blob: ed39313c408534ee832e4ff61c428ba46c8b2c12 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001079 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 rq->tag = -1;
1081
1082 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001083 printk(KERN_ERR "%s: tag %d is missing\n",
1084 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
1086 bqt->tag_index[tag] = NULL;
Jens Axboef3da54b2007-09-13 14:26:53 +02001087
Nick Piggindd941252007-09-14 08:41:12 +02001088 /*
1089 * We use test_and_clear_bit's memory ordering properties here.
1090 * The tag_map bit acts as a lock for tag_index[bit], so we need
1091 * a barrer before clearing the bit (precisely: release semantics).
1092 * Could use clear_bit_unlock when it is merged.
1093 */
Jens Axboef3da54b2007-09-13 14:26:53 +02001094 if (unlikely(!test_and_clear_bit(tag, bqt->tag_map))) {
1095 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1096 __FUNCTION__, tag);
1097 return;
1098 }
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 bqt->busy--;
1101}
1102
1103EXPORT_SYMBOL(blk_queue_end_tag);
1104
1105/**
1106 * blk_queue_start_tag - find a free tag and assign it
1107 * @q: the request queue for the device
1108 * @rq: the block request that needs tagging
1109 *
1110 * Description:
1111 * This can either be used as a stand-alone helper, or possibly be
1112 * assigned as the queue &prep_rq_fn (in which case &struct request
1113 * automagically gets a tag assigned). Note that this function
1114 * assumes that any type of request can be queued! if this is not
1115 * true for your device, you must check the request type before
1116 * calling this function. The request will also be removed from
1117 * the request queue, so it's the drivers responsibility to readd
1118 * it if it should need to be restarted for some reason.
1119 *
1120 * Notes:
1121 * queue lock must be held.
1122 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001123int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
1125 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001126 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Jens Axboe4aff5e22006-08-10 08:44:47 +02001128 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001130 "%s: request %p for device [%s] already tagged %d",
1131 __FUNCTION__, rq,
1132 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 BUG();
1134 }
1135
Jens Axboe059af492006-09-21 20:37:22 +02001136 /*
1137 * Protect against shared tag maps, as we may not have exclusive
1138 * access to the tag map.
1139 */
1140 do {
1141 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1142 if (tag >= bqt->max_depth)
1143 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Jens Axboe059af492006-09-21 20:37:22 +02001145 } while (test_and_set_bit(tag, bqt->tag_map));
Nick Piggindd941252007-09-14 08:41:12 +02001146 /*
1147 * We rely on test_and_set_bit providing lock memory ordering semantics
1148 * (could use test_and_set_bit_lock when it is merged).
1149 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
Jens Axboe4aff5e22006-08-10 08:44:47 +02001151 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 rq->tag = tag;
1153 bqt->tag_index[tag] = rq;
1154 blkdev_dequeue_request(rq);
1155 list_add(&rq->queuelist, &bqt->busy_list);
1156 bqt->busy++;
1157 return 0;
1158}
1159
1160EXPORT_SYMBOL(blk_queue_start_tag);
1161
1162/**
1163 * blk_queue_invalidate_tags - invalidate all pending tags
1164 * @q: the request queue for the device
1165 *
1166 * Description:
1167 * Hardware conditions may dictate a need to stop all pending requests.
1168 * In this case, we will safely clear the block side of the tag queue and
1169 * readd all requests to the request queue in the right order.
1170 *
1171 * Notes:
1172 * queue lock must be held.
1173 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001174void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
1176 struct blk_queue_tag *bqt = q->queue_tags;
1177 struct list_head *tmp, *n;
1178 struct request *rq;
1179
1180 list_for_each_safe(tmp, n, &bqt->busy_list) {
1181 rq = list_entry_rq(tmp);
1182
1183 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001184 printk(KERN_ERR
1185 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001187 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 } else
1189 blk_queue_end_tag(q, rq);
1190
Jens Axboe4aff5e22006-08-10 08:44:47 +02001191 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1193 }
1194}
1195
1196EXPORT_SYMBOL(blk_queue_invalidate_tags);
1197
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198void blk_dump_rq_flags(struct request *rq, char *msg)
1199{
1200 int bit;
1201
Jens Axboe4aff5e22006-08-10 08:44:47 +02001202 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1203 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1204 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1207 rq->nr_sectors,
1208 rq->current_nr_sectors);
1209 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1210
Jens Axboe4aff5e22006-08-10 08:44:47 +02001211 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 printk("cdb: ");
1213 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1214 printk("%02x ", rq->cmd[bit]);
1215 printk("\n");
1216 }
1217}
1218
1219EXPORT_SYMBOL(blk_dump_rq_flags);
1220
Jens Axboe165125e2007-07-24 09:28:11 +02001221void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
1223 struct bio_vec *bv, *bvprv = NULL;
1224 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1225 int high, highprv = 1;
1226
1227 if (unlikely(!bio->bi_io_vec))
1228 return;
1229
1230 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1231 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1232 bio_for_each_segment(bv, bio, i) {
1233 /*
1234 * the trick here is making sure that a high page is never
1235 * considered part of another segment, since that might
1236 * change with the bounce page.
1237 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001238 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (high || highprv)
1240 goto new_hw_segment;
1241 if (cluster) {
1242 if (seg_size + bv->bv_len > q->max_segment_size)
1243 goto new_segment;
1244 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1245 goto new_segment;
1246 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1247 goto new_segment;
1248 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1249 goto new_hw_segment;
1250
1251 seg_size += bv->bv_len;
1252 hw_seg_size += bv->bv_len;
1253 bvprv = bv;
1254 continue;
1255 }
1256new_segment:
1257 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1258 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1259 hw_seg_size += bv->bv_len;
1260 } else {
1261new_hw_segment:
1262 if (hw_seg_size > bio->bi_hw_front_size)
1263 bio->bi_hw_front_size = hw_seg_size;
1264 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1265 nr_hw_segs++;
1266 }
1267
1268 nr_phys_segs++;
1269 bvprv = bv;
1270 seg_size = bv->bv_len;
1271 highprv = high;
1272 }
1273 if (hw_seg_size > bio->bi_hw_back_size)
1274 bio->bi_hw_back_size = hw_seg_size;
1275 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1276 bio->bi_hw_front_size = hw_seg_size;
1277 bio->bi_phys_segments = nr_phys_segs;
1278 bio->bi_hw_segments = nr_hw_segs;
1279 bio->bi_flags |= (1 << BIO_SEG_VALID);
1280}
Neil Brown387bb172007-02-08 14:20:29 -08001281EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Jens Axboe165125e2007-07-24 09:28:11 +02001283static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 struct bio *nxt)
1285{
1286 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1287 return 0;
1288
1289 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1290 return 0;
1291 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1292 return 0;
1293
1294 /*
1295 * bio and nxt are contigous in memory, check if the queue allows
1296 * these two to be merged into one
1297 */
1298 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1299 return 1;
1300
1301 return 0;
1302}
1303
Jens Axboe165125e2007-07-24 09:28:11 +02001304static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 struct bio *nxt)
1306{
1307 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1308 blk_recount_segments(q, bio);
1309 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1310 blk_recount_segments(q, nxt);
1311 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001312 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001314 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 return 0;
1316
1317 return 1;
1318}
1319
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320/*
1321 * map a request to scatterlist, return number of sg entries setup. Caller
1322 * must make sure sg can hold rq->nr_phys_segments entries
1323 */
Jens Axboe165125e2007-07-24 09:28:11 +02001324int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1325 struct scatterlist *sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
1327 struct bio_vec *bvec, *bvprv;
1328 struct bio *bio;
1329 int nsegs, i, cluster;
1330
1331 nsegs = 0;
1332 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1333
1334 /*
1335 * for each bio in rq
1336 */
1337 bvprv = NULL;
1338 rq_for_each_bio(bio, rq) {
1339 /*
1340 * for each segment in bio
1341 */
1342 bio_for_each_segment(bvec, bio, i) {
1343 int nbytes = bvec->bv_len;
1344
1345 if (bvprv && cluster) {
1346 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1347 goto new_segment;
1348
1349 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1350 goto new_segment;
1351 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1352 goto new_segment;
1353
1354 sg[nsegs - 1].length += nbytes;
1355 } else {
1356new_segment:
1357 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1358 sg[nsegs].page = bvec->bv_page;
1359 sg[nsegs].length = nbytes;
1360 sg[nsegs].offset = bvec->bv_offset;
1361
1362 nsegs++;
1363 }
1364 bvprv = bvec;
1365 } /* segments in bio */
1366 } /* bios in rq */
1367
1368 return nsegs;
1369}
1370
1371EXPORT_SYMBOL(blk_rq_map_sg);
1372
1373/*
1374 * the standard queue merge functions, can be overridden with device
1375 * specific ones if so desired
1376 */
1377
Jens Axboe165125e2007-07-24 09:28:11 +02001378static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 struct request *req,
1380 struct bio *bio)
1381{
1382 int nr_phys_segs = bio_phys_segments(q, bio);
1383
1384 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001385 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 if (req == q->last_merge)
1387 q->last_merge = NULL;
1388 return 0;
1389 }
1390
1391 /*
1392 * A hw segment is just getting larger, bump just the phys
1393 * counter.
1394 */
1395 req->nr_phys_segments += nr_phys_segs;
1396 return 1;
1397}
1398
Jens Axboe165125e2007-07-24 09:28:11 +02001399static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 struct request *req,
1401 struct bio *bio)
1402{
1403 int nr_hw_segs = bio_hw_segments(q, bio);
1404 int nr_phys_segs = bio_phys_segments(q, bio);
1405
1406 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1407 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001408 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 if (req == q->last_merge)
1410 q->last_merge = NULL;
1411 return 0;
1412 }
1413
1414 /*
1415 * This will form the start of a new hw segment. Bump both
1416 * counters.
