULL; |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 725 | const int rw = rw_flags & 0x01; |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 726 | int may_queue, priv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 728 | may_queue = elv_may_queue(q, rw_flags); |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 729 | if (may_queue == ELV_MQUEUE_NO) |
| 730 | goto rq_starved; |
| 731 | |
| 732 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { |
| 733 | if (rl->count[rw]+1 >= q->nr_requests) { |
Jens Axboe | b5deef9 | 2006-07-19 23:39:40 +0200 | [diff] [blame] | 734 | ioc = current_io_context(GFP_ATOMIC, q->node); |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 735 | /* |
| 736 | * The queue will fill after this allocation, so set |
| 737 | * it as full, and mark this process as "batching". |
| 738 | * This process will be allowed to complete a batch of |
| 739 | * requests, others will be blocked. |
| 740 | */ |
| 741 | if (!blk_queue_full(q, rw)) { |
| 742 | ioc_set_batching(q, ioc); |
| 743 | blk_set_queue_full(q, rw); |
| 744 | } else { |
| 745 | if (may_queue != ELV_MQUEUE_MUST |
| 746 | && !ioc_batching(q, ioc)) { |
| 747 | /* |
| 748 | * The queue is full and the allocating |
| 749 | * process is not a "batcher", and not |
| 750 | * exempted by the IO scheduler |
| 751 | */ |
| 752 | goto out; |
| 753 | } |
| 754 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | } |
Thomas Maier | 79e2de4 | 2006-10-19 23:28:15 -0700 | [diff] [blame] | 756 | blk_set_queue_congested(q, rw); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | } |
| 758 | |
Jens Axboe | 082cf69 | 2005-06-28 16:35:11 +0200 | [diff] [blame] | 759 | /* |
| 760 | * Only allow batching queuers to allocate up to 50% over the defined |
| 761 | * limit of requests, otherwise we could have thousands of requests |
| 762 | * allocated with any setting of ->nr_requests |
| 763 | */ |
Hugh Dickins | fd782a4 | 2005-06-29 15:15:40 +0100 | [diff] [blame] | 764 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
Jens Axboe | 082cf69 | 2005-06-28 16:35:11 +0200 | [diff] [blame] | 765 | goto out; |
Hugh Dickins | fd782a4 | 2005-06-29 15:15:40 +0100 | [diff] [blame] | 766 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | rl->count[rw]++; |
| 768 | rl->starved[rw] = 0; |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 769 | |
Jens Axboe | 64521d1 | 2005-10-28 08:30:39 +0200 | [diff] [blame] | 770 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 771 | if (priv) |
| 772 | rl->elvpriv++; |
| 773 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | spin_unlock_irq(q->queue_lock); |
| 775 | |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 776 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 777 | if (unlikely(!rq)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 778 | /* |
| 779 | * Allocation failed presumably due to memory. Undo anything |
| 780 | * we might have messed up. |
| 781 | * |
| 782 | * Allocating task should really be put onto the front of the |
| 783 | * wait queue, but this is pretty rare. |
| 784 | */ |
| 785 | spin_lock_irq(q->queue_lock); |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 786 | freed_request(q, rw, priv); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | |
| 788 | /* |
| 789 | * in the very unlikely event that allocation failed and no |
| 790 | * requests for this direction was pending, mark us starved |
| 791 | * so that freeing of a request in the other direction will |
| 792 | * notice us. another possible fix would be to split the |
| 793 | * rq mempool into READ and WRITE |
| 794 | */ |
| 795 | rq_starved: |
| 796 | if (unlikely(rl->count[rw] == 0)) |
| 797 | rl->starved[rw] = 1; |
| 798 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | goto out; |
| 800 | } |
| 801 | |
Jens Axboe | 88ee5ef | 2005-11-12 11:09:12 +0100 | [diff] [blame] | 802 | /* |
| 803 | * ioc may be NULL here, and ioc_batching will be false. That's |
| 804 | * OK, if the queue is under the request limit then requests need |
| 805 | * not count toward the nr_batch_requests limit. There will always |
| 806 | * be some limit enforced by BLK_BATCH_TIME. |
| 807 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | if (ioc_batching(q, ioc)) |
| 809 | ioc->nr_batch_requests--; |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 810 | |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 811 | blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | return rq; |
| 814 | } |
| 815 | |
| 816 | /* |
| 817 | * No available requests for this queue, unplug the device and wait for some |
| 818 | * requests to become available. |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 819 | * |
| 820 | * Called with q->queue_lock held, and returns with it unlocked. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 822 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 823 | struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 824 | { |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 825 | const int rw = rw_flags & 0x01; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | struct request *rq; |
| 827 | |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 828 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 829 | while (!rq) { |
| 830 | DEFINE_WAIT(wait); |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 831 | struct io_context *ioc; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | struct request_list *rl = &q->rq; |
| 833 | |
| 834 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, |
| 835 | TASK_UNINTERRUPTIBLE); |
| 836 | |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 837 | blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 839 | __generic_unplug_device(q); |
| 840 | spin_unlock_irq(q->queue_lock); |
| 841 | io_schedule(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 843 | /* |
| 844 | * After sleeping, we become a "batching" process and |
| 845 | * will be able to allocate at least one request, and |
| 846 | * up to a big batch of them for a small period time. |
| 847 | * See ioc_batching, ioc_set_batching |
| 848 | */ |
| 849 | ioc = current_io_context(GFP_NOIO, q->node); |
| 850 | ioc_set_batching(q, ioc); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 851 | |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 852 | spin_lock_irq(q->queue_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 853 | finish_wait(&rl->wait[rw], &wait); |
Zhang, Yanmin | 05caf8d | 2008-05-22 15:13:29 +0200 | [diff] [blame] | 854 | |
| 855 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
| 856 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | |
| 858 | return rq; |
| 859 | } |
| 860 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 861 | 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] | 862 | { |
| 863 | struct request *rq; |
| 864 | |
| 865 | BUG_ON(rw != READ && rw != WRITE); |
| 866 | |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 867 | spin_lock_irq(q->queue_lock); |
| 868 | if (gfp_mask & __GFP_WAIT) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 869 | rq = get_request_wait(q, rw, NULL); |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 870 | } else { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 871 | rq = get_request(q, rw, NULL, gfp_mask); |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 872 | if (!rq) |
| 873 | spin_unlock_irq(q->queue_lock); |
| 874 | } |
| 875 | /* q->queue_lock is unlocked at this point */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | |
| 877 | return rq; |
| 878 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 879 | EXPORT_SYMBOL(blk_get_request); |
| 880 | |
| 881 | /** |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 882 | * blk_start_queueing - initiate dispatch of requests to device |
| 883 | * @q: request queue to kick into gear |
| 884 | * |
| 885 | * This is basically a helper to remove the need to know whether a queue |
| 886 | * is plugged or not if someone just wants to initiate dispatch of requests |
| 887 | * for this queue. |
| 888 | * |
| 889 | * The queue lock must be held with interrupts disabled. |
| 890 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 891 | void blk_start_queueing(struct request_queue *q) |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 892 | { |
Elias Oltmanns | 336c3d8 | 2008-10-01 16:02:33 +0200 | [diff] [blame] | 893 | if (!blk_queue_plugged(q)) { |
| 894 | if (unlikely(blk_queue_stopped(q))) |
| 895 | return; |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 896 | q->request_fn(q); |
Elias Oltmanns | 336c3d8 | 2008-10-01 16:02:33 +0200 | [diff] [blame] | 897 | } else |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 898 | __generic_unplug_device(q); |
| 899 | } |
| 900 | EXPORT_SYMBOL(blk_start_queueing); |
| 901 | |
| 902 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | * blk_requeue_request - put a request back on queue |
