blob: cd9d2c5d91ae22664081e83136f9da4f3a9aae4c [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);
NeilBrown9dfa5282007-08-16 13:27:52 +020045static void blk_recalc_rq_segments(struct request *rq);
NeilBrown66846572007-08-16 13:31:28 +020046static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
47 struct bio *bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
50 * For the allocated request tables
51 */
Christoph Lametere18b8902006-12-06 20:33:20 -080052static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * For queue allocation
56 */
Christoph Lametere18b8902006-12-06 20:33:20 -080057static struct kmem_cache *requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59/*
60 * For io context allocations
61 */
Christoph Lametere18b8902006-12-06 20:33:20 -080062static struct kmem_cache *iocontext_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064/*
65 * Controlling structure to kblockd
66 */
Jens Axboeff856ba2006-01-09 16:02:34 +010067static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69unsigned long blk_max_low_pfn, blk_max_pfn;
70
71EXPORT_SYMBOL(blk_max_low_pfn);
72EXPORT_SYMBOL(blk_max_pfn);
73
Jens Axboeff856ba2006-01-09 16:02:34 +010074static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* Amount of time in which a process may batch requests */
77#define BLK_BATCH_TIME (HZ/50UL)
78
79/* Number of requests a "batching" process may submit */
80#define BLK_BATCH_REQ 32
81
82/*
83 * Return the threshold (number of used requests) at which the queue is
84 * considered to be congested. It include a little hysteresis to keep the
85 * context switch rate down.
86 */
87static inline int queue_congestion_on_threshold(struct request_queue *q)
88{
89 return q->nr_congestion_on;
90}
91
92/*
93 * The threshold at which a queue is considered to be uncongested
94 */
95static inline int queue_congestion_off_threshold(struct request_queue *q)
96{
97 return q->nr_congestion_off;
98}
99
100static void blk_queue_congestion_threshold(struct request_queue *q)
101{
102 int nr;
103
104 nr = q->nr_requests - (q->nr_requests / 8) + 1;
105 if (nr > q->nr_requests)
106 nr = q->nr_requests;
107 q->nr_congestion_on = nr;
108
109 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
110 if (nr < 1)
111 nr = 1;
112 q->nr_congestion_off = nr;
113}
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/**
116 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
117 * @bdev: device
118 *
119 * Locates the passed device's request queue and returns the address of its
120 * backing_dev_info
121 *
122 * Will return NULL if the request queue cannot be located.
123 */
124struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
125{
126 struct backing_dev_info *ret = NULL;
Jens Axboe165125e2007-07-24 09:28:11 +0200127 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129 if (q)
130 ret = &q->backing_dev_info;
131 return ret;
132}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133EXPORT_SYMBOL(blk_get_backing_dev_info);
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/**
136 * blk_queue_prep_rq - set a prepare_request function for queue
137 * @q: queue
138 * @pfn: prepare_request function
139 *
140 * It's possible for a queue to register a prepare_request callback which
141 * is invoked before the request is handed to the request_fn. The goal of
142 * the function is to prepare a request for I/O, it can be used to build a
143 * cdb from the request data for instance.
144 *
145 */
Jens Axboe165125e2007-07-24 09:28:11 +0200146void blk_queue_prep_rq(struct request_queue *q, prep_rq_fn *pfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147{
148 q->prep_rq_fn = pfn;
149}
150
151EXPORT_SYMBOL(blk_queue_prep_rq);
152
153/**
154 * blk_queue_merge_bvec - set a merge_bvec function for queue
155 * @q: queue
156 * @mbfn: merge_bvec_fn
157 *
158 * Usually queues have static limitations on the max sectors or segments that
159 * we can put in a request. Stacking drivers may have some settings that
160 * are dynamic, and thus we have to query the queue whether it is ok to
161 * add a new bio_vec to a bio at a given offset or not. If the block device
162 * has such limitations, it needs to register a merge_bvec_fn to control
163 * the size of bio's sent to it. Note that a block device *must* allow a
164 * single page to be added to an empty bio. The block device driver may want
165 * to use the bio_split() function to deal with these bio's. By default
166 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
167 * honored.