1417 */
1418 req->nr_hw_segments += nr_hw_segs;
1419 req->nr_phys_segments += nr_phys_segs;
1420 return 1;
1421}
1422
Jens Axboe165125e2007-07-24 09:28:11 +02001423int ll_back_merge_fn(struct request_queue *q, struct request *req, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
Mike Christiedefd94b2005-12-05 02:37:06 -06001425 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 int len;
1427
Mike Christiedefd94b2005-12-05 02:37:06 -06001428 if (unlikely(blk_pc_request(req)))
1429 max_sectors = q->max_hw_sectors;
1430 else
1431 max_sectors = q->max_sectors;
1432
1433 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001434 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 if (req == q->last_merge)
1436 q->last_merge = NULL;
1437 return 0;
1438 }
1439 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1440 blk_recount_segments(q, req->biotail);
1441 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1442 blk_recount_segments(q, bio);
1443 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1444 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1445 !BIOVEC_VIRT_OVERSIZE(len)) {
1446 int mergeable = ll_new_mergeable(q, req, bio);
1447
1448 if (mergeable) {
1449 if (req->nr_hw_segments == 1)
1450 req->bio->bi_hw_front_size = len;
1451 if (bio->bi_hw_segments == 1)
1452 bio->bi_hw_back_size = len;
1453 }
1454 return mergeable;
1455 }
1456
1457 return ll_new_hw_segment(q, req, bio);
1458}
Jens Axboe1aa4f242006-12-19 08:33:11 +01001459EXPORT_SYMBOL(ll_back_merge_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Jens Axboe165125e2007-07-24 09:28:11 +02001461static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 struct bio *bio)
1463{
Mike Christiedefd94b2005-12-05 02:37:06 -06001464 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 int len;
1466
Mike Christiedefd94b2005-12-05 02:37:06 -06001467 if (unlikely(blk_pc_request(req)))
1468 max_sectors = q->max_hw_sectors;
1469 else
1470 max_sectors = q->max_sectors;
1471
1472
1473 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001474 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (req == q->last_merge)
1476 q->last_merge = NULL;
1477 return 0;
1478 }
1479 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1480 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1481 blk_recount_segments(q, bio);
1482 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1483 blk_recount_segments(q, req->bio);
1484 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1485 !BIOVEC_VIRT_OVERSIZE(len)) {
1486 int mergeable = ll_new_mergeable(q, req, bio);
1487
1488 if (mergeable) {
1489 if (bio->bi_hw_segments == 1)
1490 bio->bi_hw_front_size = len;
1491 if (req->nr_hw_segments == 1)
1492 req->biotail->bi_hw_back_size = len;
1493 }
1494 return mergeable;
1495 }
1496
1497 return ll_new_hw_segment(q, req, bio);
1498}
1499
Jens Axboe165125e2007-07-24 09:28:11 +02001500static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 struct request *next)
1502{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001503 int total_phys_segments;
1504 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 /*
1507 * First check if the either of the requests are re-queued
1508 * requests. Can't merge them if they are.
1509 */
1510 if (req->special || next->special)
1511 return 0;
1512
1513 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001514 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 */
1516 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1517 return 0;
1518
1519 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1520 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1521 total_phys_segments--;
1522
1523 if (total_phys_segments > q->max_phys_segments)
1524 return 0;
1525
1526 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1527 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1528 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1529 /*
1530 * propagate the combined length to the end of the requests
1531 */
1532 if (req->nr_hw_segments == 1)
1533 req->bio->bi_hw_front_size = len;
1534 if (next->nr_hw_segments == 1)
1535 next->biotail->bi_hw_back_size = len;
1536 total_hw_segments--;
1537 }
1538
1539 if (total_hw_segments > q->max_hw_segments)
1540 return 0;
1541
1542 /* Merge is OK... */
1543 req->nr_phys_segments = total_phys_segments;
1544 req->nr_hw_segments = total_hw_segments;
1545 return 1;
1546}
1547
1548/*
1549 * "plug" the device if there are no outstanding requests: this will
1550 * force the transfer to start only after we have put all the requests
1551 * on the list.
1552 *
1553 * This is called with interrupts off and no requests on the queue and
1554 * with the queue lock held.
1555 */
Jens Axboe165125e2007-07-24 09:28:11 +02001556void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 WARN_ON(!irqs_disabled());
1559
1560 /*
1561 * don't plug a stopped queue, it must be paired with blk_start_queue()
1562 * which will restart the queueing
1563 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001564 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 return;
1566
Jens Axboe2056a782006-03-23 20:00:26 +01001567 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001569 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
1572
1573EXPORT_SYMBOL(blk_plug_device);
1574
1575/*
1576 * remove the queue from the plugged list, if present. called with
1577 * queue lock held and interrupts disabled.
1578 */
Jens Axboe165125e2007-07-24 09:28:11 +02001579int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580{
1581 WARN_ON(!irqs_disabled());
1582
1583 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1584 return 0;
1585
1586 del_timer(&q->unplug_timer);
1587 return 1;
1588}
1589
1590EXPORT_SYMBOL(blk_remove_plug);
1591
1592/*
1593 * remove the plug and let it rip..
1594 */
Jens Axboe165125e2007-07-24 09:28:11 +02001595void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001597 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return;
1599
1600 if (!blk_remove_plug(q))
1601 return;
1602
Jens Axboe22e2c502005-06-27 10:55:12 +02001603 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
1605EXPORT_SYMBOL(__generic_unplug_device);
1606
1607/**
1608 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001609 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 *
1611 * Description:
1612 * Linux uses plugging to build bigger requests queues before letting
1613 * the device have at them. If a queue is plugged, the I/O scheduler
1614 * is still adding and merging requests on the queue. Once the queue
1615 * gets unplugged, the request_fn defined for the queue is invoked and
1616 * transfers started.
1617 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001618void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
1620 spin_lock_irq(q->queue_lock);
1621 __generic_unplug_device(q);
1622 spin_unlock_irq(q->queue_lock);
1623}
1624EXPORT_SYMBOL(generic_unplug_device);
1625
1626static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1627 struct page *page)
1628{
Jens Axboe165125e2007-07-24 09:28:11 +02001629 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 /*
1632 * devices don't necessarily have an ->unplug_fn defined
1633 */
Jens Axboe2056a782006-03-23 20:00:26 +01001634 if (q->unplug_fn) {
1635 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1636 q->rq.count[READ] + q->rq.count[WRITE]);
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640}
1641
David Howells65f27f32006-11-22 14:55:48 +00001642static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
Jens Axboe165125e2007-07-24 09:28:11 +02001644 struct request_queue *q =
1645 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Jens Axboe2056a782006-03-23 20:00:26 +01001647 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1648 q->rq.count[READ] + q->rq.count[WRITE]);
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 q->unplug_fn(q);
1651}
1652
1653static void blk_unplug_timeout(unsigned long data)
1654{
Jens Axboe165125e2007-07-24 09:28:11 +02001655 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
Jens Axboe2056a782006-03-23 20:00:26 +01001657 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1658 q->rq.count[READ] + q->rq.count[WRITE]);
1659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 kblockd_schedule_work(&q->unplug_work);
1661}
1662
1663/**
1664 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001665 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 *
1667 * Description:
1668 * blk_start_queue() will clear the stop flag on the queue, and call
1669 * the request_fn for the queue if it was in a stopped state when
1670 * entered. Also see blk_stop_queue(). Queue lock must be held.
1671 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001672void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001674 WARN_ON(!irqs_disabled());
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1677
1678 /*
1679 * one level of recursion is ok and is much faster than kicking
1680 * the unplug handling
1681 */
1682 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1683 q->request_fn(q);
1684 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1685 } else {
1686 blk_plug_device(q);
1687 kblockd_schedule_work(&q->unplug_work);
1688 }
1689}
1690
1691EXPORT_SYMBOL(blk_start_queue);
1692
1693/**
1694 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001695 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 *
1697 * Description:
1698 * The Linux block layer assumes that a block driver will consume all
1699 * entries on the request queue when the request_fn strategy is called.
1700 * Often this will not happen, because of hardware limitations (queue
1701 * depth settings). If a device driver gets a 'queue full' response,
1702 * or if it simply chooses not to queue more I/O at one point, it can
1703 * call this function to prevent the request_fn from being called until
1704 * the driver has signalled it's ready to go again. This happens by calling
1705 * blk_start_queue() to restart queue operations. Queue lock must be held.
1706 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001707void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
1709 blk_remove_plug(q);
1710 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1711}
1712EXPORT_SYMBOL(blk_stop_queue);
1713
1714/**
1715 * blk_sync_queue - cancel any pending callbacks on a queue
1716 * @q: the queue
1717 *
1718 * Description:
1719 * The block layer may perform asynchronous callback activity
1720 * on a queue, such as calling the unplug function after a timeout.
1721 * A block device may call blk_sync_queue to ensure that any
1722 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001723 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 * that its ->make_request_fn will not re-add plugging prior to calling
1725 * this function.
1726 *
1727 */
1728void blk_sync_queue(struct request_queue *q)
1729{
1730 del_timer_sync(&q->unplug_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731}
1732EXPORT_SYMBOL(blk_sync_queue);
1733
1734/**
1735 * blk_run_queue - run a single device queue
1736 * @q: The queue to run
1737 */
1738void blk_run_queue(struct request_queue *q)
1739{
1740 unsigned long flags;
1741
1742 spin_lock_irqsave(q->queue_lock, flags);
1743 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001744
1745 /*
1746 * Only recurse once to avoid overrunning the stack, let the unplug
1747 * handling reinvoke the handler shortly if we already got there.
1748 */
1749 if (!elv_queue_empty(q)) {
1750 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1751 q->request_fn(q);
1752 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1753 } else {
1754 blk_plug_device(q);
1755 kblockd_schedule_work(&q->unplug_work);
1756 }
1757 }
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 spin_unlock_irqrestore(q->queue_lock, flags);
1760}
1761EXPORT_SYMBOL(blk_run_queue);
1762
1763/**
Jens Axboe165125e2007-07-24 09:28:11 +02001764 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001765 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 *
1767 * Description:
1768 * blk_cleanup_queue is the pair to blk_init_queue() or
1769 * blk_queue_make_request(). It should be called when a request queue is
1770 * being released; typically when a block device is being de-registered.
1771 * Currently, its primary task it to free all the &struct request
1772 * structures that were allocated to the queue and the queue itself.
1773 *
1774 * Caveat:
1775 * Hopefully the low level driver will have finished any
1776 * outstanding requests first...
1777 **/
Al Viro483f4af2006-03-18 18:34:37 -05001778static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779{
Jens Axboe165125e2007-07-24 09:28:11 +02001780 struct request_queue *q =
1781 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 struct request_list *rl = &q->rq;
1783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 blk_sync_queue(q);
1785
1786 if (rl->rq_pool)
1787 mempool_destroy(rl->rq_pool);
1788
1789 if (q->queue_tags)
1790 __blk_queue_free_tags(q);
1791
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001792 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001793
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 kmem_cache_free(requestq_cachep, q);
1795}
1796
Jens Axboe165125e2007-07-24 09:28:11 +02001797void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001798{
1799 kobject_put(&q->kobj);
1800}
1801EXPORT_SYMBOL(blk_put_queue);
1802
Jens Axboe165125e2007-07-24 09:28:11 +02001803void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001804{
1805 mutex_lock(&q->sysfs_lock);
1806 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1807 mutex_unlock(&q->sysfs_lock);
1808
1809 if (q->elevator)
1810 elevator_exit(q->elevator);
1811
1812 blk_put_queue(q);
1813}
1814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815EXPORT_SYMBOL(blk_cleanup_queue);
1816
Jens Axboe165125e2007-07-24 09:28:11 +02001817static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
1819 struct request_list *rl = &q->rq;
1820
1821 rl->count[READ] = rl->count[WRITE] = 0;
1822 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001823 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 init_waitqueue_head(&rl->wait[READ]);
1825 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Christoph Lameter19460892005-06-23 00:08:19 -07001827 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1828 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830 if (!rl->rq_pool)
1831 return -ENOMEM;
1832
1833 return 0;
1834}
1835
Jens Axboe165125e2007-07-24 09:28:11 +02001836struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837{
Christoph Lameter19460892005-06-23 00:08:19 -07001838 return blk_alloc_queue_node(gfp_mask, -1);
1839}
1840EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
Al Viro483f4af2006-03-18 18:34:37 -05001842static struct kobj_type queue_ktype;
1843
Jens Axboe165125e2007-07-24 09:28:11 +02001844struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001845{
Jens Axboe165125e2007-07-24 09:28:11 +02001846 struct request_queue *q;
Christoph Lameter19460892005-06-23 00:08:19 -07001847
Christoph Lameter94f60302007-07-17 04:03:29 -07001848 q = kmem_cache_alloc_node(requestq_cachep,
1849 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 if (!q)
1851 return NULL;
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001854
1855 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1856 q->kobj.ktype = &queue_ktype;
1857 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1860 q->backing_dev_info.unplug_io_data = q;
1861
Al Viro483f4af2006-03-18 18:34:37 -05001862 mutex_init(&q->sysfs_lock);
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 return q;
1865}
Christoph Lameter19460892005-06-23 00:08:19 -07001866EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
1868/**
1869 * blk_init_queue - prepare a request queue for use with a block device
1870 * @rfn: The function to be called to process requests that have been
1871 * placed on the queue.
