blob: a15845c164f298c97854e21e1c7d79019af646a6 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34/*
35 * for max sense size
36 */
37#include <scsi/scsi_cmnd.h>
38
David Howells65f27f32006-11-22 14:55:48 +000039static void blk_unplug_work(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040static void blk_unplug_timeout(unsigned long data);
Adrian Bunk93d17d32005-06-25 14:59:10 -070041static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
Tejun Heo52d9e672006-01-06 09:49:58 +010042static void init_request_from_bio(struct request *req, struct bio *bio);
Jens Axboe165125e2007-07-24 09:28:11 +020043static int __make_request(struct request_queue *q, struct bio *bio);
Jens Axboeb5deef92006-07-19 23:39:40 +020044static struct io_context *current_io_context(gfp_t gfp_flags, int node);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * For the allocated request tables
48 */
Christoph Lametere18b8902006-12-06 20:33:20 -080049static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51/*
52 * For queue allocation
53 */
Christoph Lametere18b8902006-12-06 20:33:20 -080054static struct kmem_cache *requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56/*
57 * For io context allocations
58 */
Christoph Lametere18b8902006-12-06 20:33:20 -080059static struct kmem_cache *iocontext_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061/*
62 * Controlling structure to kblockd
63 */
Jens Axboeff856ba2006-01-09 16:02:34 +010064static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66unsigned long blk_max_low_pfn, blk_max_pfn;
67
68EXPORT_SYMBOL(blk_max_low_pfn);
69EXPORT_SYMBOL(blk_max_pfn);
70
Jens Axboeff856ba2006-01-09 16:02:34 +010071static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/* Amount of time in which a process may batch requests */
74#define BLK_BATCH_TIME (HZ/50UL)
75
76/* Number of requests a "batching" process may submit */
77#define BLK_BATCH_REQ 32
78
79/*
80 * Return the threshold (number of used requests) at which the queue is
81 * considered to be congested. It include a little hysteresis to keep the
82 * context switch rate down.
83 */
84static inline int queue_congestion_on_threshold(struct request_queue *q)
85{
86 return q->nr_congestion_on;
87}
88
89/*
90 * The threshold at which a queue is considered to be uncongested
91 */
92static inline int queue_congestion_off_threshold(struct request_queue *q)
93{
94 return q->nr_congestion_off;
95}
96
97static void blk_queue_congestion_threshold(struct request_queue *q)
98{
99 int nr;
100
101 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests)
103 nr = q->nr_requests;
104 q->nr_congestion_on = nr;
105
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 if (nr < 1)
108 nr = 1;
109 q->nr_congestion_off = nr;
110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/**
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * @bdev: device
115 *
116 * Locates the passed device's request queue and returns the address of its
117 * backing_dev_info
118 *
119 * Will return NULL if the request queue cannot be located.
120 */
121struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122{
123 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200124 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126 if (q)
127 ret = &q->backing_dev_info;
128 return ret;
129}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130EXPORT_SYMBOL(blk_get_backing_dev_info);
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132/**
133 * blk_queue_prep_rq - set a prepare_request function for queue
134 * @q: queue
135 * @pfn: prepare_request function
136 *
137 * It's possible for a queue to register a prepare_request callback which
138 * is invoked before the request is handed to the request_fn. The goal of
139 * the function is to prepare a request for I/O, it can be used to build a
140 * cdb from the request data for instance.
141 *
142 */
Jens Axboe165125e2007-07-24 09:28:11 +0200143void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 q->prep_rq_fn = pfn;
146}
147
148EXPORT_SYMBOL(blk_queue_prep_rq);
149
150/**
151 * blk_queue_merge_bvec - set a merge_bvec function for queue
152 * @q: queue
153 * @mbfn: merge_bvec_fn
154 *
155 * Usually queues have static limitations on the max sectors or segments that
156 * we can put in a request. Stacking drivers may have some settings that
157 * are dynamic, and thus we have to query the queue whether it is ok to
158 * add a new bio_vec to a bio at a given offset or not. If the block device
159 * has such limitations, it needs to register a merge_bvec_fn to control
160 * the size of bio's sent to it. Note that a block device *must* allow a
161 * single page to be added to an empty bio. The block device driver may want
162 * to use the bio_split() function to deal with these bio's. By default
163 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
164 * honored.
165 */
Jens Axboe165125e2007-07-24 09:28:11 +0200166void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
168 q->merge_bvec_fn = mbfn;
169}
170
171EXPORT_SYMBOL(blk_queue_merge_bvec);
172
Jens Axboe165125e2007-07-24 09:28:11 +0200173void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
Jens Axboeff856ba2006-01-09 16:02:34 +0100174{
175 q->softirq_done_fn = fn;
176}
177
178EXPORT_SYMBOL(blk_queue_softirq_done);
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180/**
181 * blk_queue_make_request - define an alternate make_request function for a device
182 * @q: the request queue for the device to be affected
183 * @mfn: the alternate make_request function
184 *
185 * Description:
186 * The normal way for &struct bios to be passed to a device
187 * driver is for them to be collected into requests on a request
188 * queue, and then to allow the device driver to select requests
189 * off that queue when it is ready. This works well for many block
190 * devices. However some block devices (typically virtual devices
191 * such as md or lvm) do not benefit from the processing on the
192 * request queue, and are served best by having the requests passed
193 * directly to them. This can be achieved by providing a function
194 * to blk_queue_make_request().
195 *
196 * Caveat:
197 * The driver that does this *must* be able to deal appropriately
198 * with buffers in "highmemory". This can be accomplished by either calling
199 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
200 * blk_queue_bounce() to create a buffer in normal memory.
201 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200202void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 /*
205 * set defaults
206 */
207 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700208 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
209 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 q->make_request_fn = mfn;
211 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
212 q->backing_dev_info.state = 0;
213 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600214 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 blk_queue_hardsect_size(q, 512);
216 blk_queue_dma_alignment(q, 511);
217 blk_queue_congestion_threshold(q);
218 q->nr_batching = BLK_BATCH_REQ;
219
220 q->unplug_thresh = 4; /* hmm */
221 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
222 if (q->unplug_delay == 0)
223 q->unplug_delay = 1;
224
David Howells65f27f32006-11-22 14:55:48 +0000225 INIT_WORK(&q->unplug_work, blk_unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227 q->unplug_timer.function = blk_unplug_timeout;
228 q->unplug_timer.data = (unsigned long)q;
229
230 /*
231 * by default assume old behaviour and bounce for any highmem page
232 */
233 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236EXPORT_SYMBOL(blk_queue_make_request);
237
Jens Axboe165125e2007-07-24 09:28:11 +0200238static void rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100241 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200245 INIT_HLIST_NODE(&rq->hash);
246 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200247 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 rq->buffer = NULL;
249 rq->ref_count = 1;
250 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 rq->special = NULL;
252 rq->data_len = 0;
253 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200254 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 rq->sense = NULL;
256 rq->end_io = NULL;
257 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100258 rq->completion_data = NULL;
FUJITA Tomonoriabae1fd2007-07-16 08:52:14 +0200259 rq->next_rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
261
262/**
263 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100264 * @q: the request queue
265 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100266 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 *
268 * Description:
269 * For journalled file systems, doing ordered writes on a commit
270 * block instead of explicitly doing wait_on_buffer (which is bad
271 * for performance) can be a big win. Block drivers supporting this
272 * feature should call this function and indicate so.
273 *
274 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200275int blk_queue_ordered(struct request_queue *q, unsigned ordered,
Tejun Heo797e7db2006-01-06 09:51:03 +0100276 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Tejun Heo797e7db2006-01-06 09:51:03 +0100278 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
279 prepare_flush_fn == NULL) {
280 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
281 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100283
284 if (ordered != QUEUE_ORDERED_NONE &&
285 ordered != QUEUE_ORDERED_DRAIN &&
286 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
287 ordered != QUEUE_ORDERED_DRAIN_FUA &&
288 ordered != QUEUE_ORDERED_TAG &&
289 ordered != QUEUE_ORDERED_TAG_FLUSH &&
290 ordered != QUEUE_ORDERED_TAG_FUA) {
291 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
292 return -EINVAL;
293 }
294
Tetsuo Takata60481b12006-01-24 10:34:36 +0100295 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100296 q->next_ordered = ordered;
297 q->prepare_flush_fn = prepare_flush_fn;
298
299 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
302EXPORT_SYMBOL(blk_queue_ordered);
303
304/**
305 * blk_queue_issue_flush_fn - set function for issuing a flush
306 * @q: the request queue
307 * @iff: the function to be called issuing the flush
308 *
309 * Description:
310 * If a driver supports issuing a flush command, the support is notified
311 * to the block layer by defining it through this call.
312 *
313 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200314void blk_queue_issue_flush_fn(struct request_queue *q, issue_flush_fn *iff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
316 q->issue_flush_fn = iff;
317}
318
319EXPORT_SYMBOL(blk_queue_issue_flush_fn);
320
321/*
322 * Cache flushing for ordered writes handling
323 */
Jens Axboe165125e2007-07-24 09:28:11 +0200324inline unsigned blk_ordered_cur_seq(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Tejun Heo797e7db2006-01-06 09:51:03 +0100326 if (!q->ordseq)
327 return 0;
328 return 1 << ffz(q->ordseq);
329}
330
331unsigned blk_ordered_req_seq(struct request *rq)
332{
Jens Axboe165125e2007-07-24 09:28:11 +0200333 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Tejun Heo797e7db2006-01-06 09:51:03 +0100335 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200336
Tejun Heo797e7db2006-01-06 09:51:03 +0100337 if (rq == &q->pre_flush_rq)
338 return QUEUE_ORDSEQ_PREFLUSH;
339 if (rq == &q->bar_rq)
340 return QUEUE_ORDSEQ_BAR;
341 if (rq == &q->post_flush_rq)
342 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Tejun Heobc90ba02007-06-15 13:24:28 +0200344 /*
345 * !fs requests don't need to follow barrier ordering. Always
346 * put them at the front. This fixes the following deadlock.
347 *
348 * http://thread.gmane.org/gmane.linux.kernel/537473
349 */
350 if (!blk_fs_request(rq))
351 return QUEUE_ORDSEQ_DRAIN;
352
Jens Axboe4aff5e22006-08-10 08:44:47 +0200353 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
354 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100355 return QUEUE_ORDSEQ_DRAIN;
356 else
357 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
Jens Axboe165125e2007-07-24 09:28:11 +0200360void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Tejun Heo797e7db2006-01-06 09:51:03 +0100362 struct request *rq;
363 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Tejun Heo797e7db2006-01-06 09:51:03 +0100365 if (error && !q->orderr)
366 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200367
Tejun Heo797e7db2006-01-06 09:51:03 +0100368 BUG_ON(q->ordseq & seq);
369 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Tejun Heo797e7db2006-01-06 09:51:03 +0100371 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
372 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100375 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100377 rq = q->orig_bar_rq;
378 uptodate = q->orderr ? q->orderr : 1;
379
380 q->ordseq = 0;
381
382 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
383 end_that_request_last(rq, uptodate);
384}
385
386static void pre_flush_end_io(struct request *rq, int error)
387{
388 elv_completed_request(rq->q, rq);
389 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
390}
391
392static void bar_end_io(struct request *rq, int error)
393{
394 elv_completed_request(rq->q, rq);
395 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
396}
397
398static void post_flush_end_io(struct request *rq, int error)
399{
400 elv_completed_request(rq->q, rq);
401 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
402}
403
Jens Axboe165125e2007-07-24 09:28:11 +0200404static void queue_flush(struct request_queue *q, unsigned which)
Tejun Heo797e7db2006-01-06 09:51:03 +0100405{
406 struct request *rq;
407 rq_end_io_fn *end_io;
408
409 if (which == QUEUE_ORDERED_PREFLUSH) {
410 rq = &q->pre_flush_rq;
411 end_io = pre_flush_end_io;
412 } else {
413 rq = &q->post_flush_rq;
414 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
416
Jens Axboe4aff5e22006-08-10 08:44:47 +0200417 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100418 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100419 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200420 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100421 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100422 rq->end_io = end_io;
423 q->prepare_flush_fn(q, rq);
424
Tejun Heo30e96562006-02-08 01:01:31 -0800425 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100426}
427
Jens Axboe165125e2007-07-24 09:28:11 +0200428static inline struct request *start_ordered(struct request_queue *q,
Tejun Heo797e7db2006-01-06 09:51:03 +0100429 struct request *rq)
430{
431 q->bi_size = 0;
432 q->orderr = 0;
433 q->ordered = q->next_ordered;
434 q->ordseq |= QUEUE_ORDSEQ_STARTED;
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100437 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100439 blkdev_dequeue_request(rq);
440 q->orig_bar_rq = rq;
441 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200442 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100443 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200444 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
445 rq->cmd_flags |= REQ_RW;
446 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100447 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200448 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 init_request_from_bio(rq, q->orig_bar_rq->bio);
450 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Tejun Heo797e7db2006-01-06 09:51:03 +0100452 /*
453 * Queue ordered sequence. As we stack them at the head, we
454 * need to queue in reverse order. Note that we rely on that
455 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
456 * request gets inbetween ordered sequence.
457 */
458 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
459 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
460 else
461 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Tejun Heo30e96562006-02-08 01:01:31 -0800463 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100464
465 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
466 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
467 rq = &q->pre_flush_rq;
468 } else
469 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
470
471 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
472 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
473 else
474 rq = NULL;
475
476 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
Jens Axboe165125e2007-07-24 09:28:11 +0200479int blk_do_ordered(struct request_queue *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800481 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100482 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Tejun Heo797e7db2006-01-06 09:51:03 +0100484 if (!q->ordseq) {
485 if (!is_barrier)
486 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
Tejun Heo797e7db2006-01-06 09:51:03 +0100488 if (q->next_ordered != QUEUE_ORDERED_NONE) {
489 *rqp = start_ordered(q, rq);
490 return 1;
491 } else {
492 /*
493 * This can happen when the queue switches to
494 * ORDERED_NONE while this request is on it.
495 */
496 blkdev_dequeue_request(rq);
497 end_that_request_first(rq, -EOPNOTSUPP,
498 rq->hard_nr_sectors);
499 end_that_request_last(rq, -EOPNOTSUPP);
500 *rqp = NULL;
501 return 0;
502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800505 /*
506 * Ordered sequence in progress
507 */
508
509 /* Special requests are not subject to ordering rules. */
510 if (!blk_fs_request(rq) &&
511 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
512 return 1;
513
Tejun Heo797e7db2006-01-06 09:51:03 +0100514 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800515 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100516 if (is_barrier && rq != &q->bar_rq)
517 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800518 } else {
519 /* Ordered by draining. Wait for turn. */
520 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
521 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
522 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 }
524
525 return 1;
526}
527
Tejun Heo797e7db2006-01-06 09:51:03 +0100528static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529{
Jens Axboe165125e2007-07-24 09:28:11 +0200530 struct request_queue *q = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Tejun Heo797e7db2006-01-06 09:51:03 +0100532 /*
533 * This is dry run, restore bio_sector and size. We'll finish
534 * this request again with the original bi_end_io after an
535 * error occurs or post flush is complete.