168 */
Jens Axboe165125e2007-07-24 09:28:11 +0200169void blk_queue_merge_bvec(struct request_queue *q, merge_bvec_fn *mbfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
171 q->merge_bvec_fn = mbfn;
172}
173
174EXPORT_SYMBOL(blk_queue_merge_bvec);
175
Jens Axboe165125e2007-07-24 09:28:11 +0200176void blk_queue_softirq_done(struct request_queue *q, softirq_done_fn *fn)
Jens Axboeff856ba2006-01-09 16:02:34 +0100177{
178 q->softirq_done_fn = fn;
179}
180
181EXPORT_SYMBOL(blk_queue_softirq_done);
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/**
184 * blk_queue_make_request - define an alternate make_request function for a device
185 * @q: the request queue for the device to be affected
186 * @mfn: the alternate make_request function
187 *
188 * Description:
189 * The normal way for &struct bios to be passed to a device
190 * driver is for them to be collected into requests on a request
191 * queue, and then to allow the device driver to select requests
192 * off that queue when it is ready. This works well for many block
193 * devices. However some block devices (typically virtual devices
194 * such as md or lvm) do not benefit from the processing on the
195 * request queue, and are served best by having the requests passed
196 * directly to them. This can be achieved by providing a function
197 * to blk_queue_make_request().
198 *
199 * Caveat:
200 * The driver that does this *must* be able to deal appropriately
201 * with buffers in "highmemory". This can be accomplished by either calling
202 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
203 * blk_queue_bounce() to create a buffer in normal memory.
204 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200205void blk_queue_make_request(struct request_queue * q, make_request_fn * mfn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 /*
208 * set defaults
209 */
210 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700211 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
212 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 q->make_request_fn = mfn;
214 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
215 q->backing_dev_info.state = 0;
216 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600217 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 blk_queue_hardsect_size(q, 512);
219 blk_queue_dma_alignment(q, 511);
220 blk_queue_congestion_threshold(q);
221 q->nr_batching = BLK_BATCH_REQ;
222
223 q->unplug_thresh = 4; /* hmm */
224 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
225 if (q->unplug_delay == 0)
226 q->unplug_delay = 1;
227
David Howells65f27f32006-11-22 14:55:48 +0000228 INIT_WORK(&q->unplug_work, blk_unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 q->unplug_timer.function = blk_unplug_timeout;
231 q->unplug_timer.data = (unsigned long)q;
232
233 /*
234 * by default assume old behaviour and bounce for any highmem page
235 */
236 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
239EXPORT_SYMBOL(blk_queue_make_request);
240
Jens Axboe165125e2007-07-24 09:28:11 +0200241static void rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
243 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100244 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200248 INIT_HLIST_NODE(&rq->hash);
249 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200250 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 rq->buffer = NULL;
252 rq->ref_count = 1;
253 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 rq->special = NULL;
255 rq->data_len = 0;
256 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200257 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 rq->sense = NULL;
259 rq->end_io = NULL;
260 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100261 rq->completion_data = NULL;
FUJITA Tomonoriabae1fd2007-07-16 08:52:14 +0200262 rq->next_rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263}
264
265/**
266 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100267 * @q: the request queue
268 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100269 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 *
271 * Description:
272 * For journalled file systems, doing ordered writes on a commit
273 * block instead of explicitly doing wait_on_buffer (which is bad
274 * for performance) can be a big win. Block drivers supporting this
275 * feature should call this function and indicate so.
276 *
277 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200278int blk_queue_ordered(struct request_queue *q, unsigned ordered,
Tejun Heo797e7db2006-01-06 09:51:03 +0100279 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Tejun Heo797e7db2006-01-06 09:51:03 +0100281 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
282 prepare_flush_fn == NULL) {
283 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
284 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100286
287 if (ordered != QUEUE_ORDERED_NONE &&
288 ordered != QUEUE_ORDERED_DRAIN &&
289 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
290 ordered != QUEUE_ORDERED_DRAIN_FUA &&
291 ordered != QUEUE_ORDERED_TAG &&
292 ordered != QUEUE_ORDERED_TAG_FLUSH &&
293 ordered != QUEUE_ORDERED_TAG_FUA) {
294 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
295 return -EINVAL;
296 }
297
Tetsuo Takata60481b12006-01-24 10:34:36 +0100298 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100299 q->next_ordered = ordered;
300 q->prepare_flush_fn = prepare_flush_fn;
301
302 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303}
304
305EXPORT_SYMBOL(blk_queue_ordered);
306
307/**
308 * blk_queue_issue_flush_fn - set function for issuing a flush
309 * @q: the request queue
310 * @iff: the function to be called issuing the flush
311 *
312 * Description:
313 * If a driver supports issuing a flush command, the support is notified
314 * to the block layer by defining it through this call.