1872 * @lock: Request queue spin lock
1873 *
1874 * Description:
1875 * If a block device wishes to use the standard request handling procedures,
1876 * which sorts requests and coalesces adjacent requests, then it must
1877 * call blk_init_queue(). The function @rfn will be called when there
1878 * are requests on the queue that need to be processed. If the device
1879 * supports plugging, then @rfn may not be called immediately when requests
1880 * are available on the queue, but may be called at some time later instead.
1881 * Plugged queues are generally unplugged when a buffer belonging to one
1882 * of the requests on the queue is needed, or due to memory pressure.
1883 *
1884 * @rfn is not required, or even expected, to remove all requests off the
1885 * queue, but only as many as it can handle at a time. If it does leave
1886 * requests on the queue, it is responsible for arranging that the requests
1887 * get dealt with eventually.
1888 *
1889 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001890 * request queue; this lock will be taken also from interrupt context, so irq
1891 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 *
1893 * Function returns a pointer to the initialized request queue, or NULL if
1894 * it didn't succeed.
1895 *
1896 * Note:
1897 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1898 * when the block device is deactivated (such as at module unload).
1899 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001900
Jens Axboe165125e2007-07-24 09:28:11 +02001901struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Christoph Lameter19460892005-06-23 00:08:19 -07001903 return blk_init_queue_node(rfn, lock, -1);
1904}
1905EXPORT_SYMBOL(blk_init_queue);
1906
Jens Axboe165125e2007-07-24 09:28:11 +02001907struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001908blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1909{
Jens Axboe165125e2007-07-24 09:28:11 +02001910 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 if (!q)
1913 return NULL;
1914
Christoph Lameter19460892005-06-23 00:08:19 -07001915 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001916 if (blk_init_free_list(q)) {
1917 kmem_cache_free(requestq_cachep, q);
1918 return NULL;
1919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
152587d2005-04-12 16:22:06 -05001921 /*
1922 * if caller didn't supply a lock, they get per-queue locking with
1923 * our embedded lock
1924 */
1925 if (!lock) {
1926 spin_lock_init(&q->__queue_lock);
1927 lock = &q->__queue_lock;
1928 }
1929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 q->prep_rq_fn = NULL;
1932 q->unplug_fn = generic_unplug_device;
1933 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1934 q->queue_lock = lock;
1935
1936 blk_queue_segment_boundary(q, 0xffffffff);
1937
1938 blk_queue_make_request(q, __make_request);
1939 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1940
1941 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1942 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1943
Alan Stern44ec9542007-02-20 11:01:57 -05001944 q->sg_reserved_size = INT_MAX;
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 /*
1947 * all done
1948 */
1949 if (!elevator_init(q, NULL)) {
1950 blk_queue_congestion_threshold(q);
1951 return q;
1952 }
1953
Al Viro8669aaf2006-03-18 13:50:00 -05001954 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return NULL;
1956}
Christoph Lameter19460892005-06-23 00:08:19 -07001957EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
Jens Axboe165125e2007-07-24 09:28:11 +02001959int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001961 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001962 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 return 0;
1964 }
1965
1966 return 1;
1967}
1968
1969EXPORT_SYMBOL(blk_get_queue);
1970
Jens Axboe165125e2007-07-24 09:28:11 +02001971static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001973 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001974 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 mempool_free(rq, q->rq.rq_pool);
1976}
1977
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001978static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001979blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980{
1981 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1982
1983 if (!rq)
1984 return NULL;
1985
1986 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001987 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 * see bio.h and blkdev.h
1989 */
Jens Axboe49171e52006-08-10 08:59:11 +02001990 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Tejun Heocb98fc82005-10-28 08:29:39 +02001992 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001993 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001994 mempool_free(rq, q->rq.rq_pool);
1995 return NULL;
1996 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02001997 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02001998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Tejun Heocb98fc82005-10-28 08:29:39 +02002000 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001}
2002
2003/*
2004 * ioc_batching returns true if the ioc is a valid batching request and
2005 * should be given priority access to a request.
2006 */
Jens Axboe165125e2007-07-24 09:28:11 +02002007static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008{
2009 if (!ioc)
2010 return 0;
2011
2012 /*
2013 * Make sure the process is able to allocate at least 1 request
2014 * even if the batch times out, otherwise we could theoretically
2015 * lose wakeups.
2016 */
2017 return ioc->nr_batch_requests == q->nr_batching ||
2018 (ioc->nr_batch_requests > 0
2019 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2020}
2021
2022/*
2023 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2024 * will cause the process to be a "batcher" on all queues in the system. This
2025 * is the behaviour we want though - once it gets a wakeup it should be given
2026 * a nice run.
2027 */
Jens Axboe165125e2007-07-24 09:28:11 +02002028static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
2030 if (!ioc || ioc_batching(q, ioc))
2031 return;
2032
2033 ioc->nr_batch_requests = q->nr_batching;
2034 ioc->last_waited = jiffies;
2035}
2036
Jens Axboe165125e2007-07-24 09:28:11 +02002037static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038{
2039 struct request_list *rl = &q->rq;
2040
2041 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002042 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 if (waitqueue_active(&rl->wait[rw]))
2046 wake_up(&rl->wait[rw]);
2047
2048 blk_clear_queue_full(q, rw);
2049 }
2050}
2051
2052/*
2053 * A request has just been released. Account for it, update the full and
2054 * congestion status, wake up any waiters. Called under q->queue_lock.
2055 */
Jens Axboe165125e2007-07-24 09:28:11 +02002056static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057{
2058 struct request_list *rl = &q->rq;
2059
2060 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002061 if (priv)
2062 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063
2064 __freed_request(q, rw);
2065
2066 if (unlikely(rl->starved[rw ^ 1]))
2067 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068}
2069
2070#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2071/*
Nick Piggind6344532005-06-28 20:45:14 -07002072 * Get a free request, queue_lock must be held.
2073 * Returns NULL on failure, with queue_lock held.
2074 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 */
Jens Axboe165125e2007-07-24 09:28:11 +02002076static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002077 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078{
2079 struct request *rq = NULL;
2080 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002081 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002082 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002083 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084
Jens Axboe7749a8d2006-12-13 13:02:26 +01002085 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002086 if (may_queue == ELV_MQUEUE_NO)
2087 goto rq_starved;
2088
2089 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2090 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002091 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002092 /*
2093 * The queue will fill after this allocation, so set
2094 * it as full, and mark this process as "batching".
2095 * This process will be allowed to complete a batch of
2096 * requests, others will be blocked.
2097 */
2098 if (!blk_queue_full(q, rw)) {
2099 ioc_set_batching(q, ioc);
2100 blk_set_queue_full(q, rw);
2101 } else {
2102 if (may_queue != ELV_MQUEUE_MUST
2103 && !ioc_batching(q, ioc)) {
2104 /*
2105 * The queue is full and the allocating
2106 * process is not a "batcher", and not
2107 * exempted by the IO scheduler
2108 */
2109 goto out;
2110 }
2111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002113 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
2115
Jens Axboe082cf692005-06-28 16:35:11 +02002116 /*
2117 * Only allow batching queuers to allocate up to 50% over the defined
2118 * limit of requests, otherwise we could have thousands of requests
2119 * allocated with any setting of ->nr_requests
2120 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002121 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002122 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 rl->count[rw]++;
2125 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002126
Jens Axboe64521d12005-10-28 08:30:39 +02002127 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002128 if (priv)
2129 rl->elvpriv++;
2130
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 spin_unlock_irq(q->queue_lock);
2132
Jens Axboe7749a8d2006-12-13 13:02:26 +01002133 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002134 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 /*
2136 * Allocation failed presumably due to memory. Undo anything
2137 * we might have messed up.
2138 *
2139 * Allocating task should really be put onto the front of the
2140 * wait queue, but this is pretty rare.
2141 */
2142 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002143 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
2145 /*
2146 * in the very unlikely event that allocation failed and no
2147 * requests for this direction was pending, mark us starved
2148 * so that freeing of a request in the other direction will
2149 * notice us. another possible fix would be to split the
2150 * rq mempool into READ and WRITE
2151 */
2152rq_starved:
2153 if (unlikely(rl->count[rw] == 0))
2154 rl->starved[rw] = 1;
2155
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 goto out;
2157 }
2158
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002159 /*
2160 * ioc may be NULL here, and ioc_batching will be false. That's
2161 * OK, if the queue is under the request limit then requests need
2162 * not count toward the nr_batch_requests limit. There will always
2163 * be some limit enforced by BLK_BATCH_TIME.
2164 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 if (ioc_batching(q, ioc))
2166 ioc->nr_batch_requests--;
2167
2168 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002169
2170 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 return rq;
2173}
2174
2175/*
2176 * No available requests for this queue, unplug the device and wait for some
2177 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002178 *
2179 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 */
Jens Axboe165125e2007-07-24 09:28:11 +02002181static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002182 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002184 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 struct request *rq;
2186
Jens Axboe7749a8d2006-12-13 13:02:26 +01002187 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002188 while (!rq) {
2189 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 struct request_list *rl = &q->rq;
2191
2192 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2193 TASK_UNINTERRUPTIBLE);
2194
Jens Axboe7749a8d2006-12-13 13:02:26 +01002195 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
2197 if (!rq) {
2198 struct io_context *ioc;
2199
Jens Axboe2056a782006-03-23 20:00:26 +01002200 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2201
Nick Piggind6344532005-06-28 20:45:14 -07002202 __generic_unplug_device(q);
2203 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 io_schedule();
2205
2206 /*
2207 * After sleeping, we become a "batching" process and
2208 * will be able to allocate at least one request, and
2209 * up to a big batch of them for a small period time.
2210 * See ioc_batching, ioc_set_batching
2211 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002212 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002214
2215 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
2217 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
2220 return rq;
2221}
2222
Jens Axboe165125e2007-07-24 09:28:11 +02002223struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
2225 struct request *rq;
2226
2227 BUG_ON(rw != READ && rw != WRITE);
2228
Nick Piggind6344532005-06-28 20:45:14 -07002229 spin_lock_irq(q->queue_lock);
2230 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002231 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002232 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002233 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002234 if (!rq)
2235 spin_unlock_irq(q->queue_lock);
2236 }
2237 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238
2239 return rq;
2240}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241EXPORT_SYMBOL(blk_get_request);
2242
2243/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002244 * blk_start_queueing - initiate dispatch of requests to device
2245 * @q: request queue to kick into gear
2246 *
2247 * This is basically a helper to remove the need to know whether a queue
2248 * is plugged or not if someone just wants to initiate dispatch of requests
2249 * for this queue.
