blob: 1014d349a386b48385a75f0a98de7186d25df73d [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 Mortonfaccbd4b2006-12-10 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;
Jens Axboef5659132007-09-21 10:44:19 +02001325 struct scatterlist *next_sg, *sg;
NeilBrown5705f702007-09-25 12:35:59 +02001326 struct req_iterator iter;
1327 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 Axboef5659132007-09-21 10:44:19 +02001336 sg = next_sg = &sglist[0];
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 Axboef5659132007-09-21 10:44:19 +02001352 sg = next_sg;
1353 next_sg = sg_next(sg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Jens Axboe60573b82007-10-17 13:02:33 +02001355 memset(sg, 0, sizeof(*sg));
Jens Axboef5659132007-09-21 10:44:19 +02001356 sg->page = bvec->bv_page;
1357 sg->length = nbytes;
1358 sg->offset = bvec->bv_offset;
Jens Axboe6c92e692007-08-16 13:43:12 +02001359 nsegs++;
1360 }
1361 bvprv = bvec;
NeilBrown5705f702007-09-25 12:35:59 +02001362 } /* segments in rq */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
1364 return nsegs;
1365}
1366
1367EXPORT_SYMBOL(blk_rq_map_sg);
1368
1369/*
1370 * the standard queue merge functions, can be overridden with device
1371 * specific ones if so desired
1372 */
1373
Jens Axboe165125e2007-07-24 09:28:11 +02001374static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 struct request *req,
1376 struct bio *bio)
1377{
1378 int nr_phys_segs = bio_phys_segments(q, bio);
1379
1380 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001381 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 if (req == q->last_merge)
1383 q->last_merge = NULL;
1384 return 0;
1385 }
1386
1387 /*
1388 * A hw segment is just getting larger, bump just the phys
1389 * counter.
1390 */
1391 req->nr_phys_segments += nr_phys_segs;
1392 return 1;
1393}
1394
Jens Axboe165125e2007-07-24 09:28:11 +02001395static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 struct request *req,
1397 struct bio *bio)
1398{
1399 int nr_hw_segs = bio_hw_segments(q, bio);
1400 int nr_phys_segs = bio_phys_segments(q, bio);
1401
1402 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1403 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001404 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (req == q->last_merge)
1406 q->last_merge = NULL;
1407 return 0;
1408 }
1409
1410 /*
1411 * This will form the start of a new hw segment. Bump both
1412 * counters.
1413 */
1414 req->nr_hw_segments += nr_hw_segs;
1415 req->nr_phys_segments += nr_phys_segs;
1416 return 1;
1417}
1418
NeilBrown3001ca72007-08-16 13:31:27 +02001419static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1420 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421{
Mike Christiedefd94b2005-12-05 02:37:06 -06001422 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 int len;
1424
Mike Christiedefd94b2005-12-05 02:37:06 -06001425 if (unlikely(blk_pc_request(req)))
1426 max_sectors = q->max_hw_sectors;
1427 else
1428 max_sectors = q->max_sectors;
1429
1430 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001431 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 if (req == q->last_merge)
1433 q->last_merge = NULL;
1434 return 0;
1435 }
1436 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1437 blk_recount_segments(q, req->biotail);
1438 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1439 blk_recount_segments(q, bio);
1440 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1441 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1442 !BIOVEC_VIRT_OVERSIZE(len)) {
1443 int mergeable = ll_new_mergeable(q, req, bio);
1444
1445 if (mergeable) {
1446 if (req->nr_hw_segments == 1)
1447 req->bio->bi_hw_front_size = len;
1448 if (bio->bi_hw_segments == 1)
1449 bio->bi_hw_back_size = len;
1450 }
1451 return mergeable;
1452 }
1453
1454 return ll_new_hw_segment(q, req, bio);
1455}
1456
Jens Axboe165125e2007-07-24 09:28:11 +02001457static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 struct bio *bio)
1459{
Mike Christiedefd94b2005-12-05 02:37:06 -06001460 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 int len;
1462
Mike Christiedefd94b2005-12-05 02:37:06 -06001463 if (unlikely(blk_pc_request(req)))
1464 max_sectors = q->max_hw_sectors;
1465 else
1466 max_sectors = q->max_sectors;
1467
1468
1469 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001470 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 if (req == q->last_merge)
1472 q->last_merge = NULL;
1473 return 0;
1474 }
1475 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1476 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1477 blk_recount_segments(q, bio);
1478 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1479 blk_recount_segments(q, req->bio);
1480 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1481 !BIOVEC_VIRT_OVERSIZE(len)) {
1482 int mergeable = ll_new_mergeable(q, req, bio);
1483
1484 if (mergeable) {
1485 if (bio->bi_hw_segments == 1)
1486 bio->bi_hw_front_size = len;
1487 if (req->nr_hw_segments == 1)
1488 req->biotail->bi_hw_back_size = len;
1489 }
1490 return mergeable;
1491 }
1492
1493 return ll_new_hw_segment(q, req, bio);
1494}
1495
Jens Axboe165125e2007-07-24 09:28:11 +02001496static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 struct request *next)
1498{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001499 int total_phys_segments;
1500 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502 /*
1503 * First check if the either of the requests are re-queued
1504 * requests. Can't merge them if they are.
1505 */
1506 if (req->special || next->special)
1507 return 0;
1508
1509 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001510 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 */
1512 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1513 return 0;
1514
1515 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1516 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1517 total_phys_segments--;
1518
1519 if (total_phys_segments > q->max_phys_segments)
1520 return 0;
1521
1522 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1523 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1524 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1525 /*
1526 * propagate the combined length to the end of the requests
1527 */
1528 if (req->nr_hw_segments == 1)
1529 req->bio->bi_hw_front_size = len;
1530 if (next->nr_hw_segments == 1)
1531 next->biotail->bi_hw_back_size = len;
1532 total_hw_segments--;
1533 }
1534
1535 if (total_hw_segments > q->max_hw_segments)
1536 return 0;
1537
1538 /* Merge is OK... */
1539 req->nr_phys_segments = total_phys_segments;
1540 req->nr_hw_segments = total_hw_segments;
1541 return 1;
1542}
1543
1544/*
1545 * "plug" the device if there are no outstanding requests: this will
1546 * force the transfer to start only after we have put all the requests
1547 * on the list.
1548 *
1549 * This is called with interrupts off and no requests on the queue and
1550 * with the queue lock held.
1551 */
Jens Axboe165125e2007-07-24 09:28:11 +02001552void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553{
1554 WARN_ON(!irqs_disabled());
1555
1556 /*
1557 * don't plug a stopped queue, it must be paired with blk_start_queue()
1558 * which will restart the queueing
1559 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001560 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 return;
1562
Jens Axboe2056a782006-03-23 20:00:26 +01001563 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001565 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567}
1568
1569EXPORT_SYMBOL(blk_plug_device);
1570
1571/*
1572 * remove the queue from the plugged list, if present. called with
1573 * queue lock held and interrupts disabled.
1574 */
Jens Axboe165125e2007-07-24 09:28:11 +02001575int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576{
1577 WARN_ON(!irqs_disabled());
1578
1579 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1580 return 0;
1581
1582 del_timer(&q->unplug_timer);
1583 return 1;
1584}
1585
1586EXPORT_SYMBOL(blk_remove_plug);
1587
1588/*
1589 * remove the plug and let it rip..
1590 */
Jens Axboe165125e2007-07-24 09:28:11 +02001591void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001593 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 return;
1595
1596 if (!blk_remove_plug(q))
1597 return;
1598
Jens Axboe22e2c502005-06-27 10:55:12 +02001599 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
1601EXPORT_SYMBOL(__generic_unplug_device);
1602
1603/**
1604 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001605 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 *
1607 * Description:
1608 * Linux uses plugging to build bigger requests queues before letting
1609 * the device have at them. If a queue is plugged, the I/O scheduler
1610 * is still adding and merging requests on the queue. Once the queue
1611 * gets unplugged, the request_fn defined for the queue is invoked and
1612 * transfers started.
1613 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001614void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
1616 spin_lock_irq(q->queue_lock);
1617 __generic_unplug_device(q);
1618 spin_unlock_irq(q->queue_lock);
1619}
1620EXPORT_SYMBOL(generic_unplug_device);
1621
1622static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1623 struct page *page)
1624{
Jens Axboe165125e2007-07-24 09:28:11 +02001625 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
1627 /*
1628 * devices don't necessarily have an ->unplug_fn defined
1629 */
Jens Axboe2056a782006-03-23 20:00:26 +01001630 if (q->unplug_fn) {
1631 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1632 q->rq.count[READ] + q->rq.count[WRITE]);
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636}
1637
David Howells65f27f32006-11-22 14:55:48 +00001638static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639{
Jens Axboe165125e2007-07-24 09:28:11 +02001640 struct request_queue *q =
1641 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
Jens Axboe2056a782006-03-23 20:00:26 +01001643 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1644 q->rq.count[READ] + q->rq.count[WRITE]);
1645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 q->unplug_fn(q);
1647}
1648
1649static void blk_unplug_timeout(unsigned long data)
1650{
Jens Axboe165125e2007-07-24 09:28:11 +02001651 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
Jens Axboe2056a782006-03-23 20:00:26 +01001653 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1654 q->rq.count[READ] + q->rq.count[WRITE]);
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 kblockd_schedule_work(&q->unplug_work);
1657}
1658
1659/**
1660 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001661 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 *
1663 * Description:
1664 * blk_start_queue() will clear the stop flag on the queue, and call
1665 * the request_fn for the queue if it was in a stopped state when
1666 * entered. Also see blk_stop_queue(). Queue lock must be held.
1667 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001668void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001670 WARN_ON(!irqs_disabled());
1671
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1673
1674 /*
1675 * one level of recursion is ok and is much faster than kicking
1676 * the unplug handling
1677 */
1678 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1679 q->request_fn(q);
1680 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1681 } else {
1682 blk_plug_device(q);
1683 kblockd_schedule_work(&q->unplug_work);
1684 }
1685}
1686
1687EXPORT_SYMBOL(blk_start_queue);
1688
1689/**
1690 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001691 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 *
1693 * Description:
1694 * The Linux block layer assumes that a block driver will consume all
1695 * entries on the request queue when the request_fn strategy is called.
1696 * Often this will not happen, because of hardware limitations (queue
1697 * depth settings). If a device driver gets a 'queue full' response,
1698 * or if it simply chooses not to queue more I/O at one point, it can
1699 * call this function to prevent the request_fn from being called until
1700 * the driver has signalled it's ready to go again. This happens by calling
1701 * blk_start_queue() to restart queue operations. Queue lock must be held.