| 904 | * @q: request queue where request should be inserted |
| 905 | * @rq: request to be inserted |
| 906 | * |
| 907 | * Description: |
| 908 | * Drivers often keep queueing requests until the hardware cannot accept |
| 909 | * more, when that condition happens we need to put the request back |
| 910 | * on the queue. Must be called with queue lock held. |
| 911 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 912 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 913 | { |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 914 | blk_delete_timer(rq); |
| 915 | blk_clear_rq_complete(rq); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 916 | blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); |
| 917 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | if (blk_rq_tagged(rq)) |
| 919 | blk_queue_end_tag(q, rq); |
| 920 | |
| 921 | elv_requeue_request(q, rq); |
| 922 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | EXPORT_SYMBOL(blk_requeue_request); |
| 924 | |
| 925 | /** |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 926 | * blk_insert_request - insert a special request into a request queue |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | * @q: request queue where request should be inserted |
| 928 | * @rq: request to be inserted |
| 929 | * @at_head: insert request at head or tail of queue |
| 930 | * @data: private data |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 931 | * |
| 932 | * Description: |
| 933 | * Many block devices need to execute commands asynchronously, so they don't |
| 934 | * block the whole kernel from preemption during request execution. This is |
| 935 | * accomplished normally by inserting aritficial requests tagged as |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 936 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
| 937 | * be scheduled for actual execution by the request queue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | * |
| 939 | * We have the option of inserting the head or the tail of the queue. |
| 940 | * Typically we use the tail for new ioctls and so forth. We use the head |
| 941 | * of the queue for things like a QUEUE_FULL message from a device, or a |
| 942 | * host that is unable to accept a particular command. |
| 943 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 944 | void blk_insert_request(struct request_queue *q, struct request *rq, |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 945 | int at_head, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | { |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 947 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | unsigned long flags; |
| 949 | |
| 950 | /* |
| 951 | * tell I/O scheduler that this isn't a regular read/write (ie it |
| 952 | * must not attempt merges on this) and that it acts as a soft |
| 953 | * barrier |
| 954 | */ |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 955 | rq->cmd_type = REQ_TYPE_SPECIAL; |
| 956 | rq->cmd_flags |= REQ_SOFTBARRIER; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | |
| 958 | rq->special = data; |
| 959 | |
| 960 | spin_lock_irqsave(q->queue_lock, flags); |
| 961 | |
| 962 | /* |
| 963 | * If command is tagged, release the tag |
| 964 | */ |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 965 | if (blk_rq_tagged(rq)) |
| 966 | blk_queue_end_tag(q, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | |
Jerome Marchand | b238b3d | 2007-10-23 15:05:46 +0200 | [diff] [blame] | 968 | drive_stat_acct(rq, 1); |
Tejun Heo | 867d119 | 2005-04-24 02:06:05 -0500 | [diff] [blame] | 969 | __elv_add_request(q, rq, where, 0); |
Jens Axboe | dc72ef4 | 2006-07-20 14:54:05 +0200 | [diff] [blame] | 970 | blk_start_queueing(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 972 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | EXPORT_SYMBOL(blk_insert_request); |
| 974 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 975 | /* |
| 976 | * add-request adds a request to the linked list. |
| 977 | * queue lock is held and interrupts disabled, as we muck with the |
| 978 | * request queue list. |
| 979 | */ |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 980 | static inline void add_request(struct request_queue *q, struct request *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 981 | { |
Jerome Marchand | b238b3d | 2007-10-23 15:05:46 +0200 | [diff] [blame] | 982 | drive_stat_acct(req, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 983 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | /* |
| 985 | * elevator indicated where it wants this request to be |
| 986 | * inserted at elevator_merge time |
| 987 | */ |
| 988 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); |
| 989 | } |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 990 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 991 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
| 992 | unsigned long now) |
| 993 | { |
| 994 | if (now == part->stamp) |
| 995 | return; |
| 996 | |
| 997 | if (part->in_flight) { |
| 998 | __part_stat_add(cpu, part, time_in_queue, |
| 999 | part->in_flight * (now - part->stamp)); |
| 1000 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); |
| 1001 | } |
| 1002 | part->stamp = now; |
| 1003 | } |
| 1004 | |
| 1005 | /** |
| 1006 | * part_round_stats() - Round off the performance stats on a struct |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | * disk_stats. |
| 1008 | * |
| 1009 | * The average IO queue length and utilisation statistics are maintained |
| 1010 | * by observing the current state of the queue length and the amount of |
| 1011 | * time it has been in this state for. |
| 1012 | * |
| 1013 | * Normally, that accounting is done on IO completion, but that can result |
| 1014 | * in more than a second's worth of IO being accounted for within any one |
| 1015 | * second, leading to >100% utilisation. To deal with that, we call this |
| 1016 | * function to do a round-off before returning the results when reading |
| 1017 | * /proc/diskstats. This accounts immediately for all queue usage up to |
| 1018 | * the current jiffies and restarts the counters again. |
| 1019 | */ |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 1020 | void part_round_stats(int cpu, struct hd_struct *part) |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1021 | { |
| 1022 | unsigned long now = jiffies; |
| 1023 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1024 | if (part->partno) |
| 1025 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); |
| 1026 | part_round_stats_single(cpu, part, now); |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1027 | } |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1028 | EXPORT_SYMBOL_GPL(part_round_stats); |
Jerome Marchand | 6f2576a | 2008-02-08 11:04:35 +0100 | [diff] [blame] | 1029 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1030 | /* |
| 1031 | * queue lock must be held |
| 1032 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1033 | void __blk_put_request(struct request_queue *q, struct request *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | if (unlikely(!q)) |
| 1036 | return; |
| 1037 | if (unlikely(--req->ref_count)) |
| 1038 | return; |
| 1039 | |
Tejun Heo | 8922e16 | 2005-10-20 16:23:44 +0200 | [diff] [blame] | 1040 | elv_completed_request(q, req); |
| 1041 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1042 | /* |
| 1043 | * Request may not have originated from ll_rw_blk. if not, |
| 1044 | * it didn't come out of our reserved rq pools |
| 1045 | */ |
Jens Axboe | 49171e5 | 2006-08-10 08:59:11 +0200 | [diff] [blame] | 1046 | if (req->cmd_flags & REQ_ALLOCED) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1047 | int rw = rq_data_dir(req); |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 1048 | int priv = req->cmd_flags & REQ_ELVPRIV; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1050 | BUG_ON(!list_empty(&req->queuelist)); |
Jens Axboe | 9817064 | 2006-07-28 09:23:08 +0200 | [diff] [blame] | 1051 | BUG_ON(!hlist_unhashed(&req->hash)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | |
| 1053 | blk_free_request(q, req); |
Tejun Heo | cb98fc8 | 2005-10-28 08:29:39 +0200 | [diff] [blame] | 1054 | freed_request(q, rw, priv); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | } |
| 1056 | } |
Mike Christie | 6e39b69e | 2005-11-11 05:30:24 -0600 | [diff] [blame] | 1057 | EXPORT_SYMBOL_GPL(__blk_put_request); |
| 1058 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | void blk_put_request(struct request *req) |
| 1060 | { |
Tejun Heo | 8922e16 | 2005-10-20 16:23:44 +0200 | [diff] [blame] | 1061 | unsigned long flags; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1062 | struct request_queue *q = req->q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1063 | |
FUJITA Tomonori | 52a93ba | 2008-07-15 21:21:45 +0200 | [diff] [blame] | 1064 | spin_lock_irqsave(q->queue_lock, flags); |
