blob: 8025d646ab3085ff5627fc0f082465eba8329ce7 [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);
Adrian Bunk93d17d32005-06-25 14:59:10 -070042static void drive_stat_acct(struct request *rq, int nr_sectors, 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/**
763 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * @q: The request queue for the device
765 * @tag: The tag of the request
766 *
767 * Notes:
768 * Should be used when a device returns a tag and you want to match
769 * it with a request.
770 *
771 * no locks need be held.
772 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200773struct request *blk_queue_find_tag(struct request_queue *q, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
David C Somayajuluf583f492006-10-04 08:27:25 +0200775 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
778EXPORT_SYMBOL(blk_queue_find_tag);
779
780/**
James Bottomley492dfb42006-08-30 15:48:45 -0400781 * __blk_free_tags - release a given set of tag maintenance info
782 * @bqt: the tag map to free
783 *
784 * Tries to free the specified @bqt@. Returns true if it was
785 * actually freed and false if there are still references using it
786 */
787static int __blk_free_tags(struct blk_queue_tag *bqt)
788{
789 int retval;
790
791 retval = atomic_dec_and_test(&bqt->refcnt);
792 if (retval) {
793 BUG_ON(bqt->busy);
794 BUG_ON(!list_empty(&bqt->busy_list));
795
796 kfree(bqt->tag_index);
797 bqt->tag_index = NULL;
798
799 kfree(bqt->tag_map);
800 bqt->tag_map = NULL;
801
802 kfree(bqt);
803
804 }
805
806 return retval;
807}
808
809/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 * __blk_queue_free_tags - release tag maintenance info
811 * @q: the request queue for the device
812 *
813 * Notes:
814 * blk_cleanup_queue() will take care of calling this function, if tagging
815 * has been used. So there's no need to call this directly.
816 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200817static void __blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
819 struct blk_queue_tag *bqt = q->queue_tags;
820
821 if (!bqt)
822 return;
823
James Bottomley492dfb42006-08-30 15:48:45 -0400824 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 q->queue_tags = NULL;
827 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
828}
829
James Bottomley492dfb42006-08-30 15:48:45 -0400830
831/**
832 * blk_free_tags - release a given set of tag maintenance info
833 * @bqt: the tag map to free
834 *
835 * For externally managed @bqt@ frees the map. Callers of this
836 * function must guarantee to have released all the queues that
837 * might have been using this tag map.
838 */
839void blk_free_tags(struct blk_queue_tag *bqt)
840{
841 if (unlikely(!__blk_free_tags(bqt)))
842 BUG();
843}
844EXPORT_SYMBOL(blk_free_tags);
845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846/**
847 * blk_queue_free_tags - release tag maintenance info
848 * @q: the request queue for the device
849 *
850 * Notes:
851 * This is used to disabled tagged queuing to a device, yet leave
852 * queue in function.
853 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200854void blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
856 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
857}
858
859EXPORT_SYMBOL(blk_queue_free_tags);
860
861static int
Jens Axboe165125e2007-07-24 09:28:11 +0200862init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct request **tag_index;
865 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700866 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
James Bottomley492dfb42006-08-30 15:48:45 -0400868 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 depth = q->nr_requests * 2;
870 printk(KERN_ERR "%s: adjusted depth to %d\n",
871 __FUNCTION__, depth);
872 }
873
Jens Axboef68110f2006-03-08 13:31:44 +0100874 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (!tag_index)
876 goto fail;
877
Tejun Heof7d37d02005-06-23 00:08:50 -0700878 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100879 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 if (!tag_map)
881 goto fail;
882
Tejun Heoba025082005-08-05 13:28:11 -0700883 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 tags->tag_index = tag_index;
886 tags->tag_map = tag_map;
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 return 0;
889fail:
890 kfree(tag_index);
891 return -ENOMEM;
892}
893
James Bottomley492dfb42006-08-30 15:48:45 -0400894static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
895 int depth)
896{
897 struct blk_queue_tag *tags;
898
899 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
900 if (!tags)
901 goto fail;
902
903 if (init_tag_map(q, tags, depth))
904 goto fail;
905
906 INIT_LIST_HEAD(&tags->busy_list);
907 tags->busy = 0;
908 atomic_set(&tags->refcnt, 1);
909 return tags;
910fail:
911 kfree(tags);
912 return NULL;
913}
914
915/**
916 * blk_init_tags - initialize the tag info for an external tag map
917 * @depth: the maximum queue depth supported
918 * @tags: the tag to use
919 **/
920struct blk_queue_tag *blk_init_tags(int depth)
921{
922 return __blk_queue_init_tags(NULL, depth);
923}
924EXPORT_SYMBOL(blk_init_tags);
925
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926/**
927 * blk_queue_init_tags - initialize the queue tag info
928 * @q: the request queue for the device
929 * @depth: the maximum queue depth supported
930 * @tags: the tag to use
931 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200932int blk_queue_init_tags(struct request_queue *q, int depth,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 struct blk_queue_tag *tags)
934{
935 int rc;
936
937 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
938
939 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400940 tags = __blk_queue_init_tags(q, depth);
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (!tags)
943 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 } else if (q->queue_tags) {
945 if ((rc = blk_queue_resize_tags(q, depth)))
946 return rc;
947 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
948 return 0;
949 } else
950 atomic_inc(&tags->refcnt);
951
952 /*
953 * assign it, all done
954 */
955 q->queue_tags = tags;
956 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
957 return 0;
958fail:
959 kfree(tags);
960 return -ENOMEM;
961}
962
963EXPORT_SYMBOL(blk_queue_init_tags);
964
965/**
966 * blk_queue_resize_tags - change the queueing depth
967 * @q: the request queue for the device
968 * @new_depth: the new max command queueing depth
969 *
970 * Notes:
971 * Must be called with the queue lock held.
972 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200973int blk_queue_resize_tags(struct request_queue *q, int new_depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
975 struct blk_queue_tag *bqt = q->queue_tags;
976 struct request **tag_index;
977 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700978 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 if (!bqt)
981 return -ENXIO;
982
983 /*
Tejun Heoba025082005-08-05 13:28:11 -0700984 * if we already have large enough real_max_depth. just
985 * adjust max_depth. *NOTE* as requests with tag value
986 * between new_depth and real_max_depth can be in-flight, tag
987 * map can not be shrunk blindly here.
988 */
989 if (new_depth <= bqt->real_max_depth) {
990 bqt->max_depth = new_depth;
991 return 0;
992 }
993
994 /*
James Bottomley492dfb42006-08-30 15:48:45 -0400995 * Currently cannot replace a shared tag map with a new
996 * one, so error out if this is the case
997 */
998 if (atomic_read(&bqt->refcnt) != 1)
999 return -EBUSY;
1000
1001 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 * save the old state info, so we can copy it back
1003 */
1004 tag_index = bqt->tag_index;
1005 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001006 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 if (init_tag_map(q, bqt, new_depth))
1009 return -ENOMEM;
1010
1011 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001012 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001013 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
1015 kfree(tag_index);
1016 kfree(tag_map);
1017 return 0;
1018}
1019
1020EXPORT_SYMBOL(blk_queue_resize_tags);
1021
1022/**
1023 * blk_queue_end_tag - end tag operations for a request
1024 * @q: the request queue for the device
1025 * @rq: the request that has completed
1026 *
1027 * Description:
1028 * Typically called when end_that_request_first() returns 0, meaning
1029 * all transfers have been done for a request. It's important to call
1030 * this function before end_that_request_last(), as that will put the
1031 * request back on the free list thus corrupting the internal tag list.
1032 *
1033 * Notes:
1034 * queue lock must be held.
1035 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001036void blk_queue_end_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 struct blk_queue_tag *bqt = q->queue_tags;
1039 int tag = rq->tag;
1040
1041 BUG_ON(tag == -1);
1042
Tejun Heoba025082005-08-05 13:28:11 -07001043 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001044 /*
1045 * This can happen after tag depth has been reduced.
1046 * FIXME: how about a warning or info message here?
1047 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return;
1049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001051 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 rq->tag = -1;
1053
1054 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001055 printk(KERN_ERR "%s: tag %d is missing\n",
1056 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058 bqt->tag_index[tag] = NULL;
Jens Axboef3da54b2007-09-13 14:26:53 +02001059
Nick Piggindd941252007-09-14 08:41:12 +02001060 /*
1061 * We use test_and_clear_bit's memory ordering properties here.
1062 * The tag_map bit acts as a lock for tag_index[bit], so we need
1063 * a barrer before clearing the bit (precisely: release semantics).
1064 * Could use clear_bit_unlock when it is merged.
1065 */
Jens Axboef3da54b2007-09-13 14:26:53 +02001066 if (unlikely(!test_and_clear_bit(tag, bqt->tag_map))) {
1067 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1068 __FUNCTION__, tag);
1069 return;
1070 }
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 bqt->busy--;
1073}
1074
1075EXPORT_SYMBOL(blk_queue_end_tag);
1076
1077/**
1078 * blk_queue_start_tag - find a free tag and assign it
1079 * @q: the request queue for the device
1080 * @rq: the block request that needs tagging
1081 *
1082 * Description:
1083 * This can either be used as a stand-alone helper, or possibly be
1084 * assigned as the queue &prep_rq_fn (in which case &struct request
1085 * automagically gets a tag assigned). Note that this function
1086 * assumes that any type of request can be queued! if this is not
1087 * true for your device, you must check the request type before
1088 * calling this function. The request will also be removed from
1089 * the request queue, so it's the drivers responsibility to readd
1090 * it if it should need to be restarted for some reason.
1091 *
1092 * Notes:
1093 * queue lock must be held.
1094 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001095int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
1097 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001098 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Jens Axboe4aff5e22006-08-10 08:44:47 +02001100 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001102 "%s: request %p for device [%s] already tagged %d",
1103 __FUNCTION__, rq,
1104 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 BUG();
1106 }
1107
Jens Axboe059af492006-09-21 20:37:22 +02001108 /*
1109 * Protect against shared tag maps, as we may not have exclusive
1110 * access to the tag map.
1111 */
1112 do {
1113 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1114 if (tag >= bqt->max_depth)
1115 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Jens Axboe059af492006-09-21 20:37:22 +02001117 } while (test_and_set_bit(tag, bqt->tag_map));
Nick Piggindd941252007-09-14 08:41:12 +02001118 /*
1119 * We rely on test_and_set_bit providing lock memory ordering semantics
1120 * (could use test_and_set_bit_lock when it is merged).
1121 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Jens Axboe4aff5e22006-08-10 08:44:47 +02001123 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 rq->tag = tag;
1125 bqt->tag_index[tag] = rq;
1126 blkdev_dequeue_request(rq);
1127 list_add(&rq->queuelist, &bqt->busy_list);
1128 bqt->busy++;
1129 return 0;
1130}
1131
1132EXPORT_SYMBOL(blk_queue_start_tag);
1133
1134/**
1135 * blk_queue_invalidate_tags - invalidate all pending tags
1136 * @q: the request queue for the device
1137 *
1138 * Description:
1139 * Hardware conditions may dictate a need to stop all pending requests.
1140 * In this case, we will safely clear the block side of the tag queue and
1141 * readd all requests to the request queue in the right order.
1142 *
1143 * Notes:
1144 * queue lock must be held.
1145 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001146void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147{
1148 struct blk_queue_tag *bqt = q->queue_tags;
1149 struct list_head *tmp, *n;
1150 struct request *rq;
1151
1152 list_for_each_safe(tmp, n, &bqt->busy_list) {
1153 rq = list_entry_rq(tmp);
1154
1155 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001156 printk(KERN_ERR
1157 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001159 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 } else
1161 blk_queue_end_tag(q, rq);
1162
Jens Axboe4aff5e22006-08-10 08:44:47 +02001163 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1165 }
1166}
1167
1168EXPORT_SYMBOL(blk_queue_invalidate_tags);
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170void blk_dump_rq_flags(struct request *rq, char *msg)
1171{
1172 int bit;
1173
Jens Axboe4aff5e22006-08-10 08:44:47 +02001174 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1175 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1176 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
1178 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1179 rq->nr_sectors,
1180 rq->current_nr_sectors);
1181 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1182
Jens Axboe4aff5e22006-08-10 08:44:47 +02001183 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 printk("cdb: ");
1185 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1186 printk("%02x ", rq->cmd[bit]);
1187 printk("\n");
1188 }
1189}
1190
1191EXPORT_SYMBOL(blk_dump_rq_flags);
1192
Jens Axboe165125e2007-07-24 09:28:11 +02001193void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194{
NeilBrown9dfa5282007-08-16 13:27:52 +02001195 struct request rq;
1196 struct bio *nxt = bio->bi_next;
1197 rq.q = q;
1198 rq.bio = rq.biotail = bio;
1199 bio->bi_next = NULL;
1200 blk_recalc_rq_segments(&rq);
1201 bio->bi_next = nxt;
1202 bio->bi_phys_segments = rq.nr_phys_segments;
1203 bio->bi_hw_segments = rq.nr_hw_segments;
1204 bio->bi_flags |= (1 << BIO_SEG_VALID);
1205}
1206EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
NeilBrown9dfa5282007-08-16 13:27:52 +02001208static void blk_recalc_rq_segments(struct request *rq)
1209{
1210 int nr_phys_segs;
1211 int nr_hw_segs;
1212 unsigned int phys_size;
1213 unsigned int hw_size;
1214 struct bio_vec *bv, *bvprv = NULL;
1215 int seg_size;
1216 int hw_seg_size;
1217 int cluster;
NeilBrown5705f702007-09-25 12:35:59 +02001218 struct req_iterator iter;
NeilBrown9dfa5282007-08-16 13:27:52 +02001219 int high, highprv = 1;
1220 struct request_queue *q = rq->q;
1221
1222 if (!rq->bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return;
1224
1225 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
NeilBrown9dfa5282007-08-16 13:27:52 +02001226 hw_seg_size = seg_size = 0;
1227 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
NeilBrown5705f702007-09-25 12:35:59 +02001228 rq_for_each_segment(bv, rq, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 /*
1230 * the trick here is making sure that a high page is never
1231 * considered part of another segment, since that might
1232 * change with the bounce page.