2250 *
2251 * The queue lock must be held with interrupts disabled.
2252 */
Jens Axboe165125e2007-07-24 09:28:11 +02002253void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002254{
2255 if (!blk_queue_plugged(q))
2256 q->request_fn(q);
2257 else
2258 __generic_unplug_device(q);
2259}
2260EXPORT_SYMBOL(blk_start_queueing);
2261
2262/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 * blk_requeue_request - put a request back on queue
2264 * @q: request queue where request should be inserted
2265 * @rq: request to be inserted
2266 *
2267 * Description:
2268 * Drivers often keep queueing requests until the hardware cannot accept
2269 * more, when that condition happens we need to put the request back
2270 * on the queue. Must be called with queue lock held.
2271 */
Jens Axboe165125e2007-07-24 09:28:11 +02002272void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273{
Jens Axboe2056a782006-03-23 20:00:26 +01002274 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 if (blk_rq_tagged(rq))
2277 blk_queue_end_tag(q, rq);
2278
2279 elv_requeue_request(q, rq);
2280}
2281
2282EXPORT_SYMBOL(blk_requeue_request);
2283
2284/**
2285 * blk_insert_request - insert a special request in to a request queue
2286 * @q: request queue where request should be inserted
2287 * @rq: request to be inserted
2288 * @at_head: insert request at head or tail of queue
2289 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 *
2291 * Description:
2292 * Many block devices need to execute commands asynchronously, so they don't
2293 * block the whole kernel from preemption during request execution. This is
2294 * accomplished normally by inserting aritficial requests tagged as
2295 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2296 * scheduled for actual execution by the request queue.
2297 *
2298 * We have the option of inserting the head or the tail of the queue.
2299 * Typically we use the tail for new ioctls and so forth. We use the head
2300 * of the queue for things like a QUEUE_FULL message from a device, or a
2301 * host that is unable to accept a particular command.
2302 */
Jens Axboe165125e2007-07-24 09:28:11 +02002303void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002304 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305{
Tejun Heo 867d1192005-04-24 02:06:05 -05002306 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 unsigned long flags;
2308
2309 /*
2310 * tell I/O scheduler that this isn't a regular read/write (ie it
2311 * must not attempt merges on this) and that it acts as a soft
2312 * barrier
2313 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002314 rq->cmd_type = REQ_TYPE_SPECIAL;
2315 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316
2317 rq->special = data;
2318
2319 spin_lock_irqsave(q->queue_lock, flags);
2320
2321 /*
2322 * If command is tagged, release the tag
2323 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002324 if (blk_rq_tagged(rq))
2325 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Tejun Heo 867d1192005-04-24 02:06:05 -05002327 drive_stat_acct(rq, rq->nr_sectors, 1);
2328 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002329 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 spin_unlock_irqrestore(q->queue_lock, flags);
2331}
2332
2333EXPORT_SYMBOL(blk_insert_request);
2334
Mike Christie0e75f902006-12-01 10:40:55 +01002335static int __blk_rq_unmap_user(struct bio *bio)
2336{
2337 int ret = 0;
2338
2339 if (bio) {
2340 if (bio_flagged(bio, BIO_USER_MAPPED))
2341 bio_unmap_user(bio);
2342 else
2343 ret = bio_uncopy_user(bio);
2344 }
2345
2346 return ret;
2347}
2348
Jens Axboe165125e2007-07-24 09:28:11 +02002349static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002350 void __user *ubuf, unsigned int len)
2351{
2352 unsigned long uaddr;
2353 struct bio *bio, *orig_bio;
2354 int reading, ret;
2355
2356 reading = rq_data_dir(rq) == READ;
2357
2358 /*
2359 * if alignment requirement is satisfied, map in user pages for
2360 * direct dma. else, set up kernel bounce buffers
2361 */
2362 uaddr = (unsigned long) ubuf;
2363 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2364 bio = bio_map_user(q, NULL, uaddr, len, reading);
2365 else
2366 bio = bio_copy_user(q, uaddr, len, reading);
2367
Jens Axboe29852592006-12-19 08:27:31 +01002368 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002369 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002370
2371 orig_bio = bio;
2372 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002373
Mike Christie0e75f902006-12-01 10:40:55 +01002374 /*
2375 * We link the bounce buffer in and could have to traverse it
2376 * later so we have to get a ref to prevent it from being freed
2377 */
2378 bio_get(bio);
2379
Mike Christie0e75f902006-12-01 10:40:55 +01002380 if (!rq->bio)
2381 blk_rq_bio_prep(q, rq, bio);
Jens Axboe1aa4f242006-12-19 08:33:11 +01002382 else if (!ll_back_merge_fn(q, rq, bio)) {
Mike Christie0e75f902006-12-01 10:40:55 +01002383 ret = -EINVAL;
Mike Christie0e75f902006-12-01 10:40:55 +01002384 goto unmap_bio;
2385 } else {
2386 rq->biotail->bi_next = bio;
2387 rq->biotail = bio;
2388
Mike Christie0e75f902006-12-01 10:40:55 +01002389 rq->data_len += bio->bi_size;
2390 }
Mike Christie0e75f902006-12-01 10:40:55 +01002391
2392 return bio->bi_size;
2393
2394unmap_bio:
2395 /* if it was boucned we must call the end io function */
2396 bio_endio(bio, bio->bi_size, 0);
2397 __blk_rq_unmap_user(orig_bio);
2398 bio_put(bio);
2399 return ret;
2400}
2401
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402/**
2403 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2404 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002405 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 * @ubuf: the user buffer
2407 * @len: length of user data
2408 *
2409 * Description:
2410 * Data will be mapped directly for zero copy io, if possible. Otherwise
2411 * a kernel bounce buffer is used.
2412 *
2413 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2414 * still in process context.
2415 *
2416 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2417 * before being submitted to the device, as pages mapped may be out of
2418 * reach. It's the callers responsibility to make sure this happens. The
2419 * original bio must be passed back in to blk_rq_unmap_user() for proper
2420 * unmapping.
2421 */
Jens Axboe165125e2007-07-24 09:28:11 +02002422int blk_rq_map_user(struct request_queue *q, struct request *rq,
2423 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424{
Mike Christie0e75f902006-12-01 10:40:55 +01002425 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002426 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002427 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Mike Christiedefd94b2005-12-05 02:37:06 -06002429 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002430 return -EINVAL;
2431 if (!len || !ubuf)
2432 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Mike Christie0e75f902006-12-01 10:40:55 +01002434 while (bytes_read != len) {
2435 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Mike Christie0e75f902006-12-01 10:40:55 +01002437 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2438 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2439 >> PAGE_SHIFT;
2440 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Mike Christie0e75f902006-12-01 10:40:55 +01002442 /*
2443 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2444 * pages. If this happens we just lower the requested
2445 * mapping len by a page so that we can fit
2446 */
2447 if (end - start > BIO_MAX_PAGES)
2448 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Mike Christie0e75f902006-12-01 10:40:55 +01002450 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2451 if (ret < 0)
2452 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002453 if (!bio)
2454 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002455 bytes_read += ret;
2456 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 }
2458
Mike Christie0e75f902006-12-01 10:40:55 +01002459 rq->buffer = rq->data = NULL;
2460 return 0;
2461unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002462 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002463 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464}
2465
2466EXPORT_SYMBOL(blk_rq_map_user);
2467
2468/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002469 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2470 * @q: request queue where request should be inserted
2471 * @rq: request to map data to
2472 * @iov: pointer to the iovec
2473 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002474 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002475 *
2476 * Description:
2477 * Data will be mapped directly for zero copy io, if possible. Otherwise
2478 * a kernel bounce buffer is used.
2479 *
2480 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2481 * still in process context.
2482 *
2483 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2484 * before being submitted to the device, as pages mapped may be out of
2485 * reach. It's the callers responsibility to make sure this happens. The
2486 * original bio must be passed back in to blk_rq_unmap_user() for proper
2487 * unmapping.
2488 */
Jens Axboe165125e2007-07-24 09:28:11 +02002489int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002490 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002491{
2492 struct bio *bio;
2493
2494 if (!iov || iov_count <= 0)
2495 return -EINVAL;
2496
2497 /* we don't allow misaligned data like bio_map_user() does. If the
2498 * user is using sg, they're expected to know the alignment constraints
2499 * and respect them accordingly */
2500 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2501 if (IS_ERR(bio))
2502 return PTR_ERR(bio);
2503
Mike Christie0e75f902006-12-01 10:40:55 +01002504 if (bio->bi_size != len) {
2505 bio_endio(bio, bio->bi_size, 0);
2506 bio_unmap_user(bio);
2507 return -EINVAL;
2508 }
2509
2510 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002511 blk_rq_bio_prep(q, rq, bio);
2512 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002513 return 0;
2514}
2515
2516EXPORT_SYMBOL(blk_rq_map_user_iov);
2517
2518/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002520 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 *
2522 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002523 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2524 * supply the original rq->bio from the blk_rq_map_user() return, since
2525 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002527int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002529 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002530 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002532 while (bio) {
2533 mapped_bio = bio;
2534 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002535 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
Jens Axboe48785bb2006-12-19 11:07:59 +01002537 ret2 = __blk_rq_unmap_user(mapped_bio);
2538 if (ret2 && !ret)
2539 ret = ret2;
2540
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002541 mapped_bio = bio;
2542 bio = bio->bi_next;
2543 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002544 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002545
2546 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547}
2548
2549EXPORT_SYMBOL(blk_rq_unmap_user);
2550
2551/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002552 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2553 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002554 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002555 * @kbuf: the kernel buffer
2556 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002557 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002558 */
Jens Axboe165125e2007-07-24 09:28:11 +02002559int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002560 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002561{
Mike Christie df46b9a2005-06-20 14:04:44 +02002562 struct bio *bio;
2563
Mike Christiedefd94b2005-12-05 02:37:06 -06002564 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002565 return -EINVAL;
2566 if (!len || !kbuf)
2567 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002568
2569 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002570 if (IS_ERR(bio))
2571 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002572
Jens Axboedd1cab92005-06-20 14:06:01 +02002573 if (rq_data_dir(rq) == WRITE)
2574 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002575
Jens Axboedd1cab92005-06-20 14:06:01 +02002576 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002577 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002578 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002579 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002580}
2581
2582EXPORT_SYMBOL(blk_rq_map_kern);
2583
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002584/**
2585 * blk_execute_rq_nowait - insert a request into queue for execution
2586 * @q: queue to insert the request in
2587 * @bd_disk: matching gendisk
2588 * @rq: request to insert
2589 * @at_head: insert request at head or tail of queue
2590 * @done: I/O completion handler
2591 *
2592 * Description:
2593 * Insert a fully prepared request at the back of the io scheduler queue
2594 * for execution. Don't wait for completion.