| 1065 | __blk_put_request(q, req); |
| 1066 | spin_unlock_irqrestore(q->queue_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1067 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | EXPORT_SYMBOL(blk_put_request); |
| 1069 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 1070 | void init_request_from_bio(struct request *req, struct bio *bio) |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1071 | { |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 1072 | req->cpu = bio->bi_comp_cpu; |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 1073 | req->cmd_type = REQ_TYPE_FS; |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1074 | |
| 1075 | /* |
| 1076 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) |
| 1077 | */ |
| 1078 | if (bio_rw_ahead(bio) || bio_failfast(bio)) |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 1079 | req->cmd_flags |= REQ_FAILFAST; |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1080 | |
| 1081 | /* |
| 1082 | * REQ_BARRIER implies no merging, but lets make it explicit |
| 1083 | */ |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 1084 | if (unlikely(bio_discard(bio))) { |
David Woodhouse | e17fc0a | 2008-08-09 16:42:20 +0100 | [diff] [blame] | 1085 | req->cmd_flags |= REQ_DISCARD; |
| 1086 | if (bio_barrier(bio)) |
| 1087 | req->cmd_flags |= REQ_SOFTBARRIER; |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 1088 | req->q->prepare_discard_fn(req->q, req); |
David Woodhouse | e17fc0a | 2008-08-09 16:42:20 +0100 | [diff] [blame] | 1089 | } else if (unlikely(bio_barrier(bio))) |
| 1090 | req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1091 | |
Jens Axboe | b31dc66 | 2006-06-13 08:26:10 +0200 | [diff] [blame] | 1092 | if (bio_sync(bio)) |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 1093 | req->cmd_flags |= REQ_RW_SYNC; |
Jens Axboe | 5404bc7 | 2006-08-10 09:01:02 +0200 | [diff] [blame] | 1094 | if (bio_rw_meta(bio)) |
| 1095 | req->cmd_flags |= REQ_RW_META; |
Jens Axboe | b31dc66 | 2006-06-13 08:26:10 +0200 | [diff] [blame] | 1096 | |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1097 | req->errors = 0; |
| 1098 | req->hard_sector = req->sector = bio->bi_sector; |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1099 | req->ioprio = bio_prio(bio); |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1100 | req->start_time = jiffies; |
NeilBrown | bc1c56f | 2007-08-16 13:31:30 +0200 | [diff] [blame] | 1101 | blk_rq_bio_prep(req->q, req, bio); |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1102 | } |
| 1103 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1104 | static int __make_request(struct request_queue *q, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1105 | { |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1106 | struct request *req; |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 1107 | int el_ret, nr_sectors, barrier, discard, err; |
Jens Axboe | 51da90f | 2006-07-18 04:14:45 +0200 | [diff] [blame] | 1108 | const unsigned short prio = bio_prio(bio); |
| 1109 | const int sync = bio_sync(bio); |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 1110 | int rw_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1112 | nr_sectors = bio_sectors(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | |
| 1114 | /* |
| 1115 | * low level driver can indicate that it wants pages above a |
| 1116 | * certain limit bounced to low memory (ie for highmem, or even |
| 1117 | * ISA dma in theory) |
| 1118 | */ |
| 1119 | blk_queue_bounce(q, &bio); |
| 1120 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | barrier = bio_barrier(bio); |
David Woodhouse | e17fc0a | 2008-08-09 16:42:20 +0100 | [diff] [blame] | 1122 | if (unlikely(barrier) && bio_has_data(bio) && |
| 1123 | (q->next_ordered == QUEUE_ORDERED_NONE)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | err = -EOPNOTSUPP; |
| 1125 | goto end_io; |
| 1126 | } |
| 1127 | |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 1128 | discard = bio_discard(bio); |
| 1129 | if (unlikely(discard) && !q->prepare_discard_fn) { |
| 1130 | err = -EOPNOTSUPP; |
| 1131 | goto end_io; |
| 1132 | } |
| 1133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | spin_lock_irq(q->queue_lock); |
| 1135 | |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1136 | if (unlikely(barrier) || elv_queue_empty(q)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | goto get_rq; |
| 1138 | |
| 1139 | el_ret = elv_merge(q, &req, bio); |
| 1140 | switch (el_ret) { |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1141 | case ELEVATOR_BACK_MERGE: |
| 1142 | BUG_ON(!rq_mergeable(req)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1144 | if (!ll_back_merge_fn(q, req, bio)) |
| 1145 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1147 | blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1148 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1149 | req->biotail->bi_next = bio; |
| 1150 | req->biotail = bio; |
| 1151 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
| 1152 | req->ioprio = ioprio_best(req->ioprio, prio); |
Jens Axboe | ab780f1 | 2008-08-26 10:25:02 +0200 | [diff] [blame] | 1153 | if (!blk_rq_cpu_valid(req)) |
| 1154 | req->cpu = bio->bi_comp_cpu; |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1155 | drive_stat_acct(req, 0); |
| 1156 | if (!attempt_back_merge(q, req)) |
| 1157 | elv_merged_request(q, req, el_ret); |
| 1158 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1160 | case ELEVATOR_FRONT_MERGE: |
| 1161 | BUG_ON(!rq_mergeable(req)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1163 | if (!ll_front_merge_fn(q, req, bio)) |
| 1164 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1166 | blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1167 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1168 | bio->bi_next = req->bio; |
| 1169 | req->bio = bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1170 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1171 | /* |
| 1172 | * may not be valid. if the low level driver said |
| 1173 | * it didn't need a bounce buffer then it better |
| 1174 | * not touch req->buffer either... |
| 1175 | */ |
| 1176 | req->buffer = bio_data(bio); |
| 1177 | req->current_nr_sectors = bio_cur_sectors(bio); |
| 1178 | req->hard_cur_sectors = req->current_nr_sectors; |
| 1179 | req->sector = req->hard_sector = bio->bi_sector; |
| 1180 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
| 1181 | req->ioprio = ioprio_best(req->ioprio, prio); |
Jens Axboe | ab780f1 | 2008-08-26 10:25:02 +0200 | [diff] [blame] | 1182 | if (!blk_rq_cpu_valid(req)) |
| 1183 | req->cpu = bio->bi_comp_cpu; |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1184 | drive_stat_acct(req, 0); |
| 1185 | if (!attempt_front_merge(q, req)) |
| 1186 | elv_merged_request(q, req, el_ret); |
| 1187 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1189 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
| 1190 | default: |
| 1191 | ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1192 | } |
| 1193 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1194 | get_rq: |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1195 | /* |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 1196 | * This sync check and mask will be re-done in init_request_from_bio(), |
| 1197 | * but we need to set it earlier to expose the sync flag to the |
| 1198 | * rq allocator and io schedulers. |
| 1199 | */ |
| 1200 | rw_flags = bio_data_dir(bio); |
| 1201 | if (sync) |
| 1202 | rw_flags |= REQ_RW_SYNC; |
| 1203 | |
| 1204 | /* |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1205 | * Grab a free request. This is might sleep but can not fail. |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1206 | * Returns with the queue unlocked. |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1207 | */ |
Jens Axboe | 7749a8d | 2006-12-13 13:02:26 +0100 | [diff] [blame] | 1208 | req = get_request_wait(q, rw_flags, bio); |
Nick Piggin | d634453 | 2005-06-28 20:45:14 -0700 | [diff] [blame] | 1209 | |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1210 | /* |
| 1211 | * After dropping the lock and possibly sleeping here, our request |
| 1212 | * may now be mergeable after it had proven unmergeable (above). |
| 1213 | * We don't worry about that case for efficiency. It won't happen |
| 1214 | * often, and the elevators are able to handle it. |
| 1215 | */ |
Tejun Heo | 52d9e67 | 2006-01-06 09:49:58 +0100 | [diff] [blame] | 1216 | init_request_from_bio(req, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1217 | |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1218 | spin_lock_irq(q->queue_lock); |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 1219 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
| 1220 | bio_flagged(bio, BIO_CPU_AFFINE)) |
| 1221 | req->cpu = blk_cpu_to_group(smp_processor_id()); |
