blob: b901db63f6ae082f1df23e3227c9ddd2e56d61d1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
8 */
9
10/*
11 * This handles all read/write requests to block devices
12 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/bio.h>
17#include <linux/blkdev.h>
18#include <linux/highmem.h>
19#include <linux/mm.h>
20#include <linux/kernel_stat.h>
21#include <linux/string.h>
22#include <linux/init.h>
23#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Andrew Mortonfaccbd42006-12-10 02:19:35 -080028#include <linux/task_io_accounting_ops.h>
Jens Axboeff856ba2006-01-09 16:02:34 +010029#include <linux/interrupt.h>
30#include <linux/cpu.h>
Jens Axboe2056a782006-03-23 20:00:26 +010031#include <linux/blktrace_api.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080032#include <linux/fault-inject.h>
Jens Axboef5659132007-09-21 10:44:19 +020033#include <linux/scatterlist.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35/*
36 * for max sense size
37 */
38#include <scsi/scsi_cmnd.h>
39
David Howells65f27f32006-11-22 14:55:48 +000040static void blk_unplug_work(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041static void blk_unplug_timeout(unsigned long data);
Jerome Marchandb238b3d2007-10-23 15:05:46 +020042static void drive_stat_acct(struct request *rq, int new_io);
Tejun Heo52d9e672006-01-06 09:49:58 +010043static void init_request_from_bio(struct request *req, struct bio *bio);
Jens Axboe165125e2007-07-24 09:28:11 +020044static int __make_request(struct request_queue *q, struct bio *bio);
Jens Axboeb5deef92006-07-19 23:39:40 +020045static struct io_context *current_io_context(gfp_t gfp_flags, int node);
NeilBrown9dfa5282007-08-16 13:27:52 +020046static void blk_recalc_rq_segments(struct request *rq);
NeilBrown66846572007-08-16 13:31:28 +020047static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
48 struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
50/*
51 * For the allocated request tables
52 */
Christoph Lametere18b8902006-12-06 20:33:20 -080053static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55/*
56 * For queue allocation
57 */
Christoph Lametere18b8902006-12-06 20:33:20 -080058static struct kmem_cache *requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
61 * For io context allocations
62 */
Christoph Lametere18b8902006-12-06 20:33:20 -080063static struct kmem_cache *iocontext_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
66 * Controlling structure to kblockd
67 */
Jens Axboeff856ba2006-01-09 16:02:34 +010068static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70unsigned long blk_max_low_pfn, blk_max_pfn;
71
72EXPORT_SYMBOL(blk_max_low_pfn);
73EXPORT_SYMBOL(blk_max_pfn);
74
Jens Axboeff856ba2006-01-09 16:02:34 +010075static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* Amount of time in which a process may batch requests */
78#define BLK_BATCH_TIME (HZ/50UL)
79
80/* Number of requests a "batching" process may submit */
81#define BLK_BATCH_REQ 32
82
83/*
84 * Return the threshold (number of used requests) at which the queue is
85 * considered to be congested. It include a little hysteresis to keep the
86 * context switch rate down.
87 */
88static inline int queue_congestion_on_threshold(struct request_queue *q)
89{
90 return q->nr_congestion_on;
91}
92
93/*
94 * The threshold at which a queue is considered to be uncongested
95 */
96static inline int queue_congestion_off_threshold(struct request_queue *q)
97{
98 return q->nr_congestion_off;
99}
100
101static void blk_queue_congestion_threshold(struct request_queue *q)
102{
103 int nr;
104
105 nr = q->nr_requests - (q->nr_requests / 8) + 1;
106 if (nr > q->nr_requests)
107 nr = q->nr_requests;
108 q->nr_congestion_on = nr;
109
110 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
111 if (nr < 1)
112 nr = 1;
113 q->nr_congestion_off = nr;
114}
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116/**
117 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
118 * @bdev: device
119 *
120 * Locates the passed device's request queue and returns the address of its
121 * backing_dev_info
122 *
123 * Will return NULL if the request queue cannot be located.
124 */
125struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
126{
127 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200128 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130 if (q)
131 ret = &q->backing_dev_info;
132 return ret;
133}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134EXPORT_SYMBOL(blk_get_backing_dev_info);
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/**
137 * blk_queue_prep_rq - set a prepare_request function for queue
138 * @q: queue
139 * @pfn: prepare_request function
140 *
141 * It's possible for a queue to register a prepare_request callback which
142 * is invoked before the request is handed to the request_fn. The goal of
143 * the function is to prepare a request for I/O, it can be used to build a
144 * cdb from the request data for instance.
145 *
146 */
Jens Axboe165125e2007-07-24 09:28:11 +0200147void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 q->prep_rq_fn = pfn;
150}
151
152EXPORT_SYMBOL(blk_queue_prep_rq);
153
154/**
155 * blk_queue_merge_bvec - set a merge_bvec function for queue
156 * @q: queue
157 * @mbfn: merge_bvec_fn
158 *
159 * Usually queues have static limitations on the max sectors or segments that
160 * we can put in a request. Stacking drivers may have some settings that
161 * are dynamic, and thus we have to query the queue whether it is ok to
162 * add a new bio_vec to a bio at a given offset or not. If the block device
163 * has such limitations, it needs to register a merge_bvec_fn to control
164 * the size of bio's sent to it. Note that a block device *must* allow a
165 * single page to be added to an empty bio. The block device driver may want
166 * to use the bio_split() function to deal with these bio's. By default
167 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
168 * honored.
169 */
Jens Axboe165125e2007-07-24 09:28:11 +0200170void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
172 q->merge_bvec_fn = mbfn;
173}
174
175EXPORT_SYMBOL(blk_queue_merge_bvec);
176
Jens Axboe165125e2007-07-24 09:28:11 +0200177void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
Jens Axboeff856ba2006-01-09 16:02:34 +0100178{
179 q->softirq_done_fn = fn;
180}
181
182EXPORT_SYMBOL(blk_queue_softirq_done);
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/**
185 * blk_queue_make_request - define an alternate make_request function for a device
186 * @q: the request queue for the device to be affected
187 * @mfn: the alternate make_request function
188 *
189 * Description:
190 * The normal way for &struct bios to be passed to a device
191 * driver is for them to be collected into requests on a request
192 * queue, and then to allow the device driver to select requests
193 * off that queue when it is ready. This works well for many block
194 * devices. However some block devices (typically virtual devices
195 * such as md or lvm) do not benefit from the processing on the
196 * request queue, and are served best by having the requests passed
197 * directly to them. This can be achieved by providing a function
198 * to blk_queue_make_request().
199 *
200 * Caveat:
201 * The driver that does this *must* be able to deal appropriately
202 * with buffers in "highmemory". This can be accomplished by either calling
203 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
204 * blk_queue_bounce() to create a buffer in normal memory.
205 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200206void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 /*
209 * set defaults
210 */
211 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700212 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
213 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 q->make_request_fn = mfn;
215 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
216 q->backing_dev_info.state = 0;
217 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600218 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 blk_queue_hardsect_size(q, 512);
220 blk_queue_dma_alignment(q, 511);
221 blk_queue_congestion_threshold(q);
222 q->nr_batching = BLK_BATCH_REQ;
223
224 q->unplug_thresh = 4; /* hmm */
225 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
226 if (q->unplug_delay == 0)
227 q->unplug_delay = 1;
228
David Howells65f27f32006-11-22 14:55:48 +0000229 INIT_WORK(&q->unplug_work, blk_unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
231 q->unplug_timer.function = blk_unplug_timeout;
232 q->unplug_timer.data = (unsigned long)q;
233
234 /*
235 * by default assume old behaviour and bounce for any highmem page
236 */
237 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240EXPORT_SYMBOL(blk_queue_make_request);
241
Jens Axboe165125e2007-07-24 09:28:11 +0200242static void rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
244 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100245 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200249 INIT_HLIST_NODE(&rq->hash);
250 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200251 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 rq->buffer = NULL;
253 rq->ref_count = 1;
254 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 rq->special = NULL;
256 rq->data_len = 0;
257 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200258 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 rq->sense = NULL;
260 rq->end_io = NULL;
261 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100262 rq->completion_data = NULL;
FUJITA Tomonoriabae1fd2007-07-16 08:52:14 +0200263 rq->next_rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
266/**
267 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100268 * @q: the request queue
269 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100270 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 *
272 * Description:
273 * For journalled file systems, doing ordered writes on a commit
274 * block instead of explicitly doing wait_on_buffer (which is bad
275 * for performance) can be a big win. Block drivers supporting this
276 * feature should call this function and indicate so.
277 *
278 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200279int blk_queue_ordered(struct request_queue *q, unsigned ordered,
Tejun Heo797e7db2006-01-06 09:51:03 +0100280 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Tejun Heo797e7db2006-01-06 09:51:03 +0100282 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
283 prepare_flush_fn == NULL) {
284 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
285 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100287
288 if (ordered != QUEUE_ORDERED_NONE &&
289 ordered != QUEUE_ORDERED_DRAIN &&
290 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
291 ordered != QUEUE_ORDERED_DRAIN_FUA &&
292 ordered != QUEUE_ORDERED_TAG &&
293 ordered != QUEUE_ORDERED_TAG_FLUSH &&
294 ordered != QUEUE_ORDERED_TAG_FUA) {
295 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
296 return -EINVAL;
297 }
298
Tetsuo Takata60481b12006-01-24 10:34:36 +0100299 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100300 q->next_ordered = ordered;
301 q->prepare_flush_fn = prepare_flush_fn;
302
303 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
306EXPORT_SYMBOL(blk_queue_ordered);
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * Cache flushing for ordered writes handling
310 */
Jens Axboe165125e2007-07-24 09:28:11 +0200311inline unsigned blk_ordered_cur_seq(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Tejun Heo797e7db2006-01-06 09:51:03 +0100313 if (!q->ordseq)
314 return 0;
315 return 1 << ffz(q->ordseq);
316}
317
318unsigned blk_ordered_req_seq(struct request *rq)
319{
Jens Axboe165125e2007-07-24 09:28:11 +0200320 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Tejun Heo797e7db2006-01-06 09:51:03 +0100322 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200323
Tejun Heo797e7db2006-01-06 09:51:03 +0100324 if (rq == &q->pre_flush_rq)
325 return QUEUE_ORDSEQ_PREFLUSH;
326 if (rq == &q->bar_rq)
327 return QUEUE_ORDSEQ_BAR;
328 if (rq == &q->post_flush_rq)
329 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Tejun Heobc90ba02007-06-15 13:24:28 +0200331 /*
332 * !fs requests don't need to follow barrier ordering. Always
333 * put them at the front. This fixes the following deadlock.
334 *
335 * http://thread.gmane.org/gmane.linux.kernel/537473
336 */
337 if (!blk_fs_request(rq))
338 return QUEUE_ORDSEQ_DRAIN;
339
Jens Axboe4aff5e22006-08-10 08:44:47 +0200340 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
341 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100342 return QUEUE_ORDSEQ_DRAIN;
343 else
344 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
Jens Axboe165125e2007-07-24 09:28:11 +0200347void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Tejun Heo797e7db2006-01-06 09:51:03 +0100349 struct request *rq;
350 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Tejun Heo797e7db2006-01-06 09:51:03 +0100352 if (error && !q->orderr)
353 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200354
Tejun Heo797e7db2006-01-06 09:51:03 +0100355 BUG_ON(q->ordseq & seq);
356 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Tejun Heo797e7db2006-01-06 09:51:03 +0100358 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
359 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
361 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100362 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 */
Jens Axboe4fa253f2007-07-18 13:13:10 +0200364 uptodate = 1;
365 if (q->orderr)
366 uptodate = q->orderr;
Tejun Heo797e7db2006-01-06 09:51:03 +0100367
368 q->ordseq = 0;
Jens Axboe4fa253f2007-07-18 13:13:10 +0200369 rq = q->orig_bar_rq;
Tejun Heo797e7db2006-01-06 09:51:03 +0100370
371 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
372 end_that_request_last(rq, uptodate);
373}
374
375static void pre_flush_end_io(struct request *rq, int error)
376{
377 elv_completed_request(rq->q, rq);
378 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
379}
380
381static void bar_end_io(struct request *rq, int error)
382{
383 elv_completed_request(rq->q, rq);
384 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
385}
386
387static void post_flush_end_io(struct request *rq, int error)
388{
389 elv_completed_request(rq->q, rq);
390 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
391}
392
Jens Axboe165125e2007-07-24 09:28:11 +0200393static void queue_flush(struct request_queue *q, unsigned which)
Tejun Heo797e7db2006-01-06 09:51:03 +0100394{
395 struct request *rq;
396 rq_end_io_fn *end_io;
397
398 if (which == QUEUE_ORDERED_PREFLUSH) {
399 rq = &q->pre_flush_rq;
400 end_io = pre_flush_end_io;
401 } else {
402 rq = &q->post_flush_rq;
403 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
405
Jens Axboe4aff5e22006-08-10 08:44:47 +0200406 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100407 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100408 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200409 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100410 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100411 rq->end_io = end_io;
412 q->prepare_flush_fn(q, rq);
413
Tejun Heo30e96562006-02-08 01:01:31 -0800414 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100415}
416
Jens Axboe165125e2007-07-24 09:28:11 +0200417static inline struct request *start_ordered(struct request_queue *q,
Tejun Heo797e7db2006-01-06 09:51:03 +0100418 struct request *rq)
419{
Tejun Heo797e7db2006-01-06 09:51:03 +0100420 q->orderr = 0;
421 q->ordered = q->next_ordered;
422 q->ordseq |= QUEUE_ORDSEQ_STARTED;
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100425 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100427 blkdev_dequeue_request(rq);
428 q->orig_bar_rq = rq;
429 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200430 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100431 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200432 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
433 rq->cmd_flags |= REQ_RW;
Jens Axboe4fa253f2007-07-18 13:13:10 +0200434 if (q->ordered & QUEUE_ORDERED_FUA)
435 rq->cmd_flags |= REQ_FUA;
Tejun Heo797e7db2006-01-06 09:51:03 +0100436 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200437 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100438 init_request_from_bio(rq, q->orig_bar_rq->bio);
439 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Tejun Heo797e7db2006-01-06 09:51:03 +0100441 /*
442 * Queue ordered sequence. As we stack them at the head, we
443 * need to queue in reverse order. Note that we rely on that
444 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
Jens Axboebf2de6f2007-09-27 13:01:25 +0200445 * request gets inbetween ordered sequence. If this request is
446 * an empty barrier, we don't need to do a postflush ever since
447 * there will be no data written between the pre and post flush.
448 * Hence a single flush will suffice.
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 */
Jens Axboebf2de6f2007-09-27 13:01:25 +0200450 if ((q->ordered & QUEUE_ORDERED_POSTFLUSH) && !blk_empty_barrier(rq))
Tejun Heo797e7db2006-01-06 09:51:03 +0100451 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
452 else
453 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Tejun Heo30e96562006-02-08 01:01:31 -0800455 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100456
457 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
458 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
459 rq = &q->pre_flush_rq;
460 } else
461 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
462
463 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
464 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
465 else
466 rq = NULL;
467
468 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
Jens Axboe165125e2007-07-24 09:28:11 +0200471int blk_do_ordered(struct request_queue *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800473 struct request *rq = *rqp;
Jens Axboebf2de6f2007-09-27 13:01:25 +0200474 const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Tejun Heo797e7db2006-01-06 09:51:03 +0100476 if (!q->ordseq) {
477 if (!is_barrier)
478 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Tejun Heo797e7db2006-01-06 09:51:03 +0100480 if (q->next_ordered != QUEUE_ORDERED_NONE) {
481 *rqp = start_ordered(q, rq);
482 return 1;
483 } else {
484 /*
485 * This can happen when the queue switches to
486 * ORDERED_NONE while this request is on it.