2595 */
Jens Axboe165125e2007-07-24 09:28:11 +02002596void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002597 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002598 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002599{
2600 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2601
2602 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002603 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002604 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002605 WARN_ON(irqs_disabled());
2606 spin_lock_irq(q->queue_lock);
2607 __elv_add_request(q, rq, where, 1);
2608 __generic_unplug_device(q);
2609 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002610}
Mike Christie6e39b692005-11-11 05:30:24 -06002611EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2612
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613/**
2614 * blk_execute_rq - insert a request into queue for execution
2615 * @q: queue to insert the request in
2616 * @bd_disk: matching gendisk
2617 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002618 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 *
2620 * Description:
2621 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002622 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 */
Jens Axboe165125e2007-07-24 09:28:11 +02002624int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002625 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002627 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 char sense[SCSI_SENSE_BUFFERSIZE];
2629 int err = 0;
2630
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 /*
2632 * we need an extra reference to the request, so we can look at
2633 * it after io completion
2634 */
2635 rq->ref_count++;
2636
2637 if (!rq->sense) {
2638 memset(sense, 0, sizeof(sense));
2639 rq->sense = sense;
2640 rq->sense_len = 0;
2641 }
2642
Jens Axboec00895a2006-09-30 20:29:12 +02002643 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002644 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646
2647 if (rq->errors)
2648 err = -EIO;
2649
2650 return err;
2651}
2652
2653EXPORT_SYMBOL(blk_execute_rq);
2654
2655/**
2656 * blkdev_issue_flush - queue a flush
2657 * @bdev: blockdev to issue flush for
2658 * @error_sector: error sector
2659 *
2660 * Description:
2661 * Issue a flush for the block device in question. Caller can supply
2662 * room for storing the error offset in case of a flush error, if they
2663 * wish to. Caller must run wait_for_completion() on its own.
2664 */
2665int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2666{
Jens Axboe165125e2007-07-24 09:28:11 +02002667 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
2669 if (bdev->bd_disk == NULL)
2670 return -ENXIO;
2671
2672 q = bdev_get_queue(bdev);
2673 if (!q)
2674 return -ENXIO;
2675 if (!q->issue_flush_fn)
2676 return -EOPNOTSUPP;
2677
2678 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2679}
2680
2681EXPORT_SYMBOL(blkdev_issue_flush);
2682
Adrian Bunk93d17d32005-06-25 14:59:10 -07002683static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684{
2685 int rw = rq_data_dir(rq);
2686
2687 if (!blk_fs_request(rq) || !rq->rq_disk)
2688 return;
2689
Jens Axboed72d9042005-11-01 08:35:42 +01002690 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002691 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002692 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 disk_round_stats(rq->rq_disk);
2694 rq->rq_disk->in_flight++;
2695 }
2696}
2697
2698/*
2699 * add-request adds a request to the linked list.
2700 * queue lock is held and interrupts disabled, as we muck with the
2701 * request queue list.
2702 */
Jens Axboe165125e2007-07-24 09:28:11 +02002703static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704{
2705 drive_stat_acct(req, req->nr_sectors, 1);
2706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 /*
2708 * elevator indicated where it wants this request to be
2709 * inserted at elevator_merge time
2710 */
2711 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2712}
2713
2714/*
2715 * disk_round_stats() - Round off the performance stats on a struct
2716 * disk_stats.
2717 *
2718 * The average IO queue length and utilisation statistics are maintained
2719 * by observing the current state of the queue length and the amount of
2720 * time it has been in this state for.
2721 *
2722 * Normally, that accounting is done on IO completion, but that can result
2723 * in more than a second's worth of IO being accounted for within any one
2724 * second, leading to >100% utilisation. To deal with that, we call this
2725 * function to do a round-off before returning the results when reading
2726 * /proc/diskstats. This accounts immediately for all queue usage up to
2727 * the current jiffies and restarts the counters again.
2728 */
2729void disk_round_stats(struct gendisk *disk)
2730{
2731 unsigned long now = jiffies;
2732
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002733 if (now == disk->stamp)
2734 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002736 if (disk->in_flight) {
2737 __disk_stat_add(disk, time_in_queue,
2738 disk->in_flight * (now - disk->stamp));
2739 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742}
2743
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002744EXPORT_SYMBOL_GPL(disk_round_stats);
2745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746/*
2747 * queue lock must be held
2748 */
Jens Axboe165125e2007-07-24 09:28:11 +02002749void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 if (unlikely(!q))
2752 return;
2753 if (unlikely(--req->ref_count))
2754 return;
2755
Tejun Heo8922e162005-10-20 16:23:44 +02002756 elv_completed_request(q, req);
2757
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 /*
2759 * Request may not have originated from ll_rw_blk. if not,
2760 * it didn't come out of our reserved rq pools
2761 */
Jens Axboe49171e52006-08-10 08:59:11 +02002762 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002764 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002767 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768
2769 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002770 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 }
2772}
2773
Mike Christie6e39b692005-11-11 05:30:24 -06002774EXPORT_SYMBOL_GPL(__blk_put_request);
2775
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776void blk_put_request(struct request *req)
2777{
Tejun Heo8922e162005-10-20 16:23:44 +02002778 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002779 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Tejun Heo8922e162005-10-20 16:23:44 +02002781 /*
2782 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2783 * following if (q) test.
2784 */
2785 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 spin_lock_irqsave(q->queue_lock, flags);
2787 __blk_put_request(q, req);
2788 spin_unlock_irqrestore(q->queue_lock, flags);
2789 }
2790}
2791
2792EXPORT_SYMBOL(blk_put_request);
2793
2794/**
2795 * blk_end_sync_rq - executes a completion event on a request
2796 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002797 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002799void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800{
Jens Axboec00895a2006-09-30 20:29:12 +02002801 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802
Jens Axboec00895a2006-09-30 20:29:12 +02002803 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 __blk_put_request(rq->q, rq);
2805
2806 /*
2807 * complete last, if this is a stack request the process (and thus
2808 * the rq pointer) could be invalid right after this complete()
2809 */
2810 complete(waiting);
2811}
2812EXPORT_SYMBOL(blk_end_sync_rq);
2813
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814/*
2815 * Has to be called with the request spinlock acquired
2816 */
Jens Axboe165125e2007-07-24 09:28:11 +02002817static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 struct request *next)
2819{
2820 if (!rq_mergeable(req) || !rq_mergeable(next))
2821 return 0;
2822
2823 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002824 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 */
2826 if (req->sector + req->nr_sectors != next->sector)
2827 return 0;
2828
2829 if (rq_data_dir(req) != rq_data_dir(next)
2830 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002831 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 return 0;
2833
2834 /*
2835 * If we are allowed to merge, then append bio list
2836 * from next to rq and release next. merge_requests_fn
2837 * will have updated segment counts, update sector
2838 * counts here.
2839 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002840 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 return 0;
2842
2843 /*
2844 * At this point we have either done a back merge
2845 * or front merge. We need the smaller start_time of
2846 * the merged requests to be the current request
2847 * for accounting purposes.
2848 */
2849 if (time_after(req->start_time, next->start_time))
2850 req->start_time = next->start_time;
2851
2852 req->biotail->bi_next = next->bio;
2853 req->biotail = next->biotail;
2854
2855 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2856
2857 elv_merge_requests(q, req, next);
2858
2859 if (req->rq_disk) {
2860 disk_round_stats(req->rq_disk);
2861 req->rq_disk->in_flight--;
2862 }
2863
Jens Axboe22e2c502005-06-27 10:55:12 +02002864 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2865
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 __blk_put_request(q, next);
2867 return 1;
2868}
2869
Jens Axboe165125e2007-07-24 09:28:11 +02002870static inline int attempt_back_merge(struct request_queue *q,
2871 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872{
2873 struct request *next = elv_latter_request(q, rq);
2874
2875 if (next)
2876 return attempt_merge(q, rq, next);
2877
2878 return 0;
2879}
2880
Jens Axboe165125e2007-07-24 09:28:11 +02002881static inline int attempt_front_merge(struct request_queue *q,
2882 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883{
2884 struct request *prev = elv_former_request(q, rq);
2885
2886 if (prev)
2887 return attempt_merge(q, prev, rq);
2888
2889 return 0;
2890}
2891
Tejun Heo52d9e672006-01-06 09:49:58 +01002892static void init_request_from_bio(struct request *req, struct bio *bio)
2893{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002894 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002895
2896 /*
2897 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2898 */
2899 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002900 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002901
2902 /*
2903 * REQ_BARRIER implies no merging, but lets make it explicit
2904 */
2905 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002906 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002907
Jens Axboeb31dc662006-06-13 08:26:10 +02002908 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002909 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002910 if (bio_rw_meta(bio))
2911 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002912
Tejun Heo52d9e672006-01-06 09:49:58 +01002913 req->errors = 0;
2914 req->hard_sector = req->sector = bio->bi_sector;
2915 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2916 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2917 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2918 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2919 req->buffer = bio_data(bio); /* see ->buffer comment above */
Tejun Heo52d9e672006-01-06 09:49:58 +01002920 req->bio = req->biotail = bio;
2921 req->ioprio = bio_prio(bio);
2922 req->rq_disk = bio->bi_bdev->bd_disk;
2923 req->start_time = jiffies;
2924}
2925
Jens Axboe165125e2007-07-24 09:28:11 +02002926static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927{
Nick Piggin450991b2005-06-28 20:45:13 -07002928 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002929 int el_ret, nr_sectors, barrier, err;
2930 const unsigned short prio = bio_prio(bio);
2931 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002932 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935
2936 /*
2937 * low level driver can indicate that it wants pages above a
2938 * certain limit bounced to low memory (ie for highmem, or even
2939 * ISA dma in theory)
2940 */
2941 blk_queue_bounce(q, &bio);
2942
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002944 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 err = -EOPNOTSUPP;
2946 goto end_io;
2947 }
2948
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 spin_lock_irq(q->queue_lock);
2950
Nick Piggin450991b2005-06-28 20:45:13 -07002951 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 goto get_rq;
2953
2954 el_ret = elv_merge(q, &req, bio);
2955 switch (el_ret) {
2956 case ELEVATOR_BACK_MERGE:
2957 BUG_ON(!rq_mergeable(req));
2958
Jens Axboe1aa4f242006-12-19 08:33:11 +01002959 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 break;
2961
Jens Axboe2056a782006-03-23 20:00:26 +01002962 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2963
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 req->biotail->bi_next = bio;
2965 req->biotail = bio;
2966 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002967 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 drive_stat_acct(req, nr_sectors, 0);
2969 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002970 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 goto out;
2972
2973 case ELEVATOR_FRONT_MERGE:
2974 BUG_ON(!rq_mergeable(req));
2975
Jens Axboe1aa4f242006-12-19 08:33:11 +01002976 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 break;
2978
Jens Axboe2056a782006-03-23 20:00:26 +01002979 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2980
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 bio->bi_next = req->bio;
2982 req->bio = bio;
2983
2984 /*
2985 * may not be valid. if the low level driver said
2986 * it didn't need a bounce buffer then it better
2987 * not touch req->buffer either...