Nick Piggin | 450991b | 2005-06-28 20:45:13 -0700 | [diff] [blame] | 1222 | if (elv_queue_empty(q)) |
| 1223 | blk_plug_device(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | add_request(q, req); |
| 1225 | out: |
Jens Axboe | 4a534f9 | 2005-04-16 15:25:40 -0700 | [diff] [blame] | 1226 | if (sync) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1227 | __generic_unplug_device(q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1228 | spin_unlock_irq(q->queue_lock); |
| 1229 | return 0; |
| 1230 | |
| 1231 | end_io: |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1232 | bio_endio(bio, err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | return 0; |
| 1234 | } |
| 1235 | |
| 1236 | /* |
| 1237 | * If bio->bi_dev is a partition, remap the location |
| 1238 | */ |
| 1239 | static inline void blk_partition_remap(struct bio *bio) |
| 1240 | { |
| 1241 | struct block_device *bdev = bio->bi_bdev; |
| 1242 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1243 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1244 | struct hd_struct *p = bdev->bd_part; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1245 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | bio->bi_sector += p->start_sect; |
| 1247 | bio->bi_bdev = bdev->bd_contains; |
Alan D. Brunelle | c7149d6 | 2007-08-07 15:30:23 +0200 | [diff] [blame] | 1248 | |
| 1249 | blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio, |
| 1250 | bdev->bd_dev, bio->bi_sector, |
| 1251 | bio->bi_sector - p->start_sect); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | } |
| 1253 | } |
| 1254 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | static void handle_bad_sector(struct bio *bio) |
| 1256 | { |
| 1257 | char b[BDEVNAME_SIZE]; |
| 1258 | |
| 1259 | printk(KERN_INFO "attempt to access beyond end of device\n"); |
| 1260 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", |
| 1261 | bdevname(bio->bi_bdev, b), |
| 1262 | bio->bi_rw, |
| 1263 | (unsigned long long)bio->bi_sector + bio_sectors(bio), |
| 1264 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); |
| 1265 | |
| 1266 | set_bit(BIO_EOF, &bio->bi_flags); |
| 1267 | } |
| 1268 | |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1269 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
| 1270 | |
| 1271 | static DECLARE_FAULT_ATTR(fail_make_request); |
| 1272 | |
| 1273 | static int __init setup_fail_make_request(char *str) |
| 1274 | { |
| 1275 | return setup_fault_attr(&fail_make_request, str); |
| 1276 | } |
| 1277 | __setup("fail_make_request=", setup_fail_make_request); |
| 1278 | |
| 1279 | static int should_fail_request(struct bio *bio) |
| 1280 | { |
Tejun Heo | eddb2e2 | 2008-08-25 19:56:13 +0900 | [diff] [blame] | 1281 | struct hd_struct *part = bio->bi_bdev->bd_part; |
| 1282 | |
| 1283 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1284 | return should_fail(&fail_make_request, bio->bi_size); |
| 1285 | |
| 1286 | return 0; |
| 1287 | } |
| 1288 | |
| 1289 | static int __init fail_make_request_debugfs(void) |
| 1290 | { |
| 1291 | return init_fault_attr_dentries(&fail_make_request, |
| 1292 | "fail_make_request"); |
| 1293 | } |
| 1294 | |
| 1295 | late_initcall(fail_make_request_debugfs); |
| 1296 | |
| 1297 | #else /* CONFIG_FAIL_MAKE_REQUEST */ |
| 1298 | |
| 1299 | static inline int should_fail_request(struct bio *bio) |
| 1300 | { |
| 1301 | return 0; |
| 1302 | } |
| 1303 | |
| 1304 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
| 1305 | |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1306 | /* |
| 1307 | * Check whether this bio extends beyond the end of the device. |
| 1308 | */ |
| 1309 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) |
| 1310 | { |
| 1311 | sector_t maxsector; |
| 1312 | |
| 1313 | if (!nr_sectors) |
| 1314 | return 0; |
| 1315 | |
| 1316 | /* Test device or partition size, when known. */ |
| 1317 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
| 1318 | if (maxsector) { |
| 1319 | sector_t sector = bio->bi_sector; |
| 1320 | |
| 1321 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { |
| 1322 | /* |
| 1323 | * This may well happen - the kernel calls bread() |
| 1324 | * without checking the size of the device, e.g., when |
| 1325 | * mounting a device. |
| 1326 | */ |
| 1327 | handle_bad_sector(bio); |
| 1328 | return 1; |
| 1329 | } |
| 1330 | } |
| 1331 | |
| 1332 | return 0; |
| 1333 | } |
| 1334 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | /** |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1336 | * generic_make_request - hand a buffer to its device driver for I/O |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | * @bio: The bio describing the location in memory and on the device. |
| 1338 | * |
| 1339 | * generic_make_request() is used to make I/O requests of block |
| 1340 | * devices. It is passed a &struct bio, which describes the I/O that needs |
| 1341 | * to be done. |
| 1342 | * |
| 1343 | * generic_make_request() does not return any status. The |
| 1344 | * success/failure status of the request, along with notification of |
| 1345 | * completion, is delivered asynchronously through the bio->bi_end_io |
| 1346 | * function described (one day) else where. |
| 1347 | * |
| 1348 | * The caller of generic_make_request must make sure that bi_io_vec |
| 1349 | * are set to describe the memory buffer, and that bi_dev and bi_sector are |
| 1350 | * set to describe the device address, and the |
| 1351 | * bi_end_io and optionally bi_private are set to describe how |
| 1352 | * completion notification should be signaled. |
| 1353 | * |
| 1354 | * generic_make_request and the drivers it calls may use bi_next if this |
| 1355 | * bio happens to be merged with someone else, and may change bi_dev and |
| 1356 | * bi_sector for remaps as it sees fit. So the values of these fields |
| 1357 | * should NOT be depended on after the call to generic_make_request. |
| 1358 | */ |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1359 | static inline void __generic_make_request(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1360 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1361 | struct request_queue *q; |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1362 | sector_t old_sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | int ret, nr_sectors = bio_sectors(bio); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1364 | dev_t old_dev; |
Jens Axboe | 51fd77b | 2007-11-02 08:49:08 +0100 | [diff] [blame] | 1365 | int err = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1366 | |
| 1367 | might_sleep(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1369 | if (bio_check_eod(bio, nr_sectors)) |
| 1370 | goto end_io; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1371 | |
| 1372 | /* |
| 1373 | * Resolve the mapping until finished. (drivers are |
| 1374 | * still free to implement/resolve their own stacking |
| 1375 | * by explicitly returning 0) |
| 1376 | * |
| 1377 | * NOTE: we don't repeat the blk_size check for each new device. |
| 1378 | * Stacking drivers are expected to know what they are doing. |
| 1379 | */ |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1380 | old_sector = -1; |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1381 | old_dev = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | do { |
| 1383 | char b[BDEVNAME_SIZE]; |
| 1384 | |
| 1385 | q = bdev_get_queue(bio->bi_bdev); |
| 1386 | if (!q) { |
| 1387 | printk(KERN_ERR |
| 1388 | "generic_make_request: Trying to access " |
| 1389 | "nonexistent block-device %s (%Lu)\n", |
| 1390 | bdevname(bio->bi_bdev, b), |
| 1391 | (long long) bio->bi_sector); |
| 1392 | end_io: |
Jens Axboe | 51fd77b | 2007-11-02 08:49:08 +0100 | [diff] [blame] | 1393 | bio_endio(bio, err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | break; |
| 1395 | } |
| 1396 | |
Jens Axboe | 4fa253f | 2007-07-18 13:13:10 +0200 | [diff] [blame] | 1397 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1398 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1399 | bdevname(bio->bi_bdev, b), |
| 1400 | bio_sectors(bio), |
| 1401 | q->max_hw_sectors); |
| 1402 | goto end_io; |
| 1403 | } |
| 1404 | |
Nick Piggin | fde6ad2 | 2005-06-23 00:08:53 -0700 | [diff] [blame] | 1405 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1406 | goto end_io; |
| 1407 | |
Akinobu Mita | c17bb49 | 2006-12-08 02:39:46 -0800 | [diff] [blame] | 1408 | if (should_fail_request(bio)) |
| 1409 | goto end_io; |
| 1410 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | /* |
| 1412 | * If this device has partitions, remap block n |
| 1413 | * of partition p to block n+start(p) of the disk. |
| 1414 | */ |
| 1415 | blk_partition_remap(bio); |
| 1416 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 1417 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
| 1418 | goto end_io; |
| 1419 | |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1420 | if (old_sector != -1) |