536 */
537 q->bi_size += bytes;
538
539 if (bio->bi_size)
540 return 1;
541
Tejun Heo797e7db2006-01-06 09:51:03 +0100542 /* Reset bio */
543 set_bit(BIO_UPTODATE, &bio->bi_flags);
544 bio->bi_size = q->bi_size;
545 bio->bi_sector -= (q->bi_size >> 9);
546 q->bi_size = 0;
547
548 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
Tejun Heo797e7db2006-01-06 09:51:03 +0100550
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200551static int ordered_bio_endio(struct request *rq, struct bio *bio,
552 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100553{
Jens Axboe165125e2007-07-24 09:28:11 +0200554 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100555 bio_end_io_t *endio;
556 void *private;
557
558 if (&q->bar_rq != rq)
559 return 0;
560
561 /*
562 * Okay, this is the barrier request in progress, dry finish it.
563 */
564 if (error && !q->orderr)
565 q->orderr = error;
566
567 endio = bio->bi_end_io;
568 private = bio->bi_private;
569 bio->bi_end_io = flush_dry_bio_endio;
570 bio->bi_private = q;
571
572 bio_endio(bio, nbytes, error);
573
574 bio->bi_end_io = endio;
575 bio->bi_private = private;
576
577 return 1;
578}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580/**
581 * blk_queue_bounce_limit - set bounce buffer limit for queue
582 * @q: the request queue for the device
583 * @dma_addr: bus address limit
584 *
585 * Description:
586 * Different hardware can have different requirements as to what pages
587 * it can do I/O directly to. A low level driver can call
588 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800589 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200591void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592{
593 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800594 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Andi Kleen5ee1af92006-03-08 17:57:26 -0800596 q->bounce_gfp = GFP_NOIO;
597#if BITS_PER_LONG == 64
598 /* Assume anything <= 4GB can be handled by IOMMU.
599 Actually some IOMMUs can handle everything, but I don't
600 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200601 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800602 dma = 1;
603 q->bounce_pfn = max_low_pfn;
604#else
605 if (bounce_pfn < blk_max_low_pfn)
606 dma = 1;
607 q->bounce_pfn = bounce_pfn;
608#endif
609 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 init_emergency_isa_pool();
611 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800612 q->bounce_pfn = bounce_pfn;
613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614}
615
616EXPORT_SYMBOL(blk_queue_bounce_limit);
617
618/**
619 * blk_queue_max_sectors - set max sectors for a request for this queue
620 * @q: the request queue for the device
621 * @max_sectors: max sectors in the usual 512b unit
622 *
623 * Description:
624 * Enables a low level driver to set an upper limit on the size of
625 * received requests.
626 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200627void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
629 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
630 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
631 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
632 }
633
Mike Christiedefd94b2005-12-05 02:37:06 -0600634 if (BLK_DEF_MAX_SECTORS > max_sectors)
635 q->max_hw_sectors = q->max_sectors = max_sectors;
636 else {
637 q->max_sectors = BLK_DEF_MAX_SECTORS;
638 q->max_hw_sectors = max_sectors;
639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
641
642EXPORT_SYMBOL(blk_queue_max_sectors);
643
644/**
645 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
646 * @q: the request queue for the device
647 * @max_segments: max number of segments
648 *
649 * Description:
650 * Enables a low level driver to set an upper limit on the number of
651 * physical data segments in a request. This would be the largest sized
652 * scatter list the driver could handle.
653 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200654void blk_queue_max_phys_segments(struct request_queue *q,
655 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
657 if (!max_segments) {
658 max_segments = 1;
659 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
660 }
661
662 q->max_phys_segments = max_segments;
663}
664
665EXPORT_SYMBOL(blk_queue_max_phys_segments);
666
667/**
668 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
669 * @q: the request queue for the device
670 * @max_segments: max number of segments
671 *
672 * Description:
673 * Enables a low level driver to set an upper limit on the number of
674 * hw data segments in a request. This would be the largest number of
675 * address/length pairs the host adapter can actually give as once
676 * to the device.
677 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200678void blk_queue_max_hw_segments(struct request_queue *q,
679 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 if (!max_segments) {
682 max_segments = 1;
683 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
684 }
685
686 q->max_hw_segments = max_segments;
687}
688
689EXPORT_SYMBOL(blk_queue_max_hw_segments);
690
691/**
692 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
693 * @q: the request queue for the device
694 * @max_size: max size of segment in bytes
695 *
696 * Description:
697 * Enables a low level driver to set an upper limit on the size of a
698 * coalesced segment
699 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200700void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 if (max_size < PAGE_CACHE_SIZE) {
703 max_size = PAGE_CACHE_SIZE;
704 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
705 }
706
707 q->max_segment_size = max_size;
708}
709
710EXPORT_SYMBOL(blk_queue_max_segment_size);
711
712/**
713 * blk_queue_hardsect_size - set hardware sector size for the queue
714 * @q: the request queue for the device
715 * @size: the hardware sector size, in bytes
716 *
717 * Description:
718 * This should typically be set to the lowest possible sector size
719 * that the hardware can operate on (possible without reverting to
720 * even internal read-modify-write operations). Usually the default
721 * of 512 covers most hardware.
722 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200723void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
725 q->hardsect_size = size;
726}
727
728EXPORT_SYMBOL(blk_queue_hardsect_size);
729
730/*
731 * Returns the minimum that is _not_ zero, unless both are zero.
732 */
733#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
734
735/**
736 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
737 * @t: the stacking driver (top)
738 * @b: the underlying device (bottom)
739 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200740void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600743 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
744 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
747 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
748 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
749 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800750 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
751 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
754EXPORT_SYMBOL(blk_queue_stack_limits);
755
756/**
757 * blk_queue_segment_boundary - set boundary rules for segment merging
758 * @q: the request queue for the device
759 * @mask: the memory boundary mask
760 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200761void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
763 if (mask < PAGE_CACHE_SIZE - 1) {
764 mask = PAGE_CACHE_SIZE - 1;
765 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
766 }
767
768 q->seg_boundary_mask = mask;
769}
770
771EXPORT_SYMBOL(blk_queue_segment_boundary);
772
773/**
774 * blk_queue_dma_alignment - set dma length and memory alignment
775 * @q: the request queue for the device
776 * @mask: alignment mask
777 *
778 * description:
779 * set required memory and length aligment for direct dma transactions.
780 * this is used when buiding direct io requests for the queue.
781 *
782 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200783void blk_queue_dma_alignment(struct request_queue *q, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
785 q->dma_alignment = mask;
786}
787
788EXPORT_SYMBOL(blk_queue_dma_alignment);
789
790/**
791 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 * @q: The request queue for the device
793 * @tag: The tag of the request
794 *
795 * Notes:
796 * Should be used when a device returns a tag and you want to match
797 * it with a request.
798 *
799 * no locks need be held.
800 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200801struct request *blk_queue_find_tag(struct request_queue *q, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
David C Somayajuluf583f492006-10-04 08:27:25 +0200803 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
806EXPORT_SYMBOL(blk_queue_find_tag);
807
808/**
James Bottomley492dfb42006-08-30 15:48:45 -0400809 * __blk_free_tags - release a given set of tag maintenance info
810 * @bqt: the tag map to free
811 *
812 * Tries to free the specified @bqt@. Returns true if it was
813 * actually freed and false if there are still references using it
814 */
815static int __blk_free_tags(struct blk_queue_tag *bqt)
816{
817 int retval;
818
819 retval = atomic_dec_and_test(&bqt->refcnt);
820 if (retval) {
821 BUG_ON(bqt->busy);
822 BUG_ON(!list_empty(&bqt->busy_list));
823
824 kfree(bqt->tag_index);
825 bqt->tag_index = NULL;
826
827 kfree(bqt->tag_map);
828 bqt->tag_map = NULL;
829
830 kfree(bqt);
831
832 }
833
834 return retval;
835}
836
837/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 * __blk_queue_free_tags - release tag maintenance info
839 * @q: the request queue for the device
840 *
841 * Notes:
842 * blk_cleanup_queue() will take care of calling this function, if tagging
843 * has been used. So there's no need to call this directly.
844 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200845static void __blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
847 struct blk_queue_tag *bqt = q->queue_tags;
848
849 if (!bqt)
850 return;
851
James Bottomley492dfb42006-08-30 15:48:45 -0400852 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 q->queue_tags = NULL;
855 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
856}
857
James Bottomley492dfb42006-08-30 15:48:45 -0400858
859/**
860 * blk_free_tags - release a given set of tag maintenance info
861 * @bqt: the tag map to free
862 *
863 * For externally managed @bqt@ frees the map. Callers of this
864 * function must guarantee to have released all the queues that
865 * might have been using this tag map.
866 */
867void blk_free_tags(struct blk_queue_tag *bqt)
868{
869 if (unlikely(!__blk_free_tags(bqt)))
870 BUG();
871}
872EXPORT_SYMBOL(blk_free_tags);
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874/**
875 * blk_queue_free_tags - release tag maintenance info
876 * @q: the request queue for the device
877 *
878 * Notes:
879 * This is used to disabled tagged queuing to a device, yet leave
880 * queue in function.
881 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200882void blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883{
884 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
885}
886
887EXPORT_SYMBOL(blk_queue_free_tags);
888
889static int
Jens Axboe165125e2007-07-24 09:28:11 +0200890init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 struct request **tag_index;
893 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700894 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
James Bottomley492dfb42006-08-30 15:48:45 -0400896 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 depth = q->nr_requests * 2;
898 printk(KERN_ERR "%s: adjusted depth to %d\n",
899 __FUNCTION__, depth);
900 }
901
Jens Axboef68110f2006-03-08 13:31:44 +0100902 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!tag_index)
904 goto fail;
905
Tejun Heof7d37d02005-06-23 00:08:50 -0700906 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100907 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (!tag_map)
909 goto fail;
910
Tejun Heoba025082005-08-05 13:28:11 -0700911 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 tags->tag_index = tag_index;
914 tags->tag_map = tag_map;
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 return 0;
917fail:
918 kfree(tag_index);
919 return -ENOMEM;
920}
921
James Bottomley492dfb42006-08-30 15:48:45 -0400922static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
923 int depth)
924{
925 struct blk_queue_tag *tags;
926
927 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
928 if (!tags)
929 goto fail;
930
931 if (init_tag_map(q, tags, depth))
932 goto fail;
933
934 INIT_LIST_HEAD(&tags->busy_list);
935 tags->busy = 0;
936 atomic_set(&tags->refcnt, 1);
937 return tags;
938fail:
939 kfree(tags);
940 return NULL;
941}
942
943/**
944 * blk_init_tags - initialize the tag info for an external tag map
945 * @depth: the maximum queue depth supported
946 * @tags: the tag to use
947 **/
948struct blk_queue_tag *blk_init_tags(int depth)
949{
950 return __blk_queue_init_tags(NULL, depth);
951}
952EXPORT_SYMBOL(blk_init_tags);
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954/**
955 * blk_queue_init_tags - initialize the queue tag info
956 * @q: the request queue for the device
957 * @depth: the maximum queue depth supported
958 * @tags: the tag to use
959 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200960int blk_queue_init_tags(struct request_queue *q, int depth,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 struct blk_queue_tag *tags)
962{
963 int rc;
964
965 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
966
967 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400968 tags = __blk_queue_init_tags(q, depth);
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (!tags)
971 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 } else if (q->queue_tags) {
973 if ((rc = blk_queue_resize_tags(q, depth)))
974 return rc;
975 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
976 return 0;
977 } else
978 atomic_inc(&tags->refcnt);
979
980 /*
981 * assign it, all done
982 */
983 q->queue_tags = tags;
984 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
985 return 0;
986fail:
987 kfree(tags);
988 return -ENOMEM;
989}
990
991EXPORT_SYMBOL(blk_queue_init_tags);
992
993/**
994 * blk_queue_resize_tags - change the queueing depth
995 * @q: the request queue for the device
996 * @new_depth: the new max command queueing depth
997 *
998 * Notes:
999 * Must be called with the queue lock held.
1000 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001001int blk_queue_resize_tags(struct request_queue *q, int new_depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002{
1003 struct blk_queue_tag *bqt = q->queue_tags;
1004 struct request **tag_index;
1005 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -07001006 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 if (!bqt)
1009 return -ENXIO;
1010
1011 /*
Tejun Heoba025082005-08-05 13:28:11 -07001012 * if we already have large enough real_max_depth. just
1013 * adjust max_depth. *NOTE* as requests with tag value
1014 * between new_depth and real_max_depth can be in-flight, tag
1015 * map can not be shrunk blindly here.
1016 */
1017 if (new_depth <= bqt->real_max_depth) {
1018 bqt->max_depth = new_depth;
1019 return 0;
1020 }
1021
1022 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001023 * Currently cannot replace a shared tag map with a new
1024 * one, so error out if this is the case
1025 */
1026 if (atomic_read(&bqt->refcnt) != 1)
1027 return -EBUSY;
1028
1029 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 * save the old state info, so we can copy it back
1031 */
1032 tag_index = bqt->tag_index;
1033 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001034 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036 if (init_tag_map(q, bqt, new_depth))
1037 return -ENOMEM;
1038
1039 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001040 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001041 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 kfree(tag_index);
1044 kfree(tag_map);
1045 return 0;
1046}
1047
1048EXPORT_SYMBOL(blk_queue_resize_tags);
1049
1050/**
1051 * blk_queue_end_tag - end tag operations for a request
1052 * @q: the request queue for the device
1053 * @rq: the request that has completed
1054 *
1055 * Description:
1056 * Typically called when end_that_request_first() returns 0, meaning
1057 * all transfers have been done for a request. It's important to call
1058 * this function before end_that_request_last(), as that will put the
1059 * request back on the free list thus corrupting the internal tag list.
1060 *
1061 * Notes:
1062 * queue lock must be held.
1063 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001064void blk_queue_end_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065{
1066 struct blk_queue_tag *bqt = q->queue_tags;
1067 int tag = rq->tag;
1068
1069 BUG_ON(tag == -1);
1070
Tejun Heoba025082005-08-05 13:28:11 -07001071 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001072 /*
1073 * This can happen after tag depth has been reduced.