1233 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001234 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (high || highprv)
1236 goto new_hw_segment;
1237 if (cluster) {
1238 if (seg_size + bv->bv_len > q->max_segment_size)
1239 goto new_segment;
1240 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1241 goto new_segment;
1242 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1243 goto new_segment;
1244 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1245 goto new_hw_segment;
1246
1247 seg_size += bv->bv_len;
1248 hw_seg_size += bv->bv_len;
1249 bvprv = bv;
1250 continue;
1251 }
1252new_segment:
1253 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
NeilBrown9dfa5282007-08-16 13:27:52 +02001254 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 hw_seg_size += bv->bv_len;
NeilBrown9dfa5282007-08-16 13:27:52 +02001256 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257new_hw_segment:
NeilBrown9dfa5282007-08-16 13:27:52 +02001258 if (nr_hw_segs == 1 &&
1259 hw_seg_size > rq->bio->bi_hw_front_size)
1260 rq->bio->bi_hw_front_size = hw_seg_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1262 nr_hw_segs++;
1263 }
1264
1265 nr_phys_segs++;
1266 bvprv = bv;
1267 seg_size = bv->bv_len;
1268 highprv = high;
1269 }
NeilBrown9dfa5282007-08-16 13:27:52 +02001270
1271 if (nr_hw_segs == 1 &&
1272 hw_seg_size > rq->bio->bi_hw_front_size)
1273 rq->bio->bi_hw_front_size = hw_seg_size;
1274 if (hw_seg_size > rq->biotail->bi_hw_back_size)
1275 rq->biotail->bi_hw_back_size = hw_seg_size;
1276 rq->nr_phys_segments = nr_phys_segs;
1277 rq->nr_hw_segments = nr_hw_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
Jens Axboe165125e2007-07-24 09:28:11 +02001280static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 struct bio *nxt)
1282{
1283 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1284 return 0;
1285
1286 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1287 return 0;
1288 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1289 return 0;
1290
1291 /*
1292 * bio and nxt are contigous in memory, check if the queue allows
1293 * these two to be merged into one
1294 */
1295 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1296 return 1;
1297
1298 return 0;
1299}
1300
Jens Axboe165125e2007-07-24 09:28:11 +02001301static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 struct bio *nxt)
1303{
1304 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1305 blk_recount_segments(q, bio);
1306 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1307 blk_recount_segments(q, nxt);
1308 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001309 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001311 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 return 0;
1313
1314 return 1;
1315}
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317/*
1318 * map a request to scatterlist, return number of sg entries setup. Caller
1319 * must make sure sg can hold rq->nr_phys_segments entries
1320 */
Jens Axboe165125e2007-07-24 09:28:11 +02001321int blk_rq_map_sg(struct request_queue *q, struct request *rq,
Jens Axboef5659132007-09-21 10:44:19 +02001322 struct scatterlist *sglist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
1324 struct bio_vec *bvec, *bvprv;
NeilBrown5705f702007-09-25 12:35:59 +02001325 struct req_iterator iter;
Jens Axboeba951842007-10-17 19:34:11 +02001326 struct scatterlist *sg;
NeilBrown5705f702007-09-25 12:35:59 +02001327 int nsegs, cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329 nsegs = 0;
1330 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1331
1332 /*
1333 * for each bio in rq
1334 */
1335 bvprv = NULL;
Jens Axboeba951842007-10-17 19:34:11 +02001336 sg = NULL;
NeilBrown5705f702007-09-25 12:35:59 +02001337 rq_for_each_segment(bvec, rq, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +02001338 int nbytes = bvec->bv_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Jens Axboe6c92e692007-08-16 13:43:12 +02001340 if (bvprv && cluster) {
Jens Axboef5659132007-09-21 10:44:19 +02001341 if (sg->length + nbytes > q->max_segment_size)
Jens Axboe6c92e692007-08-16 13:43:12 +02001342 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Jens Axboe6c92e692007-08-16 13:43:12 +02001344 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1345 goto new_segment;
1346 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1347 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Jens Axboef5659132007-09-21 10:44:19 +02001349 sg->length += nbytes;
Jens Axboe6c92e692007-08-16 13:43:12 +02001350 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351new_segment:
Jens Axboeba951842007-10-17 19:34:11 +02001352 if (!sg)
1353 sg = sglist;
1354 else
1355 sg = sg_next(sg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Jens Axboe60573b82007-10-17 13:02:33 +02001357 memset(sg, 0, sizeof(*sg));
Jens Axboef5659132007-09-21 10:44:19 +02001358 sg->page = bvec->bv_page;
1359 sg->length = nbytes;
1360 sg->offset = bvec->bv_offset;
Jens Axboe6c92e692007-08-16 13:43:12 +02001361 nsegs++;
1362 }
1363 bvprv = bvec;
NeilBrown5705f702007-09-25 12:35:59 +02001364 } /* segments in rq */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
1366 return nsegs;
1367}
1368
1369EXPORT_SYMBOL(blk_rq_map_sg);
1370
1371/*
1372 * the standard queue merge functions, can be overridden with device
1373 * specific ones if so desired
1374 */
1375
Jens Axboe165125e2007-07-24 09:28:11 +02001376static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 struct request *req,
1378 struct bio *bio)
1379{
1380 int nr_phys_segs = bio_phys_segments(q, bio);
1381
1382 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001383 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 if (req == q->last_merge)
1385 q->last_merge = NULL;
1386 return 0;
1387 }
1388
1389 /*
1390 * A hw segment is just getting larger, bump just the phys
1391 * counter.
1392 */
1393 req->nr_phys_segments += nr_phys_segs;
1394 return 1;
1395}
1396
Jens Axboe165125e2007-07-24 09:28:11 +02001397static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 struct request *req,
1399 struct bio *bio)
1400{
1401 int nr_hw_segs = bio_hw_segments(q, bio);
1402 int nr_phys_segs = bio_phys_segments(q, bio);
1403
1404 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1405 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001406 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (req == q->last_merge)
1408 q->last_merge = NULL;
1409 return 0;
1410 }
1411
1412 /*
1413 * This will form the start of a new hw segment. Bump both
1414 * counters.
1415 */
1416 req->nr_hw_segments += nr_hw_segs;
1417 req->nr_phys_segments += nr_phys_segs;
1418 return 1;
1419}
1420
NeilBrown3001ca72007-08-16 13:31:27 +02001421static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1422 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
Mike Christiedefd94b2005-12-05 02:37:06 -06001424 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 int len;
1426
Mike Christiedefd94b2005-12-05 02:37:06 -06001427 if (unlikely(blk_pc_request(req)))
1428 max_sectors = q->max_hw_sectors;
1429 else
1430 max_sectors = q->max_sectors;
1431
1432 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001433 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 if (req == q->last_merge)
1435 q->last_merge = NULL;
1436 return 0;
1437 }
1438 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1439 blk_recount_segments(q, req->biotail);
1440 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1441 blk_recount_segments(q, bio);
1442 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1443 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1444 !BIOVEC_VIRT_OVERSIZE(len)) {
1445 int mergeable = ll_new_mergeable(q, req, bio);
1446
1447 if (mergeable) {
1448 if (req->nr_hw_segments == 1)
1449 req->bio->bi_hw_front_size = len;
1450 if (bio->bi_hw_segments == 1)
1451 bio->bi_hw_back_size = len;
1452 }
1453 return mergeable;
1454 }
1455
1456 return ll_new_hw_segment(q, req, bio);
1457}
1458
Jens Axboe165125e2007-07-24 09:28:11 +02001459static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 struct bio *bio)
1461{
Mike Christiedefd94b2005-12-05 02:37:06 -06001462 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 int len;
1464
Mike Christiedefd94b2005-12-05 02:37:06 -06001465 if (unlikely(blk_pc_request(req)))
1466 max_sectors = q->max_hw_sectors;
1467 else
1468 max_sectors = q->max_sectors;
1469
1470
1471 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001472 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (req == q->last_merge)
1474 q->last_merge = NULL;
1475 return 0;
1476 }
1477 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1478 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1479 blk_recount_segments(q, bio);
1480 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1481 blk_recount_segments(q, req->bio);
1482 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1483 !BIOVEC_VIRT_OVERSIZE(len)) {
1484 int mergeable = ll_new_mergeable(q, req, bio);
1485
1486 if (mergeable) {
1487 if (bio->bi_hw_segments == 1)
1488 bio->bi_hw_front_size = len;
1489 if (req->nr_hw_segments == 1)
1490 req->biotail->bi_hw_back_size = len;
1491 }
1492 return mergeable;
1493 }
1494
1495 return ll_new_hw_segment(q, req, bio);
1496}
1497
Jens Axboe165125e2007-07-24 09:28:11 +02001498static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 struct request *next)
1500{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001501 int total_phys_segments;
1502 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504 /*
1505 * First check if the either of the requests are re-queued
1506 * requests. Can't merge them if they are.
1507 */
1508 if (req->special || next->special)
1509 return 0;
1510
1511 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001512 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 */
1514 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1515 return 0;
1516
1517 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1518 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1519 total_phys_segments--;
1520
1521 if (total_phys_segments > q->max_phys_segments)
1522 return 0;
1523
1524 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1525 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1526 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1527 /*
1528 * propagate the combined length to the end of the requests
1529 */
1530 if (req->nr_hw_segments == 1)
1531 req->bio->bi_hw_front_size = len;
1532 if (next->nr_hw_segments == 1)
1533 next->biotail->bi_hw_back_size = len;
1534 total_hw_segments--;
1535 }
1536
1537 if (total_hw_segments > q->max_hw_segments)
1538 return 0;
1539
1540 /* Merge is OK... */
1541 req->nr_phys_segments = total_phys_segments;
1542 req->nr_hw_segments = total_hw_segments;
1543 return 1;
1544}
1545
1546/*
1547 * "plug" the device if there are no outstanding requests: this will
1548 * force the transfer to start only after we have put all the requests
1549 * on the list.
1550 *
1551 * This is called with interrupts off and no requests on the queue and
1552 * with the queue lock held.
1553 */
Jens Axboe165125e2007-07-24 09:28:11 +02001554void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555{
1556 WARN_ON(!irqs_disabled());
1557
1558 /*
1559 * don't plug a stopped queue, it must be paired with blk_start_queue()
1560 * which will restart the queueing
1561 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001562 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 return;
1564
Jens Axboe2056a782006-03-23 20:00:26 +01001565 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001567 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
1571EXPORT_SYMBOL(blk_plug_device);
1572
1573/*
1574 * remove the queue from the plugged list, if present. called with
1575 * queue lock held and interrupts disabled.
1576 */
Jens Axboe165125e2007-07-24 09:28:11 +02001577int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
1579 WARN_ON(!irqs_disabled());
1580
1581 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1582 return 0;
1583
1584 del_timer(&q->unplug_timer);
1585 return 1;
1586}
1587
1588EXPORT_SYMBOL(blk_remove_plug);
1589
1590/*
1591 * remove the plug and let it rip..
1592 */
Jens Axboe165125e2007-07-24 09:28:11 +02001593void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001595 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 return;
1597
1598 if (!blk_remove_plug(q))
1599 return;
1600
Jens Axboe22e2c502005-06-27 10:55:12 +02001601 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602}
1603EXPORT_SYMBOL(__generic_unplug_device);
1604
1605/**
1606 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001607 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 *
1609 * Description:
1610 * Linux uses plugging to build bigger requests queues before letting
1611 * the device have at them. If a queue is plugged, the I/O scheduler
1612 * is still adding and merging requests on the queue. Once the queue
1613 * gets unplugged, the request_fn defined for the queue is invoked and
1614 * transfers started.