315 *
316 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200317void blk_queue_issue_flush_fn(struct request_queue *q, issue_flush_fn *iff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 q->issue_flush_fn = iff;
320}
321
322EXPORT_SYMBOL(blk_queue_issue_flush_fn);
323
324/*
325 * Cache flushing for ordered writes handling
326 */
Jens Axboe165125e2007-07-24 09:28:11 +0200327inline unsigned blk_ordered_cur_seq(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
Tejun Heo797e7db2006-01-06 09:51:03 +0100329 if (!q->ordseq)
330 return 0;
331 return 1 << ffz(q->ordseq);
332}
333
334unsigned blk_ordered_req_seq(struct request *rq)
335{
Jens Axboe165125e2007-07-24 09:28:11 +0200336 struct request_queue *q = rq->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Tejun Heo797e7db2006-01-06 09:51:03 +0100338 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200339
Tejun Heo797e7db2006-01-06 09:51:03 +0100340 if (rq == &q->pre_flush_rq)
341 return QUEUE_ORDSEQ_PREFLUSH;
342 if (rq == &q->bar_rq)
343 return QUEUE_ORDSEQ_BAR;
344 if (rq == &q->post_flush_rq)
345 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Tejun Heobc90ba02007-06-15 13:24:28 +0200347 /*
348 * !fs requests don't need to follow barrier ordering. Always
349 * put them at the front. This fixes the following deadlock.
350 *
351 * http://thread.gmane.org/gmane.linux.kernel/537473
352 */
353 if (!blk_fs_request(rq))
354 return QUEUE_ORDSEQ_DRAIN;
355
Jens Axboe4aff5e22006-08-10 08:44:47 +0200356 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
357 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100358 return QUEUE_ORDSEQ_DRAIN;
359 else
360 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
Jens Axboe165125e2007-07-24 09:28:11 +0200363void blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Tejun Heo797e7db2006-01-06 09:51:03 +0100365 struct request *rq;
366 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Tejun Heo797e7db2006-01-06 09:51:03 +0100368 if (error && !q->orderr)
369 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200370
Tejun Heo797e7db2006-01-06 09:51:03 +0100371 BUG_ON(q->ordseq & seq);
372 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Tejun Heo797e7db2006-01-06 09:51:03 +0100374 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
375 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
377 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100378 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100380 rq = q->orig_bar_rq;
381 uptodate = q->orderr ? q->orderr : 1;
382
383 q->ordseq = 0;
384
385 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
386 end_that_request_last(rq, uptodate);
387}
388
389static void pre_flush_end_io(struct request *rq, int error)
390{
391 elv_completed_request(rq->q, rq);
392 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
393}
394
395static void bar_end_io(struct request *rq, int error)
396{
397 elv_completed_request(rq->q, rq);
398 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
399}
400
401static void post_flush_end_io(struct request *rq, int error)
402{
403 elv_completed_request(rq->q, rq);
404 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
405}
406
Jens Axboe165125e2007-07-24 09:28:11 +0200407static void queue_flush(struct request_queue *q, unsigned which)
Tejun Heo797e7db2006-01-06 09:51:03 +0100408{
409 struct request *rq;
410 rq_end_io_fn *end_io;
411
412 if (which == QUEUE_ORDERED_PREFLUSH) {
413 rq = &q->pre_flush_rq;
414 end_io = pre_flush_end_io;
415 } else {
416 rq = &q->post_flush_rq;
417 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
419
Jens Axboe4aff5e22006-08-10 08:44:47 +0200420 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100421 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100422 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200423 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100424 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100425 rq->end_io = end_io;
426 q->prepare_flush_fn(q, rq);
427
Tejun Heo30e96562006-02-08 01:01:31 -0800428 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100429}
430
Jens Axboe165125e2007-07-24 09:28:11 +0200431static inline struct request *start_ordered(struct request_queue *q,
Tejun Heo797e7db2006-01-06 09:51:03 +0100432 struct request *rq)
433{
Tejun Heo797e7db2006-01-06 09:51:03 +0100434 q->orderr = 0;
435 q->ordered = q->next_ordered;
436 q->ordseq |= QUEUE_ORDSEQ_STARTED;
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100439 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100441 blkdev_dequeue_request(rq);
442 q->orig_bar_rq = rq;
443 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200444 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100445 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200446 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
447 rq->cmd_flags |= REQ_RW;
448 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200450 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100451 init_request_from_bio(rq, q->orig_bar_rq->bio);
452 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
Tejun Heo797e7db2006-01-06 09:51:03 +0100454 /*
455 * Queue ordered sequence. As we stack them at the head, we
456 * need to queue in reverse order. Note that we rely on that
457 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
458 * request gets inbetween ordered sequence.
459 */
460 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
461 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
462 else
463 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Tejun Heo30e96562006-02-08 01:01:31 -0800465 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100466
467 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
468 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
469 rq = &q->pre_flush_rq;
470 } else
471 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
472
473 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
474 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
475 else
476 rq = NULL;
477
478 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
Jens Axboe165125e2007-07-24 09:28:11 +0200481int blk_do_ordered(struct request_queue *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800483 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100484 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Tejun Heo797e7db2006-01-06 09:51:03 +0100486 if (!q->ordseq) {
487 if (!is_barrier)
488 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Tejun Heo797e7db2006-01-06 09:51:03 +0100490 if (q->next_ordered != QUEUE_ORDERED_NONE) {
491 *rqp = start_ordered(q, rq);
492 return 1;
493 } else {
494 /*
495 * This can happen when the queue switches to
496 * ORDERED_NONE while this request is on it.