1074 * FIXME: how about a warning or info message here?
1075 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return;
1077
1078 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001079 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1080 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return;
1082 }
1083
1084 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001085 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 rq->tag = -1;
1087
1088 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001089 printk(KERN_ERR "%s: tag %d is missing\n",
1090 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 bqt->tag_index[tag] = NULL;
1093 bqt->busy--;
1094}
1095
1096EXPORT_SYMBOL(blk_queue_end_tag);
1097
1098/**
1099 * blk_queue_start_tag - find a free tag and assign it
1100 * @q: the request queue for the device
1101 * @rq: the block request that needs tagging
1102 *
1103 * Description:
1104 * This can either be used as a stand-alone helper, or possibly be
1105 * assigned as the queue &prep_rq_fn (in which case &struct request
1106 * automagically gets a tag assigned). Note that this function
1107 * assumes that any type of request can be queued! if this is not
1108 * true for your device, you must check the request type before
1109 * calling this function. The request will also be removed from
1110 * the request queue, so it's the drivers responsibility to readd
1111 * it if it should need to be restarted for some reason.
1112 *
1113 * Notes:
1114 * queue lock must be held.
1115 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001116int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001119 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Jens Axboe4aff5e22006-08-10 08:44:47 +02001121 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001123 "%s: request %p for device [%s] already tagged %d",
1124 __FUNCTION__, rq,
1125 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 BUG();
1127 }
1128
Jens Axboe059af492006-09-21 20:37:22 +02001129 /*
1130 * Protect against shared tag maps, as we may not have exclusive
1131 * access to the tag map.
1132 */
1133 do {
1134 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1135 if (tag >= bqt->max_depth)
1136 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Jens Axboe059af492006-09-21 20:37:22 +02001138 } while (test_and_set_bit(tag, bqt->tag_map));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Jens Axboe4aff5e22006-08-10 08:44:47 +02001140 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 rq->tag = tag;
1142 bqt->tag_index[tag] = rq;
1143 blkdev_dequeue_request(rq);
1144 list_add(&rq->queuelist, &bqt->busy_list);
1145 bqt->busy++;
1146 return 0;
1147}
1148
1149EXPORT_SYMBOL(blk_queue_start_tag);
1150
1151/**
1152 * blk_queue_invalidate_tags - invalidate all pending tags
1153 * @q: the request queue for the device
1154 *
1155 * Description:
1156 * Hardware conditions may dictate a need to stop all pending requests.
1157 * In this case, we will safely clear the block side of the tag queue and
1158 * readd all requests to the request queue in the right order.
1159 *
1160 * Notes:
1161 * queue lock must be held.
1162 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001163void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164{
1165 struct blk_queue_tag *bqt = q->queue_tags;
1166 struct list_head *tmp, *n;
1167 struct request *rq;
1168
1169 list_for_each_safe(tmp, n, &bqt->busy_list) {
1170 rq = list_entry_rq(tmp);
1171
1172 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001173 printk(KERN_ERR
1174 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001176 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 } else
1178 blk_queue_end_tag(q, rq);
1179
Jens Axboe4aff5e22006-08-10 08:44:47 +02001180 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1182 }
1183}
1184
1185EXPORT_SYMBOL(blk_queue_invalidate_tags);
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187void blk_dump_rq_flags(struct request *rq, char *msg)
1188{
1189 int bit;
1190
Jens Axboe4aff5e22006-08-10 08:44:47 +02001191 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1192 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1193 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1196 rq->nr_sectors,
1197 rq->current_nr_sectors);
1198 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1199
Jens Axboe4aff5e22006-08-10 08:44:47 +02001200 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 printk("cdb: ");
1202 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1203 printk("%02x ", rq->cmd[bit]);
1204 printk("\n");
1205 }
1206}
1207
1208EXPORT_SYMBOL(blk_dump_rq_flags);
1209
Jens Axboe165125e2007-07-24 09:28:11 +02001210void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 struct bio_vec *bv, *bvprv = NULL;
1213 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1214 int high, highprv = 1;
1215
1216 if (unlikely(!bio->bi_io_vec))
1217 return;
1218
1219 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1220 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1221 bio_for_each_segment(bv, bio, i) {
1222 /*
1223 * the trick here is making sure that a high page is never
1224 * considered part of another segment, since that might
1225 * change with the bounce page.
1226 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001227 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (high || highprv)
1229 goto new_hw_segment;
1230 if (cluster) {
1231 if (seg_size + bv->bv_len > q->max_segment_size)
1232 goto new_segment;
1233 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1234 goto new_segment;
1235 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1236 goto new_segment;
1237 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1238 goto new_hw_segment;
1239
1240 seg_size += bv->bv_len;
1241 hw_seg_size += bv->bv_len;
1242 bvprv = bv;
1243 continue;
1244 }
1245new_segment:
1246 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1247 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1248 hw_seg_size += bv->bv_len;
1249 } else {
1250new_hw_segment:
1251 if (hw_seg_size > bio->bi_hw_front_size)
1252 bio->bi_hw_front_size = hw_seg_size;
1253 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1254 nr_hw_segs++;
1255 }
1256
1257 nr_phys_segs++;
1258 bvprv = bv;
1259 seg_size = bv->bv_len;
1260 highprv = high;
1261 }
1262 if (hw_seg_size > bio->bi_hw_back_size)
1263 bio->bi_hw_back_size = hw_seg_size;
1264 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1265 bio->bi_hw_front_size = hw_seg_size;
1266 bio->bi_phys_segments = nr_phys_segs;
1267 bio->bi_hw_segments = nr_hw_segs;
1268 bio->bi_flags |= (1 << BIO_SEG_VALID);
1269}
Neil Brown387bb172007-02-08 14:20:29 -08001270EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Jens Axboe165125e2007-07-24 09:28:11 +02001272static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 struct bio *nxt)
1274{
1275 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1276 return 0;
1277
1278 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1279 return 0;
1280 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1281 return 0;
1282
1283 /*
1284 * bio and nxt are contigous in memory, check if the queue allows
1285 * these two to be merged into one
1286 */
1287 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1288 return 1;
1289
1290 return 0;
1291}
1292
Jens Axboe165125e2007-07-24 09:28:11 +02001293static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 struct bio *nxt)
1295{
1296 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1297 blk_recount_segments(q, bio);
1298 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1299 blk_recount_segments(q, nxt);
1300 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001301 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001303 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 return 0;
1305
1306 return 1;
1307}
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309/*
1310 * map a request to scatterlist, return number of sg entries setup. Caller
1311 * must make sure sg can hold rq->nr_phys_segments entries
1312 */
Jens Axboe165125e2007-07-24 09:28:11 +02001313int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1314 struct scatterlist *sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 struct bio_vec *bvec, *bvprv;
1317 struct bio *bio;
1318 int nsegs, i, cluster;
1319
1320 nsegs = 0;
1321 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1322
1323 /*
1324 * for each bio in rq
1325 */
1326 bvprv = NULL;
1327 rq_for_each_bio(bio, rq) {
1328 /*
1329 * for each segment in bio
1330 */
1331 bio_for_each_segment(bvec, bio, i) {
1332 int nbytes = bvec->bv_len;
1333
1334 if (bvprv && cluster) {
1335 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1336 goto new_segment;
1337
1338 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1339 goto new_segment;
1340 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1341 goto new_segment;
1342
1343 sg[nsegs - 1].length += nbytes;
1344 } else {
1345new_segment:
1346 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1347 sg[nsegs].page = bvec->bv_page;
1348 sg[nsegs].length = nbytes;
1349 sg[nsegs].offset = bvec->bv_offset;
1350
1351 nsegs++;
1352 }
1353 bvprv = bvec;
1354 } /* segments in bio */
1355 } /* bios in rq */
1356
1357 return nsegs;
1358}
1359
1360EXPORT_SYMBOL(blk_rq_map_sg);
1361
1362/*
1363 * the standard queue merge functions, can be overridden with device
1364 * specific ones if so desired
1365 */
1366
Jens Axboe165125e2007-07-24 09:28:11 +02001367static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 struct request *req,
1369 struct bio *bio)
1370{
1371 int nr_phys_segs = bio_phys_segments(q, bio);
1372
1373 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001374 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 if (req == q->last_merge)
1376 q->last_merge = NULL;
1377 return 0;
1378 }
1379
1380 /*
1381 * A hw segment is just getting larger, bump just the phys
1382 * counter.
1383 */
1384 req->nr_phys_segments += nr_phys_segs;
1385 return 1;
1386}
1387
Jens Axboe165125e2007-07-24 09:28:11 +02001388static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct request *req,
1390 struct bio *bio)
1391{
1392 int nr_hw_segs = bio_hw_segments(q, bio);
1393 int nr_phys_segs = bio_phys_segments(q, bio);
1394
1395 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1396 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001397 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 if (req == q->last_merge)
1399 q->last_merge = NULL;
1400 return 0;
1401 }
1402
1403 /*
1404 * This will form the start of a new hw segment. Bump both
1405 * counters.
1406 */
1407 req->nr_hw_segments += nr_hw_segs;
1408 req->nr_phys_segments += nr_phys_segs;
1409 return 1;
1410}
1411
Jens Axboe165125e2007-07-24 09:28:11 +02001412int ll_back_merge_fn(struct request_queue *q, struct request *req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
Mike Christiedefd94b2005-12-05 02:37:06 -06001414 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 int len;
1416
Mike Christiedefd94b2005-12-05 02:37:06 -06001417 if (unlikely(blk_pc_request(req)))
1418 max_sectors = q->max_hw_sectors;
1419 else
1420 max_sectors = q->max_sectors;
1421
1422 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001423 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (req == q->last_merge)
1425 q->last_merge = NULL;
1426 return 0;
1427 }
1428 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1429 blk_recount_segments(q, req->biotail);
1430 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1431 blk_recount_segments(q, bio);
1432 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1433 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1434 !BIOVEC_VIRT_OVERSIZE(len)) {
1435 int mergeable = ll_new_mergeable(q, req, bio);
1436
1437 if (mergeable) {
1438 if (req->nr_hw_segments == 1)
1439 req->bio->bi_hw_front_size = len;
1440 if (bio->bi_hw_segments == 1)
1441 bio->bi_hw_back_size = len;
1442 }
1443 return mergeable;
1444 }
1445
1446 return ll_new_hw_segment(q, req, bio);
1447}
Jens Axboe1aa4f242006-12-19 08:33:11 +01001448EXPORT_SYMBOL(ll_back_merge_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Jens Axboe165125e2007-07-24 09:28:11 +02001450static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 struct bio *bio)
1452{
Mike Christiedefd94b2005-12-05 02:37:06 -06001453 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 int len;
1455
Mike Christiedefd94b2005-12-05 02:37:06 -06001456 if (unlikely(blk_pc_request(req)))
1457 max_sectors = q->max_hw_sectors;
1458 else
1459 max_sectors = q->max_sectors;
1460
1461
1462 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001463 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (req == q->last_merge)
1465 q->last_merge = NULL;
1466 return 0;
1467 }
1468 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1469 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1470 blk_recount_segments(q, bio);
1471 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1472 blk_recount_segments(q, req->bio);
1473 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1474 !BIOVEC_VIRT_OVERSIZE(len)) {
1475 int mergeable = ll_new_mergeable(q, req, bio);
1476
1477 if (mergeable) {
1478 if (bio->bi_hw_segments == 1)
1479 bio->bi_hw_front_size = len;
1480 if (req->nr_hw_segments == 1)
1481 req->biotail->bi_hw_back_size = len;
1482 }
1483 return mergeable;
1484 }
1485
1486 return ll_new_hw_segment(q, req, bio);
1487}
1488
Jens Axboe165125e2007-07-24 09:28:11 +02001489static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 struct request *next)
1491{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001492 int total_phys_segments;
1493 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 /*
1496 * First check if the either of the requests are re-queued
1497 * requests. Can't merge them if they are.
1498 */
1499 if (req->special || next->special)
1500 return 0;
1501
1502 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001503 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 */
1505 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1506 return 0;
1507
1508 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1509 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1510 total_phys_segments--;
1511
1512 if (total_phys_segments > q->max_phys_segments)
1513 return 0;
1514
1515 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1516 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1517 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1518 /*
1519 * propagate the combined length to the end of the requests
1520 */
1521 if (req->nr_hw_segments == 1)
1522 req->bio->bi_hw_front_size = len;
1523 if (next->nr_hw_segments == 1)
1524 next->biotail->bi_hw_back_size = len;
1525 total_hw_segments--;
1526 }
1527
1528 if (total_hw_segments > q->max_hw_segments)
1529 return 0;
1530
1531 /* Merge is OK... */
1532 req->nr_phys_segments = total_phys_segments;
1533 req->nr_hw_segments = total_hw_segments;
1534 return 1;
1535}
1536
1537/*
1538 * "plug" the device if there are no outstanding requests: this will
1539 * force the transfer to start only after we have put all the requests
1540 * on the list.
1541 *
1542 * This is called with interrupts off and no requests on the queue and
1543 * with the queue lock held.
1544 */
Jens Axboe165125e2007-07-24 09:28:11 +02001545void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
1547 WARN_ON(!irqs_disabled());
1548
1549 /*
1550 * don't plug a stopped queue, it must be paired with blk_start_queue()
1551 * which will restart the queueing
1552 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001553 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 return;
1555
Jens Axboe2056a782006-03-23 20:00:26 +01001556 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001558 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
1562EXPORT_SYMBOL(blk_plug_device);
1563
1564/*
1565 * remove the queue from the plugged list, if present. called with
1566 * queue lock held and interrupts disabled.
1567 */
Jens Axboe165125e2007-07-24 09:28:11 +02001568int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
1570 WARN_ON(!irqs_disabled());
1571
1572 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1573 return 0;
1574
1575 del_timer(&q->unplug_timer);
1576 return 1;
1577}
1578
1579EXPORT_SYMBOL(blk_remove_plug);
1580
1581/*
1582 * remove the plug and let it rip..
1583 */
Jens Axboe165125e2007-07-24 09:28:11 +02001584void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001586 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 return;
1588
1589 if (!blk_remove_plug(q))
1590 return;
1591
Jens Axboe22e2c502005-06-27 10:55:12 +02001592 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593}
1594EXPORT_SYMBOL(__generic_unplug_device);
1595
1596/**
1597 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001598 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 *
1600 * Description:
1601 * Linux uses plugging to build bigger requests queues before letting
1602 * the device have at them. If a queue is plugged, the I/O scheduler
1603 * is still adding and merging requests on the queue. Once the queue
1604 * gets unplugged, the request_fn defined for the queue is invoked and
1605 * transfers started.