487 */
488 blkdev_dequeue_request(rq);
489 end_that_request_first(rq, -EOPNOTSUPP,
490 rq->hard_nr_sectors);
491 end_that_request_last(rq, -EOPNOTSUPP);
492 *rqp = NULL;
493 return 0;
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800497 /*
498 * Ordered sequence in progress
499 */
500
501 /* Special requests are not subject to ordering rules. */
502 if (!blk_fs_request(rq) &&
503 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
504 return 1;
505
Tejun Heo797e7db2006-01-06 09:51:03 +0100506 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800507 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100508 if (is_barrier && rq != &q->bar_rq)
509 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800510 } else {
511 /* Ordered by draining. Wait for turn. */
512 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
513 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
514 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516
517 return 1;
518}
519
NeilBrown5bb23a62007-09-27 12:46:13 +0200520static void req_bio_endio(struct request *rq, struct bio *bio,
521 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100522{
Jens Axboe165125e2007-07-24 09:28:11 +0200523 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100524
NeilBrown5bb23a62007-09-27 12:46:13 +0200525 if (&q->bar_rq != rq) {
526 if (error)
527 clear_bit(BIO_UPTODATE, &bio->bi_flags);
528 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
529 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100530
NeilBrown5bb23a62007-09-27 12:46:13 +0200531 if (unlikely(nbytes > bio->bi_size)) {
532 printk("%s: want %u bytes done, only %u left\n",
533 __FUNCTION__, nbytes, bio->bi_size);
534 nbytes = bio->bi_size;
535 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100536
NeilBrown5bb23a62007-09-27 12:46:13 +0200537 bio->bi_size -= nbytes;
538 bio->bi_sector += (nbytes >> 9);
539 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200540 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200541 } else {
542
543 /*
544 * Okay, this is the barrier request in progress, just
545 * record the error;
546 */
547 if (error && !q->orderr)
548 q->orderr = error;
549 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100550}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552/**
553 * blk_queue_bounce_limit - set bounce buffer limit for queue
554 * @q: the request queue for the device
555 * @dma_addr: bus address limit
556 *
557 * Description:
558 * Different hardware can have different requirements as to what pages
559 * it can do I/O directly to. A low level driver can call
560 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800561 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200563void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
565 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800566 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Andi Kleen5ee1af92006-03-08 17:57:26 -0800568 q->bounce_gfp = GFP_NOIO;
569#if BITS_PER_LONG == 64
570 /* Assume anything <= 4GB can be handled by IOMMU.
571 Actually some IOMMUs can handle everything, but I don't
572 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200573 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800574 dma = 1;
575 q->bounce_pfn = max_low_pfn;
576#else
577 if (bounce_pfn < blk_max_low_pfn)
578 dma = 1;
579 q->bounce_pfn = bounce_pfn;
580#endif
581 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 init_emergency_isa_pool();
583 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800584 q->bounce_pfn = bounce_pfn;
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
588EXPORT_SYMBOL(blk_queue_bounce_limit);
589
590/**
591 * blk_queue_max_sectors - set max sectors for a request for this queue
592 * @q: the request queue for the device
593 * @max_sectors: max sectors in the usual 512b unit
594 *
595 * Description:
596 * Enables a low level driver to set an upper limit on the size of
597 * received requests.
598 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200599void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
602 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
603 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
604 }
605
Mike Christiedefd94b2005-12-05 02:37:06 -0600606 if (BLK_DEF_MAX_SECTORS > max_sectors)
607 q->max_hw_sectors = q->max_sectors = max_sectors;
608 else {
609 q->max_sectors = BLK_DEF_MAX_SECTORS;
610 q->max_hw_sectors = max_sectors;
611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614EXPORT_SYMBOL(blk_queue_max_sectors);
615
616/**
617 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
618 * @q: the request queue for the device
619 * @max_segments: max number of segments
620 *
621 * Description:
622 * Enables a low level driver to set an upper limit on the number of
623 * physical data segments in a request. This would be the largest sized
624 * scatter list the driver could handle.
625 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200626void blk_queue_max_phys_segments(struct request_queue *q,
627 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 if (!max_segments) {
630 max_segments = 1;
631 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
632 }
633
634 q->max_phys_segments = max_segments;
635}
636
637EXPORT_SYMBOL(blk_queue_max_phys_segments);
638
639/**
640 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
641 * @q: the request queue for the device
642 * @max_segments: max number of segments
643 *
644 * Description:
645 * Enables a low level driver to set an upper limit on the number of
646 * hw data segments in a request. This would be the largest number of
647 * address/length pairs the host adapter can actually give as once
648 * to the device.
649 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200650void blk_queue_max_hw_segments(struct request_queue *q,
651 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 if (!max_segments) {
654 max_segments = 1;
655 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
656 }
657
658 q->max_hw_segments = max_segments;
659}
660
661EXPORT_SYMBOL(blk_queue_max_hw_segments);
662
663/**
664 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
665 * @q: the request queue for the device
666 * @max_size: max size of segment in bytes
667 *
668 * Description:
669 * Enables a low level driver to set an upper limit on the size of a
670 * coalesced segment
671 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200672void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
674 if (max_size < PAGE_CACHE_SIZE) {
675 max_size = PAGE_CACHE_SIZE;
676 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
677 }
678
679 q->max_segment_size = max_size;
680}
681
682EXPORT_SYMBOL(blk_queue_max_segment_size);
683
684/**
685 * blk_queue_hardsect_size - set hardware sector size for the queue
686 * @q: the request queue for the device
687 * @size: the hardware sector size, in bytes
688 *
689 * Description:
690 * This should typically be set to the lowest possible sector size
691 * that the hardware can operate on (possible without reverting to
692 * even internal read-modify-write operations). Usually the default
693 * of 512 covers most hardware.
694 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200695void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 q->hardsect_size = size;
698}
699
700EXPORT_SYMBOL(blk_queue_hardsect_size);
701
702/*
703 * Returns the minimum that is _not_ zero, unless both are zero.
704 */
705#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
706
707/**
708 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
709 * @t: the stacking driver (top)
710 * @b: the underlying device (bottom)
711 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200712void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600715 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
716 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
719 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
720 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
721 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800722 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
723 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
726EXPORT_SYMBOL(blk_queue_stack_limits);
727
728/**
729 * blk_queue_segment_boundary - set boundary rules for segment merging
730 * @q: the request queue for the device
731 * @mask: the memory boundary mask
732 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200733void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
735 if (mask < PAGE_CACHE_SIZE - 1) {
736 mask = PAGE_CACHE_SIZE - 1;
737 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
738 }
739
740 q->seg_boundary_mask = mask;
741}
742
743EXPORT_SYMBOL(blk_queue_segment_boundary);
744
745/**
746 * blk_queue_dma_alignment - set dma length and memory alignment
747 * @q: the request queue for the device
748 * @mask: alignment mask
749 *
750 * description:
751 * set required memory and length aligment for direct dma transactions.
752 * this is used when buiding direct io requests for the queue.
753 *
754 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200755void blk_queue_dma_alignment(struct request_queue *q, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756{
757 q->dma_alignment = mask;
758}
759
760EXPORT_SYMBOL(blk_queue_dma_alignment);
761
762/**
James Bottomley11c3e682007-12-31 16:37:00 -0600763 * blk_queue_update_dma_alignment - update dma length and memory alignment
764 * @q: the request queue for the device
765 * @mask: alignment mask
766 *
767 * description:
768 * update required memory and length aligment for direct dma transactions.
769 * If the requested alignment is larger than the current alignment, then
770 * the current queue alignment is updated to the new value, otherwise it
771 * is left alone. The design of this is to allow multiple objects
772 * (driver, device, transport etc) to set their respective
773 * alignments without having them interfere.
774 *
775 **/
776void blk_queue_update_dma_alignment(struct request_queue *q, int mask)
777{
778 BUG_ON(mask > PAGE_SIZE);
779
780 if (mask > q->dma_alignment)
781 q->dma_alignment = mask;
782}
783
784EXPORT_SYMBOL(blk_queue_update_dma_alignment);
785
786/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * @q: The request queue for the device
789 * @tag: The tag of the request
790 *
791 * Notes:
792 * Should be used when a device returns a tag and you want to match
793 * it with a request.
794 *
795 * no locks need be held.
796 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200797struct request *blk_queue_find_tag(struct request_queue *q, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
David C Somayajuluf583f492006-10-04 08:27:25 +0200799 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
802EXPORT_SYMBOL(blk_queue_find_tag);
803
804/**
James Bottomley492dfb42006-08-30 15:48:45 -0400805 * __blk_free_tags - release a given set of tag maintenance info
806 * @bqt: the tag map to free
807 *
808 * Tries to free the specified @bqt@. Returns true if it was
809 * actually freed and false if there are still references using it
810 */
811static int __blk_free_tags(struct blk_queue_tag *bqt)
812{
813 int retval;
814
815 retval = atomic_dec_and_test(&bqt->refcnt);
816 if (retval) {
817 BUG_ON(bqt->busy);
James Bottomley492dfb42006-08-30 15:48:45 -0400818
819 kfree(bqt->tag_index);
820 bqt->tag_index = NULL;
821
822 kfree(bqt->tag_map);
823 bqt->tag_map = NULL;
824
825 kfree(bqt);
826
827 }
828
829 return retval;
830}
831
832/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 * __blk_queue_free_tags - release tag maintenance info
834 * @q: the request queue for the device
835 *
836 * Notes:
837 * blk_cleanup_queue() will take care of calling this function, if tagging
838 * has been used. So there's no need to call this directly.
839 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200840static void __blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
842 struct blk_queue_tag *bqt = q->queue_tags;
843
844 if (!bqt)
845 return;
846
James Bottomley492dfb42006-08-30 15:48:45 -0400847 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 q->queue_tags = NULL;
850 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
851}
852
James Bottomley492dfb42006-08-30 15:48:45 -0400853
854/**
855 * blk_free_tags - release a given set of tag maintenance info
856 * @bqt: the tag map to free
857 *
858 * For externally managed @bqt@ frees the map. Callers of this
859 * function must guarantee to have released all the queues that
860 * might have been using this tag map.
861 */
862void blk_free_tags(struct blk_queue_tag *bqt)
863{
864 if (unlikely(!__blk_free_tags(bqt)))
865 BUG();
866}
867EXPORT_SYMBOL(blk_free_tags);
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869/**
870 * blk_queue_free_tags - release tag maintenance info
871 * @q: the request queue for the device
872 *
873 * Notes:
874 * This is used to disabled tagged queuing to a device, yet leave
875 * queue in function.
876 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200877void blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
879 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
880}
881
882EXPORT_SYMBOL(blk_queue_free_tags);
883
884static int
Jens Axboe165125e2007-07-24 09:28:11 +0200885init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 struct request **tag_index;
888 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700889 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
James Bottomley492dfb42006-08-30 15:48:45 -0400891 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 depth = q->nr_requests * 2;
893 printk(KERN_ERR "%s: adjusted depth to %d\n",
894 __FUNCTION__, depth);
895 }
896
Jens Axboef68110f2006-03-08 13:31:44 +0100897 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 if (!tag_index)
899 goto fail;
900
Tejun Heof7d37d02005-06-23 00:08:50 -0700901 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100902 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!tag_map)
904 goto fail;
905
Tejun Heoba025082005-08-05 13:28:11 -0700906 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 tags->tag_index = tag_index;
909 tags->tag_map = tag_map;
910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 return 0;
912fail:
913 kfree(tag_index);
914 return -ENOMEM;
915}
916
James Bottomley492dfb42006-08-30 15:48:45 -0400917static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
918 int depth)
919{
920 struct blk_queue_tag *tags;
921
922 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
923 if (!tags)
924 goto fail;
925
926 if (init_tag_map(q, tags, depth))
927 goto fail;
928
James Bottomley492dfb42006-08-30 15:48:45 -0400929 tags->busy = 0;
930 atomic_set(&tags->refcnt, 1);
931 return tags;
932fail:
933 kfree(tags);
934 return NULL;
935}
936
937/**
938 * blk_init_tags - initialize the tag info for an external tag map
939 * @depth: the maximum queue depth supported
940 * @tags: the tag to use
941 **/
942struct blk_queue_tag *blk_init_tags(int depth)
943{
944 return __blk_queue_init_tags(NULL, depth);
945}
946EXPORT_SYMBOL(blk_init_tags);
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948/**
949 * blk_queue_init_tags - initialize the queue tag info
950 * @q: the request queue for the device
951 * @depth: the maximum queue depth supported
952 * @tags: the tag to use
953 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200954int blk_queue_init_tags(struct request_queue *q, int depth,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 struct blk_queue_tag *tags)
956{
957 int rc;
958
959 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
960
961 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400962 tags = __blk_queue_init_tags(q, depth);
963
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 if (!tags)
965 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 } else if (q->queue_tags) {
967 if ((rc = blk_queue_resize_tags(q, depth)))
968 return rc;
969 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
970 return 0;
971 } else
972 atomic_inc(&tags->refcnt);
973
974 /*
975 * assign it, all done
976 */
977 q->queue_tags = tags;
978 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
Jens Axboe6eca9002007-10-25 10:14:47 +0200979 INIT_LIST_HEAD(&q->tag_busy_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 return 0;
981fail:
982 kfree(tags);
983 return -ENOMEM;
984}
985
986EXPORT_SYMBOL(blk_queue_init_tags);
987
988/**
989 * blk_queue_resize_tags - change the queueing depth
990 * @q: the request queue for the device
991 * @new_depth: the new max command queueing depth
992 *
993 * Notes:
994 * Must be called with the queue lock held.
995 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200996int blk_queue_resize_tags(struct request_queue *q, int new_depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
998 struct blk_queue_tag *bqt = q->queue_tags;
999 struct request **tag_index;
1000 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -07001001 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
1003 if (!bqt)
1004 return -ENXIO;
1005
1006 /*
Tejun Heoba025082005-08-05 13:28:11 -07001007 * if we already have large enough real_max_depth. just
1008 * adjust max_depth. *NOTE* as requests with tag value
1009 * between new_depth and real_max_depth can be in-flight, tag
1010 * map can not be shrunk blindly here.
1011 */
1012 if (new_depth <= bqt->real_max_depth) {
1013 bqt->max_depth = new_depth;
1014 return 0;
1015 }
1016
1017 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001018 * Currently cannot replace a shared tag map with a new
1019 * one, so error out if this is the case
1020 */
1021 if (atomic_read(&bqt->refcnt) != 1)
1022 return -EBUSY;
1023
1024 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 * save the old state info, so we can copy it back
1026 */
1027 tag_index = bqt->tag_index;
1028 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001029 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
1031 if (init_tag_map(q, bqt, new_depth))
1032 return -ENOMEM;
1033
1034 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001035 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001036 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 kfree(tag_index);
1039 kfree(tag_map);
1040 return 0;
1041}
1042
1043EXPORT_SYMBOL(blk_queue_resize_tags);
1044
1045/**
1046 * blk_queue_end_tag - end tag operations for a request
1047 * @q: the request queue for the device
1048 * @rq: the request that has completed
1049 *
1050 * Description:
1051 * Typically called when end_that_request_first() returns 0, meaning
1052 * all transfers have been done for a request. It's important to call
1053 * this function before end_that_request_last(), as that will put the
1054 * request back on the free list thus corrupting the internal tag list.
1055 *
1056 * Notes:
1057 * queue lock must be held.
1058 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001059void blk_queue_end_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060{
1061 struct blk_queue_tag *bqt = q->queue_tags;
1062 int tag = rq->tag;
1063
1064 BUG_ON(tag == -1);
1065
Tejun Heoba025082005-08-05 13:28:11 -07001066 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001067 /*
1068 * This can happen after tag depth has been reduced.
1069 * FIXME: how about a warning or info message here?
1070 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 return;
1072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001074 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 rq->tag = -1;
1076
1077 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001078 printk(KERN_ERR "%s: tag %d is missing\n",
1079 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 bqt->tag_index[tag] = NULL;
Jens Axboef3da54b2007-09-13 14:26:53 +02001082
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001083 if (unlikely(!test_bit(tag, bqt->tag_map))) {
Jens Axboef3da54b2007-09-13 14:26:53 +02001084 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1085 __FUNCTION__, tag);
1086 return;
1087 }
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001088 /*
1089 * The tag_map bit acts as a lock for tag_index[bit], so we need
1090 * unlock memory barrier semantics.