497 */
498 blkdev_dequeue_request(rq);
499 end_that_request_first(rq, -EOPNOTSUPP,
500 rq->hard_nr_sectors);
501 end_that_request_last(rq, -EOPNOTSUPP);
502 *rqp = NULL;
503 return 0;
504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800507 /*
508 * Ordered sequence in progress
509 */
510
511 /* Special requests are not subject to ordering rules. */
512 if (!blk_fs_request(rq) &&
513 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
514 return 1;
515
Tejun Heo797e7db2006-01-06 09:51:03 +0100516 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800517 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100518 if (is_barrier && rq != &q->bar_rq)
519 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800520 } else {
521 /* Ordered by draining. Wait for turn. */
522 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
523 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
524 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 }
526
527 return 1;
528}
529
NeilBrown5bb23a62007-09-27 12:46:13 +0200530static void req_bio_endio(struct request *rq, struct bio *bio,
531 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100532{
Jens Axboe165125e2007-07-24 09:28:11 +0200533 struct request_queue *q = rq->q;
Tejun Heo797e7db2006-01-06 09:51:03 +0100534
NeilBrown5bb23a62007-09-27 12:46:13 +0200535 if (&q->bar_rq != rq) {
536 if (error)
537 clear_bit(BIO_UPTODATE, &bio->bi_flags);
538 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
539 error = -EIO;
Tejun Heo797e7db2006-01-06 09:51:03 +0100540
NeilBrown5bb23a62007-09-27 12:46:13 +0200541 if (unlikely(nbytes > bio->bi_size)) {
542 printk("%s: want %u bytes done, only %u left\n",
543 __FUNCTION__, nbytes, bio->bi_size);
544 nbytes = bio->bi_size;
545 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100546
NeilBrown5bb23a62007-09-27 12:46:13 +0200547 bio->bi_size -= nbytes;
548 bio->bi_sector += (nbytes >> 9);
549 if (bio->bi_size == 0)
NeilBrown6712ecf2007-09-27 12:47:43 +0200550 bio_endio(bio, error);
NeilBrown5bb23a62007-09-27 12:46:13 +0200551 } else {
552
553 /*
554 * Okay, this is the barrier request in progress, just
555 * record the error;
556 */
557 if (error && !q->orderr)
558 q->orderr = error;
559 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100560}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562/**
563 * blk_queue_bounce_limit - set bounce buffer limit for queue
564 * @q: the request queue for the device
565 * @dma_addr: bus address limit
566 *
567 * Description:
568 * Different hardware can have different requirements as to what pages
569 * it can do I/O directly to. A low level driver can call
570 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800571 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200573void blk_queue_bounce_limit(struct request_queue *q, u64 dma_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
575 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800576 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Andi Kleen5ee1af92006-03-08 17:57:26 -0800578 q->bounce_gfp = GFP_NOIO;
579#if BITS_PER_LONG == 64
580 /* Assume anything <= 4GB can be handled by IOMMU.
581 Actually some IOMMUs can handle everything, but I don't
582 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200583 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800584 dma = 1;
585 q->bounce_pfn = max_low_pfn;
586#else
587 if (bounce_pfn < blk_max_low_pfn)
588 dma = 1;
589 q->bounce_pfn = bounce_pfn;
590#endif
591 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 init_emergency_isa_pool();
593 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800594 q->bounce_pfn = bounce_pfn;
595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
598EXPORT_SYMBOL(blk_queue_bounce_limit);
599
600/**
601 * blk_queue_max_sectors - set max sectors for a request for this queue
602 * @q: the request queue for the device
603 * @max_sectors: max sectors in the usual 512b unit
604 *
605 * Description:
606 * Enables a low level driver to set an upper limit on the size of
607 * received requests.
608 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200609void blk_queue_max_sectors(struct request_queue *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610{
611 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
612 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
613 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
614 }
615
Mike Christiedefd94b2005-12-05 02:37:06 -0600616 if (BLK_DEF_MAX_SECTORS > max_sectors)
617 q->max_hw_sectors = q->max_sectors = max_sectors;
618 else {
619 q->max_sectors = BLK_DEF_MAX_SECTORS;
620 q->max_hw_sectors = max_sectors;
621 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
624EXPORT_SYMBOL(blk_queue_max_sectors);
625
626/**
627 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
628 * @q: the request queue for the device
629 * @max_segments: max number of segments
630 *
631 * Description:
632 * Enables a low level driver to set an upper limit on the number of
633 * physical data segments in a request. This would be the largest sized
634 * scatter list the driver could handle.