1702 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001703void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
1705 blk_remove_plug(q);
1706 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1707}
1708EXPORT_SYMBOL(blk_stop_queue);
1709
1710/**
1711 * blk_sync_queue - cancel any pending callbacks on a queue
1712 * @q: the queue
1713 *
1714 * Description:
1715 * The block layer may perform asynchronous callback activity
1716 * on a queue, such as calling the unplug function after a timeout.
1717 * A block device may call blk_sync_queue to ensure that any
1718 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001719 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * that its ->make_request_fn will not re-add plugging prior to calling
1721 * this function.
1722 *
1723 */
1724void blk_sync_queue(struct request_queue *q)
1725{
1726 del_timer_sync(&q->unplug_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727}
1728EXPORT_SYMBOL(blk_sync_queue);
1729
1730/**
1731 * blk_run_queue - run a single device queue
1732 * @q: The queue to run
1733 */
1734void blk_run_queue(struct request_queue *q)
1735{
1736 unsigned long flags;
1737
1738 spin_lock_irqsave(q->queue_lock, flags);
1739 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001740
1741 /*
1742 * Only recurse once to avoid overrunning the stack, let the unplug
1743 * handling reinvoke the handler shortly if we already got there.
1744 */
1745 if (!elv_queue_empty(q)) {
1746 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1747 q->request_fn(q);
1748 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1749 } else {
1750 blk_plug_device(q);
1751 kblockd_schedule_work(&q->unplug_work);
1752 }
1753 }
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 spin_unlock_irqrestore(q->queue_lock, flags);
1756}
1757EXPORT_SYMBOL(blk_run_queue);
1758
1759/**
Jens Axboe165125e2007-07-24 09:28:11 +02001760 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001761 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 *
1763 * Description:
1764 * blk_cleanup_queue is the pair to blk_init_queue() or
1765 * blk_queue_make_request(). It should be called when a request queue is
1766 * being released; typically when a block device is being de-registered.
1767 * Currently, its primary task it to free all the &struct request
1768 * structures that were allocated to the queue and the queue itself.
1769 *
1770 * Caveat:
1771 * Hopefully the low level driver will have finished any
1772 * outstanding requests first...
1773 **/
Al Viro483f4af2006-03-18 18:34:37 -05001774static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
Jens Axboe165125e2007-07-24 09:28:11 +02001776 struct request_queue *q =
1777 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 struct request_list *rl = &q->rq;
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 blk_sync_queue(q);
1781
1782 if (rl->rq_pool)
1783 mempool_destroy(rl->rq_pool);
1784
1785 if (q->queue_tags)
1786 __blk_queue_free_tags(q);
1787
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001788 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 kmem_cache_free(requestq_cachep, q);
1791}
1792
Jens Axboe165125e2007-07-24 09:28:11 +02001793void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001794{
1795 kobject_put(&q->kobj);
1796}
1797EXPORT_SYMBOL(blk_put_queue);
1798
Jens Axboe165125e2007-07-24 09:28:11 +02001799void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001800{
1801 mutex_lock(&q->sysfs_lock);
1802 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1803 mutex_unlock(&q->sysfs_lock);
1804
1805 if (q->elevator)
1806 elevator_exit(q->elevator);
1807
1808 blk_put_queue(q);
1809}
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811EXPORT_SYMBOL(blk_cleanup_queue);
1812
Jens Axboe165125e2007-07-24 09:28:11 +02001813static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
1815 struct request_list *rl = &q->rq;
1816
1817 rl->count[READ] = rl->count[WRITE] = 0;
1818 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001819 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 init_waitqueue_head(&rl->wait[READ]);
1821 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Christoph Lameter19460892005-06-23 00:08:19 -07001823 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1824 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
1826 if (!rl->rq_pool)
1827 return -ENOMEM;
1828
1829 return 0;
1830}
1831
Jens Axboe165125e2007-07-24 09:28:11 +02001832struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
Christoph Lameter19460892005-06-23 00:08:19 -07001834 return blk_alloc_queue_node(gfp_mask, -1);
1835}
1836EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Al Viro483f4af2006-03-18 18:34:37 -05001838static struct kobj_type queue_ktype;
1839
Jens Axboe165125e2007-07-24 09:28:11 +02001840struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001841{
Jens Axboe165125e2007-07-24 09:28:11 +02001842 struct request_queue *q;
Christoph Lameter19460892005-06-23 00:08:19 -07001843
Christoph Lameter94f60302007-07-17 04:03:29 -07001844 q = kmem_cache_alloc_node(requestq_cachep,
1845 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 if (!q)
1847 return NULL;
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001850
Greg Kroah-Hartman19c38de2007-09-12 15:06:57 -07001851 kobject_set_name(&q->kobj, "%s", "queue");
Al Viro483f4af2006-03-18 18:34:37 -05001852 q->kobj.ktype = &queue_ktype;
1853 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
1855 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1856 q->backing_dev_info.unplug_io_data = q;
1857
Al Viro483f4af2006-03-18 18:34:37 -05001858 mutex_init(&q->sysfs_lock);
1859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 return q;
1861}
Christoph Lameter19460892005-06-23 00:08:19 -07001862EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
1864/**
1865 * blk_init_queue - prepare a request queue for use with a block device
1866 * @rfn: The function to be called to process requests that have been
1867 * placed on the queue.
1868 * @lock: Request queue spin lock
1869 *
1870 * Description:
1871 * If a block device wishes to use the standard request handling procedures,
1872 * which sorts requests and coalesces adjacent requests, then it must
1873 * call blk_init_queue(). The function @rfn will be called when there
1874 * are requests on the queue that need to be processed. If the device
1875 * supports plugging, then @rfn may not be called immediately when requests
1876 * are available on the queue, but may be called at some time later instead.
1877 * Plugged queues are generally unplugged when a buffer belonging to one
1878 * of the requests on the queue is needed, or due to memory pressure.
1879 *
1880 * @rfn is not required, or even expected, to remove all requests off the
1881 * queue, but only as many as it can handle at a time. If it does leave
1882 * requests on the queue, it is responsible for arranging that the requests
1883 * get dealt with eventually.
1884 *
1885 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001886 * request queue; this lock will be taken also from interrupt context, so irq
1887 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 *
1889 * Function returns a pointer to the initialized request queue, or NULL if
1890 * it didn't succeed.
1891 *
1892 * Note:
1893 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1894 * when the block device is deactivated (such as at module unload).
1895 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001896
Jens Axboe165125e2007-07-24 09:28:11 +02001897struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Christoph Lameter19460892005-06-23 00:08:19 -07001899 return blk_init_queue_node(rfn, lock, -1);
1900}
1901EXPORT_SYMBOL(blk_init_queue);
1902
Jens Axboe165125e2007-07-24 09:28:11 +02001903struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001904blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1905{
Jens Axboe165125e2007-07-24 09:28:11 +02001906 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
1908 if (!q)
1909 return NULL;
1910
Christoph Lameter19460892005-06-23 00:08:19 -07001911 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001912 if (blk_init_free_list(q)) {
1913 kmem_cache_free(requestq_cachep, q);
1914 return NULL;
1915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
152587d2005-04-12 16:22:06 -05001917 /*
1918 * if caller didn't supply a lock, they get per-queue locking with
1919 * our embedded lock
1920 */
1921 if (!lock) {
1922 spin_lock_init(&q->__queue_lock);
1923 lock = &q->__queue_lock;
1924 }
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 q->prep_rq_fn = NULL;
1928 q->unplug_fn = generic_unplug_device;
1929 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1930 q->queue_lock = lock;
1931
1932 blk_queue_segment_boundary(q, 0xffffffff);
1933
1934 blk_queue_make_request(q, __make_request);
1935 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1936
1937 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1938 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1939
Alan Stern44ec9542007-02-20 11:01:57 -05001940 q->sg_reserved_size = INT_MAX;
1941
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 /*
1943 * all done
1944 */
1945 if (!elevator_init(q, NULL)) {
1946 blk_queue_congestion_threshold(q);
1947 return q;
1948 }
1949
Al Viro8669aaf2006-03-18 13:50:00 -05001950 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 return NULL;
1952}
Christoph Lameter19460892005-06-23 00:08:19 -07001953EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Jens Axboe165125e2007-07-24 09:28:11 +02001955int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001957 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001958 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 return 0;
1960 }
1961
1962 return 1;
1963}
1964
1965EXPORT_SYMBOL(blk_get_queue);
1966
Jens Axboe165125e2007-07-24 09:28:11 +02001967static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001969 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001970 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 mempool_free(rq, q->rq.rq_pool);
1972}
1973
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001974static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001975blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976{
1977 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1978
1979 if (!rq)
1980 return NULL;
1981
1982 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001983 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 * see bio.h and blkdev.h
1985 */
Jens Axboe49171e52006-08-10 08:59:11 +02001986 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Tejun Heocb98fc82005-10-28 08:29:39 +02001988 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001989 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001990 mempool_free(rq, q->rq.rq_pool);
1991 return NULL;
1992 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02001993 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02001994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
Tejun Heocb98fc82005-10-28 08:29:39 +02001996 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997}
1998
1999/*
2000 * ioc_batching returns true if the ioc is a valid batching request and
2001 * should be given priority access to a request.
2002 */
Jens Axboe165125e2007-07-24 09:28:11 +02002003static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004{
2005 if (!ioc)
2006 return 0;
2007
2008 /*
2009 * Make sure the process is able to allocate at least 1 request
2010 * even if the batch times out, otherwise we could theoretically
2011 * lose wakeups.
2012 */
2013 return ioc->nr_batch_requests == q->nr_batching ||
2014 (ioc->nr_batch_requests > 0
2015 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2016}
2017
2018/*
2019 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2020 * will cause the process to be a "batcher" on all queues in the system. This
2021 * is the behaviour we want though - once it gets a wakeup it should be given
2022 * a nice run.
2023 */
Jens Axboe165125e2007-07-24 09:28:11 +02002024static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
2026 if (!ioc || ioc_batching(q, ioc))
2027 return;
2028
2029 ioc->nr_batch_requests = q->nr_batching;
2030 ioc->last_waited = jiffies;
2031}
2032
Jens Axboe165125e2007-07-24 09:28:11 +02002033static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
2035 struct request_list *rl = &q->rq;
2036
2037 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002038 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
2040 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 if (waitqueue_active(&rl->wait[rw]))
2042 wake_up(&rl->wait[rw]);
2043
2044 blk_clear_queue_full(q, rw);
2045 }
2046}
2047
2048/*
2049 * A request has just been released. Account for it, update the full and
2050 * congestion status, wake up any waiters. Called under q->queue_lock.