1606 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001607void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
1609 spin_lock_irq(q->queue_lock);
1610 __generic_unplug_device(q);
1611 spin_unlock_irq(q->queue_lock);
1612}
1613EXPORT_SYMBOL(generic_unplug_device);
1614
1615static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1616 struct page *page)
1617{
Jens Axboe165125e2007-07-24 09:28:11 +02001618 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620 /*
1621 * devices don't necessarily have an ->unplug_fn defined
1622 */
Jens Axboe2056a782006-03-23 20:00:26 +01001623 if (q->unplug_fn) {
1624 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1625 q->rq.count[READ] + q->rq.count[WRITE]);
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629}
1630
David Howells65f27f32006-11-22 14:55:48 +00001631static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
Jens Axboe165125e2007-07-24 09:28:11 +02001633 struct request_queue *q =
1634 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Jens Axboe2056a782006-03-23 20:00:26 +01001636 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1637 q->rq.count[READ] + q->rq.count[WRITE]);
1638
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 q->unplug_fn(q);
1640}
1641
1642static void blk_unplug_timeout(unsigned long data)
1643{
Jens Axboe165125e2007-07-24 09:28:11 +02001644 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
Jens Axboe2056a782006-03-23 20:00:26 +01001646 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1647 q->rq.count[READ] + q->rq.count[WRITE]);
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 kblockd_schedule_work(&q->unplug_work);
1650}
1651
1652/**
1653 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001654 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 *
1656 * Description:
1657 * blk_start_queue() will clear the stop flag on the queue, and call
1658 * the request_fn for the queue if it was in a stopped state when
1659 * entered. Also see blk_stop_queue(). Queue lock must be held.
1660 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001661void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001663 WARN_ON(!irqs_disabled());
1664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1666
1667 /*
1668 * one level of recursion is ok and is much faster than kicking
1669 * the unplug handling
1670 */
1671 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1672 q->request_fn(q);
1673 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1674 } else {
1675 blk_plug_device(q);
1676 kblockd_schedule_work(&q->unplug_work);
1677 }
1678}
1679
1680EXPORT_SYMBOL(blk_start_queue);
1681
1682/**
1683 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001684 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 *
1686 * Description:
1687 * The Linux block layer assumes that a block driver will consume all
1688 * entries on the request queue when the request_fn strategy is called.
1689 * Often this will not happen, because of hardware limitations (queue
1690 * depth settings). If a device driver gets a 'queue full' response,
1691 * or if it simply chooses not to queue more I/O at one point, it can
1692 * call this function to prevent the request_fn from being called until
1693 * the driver has signalled it's ready to go again. This happens by calling
1694 * blk_start_queue() to restart queue operations. Queue lock must be held.
1695 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001696void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697{
1698 blk_remove_plug(q);
1699 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1700}
1701EXPORT_SYMBOL(blk_stop_queue);
1702
1703/**
1704 * blk_sync_queue - cancel any pending callbacks on a queue
1705 * @q: the queue
1706 *
1707 * Description:
1708 * The block layer may perform asynchronous callback activity
1709 * on a queue, such as calling the unplug function after a timeout.
1710 * A block device may call blk_sync_queue to ensure that any
1711 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001712 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 * that its ->make_request_fn will not re-add plugging prior to calling
1714 * this function.
1715 *
1716 */
1717void blk_sync_queue(struct request_queue *q)
1718{
1719 del_timer_sync(&q->unplug_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720}
1721EXPORT_SYMBOL(blk_sync_queue);
1722
1723/**
1724 * blk_run_queue - run a single device queue
1725 * @q: The queue to run
1726 */
1727void blk_run_queue(struct request_queue *q)
1728{
1729 unsigned long flags;
1730
1731 spin_lock_irqsave(q->queue_lock, flags);
1732 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001733
1734 /*
1735 * Only recurse once to avoid overrunning the stack, let the unplug
1736 * handling reinvoke the handler shortly if we already got there.
1737 */
1738 if (!elv_queue_empty(q)) {
1739 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1740 q->request_fn(q);
1741 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1742 } else {
1743 blk_plug_device(q);
1744 kblockd_schedule_work(&q->unplug_work);
1745 }
1746 }
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 spin_unlock_irqrestore(q->queue_lock, flags);
1749}
1750EXPORT_SYMBOL(blk_run_queue);
1751
1752/**
Jens Axboe165125e2007-07-24 09:28:11 +02001753 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001754 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 *
1756 * Description:
1757 * blk_cleanup_queue is the pair to blk_init_queue() or
1758 * blk_queue_make_request(). It should be called when a request queue is
1759 * being released; typically when a block device is being de-registered.
1760 * Currently, its primary task it to free all the &struct request
1761 * structures that were allocated to the queue and the queue itself.
1762 *
1763 * Caveat:
1764 * Hopefully the low level driver will have finished any
1765 * outstanding requests first...
1766 **/
Al Viro483f4af2006-03-18 18:34:37 -05001767static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
Jens Axboe165125e2007-07-24 09:28:11 +02001769 struct request_queue *q =
1770 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 struct request_list *rl = &q->rq;
1772
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 blk_sync_queue(q);
1774
1775 if (rl->rq_pool)
1776 mempool_destroy(rl->rq_pool);
1777
1778 if (q->queue_tags)
1779 __blk_queue_free_tags(q);
1780
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001781 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 kmem_cache_free(requestq_cachep, q);
1784}
1785
Jens Axboe165125e2007-07-24 09:28:11 +02001786void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001787{
1788 kobject_put(&q->kobj);
1789}
1790EXPORT_SYMBOL(blk_put_queue);
1791
Jens Axboe165125e2007-07-24 09:28:11 +02001792void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001793{
1794 mutex_lock(&q->sysfs_lock);
1795 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1796 mutex_unlock(&q->sysfs_lock);
1797
1798 if (q->elevator)
1799 elevator_exit(q->elevator);
1800
1801 blk_put_queue(q);
1802}
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804EXPORT_SYMBOL(blk_cleanup_queue);
1805
Jens Axboe165125e2007-07-24 09:28:11 +02001806static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807{
1808 struct request_list *rl = &q->rq;
1809
1810 rl->count[READ] = rl->count[WRITE] = 0;
1811 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001812 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 init_waitqueue_head(&rl->wait[READ]);
1814 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Christoph Lameter19460892005-06-23 00:08:19 -07001816 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1817 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819 if (!rl->rq_pool)
1820 return -ENOMEM;
1821
1822 return 0;
1823}
1824
Jens Axboe165125e2007-07-24 09:28:11 +02001825struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826{
Christoph Lameter19460892005-06-23 00:08:19 -07001827 return blk_alloc_queue_node(gfp_mask, -1);
1828}
1829EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Al Viro483f4af2006-03-18 18:34:37 -05001831static struct kobj_type queue_ktype;
1832
Jens Axboe165125e2007-07-24 09:28:11 +02001833struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001834{
Jens Axboe165125e2007-07-24 09:28:11 +02001835 struct request_queue *q;
Christoph Lameter19460892005-06-23 00:08:19 -07001836
Christoph Lameter94f60302007-07-17 04:03:29 -07001837 q = kmem_cache_alloc_node(requestq_cachep,
1838 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 if (!q)
1840 return NULL;
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001843
1844 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1845 q->kobj.ktype = &queue_ktype;
1846 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1849 q->backing_dev_info.unplug_io_data = q;
1850
Al Viro483f4af2006-03-18 18:34:37 -05001851 mutex_init(&q->sysfs_lock);
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 return q;
1854}
Christoph Lameter19460892005-06-23 00:08:19 -07001855EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
1857/**
1858 * blk_init_queue - prepare a request queue for use with a block device
1859 * @rfn: The function to be called to process requests that have been
1860 * placed on the queue.
1861 * @lock: Request queue spin lock
1862 *
1863 * Description:
1864 * If a block device wishes to use the standard request handling procedures,
1865 * which sorts requests and coalesces adjacent requests, then it must
1866 * call blk_init_queue(). The function @rfn will be called when there
1867 * are requests on the queue that need to be processed. If the device
1868 * supports plugging, then @rfn may not be called immediately when requests
1869 * are available on the queue, but may be called at some time later instead.
1870 * Plugged queues are generally unplugged when a buffer belonging to one
1871 * of the requests on the queue is needed, or due to memory pressure.
1872 *
1873 * @rfn is not required, or even expected, to remove all requests off the
1874 * queue, but only as many as it can handle at a time. If it does leave
1875 * requests on the queue, it is responsible for arranging that the requests
1876 * get dealt with eventually.
1877 *
1878 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001879 * request queue; this lock will be taken also from interrupt context, so irq
1880 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 *
1882 * Function returns a pointer to the initialized request queue, or NULL if
1883 * it didn't succeed.
1884 *
1885 * Note:
1886 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1887 * when the block device is deactivated (such as at module unload).
1888 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001889
Jens Axboe165125e2007-07-24 09:28:11 +02001890struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891{
Christoph Lameter19460892005-06-23 00:08:19 -07001892 return blk_init_queue_node(rfn, lock, -1);
1893}
1894EXPORT_SYMBOL(blk_init_queue);
1895
Jens Axboe165125e2007-07-24 09:28:11 +02001896struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001897blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1898{
Jens Axboe165125e2007-07-24 09:28:11 +02001899 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
1901 if (!q)
1902 return NULL;
1903
Christoph Lameter19460892005-06-23 00:08:19 -07001904 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001905 if (blk_init_free_list(q)) {
1906 kmem_cache_free(requestq_cachep, q);
1907 return NULL;
1908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
152587d2005-04-12 16:22:06 -05001910 /*
1911 * if caller didn't supply a lock, they get per-queue locking with
1912 * our embedded lock
1913 */
1914 if (!lock) {
1915 spin_lock_init(&q->__queue_lock);
1916 lock = &q->__queue_lock;
1917 }
1918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 q->prep_rq_fn = NULL;
1921 q->unplug_fn = generic_unplug_device;
1922 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1923 q->queue_lock = lock;
1924
1925 blk_queue_segment_boundary(q, 0xffffffff);
1926
1927 blk_queue_make_request(q, __make_request);
1928 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1929
1930 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1931 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1932
Alan Stern44ec9542007-02-20 11:01:57 -05001933 q->sg_reserved_size = INT_MAX;
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 /*
1936 * all done
1937 */
1938 if (!elevator_init(q, NULL)) {
1939 blk_queue_congestion_threshold(q);
1940 return q;
1941 }
1942
Al Viro8669aaf2006-03-18 13:50:00 -05001943 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 return NULL;
1945}
Christoph Lameter19460892005-06-23 00:08:19 -07001946EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Jens Axboe165125e2007-07-24 09:28:11 +02001948int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001950 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001951 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 return 0;
1953 }
1954
1955 return 1;
1956}
1957
1958EXPORT_SYMBOL(blk_get_queue);
1959
Jens Axboe165125e2007-07-24 09:28:11 +02001960static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001962 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001963 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 mempool_free(rq, q->rq.rq_pool);
1965}
1966
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001967static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001968blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
1970 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1971
1972 if (!rq)
1973 return NULL;
1974
1975 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001976 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 * see bio.h and blkdev.h
1978 */
Jens Axboe49171e52006-08-10 08:59:11 +02001979 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Tejun Heocb98fc82005-10-28 08:29:39 +02001981 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001982 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001983 mempool_free(rq, q->rq.rq_pool);
1984 return NULL;
1985 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02001986 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02001987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Tejun Heocb98fc82005-10-28 08:29:39 +02001989 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990}
1991
1992/*
1993 * ioc_batching returns true if the ioc is a valid batching request and
1994 * should be given priority access to a request.
1995 */
Jens Axboe165125e2007-07-24 09:28:11 +02001996static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997{
1998 if (!ioc)
1999 return 0;
2000
2001 /*
2002 * Make sure the process is able to allocate at least 1 request
2003 * even if the batch times out, otherwise we could theoretically
2004 * lose wakeups.
2005 */
2006 return ioc->nr_batch_requests == q->nr_batching ||
2007 (ioc->nr_batch_requests > 0
2008 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2009}
2010
2011/*
2012 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2013 * will cause the process to be a "batcher" on all queues in the system. This
2014 * is the behaviour we want though - once it gets a wakeup it should be given
2015 * a nice run.
2016 */
Jens Axboe165125e2007-07-24 09:28:11 +02002017static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018{
2019 if (!ioc || ioc_batching(q, ioc))
2020 return;
2021
2022 ioc->nr_batch_requests = q->nr_batching;
2023 ioc->last_waited = jiffies;
2024}
2025
Jens Axboe165125e2007-07-24 09:28:11 +02002026static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
2028 struct request_list *rl = &q->rq;
2029
2030 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002031 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 if (waitqueue_active(&rl->wait[rw]))
2035 wake_up(&rl->wait[rw]);
2036
2037 blk_clear_queue_full(q, rw);
2038 }
2039}
2040
2041/*
2042 * A request has just been released. Account for it, update the full and
2043 * congestion status, wake up any waiters. Called under q->queue_lock.
2044 */
Jens Axboe165125e2007-07-24 09:28:11 +02002045static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046{
2047 struct request_list *rl = &q->rq;
2048
2049 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002050 if (priv)
2051 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
2053 __freed_request(q, rw);
2054
2055 if (unlikely(rl->starved[rw ^ 1]))
2056 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057}
2058
2059#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2060/*
Nick Piggind6344532005-06-28 20:45:14 -07002061 * Get a free request, queue_lock must be held.
2062 * Returns NULL on failure, with queue_lock held.
2063 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 */
Jens Axboe165125e2007-07-24 09:28:11 +02002065static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002066 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067{
2068 struct request *rq = NULL;
2069 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002070 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002071 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002072 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Jens Axboe7749a8d2006-12-13 13:02:26 +01002074 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002075 if (may_queue == ELV_MQUEUE_NO)
2076 goto rq_starved;
2077
2078 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2079 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002080 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002081 /*
2082 * The queue will fill after this allocation, so set
2083 * it as full, and mark this process as "batching".
2084 * This process will be allowed to complete a batch of
2085 * requests, others will be blocked.