2988 */
2989 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02002990 req->current_nr_sectors = bio_cur_sectors(bio);
2991 req->hard_cur_sectors = req->current_nr_sectors;
2992 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002994 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 drive_stat_acct(req, nr_sectors, 0);
2996 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002997 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 goto out;
2999
Nick Piggin450991b2005-06-28 20:45:13 -07003000 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003002 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 }
3004
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003006 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003007 * This sync check and mask will be re-done in init_request_from_bio(),
3008 * but we need to set it earlier to expose the sync flag to the
3009 * rq allocator and io schedulers.
3010 */
3011 rw_flags = bio_data_dir(bio);
3012 if (sync)
3013 rw_flags |= REQ_RW_SYNC;
3014
3015 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003016 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003017 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003018 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003019 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003020
Nick Piggin450991b2005-06-28 20:45:13 -07003021 /*
3022 * After dropping the lock and possibly sleeping here, our request
3023 * may now be mergeable after it had proven unmergeable (above).
3024 * We don't worry about that case for efficiency. It won't happen
3025 * often, and the elevators are able to handle it.
3026 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003027 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
Nick Piggin450991b2005-06-28 20:45:13 -07003029 spin_lock_irq(q->queue_lock);
3030 if (elv_queue_empty(q))
3031 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 add_request(q, req);
3033out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003034 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 __generic_unplug_device(q);
3036
3037 spin_unlock_irq(q->queue_lock);
3038 return 0;
3039
3040end_io:
3041 bio_endio(bio, nr_sectors << 9, err);
3042 return 0;
3043}
3044
3045/*
3046 * If bio->bi_dev is a partition, remap the location
3047 */
3048static inline void blk_partition_remap(struct bio *bio)
3049{
3050 struct block_device *bdev = bio->bi_bdev;
3051
3052 if (bdev != bdev->bd_contains) {
3053 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003054 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055
Jens Axboea3623572005-11-01 09:26:16 +01003056 p->sectors[rw] += bio_sectors(bio);
3057 p->ios[rw]++;
3058
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 bio->bi_sector += p->start_sect;
3060 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003061
3062 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3063 bdev->bd_dev, bio->bi_sector,
3064 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 }
3066}
3067
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068static void handle_bad_sector(struct bio *bio)
3069{
3070 char b[BDEVNAME_SIZE];
3071
3072 printk(KERN_INFO "attempt to access beyond end of device\n");
3073 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3074 bdevname(bio->bi_bdev, b),
3075 bio->bi_rw,
3076 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3077 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3078
3079 set_bit(BIO_EOF, &bio->bi_flags);
3080}
3081
Akinobu Mitac17bb492006-12-08 02:39:46 -08003082#ifdef CONFIG_FAIL_MAKE_REQUEST
3083
3084static DECLARE_FAULT_ATTR(fail_make_request);
3085
3086static int __init setup_fail_make_request(char *str)
3087{
3088 return setup_fault_attr(&fail_make_request, str);
3089}
3090__setup("fail_make_request=", setup_fail_make_request);
3091
3092static int should_fail_request(struct bio *bio)
3093{
3094 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3095 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3096 return should_fail(&fail_make_request, bio->bi_size);
3097
3098 return 0;
3099}
3100
3101static int __init fail_make_request_debugfs(void)
3102{
3103 return init_fault_attr_dentries(&fail_make_request,
3104 "fail_make_request");
3105}
3106
3107late_initcall(fail_make_request_debugfs);
3108
3109#else /* CONFIG_FAIL_MAKE_REQUEST */
3110
3111static inline int should_fail_request(struct bio *bio)
3112{
3113 return 0;
3114}
3115
3116#endif /* CONFIG_FAIL_MAKE_REQUEST */
3117
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118/**
3119 * generic_make_request: hand a buffer to its device driver for I/O
3120 * @bio: The bio describing the location in memory and on the device.
3121 *
3122 * generic_make_request() is used to make I/O requests of block
3123 * devices. It is passed a &struct bio, which describes the I/O that needs
3124 * to be done.
3125 *
3126 * generic_make_request() does not return any status. The
3127 * success/failure status of the request, along with notification of
3128 * completion, is delivered asynchronously through the bio->bi_end_io
3129 * function described (one day) else where.
3130 *
3131 * The caller of generic_make_request must make sure that bi_io_vec
3132 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3133 * set to describe the device address, and the
3134 * bi_end_io and optionally bi_private are set to describe how
3135 * completion notification should be signaled.
3136 *
3137 * generic_make_request and the drivers it calls may use bi_next if this
3138 * bio happens to be merged with someone else, and may change bi_dev and
3139 * bi_sector for remaps as it sees fit. So the values of these fields
3140 * should NOT be depended on after the call to generic_make_request.
3141 */
Neil Brownd89d8792007-05-01 09:53:42 +02003142static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143{
Jens Axboe165125e2007-07-24 09:28:11 +02003144 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 sector_t maxsector;
NeilBrown5ddfe962006-10-30 22:07:21 -08003146 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003148 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149
3150 might_sleep();
3151 /* Test device or partition size, when known. */
3152 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3153 if (maxsector) {
3154 sector_t sector = bio->bi_sector;
3155
3156 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3157 /*
3158 * This may well happen - the kernel calls bread()
3159 * without checking the size of the device, e.g., when
3160 * mounting a device.
3161 */
3162 handle_bad_sector(bio);
3163 goto end_io;
3164 }
3165 }
3166
3167 /*
3168 * Resolve the mapping until finished. (drivers are
3169 * still free to implement/resolve their own stacking
3170 * by explicitly returning 0)
3171 *
3172 * NOTE: we don't repeat the blk_size check for each new device.
3173 * Stacking drivers are expected to know what they are doing.
3174 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003175 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003176 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 do {
3178 char b[BDEVNAME_SIZE];
3179
3180 q = bdev_get_queue(bio->bi_bdev);
3181 if (!q) {
3182 printk(KERN_ERR
3183 "generic_make_request: Trying to access "
3184 "nonexistent block-device %s (%Lu)\n",
3185 bdevname(bio->bi_bdev, b),
3186 (long long) bio->bi_sector);
3187end_io:
3188 bio_endio(bio, bio->bi_size, -EIO);
3189 break;
3190 }
3191
3192 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3193 printk("bio too big device %s (%u > %u)\n",
3194 bdevname(bio->bi_bdev, b),
3195 bio_sectors(bio),
3196 q->max_hw_sectors);
3197 goto end_io;
3198 }
3199
Nick Pigginfde6ad22005-06-23 00:08:53 -07003200 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 goto end_io;
3202
Akinobu Mitac17bb492006-12-08 02:39:46 -08003203 if (should_fail_request(bio))
3204 goto end_io;
3205
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 /*
3207 * If this device has partitions, remap block n
3208 * of partition p to block n+start(p) of the disk.
3209 */
3210 blk_partition_remap(bio);
3211
NeilBrown5ddfe962006-10-30 22:07:21 -08003212 if (old_sector != -1)
Jens Axboe2056a782006-03-23 20:00:26 +01003213 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003214 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003215
3216 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3217
NeilBrown5ddfe962006-10-30 22:07:21 -08003218 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003219 old_dev = bio->bi_bdev->bd_dev;
3220
NeilBrown5ddfe962006-10-30 22:07:21 -08003221 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3222 if (maxsector) {
3223 sector_t sector = bio->bi_sector;
3224
Andrew Mortondf66b852006-11-02 22:06:56 -08003225 if (maxsector < nr_sectors ||
3226 maxsector - nr_sectors < sector) {
NeilBrown5ddfe962006-10-30 22:07:21 -08003227 /*
Andrew Mortondf66b852006-11-02 22:06:56 -08003228 * This may well happen - partitions are not
3229 * checked to make sure they are within the size
3230 * of the whole device.
NeilBrown5ddfe962006-10-30 22:07:21 -08003231 */
3232 handle_bad_sector(bio);
3233 goto end_io;
3234 }
3235 }
3236
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 ret = q->make_request_fn(q, bio);
3238 } while (ret);
3239}
3240
Neil Brownd89d8792007-05-01 09:53:42 +02003241/*
3242 * We only want one ->make_request_fn to be active at a time,
3243 * else stack usage with stacked devices could be a problem.
3244 * So use current->bio_{list,tail} to keep a list of requests
3245 * submited by a make_request_fn function.
3246 * current->bio_tail is also used as a flag to say if
3247 * generic_make_request is currently active in this task or not.
3248 * If it is NULL, then no make_request is active. If it is non-NULL,
3249 * then a make_request is active, and new requests should be added
3250 * at the tail
3251 */
3252void generic_make_request(struct bio *bio)
3253{
3254 if (current->bio_tail) {
3255 /* make_request is active */
3256 *(current->bio_tail) = bio;
3257 bio->bi_next = NULL;
3258 current->bio_tail = &bio->bi_next;
3259 return;
3260 }
3261 /* following loop may be a bit non-obvious, and so deserves some
3262 * explanation.
3263 * Before entering the loop, bio->bi_next is NULL (as all callers
3264 * ensure that) so we have a list with a single bio.
3265 * We pretend that we have just taken it off a longer list, so
3266 * we assign bio_list to the next (which is NULL) and bio_tail
3267 * to &bio_list, thus initialising the bio_list of new bios to be
3268 * added. __generic_make_request may indeed add some more bios
3269 * through a recursive call to generic_make_request. If it
3270 * did, we find a non-NULL value in bio_list and re-enter the loop
3271 * from the top. In this case we really did just take the bio
3272 * of the top of the list (no pretending) and so fixup bio_list and
3273 * bio_tail or bi_next, and call into __generic_make_request again.
3274 *
3275 * The loop was structured like this to make only one call to
3276 * __generic_make_request (which is important as it is large and
3277 * inlined) and to keep the structure simple.
3278 */
3279 BUG_ON(bio->bi_next);
3280 do {
3281 current->bio_list = bio->bi_next;
3282 if (bio->bi_next == NULL)
3283 current->bio_tail = &current->bio_list;
3284 else
3285 bio->bi_next = NULL;
3286 __generic_make_request(bio);
3287 bio = current->bio_list;
3288 } while (bio);
3289 current->bio_tail = NULL; /* deactivate */
3290}
3291
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292EXPORT_SYMBOL(generic_make_request);
3293
3294/**
3295 * submit_bio: submit a bio to the block device layer for I/O
3296 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3297 * @bio: The &struct bio which describes the I/O
3298 *
3299 * submit_bio() is very similar in purpose to generic_make_request(), and
3300 * uses that function to do most of the work. Both are fairly rough
3301 * interfaces, @bio must be presetup and ready for I/O.