2051 */
Jens Axboe165125e2007-07-24 09:28:11 +02002052static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053{
2054 struct request_list *rl = &q->rq;
2055
2056 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002057 if (priv)
2058 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
2060 __freed_request(q, rw);
2061
2062 if (unlikely(rl->starved[rw ^ 1]))
2063 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064}
2065
2066#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2067/*
Nick Piggind6344532005-06-28 20:45:14 -07002068 * Get a free request, queue_lock must be held.
2069 * Returns NULL on failure, with queue_lock held.
2070 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 */
Jens Axboe165125e2007-07-24 09:28:11 +02002072static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002073 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
2075 struct request *rq = NULL;
2076 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002077 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002078 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002079 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
Jens Axboe7749a8d2006-12-13 13:02:26 +01002081 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002082 if (may_queue == ELV_MQUEUE_NO)
2083 goto rq_starved;
2084
2085 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2086 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002087 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002088 /*
2089 * The queue will fill after this allocation, so set
2090 * it as full, and mark this process as "batching".
2091 * This process will be allowed to complete a batch of
2092 * requests, others will be blocked.
2093 */
2094 if (!blk_queue_full(q, rw)) {
2095 ioc_set_batching(q, ioc);
2096 blk_set_queue_full(q, rw);
2097 } else {
2098 if (may_queue != ELV_MQUEUE_MUST
2099 && !ioc_batching(q, ioc)) {
2100 /*
2101 * The queue is full and the allocating
2102 * process is not a "batcher", and not
2103 * exempted by the IO scheduler
2104 */
2105 goto out;
2106 }
2107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002109 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
2111
Jens Axboe082cf692005-06-28 16:35:11 +02002112 /*
2113 * Only allow batching queuers to allocate up to 50% over the defined
2114 * limit of requests, otherwise we could have thousands of requests
2115 * allocated with any setting of ->nr_requests
2116 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002117 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002118 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002119
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 rl->count[rw]++;
2121 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002122
Jens Axboe64521d12005-10-28 08:30:39 +02002123 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002124 if (priv)
2125 rl->elvpriv++;
2126
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 spin_unlock_irq(q->queue_lock);
2128
Jens Axboe7749a8d2006-12-13 13:02:26 +01002129 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002130 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 /*
2132 * Allocation failed presumably due to memory. Undo anything
2133 * we might have messed up.
2134 *
2135 * Allocating task should really be put onto the front of the
2136 * wait queue, but this is pretty rare.
2137 */
2138 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002139 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 /*
2142 * in the very unlikely event that allocation failed and no
2143 * requests for this direction was pending, mark us starved
2144 * so that freeing of a request in the other direction will
2145 * notice us. another possible fix would be to split the
2146 * rq mempool into READ and WRITE
2147 */
2148rq_starved:
2149 if (unlikely(rl->count[rw] == 0))
2150 rl->starved[rw] = 1;
2151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 goto out;
2153 }
2154
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002155 /*
2156 * ioc may be NULL here, and ioc_batching will be false. That's
2157 * OK, if the queue is under the request limit then requests need
2158 * not count toward the nr_batch_requests limit. There will always
2159 * be some limit enforced by BLK_BATCH_TIME.
2160 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 if (ioc_batching(q, ioc))
2162 ioc->nr_batch_requests--;
2163
2164 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002165
2166 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 return rq;
2169}
2170
2171/*
2172 * No available requests for this queue, unplug the device and wait for some
2173 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002174 *
2175 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 */
Jens Axboe165125e2007-07-24 09:28:11 +02002177static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002178 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002180 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 struct request *rq;
2182
Jens Axboe7749a8d2006-12-13 13:02:26 +01002183 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002184 while (!rq) {
2185 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 struct request_list *rl = &q->rq;
2187
2188 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2189 TASK_UNINTERRUPTIBLE);
2190
Jens Axboe7749a8d2006-12-13 13:02:26 +01002191 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 if (!rq) {
2194 struct io_context *ioc;
2195
Jens Axboe2056a782006-03-23 20:00:26 +01002196 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2197
Nick Piggind6344532005-06-28 20:45:14 -07002198 __generic_unplug_device(q);
2199 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 io_schedule();
2201
2202 /*
2203 * After sleeping, we become a "batching" process and
2204 * will be able to allocate at least one request, and
2205 * up to a big batch of them for a small period time.
2206 * See ioc_batching, ioc_set_batching
2207 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002208 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002210
2211 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 }
2213 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216 return rq;
2217}
2218
Jens Axboe165125e2007-07-24 09:28:11 +02002219struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220{
2221 struct request *rq;
2222
2223 BUG_ON(rw != READ && rw != WRITE);
2224
Nick Piggind6344532005-06-28 20:45:14 -07002225 spin_lock_irq(q->queue_lock);
2226 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002227 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002228 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002229 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002230 if (!rq)
2231 spin_unlock_irq(q->queue_lock);
2232 }
2233 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
2235 return rq;
2236}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237EXPORT_SYMBOL(blk_get_request);
2238
2239/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002240 * blk_start_queueing - initiate dispatch of requests to device
2241 * @q: request queue to kick into gear
2242 *
2243 * This is basically a helper to remove the need to know whether a queue
2244 * is plugged or not if someone just wants to initiate dispatch of requests
2245 * for this queue.
2246 *
2247 * The queue lock must be held with interrupts disabled.
2248 */
Jens Axboe165125e2007-07-24 09:28:11 +02002249void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002250{
2251 if (!blk_queue_plugged(q))
2252 q->request_fn(q);
2253 else
2254 __generic_unplug_device(q);
2255}
2256EXPORT_SYMBOL(blk_start_queueing);
2257
2258/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 * blk_requeue_request - put a request back on queue
2260 * @q: request queue where request should be inserted
2261 * @rq: request to be inserted
2262 *
2263 * Description:
2264 * Drivers often keep queueing requests until the hardware cannot accept
2265 * more, when that condition happens we need to put the request back
2266 * on the queue. Must be called with queue lock held.
2267 */
Jens Axboe165125e2007-07-24 09:28:11 +02002268void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269{
Jens Axboe2056a782006-03-23 20:00:26 +01002270 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (blk_rq_tagged(rq))
2273 blk_queue_end_tag(q, rq);
2274
2275 elv_requeue_request(q, rq);
2276}
2277
2278EXPORT_SYMBOL(blk_requeue_request);
2279
2280/**
2281 * blk_insert_request - insert a special request in to a request queue
2282 * @q: request queue where request should be inserted
2283 * @rq: request to be inserted
2284 * @at_head: insert request at head or tail of queue
2285 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 *
2287 * Description:
2288 * Many block devices need to execute commands asynchronously, so they don't
2289 * block the whole kernel from preemption during request execution. This is
2290 * accomplished normally by inserting aritficial requests tagged as
2291 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2292 * scheduled for actual execution by the request queue.
2293 *
2294 * We have the option of inserting the head or the tail of the queue.
2295 * Typically we use the tail for new ioctls and so forth. We use the head
2296 * of the queue for things like a QUEUE_FULL message from a device, or a
2297 * host that is unable to accept a particular command.
2298 */
Jens Axboe165125e2007-07-24 09:28:11 +02002299void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002300 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
Tejun Heo 867d1192005-04-24 02:06:05 -05002302 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 unsigned long flags;
2304
2305 /*
2306 * tell I/O scheduler that this isn't a regular read/write (ie it
2307 * must not attempt merges on this) and that it acts as a soft
2308 * barrier
2309 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002310 rq->cmd_type = REQ_TYPE_SPECIAL;
2311 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
2313 rq->special = data;
2314
2315 spin_lock_irqsave(q->queue_lock, flags);
2316
2317 /*
2318 * If command is tagged, release the tag
2319 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002320 if (blk_rq_tagged(rq))
2321 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
Tejun Heo 867d1192005-04-24 02:06:05 -05002323 drive_stat_acct(rq, rq->nr_sectors, 1);
2324 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002325 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 spin_unlock_irqrestore(q->queue_lock, flags);
2327}
2328
2329EXPORT_SYMBOL(blk_insert_request);
2330
Mike Christie0e75f902006-12-01 10:40:55 +01002331static int __blk_rq_unmap_user(struct bio *bio)
2332{
2333 int ret = 0;
2334
2335 if (bio) {
2336 if (bio_flagged(bio, BIO_USER_MAPPED))
2337 bio_unmap_user(bio);
2338 else
2339 ret = bio_uncopy_user(bio);
2340 }
2341
2342 return ret;
2343}
2344
NeilBrown3001ca72007-08-16 13:31:27 +02002345int blk_rq_append_bio(struct request_queue *q, struct request *rq,
2346 struct bio *bio)
2347{
2348 if (!rq->bio)
2349 blk_rq_bio_prep(q, rq, bio);
2350 else if (!ll_back_merge_fn(q, rq, bio))
2351 return -EINVAL;
2352 else {
2353 rq->biotail->bi_next = bio;
2354 rq->biotail = bio;
2355
2356 rq->data_len += bio->bi_size;
2357 }
2358 return 0;
2359}
2360EXPORT_SYMBOL(blk_rq_append_bio);
2361
Jens Axboe165125e2007-07-24 09:28:11 +02002362static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002363 void __user *ubuf, unsigned int len)
2364{
2365 unsigned long uaddr;
2366 struct bio *bio, *orig_bio;
2367 int reading, ret;
2368
2369 reading = rq_data_dir(rq) == READ;
2370
2371 /*
2372 * if alignment requirement is satisfied, map in user pages for
2373 * direct dma. else, set up kernel bounce buffers
2374 */
2375 uaddr = (unsigned long) ubuf;
2376 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2377 bio = bio_map_user(q, NULL, uaddr, len, reading);
2378 else
2379 bio = bio_copy_user(q, uaddr, len, reading);
2380
Jens Axboe29852592006-12-19 08:27:31 +01002381 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002382 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002383
2384 orig_bio = bio;
2385 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002386
Mike Christie0e75f902006-12-01 10:40:55 +01002387 /*
2388 * We link the bounce buffer in and could have to traverse it
2389 * later so we have to get a ref to prevent it from being freed
2390 */
2391 bio_get(bio);
2392
NeilBrown3001ca72007-08-16 13:31:27 +02002393 ret = blk_rq_append_bio(q, rq, bio);
2394 if (!ret)
2395 return bio->bi_size;
Mike Christie0e75f902006-12-01 10:40:55 +01002396
Mike Christie0e75f902006-12-01 10:40:55 +01002397 /* if it was boucned we must call the end io function */
NeilBrown6712ecf2007-09-27 12:47:43 +02002398 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002399 __blk_rq_unmap_user(orig_bio);
2400 bio_put(bio);
2401 return ret;
2402}
2403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404/**
2405 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2406 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002407 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 * @ubuf: the user buffer
2409 * @len: length of user data
2410 *
2411 * Description:
2412 * Data will be mapped directly for zero copy io, if possible. Otherwise
2413 * a kernel bounce buffer is used.