1091 */
1092 clear_bit_unlock(tag, bqt->tag_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 bqt->busy--;
1094}
1095
1096EXPORT_SYMBOL(blk_queue_end_tag);
1097
1098/**
1099 * blk_queue_start_tag - find a free tag and assign it
1100 * @q: the request queue for the device
1101 * @rq: the block request that needs tagging
1102 *
1103 * Description:
1104 * This can either be used as a stand-alone helper, or possibly be
1105 * assigned as the queue &prep_rq_fn (in which case &struct request
1106 * automagically gets a tag assigned). Note that this function
1107 * assumes that any type of request can be queued! if this is not
1108 * true for your device, you must check the request type before
1109 * calling this function. The request will also be removed from
1110 * the request queue, so it's the drivers responsibility to readd
1111 * it if it should need to be restarted for some reason.
1112 *
1113 * Notes:
1114 * queue lock must be held.
1115 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001116int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001119 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Jens Axboe4aff5e22006-08-10 08:44:47 +02001121 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001123 "%s: request %p for device [%s] already tagged %d",
1124 __FUNCTION__, rq,
1125 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 BUG();
1127 }
1128
Jens Axboe059af492006-09-21 20:37:22 +02001129 /*
1130 * Protect against shared tag maps, as we may not have exclusive
1131 * access to the tag map.
1132 */
1133 do {
1134 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1135 if (tag >= bqt->max_depth)
1136 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001138 } while (test_and_set_bit_lock(tag, bqt->tag_map));
Nick Piggindd941252007-09-14 08:41:12 +02001139 /*
Nick Pigginadb4ddb2007-10-24 10:54:38 +02001140 * We need lock ordering semantics given by test_and_set_bit_lock.
1141 * See blk_queue_end_tag for details.
Nick Piggindd941252007-09-14 08:41:12 +02001142 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Jens Axboe4aff5e22006-08-10 08:44:47 +02001144 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 rq->tag = tag;
1146 bqt->tag_index[tag] = rq;
1147 blkdev_dequeue_request(rq);
Jens Axboe6eca9002007-10-25 10:14:47 +02001148 list_add(&rq->queuelist, &q->tag_busy_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 bqt->busy++;
1150 return 0;
1151}
1152
1153EXPORT_SYMBOL(blk_queue_start_tag);
1154
1155/**
1156 * blk_queue_invalidate_tags - invalidate all pending tags
1157 * @q: the request queue for the device
1158 *
1159 * Description:
1160 * Hardware conditions may dictate a need to stop all pending requests.
1161 * In this case, we will safely clear the block side of the tag queue and
1162 * readd all requests to the request queue in the right order.
1163 *
1164 * Notes:
1165 * queue lock must be held.
1166 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001167void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 struct list_head *tmp, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Jens Axboed85532e2007-11-09 12:52:45 +01001171 list_for_each_safe(tmp, n, &q->tag_busy_list)
1172 blk_requeue_request(q, list_entry_rq(tmp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175EXPORT_SYMBOL(blk_queue_invalidate_tags);
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177void blk_dump_rq_flags(struct request *rq, char *msg)
1178{
1179 int bit;
1180
Jens Axboe4aff5e22006-08-10 08:44:47 +02001181 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1182 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1183 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
1185 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1186 rq->nr_sectors,
1187 rq->current_nr_sectors);
1188 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1189
Jens Axboe4aff5e22006-08-10 08:44:47 +02001190 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 printk("cdb: ");
1192 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1193 printk("%02x ", rq->cmd[bit]);
1194 printk("\n");
1195 }
1196}
1197
1198EXPORT_SYMBOL(blk_dump_rq_flags);
1199
Jens Axboe165125e2007-07-24 09:28:11 +02001200void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
NeilBrown9dfa5282007-08-16 13:27:52 +02001202 struct request rq;
1203 struct bio *nxt = bio->bi_next;
1204 rq.q = q;
1205 rq.bio = rq.biotail = bio;
1206 bio->bi_next = NULL;
1207 blk_recalc_rq_segments(&rq);
1208 bio->bi_next = nxt;
1209 bio->bi_phys_segments = rq.nr_phys_segments;
1210 bio->bi_hw_segments = rq.nr_hw_segments;
1211 bio->bi_flags |= (1 << BIO_SEG_VALID);
1212}
1213EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
NeilBrown9dfa5282007-08-16 13:27:52 +02001215static void blk_recalc_rq_segments(struct request *rq)
1216{
1217 int nr_phys_segs;
1218 int nr_hw_segs;
1219 unsigned int phys_size;
1220 unsigned int hw_size;
1221 struct bio_vec *bv, *bvprv = NULL;
1222 int seg_size;
1223 int hw_seg_size;
1224 int cluster;
NeilBrown5705f702007-09-25 12:35:59 +02001225 struct req_iterator iter;
NeilBrown9dfa5282007-08-16 13:27:52 +02001226 int high, highprv = 1;
1227 struct request_queue *q = rq->q;
1228
1229 if (!rq->bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return;
1231
1232 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
NeilBrown9dfa5282007-08-16 13:27:52 +02001233 hw_seg_size = seg_size = 0;
1234 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
NeilBrown5705f702007-09-25 12:35:59 +02001235 rq_for_each_segment(bv, rq, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 /*
1237 * the trick here is making sure that a high page is never
1238 * considered part of another segment, since that might
1239 * change with the bounce page.
1240 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001241 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (high || highprv)
1243 goto new_hw_segment;
1244 if (cluster) {
1245 if (seg_size + bv->bv_len > q->max_segment_size)
1246 goto new_segment;
1247 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1248 goto new_segment;
1249 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1250 goto new_segment;
1251 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1252 goto new_hw_segment;
1253
1254 seg_size += bv->bv_len;
1255 hw_seg_size += bv->bv_len;
1256 bvprv = bv;
1257 continue;
1258 }
1259new_segment:
1260 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
NeilBrown9dfa5282007-08-16 13:27:52 +02001261 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 hw_seg_size += bv->bv_len;
NeilBrown9dfa5282007-08-16 13:27:52 +02001263 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264new_hw_segment:
NeilBrown9dfa5282007-08-16 13:27:52 +02001265 if (nr_hw_segs == 1 &&
1266 hw_seg_size > rq->bio->bi_hw_front_size)
1267 rq->bio->bi_hw_front_size = hw_seg_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1269 nr_hw_segs++;
1270 }
1271
1272 nr_phys_segs++;
1273 bvprv = bv;
1274 seg_size = bv->bv_len;
1275 highprv = high;
1276 }
NeilBrown9dfa5282007-08-16 13:27:52 +02001277
1278 if (nr_hw_segs == 1 &&
1279 hw_seg_size > rq->bio->bi_hw_front_size)
1280 rq->bio->bi_hw_front_size = hw_seg_size;
1281 if (hw_seg_size > rq->biotail->bi_hw_back_size)
1282 rq->biotail->bi_hw_back_size = hw_seg_size;
1283 rq->nr_phys_segments = nr_phys_segs;
1284 rq->nr_hw_segments = nr_hw_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Jens Axboe165125e2007-07-24 09:28:11 +02001287static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 struct bio *nxt)
1289{
1290 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1291 return 0;
1292
1293 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1294 return 0;
1295 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1296 return 0;
1297
1298 /*
1299 * bio and nxt are contigous in memory, check if the queue allows
1300 * these two to be merged into one
1301 */
1302 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1303 return 1;
1304
1305 return 0;
1306}
1307
Jens Axboe165125e2007-07-24 09:28:11 +02001308static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 struct bio *nxt)
1310{
1311 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1312 blk_recount_segments(q, bio);
1313 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1314 blk_recount_segments(q, nxt);
1315 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001316 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001318 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 return 0;
1320
1321 return 1;
1322}
1323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324/*
1325 * map a request to scatterlist, return number of sg entries setup. Caller
1326 * must make sure sg can hold rq->nr_phys_segments entries
1327 */
Jens Axboe165125e2007-07-24 09:28:11 +02001328int blk_rq_map_sg(struct request_queue *q, struct request *rq,
Jens Axboef5659132007-09-21 10:44:19 +02001329 struct scatterlist *sglist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
1331 struct bio_vec *bvec, *bvprv;
NeilBrown5705f702007-09-25 12:35:59 +02001332 struct req_iterator iter;
Jens Axboeba951842007-10-17 19:34:11 +02001333 struct scatterlist *sg;
NeilBrown5705f702007-09-25 12:35:59 +02001334 int nsegs, cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
1336 nsegs = 0;
1337 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1338
1339 /*
1340 * for each bio in rq
1341 */
1342 bvprv = NULL;
Jens Axboeba951842007-10-17 19:34:11 +02001343 sg = NULL;
NeilBrown5705f702007-09-25 12:35:59 +02001344 rq_for_each_segment(bvec, rq, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +02001345 int nbytes = bvec->bv_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Jens Axboe6c92e692007-08-16 13:43:12 +02001347 if (bvprv && cluster) {
Jens Axboef5659132007-09-21 10:44:19 +02001348 if (sg->length + nbytes > q->max_segment_size)
Jens Axboe6c92e692007-08-16 13:43:12 +02001349 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Jens Axboe6c92e692007-08-16 13:43:12 +02001351 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1352 goto new_segment;
1353 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1354 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Jens Axboef5659132007-09-21 10:44:19 +02001356 sg->length += nbytes;
Jens Axboe6c92e692007-08-16 13:43:12 +02001357 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358new_segment:
Jens Axboeba951842007-10-17 19:34:11 +02001359 if (!sg)
1360 sg = sglist;
Jens Axboe7aeacf92007-10-23 09:49:25 +02001361 else {
1362 /*
1363 * If the driver previously mapped a shorter
1364 * list, we could see a termination bit
1365 * prematurely unless it fully inits the sg
1366 * table on each mapping. We KNOW that there
1367 * must be more entries here or the driver
1368 * would be buggy, so force clear the
1369 * termination bit to avoid doing a full
1370 * sg_init_table() in drivers for each command.
1371 */
1372 sg->page_link &= ~0x02;
Jens Axboeba951842007-10-17 19:34:11 +02001373 sg = sg_next(sg);
Jens Axboe7aeacf92007-10-23 09:49:25 +02001374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Jens Axboe642f1492007-10-24 11:20:47 +02001376 sg_set_page(sg, bvec->bv_page, nbytes, bvec->bv_offset);
Jens Axboe6c92e692007-08-16 13:43:12 +02001377 nsegs++;
1378 }
1379 bvprv = bvec;
NeilBrown5705f702007-09-25 12:35:59 +02001380 } /* segments in rq */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Jens Axboe9b617642007-10-22 19:39:33 +02001382 if (sg)
Jens Axboec46f2332007-10-31 12:06:37 +01001383 sg_mark_end(sg);
Jens Axboe9b617642007-10-22 19:39:33 +02001384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return nsegs;
1386}
1387
1388EXPORT_SYMBOL(blk_rq_map_sg);
1389
1390/*
1391 * the standard queue merge functions, can be overridden with device
1392 * specific ones if so desired
1393 */
1394
Jens Axboe165125e2007-07-24 09:28:11 +02001395static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 struct request *req,
1397 struct bio *bio)
1398{
1399 int nr_phys_segs = bio_phys_segments(q, bio);
1400
1401 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001402 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (req == q->last_merge)
1404 q->last_merge = NULL;
1405 return 0;
1406 }
1407
1408 /*
1409 * A hw segment is just getting larger, bump just the phys
1410 * counter.
1411 */
1412 req->nr_phys_segments += nr_phys_segs;
1413 return 1;
1414}
1415
Jens Axboe165125e2007-07-24 09:28:11 +02001416static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 struct request *req,
1418 struct bio *bio)
1419{
1420 int nr_hw_segs = bio_hw_segments(q, bio);
1421 int nr_phys_segs = bio_phys_segments(q, bio);
1422
1423 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1424 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001425 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (req == q->last_merge)
1427 q->last_merge = NULL;
1428 return 0;
1429 }
1430
1431 /*
1432 * This will form the start of a new hw segment. Bump both
1433 * counters.
1434 */
1435 req->nr_hw_segments += nr_hw_segs;
1436 req->nr_phys_segments += nr_phys_segs;
1437 return 1;
1438}
1439
NeilBrown3001ca72007-08-16 13:31:27 +02001440static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1441 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
Mike Christiedefd94b2005-12-05 02:37:06 -06001443 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 int len;
1445
Mike Christiedefd94b2005-12-05 02:37:06 -06001446 if (unlikely(blk_pc_request(req)))
1447 max_sectors = q->max_hw_sectors;
1448 else
1449 max_sectors = q->max_sectors;
1450
1451 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001452 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 if (req == q->last_merge)
1454 q->last_merge = NULL;
1455 return 0;
1456 }
1457 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1458 blk_recount_segments(q, req->biotail);
1459 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1460 blk_recount_segments(q, bio);
1461 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1462 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1463 !BIOVEC_VIRT_OVERSIZE(len)) {
1464 int mergeable = ll_new_mergeable(q, req, bio);
1465
1466 if (mergeable) {
1467 if (req->nr_hw_segments == 1)
1468 req->bio->bi_hw_front_size = len;
1469 if (bio->bi_hw_segments == 1)
1470 bio->bi_hw_back_size = len;
1471 }
1472 return mergeable;
1473 }
1474
1475 return ll_new_hw_segment(q, req, bio);
1476}
1477
Jens Axboe165125e2007-07-24 09:28:11 +02001478static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 struct bio *bio)
1480{
Mike Christiedefd94b2005-12-05 02:37:06 -06001481 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 int len;
1483
Mike Christiedefd94b2005-12-05 02:37:06 -06001484 if (unlikely(blk_pc_request(req)))
1485 max_sectors = q->max_hw_sectors;
1486 else
1487 max_sectors = q->max_sectors;
1488
1489
1490 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001491 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 if (req == q->last_merge)
1493 q->last_merge = NULL;
1494 return 0;
1495 }
1496 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1497 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1498 blk_recount_segments(q, bio);
1499 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1500 blk_recount_segments(q, req->bio);
1501 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1502 !BIOVEC_VIRT_OVERSIZE(len)) {
1503 int mergeable = ll_new_mergeable(q, req, bio);
1504
1505 if (mergeable) {
1506 if (bio->bi_hw_segments == 1)
1507 bio->bi_hw_front_size = len;
1508 if (req->nr_hw_segments == 1)
1509 req->biotail->bi_hw_back_size = len;
1510 }
1511 return mergeable;
1512 }
1513
1514 return ll_new_hw_segment(q, req, bio);
1515}
1516
Jens Axboe165125e2007-07-24 09:28:11 +02001517static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 struct request *next)
1519{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001520 int total_phys_segments;
1521 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
1523 /*
1524 * First check if the either of the requests are re-queued
1525 * requests. Can't merge them if they are.
1526 */
1527 if (req->special || next->special)
1528 return 0;
1529
1530 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001531 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 */
1533 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1534 return 0;
1535
1536 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1537 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1538 total_phys_segments--;
1539
1540 if (total_phys_segments > q->max_phys_segments)
1541 return 0;
1542
1543 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1544 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1545 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1546 /*
1547 * propagate the combined length to the end of the requests
1548 */
1549 if (req->nr_hw_segments == 1)
1550 req->bio->bi_hw_front_size = len;
1551 if (next->nr_hw_segments == 1)
1552 next->biotail->bi_hw_back_size = len;
1553 total_hw_segments--;
1554 }
1555
1556 if (total_hw_segments > q->max_hw_segments)
1557 return 0;
1558
1559 /* Merge is OK... */
1560 req->nr_phys_segments = total_phys_segments;
1561 req->nr_hw_segments = total_hw_segments;
1562 return 1;
1563}
1564
1565/*
1566 * "plug" the device if there are no outstanding requests: this will
1567 * force the transfer to start only after we have put all the requests
1568 * on the list.
1569 *
1570 * This is called with interrupts off and no requests on the queue and
1571 * with the queue lock held.
1572 */
Jens Axboe165125e2007-07-24 09:28:11 +02001573void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574{
1575 WARN_ON(!irqs_disabled());
1576
1577 /*
1578 * don't plug a stopped queue, it must be paired with blk_start_queue()
1579 * which will restart the queueing
1580 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001581 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 return;
1583
Jens Axboe2056a782006-03-23 20:00:26 +01001584 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001586 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588}
1589
1590EXPORT_SYMBOL(blk_plug_device);
1591
1592/*
1593 * remove the queue from the plugged list, if present. called with
1594 * queue lock held and interrupts disabled.