635 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200636void blk_queue_max_phys_segments(struct request_queue *q,
637 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
639 if (!max_segments) {
640 max_segments = 1;
641 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
642 }
643
644 q->max_phys_segments = max_segments;
645}
646
647EXPORT_SYMBOL(blk_queue_max_phys_segments);
648
649/**
650 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
651 * @q: the request queue for the device
652 * @max_segments: max number of segments
653 *
654 * Description:
655 * Enables a low level driver to set an upper limit on the number of
656 * hw data segments in a request. This would be the largest number of
657 * address/length pairs the host adapter can actually give as once
658 * to the device.
659 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200660void blk_queue_max_hw_segments(struct request_queue *q,
661 unsigned short max_segments)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 if (!max_segments) {
664 max_segments = 1;
665 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
666 }
667
668 q->max_hw_segments = max_segments;
669}
670
671EXPORT_SYMBOL(blk_queue_max_hw_segments);
672
673/**
674 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
675 * @q: the request queue for the device
676 * @max_size: max size of segment in bytes
677 *
678 * Description:
679 * Enables a low level driver to set an upper limit on the size of a
680 * coalesced segment
681 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200682void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 if (max_size < PAGE_CACHE_SIZE) {
685 max_size = PAGE_CACHE_SIZE;
686 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
687 }
688
689 q->max_segment_size = max_size;
690}
691
692EXPORT_SYMBOL(blk_queue_max_segment_size);
693
694/**
695 * blk_queue_hardsect_size - set hardware sector size for the queue
696 * @q: the request queue for the device
697 * @size: the hardware sector size, in bytes
698 *
699 * Description:
700 * This should typically be set to the lowest possible sector size
701 * that the hardware can operate on (possible without reverting to
702 * even internal read-modify-write operations). Usually the default
703 * of 512 covers most hardware.
704 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200705void blk_queue_hardsect_size(struct request_queue *q, unsigned short size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 q->hardsect_size = size;
708}
709
710EXPORT_SYMBOL(blk_queue_hardsect_size);
711
712/*
713 * Returns the minimum that is _not_ zero, unless both are zero.
714 */
715#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
716
717/**
718 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
719 * @t: the stacking driver (top)
720 * @b: the underlying device (bottom)
721 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200722void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
724 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600725 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
726 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
729 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
730 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
731 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800732 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
733 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734}
735
736EXPORT_SYMBOL(blk_queue_stack_limits);
737
738/**
739 * blk_queue_segment_boundary - set boundary rules for segment merging
740 * @q: the request queue for the device
741 * @mask: the memory boundary mask
742 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200743void blk_queue_segment_boundary(struct request_queue *q, unsigned long mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 if (mask < PAGE_CACHE_SIZE - 1) {
746 mask = PAGE_CACHE_SIZE - 1;
747 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
748 }
749
750 q->seg_boundary_mask = mask;
751}
752
753EXPORT_SYMBOL(blk_queue_segment_boundary);
754
755/**
756 * blk_queue_dma_alignment - set dma length and memory alignment
757 * @q: the request queue for the device
758 * @mask: alignment mask
759 *
760 * description:
761 * set required memory and length aligment for direct dma transactions.
762 * this is used when buiding direct io requests for the queue.
763 *
764 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200765void blk_queue_dma_alignment(struct request_queue *q, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
767 q->dma_alignment = mask;
768}
769
770EXPORT_SYMBOL(blk_queue_dma_alignment);
771
772/**
773 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 * @q: The request queue for the device
775 * @tag: The tag of the request
776 *
777 * Notes:
778 * Should be used when a device returns a tag and you want to match
779 * it with a request.
780 *
781 * no locks need be held.
782 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200783struct request *blk_queue_find_tag(struct request_queue *q, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
David C Somayajuluf583f492006-10-04 08:27:25 +0200785 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786}
787
788EXPORT_SYMBOL(blk_queue_find_tag);
789
790/**
James Bottomley492dfb42006-08-30 15:48:45 -0400791 * __blk_free_tags - release a given set of tag maintenance info
792 * @bqt: the tag map to free
793 *
794 * Tries to free the specified @bqt@. Returns true if it was
795 * actually freed and false if there are still references using it
796 */
797static int __blk_free_tags(struct blk_queue_tag *bqt)
798{
799 int retval;
800
801 retval = atomic_dec_and_test(&bqt->refcnt);
802 if (retval) {
803 BUG_ON(bqt->busy);
804 BUG_ON(!list_empty(&bqt->busy_list));
805
806 kfree(bqt->tag_index);
807 bqt->tag_index = NULL;
808
809 kfree(bqt->tag_map);
810 bqt->tag_map = NULL;
811
812 kfree(bqt);
813
814 }
815
816 return retval;
817}
818
819/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 * __blk_queue_free_tags - release tag maintenance info
821 * @q: the request queue for the device
822 *
823 * Notes:
824 * blk_cleanup_queue() will take care of calling this function, if tagging
825 * has been used. So there's no need to call this directly.