2086 */
2087 if (!blk_queue_full(q, rw)) {
2088 ioc_set_batching(q, ioc);
2089 blk_set_queue_full(q, rw);
2090 } else {
2091 if (may_queue != ELV_MQUEUE_MUST
2092 && !ioc_batching(q, ioc)) {
2093 /*
2094 * The queue is full and the allocating
2095 * process is not a "batcher", and not
2096 * exempted by the IO scheduler
2097 */
2098 goto out;
2099 }
2100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002102 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
2104
Jens Axboe082cf692005-06-28 16:35:11 +02002105 /*
2106 * Only allow batching queuers to allocate up to 50% over the defined
2107 * limit of requests, otherwise we could have thousands of requests
2108 * allocated with any setting of ->nr_requests
2109 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002110 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002111 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002112
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 rl->count[rw]++;
2114 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002115
Jens Axboe64521d12005-10-28 08:30:39 +02002116 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002117 if (priv)
2118 rl->elvpriv++;
2119
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 spin_unlock_irq(q->queue_lock);
2121
Jens Axboe7749a8d2006-12-13 13:02:26 +01002122 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002123 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 /*
2125 * Allocation failed presumably due to memory. Undo anything
2126 * we might have messed up.
2127 *
2128 * Allocating task should really be put onto the front of the
2129 * wait queue, but this is pretty rare.
2130 */
2131 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002132 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
2134 /*
2135 * in the very unlikely event that allocation failed and no
2136 * requests for this direction was pending, mark us starved
2137 * so that freeing of a request in the other direction will
2138 * notice us. another possible fix would be to split the
2139 * rq mempool into READ and WRITE
2140 */
2141rq_starved:
2142 if (unlikely(rl->count[rw] == 0))
2143 rl->starved[rw] = 1;
2144
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 goto out;
2146 }
2147
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002148 /*
2149 * ioc may be NULL here, and ioc_batching will be false. That's
2150 * OK, if the queue is under the request limit then requests need
2151 * not count toward the nr_batch_requests limit. There will always
2152 * be some limit enforced by BLK_BATCH_TIME.
2153 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 if (ioc_batching(q, ioc))
2155 ioc->nr_batch_requests--;
2156
2157 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002158
2159 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 return rq;
2162}
2163
2164/*
2165 * No available requests for this queue, unplug the device and wait for some
2166 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002167 *
2168 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 */
Jens Axboe165125e2007-07-24 09:28:11 +02002170static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002171 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002173 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 struct request *rq;
2175
Jens Axboe7749a8d2006-12-13 13:02:26 +01002176 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002177 while (!rq) {
2178 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 struct request_list *rl = &q->rq;
2180
2181 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2182 TASK_UNINTERRUPTIBLE);
2183
Jens Axboe7749a8d2006-12-13 13:02:26 +01002184 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185
2186 if (!rq) {
2187 struct io_context *ioc;
2188
Jens Axboe2056a782006-03-23 20:00:26 +01002189 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2190
Nick Piggind6344532005-06-28 20:45:14 -07002191 __generic_unplug_device(q);
2192 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 io_schedule();
2194
2195 /*
2196 * After sleeping, we become a "batching" process and
2197 * will be able to allocate at least one request, and
2198 * up to a big batch of them for a small period time.
2199 * See ioc_batching, ioc_set_batching
2200 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002201 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002203
2204 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 }
2206 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208
2209 return rq;
2210}
2211
Jens Axboe165125e2007-07-24 09:28:11 +02002212struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213{
2214 struct request *rq;
2215
2216 BUG_ON(rw != READ && rw != WRITE);
2217
Nick Piggind6344532005-06-28 20:45:14 -07002218 spin_lock_irq(q->queue_lock);
2219 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002220 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002221 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002222 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002223 if (!rq)
2224 spin_unlock_irq(q->queue_lock);
2225 }
2226 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228 return rq;
2229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230EXPORT_SYMBOL(blk_get_request);
2231
2232/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002233 * blk_start_queueing - initiate dispatch of requests to device
2234 * @q: request queue to kick into gear
2235 *
2236 * This is basically a helper to remove the need to know whether a queue
2237 * is plugged or not if someone just wants to initiate dispatch of requests
2238 * for this queue.
2239 *
2240 * The queue lock must be held with interrupts disabled.
2241 */
Jens Axboe165125e2007-07-24 09:28:11 +02002242void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002243{
2244 if (!blk_queue_plugged(q))
2245 q->request_fn(q);
2246 else
2247 __generic_unplug_device(q);
2248}
2249EXPORT_SYMBOL(blk_start_queueing);
2250
2251/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 * blk_requeue_request - put a request back on queue
2253 * @q: request queue where request should be inserted
2254 * @rq: request to be inserted
2255 *
2256 * Description:
2257 * Drivers often keep queueing requests until the hardware cannot accept
2258 * more, when that condition happens we need to put the request back
2259 * on the queue. Must be called with queue lock held.
2260 */
Jens Axboe165125e2007-07-24 09:28:11 +02002261void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
Jens Axboe2056a782006-03-23 20:00:26 +01002263 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2264
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 if (blk_rq_tagged(rq))
2266 blk_queue_end_tag(q, rq);
2267
2268 elv_requeue_request(q, rq);
2269}
2270
2271EXPORT_SYMBOL(blk_requeue_request);
2272
2273/**
2274 * blk_insert_request - insert a special request in to a request queue
2275 * @q: request queue where request should be inserted
2276 * @rq: request to be inserted
2277 * @at_head: insert request at head or tail of queue
2278 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 *
2280 * Description:
2281 * Many block devices need to execute commands asynchronously, so they don't
2282 * block the whole kernel from preemption during request execution. This is
2283 * accomplished normally by inserting aritficial requests tagged as
2284 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2285 * scheduled for actual execution by the request queue.
2286 *
2287 * We have the option of inserting the head or the tail of the queue.
2288 * Typically we use the tail for new ioctls and so forth. We use the head
2289 * of the queue for things like a QUEUE_FULL message from a device, or a
2290 * host that is unable to accept a particular command.
2291 */
Jens Axboe165125e2007-07-24 09:28:11 +02002292void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002293 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294{
Tejun Heo 867d1192005-04-24 02:06:05 -05002295 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 unsigned long flags;
2297
2298 /*
2299 * tell I/O scheduler that this isn't a regular read/write (ie it
2300 * must not attempt merges on this) and that it acts as a soft
2301 * barrier
2302 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002303 rq->cmd_type = REQ_TYPE_SPECIAL;
2304 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305
2306 rq->special = data;
2307
2308 spin_lock_irqsave(q->queue_lock, flags);
2309
2310 /*
2311 * If command is tagged, release the tag
2312 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002313 if (blk_rq_tagged(rq))
2314 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
Tejun Heo 867d1192005-04-24 02:06:05 -05002316 drive_stat_acct(rq, rq->nr_sectors, 1);
2317 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002318 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 spin_unlock_irqrestore(q->queue_lock, flags);
2320}
2321
2322EXPORT_SYMBOL(blk_insert_request);
2323
Mike Christie0e75f902006-12-01 10:40:55 +01002324static int __blk_rq_unmap_user(struct bio *bio)
2325{
2326 int ret = 0;
2327
2328 if (bio) {
2329 if (bio_flagged(bio, BIO_USER_MAPPED))
2330 bio_unmap_user(bio);
2331 else
2332 ret = bio_uncopy_user(bio);
2333 }
2334
2335 return ret;
2336}
2337
Jens Axboe165125e2007-07-24 09:28:11 +02002338static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002339 void __user *ubuf, unsigned int len)
2340{
2341 unsigned long uaddr;
2342 struct bio *bio, *orig_bio;
2343 int reading, ret;
2344
2345 reading = rq_data_dir(rq) == READ;
2346
2347 /*
2348 * if alignment requirement is satisfied, map in user pages for
2349 * direct dma. else, set up kernel bounce buffers
2350 */
2351 uaddr = (unsigned long) ubuf;
2352 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2353 bio = bio_map_user(q, NULL, uaddr, len, reading);
2354 else
2355 bio = bio_copy_user(q, uaddr, len, reading);
2356
Jens Axboe29852592006-12-19 08:27:31 +01002357 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002358 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002359
2360 orig_bio = bio;
2361 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002362
Mike Christie0e75f902006-12-01 10:40:55 +01002363 /*
2364 * We link the bounce buffer in and could have to traverse it
2365 * later so we have to get a ref to prevent it from being freed
2366 */
2367 bio_get(bio);
2368
Mike Christie0e75f902006-12-01 10:40:55 +01002369 if (!rq->bio)
2370 blk_rq_bio_prep(q, rq, bio);
Jens Axboe1aa4f242006-12-19 08:33:11 +01002371 else if (!ll_back_merge_fn(q, rq, bio)) {
Mike Christie0e75f902006-12-01 10:40:55 +01002372 ret = -EINVAL;
Mike Christie0e75f902006-12-01 10:40:55 +01002373 goto unmap_bio;
2374 } else {
2375 rq->biotail->bi_next = bio;
2376 rq->biotail = bio;
2377
Mike Christie0e75f902006-12-01 10:40:55 +01002378 rq->data_len += bio->bi_size;
2379 }
Mike Christie0e75f902006-12-01 10:40:55 +01002380
2381 return bio->bi_size;
2382
2383unmap_bio:
2384 /* if it was boucned we must call the end io function */
2385 bio_endio(bio, bio->bi_size, 0);
2386 __blk_rq_unmap_user(orig_bio);
2387 bio_put(bio);
2388 return ret;
2389}
2390
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391/**
2392 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2393 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002394 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 * @ubuf: the user buffer
2396 * @len: length of user data
2397 *
2398 * Description:
2399 * Data will be mapped directly for zero copy io, if possible. Otherwise
2400 * a kernel bounce buffer is used.
2401 *
2402 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2403 * still in process context.
2404 *
2405 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2406 * before being submitted to the device, as pages mapped may be out of
2407 * reach. It's the callers responsibility to make sure this happens. The
2408 * original bio must be passed back in to blk_rq_unmap_user() for proper
2409 * unmapping.
2410 */
Jens Axboe165125e2007-07-24 09:28:11 +02002411int blk_rq_map_user(struct request_queue *q, struct request *rq,
2412 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413{
Mike Christie0e75f902006-12-01 10:40:55 +01002414 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002415 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002416 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
Mike Christiedefd94b2005-12-05 02:37:06 -06002418 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002419 return -EINVAL;
2420 if (!len || !ubuf)
2421 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Mike Christie0e75f902006-12-01 10:40:55 +01002423 while (bytes_read != len) {
2424 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Mike Christie0e75f902006-12-01 10:40:55 +01002426 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2427 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2428 >> PAGE_SHIFT;
2429 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Mike Christie0e75f902006-12-01 10:40:55 +01002431 /*
2432 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2433 * pages. If this happens we just lower the requested
2434 * mapping len by a page so that we can fit
2435 */
2436 if (end - start > BIO_MAX_PAGES)
2437 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Mike Christie0e75f902006-12-01 10:40:55 +01002439 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2440 if (ret < 0)
2441 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002442 if (!bio)
2443 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002444 bytes_read += ret;
2445 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 }
2447
Mike Christie0e75f902006-12-01 10:40:55 +01002448 rq->buffer = rq->data = NULL;
2449 return 0;
2450unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002451 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002452 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453}
2454
2455EXPORT_SYMBOL(blk_rq_map_user);
2456
2457/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002458 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2459 * @q: request queue where request should be inserted
2460 * @rq: request to map data to
2461 * @iov: pointer to the iovec
2462 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002463 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002464 *
2465 * Description:
2466 * Data will be mapped directly for zero copy io, if possible. Otherwise
2467 * a kernel bounce buffer is used.
2468 *
2469 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2470 * still in process context.
2471 *
2472 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2473 * before being submitted to the device, as pages mapped may be out of
2474 * reach. It's the callers responsibility to make sure this happens. The
2475 * original bio must be passed back in to blk_rq_unmap_user() for proper
2476 * unmapping.
2477 */
Jens Axboe165125e2007-07-24 09:28:11 +02002478int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002479 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002480{
2481 struct bio *bio;
2482
2483 if (!iov || iov_count <= 0)
2484 return -EINVAL;
2485
2486 /* we don't allow misaligned data like bio_map_user() does. If the
2487 * user is using sg, they're expected to know the alignment constraints
2488 * and respect them accordingly */
2489 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2490 if (IS_ERR(bio))
2491 return PTR_ERR(bio);
2492
Mike Christie0e75f902006-12-01 10:40:55 +01002493 if (bio->bi_size != len) {
2494 bio_endio(bio, bio->bi_size, 0);
2495 bio_unmap_user(bio);
2496 return -EINVAL;
2497 }
2498
2499 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002500 blk_rq_bio_prep(q, rq, bio);
2501 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002502 return 0;
2503}
2504
2505EXPORT_SYMBOL(blk_rq_map_user_iov);
2506
2507/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002509 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 *
2511 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002512 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2513 * supply the original rq->bio from the blk_rq_map_user() return, since
2514 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002516int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002518 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002519 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002521 while (bio) {
2522 mapped_bio = bio;
2523 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002524 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Jens Axboe48785bb2006-12-19 11:07:59 +01002526 ret2 = __blk_rq_unmap_user(mapped_bio);
2527 if (ret2 && !ret)
2528 ret = ret2;
2529
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002530 mapped_bio = bio;
2531 bio = bio->bi_next;
2532 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002533 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002534
2535 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536}
2537
2538EXPORT_SYMBOL(blk_rq_unmap_user);
2539
2540/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002541 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2542 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002543 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002544 * @kbuf: the kernel buffer
2545 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002546 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002547 */
Jens Axboe165125e2007-07-24 09:28:11 +02002548int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002549 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002550{
Mike Christie df46b9a2005-06-20 14:04:44 +02002551 struct bio *bio;
2552
Mike Christiedefd94b2005-12-05 02:37:06 -06002553 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002554 return -EINVAL;
2555 if (!len || !kbuf)
2556 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002557
2558 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002559 if (IS_ERR(bio))
2560 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002561
Jens Axboedd1cab92005-06-20 14:06:01 +02002562 if (rq_data_dir(rq) == WRITE)
2563 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002564
Jens Axboedd1cab92005-06-20 14:06:01 +02002565 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002566 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002567 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002568 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002569}
2570
2571EXPORT_SYMBOL(blk_rq_map_kern);
2572
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002573/**
2574 * blk_execute_rq_nowait - insert a request into queue for execution
2575 * @q: queue to insert the request in
2576 * @bd_disk: matching gendisk
2577 * @rq: request to insert
2578 * @at_head: insert request at head or tail of queue
2579 * @done: I/O completion handler
2580 *
2581 * Description:
2582 * Insert a fully prepared request at the back of the io scheduler queue
2583 * for execution. Don't wait for completion.