1595 */
Jens Axboe165125e2007-07-24 09:28:11 +02001596int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 WARN_ON(!irqs_disabled());
1599
1600 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1601 return 0;
1602
1603 del_timer(&q->unplug_timer);
1604 return 1;
1605}
1606
1607EXPORT_SYMBOL(blk_remove_plug);
1608
1609/*
1610 * remove the plug and let it rip..
1611 */
Jens Axboe165125e2007-07-24 09:28:11 +02001612void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001614 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 return;
1616
1617 if (!blk_remove_plug(q))
1618 return;
1619
Jens Axboe22e2c502005-06-27 10:55:12 +02001620 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622EXPORT_SYMBOL(__generic_unplug_device);
1623
1624/**
1625 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001626 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 *
1628 * Description:
1629 * Linux uses plugging to build bigger requests queues before letting
1630 * the device have at them. If a queue is plugged, the I/O scheduler
1631 * is still adding and merging requests on the queue. Once the queue
1632 * gets unplugged, the request_fn defined for the queue is invoked and
1633 * transfers started.
1634 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001635void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
1637 spin_lock_irq(q->queue_lock);
1638 __generic_unplug_device(q);
1639 spin_unlock_irq(q->queue_lock);
1640}
1641EXPORT_SYMBOL(generic_unplug_device);
1642
1643static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1644 struct page *page)
1645{
Jens Axboe165125e2007-07-24 09:28:11 +02001646 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05001648 blk_unplug(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649}
1650
David Howells65f27f32006-11-22 14:55:48 +00001651static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652{
Jens Axboe165125e2007-07-24 09:28:11 +02001653 struct request_queue *q =
1654 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Jens Axboe2056a782006-03-23 20:00:26 +01001656 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1657 q->rq.count[READ] + q->rq.count[WRITE]);
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 q->unplug_fn(q);
1660}
1661
1662static void blk_unplug_timeout(unsigned long data)
1663{
Jens Axboe165125e2007-07-24 09:28:11 +02001664 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Jens Axboe2056a782006-03-23 20:00:26 +01001666 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1667 q->rq.count[READ] + q->rq.count[WRITE]);
1668
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 kblockd_schedule_work(&q->unplug_work);
1670}
1671
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -05001672void blk_unplug(struct request_queue *q)
1673{
1674 /*
1675 * devices don't necessarily have an ->unplug_fn defined
1676 */
1677 if (q->unplug_fn) {
1678 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1679 q->rq.count[READ] + q->rq.count[WRITE]);
1680
1681 q->unplug_fn(q);
1682 }
1683}
1684EXPORT_SYMBOL(blk_unplug);
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686/**
1687 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001688 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 *
1690 * Description:
1691 * blk_start_queue() will clear the stop flag on the queue, and call
1692 * the request_fn for the queue if it was in a stopped state when
1693 * entered. Also see blk_stop_queue(). Queue lock must be held.
1694 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001695void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001697 WARN_ON(!irqs_disabled());
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1700
1701 /*
1702 * one level of recursion is ok and is much faster than kicking
1703 * the unplug handling
1704 */
1705 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1706 q->request_fn(q);
1707 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1708 } else {
1709 blk_plug_device(q);
1710 kblockd_schedule_work(&q->unplug_work);
1711 }
1712}
1713
1714EXPORT_SYMBOL(blk_start_queue);
1715
1716/**
1717 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001718 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 *
1720 * Description:
1721 * The Linux block layer assumes that a block driver will consume all
1722 * entries on the request queue when the request_fn strategy is called.
1723 * Often this will not happen, because of hardware limitations (queue
1724 * depth settings). If a device driver gets a 'queue full' response,
1725 * or if it simply chooses not to queue more I/O at one point, it can
1726 * call this function to prevent the request_fn from being called until
1727 * the driver has signalled it's ready to go again. This happens by calling
1728 * blk_start_queue() to restart queue operations. Queue lock must be held.
1729 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001730void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731{
1732 blk_remove_plug(q);
1733 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1734}
1735EXPORT_SYMBOL(blk_stop_queue);
1736
1737/**
1738 * blk_sync_queue - cancel any pending callbacks on a queue
1739 * @q: the queue
1740 *
1741 * Description:
1742 * The block layer may perform asynchronous callback activity
1743 * on a queue, such as calling the unplug function after a timeout.
1744 * A block device may call blk_sync_queue to ensure that any
1745 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001746 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * that its ->make_request_fn will not re-add plugging prior to calling
1748 * this function.
1749 *
1750 */
1751void blk_sync_queue(struct request_queue *q)
1752{
1753 del_timer_sync(&q->unplug_timer);
Oleg Nesterovabbeb882007-10-23 15:08:19 +02001754 kblockd_flush_work(&q->unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755}
1756EXPORT_SYMBOL(blk_sync_queue);
1757
1758/**
1759 * blk_run_queue - run a single device queue
1760 * @q: The queue to run
1761 */
1762void blk_run_queue(struct request_queue *q)
1763{
1764 unsigned long flags;
1765
1766 spin_lock_irqsave(q->queue_lock, flags);
1767 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001768
1769 /*
1770 * Only recurse once to avoid overrunning the stack, let the unplug
1771 * handling reinvoke the handler shortly if we already got there.
1772 */
1773 if (!elv_queue_empty(q)) {
1774 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1775 q->request_fn(q);
1776 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1777 } else {
1778 blk_plug_device(q);
1779 kblockd_schedule_work(&q->unplug_work);
1780 }
1781 }
1782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 spin_unlock_irqrestore(q->queue_lock, flags);
1784}
1785EXPORT_SYMBOL(blk_run_queue);
1786
1787/**
Jens Axboe165125e2007-07-24 09:28:11 +02001788 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001789 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 *
1791 * Description:
1792 * blk_cleanup_queue is the pair to blk_init_queue() or
1793 * blk_queue_make_request(). It should be called when a request queue is
1794 * being released; typically when a block device is being de-registered.
1795 * Currently, its primary task it to free all the &struct request
1796 * structures that were allocated to the queue and the queue itself.
1797 *
1798 * Caveat:
1799 * Hopefully the low level driver will have finished any
1800 * outstanding requests first...
1801 **/
Al Viro483f4af2006-03-18 18:34:37 -05001802static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
Jens Axboe165125e2007-07-24 09:28:11 +02001804 struct request_queue *q =
1805 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 struct request_list *rl = &q->rq;
1807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 blk_sync_queue(q);
1809
1810 if (rl->rq_pool)
1811 mempool_destroy(rl->rq_pool);
1812
1813 if (q->queue_tags)
1814 __blk_queue_free_tags(q);
1815
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001816 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001817
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001818 bdi_destroy(&q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 kmem_cache_free(requestq_cachep, q);
1820}
1821
Jens Axboe165125e2007-07-24 09:28:11 +02001822void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001823{
1824 kobject_put(&q->kobj);
1825}
1826EXPORT_SYMBOL(blk_put_queue);
1827
Jens Axboe165125e2007-07-24 09:28:11 +02001828void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001829{
1830 mutex_lock(&q->sysfs_lock);
1831 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1832 mutex_unlock(&q->sysfs_lock);
1833
1834 if (q->elevator)
1835 elevator_exit(q->elevator);
1836
1837 blk_put_queue(q);
1838}
1839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840EXPORT_SYMBOL(blk_cleanup_queue);
1841
Jens Axboe165125e2007-07-24 09:28:11 +02001842static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843{
1844 struct request_list *rl = &q->rq;
1845
1846 rl->count[READ] = rl->count[WRITE] = 0;
1847 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001848 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 init_waitqueue_head(&rl->wait[READ]);
1850 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
Christoph Lameter19460892005-06-23 00:08:19 -07001852 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1853 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
1855 if (!rl->rq_pool)
1856 return -ENOMEM;
1857
1858 return 0;
1859}
1860
Jens Axboe165125e2007-07-24 09:28:11 +02001861struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862{
Christoph Lameter19460892005-06-23 00:08:19 -07001863 return blk_alloc_queue_node(gfp_mask, -1);
1864}
1865EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
Al Viro483f4af2006-03-18 18:34:37 -05001867static struct kobj_type queue_ktype;
1868
Jens Axboe165125e2007-07-24 09:28:11 +02001869struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001870{
Jens Axboe165125e2007-07-24 09:28:11 +02001871 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001872 int err;
Christoph Lameter19460892005-06-23 00:08:19 -07001873
Christoph Lameter94f60302007-07-17 04:03:29 -07001874 q = kmem_cache_alloc_node(requestq_cachep,
1875 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 if (!q)
1877 return NULL;
1878
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001879 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1880 q->backing_dev_info.unplug_io_data = q;
1881 err = bdi_init(&q->backing_dev_info);
1882 if (err) {
1883 kmem_cache_free(requestq_cachep, q);
1884 return NULL;
1885 }
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001888
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07001889 kobject_init(&q->kobj, &queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Al Viro483f4af2006-03-18 18:34:37 -05001891 mutex_init(&q->sysfs_lock);
1892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 return q;
1894}
Christoph Lameter19460892005-06-23 00:08:19 -07001895EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897/**
1898 * blk_init_queue - prepare a request queue for use with a block device
1899 * @rfn: The function to be called to process requests that have been
1900 * placed on the queue.
1901 * @lock: Request queue spin lock
1902 *
1903 * Description:
1904 * If a block device wishes to use the standard request handling procedures,
1905 * which sorts requests and coalesces adjacent requests, then it must
1906 * call blk_init_queue(). The function @rfn will be called when there
1907 * are requests on the queue that need to be processed. If the device
1908 * supports plugging, then @rfn may not be called immediately when requests
1909 * are available on the queue, but may be called at some time later instead.
1910 * Plugged queues are generally unplugged when a buffer belonging to one
1911 * of the requests on the queue is needed, or due to memory pressure.
1912 *
1913 * @rfn is not required, or even expected, to remove all requests off the
1914 * queue, but only as many as it can handle at a time. If it does leave
1915 * requests on the queue, it is responsible for arranging that the requests
1916 * get dealt with eventually.
1917 *
1918 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001919 * request queue; this lock will be taken also from interrupt context, so irq
1920 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 *
1922 * Function returns a pointer to the initialized request queue, or NULL if
1923 * it didn't succeed.
1924 *
1925 * Note:
1926 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1927 * when the block device is deactivated (such as at module unload).
1928 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001929
Jens Axboe165125e2007-07-24 09:28:11 +02001930struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931{
Christoph Lameter19460892005-06-23 00:08:19 -07001932 return blk_init_queue_node(rfn, lock, -1);
1933}
1934EXPORT_SYMBOL(blk_init_queue);
1935
Jens Axboe165125e2007-07-24 09:28:11 +02001936struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001937blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1938{
Jens Axboe165125e2007-07-24 09:28:11 +02001939 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 if (!q)
1942 return NULL;
1943
Christoph Lameter19460892005-06-23 00:08:19 -07001944 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001945 if (blk_init_free_list(q)) {
1946 kmem_cache_free(requestq_cachep, q);
1947 return NULL;
1948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
152587d2005-04-12 16:22:06 -05001950 /*
1951 * if caller didn't supply a lock, they get per-queue locking with
1952 * our embedded lock
1953 */
1954 if (!lock) {
1955 spin_lock_init(&q->__queue_lock);
1956 lock = &q->__queue_lock;
1957 }
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 q->prep_rq_fn = NULL;
1961 q->unplug_fn = generic_unplug_device;
1962 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1963 q->queue_lock = lock;
1964
1965 blk_queue_segment_boundary(q, 0xffffffff);
1966
1967 blk_queue_make_request(q, __make_request);
1968 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1969
1970 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1971 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1972
Alan Stern44ec9542007-02-20 11:01:57 -05001973 q->sg_reserved_size = INT_MAX;
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 /*
1976 * all done
1977 */
1978 if (!elevator_init(q, NULL)) {
1979 blk_queue_congestion_threshold(q);
1980 return q;
1981 }
1982
Al Viro8669aaf2006-03-18 13:50:00 -05001983 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 return NULL;
1985}
Christoph Lameter19460892005-06-23 00:08:19 -07001986EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Jens Axboe165125e2007-07-24 09:28:11 +02001988int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001990 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001991 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return 0;
1993 }
1994
1995 return 1;
1996}
1997
1998EXPORT_SYMBOL(blk_get_queue);
1999
Jens Axboe165125e2007-07-24 09:28:11 +02002000static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002002 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02002003 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 mempool_free(rq, q->rq.rq_pool);
2005}
2006
Jens Axboe1ea25ec2006-07-18 22:24:11 +02002007static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02002008blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
2010 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
2011
2012 if (!rq)
2013 return NULL;
2014
2015 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02002016 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 * see bio.h and blkdev.h
2018 */
Jens Axboe49171e52006-08-10 08:59:11 +02002019 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
Tejun Heocb98fc82005-10-28 08:29:39 +02002021 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02002022 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02002023 mempool_free(rq, q->rq.rq_pool);
2024 return NULL;
2025 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02002026 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Tejun Heocb98fc82005-10-28 08:29:39 +02002029 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
2032/*
2033 * ioc_batching returns true if the ioc is a valid batching request and
2034 * should be given priority access to a request.
2035 */
Jens Axboe165125e2007-07-24 09:28:11 +02002036static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037{
2038 if (!ioc)
2039 return 0;
2040
2041 /*
2042 * Make sure the process is able to allocate at least 1 request
2043 * even if the batch times out, otherwise we could theoretically
2044 * lose wakeups.
2045 */
2046 return ioc->nr_batch_requests == q->nr_batching ||
2047 (ioc->nr_batch_requests > 0
2048 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2049}
2050
2051/*
2052 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2053 * will cause the process to be a "batcher" on all queues in the system. This
2054 * is the behaviour we want though - once it gets a wakeup it should be given
2055 * a nice run.
2056 */
Jens Axboe165125e2007-07-24 09:28:11 +02002057static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058{
2059 if (!ioc || ioc_batching(q, ioc))
2060 return;
2061
2062 ioc->nr_batch_requests = q->nr_batching;
2063 ioc->last_waited = jiffies;
2064}
2065
Jens Axboe165125e2007-07-24 09:28:11 +02002066static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067{
2068 struct request_list *rl = &q->rq;
2069
2070 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002071 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 if (waitqueue_active(&rl->wait[rw]))
2075 wake_up(&rl->wait[rw]);
2076
2077 blk_clear_queue_full(q, rw);
2078 }
2079}
2080
2081/*
2082 * A request has just been released. Account for it, update the full and
2083 * congestion status, wake up any waiters. Called under q->queue_lock.
2084 */
Jens Axboe165125e2007-07-24 09:28:11 +02002085static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086{
2087 struct request_list *rl = &q->rq;
2088
2089 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002090 if (priv)
2091 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092
2093 __freed_request(q, rw);
2094
2095 if (unlikely(rl->starved[rw ^ 1]))
2096 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097}
2098
2099#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2100/*
Nick Piggind6344532005-06-28 20:45:14 -07002101 * Get a free request, queue_lock must be held.
2102 * Returns NULL on failure, with queue_lock held.
2103 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 */
Jens Axboe165125e2007-07-24 09:28:11 +02002105static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002106 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107{
2108 struct request *rq = NULL;
2109 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002110 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002111 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002112 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Jens Axboe7749a8d2006-12-13 13:02:26 +01002114 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002115 if (may_queue == ELV_MQUEUE_NO)
2116 goto rq_starved;
2117
2118 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2119 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002120 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002121 /*
2122 * The queue will fill after this allocation, so set
2123 * it as full, and mark this process as "batching".
2124 * This process will be allowed to complete a batch of
2125 * requests, others will be blocked.