2414 *
2415 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2416 * still in process context.
2417 *
2418 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2419 * before being submitted to the device, as pages mapped may be out of
2420 * reach. It's the callers responsibility to make sure this happens. The
2421 * original bio must be passed back in to blk_rq_unmap_user() for proper
2422 * unmapping.
2423 */
Jens Axboe165125e2007-07-24 09:28:11 +02002424int blk_rq_map_user(struct request_queue *q, struct request *rq,
2425 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426{
Mike Christie0e75f902006-12-01 10:40:55 +01002427 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002428 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002429 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Mike Christiedefd94b2005-12-05 02:37:06 -06002431 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002432 return -EINVAL;
2433 if (!len || !ubuf)
2434 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435
Mike Christie0e75f902006-12-01 10:40:55 +01002436 while (bytes_read != len) {
2437 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Mike Christie0e75f902006-12-01 10:40:55 +01002439 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2440 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2441 >> PAGE_SHIFT;
2442 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
Mike Christie0e75f902006-12-01 10:40:55 +01002444 /*
2445 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2446 * pages. If this happens we just lower the requested
2447 * mapping len by a page so that we can fit
2448 */
2449 if (end - start > BIO_MAX_PAGES)
2450 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
Mike Christie0e75f902006-12-01 10:40:55 +01002452 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2453 if (ret < 0)
2454 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002455 if (!bio)
2456 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002457 bytes_read += ret;
2458 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 }
2460
Mike Christie0e75f902006-12-01 10:40:55 +01002461 rq->buffer = rq->data = NULL;
2462 return 0;
2463unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002464 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002465 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466}
2467
2468EXPORT_SYMBOL(blk_rq_map_user);
2469
2470/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002471 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2472 * @q: request queue where request should be inserted
2473 * @rq: request to map data to
2474 * @iov: pointer to the iovec
2475 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002476 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002477 *
2478 * Description:
2479 * Data will be mapped directly for zero copy io, if possible. Otherwise
2480 * a kernel bounce buffer is used.
2481 *
2482 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2483 * still in process context.
2484 *
2485 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2486 * before being submitted to the device, as pages mapped may be out of
2487 * reach. It's the callers responsibility to make sure this happens. The
2488 * original bio must be passed back in to blk_rq_unmap_user() for proper
2489 * unmapping.
2490 */
Jens Axboe165125e2007-07-24 09:28:11 +02002491int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002492 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002493{
2494 struct bio *bio;
2495
2496 if (!iov || iov_count <= 0)
2497 return -EINVAL;
2498
2499 /* we don't allow misaligned data like bio_map_user() does. If the
2500 * user is using sg, they're expected to know the alignment constraints
2501 * and respect them accordingly */
2502 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2503 if (IS_ERR(bio))
2504 return PTR_ERR(bio);
2505
Mike Christie0e75f902006-12-01 10:40:55 +01002506 if (bio->bi_size != len) {
NeilBrown6712ecf2007-09-27 12:47:43 +02002507 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002508 bio_unmap_user(bio);
2509 return -EINVAL;
2510 }
2511
2512 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002513 blk_rq_bio_prep(q, rq, bio);
2514 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002515 return 0;
2516}
2517
2518EXPORT_SYMBOL(blk_rq_map_user_iov);
2519
2520/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002522 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 *
2524 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002525 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2526 * supply the original rq->bio from the blk_rq_map_user() return, since
2527 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002529int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002531 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002532 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002534 while (bio) {
2535 mapped_bio = bio;
2536 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002537 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Jens Axboe48785bb2006-12-19 11:07:59 +01002539 ret2 = __blk_rq_unmap_user(mapped_bio);
2540 if (ret2 && !ret)
2541 ret = ret2;
2542
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002543 mapped_bio = bio;
2544 bio = bio->bi_next;
2545 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002546 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002547
2548 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549}
2550
2551EXPORT_SYMBOL(blk_rq_unmap_user);
2552
2553/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002554 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2555 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002556 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002557 * @kbuf: the kernel buffer
2558 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002559 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002560 */
Jens Axboe165125e2007-07-24 09:28:11 +02002561int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002562 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002563{
Mike Christie df46b9a2005-06-20 14:04:44 +02002564 struct bio *bio;
2565
Mike Christiedefd94b2005-12-05 02:37:06 -06002566 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002567 return -EINVAL;
2568 if (!len || !kbuf)
2569 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002570
2571 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002572 if (IS_ERR(bio))
2573 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002574
Jens Axboedd1cab92005-06-20 14:06:01 +02002575 if (rq_data_dir(rq) == WRITE)
2576 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002577
Jens Axboedd1cab92005-06-20 14:06:01 +02002578 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002579 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002580 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002581 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002582}
2583
2584EXPORT_SYMBOL(blk_rq_map_kern);
2585
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002586/**
2587 * blk_execute_rq_nowait - insert a request into queue for execution
2588 * @q: queue to insert the request in
2589 * @bd_disk: matching gendisk
2590 * @rq: request to insert
2591 * @at_head: insert request at head or tail of queue
2592 * @done: I/O completion handler
2593 *
2594 * Description:
2595 * Insert a fully prepared request at the back of the io scheduler queue
2596 * for execution. Don't wait for completion.
2597 */
Jens Axboe165125e2007-07-24 09:28:11 +02002598void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002599 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002600 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002601{
2602 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2603
2604 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002605 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002606 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002607 WARN_ON(irqs_disabled());
2608 spin_lock_irq(q->queue_lock);
2609 __elv_add_request(q, rq, where, 1);
2610 __generic_unplug_device(q);
2611 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002612}
Mike Christie6e39b69e2005-11-11 05:30:24 -06002613EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2614
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615/**
2616 * blk_execute_rq - insert a request into queue for execution
2617 * @q: queue to insert the request in
2618 * @bd_disk: matching gendisk
2619 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002620 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 *
2622 * Description:
2623 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002624 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 */
Jens Axboe165125e2007-07-24 09:28:11 +02002626int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002627 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002629 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 char sense[SCSI_SENSE_BUFFERSIZE];
2631 int err = 0;
2632
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 /*
2634 * we need an extra reference to the request, so we can look at
2635 * it after io completion
2636 */
2637 rq->ref_count++;
2638
2639 if (!rq->sense) {
2640 memset(sense, 0, sizeof(sense));
2641 rq->sense = sense;
2642 rq->sense_len = 0;
2643 }
2644
Jens Axboec00895a2006-09-30 20:29:12 +02002645 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002646 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648
2649 if (rq->errors)
2650 err = -EIO;
2651
2652 return err;
2653}
2654
2655EXPORT_SYMBOL(blk_execute_rq);
2656
Jens Axboefd5d8062007-10-16 11:05:02 +02002657static void bio_end_empty_barrier(struct bio *bio, int err)
2658{
2659 if (err)
2660 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2661
2662 complete(bio->bi_private);
2663}
2664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665/**
2666 * blkdev_issue_flush - queue a flush
2667 * @bdev: blockdev to issue flush for
2668 * @error_sector: error sector
2669 *
2670 * Description:
2671 * Issue a flush for the block device in question. Caller can supply
2672 * room for storing the error offset in case of a flush error, if they
2673 * wish to. Caller must run wait_for_completion() on its own.
2674 */
2675int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2676{
Jens Axboefd5d8062007-10-16 11:05:02 +02002677 DECLARE_COMPLETION_ONSTACK(wait);
Jens Axboe165125e2007-07-24 09:28:11 +02002678 struct request_queue *q;
Jens Axboefd5d8062007-10-16 11:05:02 +02002679 struct bio *bio;
2680 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
2682 if (bdev->bd_disk == NULL)
2683 return -ENXIO;
2684
2685 q = bdev_get_queue(bdev);
2686 if (!q)
2687 return -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
Jens Axboefd5d8062007-10-16 11:05:02 +02002689 bio = bio_alloc(GFP_KERNEL, 0);
2690 if (!bio)
2691 return -ENOMEM;
2692
2693 bio->bi_end_io = bio_end_empty_barrier;
2694 bio->bi_private = &wait;
2695 bio->bi_bdev = bdev;
2696 submit_bio(1 << BIO_RW_BARRIER, bio);
2697
2698 wait_for_completion(&wait);
2699
2700 /*
2701 * The driver must store the error location in ->bi_sector, if
2702 * it supports it. For non-stacked drivers, this should be copied
2703 * from rq->sector.
2704 */
2705 if (error_sector)
2706 *error_sector = bio->bi_sector;
2707
2708 ret = 0;
2709 if (!bio_flagged(bio, BIO_UPTODATE))
2710 ret = -EIO;
2711
2712 bio_put(bio);
2713 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714}
2715
2716EXPORT_SYMBOL(blkdev_issue_flush);
2717
Adrian Bunk93d17d32005-06-25 14:59:10 -07002718static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719{
2720 int rw = rq_data_dir(rq);
2721
2722 if (!blk_fs_request(rq) || !rq->rq_disk)
2723 return;
2724
Jens Axboed72d9042005-11-01 08:35:42 +01002725 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002726 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002727 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 disk_round_stats(rq->rq_disk);
2729 rq->rq_disk->in_flight++;
2730 }
2731}
2732
2733/*
2734 * add-request adds a request to the linked list.
2735 * queue lock is held and interrupts disabled, as we muck with the
2736 * request queue list.
2737 */
Jens Axboe165125e2007-07-24 09:28:11 +02002738static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739{
2740 drive_stat_acct(req, req->nr_sectors, 1);
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 /*
2743 * elevator indicated where it wants this request to be
2744 * inserted at elevator_merge time
2745 */
2746 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2747}
2748
2749/*
2750 * disk_round_stats() - Round off the performance stats on a struct
2751 * disk_stats.
2752 *
2753 * The average IO queue length and utilisation statistics are maintained
2754 * by observing the current state of the queue length and the amount of
2755 * time it has been in this state for.
2756 *
2757 * Normally, that accounting is done on IO completion, but that can result
2758 * in more than a second's worth of IO being accounted for within any one
2759 * second, leading to >100% utilisation. To deal with that, we call this
2760 * function to do a round-off before returning the results when reading
2761 * /proc/diskstats. This accounts immediately for all queue usage up to
2762 * the current jiffies and restarts the counters again.