Jens Axboe | 4fa253f | 2007-07-18 13:13:10 +0200 | [diff] [blame] | 1421 | blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1422 | old_sector); |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1423 | |
| 1424 | blk_add_trace_bio(q, bio, BLK_TA_QUEUE); |
| 1425 | |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1426 | old_sector = bio->bi_sector; |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1427 | old_dev = bio->bi_bdev->bd_dev; |
| 1428 | |
Jens Axboe | c07e2b4 | 2007-07-18 13:27:58 +0200 | [diff] [blame] | 1429 | if (bio_check_eod(bio, nr_sectors)) |
| 1430 | goto end_io; |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 1431 | if ((bio_empty_barrier(bio) && !q->prepare_flush_fn) || |
| 1432 | (bio_discard(bio) && !q->prepare_discard_fn)) { |
Jens Axboe | 51fd77b | 2007-11-02 08:49:08 +0100 | [diff] [blame] | 1433 | err = -EOPNOTSUPP; |
| 1434 | goto end_io; |
| 1435 | } |
NeilBrown | 5ddfe96 | 2006-10-30 22:07:21 -0800 | [diff] [blame] | 1436 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | ret = q->make_request_fn(q, bio); |
| 1438 | } while (ret); |
| 1439 | } |
| 1440 | |
Neil Brown | d89d879 | 2007-05-01 09:53:42 +0200 | [diff] [blame] | 1441 | /* |
| 1442 | * We only want one ->make_request_fn to be active at a time, |
| 1443 | * else stack usage with stacked devices could be a problem. |
| 1444 | * So use current->bio_{list,tail} to keep a list of requests |
| 1445 | * submited by a make_request_fn function. |
| 1446 | * current->bio_tail is also used as a flag to say if |
| 1447 | * generic_make_request is currently active in this task or not. |
| 1448 | * If it is NULL, then no make_request is active. If it is non-NULL, |
| 1449 | * then a make_request is active, and new requests should be added |
| 1450 | * at the tail |
| 1451 | */ |
| 1452 | void generic_make_request(struct bio *bio) |
| 1453 | { |
| 1454 | if (current->bio_tail) { |
| 1455 | /* make_request is active */ |
| 1456 | *(current->bio_tail) = bio; |
| 1457 | bio->bi_next = NULL; |
| 1458 | current->bio_tail = &bio->bi_next; |
| 1459 | return; |
| 1460 | } |
| 1461 | /* following loop may be a bit non-obvious, and so deserves some |
| 1462 | * explanation. |
| 1463 | * Before entering the loop, bio->bi_next is NULL (as all callers |
| 1464 | * ensure that) so we have a list with a single bio. |
| 1465 | * We pretend that we have just taken it off a longer list, so |
| 1466 | * we assign bio_list to the next (which is NULL) and bio_tail |
| 1467 | * to &bio_list, thus initialising the bio_list of new bios to be |
| 1468 | * added. __generic_make_request may indeed add some more bios |
| 1469 | * through a recursive call to generic_make_request. If it |
| 1470 | * did, we find a non-NULL value in bio_list and re-enter the loop |
| 1471 | * from the top. In this case we really did just take the bio |
| 1472 | * of the top of the list (no pretending) and so fixup bio_list and |
| 1473 | * bio_tail or bi_next, and call into __generic_make_request again. |
| 1474 | * |
| 1475 | * The loop was structured like this to make only one call to |
| 1476 | * __generic_make_request (which is important as it is large and |
| 1477 | * inlined) and to keep the structure simple. |
| 1478 | */ |
| 1479 | BUG_ON(bio->bi_next); |
| 1480 | do { |
| 1481 | current->bio_list = bio->bi_next; |
| 1482 | if (bio->bi_next == NULL) |
| 1483 | current->bio_tail = ¤t->bio_list; |
| 1484 | else |
| 1485 | bio->bi_next = NULL; |
| 1486 | __generic_make_request(bio); |
| 1487 | bio = current->bio_list; |
| 1488 | } while (bio); |
| 1489 | current->bio_tail = NULL; /* deactivate */ |
| 1490 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | EXPORT_SYMBOL(generic_make_request); |
| 1492 | |
| 1493 | /** |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1494 | * 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] | 1495 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
| 1496 | * @bio: The &struct bio which describes the I/O |
| 1497 | * |
| 1498 | * submit_bio() is very similar in purpose to generic_make_request(), and |
| 1499 | * 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] | 1500 | * interfaces; @bio must be presetup and ready for I/O. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | * |
| 1502 | */ |
| 1503 | void submit_bio(int rw, struct bio *bio) |
| 1504 | { |
| 1505 | int count = bio_sectors(bio); |
| 1506 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1507 | bio->bi_rw |= rw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1509 | /* |
| 1510 | * If it's a regular read/write or a barrier with data attached, |
| 1511 | * go through the normal accounting stuff before submission. |
| 1512 | */ |
Jens Axboe | a9c701e | 2008-08-08 11:04:44 +0200 | [diff] [blame] | 1513 | if (bio_has_data(bio)) { |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1514 | if (rw & WRITE) { |
| 1515 | count_vm_events(PGPGOUT, count); |
| 1516 | } else { |
| 1517 | task_io_account_read(bio->bi_size); |
| 1518 | count_vm_events(PGPGIN, count); |
| 1519 | } |
| 1520 | |
| 1521 | if (unlikely(block_dump)) { |
| 1522 | char b[BDEVNAME_SIZE]; |
| 1523 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", |
Pavel Emelyanov | ba25f9d | 2007-10-18 23:40:40 -0700 | [diff] [blame] | 1524 | current->comm, task_pid_nr(current), |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1525 | (rw & WRITE) ? "WRITE" : "READ", |
| 1526 | (unsigned long long)bio->bi_sector, |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1527 | bdevname(bio->bi_bdev, b)); |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1528 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | } |
| 1530 | |
| 1531 | generic_make_request(bio); |
| 1532 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1533 | EXPORT_SYMBOL(submit_bio); |
| 1534 | |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 1535 | /** |
Kiyoshi Ueda | 82124d6 | 2008-09-18 10:45:38 -0400 | [diff] [blame] | 1536 | * blk_rq_check_limits - Helper function to check a request for the queue limit |
| 1537 | * @q: the queue |
| 1538 | * @rq: the request being checked |
| 1539 | * |
| 1540 | * Description: |
| 1541 | * @rq may have been made based on weaker limitations of upper-level queues |
| 1542 | * in request stacking drivers, and it may violate the limitation of @q. |
| 1543 | * Since the block layer and the underlying device driver trust @rq |
| 1544 | * after it is inserted to @q, it should be checked against @q before |
| 1545 | * the insertion using this generic function. |
| 1546 | * |
| 1547 | * This function should also be useful for request stacking drivers |
| 1548 | * in some cases below, so export this fuction. |
| 1549 | * Request stacking drivers like request-based dm may change the queue |
| 1550 | * limits while requests are in the queue (e.g. dm's table swapping). |
| 1551 | * Such request stacking drivers should check those requests agaist |
| 1552 | * the new queue limits again when they dispatch those requests, |
| 1553 | * although such checkings are also done against the old queue limits |
| 1554 | * when submitting requests. |
| 1555 | */ |
| 1556 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) |
| 1557 | { |
| 1558 | if (rq->nr_sectors > q->max_sectors || |
| 1559 | rq->data_len > q->max_hw_sectors << 9) { |
| 1560 | printk(KERN_ERR "%s: over max size limit.\n", __func__); |
| 1561 | return -EIO; |
| 1562 | } |
| 1563 | |
| 1564 | /* |
| 1565 | * queue's settings related to segment counting like q->bounce_pfn |
| 1566 | * may differ from that of other stacking queues. |
| 1567 | * Recalculate it to check the request correctly on this queue's |
| 1568 | * limitation. |
| 1569 | */ |
| 1570 | blk_recalc_rq_segments(rq); |
| 1571 | if (rq->nr_phys_segments > q->max_phys_segments || |
| 1572 | rq->nr_phys_segments > q->max_hw_segments) { |
| 1573 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); |
| 1574 | return -EIO; |
| 1575 | } |
| 1576 | |
| 1577 | return 0; |
| 1578 | } |
| 1579 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); |
| 1580 | |
| 1581 | /** |
| 1582 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request |
| 1583 | * @q: the queue to submit the request |
| 1584 | * @rq: the request being queued |
| 1585 | */ |
| 1586 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
| 1587 | { |
| 1588 | unsigned long flags; |
| 1589 | |
| 1590 | if (blk_rq_check_limits(q, rq)) |
| 1591 | return -EIO; |
| 1592 | |
| 1593 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
| 1594 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && |
| 1595 | should_fail(&fail_make_request, blk_rq_bytes(rq))) |
| 1596 | return -EIO; |
| 1597 | #endif |
| 1598 | |
| 1599 | spin_lock_irqsave(q->queue_lock, flags); |
| 1600 | |
| 1601 | /* |
| 1602 | * Submitting request must be dequeued before calling this function |
| 1603 | * because it will be linked to another request_queue |
| 1604 | */ |