2126 */
2127 if (!blk_queue_full(q, rw)) {
2128 ioc_set_batching(q, ioc);
2129 blk_set_queue_full(q, rw);
2130 } else {
2131 if (may_queue != ELV_MQUEUE_MUST
2132 && !ioc_batching(q, ioc)) {
2133 /*
2134 * The queue is full and the allocating
2135 * process is not a "batcher", and not
2136 * exempted by the IO scheduler
2137 */
2138 goto out;
2139 }
2140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002142 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 }
2144
Jens Axboe082cf692005-06-28 16:35:11 +02002145 /*
2146 * Only allow batching queuers to allocate up to 50% over the defined
2147 * limit of requests, otherwise we could have thousands of requests
2148 * allocated with any setting of ->nr_requests
2149 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002150 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002151 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002152
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 rl->count[rw]++;
2154 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002155
Jens Axboe64521d12005-10-28 08:30:39 +02002156 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002157 if (priv)
2158 rl->elvpriv++;
2159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 spin_unlock_irq(q->queue_lock);
2161
Jens Axboe7749a8d2006-12-13 13:02:26 +01002162 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002163 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 /*
2165 * Allocation failed presumably due to memory. Undo anything
2166 * we might have messed up.
2167 *
2168 * Allocating task should really be put onto the front of the
2169 * wait queue, but this is pretty rare.
2170 */
2171 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002172 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
2174 /*
2175 * in the very unlikely event that allocation failed and no
2176 * requests for this direction was pending, mark us starved
2177 * so that freeing of a request in the other direction will
2178 * notice us. another possible fix would be to split the
2179 * rq mempool into READ and WRITE
2180 */
2181rq_starved:
2182 if (unlikely(rl->count[rw] == 0))
2183 rl->starved[rw] = 1;
2184
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 goto out;
2186 }
2187
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002188 /*
2189 * ioc may be NULL here, and ioc_batching will be false. That's
2190 * OK, if the queue is under the request limit then requests need
2191 * not count toward the nr_batch_requests limit. There will always
2192 * be some limit enforced by BLK_BATCH_TIME.
2193 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (ioc_batching(q, ioc))
2195 ioc->nr_batch_requests--;
2196
2197 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002198
2199 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 return rq;
2202}
2203
2204/*
2205 * No available requests for this queue, unplug the device and wait for some
2206 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002207 *
2208 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 */
Jens Axboe165125e2007-07-24 09:28:11 +02002210static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002211 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002213 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 struct request *rq;
2215
Jens Axboe7749a8d2006-12-13 13:02:26 +01002216 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002217 while (!rq) {
2218 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 struct request_list *rl = &q->rq;
2220
2221 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2222 TASK_UNINTERRUPTIBLE);
2223
Jens Axboe7749a8d2006-12-13 13:02:26 +01002224 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226 if (!rq) {
2227 struct io_context *ioc;
2228
Jens Axboe2056a782006-03-23 20:00:26 +01002229 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2230
Nick Piggind6344532005-06-28 20:45:14 -07002231 __generic_unplug_device(q);
2232 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 io_schedule();
2234
2235 /*
2236 * After sleeping, we become a "batching" process and
2237 * will be able to allocate at least one request, and
2238 * up to a big batch of them for a small period time.
2239 * See ioc_batching, ioc_set_batching
2240 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002241 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002243
2244 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 }
2246 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249 return rq;
2250}
2251
Jens Axboe165125e2007-07-24 09:28:11 +02002252struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253{
2254 struct request *rq;
2255
2256 BUG_ON(rw != READ && rw != WRITE);
2257
Nick Piggind6344532005-06-28 20:45:14 -07002258 spin_lock_irq(q->queue_lock);
2259 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002260 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002261 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002262 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002263 if (!rq)
2264 spin_unlock_irq(q->queue_lock);
2265 }
2266 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
2268 return rq;
2269}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270EXPORT_SYMBOL(blk_get_request);
2271
2272/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002273 * blk_start_queueing - initiate dispatch of requests to device
2274 * @q: request queue to kick into gear
2275 *
2276 * This is basically a helper to remove the need to know whether a queue
2277 * is plugged or not if someone just wants to initiate dispatch of requests
2278 * for this queue.
2279 *
2280 * The queue lock must be held with interrupts disabled.
2281 */
Jens Axboe165125e2007-07-24 09:28:11 +02002282void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002283{
2284 if (!blk_queue_plugged(q))
2285 q->request_fn(q);
2286 else
2287 __generic_unplug_device(q);
2288}
2289EXPORT_SYMBOL(blk_start_queueing);
2290
2291/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 * blk_requeue_request - put a request back on queue
2293 * @q: request queue where request should be inserted
2294 * @rq: request to be inserted
2295 *
2296 * Description:
2297 * Drivers often keep queueing requests until the hardware cannot accept
2298 * more, when that condition happens we need to put the request back
2299 * on the queue. Must be called with queue lock held.
2300 */
Jens Axboe165125e2007-07-24 09:28:11 +02002301void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302{
Jens Axboe2056a782006-03-23 20:00:26 +01002303 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 if (blk_rq_tagged(rq))
2306 blk_queue_end_tag(q, rq);
2307
2308 elv_requeue_request(q, rq);
2309}
2310
2311EXPORT_SYMBOL(blk_requeue_request);
2312
2313/**
2314 * blk_insert_request - insert a special request in to a request queue
2315 * @q: request queue where request should be inserted
2316 * @rq: request to be inserted
2317 * @at_head: insert request at head or tail of queue
2318 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 *
2320 * Description:
2321 * Many block devices need to execute commands asynchronously, so they don't
2322 * block the whole kernel from preemption during request execution. This is
2323 * accomplished normally by inserting aritficial requests tagged as
2324 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2325 * scheduled for actual execution by the request queue.
2326 *
2327 * We have the option of inserting the head or the tail of the queue.
2328 * Typically we use the tail for new ioctls and so forth. We use the head
2329 * of the queue for things like a QUEUE_FULL message from a device, or a
2330 * host that is unable to accept a particular command.
2331 */
Jens Axboe165125e2007-07-24 09:28:11 +02002332void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002333 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334{
Tejun Heo 867d1192005-04-24 02:06:05 -05002335 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 unsigned long flags;
2337
2338 /*
2339 * tell I/O scheduler that this isn't a regular read/write (ie it
2340 * must not attempt merges on this) and that it acts as a soft
2341 * barrier
2342 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002343 rq->cmd_type = REQ_TYPE_SPECIAL;
2344 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 rq->special = data;
2347
2348 spin_lock_irqsave(q->queue_lock, flags);
2349
2350 /*
2351 * If command is tagged, release the tag
2352 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002353 if (blk_rq_tagged(rq))
2354 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002356 drive_stat_acct(rq, 1);
Tejun Heo 867d1192005-04-24 02:06:05 -05002357 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002358 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 spin_unlock_irqrestore(q->queue_lock, flags);
2360}
2361
2362EXPORT_SYMBOL(blk_insert_request);
2363
Mike Christie0e75f902006-12-01 10:40:55 +01002364static int __blk_rq_unmap_user(struct bio *bio)
2365{
2366 int ret = 0;
2367
2368 if (bio) {
2369 if (bio_flagged(bio, BIO_USER_MAPPED))
2370 bio_unmap_user(bio);
2371 else
2372 ret = bio_uncopy_user(bio);
2373 }
2374
2375 return ret;
2376}
2377
NeilBrown3001ca72007-08-16 13:31:27 +02002378int blk_rq_append_bio(struct request_queue *q, struct request *rq,
2379 struct bio *bio)
2380{
2381 if (!rq->bio)
2382 blk_rq_bio_prep(q, rq, bio);
2383 else if (!ll_back_merge_fn(q, rq, bio))
2384 return -EINVAL;
2385 else {
2386 rq->biotail->bi_next = bio;
2387 rq->biotail = bio;
2388
2389 rq->data_len += bio->bi_size;
2390 }
2391 return 0;
2392}
2393EXPORT_SYMBOL(blk_rq_append_bio);
2394
Jens Axboe165125e2007-07-24 09:28:11 +02002395static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002396 void __user *ubuf, unsigned int len)
2397{
2398 unsigned long uaddr;
2399 struct bio *bio, *orig_bio;
2400 int reading, ret;
2401
2402 reading = rq_data_dir(rq) == READ;
2403
2404 /*
2405 * if alignment requirement is satisfied, map in user pages for
2406 * direct dma. else, set up kernel bounce buffers
2407 */
2408 uaddr = (unsigned long) ubuf;
2409 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2410 bio = bio_map_user(q, NULL, uaddr, len, reading);
2411 else
2412 bio = bio_copy_user(q, uaddr, len, reading);
2413
Jens Axboe29852592006-12-19 08:27:31 +01002414 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002415 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002416
2417 orig_bio = bio;
2418 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002419
Mike Christie0e75f902006-12-01 10:40:55 +01002420 /*
2421 * We link the bounce buffer in and could have to traverse it
2422 * later so we have to get a ref to prevent it from being freed
2423 */
2424 bio_get(bio);
2425
NeilBrown3001ca72007-08-16 13:31:27 +02002426 ret = blk_rq_append_bio(q, rq, bio);
2427 if (!ret)
2428 return bio->bi_size;
Mike Christie0e75f902006-12-01 10:40:55 +01002429
Mike Christie0e75f902006-12-01 10:40:55 +01002430 /* if it was boucned we must call the end io function */
NeilBrown6712ecf2007-09-27 12:47:43 +02002431 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002432 __blk_rq_unmap_user(orig_bio);
2433 bio_put(bio);
2434 return ret;
2435}
2436
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437/**
2438 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2439 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002440 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 * @ubuf: the user buffer
2442 * @len: length of user data
2443 *
2444 * Description:
2445 * Data will be mapped directly for zero copy io, if possible. Otherwise
2446 * a kernel bounce buffer is used.
2447 *
2448 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2449 * still in process context.
2450 *
2451 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2452 * before being submitted to the device, as pages mapped may be out of
2453 * reach. It's the callers responsibility to make sure this happens. The
2454 * original bio must be passed back in to blk_rq_unmap_user() for proper
2455 * unmapping.
2456 */
Jens Axboe165125e2007-07-24 09:28:11 +02002457int blk_rq_map_user(struct request_queue *q, struct request *rq,
2458 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459{
Mike Christie0e75f902006-12-01 10:40:55 +01002460 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002461 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002462 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Mike Christiedefd94b2005-12-05 02:37:06 -06002464 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002465 return -EINVAL;
2466 if (!len || !ubuf)
2467 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Mike Christie0e75f902006-12-01 10:40:55 +01002469 while (bytes_read != len) {
2470 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Mike Christie0e75f902006-12-01 10:40:55 +01002472 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2473 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2474 >> PAGE_SHIFT;
2475 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Mike Christie0e75f902006-12-01 10:40:55 +01002477 /*
2478 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2479 * pages. If this happens we just lower the requested
2480 * mapping len by a page so that we can fit
2481 */
2482 if (end - start > BIO_MAX_PAGES)
2483 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Mike Christie0e75f902006-12-01 10:40:55 +01002485 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2486 if (ret < 0)
2487 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002488 if (!bio)
2489 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002490 bytes_read += ret;
2491 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 }
2493
Mike Christie0e75f902006-12-01 10:40:55 +01002494 rq->buffer = rq->data = NULL;
2495 return 0;
2496unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002497 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002498 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499}
2500
2501EXPORT_SYMBOL(blk_rq_map_user);
2502
2503/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002504 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2505 * @q: request queue where request should be inserted
2506 * @rq: request to map data to
2507 * @iov: pointer to the iovec
2508 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002509 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002510 *
2511 * Description:
2512 * Data will be mapped directly for zero copy io, if possible. Otherwise
2513 * a kernel bounce buffer is used.
2514 *
2515 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2516 * still in process context.
2517 *
2518 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2519 * before being submitted to the device, as pages mapped may be out of
2520 * reach. It's the callers responsibility to make sure this happens. The
2521 * original bio must be passed back in to blk_rq_unmap_user() for proper
2522 * unmapping.
2523 */
Jens Axboe165125e2007-07-24 09:28:11 +02002524int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002525 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002526{
2527 struct bio *bio;
2528
2529 if (!iov || iov_count <= 0)
2530 return -EINVAL;
2531
2532 /* we don't allow misaligned data like bio_map_user() does. If the
2533 * user is using sg, they're expected to know the alignment constraints
2534 * and respect them accordingly */
2535 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2536 if (IS_ERR(bio))
2537 return PTR_ERR(bio);
2538
Mike Christie0e75f902006-12-01 10:40:55 +01002539 if (bio->bi_size != len) {
NeilBrown6712ecf2007-09-27 12:47:43 +02002540 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002541 bio_unmap_user(bio);
2542 return -EINVAL;
2543 }
2544
2545 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002546 blk_rq_bio_prep(q, rq, bio);
2547 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002548 return 0;
2549}
2550
2551EXPORT_SYMBOL(blk_rq_map_user_iov);
2552
2553/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002555 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 *
2557 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002558 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2559 * supply the original rq->bio from the blk_rq_map_user() return, since
2560 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002562int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002564 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002565 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002567 while (bio) {
2568 mapped_bio = bio;
2569 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002570 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
Jens Axboe48785bb2006-12-19 11:07:59 +01002572 ret2 = __blk_rq_unmap_user(mapped_bio);
2573 if (ret2 && !ret)
2574 ret = ret2;
2575
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002576 mapped_bio = bio;
2577 bio = bio->bi_next;
2578 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002579 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002580
2581 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582}
2583
2584EXPORT_SYMBOL(blk_rq_unmap_user);
2585
2586/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002587 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2588 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002589 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002590 * @kbuf: the kernel buffer
2591 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002592 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002593 */
Jens Axboe165125e2007-07-24 09:28:11 +02002594int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002595 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002596{
Mike Christie df46b9a2005-06-20 14:04:44 +02002597 struct bio *bio;
2598
Mike Christiedefd94b2005-12-05 02:37:06 -06002599 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002600 return -EINVAL;
2601 if (!len || !kbuf)
2602 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002603
2604 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002605 if (IS_ERR(bio))
2606 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002607
Jens Axboedd1cab92005-06-20 14:06:01 +02002608 if (rq_data_dir(rq) == WRITE)
2609 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002610
Jens Axboedd1cab92005-06-20 14:06:01 +02002611 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002612 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002613 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002614 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002615}
2616
2617EXPORT_SYMBOL(blk_rq_map_kern);
2618
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002619/**
2620 * blk_execute_rq_nowait - insert a request into queue for execution
2621 * @q: queue to insert the request in
2622 * @bd_disk: matching gendisk
2623 * @rq: request to insert
2624 * @at_head: insert request at head or tail of queue
2625 * @done: I/O completion handler
2626 *
2627 * Description:
2628 * Insert a fully prepared request at the back of the io scheduler queue
2629 * for execution. Don't wait for completion.
2630 */
Jens Axboe165125e2007-07-24 09:28:11 +02002631void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002632 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002633 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002634{
2635 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2636
2637 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002638 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002639 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002640 WARN_ON(irqs_disabled());
2641 spin_lock_irq(q->queue_lock);
2642 __elv_add_request(q, rq, where, 1);
2643 __generic_unplug_device(q);
2644 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002645}
Mike Christie6e39b692005-11-11 05:30:24 -06002646EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2647
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648/**
2649 * blk_execute_rq - insert a request into queue for execution
2650 * @q: queue to insert the request in
2651 * @bd_disk: matching gendisk
2652 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002653 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 *
2655 * Description:
2656 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002657 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 */
Jens Axboe165125e2007-07-24 09:28:11 +02002659int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002660 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002662 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 char sense[SCSI_SENSE_BUFFERSIZE];
2664 int err = 0;
2665
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 /*
2667 * we need an extra reference to the request, so we can look at
2668 * it after io completion
2669 */
2670 rq->ref_count++;
2671
2672 if (!rq->sense) {
2673 memset(sense, 0, sizeof(sense));
2674 rq->sense = sense;
2675 rq->sense_len = 0;
2676 }
2677
Jens Axboec00895a2006-09-30 20:29:12 +02002678 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002679 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
2682 if (rq->errors)
2683 err = -EIO;
2684
2685 return err;
2686}
2687
2688EXPORT_SYMBOL(blk_execute_rq);
2689
Jens Axboefd5d8062007-10-16 11:05:02 +02002690static void bio_end_empty_barrier(struct bio *bio, int err)
2691{
2692 if (err)
2693 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2694
2695 complete(bio->bi_private);
2696}
2697
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698/**
2699 * blkdev_issue_flush - queue a flush
2700 * @bdev: blockdev to issue flush for
2701 * @error_sector: error sector
2702 *
2703 * Description:
2704 * Issue a flush for the block device in question. Caller can supply
2705 * room for storing the error offset in case of a flush error, if they
2706 * wish to. Caller must run wait_for_completion() on its own.