1615 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001616void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617{
1618 spin_lock_irq(q->queue_lock);
1619 __generic_unplug_device(q);
1620 spin_unlock_irq(q->queue_lock);
1621}
1622EXPORT_SYMBOL(generic_unplug_device);
1623
1624static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1625 struct page *page)
1626{
Jens Axboe165125e2007-07-24 09:28:11 +02001627 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629 /*
1630 * devices don't necessarily have an ->unplug_fn defined
1631 */
Jens Axboe2056a782006-03-23 20:00:26 +01001632 if (q->unplug_fn) {
1633 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1634 q->rq.count[READ] + q->rq.count[WRITE]);
1635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638}
1639
David Howells65f27f32006-11-22 14:55:48 +00001640static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
Jens Axboe165125e2007-07-24 09:28:11 +02001642 struct request_queue *q =
1643 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
Jens Axboe2056a782006-03-23 20:00:26 +01001645 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1646 q->rq.count[READ] + q->rq.count[WRITE]);
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 q->unplug_fn(q);
1649}
1650
1651static void blk_unplug_timeout(unsigned long data)
1652{
Jens Axboe165125e2007-07-24 09:28:11 +02001653 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Jens Axboe2056a782006-03-23 20:00:26 +01001655 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1656 q->rq.count[READ] + q->rq.count[WRITE]);
1657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 kblockd_schedule_work(&q->unplug_work);
1659}
1660
1661/**
1662 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001663 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 *
1665 * Description:
1666 * blk_start_queue() will clear the stop flag on the queue, and call
1667 * the request_fn for the queue if it was in a stopped state when
1668 * entered. Also see blk_stop_queue(). Queue lock must be held.
1669 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001670void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001672 WARN_ON(!irqs_disabled());
1673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1675
1676 /*
1677 * one level of recursion is ok and is much faster than kicking
1678 * the unplug handling
1679 */
1680 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1681 q->request_fn(q);
1682 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1683 } else {
1684 blk_plug_device(q);
1685 kblockd_schedule_work(&q->unplug_work);
1686 }
1687}
1688
1689EXPORT_SYMBOL(blk_start_queue);
1690
1691/**
1692 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001693 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 *
1695 * Description:
1696 * The Linux block layer assumes that a block driver will consume all
1697 * entries on the request queue when the request_fn strategy is called.
1698 * Often this will not happen, because of hardware limitations (queue
1699 * depth settings). If a device driver gets a 'queue full' response,
1700 * or if it simply chooses not to queue more I/O at one point, it can
1701 * call this function to prevent the request_fn from being called until
1702 * the driver has signalled it's ready to go again. This happens by calling
1703 * blk_start_queue() to restart queue operations. Queue lock must be held.
1704 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001705void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706{
1707 blk_remove_plug(q);
1708 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1709}
1710EXPORT_SYMBOL(blk_stop_queue);
1711
1712/**
1713 * blk_sync_queue - cancel any pending callbacks on a queue
1714 * @q: the queue
1715 *
1716 * Description:
1717 * The block layer may perform asynchronous callback activity
1718 * on a queue, such as calling the unplug function after a timeout.
1719 * A block device may call blk_sync_queue to ensure that any
1720 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001721 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 * that its ->make_request_fn will not re-add plugging prior to calling
1723 * this function.
1724 *
1725 */
1726void blk_sync_queue(struct request_queue *q)
1727{
1728 del_timer_sync(&q->unplug_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729}
1730EXPORT_SYMBOL(blk_sync_queue);
1731
1732/**
1733 * blk_run_queue - run a single device queue
1734 * @q: The queue to run
1735 */
1736void blk_run_queue(struct request_queue *q)
1737{
1738 unsigned long flags;
1739
1740 spin_lock_irqsave(q->queue_lock, flags);
1741 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001742
1743 /*
1744 * Only recurse once to avoid overrunning the stack, let the unplug
1745 * handling reinvoke the handler shortly if we already got there.
1746 */
1747 if (!elv_queue_empty(q)) {
1748 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1749 q->request_fn(q);
1750 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1751 } else {
1752 blk_plug_device(q);
1753 kblockd_schedule_work(&q->unplug_work);
1754 }
1755 }
1756
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 spin_unlock_irqrestore(q->queue_lock, flags);
1758}
1759EXPORT_SYMBOL(blk_run_queue);
1760
1761/**
Jens Axboe165125e2007-07-24 09:28:11 +02001762 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001763 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 *
1765 * Description:
1766 * blk_cleanup_queue is the pair to blk_init_queue() or
1767 * blk_queue_make_request(). It should be called when a request queue is
1768 * being released; typically when a block device is being de-registered.
1769 * Currently, its primary task it to free all the &struct request
1770 * structures that were allocated to the queue and the queue itself.
1771 *
1772 * Caveat:
1773 * Hopefully the low level driver will have finished any
1774 * outstanding requests first...
1775 **/
Al Viro483f4af2006-03-18 18:34:37 -05001776static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777{
Jens Axboe165125e2007-07-24 09:28:11 +02001778 struct request_queue *q =
1779 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 struct request_list *rl = &q->rq;
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 blk_sync_queue(q);
1783
1784 if (rl->rq_pool)
1785 mempool_destroy(rl->rq_pool);
1786
1787 if (q->queue_tags)
1788 __blk_queue_free_tags(q);
1789
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001790 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001791
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001792 bdi_destroy(&q->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 kmem_cache_free(requestq_cachep, q);
1794}
1795
Jens Axboe165125e2007-07-24 09:28:11 +02001796void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001797{
1798 kobject_put(&q->kobj);
1799}
1800EXPORT_SYMBOL(blk_put_queue);
1801
Jens Axboe165125e2007-07-24 09:28:11 +02001802void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001803{
1804 mutex_lock(&q->sysfs_lock);
1805 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1806 mutex_unlock(&q->sysfs_lock);
1807
1808 if (q->elevator)
1809 elevator_exit(q->elevator);
1810
1811 blk_put_queue(q);
1812}
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814EXPORT_SYMBOL(blk_cleanup_queue);
1815
Jens Axboe165125e2007-07-24 09:28:11 +02001816static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817{
1818 struct request_list *rl = &q->rq;
1819
1820 rl->count[READ] = rl->count[WRITE] = 0;
1821 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001822 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 init_waitqueue_head(&rl->wait[READ]);
1824 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Christoph Lameter19460892005-06-23 00:08:19 -07001826 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1827 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 if (!rl->rq_pool)
1830 return -ENOMEM;
1831
1832 return 0;
1833}
1834
Jens Axboe165125e2007-07-24 09:28:11 +02001835struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836{
Christoph Lameter19460892005-06-23 00:08:19 -07001837 return blk_alloc_queue_node(gfp_mask, -1);
1838}
1839EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Al Viro483f4af2006-03-18 18:34:37 -05001841static struct kobj_type queue_ktype;
1842
Jens Axboe165125e2007-07-24 09:28:11 +02001843struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001844{
Jens Axboe165125e2007-07-24 09:28:11 +02001845 struct request_queue *q;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001846 int err;
Christoph Lameter19460892005-06-23 00:08:19 -07001847
Christoph Lameter94f60302007-07-17 04:03:29 -07001848 q = kmem_cache_alloc_node(requestq_cachep,
1849 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (!q)
1851 return NULL;
1852
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001853 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1854 q->backing_dev_info.unplug_io_data = q;
1855 err = bdi_init(&q->backing_dev_info);
1856 if (err) {
1857 kmem_cache_free(requestq_cachep, q);
1858 return NULL;
1859 }
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001862
Greg Kroah-Hartman19c38de2007-09-12 15:06:57 -07001863 kobject_set_name(&q->kobj, "%s", "queue");
Al Viro483f4af2006-03-18 18:34:37 -05001864 q->kobj.ktype = &queue_ktype;
1865 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
Al Viro483f4af2006-03-18 18:34:37 -05001867 mutex_init(&q->sysfs_lock);
1868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 return q;
1870}
Christoph Lameter19460892005-06-23 00:08:19 -07001871EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873/**
1874 * blk_init_queue - prepare a request queue for use with a block device
1875 * @rfn: The function to be called to process requests that have been
1876 * placed on the queue.
1877 * @lock: Request queue spin lock
1878 *
1879 * Description:
1880 * If a block device wishes to use the standard request handling procedures,
1881 * which sorts requests and coalesces adjacent requests, then it must
1882 * call blk_init_queue(). The function @rfn will be called when there
1883 * are requests on the queue that need to be processed. If the device
1884 * supports plugging, then @rfn may not be called immediately when requests
1885 * are available on the queue, but may be called at some time later instead.
1886 * Plugged queues are generally unplugged when a buffer belonging to one
1887 * of the requests on the queue is needed, or due to memory pressure.
1888 *
1889 * @rfn is not required, or even expected, to remove all requests off the
1890 * queue, but only as many as it can handle at a time. If it does leave
1891 * requests on the queue, it is responsible for arranging that the requests
1892 * get dealt with eventually.
1893 *
1894 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001895 * request queue; this lock will be taken also from interrupt context, so irq
1896 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 *
1898 * Function returns a pointer to the initialized request queue, or NULL if
1899 * it didn't succeed.
1900 *
1901 * Note:
1902 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1903 * when the block device is deactivated (such as at module unload).
1904 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001905
Jens Axboe165125e2007-07-24 09:28:11 +02001906struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Christoph Lameter19460892005-06-23 00:08:19 -07001908 return blk_init_queue_node(rfn, lock, -1);
1909}
1910EXPORT_SYMBOL(blk_init_queue);
1911
Jens Axboe165125e2007-07-24 09:28:11 +02001912struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001913blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1914{
Jens Axboe165125e2007-07-24 09:28:11 +02001915 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
1917 if (!q)
1918 return NULL;
1919
Christoph Lameter19460892005-06-23 00:08:19 -07001920 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001921 if (blk_init_free_list(q)) {
1922 kmem_cache_free(requestq_cachep, q);
1923 return NULL;
1924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
152587d2005-04-12 16:22:06 -05001926 /*
1927 * if caller didn't supply a lock, they get per-queue locking with
1928 * our embedded lock
1929 */
1930 if (!lock) {
1931 spin_lock_init(&q->__queue_lock);
1932 lock = &q->__queue_lock;
1933 }
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 q->prep_rq_fn = NULL;
1937 q->unplug_fn = generic_unplug_device;
1938 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1939 q->queue_lock = lock;
1940
1941 blk_queue_segment_boundary(q, 0xffffffff);
1942
1943 blk_queue_make_request(q, __make_request);
1944 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1945
1946 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1947 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1948
Alan Stern44ec9542007-02-20 11:01:57 -05001949 q->sg_reserved_size = INT_MAX;
1950
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 /*
1952 * all done
1953 */
1954 if (!elevator_init(q, NULL)) {
1955 blk_queue_congestion_threshold(q);
1956 return q;
1957 }
1958
Al Viro8669aaf2006-03-18 13:50:00 -05001959 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 return NULL;
1961}
Christoph Lameter19460892005-06-23 00:08:19 -07001962EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Jens Axboe165125e2007-07-24 09:28:11 +02001964int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001966 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001967 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 return 0;
1969 }
1970
1971 return 1;
1972}
1973
1974EXPORT_SYMBOL(blk_get_queue);
1975
Jens Axboe165125e2007-07-24 09:28:11 +02001976static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001978 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001979 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 mempool_free(rq, q->rq.rq_pool);
1981}
1982
Jens Axboe1ea25ec2006-07-18 22:24:11 +02001983static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001984blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985{
1986 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1987
1988 if (!rq)
1989 return NULL;
1990
1991 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001992 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 * see bio.h and blkdev.h
1994 */
Jens Axboe49171e52006-08-10 08:59:11 +02001995 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
Tejun Heocb98fc82005-10-28 08:29:39 +02001997 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001998 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001999 mempool_free(rq, q->rq.rq_pool);
2000 return NULL;
2001 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02002002 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002003 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Tejun Heocb98fc82005-10-28 08:29:39 +02002005 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006}
2007
2008/*
2009 * ioc_batching returns true if the ioc is a valid batching request and
2010 * should be given priority access to a request.
2011 */
Jens Axboe165125e2007-07-24 09:28:11 +02002012static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
2014 if (!ioc)
2015 return 0;
2016
2017 /*
2018 * Make sure the process is able to allocate at least 1 request
2019 * even if the batch times out, otherwise we could theoretically
2020 * lose wakeups.
2021 */
2022 return ioc->nr_batch_requests == q->nr_batching ||
2023 (ioc->nr_batch_requests > 0
2024 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2025}
2026
2027/*
2028 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2029 * will cause the process to be a "batcher" on all queues in the system. This
2030 * is the behaviour we want though - once it gets a wakeup it should be given
2031 * a nice run.