2584 */
Jens Axboe165125e2007-07-24 09:28:11 +02002585void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002586 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002587 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002588{
2589 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2590
2591 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002592 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002593 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002594 WARN_ON(irqs_disabled());
2595 spin_lock_irq(q->queue_lock);
2596 __elv_add_request(q, rq, where, 1);
2597 __generic_unplug_device(q);
2598 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002599}
Mike Christie6e39b69e2005-11-11 05:30:24 -06002600EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2601
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602/**
2603 * blk_execute_rq - insert a request into queue for execution
2604 * @q: queue to insert the request in
2605 * @bd_disk: matching gendisk
2606 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002607 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 *
2609 * Description:
2610 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002611 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 */
Jens Axboe165125e2007-07-24 09:28:11 +02002613int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002614 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002616 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 char sense[SCSI_SENSE_BUFFERSIZE];
2618 int err = 0;
2619
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 /*
2621 * we need an extra reference to the request, so we can look at
2622 * it after io completion
2623 */
2624 rq->ref_count++;
2625
2626 if (!rq->sense) {
2627 memset(sense, 0, sizeof(sense));
2628 rq->sense = sense;
2629 rq->sense_len = 0;
2630 }
2631
Jens Axboec00895a2006-09-30 20:29:12 +02002632 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002633 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
2636 if (rq->errors)
2637 err = -EIO;
2638
2639 return err;
2640}
2641
2642EXPORT_SYMBOL(blk_execute_rq);
2643
2644/**
2645 * blkdev_issue_flush - queue a flush
2646 * @bdev: blockdev to issue flush for
2647 * @error_sector: error sector
2648 *
2649 * Description:
2650 * Issue a flush for the block device in question. Caller can supply
2651 * room for storing the error offset in case of a flush error, if they
2652 * wish to. Caller must run wait_for_completion() on its own.
2653 */
2654int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2655{
Jens Axboe165125e2007-07-24 09:28:11 +02002656 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
2658 if (bdev->bd_disk == NULL)
2659 return -ENXIO;
2660
2661 q = bdev_get_queue(bdev);
2662 if (!q)
2663 return -ENXIO;
2664 if (!q->issue_flush_fn)
2665 return -EOPNOTSUPP;
2666
2667 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2668}
2669
2670EXPORT_SYMBOL(blkdev_issue_flush);
2671
Adrian Bunk93d17d32005-06-25 14:59:10 -07002672static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673{
2674 int rw = rq_data_dir(rq);
2675
2676 if (!blk_fs_request(rq) || !rq->rq_disk)
2677 return;
2678
Jens Axboed72d9042005-11-01 08:35:42 +01002679 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002680 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002681 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 disk_round_stats(rq->rq_disk);
2683 rq->rq_disk->in_flight++;
2684 }
2685}
2686
2687/*
2688 * add-request adds a request to the linked list.
2689 * queue lock is held and interrupts disabled, as we muck with the
2690 * request queue list.
2691 */
Jens Axboe165125e2007-07-24 09:28:11 +02002692static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693{
2694 drive_stat_acct(req, req->nr_sectors, 1);
2695
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 /*
2697 * elevator indicated where it wants this request to be
2698 * inserted at elevator_merge time
2699 */
2700 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2701}
2702
2703/*
2704 * disk_round_stats() - Round off the performance stats on a struct
2705 * disk_stats.
2706 *
2707 * The average IO queue length and utilisation statistics are maintained
2708 * by observing the current state of the queue length and the amount of
2709 * time it has been in this state for.
2710 *
2711 * Normally, that accounting is done on IO completion, but that can result
2712 * in more than a second's worth of IO being accounted for within any one
2713 * second, leading to >100% utilisation. To deal with that, we call this
2714 * function to do a round-off before returning the results when reading
2715 * /proc/diskstats. This accounts immediately for all queue usage up to
2716 * the current jiffies and restarts the counters again.
2717 */
2718void disk_round_stats(struct gendisk *disk)
2719{
2720 unsigned long now = jiffies;
2721
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002722 if (now == disk->stamp)
2723 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002725 if (disk->in_flight) {
2726 __disk_stat_add(disk, time_in_queue,
2727 disk->in_flight * (now - disk->stamp));
2728 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731}
2732
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002733EXPORT_SYMBOL_GPL(disk_round_stats);
2734
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735/*
2736 * queue lock must be held
2737 */
Jens Axboe165125e2007-07-24 09:28:11 +02002738void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 if (unlikely(!q))
2741 return;
2742 if (unlikely(--req->ref_count))
2743 return;
2744
Tejun Heo8922e162005-10-20 16:23:44 +02002745 elv_completed_request(q, req);
2746
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 /*
2748 * Request may not have originated from ll_rw_blk. if not,
2749 * it didn't come out of our reserved rq pools
2750 */
Jens Axboe49171e52006-08-10 08:59:11 +02002751 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002753 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002756 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
2758 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002759 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 }
2761}
2762
Mike Christie6e39b69e2005-11-11 05:30:24 -06002763EXPORT_SYMBOL_GPL(__blk_put_request);
2764
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765void blk_put_request(struct request *req)
2766{
Tejun Heo8922e162005-10-20 16:23:44 +02002767 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002768 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769
Tejun Heo8922e162005-10-20 16:23:44 +02002770 /*
2771 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2772 * following if (q) test.
2773 */
2774 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 spin_lock_irqsave(q->queue_lock, flags);
2776 __blk_put_request(q, req);
2777 spin_unlock_irqrestore(q->queue_lock, flags);
2778 }
2779}
2780
2781EXPORT_SYMBOL(blk_put_request);
2782
2783/**
2784 * blk_end_sync_rq - executes a completion event on a request
2785 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002786 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002788void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789{
Jens Axboec00895a2006-09-30 20:29:12 +02002790 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Jens Axboec00895a2006-09-30 20:29:12 +02002792 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 __blk_put_request(rq->q, rq);
2794
2795 /*
2796 * complete last, if this is a stack request the process (and thus
2797 * the rq pointer) could be invalid right after this complete()
2798 */
2799 complete(waiting);
2800}
2801EXPORT_SYMBOL(blk_end_sync_rq);
2802
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803/*
2804 * Has to be called with the request spinlock acquired
2805 */
Jens Axboe165125e2007-07-24 09:28:11 +02002806static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 struct request *next)
2808{
2809 if (!rq_mergeable(req) || !rq_mergeable(next))
2810 return 0;
2811
2812 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002813 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 */
2815 if (req->sector + req->nr_sectors != next->sector)
2816 return 0;
2817
2818 if (rq_data_dir(req) != rq_data_dir(next)
2819 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002820 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 return 0;
2822
2823 /*
2824 * If we are allowed to merge, then append bio list
2825 * from next to rq and release next. merge_requests_fn
2826 * will have updated segment counts, update sector
2827 * counts here.
2828 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002829 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 return 0;
2831
2832 /*
2833 * At this point we have either done a back merge
2834 * or front merge. We need the smaller start_time of
2835 * the merged requests to be the current request
2836 * for accounting purposes.
2837 */
2838 if (time_after(req->start_time, next->start_time))
2839 req->start_time = next->start_time;
2840
2841 req->biotail->bi_next = next->bio;
2842 req->biotail = next->biotail;
2843
2844 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2845
2846 elv_merge_requests(q, req, next);
2847
2848 if (req->rq_disk) {
2849 disk_round_stats(req->rq_disk);
2850 req->rq_disk->in_flight--;
2851 }
2852
Jens Axboe22e2c502005-06-27 10:55:12 +02002853 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2854
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 __blk_put_request(q, next);
2856 return 1;
2857}
2858
Jens Axboe165125e2007-07-24 09:28:11 +02002859static inline int attempt_back_merge(struct request_queue *q,
2860 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861{
2862 struct request *next = elv_latter_request(q, rq);
2863
2864 if (next)
2865 return attempt_merge(q, rq, next);
2866
2867 return 0;
2868}
2869
Jens Axboe165125e2007-07-24 09:28:11 +02002870static inline int attempt_front_merge(struct request_queue *q,
2871 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872{
2873 struct request *prev = elv_former_request(q, rq);
2874
2875 if (prev)
2876 return attempt_merge(q, prev, rq);
2877
2878 return 0;
2879}
2880
Tejun Heo52d9e672006-01-06 09:49:58 +01002881static void init_request_from_bio(struct request *req, struct bio *bio)
2882{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002883 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002884
2885 /*
2886 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2887 */
2888 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002889 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002890
2891 /*
2892 * REQ_BARRIER implies no merging, but lets make it explicit
2893 */
2894 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002895 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002896
Jens Axboeb31dc662006-06-13 08:26:10 +02002897 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002898 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002899 if (bio_rw_meta(bio))
2900 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002901
Tejun Heo52d9e672006-01-06 09:49:58 +01002902 req->errors = 0;
2903 req->hard_sector = req->sector = bio->bi_sector;
2904 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2905 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2906 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2907 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2908 req->buffer = bio_data(bio); /* see ->buffer comment above */
Tejun Heo52d9e672006-01-06 09:49:58 +01002909 req->bio = req->biotail = bio;
2910 req->ioprio = bio_prio(bio);
2911 req->rq_disk = bio->bi_bdev->bd_disk;
2912 req->start_time = jiffies;
2913}
2914
Jens Axboe165125e2007-07-24 09:28:11 +02002915static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916{
Nick Piggin450991b2005-06-28 20:45:13 -07002917 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002918 int el_ret, nr_sectors, barrier, err;
2919 const unsigned short prio = bio_prio(bio);
2920 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002921 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
2925 /*
2926 * low level driver can indicate that it wants pages above a
2927 * certain limit bounced to low memory (ie for highmem, or even
2928 * ISA dma in theory)
2929 */
2930 blk_queue_bounce(q, &bio);
2931
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002933 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 err = -EOPNOTSUPP;
2935 goto end_io;
2936 }
2937
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 spin_lock_irq(q->queue_lock);
2939
Nick Piggin450991b2005-06-28 20:45:13 -07002940 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 goto get_rq;
2942
2943 el_ret = elv_merge(q, &req, bio);
2944 switch (el_ret) {
2945 case ELEVATOR_BACK_MERGE:
2946 BUG_ON(!rq_mergeable(req));
2947
Jens Axboe1aa4f242006-12-19 08:33:11 +01002948 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 break;
2950
Jens Axboe2056a782006-03-23 20:00:26 +01002951 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2952
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 req->biotail->bi_next = bio;
2954 req->biotail = bio;
2955 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002956 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957 drive_stat_acct(req, nr_sectors, 0);
2958 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002959 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 goto out;
2961
2962 case ELEVATOR_FRONT_MERGE:
2963 BUG_ON(!rq_mergeable(req));
2964
Jens Axboe1aa4f242006-12-19 08:33:11 +01002965 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 break;
2967
Jens Axboe2056a782006-03-23 20:00:26 +01002968 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2969
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 bio->bi_next = req->bio;
2971 req->bio = bio;
2972
2973 /*
2974 * may not be valid. if the low level driver said
2975 * it didn't need a bounce buffer then it better
2976 * not touch req->buffer either...
2977 */
2978 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02002979 req->current_nr_sectors = bio_cur_sectors(bio);
2980 req->hard_cur_sectors = req->current_nr_sectors;
2981 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002983 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 drive_stat_acct(req, nr_sectors, 0);
2985 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002986 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 goto out;
2988
Nick Piggin450991b2005-06-28 20:45:13 -07002989 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002991 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 }
2993
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002995 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01002996 * This sync check and mask will be re-done in init_request_from_bio(),
2997 * but we need to set it earlier to expose the sync flag to the
2998 * rq allocator and io schedulers.
2999 */
3000 rw_flags = bio_data_dir(bio);
3001 if (sync)
3002 rw_flags |= REQ_RW_SYNC;
3003
3004 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003005 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003006 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003007 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003008 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003009
Nick Piggin450991b2005-06-28 20:45:13 -07003010 /*
3011 * After dropping the lock and possibly sleeping here, our request
3012 * may now be mergeable after it had proven unmergeable (above).
3013 * We don't worry about that case for efficiency. It won't happen
3014 * often, and the elevators are able to handle it.
3015 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003016 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Nick Piggin450991b2005-06-28 20:45:13 -07003018 spin_lock_irq(q->queue_lock);
3019 if (elv_queue_empty(q))
3020 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 add_request(q, req);
3022out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003023 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 __generic_unplug_device(q);
3025
3026 spin_unlock_irq(q->queue_lock);
3027 return 0;
3028
3029end_io:
3030 bio_endio(bio, nr_sectors << 9, err);
3031 return 0;
3032}
3033
3034/*
3035 * If bio->bi_dev is a partition, remap the location
3036 */
3037static inline void blk_partition_remap(struct bio *bio)
3038{
3039 struct block_device *bdev = bio->bi_bdev;
3040
3041 if (bdev != bdev->bd_contains) {
3042 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003043 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
Jens Axboea3623572005-11-01 09:26:16 +01003045 p->sectors[rw] += bio_sectors(bio);
3046 p->ios[rw]++;
3047
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 bio->bi_sector += p->start_sect;
3049 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003050
3051 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3052 bdev->bd_dev, bio->bi_sector,
3053 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 }
3055}
3056
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057static void handle_bad_sector(struct bio *bio)
3058{
3059 char b[BDEVNAME_SIZE];
3060
3061 printk(KERN_INFO "attempt to access beyond end of device\n");
3062 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3063 bdevname(bio->bi_bdev, b),
3064 bio->bi_rw,
3065 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3066 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3067
3068 set_bit(BIO_EOF, &bio->bi_flags);
3069}
3070
Akinobu Mitac17bb492006-12-08 02:39:46 -08003071#ifdef CONFIG_FAIL_MAKE_REQUEST
3072
3073static DECLARE_FAULT_ATTR(fail_make_request);
3074
3075static int __init setup_fail_make_request(char *str)
3076{
3077 return setup_fault_attr(&fail_make_request, str);
3078}
3079__setup("fail_make_request=", setup_fail_make_request);
3080
3081static int should_fail_request(struct bio *bio)
3082{
3083 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3084 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3085 return should_fail(&fail_make_request, bio->bi_size);
3086
3087 return 0;
3088}
3089
3090static int __init fail_make_request_debugfs(void)
3091{
3092 return init_fault_attr_dentries(&fail_make_request,
3093 "fail_make_request");
3094}
3095
3096late_initcall(fail_make_request_debugfs);
3097
3098#else /* CONFIG_FAIL_MAKE_REQUEST */
3099
3100static inline int should_fail_request(struct bio *bio)
3101{
3102 return 0;
3103}
3104
3105#endif /* CONFIG_FAIL_MAKE_REQUEST */
3106
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107/**
3108 * generic_make_request: hand a buffer to its device driver for I/O
3109 * @bio: The bio describing the location in memory and on the device.
3110 *
3111 * generic_make_request() is used to make I/O requests of block
3112 * devices. It is passed a &struct bio, which describes the I/O that needs
3113 * to be done.