3302 *
3303 */
3304void submit_bio(int rw, struct bio *bio)
3305{
3306 int count = bio_sectors(bio);
3307
3308 BIO_BUG_ON(!bio->bi_size);
3309 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003310 bio->bi_rw |= rw;
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003311 if (rw & WRITE) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07003312 count_vm_events(PGPGOUT, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003313 } else {
3314 task_io_account_read(bio->bi_size);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003315 count_vm_events(PGPGIN, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003316 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317
3318 if (unlikely(block_dump)) {
3319 char b[BDEVNAME_SIZE];
3320 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3321 current->comm, current->pid,
3322 (rw & WRITE) ? "WRITE" : "READ",
3323 (unsigned long long)bio->bi_sector,
3324 bdevname(bio->bi_bdev,b));
3325 }
3326
3327 generic_make_request(bio);
3328}
3329
3330EXPORT_SYMBOL(submit_bio);
3331
Adrian Bunk93d17d32005-06-25 14:59:10 -07003332static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333{
3334 struct bio *bio, *prevbio = NULL;
3335 int nr_phys_segs, nr_hw_segs;
3336 unsigned int phys_size, hw_size;
Jens Axboe165125e2007-07-24 09:28:11 +02003337 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338
3339 if (!rq->bio)
3340 return;
3341
3342 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3343 rq_for_each_bio(bio, rq) {
3344 /* Force bio hw/phys segs to be recalculated. */
3345 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3346
3347 nr_phys_segs += bio_phys_segments(q, bio);
3348 nr_hw_segs += bio_hw_segments(q, bio);
3349 if (prevbio) {
3350 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3351 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3352
3353 if (blk_phys_contig_segment(q, prevbio, bio) &&
3354 pseg <= q->max_segment_size) {
3355 nr_phys_segs--;
3356 phys_size += prevbio->bi_size + bio->bi_size;
3357 } else
3358 phys_size = 0;
3359
3360 if (blk_hw_contig_segment(q, prevbio, bio) &&
3361 hseg <= q->max_segment_size) {
3362 nr_hw_segs--;
3363 hw_size += prevbio->bi_size + bio->bi_size;
3364 } else
3365 hw_size = 0;
3366 }
3367 prevbio = bio;
3368 }
3369
3370 rq->nr_phys_segments = nr_phys_segs;
3371 rq->nr_hw_segments = nr_hw_segs;
3372}
3373
Adrian Bunk93d17d32005-06-25 14:59:10 -07003374static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375{
3376 if (blk_fs_request(rq)) {
3377 rq->hard_sector += nsect;
3378 rq->hard_nr_sectors -= nsect;
3379
3380 /*
3381 * Move the I/O submission pointers ahead if required.
3382 */
3383 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3384 (rq->sector <= rq->hard_sector)) {
3385 rq->sector = rq->hard_sector;
3386 rq->nr_sectors = rq->hard_nr_sectors;
3387 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3388 rq->current_nr_sectors = rq->hard_cur_sectors;
3389 rq->buffer = bio_data(rq->bio);
3390 }
3391
3392 /*
3393 * if total number of sectors is less than the first segment
3394 * size, something has gone terribly wrong
3395 */
3396 if (rq->nr_sectors < rq->current_nr_sectors) {
3397 printk("blk: request botched\n");
3398 rq->nr_sectors = rq->current_nr_sectors;
3399 }
3400 }
3401}
3402
3403static int __end_that_request_first(struct request *req, int uptodate,
3404 int nr_bytes)
3405{
3406 int total_bytes, bio_nbytes, error, next_idx = 0;
3407 struct bio *bio;
3408
Jens Axboe2056a782006-03-23 20:00:26 +01003409 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3410
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 /*
3412 * extend uptodate bool to allow < 0 value to be direct io error
3413 */
3414 error = 0;
3415 if (end_io_error(uptodate))
3416 error = !uptodate ? -EIO : uptodate;
3417
3418 /*
3419 * for a REQ_BLOCK_PC request, we want to carry any eventual
3420 * sense key with us all the way through
3421 */
3422 if (!blk_pc_request(req))
3423 req->errors = 0;
3424
3425 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003426 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 printk("end_request: I/O error, dev %s, sector %llu\n",
3428 req->rq_disk ? req->rq_disk->disk_name : "?",
3429 (unsigned long long)req->sector);
3430 }
3431
Jens Axboed72d9042005-11-01 08:35:42 +01003432 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003433 const int rw = rq_data_dir(req);
3434
Jens Axboe53e86062006-01-17 11:09:27 +01003435 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003436 }
3437
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 total_bytes = bio_nbytes = 0;
3439 while ((bio = req->bio) != NULL) {
3440 int nbytes;
3441
3442 if (nr_bytes >= bio->bi_size) {
3443 req->bio = bio->bi_next;
3444 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003445 if (!ordered_bio_endio(req, bio, nbytes, error))
3446 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 next_idx = 0;
3448 bio_nbytes = 0;
3449 } else {
3450 int idx = bio->bi_idx + next_idx;
3451
3452 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3453 blk_dump_rq_flags(req, "__end_that");
3454 printk("%s: bio idx %d >= vcnt %d\n",
3455 __FUNCTION__,
3456 bio->bi_idx, bio->bi_vcnt);
3457 break;
3458 }
3459
3460 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3461 BIO_BUG_ON(nbytes > bio->bi_size);
3462
3463 /*
3464 * not a complete bvec done
3465 */
3466 if (unlikely(nbytes > nr_bytes)) {
3467 bio_nbytes += nr_bytes;
3468 total_bytes += nr_bytes;
3469 break;
3470 }
3471
3472 /*
3473 * advance to the next vector
3474 */
3475 next_idx++;
3476 bio_nbytes += nbytes;
3477 }
3478
3479 total_bytes += nbytes;
3480 nr_bytes -= nbytes;
3481
3482 if ((bio = req->bio)) {
3483 /*
3484 * end more in this run, or just return 'not-done'
3485 */
3486 if (unlikely(nr_bytes <= 0))
3487 break;
3488 }
3489 }
3490
3491 /*
3492 * completely done
3493 */
3494 if (!req->bio)
3495 return 0;
3496
3497 /*
3498 * if the request wasn't completed, update state
3499 */
3500 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003501 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3502 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 bio->bi_idx += next_idx;
3504 bio_iovec(bio)->bv_offset += nr_bytes;
3505 bio_iovec(bio)->bv_len -= nr_bytes;
3506 }
3507
3508 blk_recalc_rq_sectors(req, total_bytes >> 9);
3509 blk_recalc_rq_segments(req);
3510 return 1;
3511}
3512
3513/**
3514 * end_that_request_first - end I/O on a request
3515 * @req: the request being processed
3516 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3517 * @nr_sectors: number of sectors to end I/O on
3518 *
3519 * Description:
3520 * Ends I/O on a number of sectors attached to @req, and sets it up
3521 * for the next range of segments (if any) in the cluster.
3522 *
3523 * Return:
3524 * 0 - we are done with this request, call end_that_request_last()
3525 * 1 - still buffers pending for this request
3526 **/
3527int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3528{
3529 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3530}
3531
3532EXPORT_SYMBOL(end_that_request_first);
3533
3534/**
3535 * end_that_request_chunk - end I/O on a request
3536 * @req: the request being processed
3537 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3538 * @nr_bytes: number of bytes to complete
3539 *
3540 * Description:
3541 * Ends I/O on a number of bytes attached to @req, and sets it up
3542 * for the next range of segments (if any). Like end_that_request_first(),
3543 * but deals with bytes instead of sectors.
3544 *
3545 * Return:
3546 * 0 - we are done with this request, call end_that_request_last()
3547 * 1 - still buffers pending for this request
3548 **/
3549int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3550{
3551 return __end_that_request_first(req, uptodate, nr_bytes);
3552}
3553
3554EXPORT_SYMBOL(end_that_request_chunk);
3555
3556/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003557 * splice the completion data to a local structure and hand off to
3558 * process_completion_queue() to complete the requests
3559 */
3560static void blk_done_softirq(struct softirq_action *h)
3561{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003562 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003563
3564 local_irq_disable();
3565 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003566 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003567 local_irq_enable();
3568
3569 while (!list_empty(&local_list)) {
3570 struct request *rq = list_entry(local_list.next, struct request, donelist);
3571
3572 list_del_init(&rq->donelist);
3573 rq->q->softirq_done_fn(rq);
3574 }
3575}
3576
Jens Axboeff856ba2006-01-09 16:02:34 +01003577static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3578 void *hcpu)
3579{
3580 /*
3581 * If a CPU goes away, splice its entries to the current CPU
3582 * and trigger a run of the softirq
3583 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003584 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003585 int cpu = (unsigned long) hcpu;
3586
3587 local_irq_disable();
3588 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3589 &__get_cpu_var(blk_cpu_done));
3590 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3591 local_irq_enable();
3592 }
3593
3594 return NOTIFY_OK;
3595}
3596
3597
Chandra Seetharaman054cc8a2006-06-27 02:54:07 -07003598static struct notifier_block __devinitdata blk_cpu_notifier = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003599 .notifier_call = blk_cpu_notify,
3600};
3601
Jens Axboeff856ba2006-01-09 16:02:34 +01003602/**
3603 * blk_complete_request - end I/O on a request
3604 * @req: the request being processed
3605 *
3606 * Description:
3607 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003608 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003609 * through requeueing. Theh actual completion happens out-of-order,
3610 * through a softirq handler. The user must have registered a completion
3611 * callback through blk_queue_softirq_done().
3612 **/
3613
3614void blk_complete_request(struct request *req)
3615{
3616 struct list_head *cpu_list;
3617 unsigned long flags;
3618
3619 BUG_ON(!req->q->softirq_done_fn);
3620
3621 local_irq_save(flags);
3622
3623 cpu_list = &__get_cpu_var(blk_cpu_done);
3624 list_add_tail(&req->donelist, cpu_list);
3625 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3626
3627 local_irq_restore(flags);
3628}
3629
3630EXPORT_SYMBOL(blk_complete_request);
3631
3632/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 * queue lock must be held
3634 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003635void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636{
3637 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003638 int error;
3639
3640 /*
3641 * extend uptodate bool to allow < 0 value to be direct io error
3642 */
3643 error = 0;
3644 if (end_io_error(uptodate))
3645 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646
3647 if (unlikely(laptop_mode) && blk_fs_request(req))
3648 laptop_io_completion();
3649
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003650 /*
3651 * Account IO completion. bar_rq isn't accounted as a normal
3652 * IO on queueing nor completion. Accounting the containing
3653 * request is enough.