2032 */
Jens Axboe165125e2007-07-24 09:28:11 +02002033static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
2035 if (!ioc || ioc_batching(q, ioc))
2036 return;
2037
2038 ioc->nr_batch_requests = q->nr_batching;
2039 ioc->last_waited = jiffies;
2040}
2041
Jens Axboe165125e2007-07-24 09:28:11 +02002042static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
2044 struct request_list *rl = &q->rq;
2045
2046 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002047 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
2049 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (waitqueue_active(&rl->wait[rw]))
2051 wake_up(&rl->wait[rw]);
2052
2053 blk_clear_queue_full(q, rw);
2054 }
2055}
2056
2057/*
2058 * A request has just been released. Account for it, update the full and
2059 * congestion status, wake up any waiters. Called under q->queue_lock.
2060 */
Jens Axboe165125e2007-07-24 09:28:11 +02002061static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
2063 struct request_list *rl = &q->rq;
2064
2065 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002066 if (priv)
2067 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
2069 __freed_request(q, rw);
2070
2071 if (unlikely(rl->starved[rw ^ 1]))
2072 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
2075#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2076/*
Nick Piggind6344532005-06-28 20:45:14 -07002077 * Get a free request, queue_lock must be held.
2078 * Returns NULL on failure, with queue_lock held.
2079 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 */
Jens Axboe165125e2007-07-24 09:28:11 +02002081static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002082 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
2084 struct request *rq = NULL;
2085 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002086 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002087 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002088 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Jens Axboe7749a8d2006-12-13 13:02:26 +01002090 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002091 if (may_queue == ELV_MQUEUE_NO)
2092 goto rq_starved;
2093
2094 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2095 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002096 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002097 /*
2098 * The queue will fill after this allocation, so set
2099 * it as full, and mark this process as "batching".
2100 * This process will be allowed to complete a batch of
2101 * requests, others will be blocked.
2102 */
2103 if (!blk_queue_full(q, rw)) {
2104 ioc_set_batching(q, ioc);
2105 blk_set_queue_full(q, rw);
2106 } else {
2107 if (may_queue != ELV_MQUEUE_MUST
2108 && !ioc_batching(q, ioc)) {
2109 /*
2110 * The queue is full and the allocating
2111 * process is not a "batcher", and not
2112 * exempted by the IO scheduler
2113 */
2114 goto out;
2115 }
2116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002118 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 }
2120
Jens Axboe082cf692005-06-28 16:35:11 +02002121 /*
2122 * Only allow batching queuers to allocate up to 50% over the defined
2123 * limit of requests, otherwise we could have thousands of requests
2124 * allocated with any setting of ->nr_requests
2125 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002126 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002127 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 rl->count[rw]++;
2130 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002131
Jens Axboe64521d12005-10-28 08:30:39 +02002132 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002133 if (priv)
2134 rl->elvpriv++;
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 spin_unlock_irq(q->queue_lock);
2137
Jens Axboe7749a8d2006-12-13 13:02:26 +01002138 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002139 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 /*
2141 * Allocation failed presumably due to memory. Undo anything
2142 * we might have messed up.
2143 *
2144 * Allocating task should really be put onto the front of the
2145 * wait queue, but this is pretty rare.
2146 */
2147 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002148 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
2150 /*
2151 * in the very unlikely event that allocation failed and no
2152 * requests for this direction was pending, mark us starved
2153 * so that freeing of a request in the other direction will
2154 * notice us. another possible fix would be to split the
2155 * rq mempool into READ and WRITE
2156 */
2157rq_starved:
2158 if (unlikely(rl->count[rw] == 0))
2159 rl->starved[rw] = 1;
2160
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 goto out;
2162 }
2163
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002164 /*
2165 * ioc may be NULL here, and ioc_batching will be false. That's
2166 * OK, if the queue is under the request limit then requests need
2167 * not count toward the nr_batch_requests limit. There will always
2168 * be some limit enforced by BLK_BATCH_TIME.
2169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 if (ioc_batching(q, ioc))
2171 ioc->nr_batch_requests--;
2172
2173 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002174
2175 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 return rq;
2178}
2179
2180/*
2181 * No available requests for this queue, unplug the device and wait for some
2182 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002183 *
2184 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 */
Jens Axboe165125e2007-07-24 09:28:11 +02002186static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002187 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002189 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 struct request *rq;
2191
Jens Axboe7749a8d2006-12-13 13:02:26 +01002192 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002193 while (!rq) {
2194 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 struct request_list *rl = &q->rq;
2196
2197 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2198 TASK_UNINTERRUPTIBLE);
2199
Jens Axboe7749a8d2006-12-13 13:02:26 +01002200 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
2202 if (!rq) {
2203 struct io_context *ioc;
2204
Jens Axboe2056a782006-03-23 20:00:26 +01002205 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2206
Nick Piggind6344532005-06-28 20:45:14 -07002207 __generic_unplug_device(q);
2208 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 io_schedule();
2210
2211 /*
2212 * After sleeping, we become a "batching" process and
2213 * will be able to allocate at least one request, and
2214 * up to a big batch of them for a small period time.
2215 * See ioc_batching, ioc_set_batching
2216 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002217 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002219
2220 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
2222 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
2225 return rq;
2226}
2227
Jens Axboe165125e2007-07-24 09:28:11 +02002228struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229{
2230 struct request *rq;
2231
2232 BUG_ON(rw != READ && rw != WRITE);
2233
Nick Piggind6344532005-06-28 20:45:14 -07002234 spin_lock_irq(q->queue_lock);
2235 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002236 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002237 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002238 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002239 if (!rq)
2240 spin_unlock_irq(q->queue_lock);
2241 }
2242 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
2244 return rq;
2245}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246EXPORT_SYMBOL(blk_get_request);
2247
2248/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002249 * blk_start_queueing - initiate dispatch of requests to device
2250 * @q: request queue to kick into gear
2251 *
2252 * This is basically a helper to remove the need to know whether a queue
2253 * is plugged or not if someone just wants to initiate dispatch of requests
2254 * for this queue.
2255 *
2256 * The queue lock must be held with interrupts disabled.
2257 */
Jens Axboe165125e2007-07-24 09:28:11 +02002258void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002259{
2260 if (!blk_queue_plugged(q))
2261 q->request_fn(q);
2262 else
2263 __generic_unplug_device(q);
2264}
2265EXPORT_SYMBOL(blk_start_queueing);
2266
2267/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 * blk_requeue_request - put a request back on queue
2269 * @q: request queue where request should be inserted
2270 * @rq: request to be inserted
2271 *
2272 * Description:
2273 * Drivers often keep queueing requests until the hardware cannot accept
2274 * more, when that condition happens we need to put the request back
2275 * on the queue. Must be called with queue lock held.
2276 */
Jens Axboe165125e2007-07-24 09:28:11 +02002277void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278{
Jens Axboe2056a782006-03-23 20:00:26 +01002279 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 if (blk_rq_tagged(rq))
2282 blk_queue_end_tag(q, rq);
2283
2284 elv_requeue_request(q, rq);
2285}
2286
2287EXPORT_SYMBOL(blk_requeue_request);
2288
2289/**
2290 * blk_insert_request - insert a special request in to a request queue
2291 * @q: request queue where request should be inserted
2292 * @rq: request to be inserted
2293 * @at_head: insert request at head or tail of queue
2294 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 *
2296 * Description:
2297 * Many block devices need to execute commands asynchronously, so they don't
2298 * block the whole kernel from preemption during request execution. This is
2299 * accomplished normally by inserting aritficial requests tagged as
2300 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2301 * scheduled for actual execution by the request queue.
2302 *
2303 * We have the option of inserting the head or the tail of the queue.
2304 * Typically we use the tail for new ioctls and so forth. We use the head
2305 * of the queue for things like a QUEUE_FULL message from a device, or a
2306 * host that is unable to accept a particular command.
2307 */
Jens Axboe165125e2007-07-24 09:28:11 +02002308void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002309 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Tejun Heo 867d1192005-04-24 02:06:05 -05002311 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 unsigned long flags;
2313
2314 /*
2315 * tell I/O scheduler that this isn't a regular read/write (ie it
2316 * must not attempt merges on this) and that it acts as a soft
2317 * barrier
2318 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002319 rq->cmd_type = REQ_TYPE_SPECIAL;
2320 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
2322 rq->special = data;
2323
2324 spin_lock_irqsave(q->queue_lock, flags);
2325
2326 /*
2327 * If command is tagged, release the tag
2328 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002329 if (blk_rq_tagged(rq))
2330 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Tejun Heo 867d1192005-04-24 02:06:05 -05002332 drive_stat_acct(rq, rq->nr_sectors, 1);
2333 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002334 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 spin_unlock_irqrestore(q->queue_lock, flags);
2336}
2337
2338EXPORT_SYMBOL(blk_insert_request);
2339
Mike Christie0e75f902006-12-01 10:40:55 +01002340static int __blk_rq_unmap_user(struct bio *bio)
2341{
2342 int ret = 0;
2343
2344 if (bio) {
2345 if (bio_flagged(bio, BIO_USER_MAPPED))
2346 bio_unmap_user(bio);
2347 else
2348 ret = bio_uncopy_user(bio);
2349 }
2350
2351 return ret;
2352}
2353
NeilBrown3001ca72007-08-16 13:31:27 +02002354int blk_rq_append_bio(struct request_queue *q, struct request *rq,
2355 struct bio *bio)
2356{
2357 if (!rq->bio)
2358 blk_rq_bio_prep(q, rq, bio);
2359 else if (!ll_back_merge_fn(q, rq, bio))
2360 return -EINVAL;
2361 else {
2362 rq->biotail->bi_next = bio;
2363 rq->biotail = bio;
2364
2365 rq->data_len += bio->bi_size;
2366 }
2367 return 0;
2368}
2369EXPORT_SYMBOL(blk_rq_append_bio);
2370
Jens Axboe165125e2007-07-24 09:28:11 +02002371static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002372 void __user *ubuf, unsigned int len)
2373{
2374 unsigned long uaddr;
2375 struct bio *bio, *orig_bio;
2376 int reading, ret;
2377
2378 reading = rq_data_dir(rq) == READ;
2379
2380 /*
2381 * if alignment requirement is satisfied, map in user pages for
2382 * direct dma. else, set up kernel bounce buffers
2383 */
2384 uaddr = (unsigned long) ubuf;
2385 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2386 bio = bio_map_user(q, NULL, uaddr, len, reading);
2387 else
2388 bio = bio_copy_user(q, uaddr, len, reading);
2389
Jens Axboe29852592006-12-19 08:27:31 +01002390 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002391 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002392
2393 orig_bio = bio;
2394 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002395
Mike Christie0e75f902006-12-01 10:40:55 +01002396 /*
2397 * We link the bounce buffer in and could have to traverse it
2398 * later so we have to get a ref to prevent it from being freed
2399 */
2400 bio_get(bio);
2401
NeilBrown3001ca72007-08-16 13:31:27 +02002402 ret = blk_rq_append_bio(q, rq, bio);
2403 if (!ret)
2404 return bio->bi_size;
Mike Christie0e75f902006-12-01 10:40:55 +01002405
Mike Christie0e75f902006-12-01 10:40:55 +01002406 /* if it was boucned we must call the end io function */
NeilBrown6712ecf2007-09-27 12:47:43 +02002407 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002408 __blk_rq_unmap_user(orig_bio);
2409 bio_put(bio);
2410 return ret;
2411}
2412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413/**
2414 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2415 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002416 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 * @ubuf: the user buffer
2418 * @len: length of user data
2419 *
2420 * Description:
2421 * Data will be mapped directly for zero copy io, if possible. Otherwise
2422 * a kernel bounce buffer is used.
2423 *
2424 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2425 * still in process context.
2426 *
2427 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2428 * before being submitted to the device, as pages mapped may be out of
2429 * reach. It's the callers responsibility to make sure this happens. The
2430 * original bio must be passed back in to blk_rq_unmap_user() for proper
2431 * unmapping.
2432 */
Jens Axboe165125e2007-07-24 09:28:11 +02002433int blk_rq_map_user(struct request_queue *q, struct request *rq,
2434 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435{
Mike Christie0e75f902006-12-01 10:40:55 +01002436 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002437 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002438 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
Mike Christiedefd94b2005-12-05 02:37:06 -06002440 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002441 return -EINVAL;
2442 if (!len || !ubuf)
2443 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Mike Christie0e75f902006-12-01 10:40:55 +01002445 while (bytes_read != len) {
2446 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Mike Christie0e75f902006-12-01 10:40:55 +01002448 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2449 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2450 >> PAGE_SHIFT;
2451 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Mike Christie0e75f902006-12-01 10:40:55 +01002453 /*
2454 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2455 * pages. If this happens we just lower the requested
2456 * mapping len by a page so that we can fit
2457 */
2458 if (end - start > BIO_MAX_PAGES)
2459 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460
Mike Christie0e75f902006-12-01 10:40:55 +01002461 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2462 if (ret < 0)
2463 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002464 if (!bio)
2465 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002466 bytes_read += ret;
2467 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 }
2469
Mike Christie0e75f902006-12-01 10:40:55 +01002470 rq->buffer = rq->data = NULL;
2471 return 0;
2472unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002473 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002474 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475}
2476
2477EXPORT_SYMBOL(blk_rq_map_user);
2478
2479/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002480 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2481 * @q: request queue where request should be inserted
2482 * @rq: request to map data to
2483 * @iov: pointer to the iovec
2484 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002485 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002486 *
2487 * Description:
2488 * Data will be mapped directly for zero copy io, if possible. Otherwise
2489 * a kernel bounce buffer is used.