2763 */
2764void disk_round_stats(struct gendisk *disk)
2765{
2766 unsigned long now = jiffies;
2767
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002768 if (now == disk->stamp)
2769 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002771 if (disk->in_flight) {
2772 __disk_stat_add(disk, time_in_queue,
2773 disk->in_flight * (now - disk->stamp));
2774 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777}
2778
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002779EXPORT_SYMBOL_GPL(disk_round_stats);
2780
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781/*
2782 * queue lock must be held
2783 */
Jens Axboe165125e2007-07-24 09:28:11 +02002784void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (unlikely(!q))
2787 return;
2788 if (unlikely(--req->ref_count))
2789 return;
2790
Tejun Heo8922e162005-10-20 16:23:44 +02002791 elv_completed_request(q, req);
2792
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 /*
2794 * Request may not have originated from ll_rw_blk. if not,
2795 * it didn't come out of our reserved rq pools
2796 */
Jens Axboe49171e52006-08-10 08:59:11 +02002797 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002799 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002802 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
2804 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002805 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 }
2807}
2808
Mike Christie6e39b69e2005-11-11 05:30:24 -06002809EXPORT_SYMBOL_GPL(__blk_put_request);
2810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811void blk_put_request(struct request *req)
2812{
Tejun Heo8922e162005-10-20 16:23:44 +02002813 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002814 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
Tejun Heo8922e162005-10-20 16:23:44 +02002816 /*
2817 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2818 * following if (q) test.
2819 */
2820 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 spin_lock_irqsave(q->queue_lock, flags);
2822 __blk_put_request(q, req);
2823 spin_unlock_irqrestore(q->queue_lock, flags);
2824 }
2825}
2826
2827EXPORT_SYMBOL(blk_put_request);
2828
2829/**
2830 * blk_end_sync_rq - executes a completion event on a request
2831 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002832 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002834void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835{
Jens Axboec00895a2006-09-30 20:29:12 +02002836 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Jens Axboec00895a2006-09-30 20:29:12 +02002838 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 __blk_put_request(rq->q, rq);
2840
2841 /*
2842 * complete last, if this is a stack request the process (and thus
2843 * the rq pointer) could be invalid right after this complete()
2844 */
2845 complete(waiting);
2846}
2847EXPORT_SYMBOL(blk_end_sync_rq);
2848
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849/*
2850 * Has to be called with the request spinlock acquired
2851 */
Jens Axboe165125e2007-07-24 09:28:11 +02002852static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 struct request *next)
2854{
2855 if (!rq_mergeable(req) || !rq_mergeable(next))
2856 return 0;
2857
2858 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002859 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 */
2861 if (req->sector + req->nr_sectors != next->sector)
2862 return 0;
2863
2864 if (rq_data_dir(req) != rq_data_dir(next)
2865 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002866 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 return 0;
2868
2869 /*
2870 * If we are allowed to merge, then append bio list
2871 * from next to rq and release next. merge_requests_fn
2872 * will have updated segment counts, update sector
2873 * counts here.
2874 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002875 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 return 0;
2877
2878 /*
2879 * At this point we have either done a back merge
2880 * or front merge. We need the smaller start_time of
2881 * the merged requests to be the current request
2882 * for accounting purposes.
2883 */
2884 if (time_after(req->start_time, next->start_time))
2885 req->start_time = next->start_time;
2886
2887 req->biotail->bi_next = next->bio;
2888 req->biotail = next->biotail;
2889
2890 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2891
2892 elv_merge_requests(q, req, next);
2893
2894 if (req->rq_disk) {
2895 disk_round_stats(req->rq_disk);
2896 req->rq_disk->in_flight--;
2897 }
2898
Jens Axboe22e2c502005-06-27 10:55:12 +02002899 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2900
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 __blk_put_request(q, next);
2902 return 1;
2903}
2904
Jens Axboe165125e2007-07-24 09:28:11 +02002905static inline int attempt_back_merge(struct request_queue *q,
2906 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907{
2908 struct request *next = elv_latter_request(q, rq);
2909
2910 if (next)
2911 return attempt_merge(q, rq, next);
2912
2913 return 0;
2914}
2915
Jens Axboe165125e2007-07-24 09:28:11 +02002916static inline int attempt_front_merge(struct request_queue *q,
2917 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918{
2919 struct request *prev = elv_former_request(q, rq);
2920
2921 if (prev)
2922 return attempt_merge(q, prev, rq);
2923
2924 return 0;
2925}
2926
Tejun Heo52d9e672006-01-06 09:49:58 +01002927static void init_request_from_bio(struct request *req, struct bio *bio)
2928{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002929 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002930
2931 /*
2932 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2933 */
2934 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002935 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002936
2937 /*
2938 * REQ_BARRIER implies no merging, but lets make it explicit
2939 */
2940 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002941 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002942
Jens Axboeb31dc662006-06-13 08:26:10 +02002943 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002944 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002945 if (bio_rw_meta(bio))
2946 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002947
Tejun Heo52d9e672006-01-06 09:49:58 +01002948 req->errors = 0;
2949 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01002950 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002951 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02002952 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002953}
2954
Jens Axboe165125e2007-07-24 09:28:11 +02002955static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956{
Nick Piggin450991b2005-06-28 20:45:13 -07002957 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002958 int el_ret, nr_sectors, barrier, err;
2959 const unsigned short prio = bio_prio(bio);
2960 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002961 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
2965 /*
2966 * low level driver can indicate that it wants pages above a
2967 * certain limit bounced to low memory (ie for highmem, or even
2968 * ISA dma in theory)
2969 */
2970 blk_queue_bounce(q, &bio);
2971
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002973 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 err = -EOPNOTSUPP;
2975 goto end_io;
2976 }
2977
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 spin_lock_irq(q->queue_lock);
2979
Nick Piggin450991b2005-06-28 20:45:13 -07002980 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 goto get_rq;
2982
2983 el_ret = elv_merge(q, &req, bio);
2984 switch (el_ret) {
2985 case ELEVATOR_BACK_MERGE:
2986 BUG_ON(!rq_mergeable(req));
2987
Jens Axboe1aa4f242006-12-19 08:33:11 +01002988 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 break;
2990
Jens Axboe2056a782006-03-23 20:00:26 +01002991 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2992
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 req->biotail->bi_next = bio;
2994 req->biotail = bio;
2995 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002996 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 drive_stat_acct(req, nr_sectors, 0);
2998 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002999 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 goto out;
3001
3002 case ELEVATOR_FRONT_MERGE:
3003 BUG_ON(!rq_mergeable(req));
3004
Jens Axboe1aa4f242006-12-19 08:33:11 +01003005 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 break;
3007
Jens Axboe2056a782006-03-23 20:00:26 +01003008 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
3009
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 bio->bi_next = req->bio;
3011 req->bio = bio;
3012
3013 /*
3014 * may not be valid. if the low level driver said
3015 * it didn't need a bounce buffer then it better
3016 * not touch req->buffer either...
3017 */
3018 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02003019 req->current_nr_sectors = bio_cur_sectors(bio);
3020 req->hard_cur_sectors = req->current_nr_sectors;
3021 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02003023 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 drive_stat_acct(req, nr_sectors, 0);
3025 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02003026 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 goto out;
3028
Nick Piggin450991b2005-06-28 20:45:13 -07003029 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003031 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 }
3033
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003035 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003036 * This sync check and mask will be re-done in init_request_from_bio(),
3037 * but we need to set it earlier to expose the sync flag to the
3038 * rq allocator and io schedulers.
3039 */
3040 rw_flags = bio_data_dir(bio);
3041 if (sync)
3042 rw_flags |= REQ_RW_SYNC;
3043
3044 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003045 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003046 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003047 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003048 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003049
Nick Piggin450991b2005-06-28 20:45:13 -07003050 /*
3051 * After dropping the lock and possibly sleeping here, our request
3052 * may now be mergeable after it had proven unmergeable (above).
3053 * We don't worry about that case for efficiency. It won't happen
3054 * often, and the elevators are able to handle it.
3055 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003056 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Nick Piggin450991b2005-06-28 20:45:13 -07003058 spin_lock_irq(q->queue_lock);
3059 if (elv_queue_empty(q))
3060 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 add_request(q, req);
3062out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003063 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 __generic_unplug_device(q);
3065
3066 spin_unlock_irq(q->queue_lock);
3067 return 0;
3068
3069end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003070 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 return 0;
3072}
3073
3074/*
3075 * If bio->bi_dev is a partition, remap the location
3076 */
3077static inline void blk_partition_remap(struct bio *bio)
3078{
3079 struct block_device *bdev = bio->bi_bdev;
3080
Jens Axboebf2de6f2007-09-27 13:01:25 +02003081 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003083 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084
Jens Axboea3623572005-11-01 09:26:16 +01003085 p->sectors[rw] += bio_sectors(bio);
3086 p->ios[rw]++;
3087
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 bio->bi_sector += p->start_sect;
3089 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003090
3091 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3092 bdev->bd_dev, bio->bi_sector,
3093 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 }
3095}
3096
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097static void handle_bad_sector(struct bio *bio)
3098{
3099 char b[BDEVNAME_SIZE];
3100
3101 printk(KERN_INFO "attempt to access beyond end of device\n");
3102 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3103 bdevname(bio->bi_bdev, b),
3104 bio->bi_rw,
3105 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3106 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3107
3108 set_bit(BIO_EOF, &bio->bi_flags);
3109}
3110
Akinobu Mitac17bb492006-12-08 02:39:46 -08003111#ifdef CONFIG_FAIL_MAKE_REQUEST
3112
3113static DECLARE_FAULT_ATTR(fail_make_request);
3114
3115static int __init setup_fail_make_request(char *str)
3116{
3117 return setup_fault_attr(&fail_make_request, str);
3118}
3119__setup("fail_make_request=", setup_fail_make_request);
3120
3121static int should_fail_request(struct bio *bio)
3122{
3123 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3124 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3125 return should_fail(&fail_make_request, bio->bi_size);
3126
3127 return 0;
3128}
3129
3130static int __init fail_make_request_debugfs(void)
3131{
3132 return init_fault_attr_dentries(&fail_make_request,
3133 "fail_make_request");
3134}
3135
3136late_initcall(fail_make_request_debugfs);
3137
3138#else /* CONFIG_FAIL_MAKE_REQUEST */
3139
3140static inline int should_fail_request(struct bio *bio)
3141{
3142 return 0;
3143}
3144
3145#endif /* CONFIG_FAIL_MAKE_REQUEST */
3146
Jens Axboec07e2b42007-07-18 13:27:58 +02003147/*
3148 * Check whether this bio extends beyond the end of the device.
3149 */
3150static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
3151{
3152 sector_t maxsector;
3153
3154 if (!nr_sectors)
3155 return 0;
3156
3157 /* Test device or partition size, when known. */
3158 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3159 if (maxsector) {
3160 sector_t sector = bio->bi_sector;
3161
3162 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3163 /*
3164 * This may well happen - the kernel calls bread()
3165 * without checking the size of the device, e.g., when
3166 * mounting a device.