3654 */
3655 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003657 const int rw = rq_data_dir(req);
3658
3659 __disk_stat_inc(disk, ios[rw]);
3660 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 disk_round_stats(disk);
3662 disk->in_flight--;
3663 }
3664 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003665 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666 else
3667 __blk_put_request(req->q, req);
3668}
3669
3670EXPORT_SYMBOL(end_that_request_last);
3671
3672void end_request(struct request *req, int uptodate)
3673{
3674 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3675 add_disk_randomness(req->rq_disk);
3676 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003677 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 }
3679}
3680
3681EXPORT_SYMBOL(end_request);
3682
Jens Axboe165125e2007-07-24 09:28:11 +02003683void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3684 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003686 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3687 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688
3689 rq->nr_phys_segments = bio_phys_segments(q, bio);
3690 rq->nr_hw_segments = bio_hw_segments(q, bio);
3691 rq->current_nr_sectors = bio_cur_sectors(bio);
3692 rq->hard_cur_sectors = rq->current_nr_sectors;
3693 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3694 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003695 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697 rq->bio = rq->biotail = bio;
3698}
3699
3700EXPORT_SYMBOL(blk_rq_bio_prep);
3701
3702int kblockd_schedule_work(struct work_struct *work)
3703{
3704 return queue_work(kblockd_workqueue, work);
3705}
3706
3707EXPORT_SYMBOL(kblockd_schedule_work);
3708
Andrew Morton19a75d82007-05-09 02:33:56 -07003709void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003711 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712}
Andrew Morton19a75d82007-05-09 02:33:56 -07003713EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714
3715int __init blk_dev_init(void)
3716{
Jens Axboeff856ba2006-01-09 16:02:34 +01003717 int i;
3718
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 kblockd_workqueue = create_workqueue("kblockd");
3720 if (!kblockd_workqueue)
3721 panic("Failed to create kblockd\n");
3722
3723 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003724 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725
3726 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003727 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728
3729 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003730 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003732 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003733 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3734
3735 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003736 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003737
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003738 blk_max_low_pfn = max_low_pfn - 1;
3739 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740
3741 return 0;
3742}
3743
3744/*
3745 * IO Context helper functions
3746 */
3747void put_io_context(struct io_context *ioc)
3748{
3749 if (ioc == NULL)
3750 return;
3751
3752 BUG_ON(atomic_read(&ioc->refcount) == 0);
3753
3754 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003755 struct cfq_io_context *cic;
3756
Al Viro334e94d2006-03-18 15:05:53 -05003757 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758 if (ioc->aic && ioc->aic->dtor)
3759 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003760 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003761 struct rb_node *n = rb_first(&ioc->cic_root);
3762
3763 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003764 cic->dtor(ioc);
3765 }
Al Viro334e94d2006-03-18 15:05:53 -05003766 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767
3768 kmem_cache_free(iocontext_cachep, ioc);
3769 }
3770}
3771EXPORT_SYMBOL(put_io_context);
3772
3773/* Called by the exitting task */
3774void exit_io_context(void)
3775{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003777 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
Jens Axboe22e2c502005-06-27 10:55:12 +02003779 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 ioc = current->io_context;
3781 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003782 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783
Oleg Nesterov25034d72006-08-29 09:15:14 +02003784 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 if (ioc->aic && ioc->aic->exit)
3786 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003787 if (ioc->cic_root.rb_node != NULL) {
3788 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3789 cic->exit(ioc);
3790 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003791
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 put_io_context(ioc);
3793}
3794
3795/*
3796 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003797 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003799 * This returned IO context doesn't have a specifically elevated refcount,
3800 * but since the current task itself holds a reference, the context can be
3801 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003803static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804{
3805 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 struct io_context *ret;
3807
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003809 if (likely(ret))
3810 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
Jens Axboeb5deef92006-07-19 23:39:40 +02003812 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813 if (ret) {
3814 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003815 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003816 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 ret->last_waited = jiffies; /* doesn't matter... */
3818 ret->nr_batch_requests = 0; /* because this is 0 */
3819 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003820 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003821 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003822 /* make sure set_task_ioprio() sees the settings above */
3823 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003824 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 }
3826
3827 return ret;
3828}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003829
3830/*
3831 * If the current task has no IO context then create one and initialise it.
3832 * If it does have a context, take a ref on it.
3833 *
3834 * This is always called in the context of the task which submitted the I/O.
3835 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003836struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003837{
3838 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003839 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003840 if (likely(ret))
3841 atomic_inc(&ret->refcount);
3842 return ret;
3843}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844EXPORT_SYMBOL(get_io_context);
3845
3846void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3847{
3848 struct io_context *src = *psrc;
3849 struct io_context *dst = *pdst;
3850
3851 if (src) {
3852 BUG_ON(atomic_read(&src->refcount) == 0);
3853 atomic_inc(&src->refcount);
3854 put_io_context(dst);
3855 *pdst = src;
3856 }
3857}
3858EXPORT_SYMBOL(copy_io_context);
3859
3860void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3861{
3862 struct io_context *temp;
3863 temp = *ioc1;
3864 *ioc1 = *ioc2;
3865 *ioc2 = temp;
3866}
3867EXPORT_SYMBOL(swap_io_context);
3868
3869/*
3870 * sysfs parts below
3871 */
3872struct queue_sysfs_entry {
3873 struct attribute attr;
3874 ssize_t (*show)(struct request_queue *, char *);
3875 ssize_t (*store)(struct request_queue *, const char *, size_t);
3876};
3877
3878static ssize_t
3879queue_var_show(unsigned int var, char *page)
3880{
3881 return sprintf(page, "%d\n", var);
3882}
3883
3884static ssize_t
3885queue_var_store(unsigned long *var, const char *page, size_t count)
3886{
3887 char *p = (char *) page;
3888
3889 *var = simple_strtoul(p, &p, 10);
3890 return count;
3891}
3892
3893static ssize_t queue_requests_show(struct request_queue *q, char *page)
3894{
3895 return queue_var_show(q->nr_requests, (page));
3896}
3897
3898static ssize_t
3899queue_requests_store(struct request_queue *q, const char *page, size_t count)
3900{
3901 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003902 unsigned long nr;
3903 int ret = queue_var_store(&nr, page, count);
3904 if (nr < BLKDEV_MIN_RQ)
3905 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003906
Al Viroc981ff92006-03-18 13:51:29 -05003907 spin_lock_irq(q->queue_lock);
3908 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909 blk_queue_congestion_threshold(q);
3910
3911 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003912 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003914 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915
3916 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003917 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003919 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920
3921 if (rl->count[READ] >= q->nr_requests) {
3922 blk_set_queue_full(q, READ);
3923 } else if (rl->count[READ]+1 <= q->nr_requests) {
3924 blk_clear_queue_full(q, READ);
3925 wake_up(&rl->wait[READ]);
3926 }
3927
3928 if (rl->count[WRITE] >= q->nr_requests) {
3929 blk_set_queue_full(q, WRITE);
3930 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3931 blk_clear_queue_full(q, WRITE);
3932 wake_up(&rl->wait[WRITE]);
3933 }
Al Viroc981ff92006-03-18 13:51:29 -05003934 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 return ret;
3936}
3937
3938static ssize_t queue_ra_show(struct request_queue *q, char *page)
3939{
3940 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3941
3942 return queue_var_show(ra_kb, (page));
3943}
3944
3945static ssize_t
3946queue_ra_store(struct request_queue *q, const char *page, size_t count)
3947{
3948 unsigned long ra_kb;
3949 ssize_t ret = queue_var_store(&ra_kb, page, count);
3950
3951 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3953 spin_unlock_irq(q->queue_lock);
3954
3955 return ret;
3956}
3957
3958static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3959{
3960 int max_sectors_kb = q->max_sectors >> 1;
3961
3962 return queue_var_show(max_sectors_kb, (page));
3963}
3964
3965static ssize_t
3966queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3967{
3968 unsigned long max_sectors_kb,
3969 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3970 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3971 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3972 int ra_kb;
3973
3974 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3975 return -EINVAL;
3976 /*
3977 * Take the queue lock to update the readahead and max_sectors
3978 * values synchronously:
3979 */
3980 spin_lock_irq(q->queue_lock);
3981 /*
3982 * Trim readahead window as well, if necessary:
3983 */
3984 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3985 if (ra_kb > max_sectors_kb)
3986 q->backing_dev_info.ra_pages =
3987 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3988
3989 q->max_sectors = max_sectors_kb << 1;
3990 spin_unlock_irq(q->queue_lock);
3991
3992 return ret;
3993}
3994
3995static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3996{
3997 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3998
3999 return queue_var_show(max_hw_sectors_kb, (page));
4000}
4001
4002
4003static struct queue_sysfs_entry queue_requests_entry = {
4004 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
4005 .show = queue_requests_show,
4006 .store = queue_requests_store,
4007};
4008
4009static struct queue_sysfs_entry queue_ra_entry = {
4010 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
4011 .show = queue_ra_show,
4012 .store = queue_ra_store,
4013};
4014
4015static struct queue_sysfs_entry queue_max_sectors_entry = {
4016 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
4017 .show = queue_max_sectors_show,
4018 .store = queue_max_sectors_store,
4019};
4020
4021static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
4022 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
4023 .show = queue_max_hw_sectors_show,
4024};
4025
4026static struct queue_sysfs_entry queue_iosched_entry = {
4027 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
4028 .show = elv_iosched_show,
4029 .store = elv_iosched_store,
4030};
4031
4032static struct attribute *default_attrs[] = {
4033 &queue_requests_entry.attr,
4034 &queue_ra_entry.attr,
4035 &queue_max_hw_sectors_entry.attr,
4036 &queue_max_sectors_entry.attr,
4037 &queue_iosched_entry.attr,
4038 NULL,
4039};
4040
4041#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4042
4043static ssize_t
4044queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4045{
4046 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004047 struct request_queue *q =
4048 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004049 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050
Linus Torvalds1da177e2005-04-16 15:20:36 -07004051 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004052 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004053 mutex_lock(&q->sysfs_lock);
4054 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4055 mutex_unlock(&q->sysfs_lock);
4056 return -ENOENT;
4057 }
4058 res = entry->show(q, page);
4059 mutex_unlock(&q->sysfs_lock);
4060 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061}
4062
4063static ssize_t
4064queue_attr_store(struct kobject *kobj, struct attribute *attr,
4065 const char *page, size_t length)
4066{
4067 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004068 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
Al Viro483f4af2006-03-18 18:34:37 -05004070 ssize_t res;
4071
Linus Torvalds1da177e2005-04-16 15:20:36 -07004072 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004073 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004074 mutex_lock(&q->sysfs_lock);
4075 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4076 mutex_unlock(&q->sysfs_lock);
4077 return -ENOENT;
4078 }
4079 res = entry->store(q, page, length);
4080 mutex_unlock(&q->sysfs_lock);
4081 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082}
4083
4084static struct sysfs_ops queue_sysfs_ops = {
4085 .show = queue_attr_show,
4086 .store = queue_attr_store,
4087};
4088
Adrian Bunk93d17d32005-06-25 14:59:10 -07004089static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 .sysfs_ops = &queue_sysfs_ops,
4091 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004092 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093};
4094
4095int blk_register_queue(struct gendisk *disk)
4096{
4097 int ret;
4098
Jens Axboe165125e2007-07-24 09:28:11 +02004099 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100
4101 if (!q || !q->request_fn)
4102 return -ENXIO;
4103
4104 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105
Al Viro483f4af2006-03-18 18:34:37 -05004106 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 if (ret < 0)
4108 return ret;
4109
Al Viro483f4af2006-03-18 18:34:37 -05004110 kobject_uevent(&q->kobj, KOBJ_ADD);
4111
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112 ret = elv_register_queue(q);
4113 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004114 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4115 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 return ret;
4117 }
4118
4119 return 0;
4120}
4121
4122void blk_unregister_queue(struct gendisk *disk)
4123{
Jens Axboe165125e2007-07-24 09:28:11 +02004124 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125
4126 if (q && q->request_fn) {
4127 elv_unregister_queue(q);
4128
Al Viro483f4af2006-03-18 18:34:37 -05004129 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4130 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004131 kobject_put(&disk->kobj);
4132 }
4133}