2490 *
2491 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2492 * still in process context.
2493 *
2494 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2495 * before being submitted to the device, as pages mapped may be out of
2496 * reach. It's the callers responsibility to make sure this happens. The
2497 * original bio must be passed back in to blk_rq_unmap_user() for proper
2498 * unmapping.
2499 */
Jens Axboe165125e2007-07-24 09:28:11 +02002500int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002501 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002502{
2503 struct bio *bio;
2504
2505 if (!iov || iov_count <= 0)
2506 return -EINVAL;
2507
2508 /* we don't allow misaligned data like bio_map_user() does. If the
2509 * user is using sg, they're expected to know the alignment constraints
2510 * and respect them accordingly */
2511 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2512 if (IS_ERR(bio))
2513 return PTR_ERR(bio);
2514
Mike Christie0e75f902006-12-01 10:40:55 +01002515 if (bio->bi_size != len) {
NeilBrown6712ecf2007-09-27 12:47:43 +02002516 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002517 bio_unmap_user(bio);
2518 return -EINVAL;
2519 }
2520
2521 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002522 blk_rq_bio_prep(q, rq, bio);
2523 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002524 return 0;
2525}
2526
2527EXPORT_SYMBOL(blk_rq_map_user_iov);
2528
2529/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002531 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 *
2533 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002534 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2535 * supply the original rq->bio from the blk_rq_map_user() return, since
2536 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002538int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002540 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002541 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002543 while (bio) {
2544 mapped_bio = bio;
2545 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002546 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
Jens Axboe48785bb2006-12-19 11:07:59 +01002548 ret2 = __blk_rq_unmap_user(mapped_bio);
2549 if (ret2 && !ret)
2550 ret = ret2;
2551
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002552 mapped_bio = bio;
2553 bio = bio->bi_next;
2554 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002555 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002556
2557 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
2559
2560EXPORT_SYMBOL(blk_rq_unmap_user);
2561
2562/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002563 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2564 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002565 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002566 * @kbuf: the kernel buffer
2567 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002568 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002569 */
Jens Axboe165125e2007-07-24 09:28:11 +02002570int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002571 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002572{
Mike Christie df46b9a2005-06-20 14:04:44 +02002573 struct bio *bio;
2574
Mike Christiedefd94b2005-12-05 02:37:06 -06002575 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002576 return -EINVAL;
2577 if (!len || !kbuf)
2578 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002579
2580 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002581 if (IS_ERR(bio))
2582 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002583
Jens Axboedd1cab92005-06-20 14:06:01 +02002584 if (rq_data_dir(rq) == WRITE)
2585 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002586
Jens Axboedd1cab92005-06-20 14:06:01 +02002587 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002588 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002589 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002590 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002591}
2592
2593EXPORT_SYMBOL(blk_rq_map_kern);
2594
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002595/**
2596 * blk_execute_rq_nowait - insert a request into queue for execution
2597 * @q: queue to insert the request in
2598 * @bd_disk: matching gendisk
2599 * @rq: request to insert
2600 * @at_head: insert request at head or tail of queue
2601 * @done: I/O completion handler
2602 *
2603 * Description:
2604 * Insert a fully prepared request at the back of the io scheduler queue
2605 * for execution. Don't wait for completion.
2606 */
Jens Axboe165125e2007-07-24 09:28:11 +02002607void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002608 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002609 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002610{
2611 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2612
2613 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002614 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002615 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002616 WARN_ON(irqs_disabled());
2617 spin_lock_irq(q->queue_lock);
2618 __elv_add_request(q, rq, where, 1);
2619 __generic_unplug_device(q);
2620 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002621}
Mike Christie6e39b692005-11-11 05:30:24 -06002622EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2623
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624/**
2625 * blk_execute_rq - insert a request into queue for execution
2626 * @q: queue to insert the request in
2627 * @bd_disk: matching gendisk
2628 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002629 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 *
2631 * Description:
2632 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002633 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 */
Jens Axboe165125e2007-07-24 09:28:11 +02002635int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002636 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002638 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 char sense[SCSI_SENSE_BUFFERSIZE];
2640 int err = 0;
2641
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 /*
2643 * we need an extra reference to the request, so we can look at
2644 * it after io completion
2645 */
2646 rq->ref_count++;
2647
2648 if (!rq->sense) {
2649 memset(sense, 0, sizeof(sense));
2650 rq->sense = sense;
2651 rq->sense_len = 0;
2652 }
2653
Jens Axboec00895a2006-09-30 20:29:12 +02002654 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002655 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
2658 if (rq->errors)
2659 err = -EIO;
2660
2661 return err;
2662}
2663
2664EXPORT_SYMBOL(blk_execute_rq);
2665
Jens Axboefd5d8062007-10-16 11:05:02 +02002666static void bio_end_empty_barrier(struct bio *bio, int err)
2667{
2668 if (err)
2669 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2670
2671 complete(bio->bi_private);
2672}
2673
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674/**
2675 * blkdev_issue_flush - queue a flush
2676 * @bdev: blockdev to issue flush for
2677 * @error_sector: error sector
2678 *
2679 * Description:
2680 * Issue a flush for the block device in question. Caller can supply
2681 * room for storing the error offset in case of a flush error, if they
2682 * wish to. Caller must run wait_for_completion() on its own.
2683 */
2684int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2685{
Jens Axboefd5d8062007-10-16 11:05:02 +02002686 DECLARE_COMPLETION_ONSTACK(wait);
Jens Axboe165125e2007-07-24 09:28:11 +02002687 struct request_queue *q;
Jens Axboefd5d8062007-10-16 11:05:02 +02002688 struct bio *bio;
2689 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690
2691 if (bdev->bd_disk == NULL)
2692 return -ENXIO;
2693
2694 q = bdev_get_queue(bdev);
2695 if (!q)
2696 return -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
Jens Axboefd5d8062007-10-16 11:05:02 +02002698 bio = bio_alloc(GFP_KERNEL, 0);
2699 if (!bio)
2700 return -ENOMEM;
2701
2702 bio->bi_end_io = bio_end_empty_barrier;
2703 bio->bi_private = &wait;
2704 bio->bi_bdev = bdev;
2705 submit_bio(1 << BIO_RW_BARRIER, bio);
2706
2707 wait_for_completion(&wait);
2708
2709 /*
2710 * The driver must store the error location in ->bi_sector, if
2711 * it supports it. For non-stacked drivers, this should be copied
2712 * from rq->sector.
2713 */
2714 if (error_sector)
2715 *error_sector = bio->bi_sector;
2716
2717 ret = 0;
2718 if (!bio_flagged(bio, BIO_UPTODATE))
2719 ret = -EIO;
2720
2721 bio_put(bio);
2722 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723}
2724
2725EXPORT_SYMBOL(blkdev_issue_flush);
2726
Adrian Bunk93d17d32005-06-25 14:59:10 -07002727static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728{
2729 int rw = rq_data_dir(rq);
2730
2731 if (!blk_fs_request(rq) || !rq->rq_disk)
2732 return;
2733
Jens Axboed72d9042005-11-01 08:35:42 +01002734 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002735 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002736 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 disk_round_stats(rq->rq_disk);
2738 rq->rq_disk->in_flight++;
2739 }
2740}
2741
2742/*
2743 * add-request adds a request to the linked list.
2744 * queue lock is held and interrupts disabled, as we muck with the
2745 * request queue list.
2746 */
Jens Axboe165125e2007-07-24 09:28:11 +02002747static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748{
2749 drive_stat_acct(req, req->nr_sectors, 1);
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 /*
2752 * elevator indicated where it wants this request to be
2753 * inserted at elevator_merge time
2754 */
2755 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2756}
2757
2758/*
2759 * disk_round_stats() - Round off the performance stats on a struct
2760 * disk_stats.
2761 *
2762 * The average IO queue length and utilisation statistics are maintained
2763 * by observing the current state of the queue length and the amount of
2764 * time it has been in this state for.
2765 *
2766 * Normally, that accounting is done on IO completion, but that can result
2767 * in more than a second's worth of IO being accounted for within any one
2768 * second, leading to >100% utilisation. To deal with that, we call this
2769 * function to do a round-off before returning the results when reading
2770 * /proc/diskstats. This accounts immediately for all queue usage up to
2771 * the current jiffies and restarts the counters again.
2772 */
2773void disk_round_stats(struct gendisk *disk)
2774{
2775 unsigned long now = jiffies;
2776
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002777 if (now == disk->stamp)
2778 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002780 if (disk->in_flight) {
2781 __disk_stat_add(disk, time_in_queue,
2782 disk->in_flight * (now - disk->stamp));
2783 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2784 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786}
2787
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002788EXPORT_SYMBOL_GPL(disk_round_stats);
2789
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790/*
2791 * queue lock must be held
2792 */
Jens Axboe165125e2007-07-24 09:28:11 +02002793void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 if (unlikely(!q))
2796 return;
2797 if (unlikely(--req->ref_count))
2798 return;
2799
Tejun Heo8922e162005-10-20 16:23:44 +02002800 elv_completed_request(q, req);
2801
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 /*
2803 * Request may not have originated from ll_rw_blk. if not,
2804 * it didn't come out of our reserved rq pools
2805 */
Jens Axboe49171e52006-08-10 08:59:11 +02002806 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002808 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002811 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812
2813 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002814 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 }
2816}
2817
Mike Christie6e39b692005-11-11 05:30:24 -06002818EXPORT_SYMBOL_GPL(__blk_put_request);
2819
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820void blk_put_request(struct request *req)
2821{
Tejun Heo8922e162005-10-20 16:23:44 +02002822 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002823 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
Tejun Heo8922e162005-10-20 16:23:44 +02002825 /*
2826 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2827 * following if (q) test.
2828 */
2829 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 spin_lock_irqsave(q->queue_lock, flags);
2831 __blk_put_request(q, req);
2832 spin_unlock_irqrestore(q->queue_lock, flags);
2833 }
2834}
2835
2836EXPORT_SYMBOL(blk_put_request);
2837
2838/**
2839 * blk_end_sync_rq - executes a completion event on a request
2840 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002841 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002843void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844{
Jens Axboec00895a2006-09-30 20:29:12 +02002845 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846
Jens Axboec00895a2006-09-30 20:29:12 +02002847 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 __blk_put_request(rq->q, rq);
2849
2850 /*
2851 * complete last, if this is a stack request the process (and thus
2852 * the rq pointer) could be invalid right after this complete()
2853 */
2854 complete(waiting);
2855}
2856EXPORT_SYMBOL(blk_end_sync_rq);
2857
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858/*
2859 * Has to be called with the request spinlock acquired
2860 */
Jens Axboe165125e2007-07-24 09:28:11 +02002861static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 struct request *next)
2863{
2864 if (!rq_mergeable(req) || !rq_mergeable(next))
2865 return 0;
2866
2867 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002868 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 */
2870 if (req->sector + req->nr_sectors != next->sector)
2871 return 0;
2872
2873 if (rq_data_dir(req) != rq_data_dir(next)
2874 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002875 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 return 0;
2877
2878 /*
2879 * If we are allowed to merge, then append bio list
2880 * from next to rq and release next. merge_requests_fn
2881 * will have updated segment counts, update sector
2882 * counts here.
2883 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002884 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 return 0;
2886
2887 /*
2888 * At this point we have either done a back merge
2889 * or front merge. We need the smaller start_time of
2890 * the merged requests to be the current request
2891 * for accounting purposes.