3114 *
3115 * generic_make_request() does not return any status. The
3116 * success/failure status of the request, along with notification of
3117 * completion, is delivered asynchronously through the bio->bi_end_io
3118 * function described (one day) else where.
3119 *
3120 * The caller of generic_make_request must make sure that bi_io_vec
3121 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3122 * set to describe the device address, and the
3123 * bi_end_io and optionally bi_private are set to describe how
3124 * completion notification should be signaled.
3125 *
3126 * generic_make_request and the drivers it calls may use bi_next if this
3127 * bio happens to be merged with someone else, and may change bi_dev and
3128 * bi_sector for remaps as it sees fit. So the values of these fields
3129 * should NOT be depended on after the call to generic_make_request.
3130 */
Neil Brownd89d8792007-05-01 09:53:42 +02003131static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132{
Jens Axboe165125e2007-07-24 09:28:11 +02003133 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 sector_t maxsector;
NeilBrown5ddfe962006-10-30 22:07:21 -08003135 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003137 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138
3139 might_sleep();
3140 /* Test device or partition size, when known. */
3141 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3142 if (maxsector) {
3143 sector_t sector = bio->bi_sector;
3144
3145 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3146 /*
3147 * This may well happen - the kernel calls bread()
3148 * without checking the size of the device, e.g., when
3149 * mounting a device.
3150 */
3151 handle_bad_sector(bio);
3152 goto end_io;
3153 }
3154 }
3155
3156 /*
3157 * Resolve the mapping until finished. (drivers are
3158 * still free to implement/resolve their own stacking
3159 * by explicitly returning 0)
3160 *
3161 * NOTE: we don't repeat the blk_size check for each new device.
3162 * Stacking drivers are expected to know what they are doing.
3163 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003164 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003165 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 do {
3167 char b[BDEVNAME_SIZE];
3168
3169 q = bdev_get_queue(bio->bi_bdev);
3170 if (!q) {
3171 printk(KERN_ERR
3172 "generic_make_request: Trying to access "
3173 "nonexistent block-device %s (%Lu)\n",
3174 bdevname(bio->bi_bdev, b),
3175 (long long) bio->bi_sector);
3176end_io:
3177 bio_endio(bio, bio->bi_size, -EIO);
3178 break;
3179 }
3180
3181 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3182 printk("bio too big device %s (%u > %u)\n",
3183 bdevname(bio->bi_bdev, b),
3184 bio_sectors(bio),
3185 q->max_hw_sectors);
3186 goto end_io;
3187 }
3188
Nick Pigginfde6ad22005-06-23 00:08:53 -07003189 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190 goto end_io;
3191
Akinobu Mitac17bb492006-12-08 02:39:46 -08003192 if (should_fail_request(bio))
3193 goto end_io;
3194
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 /*
3196 * If this device has partitions, remap block n
3197 * of partition p to block n+start(p) of the disk.
3198 */
3199 blk_partition_remap(bio);
3200
NeilBrown5ddfe962006-10-30 22:07:21 -08003201 if (old_sector != -1)
Jens Axboe2056a782006-03-23 20:00:26 +01003202 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003203 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003204
3205 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3206
NeilBrown5ddfe962006-10-30 22:07:21 -08003207 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003208 old_dev = bio->bi_bdev->bd_dev;
3209
NeilBrown5ddfe962006-10-30 22:07:21 -08003210 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3211 if (maxsector) {
3212 sector_t sector = bio->bi_sector;
3213
Andrew Mortondf66b852006-11-02 22:06:56 -08003214 if (maxsector < nr_sectors ||
3215 maxsector - nr_sectors < sector) {
NeilBrown5ddfe962006-10-30 22:07:21 -08003216 /*
Andrew Mortondf66b852006-11-02 22:06:56 -08003217 * This may well happen - partitions are not
3218 * checked to make sure they are within the size
3219 * of the whole device.
NeilBrown5ddfe962006-10-30 22:07:21 -08003220 */
3221 handle_bad_sector(bio);
3222 goto end_io;
3223 }
3224 }
3225
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 ret = q->make_request_fn(q, bio);
3227 } while (ret);
3228}
3229
Neil Brownd89d8792007-05-01 09:53:42 +02003230/*
3231 * We only want one ->make_request_fn to be active at a time,
3232 * else stack usage with stacked devices could be a problem.
3233 * So use current->bio_{list,tail} to keep a list of requests
3234 * submited by a make_request_fn function.
3235 * current->bio_tail is also used as a flag to say if
3236 * generic_make_request is currently active in this task or not.
3237 * If it is NULL, then no make_request is active. If it is non-NULL,
3238 * then a make_request is active, and new requests should be added
3239 * at the tail
3240 */
3241void generic_make_request(struct bio *bio)
3242{
3243 if (current->bio_tail) {
3244 /* make_request is active */
3245 *(current->bio_tail) = bio;
3246 bio->bi_next = NULL;
3247 current->bio_tail = &bio->bi_next;
3248 return;
3249 }
3250 /* following loop may be a bit non-obvious, and so deserves some
3251 * explanation.
3252 * Before entering the loop, bio->bi_next is NULL (as all callers
3253 * ensure that) so we have a list with a single bio.
3254 * We pretend that we have just taken it off a longer list, so
3255 * we assign bio_list to the next (which is NULL) and bio_tail
3256 * to &bio_list, thus initialising the bio_list of new bios to be
3257 * added. __generic_make_request may indeed add some more bios
3258 * through a recursive call to generic_make_request. If it
3259 * did, we find a non-NULL value in bio_list and re-enter the loop
3260 * from the top. In this case we really did just take the bio
3261 * of the top of the list (no pretending) and so fixup bio_list and
3262 * bio_tail or bi_next, and call into __generic_make_request again.
3263 *
3264 * The loop was structured like this to make only one call to
3265 * __generic_make_request (which is important as it is large and
3266 * inlined) and to keep the structure simple.
3267 */
3268 BUG_ON(bio->bi_next);
3269 do {
3270 current->bio_list = bio->bi_next;
3271 if (bio->bi_next == NULL)
3272 current->bio_tail = &current->bio_list;
3273 else
3274 bio->bi_next = NULL;
3275 __generic_make_request(bio);
3276 bio = current->bio_list;
3277 } while (bio);
3278 current->bio_tail = NULL; /* deactivate */
3279}
3280
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281EXPORT_SYMBOL(generic_make_request);
3282
3283/**
3284 * submit_bio: submit a bio to the block device layer for I/O
3285 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3286 * @bio: The &struct bio which describes the I/O
3287 *
3288 * submit_bio() is very similar in purpose to generic_make_request(), and
3289 * uses that function to do most of the work. Both are fairly rough
3290 * interfaces, @bio must be presetup and ready for I/O.
3291 *
3292 */
3293void submit_bio(int rw, struct bio *bio)
3294{
3295 int count = bio_sectors(bio);
3296
3297 BIO_BUG_ON(!bio->bi_size);
3298 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003299 bio->bi_rw |= rw;
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003300 if (rw & WRITE) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07003301 count_vm_events(PGPGOUT, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003302 } else {
3303 task_io_account_read(bio->bi_size);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003304 count_vm_events(PGPGIN, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306
3307 if (unlikely(block_dump)) {
3308 char b[BDEVNAME_SIZE];
3309 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3310 current->comm, current->pid,
3311 (rw & WRITE) ? "WRITE" : "READ",
3312 (unsigned long long)bio->bi_sector,
3313 bdevname(bio->bi_bdev,b));
3314 }
3315
3316 generic_make_request(bio);
3317}
3318
3319EXPORT_SYMBOL(submit_bio);
3320
Adrian Bunk93d17d32005-06-25 14:59:10 -07003321static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322{
3323 struct bio *bio, *prevbio = NULL;
3324 int nr_phys_segs, nr_hw_segs;
3325 unsigned int phys_size, hw_size;
Jens Axboe165125e2007-07-24 09:28:11 +02003326 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327
3328 if (!rq->bio)
3329 return;
3330
3331 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3332 rq_for_each_bio(bio, rq) {
3333 /* Force bio hw/phys segs to be recalculated. */
3334 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3335
3336 nr_phys_segs += bio_phys_segments(q, bio);
3337 nr_hw_segs += bio_hw_segments(q, bio);
3338 if (prevbio) {
3339 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3340 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3341
3342 if (blk_phys_contig_segment(q, prevbio, bio) &&
3343 pseg <= q->max_segment_size) {
3344 nr_phys_segs--;
3345 phys_size += prevbio->bi_size + bio->bi_size;
3346 } else
3347 phys_size = 0;
3348
3349 if (blk_hw_contig_segment(q, prevbio, bio) &&
3350 hseg <= q->max_segment_size) {
3351 nr_hw_segs--;
3352 hw_size += prevbio->bi_size + bio->bi_size;
3353 } else
3354 hw_size = 0;
3355 }
3356 prevbio = bio;
3357 }
3358
3359 rq->nr_phys_segments = nr_phys_segs;
3360 rq->nr_hw_segments = nr_hw_segs;
3361}
3362
Adrian Bunk93d17d32005-06-25 14:59:10 -07003363static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364{
3365 if (blk_fs_request(rq)) {
3366 rq->hard_sector += nsect;
3367 rq->hard_nr_sectors -= nsect;
3368
3369 /*
3370 * Move the I/O submission pointers ahead if required.
3371 */
3372 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3373 (rq->sector <= rq->hard_sector)) {
3374 rq->sector = rq->hard_sector;
3375 rq->nr_sectors = rq->hard_nr_sectors;
3376 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3377 rq->current_nr_sectors = rq->hard_cur_sectors;
3378 rq->buffer = bio_data(rq->bio);
3379 }
3380
3381 /*
3382 * if total number of sectors is less than the first segment
3383 * size, something has gone terribly wrong
3384 */
3385 if (rq->nr_sectors < rq->current_nr_sectors) {
3386 printk("blk: request botched\n");
3387 rq->nr_sectors = rq->current_nr_sectors;
3388 }
3389 }
3390}
3391
3392static int __end_that_request_first(struct request *req, int uptodate,
3393 int nr_bytes)
3394{
3395 int total_bytes, bio_nbytes, error, next_idx = 0;
3396 struct bio *bio;
3397
Jens Axboe2056a782006-03-23 20:00:26 +01003398 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3399
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 /*
3401 * extend uptodate bool to allow < 0 value to be direct io error
3402 */
3403 error = 0;
3404 if (end_io_error(uptodate))
3405 error = !uptodate ? -EIO : uptodate;
3406
3407 /*
3408 * for a REQ_BLOCK_PC request, we want to carry any eventual
3409 * sense key with us all the way through
3410 */
3411 if (!blk_pc_request(req))
3412 req->errors = 0;
3413
3414 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003415 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 printk("end_request: I/O error, dev %s, sector %llu\n",
3417 req->rq_disk ? req->rq_disk->disk_name : "?",
3418 (unsigned long long)req->sector);
3419 }
3420
Jens Axboed72d9042005-11-01 08:35:42 +01003421 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003422 const int rw = rq_data_dir(req);
3423
Jens Axboe53e86062006-01-17 11:09:27 +01003424 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003425 }
3426
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 total_bytes = bio_nbytes = 0;
3428 while ((bio = req->bio) != NULL) {
3429 int nbytes;
3430
3431 if (nr_bytes >= bio->bi_size) {
3432 req->bio = bio->bi_next;
3433 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003434 if (!ordered_bio_endio(req, bio, nbytes, error))
3435 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 next_idx = 0;
3437 bio_nbytes = 0;
3438 } else {
3439 int idx = bio->bi_idx + next_idx;
3440
3441 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3442 blk_dump_rq_flags(req, "__end_that");
3443 printk("%s: bio idx %d >= vcnt %d\n",
3444 __FUNCTION__,
3445 bio->bi_idx, bio->bi_vcnt);
3446 break;
3447 }
3448
3449 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3450 BIO_BUG_ON(nbytes > bio->bi_size);
3451
3452 /*
3453 * not a complete bvec done
3454 */
3455 if (unlikely(nbytes > nr_bytes)) {
3456 bio_nbytes += nr_bytes;
3457 total_bytes += nr_bytes;
3458 break;
3459 }
3460
3461 /*
3462 * advance to the next vector
3463 */
3464 next_idx++;
3465 bio_nbytes += nbytes;
3466 }
3467
3468 total_bytes += nbytes;
3469 nr_bytes -= nbytes;
3470
3471 if ((bio = req->bio)) {
3472 /*
3473 * end more in this run, or just return 'not-done'
3474 */
3475 if (unlikely(nr_bytes <= 0))
3476 break;
3477 }
3478 }
3479
3480 /*
3481 * completely done
3482 */
3483 if (!req->bio)
3484 return 0;
3485
3486 /*
3487 * if the request wasn't completed, update state
3488 */
3489 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003490 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3491 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 bio->bi_idx += next_idx;
3493 bio_iovec(bio)->bv_offset += nr_bytes;
3494 bio_iovec(bio)->bv_len -= nr_bytes;
3495 }
3496
3497 blk_recalc_rq_sectors(req, total_bytes >> 9);
3498 blk_recalc_rq_segments(req);
3499 return 1;
3500}
3501
3502/**
3503 * end_that_request_first - end I/O on a request
3504 * @req: the request being processed
3505 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3506 * @nr_sectors: number of sectors to end I/O on
3507 *
3508 * Description:
3509 * Ends I/O on a number of sectors attached to @req, and sets it up
3510 * for the next range of segments (if any) in the cluster.
3511 *
3512 * Return:
3513 * 0 - we are done with this request, call end_that_request_last()
3514 * 1 - still buffers pending for this request
3515 **/
3516int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3517{
3518 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3519}
3520
3521EXPORT_SYMBOL(end_that_request_first);
3522
3523/**
3524 * end_that_request_chunk - end I/O on a request
3525 * @req: the request being processed
3526 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3527 * @nr_bytes: number of bytes to complete
3528 *
3529 * Description:
3530 * Ends I/O on a number of bytes attached to @req, and sets it up
3531 * for the next range of segments (if any). Like end_that_request_first(),
3532 * but deals with bytes instead of sectors.