2707 */
2708int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2709{
Jens Axboefd5d8062007-10-16 11:05:02 +02002710 DECLARE_COMPLETION_ONSTACK(wait);
Jens Axboe165125e2007-07-24 09:28:11 +02002711 struct request_queue *q;
Jens Axboefd5d8062007-10-16 11:05:02 +02002712 struct bio *bio;
2713 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
2715 if (bdev->bd_disk == NULL)
2716 return -ENXIO;
2717
2718 q = bdev_get_queue(bdev);
2719 if (!q)
2720 return -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721
Jens Axboefd5d8062007-10-16 11:05:02 +02002722 bio = bio_alloc(GFP_KERNEL, 0);
2723 if (!bio)
2724 return -ENOMEM;
2725
2726 bio->bi_end_io = bio_end_empty_barrier;
2727 bio->bi_private = &wait;
2728 bio->bi_bdev = bdev;
2729 submit_bio(1 << BIO_RW_BARRIER, bio);
2730
2731 wait_for_completion(&wait);
2732
2733 /*
2734 * The driver must store the error location in ->bi_sector, if
2735 * it supports it. For non-stacked drivers, this should be copied
2736 * from rq->sector.
2737 */
2738 if (error_sector)
2739 *error_sector = bio->bi_sector;
2740
2741 ret = 0;
2742 if (!bio_flagged(bio, BIO_UPTODATE))
2743 ret = -EIO;
2744
2745 bio_put(bio);
2746 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747}
2748
2749EXPORT_SYMBOL(blkdev_issue_flush);
2750
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002751static void drive_stat_acct(struct request *rq, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752{
2753 int rw = rq_data_dir(rq);
2754
2755 if (!blk_fs_request(rq) || !rq->rq_disk)
2756 return;
2757
Jens Axboed72d9042005-11-01 08:35:42 +01002758 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002759 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002760 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 disk_round_stats(rq->rq_disk);
2762 rq->rq_disk->in_flight++;
2763 }
2764}
2765
2766/*
2767 * add-request adds a request to the linked list.
2768 * queue lock is held and interrupts disabled, as we muck with the
2769 * request queue list.
2770 */
Jens Axboe165125e2007-07-24 09:28:11 +02002771static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772{
Jerome Marchandb238b3d2007-10-23 15:05:46 +02002773 drive_stat_acct(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 /*
2776 * elevator indicated where it wants this request to be
2777 * inserted at elevator_merge time
2778 */
2779 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2780}
2781
2782/*
2783 * disk_round_stats() - Round off the performance stats on a struct
2784 * disk_stats.
2785 *
2786 * The average IO queue length and utilisation statistics are maintained
2787 * by observing the current state of the queue length and the amount of
2788 * time it has been in this state for.
2789 *
2790 * Normally, that accounting is done on IO completion, but that can result
2791 * in more than a second's worth of IO being accounted for within any one
2792 * second, leading to >100% utilisation. To deal with that, we call this
2793 * function to do a round-off before returning the results when reading
2794 * /proc/diskstats. This accounts immediately for all queue usage up to
2795 * the current jiffies and restarts the counters again.
2796 */
2797void disk_round_stats(struct gendisk *disk)
2798{
2799 unsigned long now = jiffies;
2800
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002801 if (now == disk->stamp)
2802 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002804 if (disk->in_flight) {
2805 __disk_stat_add(disk, time_in_queue,
2806 disk->in_flight * (now - disk->stamp));
2807 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810}
2811
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002812EXPORT_SYMBOL_GPL(disk_round_stats);
2813
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814/*
2815 * queue lock must be held
2816 */
Jens Axboe165125e2007-07-24 09:28:11 +02002817void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 if (unlikely(!q))
2820 return;
2821 if (unlikely(--req->ref_count))
2822 return;
2823
Tejun Heo8922e162005-10-20 16:23:44 +02002824 elv_completed_request(q, req);
2825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 /*
2827 * Request may not have originated from ll_rw_blk. if not,
2828 * it didn't come out of our reserved rq pools
2829 */
Jens Axboe49171e52006-08-10 08:59:11 +02002830 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002832 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002835 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
2837 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002838 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 }
2840}
2841
Mike Christie6e39b692005-11-11 05:30:24 -06002842EXPORT_SYMBOL_GPL(__blk_put_request);
2843
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844void blk_put_request(struct request *req)
2845{
Tejun Heo8922e162005-10-20 16:23:44 +02002846 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002847 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
Tejun Heo8922e162005-10-20 16:23:44 +02002849 /*
2850 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2851 * following if (q) test.
2852 */
2853 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 spin_lock_irqsave(q->queue_lock, flags);
2855 __blk_put_request(q, req);
2856 spin_unlock_irqrestore(q->queue_lock, flags);
2857 }
2858}
2859
2860EXPORT_SYMBOL(blk_put_request);
2861
2862/**
2863 * blk_end_sync_rq - executes a completion event on a request
2864 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002865 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002867void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868{
Jens Axboec00895a2006-09-30 20:29:12 +02002869 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Jens Axboec00895a2006-09-30 20:29:12 +02002871 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 __blk_put_request(rq->q, rq);
2873
2874 /*
2875 * complete last, if this is a stack request the process (and thus
2876 * the rq pointer) could be invalid right after this complete()
2877 */
2878 complete(waiting);
2879}
2880EXPORT_SYMBOL(blk_end_sync_rq);
2881
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882/*
2883 * Has to be called with the request spinlock acquired
2884 */
Jens Axboe165125e2007-07-24 09:28:11 +02002885static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 struct request *next)
2887{
2888 if (!rq_mergeable(req) || !rq_mergeable(next))
2889 return 0;
2890
2891 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002892 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 */
2894 if (req->sector + req->nr_sectors != next->sector)
2895 return 0;
2896
2897 if (rq_data_dir(req) != rq_data_dir(next)
2898 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002899 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900 return 0;
2901
2902 /*
2903 * If we are allowed to merge, then append bio list
2904 * from next to rq and release next. merge_requests_fn
2905 * will have updated segment counts, update sector
2906 * counts here.
2907 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002908 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 return 0;
2910
2911 /*
2912 * At this point we have either done a back merge
2913 * or front merge. We need the smaller start_time of
2914 * the merged requests to be the current request
2915 * for accounting purposes.
2916 */
2917 if (time_after(req->start_time, next->start_time))
2918 req->start_time = next->start_time;
2919
2920 req->biotail->bi_next = next->bio;
2921 req->biotail = next->biotail;
2922
2923 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2924
2925 elv_merge_requests(q, req, next);
2926
2927 if (req->rq_disk) {
2928 disk_round_stats(req->rq_disk);
2929 req->rq_disk->in_flight--;
2930 }
2931
Jens Axboe22e2c502005-06-27 10:55:12 +02002932 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2933
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 __blk_put_request(q, next);
2935 return 1;
2936}
2937
Jens Axboe165125e2007-07-24 09:28:11 +02002938static inline int attempt_back_merge(struct request_queue *q,
2939 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940{
2941 struct request *next = elv_latter_request(q, rq);
2942
2943 if (next)
2944 return attempt_merge(q, rq, next);
2945
2946 return 0;
2947}
2948
Jens Axboe165125e2007-07-24 09:28:11 +02002949static inline int attempt_front_merge(struct request_queue *q,
2950 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951{
2952 struct request *prev = elv_former_request(q, rq);
2953
2954 if (prev)
2955 return attempt_merge(q, prev, rq);
2956
2957 return 0;
2958}
2959
Tejun Heo52d9e672006-01-06 09:49:58 +01002960static void init_request_from_bio(struct request *req, struct bio *bio)
2961{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002962 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002963
2964 /*
2965 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2966 */
2967 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002968 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002969
2970 /*
2971 * REQ_BARRIER implies no merging, but lets make it explicit
2972 */
2973 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002974 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002975
Jens Axboeb31dc662006-06-13 08:26:10 +02002976 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002977 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002978 if (bio_rw_meta(bio))
2979 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002980
Tejun Heo52d9e672006-01-06 09:49:58 +01002981 req->errors = 0;
2982 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01002983 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002984 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02002985 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002986}
2987
Jens Axboe165125e2007-07-24 09:28:11 +02002988static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989{
Nick Piggin450991b2005-06-28 20:45:13 -07002990 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002991 int el_ret, nr_sectors, barrier, err;
2992 const unsigned short prio = bio_prio(bio);
2993 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002994 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997
2998 /*
2999 * low level driver can indicate that it wants pages above a
3000 * certain limit bounced to low memory (ie for highmem, or even
3001 * ISA dma in theory)
3002 */
3003 blk_queue_bounce(q, &bio);
3004
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01003006 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 err = -EOPNOTSUPP;
3008 goto end_io;
3009 }
3010
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 spin_lock_irq(q->queue_lock);
3012
Nick Piggin450991b2005-06-28 20:45:13 -07003013 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 goto get_rq;
3015
3016 el_ret = elv_merge(q, &req, bio);
3017 switch (el_ret) {
3018 case ELEVATOR_BACK_MERGE:
3019 BUG_ON(!rq_mergeable(req));
3020
Jens Axboe1aa4f242006-12-19 08:33:11 +01003021 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 break;
3023
Jens Axboe2056a782006-03-23 20:00:26 +01003024 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
3025
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 req->biotail->bi_next = bio;
3027 req->biotail = bio;
3028 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003029 req->ioprio = ioprio_best(req->ioprio, prio);
Jerome Marchandb238b3d2007-10-23 15:05:46 +02003030 drive_stat_acct(req, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003032 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 goto out;
3034
3035 case ELEVATOR_FRONT_MERGE:
3036 BUG_ON(!rq_mergeable(req));
3037
Jens Axboe1aa4f242006-12-19 08:33:11 +01003038 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 break;
3040
Jens Axboe2056a782006-03-23 20:00:26 +01003041 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
3042
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 bio->bi_next = req->bio;
3044 req->bio = bio;
3045
3046 /*
3047 * may not be valid. if the low level driver said
3048 * it didn't need a bounce buffer then it better
3049 * not touch req->buffer either...
3050 */
3051 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02003052 req->current_nr_sectors = bio_cur_sectors(bio);
3053 req->hard_cur_sectors = req->current_nr_sectors;
3054 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003056 req->ioprio = ioprio_best(req->ioprio, prio);
Jerome Marchandb238b3d2007-10-23 15:05:46 +02003057 drive_stat_acct(req, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003059 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 goto out;
3061
Nick Piggin450991b2005-06-28 20:45:13 -07003062 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003064 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 }
3066
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003068 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003069 * This sync check and mask will be re-done in init_request_from_bio(),
3070 * but we need to set it earlier to expose the sync flag to the
3071 * rq allocator and io schedulers.
3072 */
3073 rw_flags = bio_data_dir(bio);
3074 if (sync)
3075 rw_flags |= REQ_RW_SYNC;
3076
3077 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003078 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003079 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003080 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003081 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003082
Nick Piggin450991b2005-06-28 20:45:13 -07003083 /*
3084 * After dropping the lock and possibly sleeping here, our request
3085 * may now be mergeable after it had proven unmergeable (above).
3086 * We don't worry about that case for efficiency. It won't happen
3087 * often, and the elevators are able to handle it.
3088 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003089 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090
Nick Piggin450991b2005-06-28 20:45:13 -07003091 spin_lock_irq(q->queue_lock);
3092 if (elv_queue_empty(q))
3093 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 add_request(q, req);
3095out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003096 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 __generic_unplug_device(q);
3098
3099 spin_unlock_irq(q->queue_lock);
3100 return 0;
3101
3102end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003103 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 return 0;
3105}
3106
3107/*
3108 * If bio->bi_dev is a partition, remap the location
3109 */
3110static inline void blk_partition_remap(struct bio *bio)
3111{
3112 struct block_device *bdev = bio->bi_bdev;
3113
Jens Axboebf2de6f2007-09-27 13:01:25 +02003114 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003116 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117
Jens Axboea3623572005-11-01 09:26:16 +01003118 p->sectors[rw] += bio_sectors(bio);
3119 p->ios[rw]++;
3120
Linus Torvalds1da177e2005-04-16 15:20:36 -07003121 bio->bi_sector += p->start_sect;
3122 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003123
3124 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3125 bdev->bd_dev, bio->bi_sector,
3126 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 }
3128}
3129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130static void handle_bad_sector(struct bio *bio)
3131{
3132 char b[BDEVNAME_SIZE];
3133
3134 printk(KERN_INFO "attempt to access beyond end of device\n");
3135 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3136 bdevname(bio->bi_bdev, b),
3137 bio->bi_rw,
3138 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3139 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3140
3141 set_bit(BIO_EOF, &bio->bi_flags);
3142}
3143
Akinobu Mitac17bb492006-12-08 02:39:46 -08003144#ifdef CONFIG_FAIL_MAKE_REQUEST
3145
3146static DECLARE_FAULT_ATTR(fail_make_request);
3147
3148static int __init setup_fail_make_request(char *str)
3149{
3150 return setup_fault_attr(&fail_make_request, str);
3151}
3152__setup("fail_make_request=", setup_fail_make_request);
3153
3154static int should_fail_request(struct bio *bio)
3155{
3156 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3157 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3158 return should_fail(&fail_make_request, bio->bi_size);
3159
3160 return 0;
3161}
3162
3163static int __init fail_make_request_debugfs(void)
3164{
3165 return init_fault_attr_dentries(&fail_make_request,
3166 "fail_make_request");
3167}
3168
3169late_initcall(fail_make_request_debugfs);
3170
3171#else /* CONFIG_FAIL_MAKE_REQUEST */
3172
3173static inline int should_fail_request(struct bio *bio)
3174{
3175 return 0;
3176}
3177
3178#endif /* CONFIG_FAIL_MAKE_REQUEST */
3179
Jens Axboec07e2b42007-07-18 13:27:58 +02003180/*
3181 * Check whether this bio extends beyond the end of the device.
3182 */
3183static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
3184{
3185 sector_t maxsector;
3186
3187 if (!nr_sectors)
3188 return 0;
3189
3190 /* Test device or partition size, when known. */
3191 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3192 if (maxsector) {
3193 sector_t sector = bio->bi_sector;
3194
3195 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3196 /*
3197 * This may well happen - the kernel calls bread()
3198 * without checking the size of the device, e.g., when
3199 * mounting a device.
3200 */
3201 handle_bad_sector(bio);
3202 return 1;
3203 }
3204 }
3205
3206 return 0;
3207}
3208
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209/**
3210 * generic_make_request: hand a buffer to its device driver for I/O
3211 * @bio: The bio describing the location in memory and on the device.
3212 *
3213 * generic_make_request() is used to make I/O requests of block
3214 * devices. It is passed a &struct bio, which describes the I/O that needs
3215 * to be done.
3216 *
3217 * generic_make_request() does not return any status. The
3218 * success/failure status of the request, along with notification of
3219 * completion, is delivered asynchronously through the bio->bi_end_io
3220 * function described (one day) else where.
3221 *
3222 * The caller of generic_make_request must make sure that bi_io_vec
3223 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3224 * set to describe the device address, and the
3225 * bi_end_io and optionally bi_private are set to describe how
3226 * completion notification should be signaled.
3227 *
3228 * generic_make_request and the drivers it calls may use bi_next if this
3229 * bio happens to be merged with someone else, and may change bi_dev and
3230 * bi_sector for remaps as it sees fit. So the values of these fields
3231 * should NOT be depended on after the call to generic_make_request.