2892 */
2893 if (time_after(req->start_time, next->start_time))
2894 req->start_time = next->start_time;
2895
2896 req->biotail->bi_next = next->bio;
2897 req->biotail = next->biotail;
2898
2899 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2900
2901 elv_merge_requests(q, req, next);
2902
2903 if (req->rq_disk) {
2904 disk_round_stats(req->rq_disk);
2905 req->rq_disk->in_flight--;
2906 }
2907
Jens Axboe22e2c502005-06-27 10:55:12 +02002908 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2909
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 __blk_put_request(q, next);
2911 return 1;
2912}
2913
Jens Axboe165125e2007-07-24 09:28:11 +02002914static inline int attempt_back_merge(struct request_queue *q,
2915 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916{
2917 struct request *next = elv_latter_request(q, rq);
2918
2919 if (next)
2920 return attempt_merge(q, rq, next);
2921
2922 return 0;
2923}
2924
Jens Axboe165125e2007-07-24 09:28:11 +02002925static inline int attempt_front_merge(struct request_queue *q,
2926 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927{
2928 struct request *prev = elv_former_request(q, rq);
2929
2930 if (prev)
2931 return attempt_merge(q, prev, rq);
2932
2933 return 0;
2934}
2935
Tejun Heo52d9e672006-01-06 09:49:58 +01002936static void init_request_from_bio(struct request *req, struct bio *bio)
2937{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002938 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002939
2940 /*
2941 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2942 */
2943 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002944 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002945
2946 /*
2947 * REQ_BARRIER implies no merging, but lets make it explicit
2948 */
2949 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002950 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002951
Jens Axboeb31dc662006-06-13 08:26:10 +02002952 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002953 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002954 if (bio_rw_meta(bio))
2955 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002956
Tejun Heo52d9e672006-01-06 09:49:58 +01002957 req->errors = 0;
2958 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01002959 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002960 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02002961 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002962}
2963
Jens Axboe165125e2007-07-24 09:28:11 +02002964static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965{
Nick Piggin450991b2005-06-28 20:45:13 -07002966 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002967 int el_ret, nr_sectors, barrier, err;
2968 const unsigned short prio = bio_prio(bio);
2969 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002970 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
2974 /*
2975 * low level driver can indicate that it wants pages above a
2976 * certain limit bounced to low memory (ie for highmem, or even
2977 * ISA dma in theory)
2978 */
2979 blk_queue_bounce(q, &bio);
2980
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002982 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 err = -EOPNOTSUPP;
2984 goto end_io;
2985 }
2986
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 spin_lock_irq(q->queue_lock);
2988
Nick Piggin450991b2005-06-28 20:45:13 -07002989 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 goto get_rq;
2991
2992 el_ret = elv_merge(q, &req, bio);
2993 switch (el_ret) {
2994 case ELEVATOR_BACK_MERGE:
2995 BUG_ON(!rq_mergeable(req));
2996
Jens Axboe1aa4f242006-12-19 08:33:11 +01002997 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 break;
2999
Jens Axboe2056a782006-03-23 20:00:26 +01003000 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
3001
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 req->biotail->bi_next = bio;
3003 req->biotail = bio;
3004 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003005 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 drive_stat_acct(req, nr_sectors, 0);
3007 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003008 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 goto out;
3010
3011 case ELEVATOR_FRONT_MERGE:
3012 BUG_ON(!rq_mergeable(req));
3013
Jens Axboe1aa4f242006-12-19 08:33:11 +01003014 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 break;
3016
Jens Axboe2056a782006-03-23 20:00:26 +01003017 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
3018
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 bio->bi_next = req->bio;
3020 req->bio = bio;
3021
3022 /*
3023 * may not be valid. if the low level driver said
3024 * it didn't need a bounce buffer then it better
3025 * not touch req->buffer either...
3026 */
3027 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02003028 req->current_nr_sectors = bio_cur_sectors(bio);
3029 req->hard_cur_sectors = req->current_nr_sectors;
3030 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003032 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 drive_stat_acct(req, nr_sectors, 0);
3034 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003035 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 goto out;
3037
Nick Piggin450991b2005-06-28 20:45:13 -07003038 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003040 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 }
3042
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003044 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003045 * This sync check and mask will be re-done in init_request_from_bio(),
3046 * but we need to set it earlier to expose the sync flag to the
3047 * rq allocator and io schedulers.
3048 */
3049 rw_flags = bio_data_dir(bio);
3050 if (sync)
3051 rw_flags |= REQ_RW_SYNC;
3052
3053 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003054 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003055 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003056 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003057 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003058
Nick Piggin450991b2005-06-28 20:45:13 -07003059 /*
3060 * After dropping the lock and possibly sleeping here, our request
3061 * may now be mergeable after it had proven unmergeable (above).
3062 * We don't worry about that case for efficiency. It won't happen
3063 * often, and the elevators are able to handle it.
3064 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003065 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066
Nick Piggin450991b2005-06-28 20:45:13 -07003067 spin_lock_irq(q->queue_lock);
3068 if (elv_queue_empty(q))
3069 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 add_request(q, req);
3071out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003072 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 __generic_unplug_device(q);
3074
3075 spin_unlock_irq(q->queue_lock);
3076 return 0;
3077
3078end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003079 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 return 0;
3081}
3082
3083/*
3084 * If bio->bi_dev is a partition, remap the location
3085 */
3086static inline void blk_partition_remap(struct bio *bio)
3087{
3088 struct block_device *bdev = bio->bi_bdev;
3089
Jens Axboebf2de6f2007-09-27 13:01:25 +02003090 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003092 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
Jens Axboea3623572005-11-01 09:26:16 +01003094 p->sectors[rw] += bio_sectors(bio);
3095 p->ios[rw]++;
3096
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 bio->bi_sector += p->start_sect;
3098 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003099
3100 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3101 bdev->bd_dev, bio->bi_sector,
3102 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 }
3104}
3105
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106static void handle_bad_sector(struct bio *bio)
3107{
3108 char b[BDEVNAME_SIZE];
3109
3110 printk(KERN_INFO "attempt to access beyond end of device\n");
3111 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3112 bdevname(bio->bi_bdev, b),
3113 bio->bi_rw,
3114 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3115 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3116
3117 set_bit(BIO_EOF, &bio->bi_flags);
3118}
3119
Akinobu Mitac17bb492006-12-08 02:39:46 -08003120#ifdef CONFIG_FAIL_MAKE_REQUEST
3121
3122static DECLARE_FAULT_ATTR(fail_make_request);
3123
3124static int __init setup_fail_make_request(char *str)
3125{
3126 return setup_fault_attr(&fail_make_request, str);
3127}
3128__setup("fail_make_request=", setup_fail_make_request);
3129
3130static int should_fail_request(struct bio *bio)
3131{
3132 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3133 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3134 return should_fail(&fail_make_request, bio->bi_size);
3135
3136 return 0;
3137}
3138
3139static int __init fail_make_request_debugfs(void)
3140{
3141 return init_fault_attr_dentries(&fail_make_request,
3142 "fail_make_request");
3143}
3144
3145late_initcall(fail_make_request_debugfs);
3146
3147#else /* CONFIG_FAIL_MAKE_REQUEST */
3148
3149static inline int should_fail_request(struct bio *bio)
3150{
3151 return 0;
3152}
3153
3154#endif /* CONFIG_FAIL_MAKE_REQUEST */
3155
Jens Axboec07e2b42007-07-18 13:27:58 +02003156/*
3157 * Check whether this bio extends beyond the end of the device.
3158 */
3159static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
3160{
3161 sector_t maxsector;
3162
3163 if (!nr_sectors)
3164 return 0;
3165
3166 /* Test device or partition size, when known. */
3167 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3168 if (maxsector) {
3169 sector_t sector = bio->bi_sector;
3170
3171 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3172 /*
3173 * This may well happen - the kernel calls bread()
3174 * without checking the size of the device, e.g., when
3175 * mounting a device.
3176 */
3177 handle_bad_sector(bio);
3178 return 1;
3179 }
3180 }
3181
3182 return 0;
3183}
3184
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185/**
3186 * generic_make_request: hand a buffer to its device driver for I/O
3187 * @bio: The bio describing the location in memory and on the device.
3188 *
3189 * generic_make_request() is used to make I/O requests of block
3190 * devices. It is passed a &struct bio, which describes the I/O that needs
3191 * to be done.
3192 *
3193 * generic_make_request() does not return any status. The
3194 * success/failure status of the request, along with notification of
3195 * completion, is delivered asynchronously through the bio->bi_end_io
3196 * function described (one day) else where.
3197 *
3198 * The caller of generic_make_request must make sure that bi_io_vec
3199 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3200 * set to describe the device address, and the
3201 * bi_end_io and optionally bi_private are set to describe how
3202 * completion notification should be signaled.
3203 *
3204 * generic_make_request and the drivers it calls may use bi_next if this
3205 * bio happens to be merged with someone else, and may change bi_dev and
3206 * bi_sector for remaps as it sees fit. So the values of these fields
3207 * should NOT be depended on after the call to generic_make_request.
3208 */
Neil Brownd89d8792007-05-01 09:53:42 +02003209static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
Jens Axboe165125e2007-07-24 09:28:11 +02003211 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08003212 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003214 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215
3216 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217
Jens Axboec07e2b42007-07-18 13:27:58 +02003218 if (bio_check_eod(bio, nr_sectors))
3219 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
3221 /*
3222 * Resolve the mapping until finished. (drivers are
3223 * still free to implement/resolve their own stacking
3224 * by explicitly returning 0)
3225 *
3226 * NOTE: we don't repeat the blk_size check for each new device.
3227 * Stacking drivers are expected to know what they are doing.
3228 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003229 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003230 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 do {
3232 char b[BDEVNAME_SIZE];
3233
3234 q = bdev_get_queue(bio->bi_bdev);
3235 if (!q) {
3236 printk(KERN_ERR
3237 "generic_make_request: Trying to access "
3238 "nonexistent block-device %s (%Lu)\n",
3239 bdevname(bio->bi_bdev, b),
3240 (long long) bio->bi_sector);
3241end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003242 bio_endio(bio, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243 break;
3244 }
3245
Jens Axboe4fa253f2007-07-18 13:13:10 +02003246 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 printk("bio too big device %s (%u > %u)\n",
3248 bdevname(bio->bi_bdev, b),
3249 bio_sectors(bio),
3250 q->max_hw_sectors);
3251 goto end_io;
3252 }
3253
Nick Pigginfde6ad22005-06-23 00:08:53 -07003254 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 goto end_io;
3256
Akinobu Mitac17bb492006-12-08 02:39:46 -08003257 if (should_fail_request(bio))
3258 goto end_io;
3259
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 /*
3261 * If this device has partitions, remap block n
3262 * of partition p to block n+start(p) of the disk.
3263 */
3264 blk_partition_remap(bio);
3265
NeilBrown5ddfe962006-10-30 22:07:21 -08003266 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02003267 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003268 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003269
3270 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3271
NeilBrown5ddfe962006-10-30 22:07:21 -08003272 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003273 old_dev = bio->bi_bdev->bd_dev;
3274
Jens Axboec07e2b42007-07-18 13:27:58 +02003275 if (bio_check_eod(bio, nr_sectors))
3276 goto end_io;
NeilBrown5ddfe962006-10-30 22:07:21 -08003277
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 ret = q->make_request_fn(q, bio);
3279 } while (ret);
3280}
3281
Neil Brownd89d8792007-05-01 09:53:42 +02003282/*
3283 * We only want one ->make_request_fn to be active at a time,
3284 * else stack usage with stacked devices could be a problem.
3285 * So use current->bio_{list,tail} to keep a list of requests
3286 * submited by a make_request_fn function.
3287 * current->bio_tail is also used as a flag to say if
3288 * generic_make_request is currently active in this task or not.
3289 * If it is NULL, then no make_request is active. If it is non-NULL,
3290 * then a make_request is active, and new requests should be added
3291 * at the tail
3292 */
3293void generic_make_request(struct bio *bio)
3294{
3295 if (current->bio_tail) {
3296 /* make_request is active */
3297 *(current->bio_tail) = bio;
3298 bio->bi_next = NULL;
3299 current->bio_tail = &bio->bi_next;
3300 return;
3301 }
3302 /* following loop may be a bit non-obvious, and so deserves some
3303 * explanation.
3304 * Before entering the loop, bio->bi_next is NULL (as all callers
3305 * ensure that) so we have a list with a single bio.
3306 * We pretend that we have just taken it off a longer list, so
3307 * we assign bio_list to the next (which is NULL) and bio_tail
3308 * to &bio_list, thus initialising the bio_list of new bios to be
3309 * added. __generic_make_request may indeed add some more bios
3310 * through a recursive call to generic_make_request. If it
3311 * did, we find a non-NULL value in bio_list and re-enter the loop
3312 * from the top. In this case we really did just take the bio
3313 * of the top of the list (no pretending) and so fixup bio_list and
3314 * bio_tail or bi_next, and call into __generic_make_request again.
3315 *
3316 * The loop was structured like this to make only one call to
3317 * __generic_make_request (which is important as it is large and
3318 * inlined) and to keep the structure simple.
3319 */
3320 BUG_ON(bio->bi_next);
3321 do {
3322 current->bio_list = bio->bi_next;
3323 if (bio->bi_next == NULL)
3324 current->bio_tail = &current->bio_list;
3325 else
3326 bio->bi_next = NULL;
3327 __generic_make_request(bio);
3328 bio = current->bio_list;
3329 } while (bio);
3330 current->bio_tail = NULL; /* deactivate */
3331}
3332
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333EXPORT_SYMBOL(generic_make_request);
3334
3335/**
3336 * submit_bio: submit a bio to the block device layer for I/O
3337 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3338 * @bio: The &struct bio which describes the I/O
3339 *
3340 * submit_bio() is very similar in purpose to generic_make_request(), and
3341 * uses that function to do most of the work. Both are fairly rough
3342 * interfaces, @bio must be presetup and ready for I/O.