3167 */
3168 handle_bad_sector(bio);
3169 return 1;
3170 }
3171 }
3172
3173 return 0;
3174}
3175
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176/**
3177 * generic_make_request: hand a buffer to its device driver for I/O
3178 * @bio: The bio describing the location in memory and on the device.
3179 *
3180 * generic_make_request() is used to make I/O requests of block
3181 * devices. It is passed a &struct bio, which describes the I/O that needs
3182 * to be done.
3183 *
3184 * generic_make_request() does not return any status. The
3185 * success/failure status of the request, along with notification of
3186 * completion, is delivered asynchronously through the bio->bi_end_io
3187 * function described (one day) else where.
3188 *
3189 * The caller of generic_make_request must make sure that bi_io_vec
3190 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3191 * set to describe the device address, and the
3192 * bi_end_io and optionally bi_private are set to describe how
3193 * completion notification should be signaled.
3194 *
3195 * generic_make_request and the drivers it calls may use bi_next if this
3196 * bio happens to be merged with someone else, and may change bi_dev and
3197 * bi_sector for remaps as it sees fit. So the values of these fields
3198 * should NOT be depended on after the call to generic_make_request.
3199 */
Neil Brownd89d8792007-05-01 09:53:42 +02003200static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201{
Jens Axboe165125e2007-07-24 09:28:11 +02003202 struct request_queue *q;
NeilBrown5ddfe962006-10-30 22:07:21 -08003203 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003205 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
3207 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208
Jens Axboec07e2b42007-07-18 13:27:58 +02003209 if (bio_check_eod(bio, nr_sectors))
3210 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211
3212 /*
3213 * Resolve the mapping until finished. (drivers are
3214 * still free to implement/resolve their own stacking
3215 * by explicitly returning 0)
3216 *
3217 * NOTE: we don't repeat the blk_size check for each new device.
3218 * Stacking drivers are expected to know what they are doing.
3219 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003220 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003221 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 do {
3223 char b[BDEVNAME_SIZE];
3224
3225 q = bdev_get_queue(bio->bi_bdev);
3226 if (!q) {
3227 printk(KERN_ERR
3228 "generic_make_request: Trying to access "
3229 "nonexistent block-device %s (%Lu)\n",
3230 bdevname(bio->bi_bdev, b),
3231 (long long) bio->bi_sector);
3232end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003233 bio_endio(bio, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 break;
3235 }
3236
Jens Axboe4fa253f2007-07-18 13:13:10 +02003237 if (unlikely(nr_sectors > q->max_hw_sectors)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 printk("bio too big device %s (%u > %u)\n",
3239 bdevname(bio->bi_bdev, b),
3240 bio_sectors(bio),
3241 q->max_hw_sectors);
3242 goto end_io;
3243 }
3244
Nick Pigginfde6ad22005-06-23 00:08:53 -07003245 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 goto end_io;
3247
Akinobu Mitac17bb492006-12-08 02:39:46 -08003248 if (should_fail_request(bio))
3249 goto end_io;
3250
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 /*
3252 * If this device has partitions, remap block n
3253 * of partition p to block n+start(p) of the disk.
3254 */
3255 blk_partition_remap(bio);
3256
NeilBrown5ddfe962006-10-30 22:07:21 -08003257 if (old_sector != -1)
Jens Axboe4fa253f2007-07-18 13:13:10 +02003258 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003259 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003260
3261 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3262
NeilBrown5ddfe962006-10-30 22:07:21 -08003263 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003264 old_dev = bio->bi_bdev->bd_dev;
3265
Jens Axboec07e2b42007-07-18 13:27:58 +02003266 if (bio_check_eod(bio, nr_sectors))
3267 goto end_io;
NeilBrown5ddfe962006-10-30 22:07:21 -08003268
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 ret = q->make_request_fn(q, bio);
3270 } while (ret);
3271}
3272
Neil Brownd89d8792007-05-01 09:53:42 +02003273/*
3274 * We only want one ->make_request_fn to be active at a time,
3275 * else stack usage with stacked devices could be a problem.
3276 * So use current->bio_{list,tail} to keep a list of requests
3277 * submited by a make_request_fn function.
3278 * current->bio_tail is also used as a flag to say if
3279 * generic_make_request is currently active in this task or not.
3280 * If it is NULL, then no make_request is active. If it is non-NULL,
3281 * then a make_request is active, and new requests should be added
3282 * at the tail
3283 */
3284void generic_make_request(struct bio *bio)
3285{
3286 if (current->bio_tail) {
3287 /* make_request is active */
3288 *(current->bio_tail) = bio;
3289 bio->bi_next = NULL;
3290 current->bio_tail = &bio->bi_next;
3291 return;
3292 }
3293 /* following loop may be a bit non-obvious, and so deserves some
3294 * explanation.
3295 * Before entering the loop, bio->bi_next is NULL (as all callers
3296 * ensure that) so we have a list with a single bio.
3297 * We pretend that we have just taken it off a longer list, so
3298 * we assign bio_list to the next (which is NULL) and bio_tail
3299 * to &bio_list, thus initialising the bio_list of new bios to be
3300 * added. __generic_make_request may indeed add some more bios
3301 * through a recursive call to generic_make_request. If it
3302 * did, we find a non-NULL value in bio_list and re-enter the loop
3303 * from the top. In this case we really did just take the bio
3304 * of the top of the list (no pretending) and so fixup bio_list and
3305 * bio_tail or bi_next, and call into __generic_make_request again.
3306 *
3307 * The loop was structured like this to make only one call to
3308 * __generic_make_request (which is important as it is large and
3309 * inlined) and to keep the structure simple.
3310 */
3311 BUG_ON(bio->bi_next);
3312 do {
3313 current->bio_list = bio->bi_next;
3314 if (bio->bi_next == NULL)
3315 current->bio_tail = &current->bio_list;
3316 else
3317 bio->bi_next = NULL;
3318 __generic_make_request(bio);
3319 bio = current->bio_list;
3320 } while (bio);
3321 current->bio_tail = NULL; /* deactivate */
3322}
3323
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324EXPORT_SYMBOL(generic_make_request);
3325
3326/**
3327 * submit_bio: submit a bio to the block device layer for I/O
3328 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3329 * @bio: The &struct bio which describes the I/O
3330 *
3331 * submit_bio() is very similar in purpose to generic_make_request(), and
3332 * uses that function to do most of the work. Both are fairly rough
3333 * interfaces, @bio must be presetup and ready for I/O.
3334 *
3335 */
3336void submit_bio(int rw, struct bio *bio)
3337{
3338 int count = bio_sectors(bio);
3339
Jens Axboe22e2c502005-06-27 10:55:12 +02003340 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341
Jens Axboebf2de6f2007-09-27 13:01:25 +02003342 /*
3343 * If it's a regular read/write or a barrier with data attached,
3344 * go through the normal accounting stuff before submission.
3345 */
3346 if (!bio_empty_barrier(bio)) {
3347
3348 BIO_BUG_ON(!bio->bi_size);
3349 BIO_BUG_ON(!bio->bi_io_vec);
3350
3351 if (rw & WRITE) {
3352 count_vm_events(PGPGOUT, count);
3353 } else {
3354 task_io_account_read(bio->bi_size);
3355 count_vm_events(PGPGIN, count);
3356 }
3357
3358 if (unlikely(block_dump)) {
3359 char b[BDEVNAME_SIZE];
3360 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3361 current->comm, current->pid,
3362 (rw & WRITE) ? "WRITE" : "READ",
3363 (unsigned long long)bio->bi_sector,
3364 bdevname(bio->bi_bdev,b));
3365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 }
3367
3368 generic_make_request(bio);
3369}
3370
3371EXPORT_SYMBOL(submit_bio);
3372
Adrian Bunk93d17d32005-06-25 14:59:10 -07003373static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374{
3375 if (blk_fs_request(rq)) {
3376 rq->hard_sector += nsect;
3377 rq->hard_nr_sectors -= nsect;
3378
3379 /*
3380 * Move the I/O submission pointers ahead if required.
3381 */
3382 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3383 (rq->sector <= rq->hard_sector)) {
3384 rq->sector = rq->hard_sector;
3385 rq->nr_sectors = rq->hard_nr_sectors;
3386 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3387 rq->current_nr_sectors = rq->hard_cur_sectors;
3388 rq->buffer = bio_data(rq->bio);
3389 }
3390
3391 /*
3392 * if total number of sectors is less than the first segment
3393 * size, something has gone terribly wrong
3394 */
3395 if (rq->nr_sectors < rq->current_nr_sectors) {
3396 printk("blk: request botched\n");
3397 rq->nr_sectors = rq->current_nr_sectors;
3398 }
3399 }
3400}
3401
3402static int __end_that_request_first(struct request *req, int uptodate,
3403 int nr_bytes)
3404{
3405 int total_bytes, bio_nbytes, error, next_idx = 0;
3406 struct bio *bio;
3407
Jens Axboe2056a782006-03-23 20:00:26 +01003408 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3409
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 /*
3411 * extend uptodate bool to allow < 0 value to be direct io error
3412 */
3413 error = 0;
3414 if (end_io_error(uptodate))
3415 error = !uptodate ? -EIO : uptodate;
3416
3417 /*
3418 * for a REQ_BLOCK_PC request, we want to carry any eventual
3419 * sense key with us all the way through
3420 */
3421 if (!blk_pc_request(req))
3422 req->errors = 0;
3423
3424 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003425 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 printk("end_request: I/O error, dev %s, sector %llu\n",
3427 req->rq_disk ? req->rq_disk->disk_name : "?",
3428 (unsigned long long)req->sector);
3429 }
3430
Jens Axboed72d9042005-11-01 08:35:42 +01003431 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003432 const int rw = rq_data_dir(req);
3433
Jens Axboe53e86062006-01-17 11:09:27 +01003434 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003435 }
3436
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 total_bytes = bio_nbytes = 0;
3438 while ((bio = req->bio) != NULL) {
3439 int nbytes;
3440
Jens Axboebf2de6f2007-09-27 13:01:25 +02003441 /*
3442 * For an empty barrier request, the low level driver must
3443 * store a potential error location in ->sector. We pass
3444 * that back up in ->bi_sector.