3232 */
Neil Brownd89d8792007-05-01 09:53:42 +02003233static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
Jens Axboe165125e2007-07-24 09:28:11 +02003235 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08003236 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003238 dev_t old_dev;
Jens Axboe51fd77b2007-11-02 08:49:08 +01003239 int err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240
3241 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242
Jens Axboec07e2b42007-07-18 13:27:58 +02003243 if (bio_check_eod(bio, nr_sectors))
3244 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245
3246 /*
3247 * Resolve the mapping until finished. (drivers are
3248 * still free to implement/resolve their own stacking
3249 * by explicitly returning 0)
3250 *
3251 * NOTE: we don't repeat the blk_size check for each new device.
3252 * Stacking drivers are expected to know what they are doing.
3253 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003254 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003255 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 do {
3257 char b[BDEVNAME_SIZE];
3258
3259 q = bdev_get_queue(bio->bi_bdev);
3260 if (!q) {
3261 printk(KERN_ERR
3262 "generic_make_request: Trying to access "
3263 "nonexistent block-device %s (%Lu)\n",
3264 bdevname(bio->bi_bdev, b),
3265 (long long) bio->bi_sector);
3266end_io:
Jens Axboe51fd77b2007-11-02 08:49:08 +01003267 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 break;
3269 }
3270
Jens Axboe4fa253f2007-07-18 13:13:10 +02003271 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272 printk("bio too big device %s (%u > %u)\n",
3273 bdevname(bio->bi_bdev, b),
3274 bio_sectors(bio),
3275 q->max_hw_sectors);
3276 goto end_io;
3277 }
3278
Nick Pigginfde6ad22005-06-23 00:08:53 -07003279 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 goto end_io;
3281
Akinobu Mitac17bb492006-12-08 02:39:46 -08003282 if (should_fail_request(bio))
3283 goto end_io;
3284
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 /*
3286 * If this device has partitions, remap block n
3287 * of partition p to block n+start(p) of the disk.
3288 */
3289 blk_partition_remap(bio);
3290
NeilBrown5ddfe962006-10-30 22:07:21 -08003291 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02003292 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003293 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003294
3295 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3296
NeilBrown5ddfe962006-10-30 22:07:21 -08003297 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003298 old_dev = bio->bi_bdev->bd_dev;
3299
Jens Axboec07e2b42007-07-18 13:27:58 +02003300 if (bio_check_eod(bio, nr_sectors))
3301 goto end_io;
Jens Axboe51fd77b2007-11-02 08:49:08 +01003302 if (bio_empty_barrier(bio) && !q->prepare_flush_fn) {
3303 err = -EOPNOTSUPP;
3304 goto end_io;
3305 }
NeilBrown5ddfe962006-10-30 22:07:21 -08003306
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 ret = q->make_request_fn(q, bio);
3308 } while (ret);
3309}
3310
Neil Brownd89d8792007-05-01 09:53:42 +02003311/*
3312 * We only want one ->make_request_fn to be active at a time,
3313 * else stack usage with stacked devices could be a problem.
3314 * So use current->bio_{list,tail} to keep a list of requests
3315 * submited by a make_request_fn function.
3316 * current->bio_tail is also used as a flag to say if
3317 * generic_make_request is currently active in this task or not.
3318 * If it is NULL, then no make_request is active. If it is non-NULL,
3319 * then a make_request is active, and new requests should be added
3320 * at the tail
3321 */
3322void generic_make_request(struct bio *bio)
3323{
3324 if (current->bio_tail) {
3325 /* make_request is active */
3326 *(current->bio_tail) = bio;
3327 bio->bi_next = NULL;
3328 current->bio_tail = &bio->bi_next;
3329 return;
3330 }
3331 /* following loop may be a bit non-obvious, and so deserves some
3332 * explanation.
3333 * Before entering the loop, bio->bi_next is NULL (as all callers
3334 * ensure that) so we have a list with a single bio.
3335 * We pretend that we have just taken it off a longer list, so
3336 * we assign bio_list to the next (which is NULL) and bio_tail
3337 * to &bio_list, thus initialising the bio_list of new bios to be
3338 * added. __generic_make_request may indeed add some more bios
3339 * through a recursive call to generic_make_request. If it
3340 * did, we find a non-NULL value in bio_list and re-enter the loop
3341 * from the top. In this case we really did just take the bio
3342 * of the top of the list (no pretending) and so fixup bio_list and
3343 * bio_tail or bi_next, and call into __generic_make_request again.
3344 *
3345 * The loop was structured like this to make only one call to
3346 * __generic_make_request (which is important as it is large and
3347 * inlined) and to keep the structure simple.
3348 */
3349 BUG_ON(bio->bi_next);
3350 do {
3351 current->bio_list = bio->bi_next;
3352 if (bio->bi_next == NULL)
3353 current->bio_tail = &current->bio_list;
3354 else
3355 bio->bi_next = NULL;
3356 __generic_make_request(bio);
3357 bio = current->bio_list;
3358 } while (bio);
3359 current->bio_tail = NULL; /* deactivate */
3360}
3361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362EXPORT_SYMBOL(generic_make_request);
3363
3364/**
3365 * submit_bio: submit a bio to the block device layer for I/O
3366 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3367 * @bio: The &struct bio which describes the I/O
3368 *
3369 * submit_bio() is very similar in purpose to generic_make_request(), and
3370 * uses that function to do most of the work. Both are fairly rough
3371 * interfaces, @bio must be presetup and ready for I/O.
3372 *
3373 */
3374void submit_bio(int rw, struct bio *bio)
3375{
3376 int count = bio_sectors(bio);
3377
Jens Axboe22e2c502005-06-27 10:55:12 +02003378 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
Jens Axboebf2de6f2007-09-27 13:01:25 +02003380 /*
3381 * If it's a regular read/write or a barrier with data attached,
3382 * go through the normal accounting stuff before submission.
3383 */
3384 if (!bio_empty_barrier(bio)) {
3385
3386 BIO_BUG_ON(!bio->bi_size);
3387 BIO_BUG_ON(!bio->bi_io_vec);
3388
3389 if (rw & WRITE) {
3390 count_vm_events(PGPGOUT, count);
3391 } else {
3392 task_io_account_read(bio->bi_size);
3393 count_vm_events(PGPGIN, count);
3394 }
3395
3396 if (unlikely(block_dump)) {
3397 char b[BDEVNAME_SIZE];
3398 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07003399 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02003400 (rw & WRITE) ? "WRITE" : "READ",
3401 (unsigned long long)bio->bi_sector,
3402 bdevname(bio->bi_bdev,b));
3403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404 }
3405
3406 generic_make_request(bio);
3407}
3408
3409EXPORT_SYMBOL(submit_bio);
3410
Adrian Bunk93d17d32005-06-25 14:59:10 -07003411static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412{
3413 if (blk_fs_request(rq)) {
3414 rq->hard_sector += nsect;
3415 rq->hard_nr_sectors -= nsect;
3416
3417 /*
3418 * Move the I/O submission pointers ahead if required.
3419 */
3420 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3421 (rq->sector <= rq->hard_sector)) {
3422 rq->sector = rq->hard_sector;
3423 rq->nr_sectors = rq->hard_nr_sectors;
3424 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3425 rq->current_nr_sectors = rq->hard_cur_sectors;
3426 rq->buffer = bio_data(rq->bio);
3427 }
3428
3429 /*
3430 * if total number of sectors is less than the first segment
3431 * size, something has gone terribly wrong
3432 */
3433 if (rq->nr_sectors < rq->current_nr_sectors) {
3434 printk("blk: request botched\n");
3435 rq->nr_sectors = rq->current_nr_sectors;
3436 }
3437 }
3438}
3439
3440static int __end_that_request_first(struct request *req, int uptodate,
3441 int nr_bytes)
3442{
3443 int total_bytes, bio_nbytes, error, next_idx = 0;
3444 struct bio *bio;
3445
Jens Axboe2056a782006-03-23 20:00:26 +01003446 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3447
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448 /*
3449 * extend uptodate bool to allow < 0 value to be direct io error
3450 */
3451 error = 0;
3452 if (end_io_error(uptodate))
3453 error = !uptodate ? -EIO : uptodate;
3454
3455 /*
3456 * for a REQ_BLOCK_PC request, we want to carry any eventual
3457 * sense key with us all the way through
3458 */
3459 if (!blk_pc_request(req))
3460 req->errors = 0;
3461
3462 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003463 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 printk("end_request: I/O error, dev %s, sector %llu\n",
3465 req->rq_disk ? req->rq_disk->disk_name : "?",
3466 (unsigned long long)req->sector);
3467 }
3468
Jens Axboed72d9042005-11-01 08:35:42 +01003469 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003470 const int rw = rq_data_dir(req);
3471
Jens Axboe53e86062006-01-17 11:09:27 +01003472 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003473 }
3474
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 total_bytes = bio_nbytes = 0;
3476 while ((bio = req->bio) != NULL) {
3477 int nbytes;
3478
Jens Axboebf2de6f2007-09-27 13:01:25 +02003479 /*
3480 * For an empty barrier request, the low level driver must
3481 * store a potential error location in ->sector. We pass
3482 * that back up in ->bi_sector.
3483 */
3484 if (blk_empty_barrier(req))
3485 bio->bi_sector = req->sector;
3486
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 if (nr_bytes >= bio->bi_size) {
3488 req->bio = bio->bi_next;
3489 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02003490 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 next_idx = 0;
3492 bio_nbytes = 0;
3493 } else {
3494 int idx = bio->bi_idx + next_idx;
3495
3496 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3497 blk_dump_rq_flags(req, "__end_that");
3498 printk("%s: bio idx %d >= vcnt %d\n",
3499 __FUNCTION__,
3500 bio->bi_idx, bio->bi_vcnt);
3501 break;
3502 }
3503
3504 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3505 BIO_BUG_ON(nbytes > bio->bi_size);
3506
3507 /*
3508 * not a complete bvec done
3509 */
3510 if (unlikely(nbytes > nr_bytes)) {
3511 bio_nbytes += nr_bytes;
3512 total_bytes += nr_bytes;
3513 break;
3514 }
3515
3516 /*
3517 * advance to the next vector
3518 */
3519 next_idx++;
3520 bio_nbytes += nbytes;
3521 }
3522
3523 total_bytes += nbytes;
3524 nr_bytes -= nbytes;
3525
3526 if ((bio = req->bio)) {
3527 /*
3528 * end more in this run, or just return 'not-done'
3529 */
3530 if (unlikely(nr_bytes <= 0))
3531 break;
3532 }
3533 }
3534
3535 /*
3536 * completely done
3537 */
3538 if (!req->bio)
3539 return 0;
3540
3541 /*
3542 * if the request wasn't completed, update state
3543 */
3544 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02003545 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546 bio->bi_idx += next_idx;
3547 bio_iovec(bio)->bv_offset += nr_bytes;
3548 bio_iovec(bio)->bv_len -= nr_bytes;
3549 }
3550
3551 blk_recalc_rq_sectors(req, total_bytes >> 9);
3552 blk_recalc_rq_segments(req);
3553 return 1;
3554}
3555
3556/**
3557 * end_that_request_first - end I/O on a request
3558 * @req: the request being processed
3559 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3560 * @nr_sectors: number of sectors to end I/O on
3561 *
3562 * Description:
3563 * Ends I/O on a number of sectors attached to @req, and sets it up
3564 * for the next range of segments (if any) in the cluster.
3565 *
3566 * Return:
3567 * 0 - we are done with this request, call end_that_request_last()
3568 * 1 - still buffers pending for this request
3569 **/
3570int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3571{
3572 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3573}
3574
3575EXPORT_SYMBOL(end_that_request_first);
3576
3577/**
3578 * end_that_request_chunk - end I/O on a request
3579 * @req: the request being processed
3580 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3581 * @nr_bytes: number of bytes to complete
3582 *
3583 * Description:
3584 * Ends I/O on a number of bytes attached to @req, and sets it up
3585 * for the next range of segments (if any). Like end_that_request_first(),
3586 * but deals with bytes instead of sectors.
3587 *
3588 * Return:
3589 * 0 - we are done with this request, call end_that_request_last()
3590 * 1 - still buffers pending for this request
3591 **/
3592int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3593{
3594 return __end_that_request_first(req, uptodate, nr_bytes);
3595}
3596
3597EXPORT_SYMBOL(end_that_request_chunk);
3598
3599/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003600 * splice the completion data to a local structure and hand off to
3601 * process_completion_queue() to complete the requests
3602 */
3603static void blk_done_softirq(struct softirq_action *h)
3604{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003605 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003606
3607 local_irq_disable();
3608 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003609 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003610 local_irq_enable();
3611
3612 while (!list_empty(&local_list)) {
3613 struct request *rq = list_entry(local_list.next, struct request, donelist);
3614
3615 list_del_init(&rq->donelist);
3616 rq->q->softirq_done_fn(rq);
3617 }
3618}
3619
Satyam Sharmadb47d472007-08-23 09:29:40 +02003620static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action,
Jens Axboeff856ba2006-01-09 16:02:34 +01003621 void *hcpu)
3622{
3623 /*
3624 * If a CPU goes away, splice its entries to the current CPU
3625 * and trigger a run of the softirq
3626 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003627 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003628 int cpu = (unsigned long) hcpu;
3629
3630 local_irq_disable();
3631 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3632 &__get_cpu_var(blk_cpu_done));
3633 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3634 local_irq_enable();
3635 }
3636
3637 return NOTIFY_OK;
3638}
3639
3640
Satyam Sharmadb47d472007-08-23 09:29:40 +02003641static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003642 .notifier_call = blk_cpu_notify,
3643};
3644
Jens Axboeff856ba2006-01-09 16:02:34 +01003645/**
3646 * blk_complete_request - end I/O on a request
3647 * @req: the request being processed
3648 *
3649 * Description:
3650 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003651 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02003652 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01003653 * through a softirq handler. The user must have registered a completion
3654 * callback through blk_queue_softirq_done().
3655 **/
3656
3657void blk_complete_request(struct request *req)
3658{
3659 struct list_head *cpu_list;
3660 unsigned long flags;
3661
3662 BUG_ON(!req->q->softirq_done_fn);
3663
3664 local_irq_save(flags);
3665
3666 cpu_list = &__get_cpu_var(blk_cpu_done);
3667 list_add_tail(&req->donelist, cpu_list);
3668 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3669
3670 local_irq_restore(flags);
3671}
3672
3673EXPORT_SYMBOL(blk_complete_request);
3674
3675/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 * queue lock must be held
3677 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003678void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679{
3680 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003681 int error;
3682
3683 /*
3684 * extend uptodate bool to allow < 0 value to be direct io error
3685 */
3686 error = 0;
3687 if (end_io_error(uptodate))
3688 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689
3690 if (unlikely(laptop_mode) && blk_fs_request(req))
3691 laptop_io_completion();
3692
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003693 /*
3694 * Account IO completion. bar_rq isn't accounted as a normal
3695 * IO on queueing nor completion. Accounting the containing
3696 * request is enough.
3697 */
3698 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003700 const int rw = rq_data_dir(req);
3701
3702 __disk_stat_inc(disk, ios[rw]);
3703 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 disk_round_stats(disk);
3705 disk->in_flight--;
3706 }
3707 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003708 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 else
3710 __blk_put_request(req->q, req);
3711}
3712
3713EXPORT_SYMBOL(end_that_request_last);
3714
Jens Axboea0cd1282007-09-21 10:41:07 +02003715static inline void __end_request(struct request *rq, int uptodate,
3716 unsigned int nr_bytes, int dequeue)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717{
Jens Axboea0cd1282007-09-21 10:41:07 +02003718 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
3719 if (dequeue)
3720 blkdev_dequeue_request(rq);
3721 add_disk_randomness(rq->rq_disk);
3722 end_that_request_last(rq, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723 }
3724}
3725
Jens Axboea0cd1282007-09-21 10:41:07 +02003726static unsigned int rq_byte_size(struct request *rq)
3727{
3728 if (blk_fs_request(rq))
3729 return rq->hard_nr_sectors << 9;
3730
3731 return rq->data_len;
3732}
3733
3734/**
3735 * end_queued_request - end all I/O on a queued request
3736 * @rq: the request being processed
3737 * @uptodate: error value or 0/1 uptodate flag
3738 *
3739 * Description:
3740 * Ends all I/O on a request, and removes it from the block layer queues.
3741 * Not suitable for normal IO completion, unless the driver still has
3742 * the request attached to the block layer.