3343 *
3344 */
3345void submit_bio(int rw, struct bio *bio)
3346{
3347 int count = bio_sectors(bio);
3348
Jens Axboe22e2c502005-06-27 10:55:12 +02003349 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350
Jens Axboebf2de6f2007-09-27 13:01:25 +02003351 /*
3352 * If it's a regular read/write or a barrier with data attached,
3353 * go through the normal accounting stuff before submission.
3354 */
3355 if (!bio_empty_barrier(bio)) {
3356
3357 BIO_BUG_ON(!bio->bi_size);
3358 BIO_BUG_ON(!bio->bi_io_vec);
3359
3360 if (rw & WRITE) {
3361 count_vm_events(PGPGOUT, count);
3362 } else {
3363 task_io_account_read(bio->bi_size);
3364 count_vm_events(PGPGIN, count);
3365 }
3366
3367 if (unlikely(block_dump)) {
3368 char b[BDEVNAME_SIZE];
3369 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07003370 current->comm, task_pid_nr(current),
Jens Axboebf2de6f2007-09-27 13:01:25 +02003371 (rw & WRITE) ? "WRITE" : "READ",
3372 (unsigned long long)bio->bi_sector,
3373 bdevname(bio->bi_bdev,b));
3374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 }
3376
3377 generic_make_request(bio);
3378}
3379
3380EXPORT_SYMBOL(submit_bio);
3381
Adrian Bunk93d17d32005-06-25 14:59:10 -07003382static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383{
3384 if (blk_fs_request(rq)) {
3385 rq->hard_sector += nsect;
3386 rq->hard_nr_sectors -= nsect;
3387
3388 /*
3389 * Move the I/O submission pointers ahead if required.
3390 */
3391 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3392 (rq->sector <= rq->hard_sector)) {
3393 rq->sector = rq->hard_sector;
3394 rq->nr_sectors = rq->hard_nr_sectors;
3395 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3396 rq->current_nr_sectors = rq->hard_cur_sectors;
3397 rq->buffer = bio_data(rq->bio);
3398 }
3399
3400 /*
3401 * if total number of sectors is less than the first segment
3402 * size, something has gone terribly wrong
3403 */
3404 if (rq->nr_sectors < rq->current_nr_sectors) {
3405 printk("blk: request botched\n");
3406 rq->nr_sectors = rq->current_nr_sectors;
3407 }
3408 }
3409}
3410
3411static int __end_that_request_first(struct request *req, int uptodate,
3412 int nr_bytes)
3413{
3414 int total_bytes, bio_nbytes, error, next_idx = 0;
3415 struct bio *bio;
3416
Jens Axboe2056a782006-03-23 20:00:26 +01003417 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3418
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 /*
3420 * extend uptodate bool to allow < 0 value to be direct io error
3421 */
3422 error = 0;
3423 if (end_io_error(uptodate))
3424 error = !uptodate ? -EIO : uptodate;
3425
3426 /*
3427 * for a REQ_BLOCK_PC request, we want to carry any eventual
3428 * sense key with us all the way through
3429 */
3430 if (!blk_pc_request(req))
3431 req->errors = 0;
3432
3433 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003434 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 printk("end_request: I/O error, dev %s, sector %llu\n",
3436 req->rq_disk ? req->rq_disk->disk_name : "?",
3437 (unsigned long long)req->sector);
3438 }
3439
Jens Axboed72d9042005-11-01 08:35:42 +01003440 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003441 const int rw = rq_data_dir(req);
3442
Jens Axboe53e86062006-01-17 11:09:27 +01003443 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003444 }
3445
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 total_bytes = bio_nbytes = 0;
3447 while ((bio = req->bio) != NULL) {
3448 int nbytes;
3449
Jens Axboebf2de6f2007-09-27 13:01:25 +02003450 /*
3451 * For an empty barrier request, the low level driver must
3452 * store a potential error location in ->sector. We pass
3453 * that back up in ->bi_sector.
3454 */
3455 if (blk_empty_barrier(req))
3456 bio->bi_sector = req->sector;
3457
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 if (nr_bytes >= bio->bi_size) {
3459 req->bio = bio->bi_next;
3460 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02003461 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462 next_idx = 0;
3463 bio_nbytes = 0;
3464 } else {
3465 int idx = bio->bi_idx + next_idx;
3466
3467 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3468 blk_dump_rq_flags(req, "__end_that");
3469 printk("%s: bio idx %d >= vcnt %d\n",
3470 __FUNCTION__,
3471 bio->bi_idx, bio->bi_vcnt);
3472 break;
3473 }
3474
3475 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3476 BIO_BUG_ON(nbytes > bio->bi_size);
3477
3478 /*
3479 * not a complete bvec done
3480 */
3481 if (unlikely(nbytes > nr_bytes)) {
3482 bio_nbytes += nr_bytes;
3483 total_bytes += nr_bytes;
3484 break;
3485 }
3486
3487 /*
3488 * advance to the next vector
3489 */
3490 next_idx++;
3491 bio_nbytes += nbytes;
3492 }
3493
3494 total_bytes += nbytes;
3495 nr_bytes -= nbytes;
3496
3497 if ((bio = req->bio)) {
3498 /*
3499 * end more in this run, or just return 'not-done'
3500 */
3501 if (unlikely(nr_bytes <= 0))
3502 break;
3503 }
3504 }
3505
3506 /*
3507 * completely done
3508 */
3509 if (!req->bio)
3510 return 0;
3511
3512 /*
3513 * if the request wasn't completed, update state
3514 */
3515 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02003516 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 bio->bi_idx += next_idx;
3518 bio_iovec(bio)->bv_offset += nr_bytes;
3519 bio_iovec(bio)->bv_len -= nr_bytes;
3520 }
3521
3522 blk_recalc_rq_sectors(req, total_bytes >> 9);
3523 blk_recalc_rq_segments(req);
3524 return 1;
3525}
3526
3527/**
3528 * end_that_request_first - end I/O on a request
3529 * @req: the request being processed
3530 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3531 * @nr_sectors: number of sectors to end I/O on
3532 *
3533 * Description:
3534 * Ends I/O on a number of sectors attached to @req, and sets it up
3535 * for the next range of segments (if any) in the cluster.
3536 *
3537 * Return:
3538 * 0 - we are done with this request, call end_that_request_last()
3539 * 1 - still buffers pending for this request
3540 **/
3541int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3542{
3543 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3544}
3545
3546EXPORT_SYMBOL(end_that_request_first);
3547
3548/**
3549 * end_that_request_chunk - end I/O on a request
3550 * @req: the request being processed
3551 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3552 * @nr_bytes: number of bytes to complete
3553 *
3554 * Description:
3555 * Ends I/O on a number of bytes attached to @req, and sets it up
3556 * for the next range of segments (if any). Like end_that_request_first(),
3557 * but deals with bytes instead of sectors.
3558 *
3559 * Return:
3560 * 0 - we are done with this request, call end_that_request_last()
3561 * 1 - still buffers pending for this request
3562 **/
3563int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3564{
3565 return __end_that_request_first(req, uptodate, nr_bytes);
3566}
3567
3568EXPORT_SYMBOL(end_that_request_chunk);
3569
3570/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003571 * splice the completion data to a local structure and hand off to
3572 * process_completion_queue() to complete the requests
3573 */
3574static void blk_done_softirq(struct softirq_action *h)
3575{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003576 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003577
3578 local_irq_disable();
3579 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003580 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003581 local_irq_enable();
3582
3583 while (!list_empty(&local_list)) {
3584 struct request *rq = list_entry(local_list.next, struct request, donelist);
3585
3586 list_del_init(&rq->donelist);
3587 rq->q->softirq_done_fn(rq);
3588 }
3589}
3590
Satyam Sharmadb47d472007-08-23 09:29:40 +02003591static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action,
Jens Axboeff856ba2006-01-09 16:02:34 +01003592 void *hcpu)
3593{
3594 /*
3595 * If a CPU goes away, splice its entries to the current CPU
3596 * and trigger a run of the softirq
3597 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003598 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003599 int cpu = (unsigned long) hcpu;
3600
3601 local_irq_disable();
3602 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3603 &__get_cpu_var(blk_cpu_done));
3604 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3605 local_irq_enable();
3606 }
3607
3608 return NOTIFY_OK;
3609}
3610
3611
Satyam Sharmadb47d472007-08-23 09:29:40 +02003612static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003613 .notifier_call = blk_cpu_notify,
3614};
3615
Jens Axboeff856ba2006-01-09 16:02:34 +01003616/**
3617 * blk_complete_request - end I/O on a request
3618 * @req: the request being processed
3619 *
3620 * Description:
3621 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003622 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02003623 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01003624 * through a softirq handler. The user must have registered a completion
3625 * callback through blk_queue_softirq_done().
3626 **/
3627
3628void blk_complete_request(struct request *req)
3629{
3630 struct list_head *cpu_list;
3631 unsigned long flags;
3632
3633 BUG_ON(!req->q->softirq_done_fn);
3634
3635 local_irq_save(flags);
3636
3637 cpu_list = &__get_cpu_var(blk_cpu_done);
3638 list_add_tail(&req->donelist, cpu_list);
3639 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3640
3641 local_irq_restore(flags);
3642}
3643
3644EXPORT_SYMBOL(blk_complete_request);
3645
3646/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647 * queue lock must be held
3648 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003649void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650{
3651 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003652 int error;
3653
3654 /*
3655 * extend uptodate bool to allow < 0 value to be direct io error
3656 */
3657 error = 0;
3658 if (end_io_error(uptodate))
3659 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660
3661 if (unlikely(laptop_mode) && blk_fs_request(req))
3662 laptop_io_completion();
3663
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003664 /*
3665 * Account IO completion. bar_rq isn't accounted as a normal
3666 * IO on queueing nor completion. Accounting the containing
3667 * request is enough.
3668 */
3669 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003671 const int rw = rq_data_dir(req);
3672
3673 __disk_stat_inc(disk, ios[rw]);
3674 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675 disk_round_stats(disk);
3676 disk->in_flight--;
3677 }
3678 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003679 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 else
3681 __blk_put_request(req->q, req);
3682}
3683
3684EXPORT_SYMBOL(end_that_request_last);
3685
Jens Axboea0cd1282007-09-21 10:41:07 +02003686static inline void __end_request(struct request *rq, int uptodate,
3687 unsigned int nr_bytes, int dequeue)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688{
Jens Axboea0cd1282007-09-21 10:41:07 +02003689 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
3690 if (dequeue)
3691 blkdev_dequeue_request(rq);
3692 add_disk_randomness(rq->rq_disk);
3693 end_that_request_last(rq, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 }
3695}
3696
Jens Axboea0cd1282007-09-21 10:41:07 +02003697static unsigned int rq_byte_size(struct request *rq)
3698{
3699 if (blk_fs_request(rq))
3700 return rq->hard_nr_sectors << 9;
3701
3702 return rq->data_len;
3703}
3704
3705/**
3706 * end_queued_request - end all I/O on a queued request
3707 * @rq: the request being processed
3708 * @uptodate: error value or 0/1 uptodate flag
3709 *
3710 * Description:
3711 * Ends all I/O on a request, and removes it from the block layer queues.
3712 * Not suitable for normal IO completion, unless the driver still has
3713 * the request attached to the block layer.
3714 *
3715 **/
3716void end_queued_request(struct request *rq, int uptodate)
3717{
3718 __end_request(rq, uptodate, rq_byte_size(rq), 1);
3719}
3720EXPORT_SYMBOL(end_queued_request);
3721
3722/**
3723 * end_dequeued_request - end all I/O on a dequeued request
3724 * @rq: the request being processed
3725 * @uptodate: error value or 0/1 uptodate flag
3726 *
3727 * Description:
3728 * Ends all I/O on a request. The request must already have been
3729 * dequeued using blkdev_dequeue_request(), as is normally the case
3730 * for most drivers.
3731 *
3732 **/
3733void end_dequeued_request(struct request *rq, int uptodate)
3734{
3735 __end_request(rq, uptodate, rq_byte_size(rq), 0);
3736}
3737EXPORT_SYMBOL(end_dequeued_request);
3738
3739
3740/**
3741 * end_request - end I/O on the current segment of the request
Randy Dunlap8f731f72007-10-18 23:39:28 -07003742 * @req: the request being processed
Jens Axboea0cd1282007-09-21 10:41:07 +02003743 * @uptodate: error value or 0/1 uptodate flag
3744 *
3745 * Description:
3746 * Ends I/O on the current segment of a request. If that is the only
3747 * remaining segment, the request is also completed and freed.
3748 *
3749 * This is a remnant of how older block drivers handled IO completions.