3533 *
3534 * Return:
3535 * 0 - we are done with this request, call end_that_request_last()
3536 * 1 - still buffers pending for this request
3537 **/
3538int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3539{
3540 return __end_that_request_first(req, uptodate, nr_bytes);
3541}
3542
3543EXPORT_SYMBOL(end_that_request_chunk);
3544
3545/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003546 * splice the completion data to a local structure and hand off to
3547 * process_completion_queue() to complete the requests
3548 */
3549static void blk_done_softirq(struct softirq_action *h)
3550{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003551 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003552
3553 local_irq_disable();
3554 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003555 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003556 local_irq_enable();
3557
3558 while (!list_empty(&local_list)) {
3559 struct request *rq = list_entry(local_list.next, struct request, donelist);
3560
3561 list_del_init(&rq->donelist);
3562 rq->q->softirq_done_fn(rq);
3563 }
3564}
3565
Jens Axboeff856ba2006-01-09 16:02:34 +01003566static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3567 void *hcpu)
3568{
3569 /*
3570 * If a CPU goes away, splice its entries to the current CPU
3571 * and trigger a run of the softirq
3572 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003573 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003574 int cpu = (unsigned long) hcpu;
3575
3576 local_irq_disable();
3577 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3578 &__get_cpu_var(blk_cpu_done));
3579 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3580 local_irq_enable();
3581 }
3582
3583 return NOTIFY_OK;
3584}
3585
3586
Chandra Seetharaman054cc8a2006-06-27 02:54:07 -07003587static struct notifier_block __devinitdata blk_cpu_notifier = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003588 .notifier_call = blk_cpu_notify,
3589};
3590
Jens Axboeff856ba2006-01-09 16:02:34 +01003591/**
3592 * blk_complete_request - end I/O on a request
3593 * @req: the request being processed
3594 *
3595 * Description:
3596 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003597 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003598 * through requeueing. Theh actual completion happens out-of-order,
3599 * through a softirq handler. The user must have registered a completion
3600 * callback through blk_queue_softirq_done().
3601 **/
3602
3603void blk_complete_request(struct request *req)
3604{
3605 struct list_head *cpu_list;
3606 unsigned long flags;
3607
3608 BUG_ON(!req->q->softirq_done_fn);
3609
3610 local_irq_save(flags);
3611
3612 cpu_list = &__get_cpu_var(blk_cpu_done);
3613 list_add_tail(&req->donelist, cpu_list);
3614 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3615
3616 local_irq_restore(flags);
3617}
3618
3619EXPORT_SYMBOL(blk_complete_request);
3620
3621/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 * queue lock must be held
3623 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003624void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625{
3626 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003627 int error;
3628
3629 /*
3630 * extend uptodate bool to allow < 0 value to be direct io error
3631 */
3632 error = 0;
3633 if (end_io_error(uptodate))
3634 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635
3636 if (unlikely(laptop_mode) && blk_fs_request(req))
3637 laptop_io_completion();
3638
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003639 /*
3640 * Account IO completion. bar_rq isn't accounted as a normal
3641 * IO on queueing nor completion. Accounting the containing
3642 * request is enough.
3643 */
3644 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003646 const int rw = rq_data_dir(req);
3647
3648 __disk_stat_inc(disk, ios[rw]);
3649 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 disk_round_stats(disk);
3651 disk->in_flight--;
3652 }
3653 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003654 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 else
3656 __blk_put_request(req->q, req);
3657}
3658
3659EXPORT_SYMBOL(end_that_request_last);
3660
3661void end_request(struct request *req, int uptodate)
3662{
3663 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3664 add_disk_randomness(req->rq_disk);
3665 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003666 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 }
3668}
3669
3670EXPORT_SYMBOL(end_request);
3671
Jens Axboe165125e2007-07-24 09:28:11 +02003672void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3673 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003675 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3676 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003677
3678 rq->nr_phys_segments = bio_phys_segments(q, bio);
3679 rq->nr_hw_segments = bio_hw_segments(q, bio);
3680 rq->current_nr_sectors = bio_cur_sectors(bio);
3681 rq->hard_cur_sectors = rq->current_nr_sectors;
3682 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3683 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003684 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685
3686 rq->bio = rq->biotail = bio;
3687}
3688
3689EXPORT_SYMBOL(blk_rq_bio_prep);
3690
3691int kblockd_schedule_work(struct work_struct *work)
3692{
3693 return queue_work(kblockd_workqueue, work);
3694}
3695
3696EXPORT_SYMBOL(kblockd_schedule_work);
3697
Andrew Morton19a75d82007-05-09 02:33:56 -07003698void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003700 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701}
Andrew Morton19a75d82007-05-09 02:33:56 -07003702EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703
3704int __init blk_dev_init(void)
3705{
Jens Axboeff856ba2006-01-09 16:02:34 +01003706 int i;
3707
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 kblockd_workqueue = create_workqueue("kblockd");
3709 if (!kblockd_workqueue)
3710 panic("Failed to create kblockd\n");
3711
3712 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003713 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714
3715 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003716 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717
3718 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003719 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003721 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003722 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3723
3724 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003725 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003726
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003727 blk_max_low_pfn = max_low_pfn - 1;
3728 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729
3730 return 0;
3731}
3732
3733/*
3734 * IO Context helper functions
3735 */
3736void put_io_context(struct io_context *ioc)
3737{
3738 if (ioc == NULL)
3739 return;
3740
3741 BUG_ON(atomic_read(&ioc->refcount) == 0);
3742
3743 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003744 struct cfq_io_context *cic;
3745
Al Viro334e94d2006-03-18 15:05:53 -05003746 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 if (ioc->aic && ioc->aic->dtor)
3748 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003749 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003750 struct rb_node *n = rb_first(&ioc->cic_root);
3751
3752 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003753 cic->dtor(ioc);
3754 }
Al Viro334e94d2006-03-18 15:05:53 -05003755 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756
3757 kmem_cache_free(iocontext_cachep, ioc);
3758 }
3759}
3760EXPORT_SYMBOL(put_io_context);
3761
3762/* Called by the exitting task */
3763void exit_io_context(void)
3764{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003766 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767
Jens Axboe22e2c502005-06-27 10:55:12 +02003768 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 ioc = current->io_context;
3770 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003771 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772
Oleg Nesterov25034d72006-08-29 09:15:14 +02003773 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774 if (ioc->aic && ioc->aic->exit)
3775 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003776 if (ioc->cic_root.rb_node != NULL) {
3777 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3778 cic->exit(ioc);
3779 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003780
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781 put_io_context(ioc);
3782}
3783
3784/*
3785 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003786 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003788 * This returned IO context doesn't have a specifically elevated refcount,
3789 * but since the current task itself holds a reference, the context can be
3790 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003792static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793{
3794 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795 struct io_context *ret;
3796
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003798 if (likely(ret))
3799 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800
Jens Axboeb5deef92006-07-19 23:39:40 +02003801 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 if (ret) {
3803 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003804 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003805 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 ret->last_waited = jiffies; /* doesn't matter... */
3807 ret->nr_batch_requests = 0; /* because this is 0 */
3808 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003809 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003810 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003811 /* make sure set_task_ioprio() sees the settings above */
3812 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003813 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 }
3815
3816 return ret;
3817}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003818
3819/*
3820 * If the current task has no IO context then create one and initialise it.
3821 * If it does have a context, take a ref on it.
3822 *
3823 * This is always called in the context of the task which submitted the I/O.
3824 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003825struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003826{
3827 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003828 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003829 if (likely(ret))
3830 atomic_inc(&ret->refcount);
3831 return ret;
3832}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833EXPORT_SYMBOL(get_io_context);
3834
3835void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3836{
3837 struct io_context *src = *psrc;
3838 struct io_context *dst = *pdst;
3839
3840 if (src) {
3841 BUG_ON(atomic_read(&src->refcount) == 0);
3842 atomic_inc(&src->refcount);
3843 put_io_context(dst);
3844 *pdst = src;
3845 }
3846}
3847EXPORT_SYMBOL(copy_io_context);
3848
3849void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3850{
3851 struct io_context *temp;
3852 temp = *ioc1;
3853 *ioc1 = *ioc2;
3854 *ioc2 = temp;
3855}
3856EXPORT_SYMBOL(swap_io_context);
3857
3858/*
3859 * sysfs parts below
3860 */
3861struct queue_sysfs_entry {
3862 struct attribute attr;
3863 ssize_t (*show)(struct request_queue *, char *);
3864 ssize_t (*store)(struct request_queue *, const char *, size_t);
3865};
3866
3867static ssize_t
3868queue_var_show(unsigned int var, char *page)
3869{
3870 return sprintf(page, "%d\n", var);
3871}
3872
3873static ssize_t
3874queue_var_store(unsigned long *var, const char *page, size_t count)
3875{
3876 char *p = (char *) page;
3877
3878 *var = simple_strtoul(p, &p, 10);
3879 return count;
3880}
3881
3882static ssize_t queue_requests_show(struct request_queue *q, char *page)
3883{
3884 return queue_var_show(q->nr_requests, (page));
3885}
3886
3887static ssize_t
3888queue_requests_store(struct request_queue *q, const char *page, size_t count)
3889{
3890 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003891 unsigned long nr;
3892 int ret = queue_var_store(&nr, page, count);
3893 if (nr < BLKDEV_MIN_RQ)
3894 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003895
Al Viroc981ff92006-03-18 13:51:29 -05003896 spin_lock_irq(q->queue_lock);
3897 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003898 blk_queue_congestion_threshold(q);
3899
3900 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003901 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003903 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904
3905 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003906 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003908 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909
3910 if (rl->count[READ] >= q->nr_requests) {
3911 blk_set_queue_full(q, READ);
3912 } else if (rl->count[READ]+1 <= q->nr_requests) {
3913 blk_clear_queue_full(q, READ);
3914 wake_up(&rl->wait[READ]);
3915 }
3916
3917 if (rl->count[WRITE] >= q->nr_requests) {
3918 blk_set_queue_full(q, WRITE);
3919 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3920 blk_clear_queue_full(q, WRITE);
3921 wake_up(&rl->wait[WRITE]);
3922 }
Al Viroc981ff92006-03-18 13:51:29 -05003923 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924 return ret;
3925}
3926
3927static ssize_t queue_ra_show(struct request_queue *q, char *page)
3928{
3929 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3930
3931 return queue_var_show(ra_kb, (page));
3932}
3933
3934static ssize_t
3935queue_ra_store(struct request_queue *q, const char *page, size_t count)
3936{
3937 unsigned long ra_kb;
3938 ssize_t ret = queue_var_store(&ra_kb, page, count);
3939
3940 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3942 spin_unlock_irq(q->queue_lock);
3943
3944 return ret;
3945}
3946
3947static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3948{
3949 int max_sectors_kb = q->max_sectors >> 1;
3950
3951 return queue_var_show(max_sectors_kb, (page));
3952}
3953
3954static ssize_t
3955queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3956{
3957 unsigned long max_sectors_kb,
3958 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3959 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3960 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3961 int ra_kb;
3962
3963 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3964 return -EINVAL;
3965 /*
3966 * Take the queue lock to update the readahead and max_sectors
3967 * values synchronously:
3968 */
3969 spin_lock_irq(q->queue_lock);
3970 /*
3971 * Trim readahead window as well, if necessary:
3972 */
3973 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3974 if (ra_kb > max_sectors_kb)
3975 q->backing_dev_info.ra_pages =
3976 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3977
3978 q->max_sectors = max_sectors_kb << 1;
3979 spin_unlock_irq(q->queue_lock);
3980
3981 return ret;
3982}
3983
3984static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3985{
3986 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3987
3988 return queue_var_show(max_hw_sectors_kb, (page));
3989}
3990
3991
3992static struct queue_sysfs_entry queue_requests_entry = {
3993 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3994 .show = queue_requests_show,
3995 .store = queue_requests_store,
3996};
3997
3998static struct queue_sysfs_entry queue_ra_entry = {
3999 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4000 .show = queue_ra_show,
4001 .store = queue_ra_store,
4002};
4003
4004static struct queue_sysfs_entry queue_max_sectors_entry = {
4005 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4006 .show = queue_max_sectors_show,
4007 .store = queue_max_sectors_store,
4008};
4009
4010static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4011 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4012 .show = queue_max_hw_sectors_show,
4013};
4014
4015static struct queue_sysfs_entry queue_iosched_entry = {
4016 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4017 .show = elv_iosched_show,
4018 .store = elv_iosched_store,
4019};
4020
4021static struct attribute *default_attrs[] = {
4022 &queue_requests_entry.attr,
4023 &queue_ra_entry.attr,
4024 &queue_max_hw_sectors_entry.attr,
4025 &queue_max_sectors_entry.attr,
4026 &queue_iosched_entry.attr,
4027 NULL,
4028};
4029
4030#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4031
4032static ssize_t
4033queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4034{
4035 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004036 struct request_queue *q =
4037 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004038 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039
Linus Torvalds1da177e2005-04-16 15:20:36 -07004040 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004041 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004042 mutex_lock(&q->sysfs_lock);
4043 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4044 mutex_unlock(&q->sysfs_lock);
4045 return -ENOENT;
4046 }
4047 res = entry->show(q, page);
4048 mutex_unlock(&q->sysfs_lock);
4049 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050}
4051
4052static ssize_t
4053queue_attr_store(struct kobject *kobj, struct attribute *attr,
4054 const char *page, size_t length)
4055{
4056 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004057 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058
Al Viro483f4af2006-03-18 18:34:37 -05004059 ssize_t res;
4060
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004062 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004063 mutex_lock(&q->sysfs_lock);
4064 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4065 mutex_unlock(&q->sysfs_lock);
4066 return -ENOENT;
4067 }
4068 res = entry->store(q, page, length);
4069 mutex_unlock(&q->sysfs_lock);
4070 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071}
4072
4073static struct sysfs_ops queue_sysfs_ops = {
4074 .show = queue_attr_show,
4075 .store = queue_attr_store,
4076};
4077
Adrian Bunk93d17d32005-06-25 14:59:10 -07004078static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 .sysfs_ops = &queue_sysfs_ops,
4080 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004081 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082};
4083
4084int blk_register_queue(struct gendisk *disk)
4085{
4086 int ret;
4087
Jens Axboe165125e2007-07-24 09:28:11 +02004088 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089
4090 if (!q || !q->request_fn)
4091 return -ENXIO;
4092
4093 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094
Al Viro483f4af2006-03-18 18:34:37 -05004095 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 if (ret < 0)
4097 return ret;
4098
Al Viro483f4af2006-03-18 18:34:37 -05004099 kobject_uevent(&q->kobj, KOBJ_ADD);
4100
Linus Torvalds1da177e2005-04-16 15:20:36 -07004101 ret = elv_register_queue(q);
4102 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004103 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4104 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105 return ret;
4106 }
4107
4108 return 0;
4109}
4110
4111void blk_unregister_queue(struct gendisk *disk)
4112{
Jens Axboe165125e2007-07-24 09:28:11 +02004113 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114
4115 if (q && q->request_fn) {
4116 elv_unregister_queue(q);
4117
Al Viro483f4af2006-03-18 18:34:37 -05004118 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4119 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 kobject_put(&disk->kobj);
4121 }
4122}