826 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200827static void __blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
829 struct blk_queue_tag *bqt = q->queue_tags;
830
831 if (!bqt)
832 return;
833
James Bottomley492dfb42006-08-30 15:48:45 -0400834 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
836 q->queue_tags = NULL;
837 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
838}
839
James Bottomley492dfb42006-08-30 15:48:45 -0400840
841/**
842 * blk_free_tags - release a given set of tag maintenance info
843 * @bqt: the tag map to free
844 *
845 * For externally managed @bqt@ frees the map. Callers of this
846 * function must guarantee to have released all the queues that
847 * might have been using this tag map.
848 */
849void blk_free_tags(struct blk_queue_tag *bqt)
850{
851 if (unlikely(!__blk_free_tags(bqt)))
852 BUG();
853}
854EXPORT_SYMBOL(blk_free_tags);
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856/**
857 * blk_queue_free_tags - release tag maintenance info
858 * @q: the request queue for the device
859 *
860 * Notes:
861 * This is used to disabled tagged queuing to a device, yet leave
862 * queue in function.
863 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200864void blk_queue_free_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
866 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
867}
868
869EXPORT_SYMBOL(blk_queue_free_tags);
870
871static int
Jens Axboe165125e2007-07-24 09:28:11 +0200872init_tag_map(struct request_queue *q, struct blk_queue_tag *tags, int depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 struct request **tag_index;
875 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700876 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
James Bottomley492dfb42006-08-30 15:48:45 -0400878 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 depth = q->nr_requests * 2;
880 printk(KERN_ERR "%s: adjusted depth to %d\n",
881 __FUNCTION__, depth);
882 }
883
Jens Axboef68110f2006-03-08 13:31:44 +0100884 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 if (!tag_index)
886 goto fail;
887
Tejun Heof7d37d02005-06-23 00:08:50 -0700888 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100889 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (!tag_map)
891 goto fail;
892
Tejun Heoba025082005-08-05 13:28:11 -0700893 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 tags->tag_index = tag_index;
896 tags->tag_map = tag_map;
897
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 return 0;
899fail:
900 kfree(tag_index);
901 return -ENOMEM;
902}
903
James Bottomley492dfb42006-08-30 15:48:45 -0400904static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
905 int depth)
906{
907 struct blk_queue_tag *tags;
908
909 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
910 if (!tags)
911 goto fail;
912
913 if (init_tag_map(q, tags, depth))
914 goto fail;
915
916 INIT_LIST_HEAD(&tags->busy_list);
917 tags->busy = 0;
918 atomic_set(&tags->refcnt, 1);
919 return tags;
920fail:
921 kfree(tags);
922 return NULL;
923}
924
925/**
926 * blk_init_tags - initialize the tag info for an external tag map
927 * @depth: the maximum queue depth supported
928 * @tags: the tag to use
929 **/
930struct blk_queue_tag *blk_init_tags(int depth)
931{
932 return __blk_queue_init_tags(NULL, depth);
933}
934EXPORT_SYMBOL(blk_init_tags);
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936/**
937 * blk_queue_init_tags - initialize the queue tag info
938 * @q: the request queue for the device
939 * @depth: the maximum queue depth supported
940 * @tags: the tag to use
941 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200942int blk_queue_init_tags(struct request_queue *q, int depth,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 struct blk_queue_tag *tags)
944{
945 int rc;
946
947 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
948
949 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400950 tags = __blk_queue_init_tags(q, depth);
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (!tags)
953 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 } else if (q->queue_tags) {
955 if ((rc = blk_queue_resize_tags(q, depth)))
956 return rc;
957 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
958 return 0;
959 } else
960 atomic_inc(&tags->refcnt);
961
962 /*
963 * assign it, all done
964 */
965 q->queue_tags = tags;
966 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
967 return 0;
968fail:
969 kfree(tags);
970 return -ENOMEM;
971}
972
973EXPORT_SYMBOL(blk_queue_init_tags);
974
975/**
976 * blk_queue_resize_tags - change the queueing depth
977 * @q: the request queue for the device
978 * @new_depth: the new max command queueing depth
979 *
980 * Notes:
981 * Must be called with the queue lock held.
982 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200983int blk_queue_resize_tags(struct request_queue *q, int new_depth)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
985 struct blk_queue_tag *bqt = q->queue_tags;
986 struct request **tag_index;
987 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700988 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 if (!bqt)
991 return -ENXIO;
992
993 /*
Tejun Heoba025082005-08-05 13:28:11 -0700994 * if we already have large enough real_max_depth. just
995 * adjust max_depth. *NOTE* as requests with tag value
996 * between new_depth and real_max_depth can be in-flight, tag
997 * map can not be shrunk blindly here.