3750 * Modern drivers typically end IO on the full request in one go, unless
3751 * they have a residual value to account for. For that case this function
3752 * isn't really useful, unless the residual just happens to be the
3753 * full current segment. In other words, don't use this function in new
3754 * code. Either use end_request_completely(), or the
3755 * end_that_request_chunk() (along with end_that_request_last()) for
3756 * partial completions.
3757 *
3758 **/
3759void end_request(struct request *req, int uptodate)
3760{
3761 __end_request(req, uptodate, req->hard_cur_sectors << 9, 1);
3762}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763EXPORT_SYMBOL(end_request);
3764
NeilBrown66846572007-08-16 13:31:28 +02003765static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3766 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003768 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3769 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770
3771 rq->nr_phys_segments = bio_phys_segments(q, bio);
3772 rq->nr_hw_segments = bio_hw_segments(q, bio);
3773 rq->current_nr_sectors = bio_cur_sectors(bio);
3774 rq->hard_cur_sectors = rq->current_nr_sectors;
3775 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3776 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003777 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
3779 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780
NeilBrown66846572007-08-16 13:31:28 +02003781 if (bio->bi_bdev)
3782 rq->rq_disk = bio->bi_bdev->bd_disk;
3783}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784
3785int kblockd_schedule_work(struct work_struct *work)
3786{
3787 return queue_work(kblockd_workqueue, work);
3788}
3789
3790EXPORT_SYMBOL(kblockd_schedule_work);
3791
Andrew Morton19a75d82007-05-09 02:33:56 -07003792void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003794 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795}
Andrew Morton19a75d82007-05-09 02:33:56 -07003796EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797
3798int __init blk_dev_init(void)
3799{
Jens Axboeff856ba2006-01-09 16:02:34 +01003800 int i;
3801
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 kblockd_workqueue = create_workqueue("kblockd");
3803 if (!kblockd_workqueue)
3804 panic("Failed to create kblockd\n");
3805
3806 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003807 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808
3809 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003810 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
3812 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003813 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003815 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003816 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3817
3818 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003819 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003820
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003821 blk_max_low_pfn = max_low_pfn - 1;
3822 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
3824 return 0;
3825}
3826
3827/*
3828 * IO Context helper functions
3829 */
3830void put_io_context(struct io_context *ioc)
3831{
3832 if (ioc == NULL)
3833 return;
3834
3835 BUG_ON(atomic_read(&ioc->refcount) == 0);
3836
3837 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003838 struct cfq_io_context *cic;
3839
Al Viro334e94d2006-03-18 15:05:53 -05003840 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 if (ioc->aic && ioc->aic->dtor)
3842 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003843 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003844 struct rb_node *n = rb_first(&ioc->cic_root);
3845
3846 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003847 cic->dtor(ioc);
3848 }
Al Viro334e94d2006-03-18 15:05:53 -05003849 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850
3851 kmem_cache_free(iocontext_cachep, ioc);
3852 }
3853}
3854EXPORT_SYMBOL(put_io_context);
3855
3856/* Called by the exitting task */
3857void exit_io_context(void)
3858{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003860 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861
Jens Axboe22e2c502005-06-27 10:55:12 +02003862 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 ioc = current->io_context;
3864 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003865 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
Oleg Nesterov25034d72006-08-29 09:15:14 +02003867 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 if (ioc->aic && ioc->aic->exit)
3869 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003870 if (ioc->cic_root.rb_node != NULL) {
3871 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3872 cic->exit(ioc);
3873 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003874
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 put_io_context(ioc);
3876}
3877
3878/*
3879 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003880 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003882 * This returned IO context doesn't have a specifically elevated refcount,
3883 * but since the current task itself holds a reference, the context can be
3884 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003886static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887{
3888 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889 struct io_context *ret;
3890
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003892 if (likely(ret))
3893 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894
Jens Axboeb5deef92006-07-19 23:39:40 +02003895 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 if (ret) {
3897 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003898 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003899 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003900 ret->last_waited = jiffies; /* doesn't matter... */
3901 ret->nr_batch_requests = 0; /* because this is 0 */
3902 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003903 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003904 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003905 /* make sure set_task_ioprio() sees the settings above */
3906 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003907 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003908 }
3909
3910 return ret;
3911}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003912
3913/*
3914 * If the current task has no IO context then create one and initialise it.
3915 * If it does have a context, take a ref on it.
3916 *
3917 * This is always called in the context of the task which submitted the I/O.
3918 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003919struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003920{
3921 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003922 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003923 if (likely(ret))
3924 atomic_inc(&ret->refcount);
3925 return ret;
3926}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927EXPORT_SYMBOL(get_io_context);
3928
3929void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3930{
3931 struct io_context *src = *psrc;
3932 struct io_context *dst = *pdst;
3933
3934 if (src) {
3935 BUG_ON(atomic_read(&src->refcount) == 0);
3936 atomic_inc(&src->refcount);
3937 put_io_context(dst);
3938 *pdst = src;
3939 }
3940}
3941EXPORT_SYMBOL(copy_io_context);
3942
3943void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3944{
3945 struct io_context *temp;
3946 temp = *ioc1;
3947 *ioc1 = *ioc2;
3948 *ioc2 = temp;
3949}
3950EXPORT_SYMBOL(swap_io_context);
3951
3952/*
3953 * sysfs parts below
3954 */
3955struct queue_sysfs_entry {
3956 struct attribute attr;
3957 ssize_t (*show)(struct request_queue *, char *);
3958 ssize_t (*store)(struct request_queue *, const char *, size_t);
3959};
3960
3961static ssize_t
3962queue_var_show(unsigned int var, char *page)
3963{
3964 return sprintf(page, "%d\n", var);
3965}
3966
3967static ssize_t
3968queue_var_store(unsigned long *var, const char *page, size_t count)
3969{
3970 char *p = (char *) page;
3971
3972 *var = simple_strtoul(p, &p, 10);
3973 return count;
3974}
3975
3976static ssize_t queue_requests_show(struct request_queue *q, char *page)
3977{
3978 return queue_var_show(q->nr_requests, (page));
3979}
3980
3981static ssize_t
3982queue_requests_store(struct request_queue *q, const char *page, size_t count)
3983{
3984 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003985 unsigned long nr;
3986 int ret = queue_var_store(&nr, page, count);
3987 if (nr < BLKDEV_MIN_RQ)
3988 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
Al Viroc981ff92006-03-18 13:51:29 -05003990 spin_lock_irq(q->queue_lock);
3991 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 blk_queue_congestion_threshold(q);
3993
3994 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003995 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003997 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998
3999 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004000 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07004002 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003
4004 if (rl->count[READ] >= q->nr_requests) {
4005 blk_set_queue_full(q, READ);
4006 } else if (rl->count[READ]+1 <= q->nr_requests) {
4007 blk_clear_queue_full(q, READ);
4008 wake_up(&rl->wait[READ]);
4009 }
4010
4011 if (rl->count[WRITE] >= q->nr_requests) {
4012 blk_set_queue_full(q, WRITE);
4013 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
4014 blk_clear_queue_full(q, WRITE);
4015 wake_up(&rl->wait[WRITE]);
4016 }
Al Viroc981ff92006-03-18 13:51:29 -05004017 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 return ret;
4019}
4020
4021static ssize_t queue_ra_show(struct request_queue *q, char *page)
4022{
4023 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
4024
4025 return queue_var_show(ra_kb, (page));
4026}
4027
4028static ssize_t
4029queue_ra_store(struct request_queue *q, const char *page, size_t count)
4030{
4031 unsigned long ra_kb;
4032 ssize_t ret = queue_var_store(&ra_kb, page, count);
4033
4034 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
4036 spin_unlock_irq(q->queue_lock);
4037
4038 return ret;
4039}
4040
4041static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
4042{
4043 int max_sectors_kb = q->max_sectors >> 1;
4044
4045 return queue_var_show(max_sectors_kb, (page));
4046}
4047
4048static ssize_t
4049queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
4050{
4051 unsigned long max_sectors_kb,
4052 max_hw_sectors_kb = q->max_hw_sectors >> 1,
4053 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
4054 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055
4056 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
4057 return -EINVAL;
4058 /*
4059 * Take the queue lock to update the readahead and max_sectors
4060 * values synchronously:
4061 */
4062 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 q->max_sectors = max_sectors_kb << 1;
4064 spin_unlock_irq(q->queue_lock);
4065
4066 return ret;
4067}
4068
4069static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
4070{
4071 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
4072
4073 return queue_var_show(max_hw_sectors_kb, (page));
4074}
4075
Jens Axboe563063a2007-05-10 13:29:51 +02004076static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
4077{
4078 return queue_var_show(q->max_phys_segments, page);
4079}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
Jens Axboe563063a2007-05-10 13:29:51 +02004081static ssize_t queue_max_segments_store(struct request_queue *q,
4082 const char *page, size_t count)
4083{
4084 unsigned long segments;
4085 ssize_t ret = queue_var_store(&segments, page, count);
4086
4087 spin_lock_irq(q->queue_lock);
4088 q->max_phys_segments = segments;
4089 spin_unlock_irq(q->queue_lock);
4090
4091 return ret;
4092}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093static struct queue_sysfs_entry queue_requests_entry = {
4094 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
4095 .show = queue_requests_show,
4096 .store = queue_requests_store,
4097};
4098
4099static struct queue_sysfs_entry queue_ra_entry = {
4100 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4101 .show = queue_ra_show,
4102 .store = queue_ra_store,
4103};
4104
4105static struct queue_sysfs_entry queue_max_sectors_entry = {
4106 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4107 .show = queue_max_sectors_show,
4108 .store = queue_max_sectors_store,
4109};
4110
4111static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4112 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4113 .show = queue_max_hw_sectors_show,
4114};
4115
Jens Axboe563063a2007-05-10 13:29:51 +02004116static struct queue_sysfs_entry queue_max_segments_entry = {
4117 .attr = {.name = "max_segments", .mode = S_IRUGO | S_IWUSR },
4118 .show = queue_max_segments_show,
4119 .store = queue_max_segments_store,
4120};
4121
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122static struct queue_sysfs_entry queue_iosched_entry = {
4123 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4124 .show = elv_iosched_show,
4125 .store = elv_iosched_store,
4126};
4127
4128static struct attribute *default_attrs[] = {
4129 &queue_requests_entry.attr,
4130 &queue_ra_entry.attr,
4131 &queue_max_hw_sectors_entry.attr,
4132 &queue_max_sectors_entry.attr,
Jens Axboe563063a2007-05-10 13:29:51 +02004133 &queue_max_segments_entry.attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134 &queue_iosched_entry.attr,
4135 NULL,
4136};
4137
4138#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4139
4140static ssize_t
4141queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4142{
4143 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004144 struct request_queue *q =
4145 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004146 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004147
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004149 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004150 mutex_lock(&q->sysfs_lock);
4151 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4152 mutex_unlock(&q->sysfs_lock);
4153 return -ENOENT;
4154 }
4155 res = entry->show(q, page);
4156 mutex_unlock(&q->sysfs_lock);
4157 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158}
4159
4160static ssize_t
4161queue_attr_store(struct kobject *kobj, struct attribute *attr,
4162 const char *page, size_t length)
4163{
4164 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004165 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166
Al Viro483f4af2006-03-18 18:34:37 -05004167 ssize_t res;
4168
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004170 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004171 mutex_lock(&q->sysfs_lock);
4172 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4173 mutex_unlock(&q->sysfs_lock);
4174 return -ENOENT;
4175 }
4176 res = entry->store(q, page, length);
4177 mutex_unlock(&q->sysfs_lock);
4178 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004179}
4180
4181static struct sysfs_ops queue_sysfs_ops = {
4182 .show = queue_attr_show,
4183 .store = queue_attr_store,
4184};
4185
Adrian Bunk93d17d32005-06-25 14:59:10 -07004186static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004187 .sysfs_ops = &queue_sysfs_ops,
4188 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004189 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190};
4191
4192int blk_register_queue(struct gendisk *disk)
4193{
4194 int ret;
4195
Jens Axboe165125e2007-07-24 09:28:11 +02004196 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197
4198 if (!q || !q->request_fn)
4199 return -ENXIO;
4200
4201 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202
Al Viro483f4af2006-03-18 18:34:37 -05004203 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 if (ret < 0)
4205 return ret;
4206
Al Viro483f4af2006-03-18 18:34:37 -05004207 kobject_uevent(&q->kobj, KOBJ_ADD);
4208
Linus Torvalds1da177e2005-04-16 15:20:36 -07004209 ret = elv_register_queue(q);
4210 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004211 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4212 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213 return ret;
4214 }
4215
4216 return 0;
4217}
4218
4219void blk_unregister_queue(struct gendisk *disk)
4220{
Jens Axboe165125e2007-07-24 09:28:11 +02004221 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004222
4223 if (q && q->request_fn) {
4224 elv_unregister_queue(q);
4225
Al Viro483f4af2006-03-18 18:34:37 -05004226 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4227 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004228 kobject_put(&disk->kobj);
4229 }
4230}