3445 */
3446 if (blk_empty_barrier(req))
3447 bio->bi_sector = req->sector;
3448
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 if (nr_bytes >= bio->bi_size) {
3450 req->bio = bio->bi_next;
3451 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02003452 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 next_idx = 0;
3454 bio_nbytes = 0;
3455 } else {
3456 int idx = bio->bi_idx + next_idx;
3457
3458 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3459 blk_dump_rq_flags(req, "__end_that");
3460 printk("%s: bio idx %d >= vcnt %d\n",
3461 __FUNCTION__,
3462 bio->bi_idx, bio->bi_vcnt);
3463 break;
3464 }
3465
3466 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3467 BIO_BUG_ON(nbytes > bio->bi_size);
3468
3469 /*
3470 * not a complete bvec done
3471 */
3472 if (unlikely(nbytes > nr_bytes)) {
3473 bio_nbytes += nr_bytes;
3474 total_bytes += nr_bytes;
3475 break;
3476 }
3477
3478 /*
3479 * advance to the next vector
3480 */
3481 next_idx++;
3482 bio_nbytes += nbytes;
3483 }
3484
3485 total_bytes += nbytes;
3486 nr_bytes -= nbytes;
3487
3488 if ((bio = req->bio)) {
3489 /*
3490 * end more in this run, or just return 'not-done'
3491 */
3492 if (unlikely(nr_bytes <= 0))
3493 break;
3494 }
3495 }
3496
3497 /*
3498 * completely done
3499 */
3500 if (!req->bio)
3501 return 0;
3502
3503 /*
3504 * if the request wasn't completed, update state
3505 */
3506 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02003507 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 bio->bi_idx += next_idx;
3509 bio_iovec(bio)->bv_offset += nr_bytes;
3510 bio_iovec(bio)->bv_len -= nr_bytes;
3511 }
3512
3513 blk_recalc_rq_sectors(req, total_bytes >> 9);
3514 blk_recalc_rq_segments(req);
3515 return 1;
3516}
3517
3518/**
3519 * end_that_request_first - end I/O on a request
3520 * @req: the request being processed
3521 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3522 * @nr_sectors: number of sectors to end I/O on
3523 *
3524 * Description:
3525 * Ends I/O on a number of sectors attached to @req, and sets it up
3526 * for the next range of segments (if any) in the cluster.
3527 *
3528 * Return:
3529 * 0 - we are done with this request, call end_that_request_last()
3530 * 1 - still buffers pending for this request
3531 **/
3532int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3533{
3534 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3535}
3536
3537EXPORT_SYMBOL(end_that_request_first);
3538
3539/**
3540 * end_that_request_chunk - end I/O on a request
3541 * @req: the request being processed
3542 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3543 * @nr_bytes: number of bytes to complete
3544 *
3545 * Description:
3546 * Ends I/O on a number of bytes attached to @req, and sets it up
3547 * for the next range of segments (if any). Like end_that_request_first(),
3548 * but deals with bytes instead of sectors.
3549 *
3550 * Return:
3551 * 0 - we are done with this request, call end_that_request_last()
3552 * 1 - still buffers pending for this request
3553 **/
3554int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3555{
3556 return __end_that_request_first(req, uptodate, nr_bytes);
3557}
3558
3559EXPORT_SYMBOL(end_that_request_chunk);
3560
3561/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003562 * splice the completion data to a local structure and hand off to
3563 * process_completion_queue() to complete the requests
3564 */
3565static void blk_done_softirq(struct softirq_action *h)
3566{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003567 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003568
3569 local_irq_disable();
3570 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003571 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003572 local_irq_enable();
3573
3574 while (!list_empty(&local_list)) {
3575 struct request *rq = list_entry(local_list.next, struct request, donelist);
3576
3577 list_del_init(&rq->donelist);
3578 rq->q->softirq_done_fn(rq);
3579 }
3580}
3581
Satyam Sharmadb47d472007-08-23 09:29:40 +02003582static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action,
Jens Axboeff856ba2006-01-09 16:02:34 +01003583 void *hcpu)
3584{
3585 /*
3586 * If a CPU goes away, splice its entries to the current CPU
3587 * and trigger a run of the softirq
3588 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003589 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003590 int cpu = (unsigned long) hcpu;
3591
3592 local_irq_disable();
3593 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3594 &__get_cpu_var(blk_cpu_done));
3595 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3596 local_irq_enable();
3597 }
3598
3599 return NOTIFY_OK;
3600}
3601
3602
Satyam Sharmadb47d472007-08-23 09:29:40 +02003603static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003604 .notifier_call = blk_cpu_notify,
3605};
3606
Jens Axboeff856ba2006-01-09 16:02:34 +01003607/**
3608 * blk_complete_request - end I/O on a request
3609 * @req: the request being processed
3610 *
3611 * Description:
3612 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003613 * unless the driver actually implements this in its completion callback
Jens Axboe4fa253f2007-07-18 13:13:10 +02003614 * through requeueing. The actual completion happens out-of-order,
Jens Axboeff856ba2006-01-09 16:02:34 +01003615 * through a softirq handler. The user must have registered a completion
3616 * callback through blk_queue_softirq_done().
3617 **/
3618
3619void blk_complete_request(struct request *req)
3620{
3621 struct list_head *cpu_list;
3622 unsigned long flags;
3623
3624 BUG_ON(!req->q->softirq_done_fn);
3625
3626 local_irq_save(flags);
3627
3628 cpu_list = &__get_cpu_var(blk_cpu_done);
3629 list_add_tail(&req->donelist, cpu_list);
3630 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3631
3632 local_irq_restore(flags);
3633}
3634
3635EXPORT_SYMBOL(blk_complete_request);
3636
3637/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 * queue lock must be held
3639 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003640void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641{
3642 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003643 int error;
3644
3645 /*
3646 * extend uptodate bool to allow < 0 value to be direct io error
3647 */
3648 error = 0;
3649 if (end_io_error(uptodate))
3650 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651
3652 if (unlikely(laptop_mode) && blk_fs_request(req))
3653 laptop_io_completion();
3654
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003655 /*
3656 * Account IO completion. bar_rq isn't accounted as a normal
3657 * IO on queueing nor completion. Accounting the containing
3658 * request is enough.
3659 */
3660 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003662 const int rw = rq_data_dir(req);
3663
3664 __disk_stat_inc(disk, ios[rw]);
3665 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 disk_round_stats(disk);
3667 disk->in_flight--;
3668 }
3669 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003670 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671 else
3672 __blk_put_request(req->q, req);
3673}
3674
3675EXPORT_SYMBOL(end_that_request_last);
3676
Jens Axboea0cd1282007-09-21 10:41:07 +02003677static inline void __end_request(struct request *rq, int uptodate,
3678 unsigned int nr_bytes, int dequeue)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679{
Jens Axboea0cd1282007-09-21 10:41:07 +02003680 if (!end_that_request_chunk(rq, uptodate, nr_bytes)) {
3681 if (dequeue)
3682 blkdev_dequeue_request(rq);
3683 add_disk_randomness(rq->rq_disk);
3684 end_that_request_last(rq, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 }
3686}
3687
Jens Axboea0cd1282007-09-21 10:41:07 +02003688static unsigned int rq_byte_size(struct request *rq)
3689{
3690 if (blk_fs_request(rq))
3691 return rq->hard_nr_sectors << 9;
3692
3693 return rq->data_len;
3694}
3695
3696/**
3697 * end_queued_request - end all I/O on a queued request
3698 * @rq: the request being processed
3699 * @uptodate: error value or 0/1 uptodate flag
3700 *
3701 * Description:
3702 * Ends all I/O on a request, and removes it from the block layer queues.
3703 * Not suitable for normal IO completion, unless the driver still has
3704 * the request attached to the block layer.
3705 *
3706 **/
3707void end_queued_request(struct request *rq, int uptodate)
3708{
3709 __end_request(rq, uptodate, rq_byte_size(rq), 1);
3710}
3711EXPORT_SYMBOL(end_queued_request);
3712
3713/**
3714 * end_dequeued_request - end all I/O on a dequeued request
3715 * @rq: the request being processed
3716 * @uptodate: error value or 0/1 uptodate flag
3717 *
3718 * Description:
3719 * Ends all I/O on a request. The request must already have been
3720 * dequeued using blkdev_dequeue_request(), as is normally the case
3721 * for most drivers.
3722 *
3723 **/
3724void end_dequeued_request(struct request *rq, int uptodate)
3725{
3726 __end_request(rq, uptodate, rq_byte_size(rq), 0);
3727}
3728EXPORT_SYMBOL(end_dequeued_request);
3729
3730
3731/**
3732 * end_request - end I/O on the current segment of the request
3733 * @rq: the request being processed
3734 * @uptodate: error value or 0/1 uptodate flag
3735 *
3736 * Description:
3737 * Ends I/O on the current segment of a request. If that is the only
3738 * remaining segment, the request is also completed and freed.
3739 *
3740 * This is a remnant of how older block drivers handled IO completions.
3741 * Modern drivers typically end IO on the full request in one go, unless
3742 * they have a residual value to account for. For that case this function
3743 * isn't really useful, unless the residual just happens to be the
3744 * full current segment. In other words, don't use this function in new
3745 * code. Either use end_request_completely(), or the
3746 * end_that_request_chunk() (along with end_that_request_last()) for
3747 * partial completions.
3748 *
3749 **/
3750void end_request(struct request *req, int uptodate)
3751{
3752 __end_request(req, uptodate, req->hard_cur_sectors << 9, 1);
3753}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754EXPORT_SYMBOL(end_request);
3755
NeilBrown66846572007-08-16 13:31:28 +02003756static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3757 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003759 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3760 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761
3762 rq->nr_phys_segments = bio_phys_segments(q, bio);
3763 rq->nr_hw_segments = bio_hw_segments(q, bio);
3764 rq->current_nr_sectors = bio_cur_sectors(bio);
3765 rq->hard_cur_sectors = rq->current_nr_sectors;
3766 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3767 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003768 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769
3770 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771
NeilBrown66846572007-08-16 13:31:28 +02003772 if (bio->bi_bdev)
3773 rq->rq_disk = bio->bi_bdev->bd_disk;
3774}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775
3776int kblockd_schedule_work(struct work_struct *work)
3777{
3778 return queue_work(kblockd_workqueue, work);
3779}
3780
3781EXPORT_SYMBOL(kblockd_schedule_work);
3782
Andrew Morton19a75d82007-05-09 02:33:56 -07003783void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003785 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786}
Andrew Morton19a75d82007-05-09 02:33:56 -07003787EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788
3789int __init blk_dev_init(void)
3790{
Jens Axboeff856ba2006-01-09 16:02:34 +01003791 int i;
3792
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 kblockd_workqueue = create_workqueue("kblockd");
3794 if (!kblockd_workqueue)
3795 panic("Failed to create kblockd\n");
3796
3797 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003798 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799
3800 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003801 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802
3803 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003804 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003806 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003807 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3808
3809 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003810 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003811
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003812 blk_max_low_pfn = max_low_pfn - 1;
3813 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
3815 return 0;
3816}
3817
3818/*
3819 * IO Context helper functions
3820 */
3821void put_io_context(struct io_context *ioc)
3822{
3823 if (ioc == NULL)
3824 return;
3825
3826 BUG_ON(atomic_read(&ioc->refcount) == 0);
3827
3828 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003829 struct cfq_io_context *cic;
3830
Al Viro334e94d2006-03-18 15:05:53 -05003831 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832 if (ioc->aic && ioc->aic->dtor)
3833 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003834 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003835 struct rb_node *n = rb_first(&ioc->cic_root);
3836
3837 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003838 cic->dtor(ioc);
3839 }
Al Viro334e94d2006-03-18 15:05:53 -05003840 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841
3842 kmem_cache_free(iocontext_cachep, ioc);
3843 }
3844}
3845EXPORT_SYMBOL(put_io_context);
3846
3847/* Called by the exitting task */
3848void exit_io_context(void)
3849{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003851 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003852
Jens Axboe22e2c502005-06-27 10:55:12 +02003853 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 ioc = current->io_context;
3855 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003856 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857
Oleg Nesterov25034d72006-08-29 09:15:14 +02003858 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859 if (ioc->aic && ioc->aic->exit)
3860 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003861 if (ioc->cic_root.rb_node != NULL) {
3862 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3863 cic->exit(ioc);
3864 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003865
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866 put_io_context(ioc);
3867}
3868
3869/*
3870 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003871 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003873 * This returned IO context doesn't have a specifically elevated refcount,
3874 * but since the current task itself holds a reference, the context can be
3875 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003877static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878{
3879 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 struct io_context *ret;
3881
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003883 if (likely(ret))
3884 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885
Jens Axboeb5deef92006-07-19 23:39:40 +02003886 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 if (ret) {
3888 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003889 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003890 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 ret->last_waited = jiffies; /* doesn't matter... */
3892 ret->nr_batch_requests = 0; /* because this is 0 */
3893 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003894 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003895 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003896 /* make sure set_task_ioprio() sees the settings above */
3897 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003898 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 }
3900
3901 return ret;
3902}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003903
3904/*
3905 * If the current task has no IO context then create one and initialise it.