998 */
999 if (new_depth <= bqt->real_max_depth) {
1000 bqt->max_depth = new_depth;
1001 return 0;
1002 }
1003
1004 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001005 * Currently cannot replace a shared tag map with a new
1006 * one, so error out if this is the case
1007 */
1008 if (atomic_read(&bqt->refcnt) != 1)
1009 return -EBUSY;
1010
1011 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 * save the old state info, so we can copy it back
1013 */
1014 tag_index = bqt->tag_index;
1015 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001016 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 if (init_tag_map(q, bqt, new_depth))
1019 return -ENOMEM;
1020
1021 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001022 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001023 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025 kfree(tag_index);
1026 kfree(tag_map);
1027 return 0;
1028}
1029
1030EXPORT_SYMBOL(blk_queue_resize_tags);
1031
1032/**
1033 * blk_queue_end_tag - end tag operations for a request
1034 * @q: the request queue for the device
1035 * @rq: the request that has completed
1036 *
1037 * Description:
1038 * Typically called when end_that_request_first() returns 0, meaning
1039 * all transfers have been done for a request. It's important to call
1040 * this function before end_that_request_last(), as that will put the
1041 * request back on the free list thus corrupting the internal tag list.
1042 *
1043 * Notes:
1044 * queue lock must be held.
1045 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001046void blk_queue_end_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
1048 struct blk_queue_tag *bqt = q->queue_tags;
1049 int tag = rq->tag;
1050
1051 BUG_ON(tag == -1);
1052
Tejun Heoba025082005-08-05 13:28:11 -07001053 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001054 /*
1055 * This can happen after tag depth has been reduced.
1056 * FIXME: how about a warning or info message here?
1057 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 return;
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001061 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 rq->tag = -1;
1063
1064 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001065 printk(KERN_ERR "%s: tag %d is missing\n",
1066 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 bqt->tag_index[tag] = NULL;
Jens Axboef3da54b2007-09-13 14:26:53 +02001069
Nick Piggindd941252007-09-14 08:41:12 +02001070 /*
1071 * We use test_and_clear_bit's memory ordering properties here.
1072 * The tag_map bit acts as a lock for tag_index[bit], so we need
1073 * a barrer before clearing the bit (precisely: release semantics).
1074 * Could use clear_bit_unlock when it is merged.
1075 */
Jens Axboef3da54b2007-09-13 14:26:53 +02001076 if (unlikely(!test_and_clear_bit(tag, bqt->tag_map))) {
1077 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1078 __FUNCTION__, tag);
1079 return;
1080 }
1081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 bqt->busy--;
1083}
1084
1085EXPORT_SYMBOL(blk_queue_end_tag);
1086
1087/**
1088 * blk_queue_start_tag - find a free tag and assign it
1089 * @q: the request queue for the device
1090 * @rq: the block request that needs tagging
1091 *
1092 * Description:
1093 * This can either be used as a stand-alone helper, or possibly be
1094 * assigned as the queue &prep_rq_fn (in which case &struct request
1095 * automagically gets a tag assigned). Note that this function
1096 * assumes that any type of request can be queued! if this is not
1097 * true for your device, you must check the request type before
1098 * calling this function. The request will also be removed from
1099 * the request queue, so it's the drivers responsibility to readd
1100 * it if it should need to be restarted for some reason.
1101 *
1102 * Notes:
1103 * queue lock must be held.
1104 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001105int blk_queue_start_tag(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
1107 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001108 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Jens Axboe4aff5e22006-08-10 08:44:47 +02001110 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001112 "%s: request %p for device [%s] already tagged %d",
1113 __FUNCTION__, rq,
1114 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 BUG();
1116 }
1117
Jens Axboe059af492006-09-21 20:37:22 +02001118 /*
1119 * Protect against shared tag maps, as we may not have exclusive
1120 * access to the tag map.
1121 */
1122 do {
1123 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1124 if (tag >= bqt->max_depth)
1125 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Jens Axboe059af492006-09-21 20:37:22 +02001127 } while (test_and_set_bit(tag, bqt->tag_map));
Nick Piggindd941252007-09-14 08:41:12 +02001128 /*
1129 * We rely on test_and_set_bit providing lock memory ordering semantics
1130 * (could use test_and_set_bit_lock when it is merged).