| 1605 | BUG_ON(blk_queued_rq(rq)); |
| 1606 | |
| 1607 | drive_stat_acct(rq, 1); |
| 1608 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
| 1609 | |
| 1610 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1611 | |
| 1612 | return 0; |
| 1613 | } |
| 1614 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); |
| 1615 | |
| 1616 | /** |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 1617 | * __end_that_request_first - end I/O on a request |
| 1618 | * @req: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1619 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 1620 | * @nr_bytes: number of bytes to complete |
| 1621 | * |
| 1622 | * Description: |
| 1623 | * Ends I/O on a number of bytes attached to @req, and sets it up |
| 1624 | * for the next range of segments (if any) in the cluster. |
| 1625 | * |
| 1626 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1627 | * %0 - we are done with this request, call end_that_request_last() |
| 1628 | * %1 - still buffers pending for this request |
Kiyoshi Ueda | 3bcddea | 2007-12-11 17:52:28 -0500 | [diff] [blame] | 1629 | **/ |
Kiyoshi Ueda | 5450d3e | 2007-12-11 17:53:03 -0500 | [diff] [blame] | 1630 | static int __end_that_request_first(struct request *req, int error, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1631 | int nr_bytes) |
| 1632 | { |
Kiyoshi Ueda | 5450d3e | 2007-12-11 17:53:03 -0500 | [diff] [blame] | 1633 | int total_bytes, bio_nbytes, next_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | struct bio *bio; |
| 1635 | |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1636 | blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); |
| 1637 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | /* |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1639 | * for a REQ_TYPE_BLOCK_PC request, we want to carry any eventual |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1640 | * sense key with us all the way through |
| 1641 | */ |
| 1642 | if (!blk_pc_request(req)) |
| 1643 | req->errors = 0; |
| 1644 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1645 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
| 1646 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1647 | req->rq_disk ? req->rq_disk->disk_name : "?", |
| 1648 | (unsigned long long)req->sector); |
| 1649 | } |
| 1650 | |
Jens Axboe | d72d904 | 2005-11-01 08:35:42 +0100 | [diff] [blame] | 1651 | if (blk_fs_request(req) && req->rq_disk) { |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1652 | const int rw = rq_data_dir(req); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1653 | struct hd_struct *part; |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 1654 | int cpu; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1655 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1656 | cpu = part_stat_lock(); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1657 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1658 | part_stat_add(cpu, part, sectors[rw], nr_bytes >> 9); |
| 1659 | part_stat_unlock(); |
Jens Axboe | d72d904 | 2005-11-01 08:35:42 +0100 | [diff] [blame] | 1660 | } |
| 1661 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1662 | total_bytes = bio_nbytes = 0; |
| 1663 | while ((bio = req->bio) != NULL) { |
| 1664 | int nbytes; |
| 1665 | |
Jens Axboe | bf2de6f | 2007-09-27 13:01:25 +0200 | [diff] [blame] | 1666 | /* |
| 1667 | * For an empty barrier request, the low level driver must |
| 1668 | * store a potential error location in ->sector. We pass |
| 1669 | * that back up in ->bi_sector. |
| 1670 | */ |
| 1671 | if (blk_empty_barrier(req)) |
| 1672 | bio->bi_sector = req->sector; |
| 1673 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | if (nr_bytes >= bio->bi_size) { |
| 1675 | req->bio = bio->bi_next; |
| 1676 | nbytes = bio->bi_size; |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 1677 | req_bio_endio(req, bio, nbytes, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1678 | next_idx = 0; |
| 1679 | bio_nbytes = 0; |
| 1680 | } else { |
| 1681 | int idx = bio->bi_idx + next_idx; |
| 1682 | |
| 1683 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { |
| 1684 | blk_dump_rq_flags(req, "__end_that"); |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1685 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
Harvey Harrison | 24c03d4 | 2008-05-01 04:35:17 -0700 | [diff] [blame] | 1686 | __func__, bio->bi_idx, bio->bi_vcnt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | break; |
| 1688 | } |
| 1689 | |
| 1690 | nbytes = bio_iovec_idx(bio, idx)->bv_len; |
| 1691 | BIO_BUG_ON(nbytes > bio->bi_size); |
| 1692 | |
| 1693 | /* |
| 1694 | * not a complete bvec done |
| 1695 | */ |
| 1696 | if (unlikely(nbytes > nr_bytes)) { |
| 1697 | bio_nbytes += nr_bytes; |
| 1698 | total_bytes += nr_bytes; |
| 1699 | break; |
| 1700 | } |
| 1701 | |
| 1702 | /* |
| 1703 | * advance to the next vector |
| 1704 | */ |
| 1705 | next_idx++; |
| 1706 | bio_nbytes += nbytes; |
| 1707 | } |
| 1708 | |
| 1709 | total_bytes += nbytes; |
| 1710 | nr_bytes -= nbytes; |
| 1711 | |
Jens Axboe | 6728cb0 | 2008-01-31 13:03:55 +0100 | [diff] [blame] | 1712 | bio = req->bio; |
| 1713 | if (bio) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1714 | /* |
| 1715 | * end more in this run, or just return 'not-done' |
| 1716 | */ |
| 1717 | if (unlikely(nr_bytes <= 0)) |
| 1718 | break; |
| 1719 | } |
| 1720 | } |
| 1721 | |
| 1722 | /* |
| 1723 | * completely done |
| 1724 | */ |
| 1725 | if (!req->bio) |
| 1726 | return 0; |
| 1727 | |
| 1728 | /* |
| 1729 | * if the request wasn't completed, update state |
| 1730 | */ |
| 1731 | if (bio_nbytes) { |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 1732 | req_bio_endio(req, bio, bio_nbytes, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1733 | bio->bi_idx += next_idx; |
| 1734 | bio_iovec(bio)->bv_offset += nr_bytes; |
| 1735 | bio_iovec(bio)->bv_len -= nr_bytes; |
| 1736 | } |
| 1737 | |
| 1738 | blk_recalc_rq_sectors(req, total_bytes >> 9); |
| 1739 | blk_recalc_rq_segments(req); |
| 1740 | return 1; |
| 1741 | } |
| 1742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1743 | /* |
| 1744 | * queue lock must be held |
| 1745 | */ |
Kiyoshi Ueda | 5450d3e | 2007-12-11 17:53:03 -0500 | [diff] [blame] | 1746 | static void end_that_request_last(struct request *req, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | { |
| 1748 | struct gendisk *disk = req->rq_disk; |
Tejun Heo | 8ffdc65 | 2006-01-06 09:49:03 +0100 | [diff] [blame] | 1749 | |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 1750 | blk_delete_timer(req); |
| 1751 | |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 1752 | if (blk_rq_tagged(req)) |
| 1753 | blk_queue_end_tag(req->q, req); |
| 1754 | |
| 1755 | if (blk_queued_rq(req)) |
| 1756 | blkdev_dequeue_request(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | |
| 1758 | if (unlikely(laptop_mode) && blk_fs_request(req)) |
| 1759 | laptop_io_completion(); |
| 1760 | |
Jens Axboe | fd0ff8a | 2006-05-23 11:23:49 +0200 | [diff] [blame] | 1761 | /* |
| 1762 | * Account IO completion. bar_rq isn't accounted as a normal |
| 1763 | * IO on queueing nor completion. Accounting the containing |
| 1764 | * request is enough. |
| 1765 | */ |
| 1766 | if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1767 | unsigned long duration = jiffies - req->start_time; |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1768 | const int rw = rq_data_dir(req); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1769 | struct hd_struct *part; |
Tejun Heo | c995905 | 2008-08-25 19:47:21 +0900 | [diff] [blame] | 1770 | int cpu; |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1771 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1772 | cpu = part_stat_lock(); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1773 | part = disk_map_sector_rcu(disk, req->sector); |
Jens Axboe | a362357 | 2005-11-01 09:26:16 +0100 | [diff] [blame] | 1774 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1775 | part_stat_inc(cpu, part, ios[rw]); |
| 1776 | part_stat_add(cpu, part, ticks[rw], duration); |
| 1777 | part_round_stats(cpu, part); |
| 1778 | part_dec_in_flight(part); |
Tejun Heo | e71bf0d | 2008-09-03 09:03:02 +0200 | [diff] [blame] | 1779 | |
Tejun Heo | 074a7ac | 2008-08-25 19:56:14 +0900 | [diff] [blame] | 1780 | part_stat_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1781 | } |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 1782 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1783 | if (req->end_io) |
Tejun Heo | 8ffdc65 | 2006-01-06 09:49:03 +0100 | [diff] [blame] | 1784 | req->end_io(req, error); |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 1785 | else { |
| 1786 | if (blk_bidi_rq(req)) |
| 1787 | __blk_put_request(req->next_rq->q, req->next_rq); |
| 1788 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1789 | __blk_put_request(req->q, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | } |
| 1791 | } |
| 1792 | |
Kiyoshi Ueda | 3b11313 | 2007-12-11 17:41:17 -0500 | [diff] [blame] | 1793 | /** |
| 1794 | * blk_rq_bytes - Returns bytes left to complete in the entire request |
Randy Dunlap | 5d87a05 | 2008-02-20 09:01:22 +0100 | [diff] [blame] | 1795 | * @rq: the request being processed |
Kiyoshi Ueda | 3b11313 | 2007-12-11 17:41:17 -0500 | [diff] [blame] | 1796 | **/ |
| 1797 | unsigned int blk_rq_bytes(struct request *rq) |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1798 | { |
| 1799 | if (blk_fs_request(rq)) |
| 1800 | return rq->hard_nr_sectors << 9; |
| 1801 | |
| 1802 | return rq->data_len; |
| 1803 | } |
Kiyoshi Ueda | 3b11313 | 2007-12-11 17:41:17 -0500 | [diff] [blame] | 1804 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
| 1805 | |
| 1806 | /** |
| 1807 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment |
Randy Dunlap | 5d87a05 | 2008-02-20 09:01:22 +0100 | [diff] [blame] | 1808 | * @rq: the request being processed |
Kiyoshi Ueda | 3b11313 | 2007-12-11 17:41:17 -0500 | [diff] [blame] | 1809 | **/ |
| 1810 | unsigned int blk_rq_cur_bytes(struct request *rq) |
| 1811 | { |
| 1812 | if (blk_fs_request(rq)) |
| 1813 | return rq->current_nr_sectors << 9; |
| 1814 | |
| 1815 | if (rq->bio) |
| 1816 | return rq->bio->bi_size; |
| 1817 | |
| 1818 | return rq->data_len; |
| 1819 | } |
| 1820 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1821 | |
| 1822 | /** |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1823 | * end_request - end I/O on the current segment of the request |
Randy Dunlap | 8f731f7 | 2007-10-18 23:39:28 -0700 | [diff] [blame] | 1824 | * @req: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1825 | * @uptodate: error value or %0/%1 uptodate flag |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1826 | * |
| 1827 | * Description: |
| 1828 | * Ends I/O on the current segment of a request. If that is the only |
| 1829 | * remaining segment, the request is also completed and freed. |
| 1830 | * |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1831 | * This is a remnant of how older block drivers handled I/O completions. |
| 1832 | * Modern drivers typically end I/O on the full request in one go, unless |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1833 | * they have a residual value to account for. For that case this function |
| 1834 | * isn't really useful, unless the residual just happens to be the |
| 1835 | * full current segment. In other words, don't use this function in new |
Kiyoshi Ueda | d00e29f | 2008-10-01 10:14:46 -0400 | [diff] [blame] | 1836 | * code. Use blk_end_request() or __blk_end_request() to end a request. |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1837 | **/ |
| 1838 | void end_request(struct request *req, int uptodate) |
| 1839 | { |
Kiyoshi Ueda | d00e29f | 2008-10-01 10:14:46 -0400 | [diff] [blame] | 1840 | int error = 0; |
| 1841 | |
| 1842 | if (uptodate <= 0) |
| 1843 | error = uptodate ? uptodate : -EIO; |
| 1844 | |
| 1845 | __blk_end_request(req, error, req->hard_cur_sectors << 9); |
Jens Axboe | a0cd128 | 2007-09-21 10:41:07 +0200 | [diff] [blame] | 1846 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | EXPORT_SYMBOL(end_request); |
| 1848 | |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 1849 | static int end_that_request_data(struct request *rq, int error, |
| 1850 | unsigned int nr_bytes, unsigned int bidi_bytes) |
| 1851 | { |
| 1852 | if (rq->bio) { |
| 1853 | if (__end_that_request_first(rq, error, nr_bytes)) |
| 1854 | return 1; |
| 1855 | |
| 1856 | /* Bidi request must be completed as a whole */ |
| 1857 | if (blk_bidi_rq(rq) && |
| 1858 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) |
| 1859 | return 1; |
| 1860 | } |
| 1861 | |
| 1862 | return 0; |
| 1863 | } |
| 1864 | |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1865 | /** |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1866 | * blk_end_io - Generic end_io function to complete a request. |
| 1867 | * @rq: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1868 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1869 | * @nr_bytes: number of bytes to complete @rq |
| 1870 | * @bidi_bytes: number of bytes to complete @rq->next_rq |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1871 | * @drv_callback: function called between completion of bios in the request |
| 1872 | * and completion of the request. |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1873 | * If the callback returns non %0, this helper returns without |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1874 | * completion of the request. |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1875 | * |
| 1876 | * Description: |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1877 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1878 | * If @rq has leftover, sets it up for the next range of segments. |
| 1879 | * |
| 1880 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1881 | * %0 - we are done with this request |
| 1882 | * %1 - this request is not freed yet, it still has pending buffers. |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1883 | **/ |
Jens Axboe | 22b1321 | 2008-01-31 12:36:19 +0100 | [diff] [blame] | 1884 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
| 1885 | unsigned int bidi_bytes, |
| 1886 | int (drv_callback)(struct request *)) |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1887 | { |
| 1888 | struct request_queue *q = rq->q; |
| 1889 | unsigned long flags = 0UL; |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1890 | |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 1891 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
| 1892 | return 1; |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1893 | |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1894 | /* Special feature for tricky drivers */ |
| 1895 | if (drv_callback && drv_callback(rq)) |
| 1896 | return 1; |
| 1897 | |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1898 | add_disk_randomness(rq->rq_disk); |
| 1899 | |
| 1900 | spin_lock_irqsave(q->queue_lock, flags); |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 1901 | end_that_request_last(rq, error); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1902 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 1903 | |
| 1904 | return 0; |
| 1905 | } |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1906 | |
| 1907 | /** |
| 1908 | * blk_end_request - Helper function for drivers to complete the request. |
| 1909 | * @rq: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1910 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1911 | * @nr_bytes: number of bytes to complete |
| 1912 | * |
| 1913 | * Description: |
| 1914 | * Ends I/O on a number of bytes attached to @rq. |
| 1915 | * If @rq has leftover, sets it up for the next range of segments. |
| 1916 | * |
| 1917 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1918 | * %0 - we are done with this request |
| 1919 | * %1 - still buffers pending for this request |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1920 | **/ |
Jens Axboe | 22b1321 | 2008-01-31 12:36:19 +0100 | [diff] [blame] | 1921 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1922 | { |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1923 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1924 | } |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1925 | EXPORT_SYMBOL_GPL(blk_end_request); |
| 1926 | |
| 1927 | /** |
| 1928 | * __blk_end_request - Helper function for drivers to complete the request. |
| 1929 | * @rq: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1930 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1931 | * @nr_bytes: number of bytes to complete |
| 1932 | * |
| 1933 | * Description: |
| 1934 | * Must be called with queue lock held unlike blk_end_request(). |
| 1935 | * |
| 1936 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1937 | * %0 - we are done with this request |
| 1938 | * %1 - still buffers pending for this request |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1939 | **/ |
Jens Axboe | 22b1321 | 2008-01-31 12:36:19 +0100 | [diff] [blame] | 1940 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1941 | { |
Jens Axboe | 6054016 | 2008-08-26 13:34:34 +0200 | [diff] [blame] | 1942 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
Jens Axboe | 051cc39 | 2008-08-08 11:06:45 +0200 | [diff] [blame] | 1943 | return 1; |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1944 | |
| 1945 | add_disk_randomness(rq->rq_disk); |
| 1946 | |
Kiyoshi Ueda | b828623 | 2007-12-11 17:53:24 -0500 | [diff] [blame] | 1947 | end_that_request_last(rq, error); |
Kiyoshi Ueda | 336cdb4 | 2007-12-11 17:40:30 -0500 | [diff] [blame] | 1948 | |
| 1949 | return 0; |
| 1950 | } |
| 1951 | EXPORT_SYMBOL_GPL(__blk_end_request); |