3906 * If it does have a context, take a ref on it.
3907 *
3908 * This is always called in the context of the task which submitted the I/O.
3909 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003910struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003911{
3912 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003913 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003914 if (likely(ret))
3915 atomic_inc(&ret->refcount);
3916 return ret;
3917}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918EXPORT_SYMBOL(get_io_context);
3919
3920void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3921{
3922 struct io_context *src = *psrc;
3923 struct io_context *dst = *pdst;
3924
3925 if (src) {
3926 BUG_ON(atomic_read(&src->refcount) == 0);
3927 atomic_inc(&src->refcount);
3928 put_io_context(dst);
3929 *pdst = src;
3930 }
3931}
3932EXPORT_SYMBOL(copy_io_context);
3933
3934void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3935{
3936 struct io_context *temp;
3937 temp = *ioc1;
3938 *ioc1 = *ioc2;
3939 *ioc2 = temp;
3940}
3941EXPORT_SYMBOL(swap_io_context);
3942
3943/*
3944 * sysfs parts below
3945 */
3946struct queue_sysfs_entry {
3947 struct attribute attr;
3948 ssize_t (*show)(struct request_queue *, char *);
3949 ssize_t (*store)(struct request_queue *, const char *, size_t);
3950};
3951
3952static ssize_t
3953queue_var_show(unsigned int var, char *page)
3954{
3955 return sprintf(page, "%d\n", var);
3956}
3957
3958static ssize_t
3959queue_var_store(unsigned long *var, const char *page, size_t count)
3960{
3961 char *p = (char *) page;
3962
3963 *var = simple_strtoul(p, &p, 10);
3964 return count;
3965}
3966
3967static ssize_t queue_requests_show(struct request_queue *q, char *page)
3968{
3969 return queue_var_show(q->nr_requests, (page));
3970}
3971
3972static ssize_t
3973queue_requests_store(struct request_queue *q, const char *page, size_t count)
3974{
3975 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003976 unsigned long nr;
3977 int ret = queue_var_store(&nr, page, count);
3978 if (nr < BLKDEV_MIN_RQ)
3979 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980
Al Viroc981ff92006-03-18 13:51:29 -05003981 spin_lock_irq(q->queue_lock);
3982 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 blk_queue_congestion_threshold(q);
3984
3985 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003986 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003988 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
3990 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003991 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003993 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994
3995 if (rl->count[READ] >= q->nr_requests) {
3996 blk_set_queue_full(q, READ);
3997 } else if (rl->count[READ]+1 <= q->nr_requests) {
3998 blk_clear_queue_full(q, READ);
3999 wake_up(&rl->wait[READ]);
4000 }
4001
4002 if (rl->count[WRITE] >= q->nr_requests) {
4003 blk_set_queue_full(q, WRITE);
4004 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
4005 blk_clear_queue_full(q, WRITE);
4006 wake_up(&rl->wait[WRITE]);
4007 }
Al Viroc981ff92006-03-18 13:51:29 -05004008 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009 return ret;
4010}
4011
4012static ssize_t queue_ra_show(struct request_queue *q, char *page)
4013{
4014 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
4015
4016 return queue_var_show(ra_kb, (page));
4017}
4018
4019static ssize_t
4020queue_ra_store(struct request_queue *q, const char *page, size_t count)
4021{
4022 unsigned long ra_kb;
4023 ssize_t ret = queue_var_store(&ra_kb, page, count);
4024
4025 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
4027 spin_unlock_irq(q->queue_lock);
4028
4029 return ret;
4030}
4031
4032static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
4033{
4034 int max_sectors_kb = q->max_sectors >> 1;
4035
4036 return queue_var_show(max_sectors_kb, (page));
4037}
4038
4039static ssize_t
4040queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
4041{
4042 unsigned long max_sectors_kb,
4043 max_hw_sectors_kb = q->max_hw_sectors >> 1,
4044 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
4045 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046
4047 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
4048 return -EINVAL;
4049 /*
4050 * Take the queue lock to update the readahead and max_sectors
4051 * values synchronously:
4052 */
4053 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004054 q->max_sectors = max_sectors_kb << 1;
4055 spin_unlock_irq(q->queue_lock);
4056
4057 return ret;
4058}
4059
4060static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
4061{
4062 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
4063
4064 return queue_var_show(max_hw_sectors_kb, (page));
4065}
4066
Jens Axboe563063a2007-05-10 13:29:51 +02004067static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
4068{
4069 return queue_var_show(q->max_phys_segments, page);
4070}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Jens Axboe563063a2007-05-10 13:29:51 +02004072static ssize_t queue_max_segments_store(struct request_queue *q,
4073 const char *page, size_t count)
4074{
4075 unsigned long segments;
4076 ssize_t ret = queue_var_store(&segments, page, count);
4077
4078 spin_lock_irq(q->queue_lock);
4079 q->max_phys_segments = segments;
4080 spin_unlock_irq(q->queue_lock);
4081
4082 return ret;
4083}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084static struct queue_sysfs_entry queue_requests_entry = {
4085 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
4086 .show = queue_requests_show,
4087 .store = queue_requests_store,
4088};
4089
4090static struct queue_sysfs_entry queue_ra_entry = {
4091 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4092 .show = queue_ra_show,
4093 .store = queue_ra_store,
4094};
4095
4096static struct queue_sysfs_entry queue_max_sectors_entry = {
4097 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4098 .show = queue_max_sectors_show,
4099 .store = queue_max_sectors_store,
4100};
4101
4102static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4103 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4104 .show = queue_max_hw_sectors_show,
4105};
4106
Jens Axboe563063a2007-05-10 13:29:51 +02004107static struct queue_sysfs_entry queue_max_segments_entry = {
4108 .attr = {.name = "max_segments", .mode = S_IRUGO | S_IWUSR },
4109 .show = queue_max_segments_show,
4110 .store = queue_max_segments_store,
4111};
4112
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113static struct queue_sysfs_entry queue_iosched_entry = {
4114 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4115 .show = elv_iosched_show,
4116 .store = elv_iosched_store,
4117};
4118
4119static struct attribute *default_attrs[] = {
4120 &queue_requests_entry.attr,
4121 &queue_ra_entry.attr,
4122 &queue_max_hw_sectors_entry.attr,
4123 &queue_max_sectors_entry.attr,
Jens Axboe563063a2007-05-10 13:29:51 +02004124 &queue_max_segments_entry.attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 &queue_iosched_entry.attr,
4126 NULL,
4127};
4128
4129#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4130
4131static ssize_t
4132queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4133{
4134 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004135 struct request_queue *q =
4136 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004137 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004140 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004141 mutex_lock(&q->sysfs_lock);
4142 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4143 mutex_unlock(&q->sysfs_lock);
4144 return -ENOENT;
4145 }
4146 res = entry->show(q, page);
4147 mutex_unlock(&q->sysfs_lock);
4148 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004149}
4150
4151static ssize_t
4152queue_attr_store(struct kobject *kobj, struct attribute *attr,
4153 const char *page, size_t length)
4154{
4155 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004156 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157
Al Viro483f4af2006-03-18 18:34:37 -05004158 ssize_t res;
4159
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004161 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004162 mutex_lock(&q->sysfs_lock);
4163 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4164 mutex_unlock(&q->sysfs_lock);
4165 return -ENOENT;
4166 }
4167 res = entry->store(q, page, length);
4168 mutex_unlock(&q->sysfs_lock);
4169 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170}
4171
4172static struct sysfs_ops queue_sysfs_ops = {
4173 .show = queue_attr_show,
4174 .store = queue_attr_store,
4175};
4176
Adrian Bunk93d17d32005-06-25 14:59:10 -07004177static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004178 .sysfs_ops = &queue_sysfs_ops,
4179 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004180 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181};
4182
4183int blk_register_queue(struct gendisk *disk)
4184{
4185 int ret;
4186
Jens Axboe165125e2007-07-24 09:28:11 +02004187 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188
4189 if (!q || !q->request_fn)
4190 return -ENXIO;
4191
4192 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193
Al Viro483f4af2006-03-18 18:34:37 -05004194 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 if (ret < 0)
4196 return ret;
4197
Al Viro483f4af2006-03-18 18:34:37 -05004198 kobject_uevent(&q->kobj, KOBJ_ADD);
4199
Linus Torvalds1da177e2005-04-16 15:20:36 -07004200 ret = elv_register_queue(q);
4201 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004202 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4203 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 return ret;
4205 }
4206
4207 return 0;
4208}
4209
4210void blk_unregister_queue(struct gendisk *disk)
4211{
Jens Axboe165125e2007-07-24 09:28:11 +02004212 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004213
4214 if (q && q->request_fn) {
4215 elv_unregister_queue(q);
4216
Al Viro483f4af2006-03-18 18:34:37 -05004217 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4218 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219 kobject_put(&disk->kobj);
4220 }
4221}