3743 *
3744 **/
3745void end_queued_request(struct request *rq, int uptodate)
3746{
3747 __end_request(rq, uptodate, rq_byte_size(rq), 1);
3748}
3749EXPORT_SYMBOL(end_queued_request);
3750
3751/**
3752 * end_dequeued_request - end all I/O on a dequeued request
3753 * @rq: the request being processed
3754 * @uptodate: error value or 0/1 uptodate flag
3755 *
3756 * Description:
3757 * Ends all I/O on a request. The request must already have been
3758 * dequeued using blkdev_dequeue_request(), as is normally the case
3759 * for most drivers.
3760 *
3761 **/
3762void end_dequeued_request(struct request *rq, int uptodate)
3763{
3764 __end_request(rq, uptodate, rq_byte_size(rq), 0);
3765}
3766EXPORT_SYMBOL(end_dequeued_request);
3767
3768
3769/**
3770 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07003771 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02003772 * @uptodate: error value or 0/1 uptodate flag
3773 *
3774 * Description:
3775 * Ends I/O on the current segment of a request. If that is the only
3776 * remaining segment, the request is also completed and freed.
3777 *
3778 * This is a remnant of how older block drivers handled IO completions.
3779 * Modern drivers typically end IO on the full request in one go, unless
3780 * they have a residual value to account for. For that case this function
3781 * isn't really useful, unless the residual just happens to be the
3782 * full current segment. In other words, don't use this function in new
3783 * code. Either use end_request_completely(), or the
3784 * end_that_request_chunk() (along with end_that_request_last()) for
3785 * partial completions.
3786 *
3787 **/
3788void end_request(struct request *req, int uptodate)
3789{
3790 __end_request(req, uptodate, req->hard_cur_sectors << 9, 1);
3791}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792EXPORT_SYMBOL(end_request);
3793
NeilBrown66846572007-08-16 13:31:28 +02003794static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3795 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003797 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3798 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799
3800 rq->nr_phys_segments = bio_phys_segments(q, bio);
3801 rq->nr_hw_segments = bio_hw_segments(q, bio);
3802 rq->current_nr_sectors = bio_cur_sectors(bio);
3803 rq->hard_cur_sectors = rq->current_nr_sectors;
3804 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3805 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003806 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003807
3808 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809
NeilBrown66846572007-08-16 13:31:28 +02003810 if (bio->bi_bdev)
3811 rq->rq_disk = bio->bi_bdev->bd_disk;
3812}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813
3814int kblockd_schedule_work(struct work_struct *work)
3815{
3816 return queue_work(kblockd_workqueue, work);
3817}
3818
3819EXPORT_SYMBOL(kblockd_schedule_work);
3820
Andrew Morton19a75d82007-05-09 02:33:56 -07003821void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003823 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824}
Andrew Morton19a75d82007-05-09 02:33:56 -07003825EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826
3827int __init blk_dev_init(void)
3828{
Jens Axboeff856ba2006-01-09 16:02:34 +01003829 int i;
3830
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 kblockd_workqueue = create_workqueue("kblockd");
3832 if (!kblockd_workqueue)
3833 panic("Failed to create kblockd\n");
3834
3835 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003836 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837
3838 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003839 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840
3841 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003842 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003844 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003845 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3846
3847 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003848 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003849
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003850 blk_max_low_pfn = max_low_pfn - 1;
3851 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852
3853 return 0;
3854}
3855
Jens Axboe4ac845a2008-01-24 08:44:49 +01003856static void cfq_dtor(struct io_context *ioc)
3857{
3858 struct cfq_io_context *cic[1];
3859 int r;
3860
3861 /*
3862 * We don't have a specific key to lookup with, so use the gang
3863 * lookup to just retrieve the first item stored. The cfq exit
3864 * function will iterate the full tree, so any member will do.
3865 */
3866 r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
3867 if (r > 0)
3868 cic[0]->dtor(ioc);
3869}
3870
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871/*
Jens Axboed38ecf92008-01-24 08:53:35 +01003872 * IO Context helper functions. put_io_context() returns 1 if there are no
3873 * more users of this io context, 0 otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 */
Jens Axboed38ecf92008-01-24 08:53:35 +01003875int put_io_context(struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876{
3877 if (ioc == NULL)
Jens Axboed38ecf92008-01-24 08:53:35 +01003878 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879
3880 BUG_ON(atomic_read(&ioc->refcount) == 0);
3881
3882 if (atomic_dec_and_test(&ioc->refcount)) {
Al Viro334e94d2006-03-18 15:05:53 -05003883 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 if (ioc->aic && ioc->aic->dtor)
3885 ioc->aic->dtor(ioc->aic);
Al Viro334e94d2006-03-18 15:05:53 -05003886 rcu_read_unlock();
Jens Axboe4ac845a2008-01-24 08:44:49 +01003887 cfq_dtor(ioc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888
3889 kmem_cache_free(iocontext_cachep, ioc);
Jens Axboed38ecf92008-01-24 08:53:35 +01003890 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 }
Jens Axboed38ecf92008-01-24 08:53:35 +01003892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893}
3894EXPORT_SYMBOL(put_io_context);
3895
Jens Axboe4ac845a2008-01-24 08:44:49 +01003896static void cfq_exit(struct io_context *ioc)
3897{
3898 struct cfq_io_context *cic[1];
3899 int r;
3900
3901 rcu_read_lock();
3902 /*
3903 * See comment for cfq_dtor()
3904 */
3905 r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
3906 rcu_read_unlock();
3907
3908 if (r > 0)
3909 cic[0]->exit(ioc);
3910}
3911
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912/* Called by the exitting task */
3913void exit_io_context(void)
3914{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 struct io_context *ioc;
3916
Jens Axboe22e2c502005-06-27 10:55:12 +02003917 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 ioc = current->io_context;
3919 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003920 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921
Jens Axboed38ecf92008-01-24 08:53:35 +01003922 if (atomic_dec_and_test(&ioc->nr_tasks)) {
3923 if (ioc->aic && ioc->aic->exit)
3924 ioc->aic->exit(ioc->aic);
Jens Axboe4ac845a2008-01-24 08:44:49 +01003925 cfq_exit(ioc);
Oleg Nesterov25034d72006-08-29 09:15:14 +02003926
Jens Axboed38ecf92008-01-24 08:53:35 +01003927 put_io_context(ioc);
3928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929}
3930
Jens Axboefd0928d2008-01-24 08:52:45 +01003931struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
3932{
3933 struct io_context *ret;
3934
3935 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
3936 if (ret) {
3937 atomic_set(&ret->refcount, 1);
Jens Axboed38ecf92008-01-24 08:53:35 +01003938 atomic_set(&ret->nr_tasks, 1);
3939 spin_lock_init(&ret->lock);
Jens Axboefd0928d2008-01-24 08:52:45 +01003940 ret->ioprio_changed = 0;
3941 ret->ioprio = 0;
3942 ret->last_waited = jiffies; /* doesn't matter... */
3943 ret->nr_batch_requests = 0; /* because this is 0 */
3944 ret->aic = NULL;
Jens Axboe4ac845a2008-01-24 08:44:49 +01003945 INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
Jens Axboefd0928d2008-01-24 08:52:45 +01003946 ret->ioc_data = NULL;
3947 }
3948
3949 return ret;
3950}
3951
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952/*
3953 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003954 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003956 * This returned IO context doesn't have a specifically elevated refcount,
3957 * but since the current task itself holds a reference, the context can be
3958 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003960static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961{
3962 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963 struct io_context *ret;
3964
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003966 if (likely(ret))
3967 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
Jens Axboefd0928d2008-01-24 08:52:45 +01003969 ret = alloc_io_context(gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 if (ret) {
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003971 /* make sure set_task_ioprio() sees the settings above */
3972 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003973 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974 }
3975
3976 return ret;
3977}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003978
3979/*
3980 * If the current task has no IO context then create one and initialise it.
3981 * If it does have a context, take a ref on it.
3982 *
3983 * This is always called in the context of the task which submitted the I/O.
3984 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003985struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003986{
Jens Axboed38ecf92008-01-24 08:53:35 +01003987 struct io_context *ret = NULL;
3988
3989 /*
3990 * Check for unlikely race with exiting task. ioc ref count is
3991 * zero when ioc is being detached.
3992 */
3993 do {
3994 ret = current_io_context(gfp_flags, node);
3995 if (unlikely(!ret))
3996 break;
3997 } while (!atomic_inc_not_zero(&ret->refcount));
3998
Nick Pigginfb3cc432005-06-28 20:45:15 -07003999 return ret;
4000}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001EXPORT_SYMBOL(get_io_context);
4002
4003void copy_io_context(struct io_context **pdst, struct io_context **psrc)
4004{
4005 struct io_context *src = *psrc;
4006 struct io_context *dst = *pdst;
4007
4008 if (src) {
4009 BUG_ON(atomic_read(&src->refcount) == 0);
4010 atomic_inc(&src->refcount);
4011 put_io_context(dst);
4012 *pdst = src;
4013 }
4014}
4015EXPORT_SYMBOL(copy_io_context);
4016
4017void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
4018{
4019 struct io_context *temp;
4020 temp = *ioc1;
4021 *ioc1 = *ioc2;
4022 *ioc2 = temp;
4023}
4024EXPORT_SYMBOL(swap_io_context);
4025
4026/*
4027 * sysfs parts below
4028 */
4029struct queue_sysfs_entry {
4030 struct attribute attr;
4031 ssize_t (*show)(struct request_queue *, char *);
4032 ssize_t (*store)(struct request_queue *, const char *, size_t);
4033};
4034
4035static ssize_t
4036queue_var_show(unsigned int var, char *page)
4037{
4038 return sprintf(page, "%d\n", var);
4039}
4040
4041static ssize_t
4042queue_var_store(unsigned long *var, const char *page, size_t count)
4043{
4044 char *p = (char *) page;
4045
4046 *var = simple_strtoul(p, &p, 10);
4047 return count;
4048}
4049
4050static ssize_t queue_requests_show(struct request_queue *q, char *page)
4051{
4052 return queue_var_show(q->nr_requests, (page));
4053}
4054
4055static ssize_t
4056queue_requests_store(struct request_queue *q, const char *page, size_t count)
4057{
4058 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05004059 unsigned long nr;
4060 int ret = queue_var_store(&nr, page, count);
4061 if (nr < BLKDEV_MIN_RQ)
4062 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063
Al Viroc981ff92006-03-18 13:51:29 -05004064 spin_lock_irq(q->queue_lock);
4065 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 blk_queue_congestion_threshold(q);
4067
4068 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004069 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004071 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004072
4073 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004074 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004076 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077
4078 if (rl->count[READ] >= q->nr_requests) {
4079 blk_set_queue_full(q, READ);
4080 } else if (rl->count[READ]+1 <= q->nr_requests) {
4081 blk_clear_queue_full(q, READ);
4082 wake_up(&rl->wait[READ]);
4083 }
4084
4085 if (rl->count[WRITE] >= q->nr_requests) {
4086 blk_set_queue_full(q, WRITE);
4087 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
4088 blk_clear_queue_full(q, WRITE);
4089 wake_up(&rl->wait[WRITE]);
4090 }
Al Viroc981ff92006-03-18 13:51:29 -05004091 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092 return ret;
4093}
4094
4095static ssize_t queue_ra_show(struct request_queue *q, char *page)
4096{
4097 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
4098
4099 return queue_var_show(ra_kb, (page));
4100}
4101
4102static ssize_t
4103queue_ra_store(struct request_queue *q, const char *page, size_t count)
4104{
4105 unsigned long ra_kb;
4106 ssize_t ret = queue_var_store(&ra_kb, page, count);
4107
4108 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
4110 spin_unlock_irq(q->queue_lock);
4111
4112 return ret;
4113}
4114
4115static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
4116{
4117 int max_sectors_kb = q->max_sectors >> 1;
4118
4119 return queue_var_show(max_sectors_kb, (page));
4120}
4121
4122static ssize_t
4123queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
4124{
4125 unsigned long max_sectors_kb,
4126 max_hw_sectors_kb = q->max_hw_sectors >> 1,
4127 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
4128 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004129
4130 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
4131 return -EINVAL;
4132 /*
4133 * Take the queue lock to update the readahead and max_sectors
4134 * values synchronously:
4135 */
4136 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137 q->max_sectors = max_sectors_kb << 1;
4138 spin_unlock_irq(q->queue_lock);
4139
4140 return ret;
4141}
4142
4143static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
4144{
4145 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
4146
4147 return queue_var_show(max_hw_sectors_kb, (page));
4148}
4149
4150
4151static struct queue_sysfs_entry queue_requests_entry = {
4152 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
4153 .show = queue_requests_show,
4154 .store = queue_requests_store,
4155};
4156
4157static struct queue_sysfs_entry queue_ra_entry = {
4158 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4159 .show = queue_ra_show,
4160 .store = queue_ra_store,
4161};
4162
4163static struct queue_sysfs_entry queue_max_sectors_entry = {
4164 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4165 .show = queue_max_sectors_show,
4166 .store = queue_max_sectors_store,
4167};
4168
4169static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4170 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4171 .show = queue_max_hw_sectors_show,
4172};
4173
4174static struct queue_sysfs_entry queue_iosched_entry = {
4175 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4176 .show = elv_iosched_show,
4177 .store = elv_iosched_store,
4178};
4179
4180static struct attribute *default_attrs[] = {
4181 &queue_requests_entry.attr,
4182 &queue_ra_entry.attr,
4183 &queue_max_hw_sectors_entry.attr,
4184 &queue_max_sectors_entry.attr,
4185 &queue_iosched_entry.attr,
4186 NULL,
4187};
4188
4189#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4190
4191static ssize_t
4192queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4193{
4194 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004195 struct request_queue *q =
4196 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004197 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004198
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004200 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004201 mutex_lock(&q->sysfs_lock);
4202 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4203 mutex_unlock(&q->sysfs_lock);
4204 return -ENOENT;
4205 }
4206 res = entry->show(q, page);
4207 mutex_unlock(&q->sysfs_lock);
4208 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209}
4210
4211static ssize_t
4212queue_attr_store(struct kobject *kobj, struct attribute *attr,
4213 const char *page, size_t length)
4214{
4215 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004216 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217
Al Viro483f4af2006-03-18 18:34:37 -05004218 ssize_t res;
4219
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004221 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004222 mutex_lock(&q->sysfs_lock);
4223 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4224 mutex_unlock(&q->sysfs_lock);
4225 return -ENOENT;
4226 }
4227 res = entry->store(q, page, length);
4228 mutex_unlock(&q->sysfs_lock);
4229 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230}
4231
4232static struct sysfs_ops queue_sysfs_ops = {
4233 .show = queue_attr_show,
4234 .store = queue_attr_store,
4235};
4236
Adrian Bunk93d17d32005-06-25 14:59:10 -07004237static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004238 .sysfs_ops = &queue_sysfs_ops,
4239 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004240 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241};
4242
4243int blk_register_queue(struct gendisk *disk)
4244{
4245 int ret;
4246
Jens Axboe165125e2007-07-24 09:28:11 +02004247 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248
4249 if (!q || !q->request_fn)
4250 return -ENXIO;
4251
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07004252 ret = kobject_add(&q->kobj, kobject_get(&disk->dev.kobj),
4253 "%s", "queue");
Linus Torvalds1da177e2005-04-16 15:20:36 -07004254 if (ret < 0)
4255 return ret;
4256
Al Viro483f4af2006-03-18 18:34:37 -05004257 kobject_uevent(&q->kobj, KOBJ_ADD);
4258
Linus Torvalds1da177e2005-04-16 15:20:36 -07004259 ret = elv_register_queue(q);
4260 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004261 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4262 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 return ret;
4264 }
4265
4266 return 0;
4267}
4268
4269void blk_unregister_queue(struct gendisk *disk)
4270{
Jens Axboe165125e2007-07-24 09:28:11 +02004271 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272
4273 if (q && q->request_fn) {
4274 elv_unregister_queue(q);
4275
Al Viro483f4af2006-03-18 18:34:37 -05004276 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4277 kobject_del(&q->kobj);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02004278 kobject_put(&disk->dev.kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004279 }
4280}