| 1952 | |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 1953 | /** |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1954 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. |
| 1955 | * @rq: the bidi request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1956 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1957 | * @nr_bytes: number of bytes to complete @rq |
| 1958 | * @bidi_bytes: number of bytes to complete @rq->next_rq |
| 1959 | * |
| 1960 | * Description: |
| 1961 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
| 1962 | * |
| 1963 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 1964 | * %0 - we are done with this request |
| 1965 | * %1 - still buffers pending for this request |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1966 | **/ |
Jens Axboe | 22b1321 | 2008-01-31 12:36:19 +0100 | [diff] [blame] | 1967 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
| 1968 | unsigned int bidi_bytes) |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 1969 | { |
| 1970 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); |
| 1971 | } |
| 1972 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); |
| 1973 | |
| 1974 | /** |
Kiyoshi Ueda | 32fab44 | 2008-09-18 10:45:09 -0400 | [diff] [blame] | 1975 | * blk_update_request - Special helper function for request stacking drivers |
| 1976 | * @rq: the request being processed |
| 1977 | * @error: %0 for success, < %0 for error |
| 1978 | * @nr_bytes: number of bytes to complete @rq |
| 1979 | * |
| 1980 | * Description: |
| 1981 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete |
| 1982 | * the request structure even if @rq doesn't have leftover. |
| 1983 | * If @rq has leftover, sets it up for the next range of segments. |
| 1984 | * |
| 1985 | * This special helper function is only for request stacking drivers |
| 1986 | * (e.g. request-based dm) so that they can handle partial completion. |
| 1987 | * Actual device drivers should use blk_end_request instead. |
| 1988 | */ |
| 1989 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) |
| 1990 | { |
| 1991 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { |
| 1992 | /* |
| 1993 | * These members are not updated in end_that_request_data() |
| 1994 | * when all bios are completed. |
| 1995 | * Update them so that the request stacking driver can find |
| 1996 | * how many bytes remain in the request later. |
| 1997 | */ |
| 1998 | rq->nr_sectors = rq->hard_nr_sectors = 0; |
| 1999 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; |
| 2000 | } |
| 2001 | } |
| 2002 | EXPORT_SYMBOL_GPL(blk_update_request); |
| 2003 | |
| 2004 | /** |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2005 | * blk_end_request_callback - Special helper function for tricky drivers |
| 2006 | * @rq: the request being processed |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 2007 | * @error: %0 for success, < %0 for error |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2008 | * @nr_bytes: number of bytes to complete |
| 2009 | * @drv_callback: function called between completion of bios in the request |
| 2010 | * and completion of the request. |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 2011 | * If the callback returns non %0, this helper returns without |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2012 | * completion of the request. |
| 2013 | * |
| 2014 | * Description: |
| 2015 | * Ends I/O on a number of bytes attached to @rq. |
| 2016 | * If @rq has leftover, sets it up for the next range of segments. |
| 2017 | * |
| 2018 | * This special helper function is used only for existing tricky drivers. |
| 2019 | * (e.g. cdrom_newpc_intr() of ide-cd) |
| 2020 | * This interface will be removed when such drivers are rewritten. |
| 2021 | * Don't use this interface in other places anymore. |
| 2022 | * |
| 2023 | * Return: |
Randy Dunlap | 710027a | 2008-08-19 20:13:11 +0200 | [diff] [blame] | 2024 | * %0 - we are done with this request |
| 2025 | * %1 - this request is not freed yet. |
| 2026 | * this request still has pending buffers or |
| 2027 | * the driver doesn't want to finish this request yet. |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2028 | **/ |
Jens Axboe | 22b1321 | 2008-01-31 12:36:19 +0100 | [diff] [blame] | 2029 | int blk_end_request_callback(struct request *rq, int error, |
| 2030 | unsigned int nr_bytes, |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2031 | int (drv_callback)(struct request *)) |
| 2032 | { |
Kiyoshi Ueda | e3a04fe | 2007-12-11 17:51:46 -0500 | [diff] [blame] | 2033 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
Kiyoshi Ueda | e19a3ab | 2007-12-11 17:51:02 -0500 | [diff] [blame] | 2034 | } |
| 2035 | EXPORT_SYMBOL_GPL(blk_end_request_callback); |
| 2036 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 2037 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
| 2038 | struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2039 | { |
David Woodhouse | d628eae | 2008-08-09 16:22:17 +0100 | [diff] [blame] | 2040 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
| 2041 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ |
Jens Axboe | 4aff5e2 | 2006-08-10 08:44:47 +0200 | [diff] [blame] | 2042 | rq->cmd_flags |= (bio->bi_rw & 3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 2044 | if (bio_has_data(bio)) { |
| 2045 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
David Woodhouse | fb2dce8 | 2008-08-05 18:01:53 +0100 | [diff] [blame] | 2046 | rq->buffer = bio_data(bio); |
| 2047 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | rq->current_nr_sectors = bio_cur_sectors(bio); |
| 2049 | rq->hard_cur_sectors = rq->current_nr_sectors; |
| 2050 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); |
Mike Christie | 0e75f90 | 2006-12-01 10:40:55 +0100 | [diff] [blame] | 2051 | rq->data_len = bio->bi_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | |
| 2053 | rq->bio = rq->biotail = bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | |
NeilBrown | 6684657 | 2007-08-16 13:31:28 +0200 | [diff] [blame] | 2055 | if (bio->bi_bdev) |
| 2056 | rq->rq_disk = bio->bi_bdev->bd_disk; |
| 2057 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2058 | |
Kiyoshi Ueda | ef9e3fa | 2008-10-01 16:12:15 +0200 | [diff] [blame] | 2059 | /** |
| 2060 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy |
| 2061 | * @q : the queue of the device being checked |
| 2062 | * |
| 2063 | * Description: |
| 2064 | * Check if underlying low-level drivers of a device are busy. |
| 2065 | * If the drivers want to export their busy state, they must set own |
| 2066 | * exporting function using blk_queue_lld_busy() first. |
| 2067 | * |
| 2068 | * Basically, this function is used only by request stacking drivers |
| 2069 | * to stop dispatching requests to underlying devices when underlying |
| 2070 | * devices are busy. This behavior helps more I/O merging on the queue |
| 2071 | * of the request stacking driver and prevents I/O throughput regression |
| 2072 | * on burst I/O load. |
| 2073 | * |
| 2074 | * Return: |
| 2075 | * 0 - Not busy (The request stacking driver should dispatch request) |
| 2076 | * 1 - Busy (The request stacking driver should stop dispatching request) |
| 2077 | */ |
| 2078 | int blk_lld_busy(struct request_queue *q) |
| 2079 | { |
| 2080 | if (q->lld_busy_fn) |
| 2081 | return q->lld_busy_fn(q); |
| 2082 | |
| 2083 | return 0; |
| 2084 | } |
| 2085 | EXPORT_SYMBOL_GPL(blk_lld_busy); |
| 2086 | |
Jens Axboe | 18887ad | 2008-07-28 13:08:45 +0200 | [diff] [blame] | 2087 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2088 | { |
| 2089 | return queue_work(kblockd_workqueue, work); |
| 2090 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2091 | EXPORT_SYMBOL(kblockd_schedule_work); |
| 2092 | |
Andrew Morton | 19a75d8 | 2007-05-09 02:33:56 -0700 | [diff] [blame] | 2093 | void kblockd_flush_work(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | { |
Oleg Nesterov | 28e53bd | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 2095 | cancel_work_sync(work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2096 | } |
Andrew Morton | 19a75d8 | 2007-05-09 02:33:56 -0700 | [diff] [blame] | 2097 | EXPORT_SYMBOL(kblockd_flush_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2098 | |
| 2099 | int __init blk_dev_init(void) |
| 2100 | { |
| 2101 | kblockd_workqueue = create_workqueue("kblockd"); |
| 2102 | if (!kblockd_workqueue) |
| 2103 | panic("Failed to create kblockd\n"); |
| 2104 | |
| 2105 | request_cachep = kmem_cache_create("blkdev_requests", |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2106 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | |
Jens Axboe | 8324aa9 | 2008-01-29 14:51:59 +0100 | [diff] [blame] | 2108 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2109 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2110 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2111 | return 0; |
| 2112 | } |
| 2113 | |