1131 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Jens Axboe4aff5e22006-08-10 08:44:47 +02001133 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 rq->tag = tag;
1135 bqt->tag_index[tag] = rq;
1136 blkdev_dequeue_request(rq);
1137 list_add(&rq->queuelist, &bqt->busy_list);
1138 bqt->busy++;
1139 return 0;
1140}
1141
1142EXPORT_SYMBOL(blk_queue_start_tag);
1143
1144/**
1145 * blk_queue_invalidate_tags - invalidate all pending tags
1146 * @q: the request queue for the device
1147 *
1148 * Description:
1149 * Hardware conditions may dictate a need to stop all pending requests.
1150 * In this case, we will safely clear the block side of the tag queue and
1151 * readd all requests to the request queue in the right order.
1152 *
1153 * Notes:
1154 * queue lock must be held.
1155 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001156void blk_queue_invalidate_tags(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157{
1158 struct blk_queue_tag *bqt = q->queue_tags;
1159 struct list_head *tmp, *n;
1160 struct request *rq;
1161
1162 list_for_each_safe(tmp, n, &bqt->busy_list) {
1163 rq = list_entry_rq(tmp);
1164
1165 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001166 printk(KERN_ERR
1167 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001169 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 } else
1171 blk_queue_end_tag(q, rq);
1172
Jens Axboe4aff5e22006-08-10 08:44:47 +02001173 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1175 }
1176}
1177
1178EXPORT_SYMBOL(blk_queue_invalidate_tags);
1179
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180void blk_dump_rq_flags(struct request *rq, char *msg)
1181{
1182 int bit;
1183
Jens Axboe4aff5e22006-08-10 08:44:47 +02001184 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1185 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1186 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
1188 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1189 rq->nr_sectors,
1190 rq->current_nr_sectors);
1191 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1192
Jens Axboe4aff5e22006-08-10 08:44:47 +02001193 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 printk("cdb: ");
1195 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1196 printk("%02x ", rq->cmd[bit]);
1197 printk("\n");
1198 }
1199}
1200
1201EXPORT_SYMBOL(blk_dump_rq_flags);
1202
Jens Axboe165125e2007-07-24 09:28:11 +02001203void blk_recount_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
NeilBrown9dfa5282007-08-16 13:27:52 +02001205 struct request rq;
1206 struct bio *nxt = bio->bi_next;
1207 rq.q = q;
1208 rq.bio = rq.biotail = bio;
1209 bio->bi_next = NULL;
1210 blk_recalc_rq_segments(&rq);
1211 bio->bi_next = nxt;
1212 bio->bi_phys_segments = rq.nr_phys_segments;
1213 bio->bi_hw_segments = rq.nr_hw_segments;
1214 bio->bi_flags |= (1 << BIO_SEG_VALID);
1215}
1216EXPORT_SYMBOL(blk_recount_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
NeilBrown9dfa5282007-08-16 13:27:52 +02001218static void blk_recalc_rq_segments(struct request *rq)
1219{
1220 int nr_phys_segs;
1221 int nr_hw_segs;
1222 unsigned int phys_size;
1223 unsigned int hw_size;
1224 struct bio_vec *bv, *bvprv = NULL;
1225 int seg_size;
1226 int hw_seg_size;
1227 int cluster;
NeilBrown5705f702007-09-25 12:35:59 +02001228 struct req_iterator iter;
NeilBrown9dfa5282007-08-16 13:27:52 +02001229 int high, highprv = 1;
1230 struct request_queue *q = rq->q;
1231
1232 if (!rq->bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return;
1234
1235 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
NeilBrown9dfa5282007-08-16 13:27:52 +02001236 hw_seg_size = seg_size = 0;
1237 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
NeilBrown5705f702007-09-25 12:35:59 +02001238 rq_for_each_segment(bv, rq, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 /*
1240 * the trick here is making sure that a high page is never
1241 * considered part of another segment, since that might
1242 * change with the bounce page.
1243 */
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02001244 high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (high || highprv)
1246 goto new_hw_segment;
1247 if (cluster) {
1248 if (seg_size + bv->bv_len > q->max_segment_size)
1249 goto new_segment;
1250 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1251 goto new_segment;
1252 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1253 goto new_segment;
1254 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1255 goto new_hw_segment;
1256
1257 seg_size += bv->bv_len;
1258 hw_seg_size += bv->bv_len;
1259 bvprv = bv;
1260 continue;
1261 }
1262new_segment:
1263 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
NeilBrown9dfa5282007-08-16 13:27:52 +02001264 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 hw_seg_size += bv->bv_len;
NeilBrown9dfa5282007-08-16 13:27:52 +02001266 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267new_hw_segment:
NeilBrown9dfa5282007-08-16 13:27:52 +02001268 if (nr_hw_segs == 1 &&
1269 hw_seg_size > rq->bio->bi_hw_front_size)
1270 rq->bio->bi_hw_front_size = hw_seg_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1272 nr_hw_segs++;
1273 }
1274
1275 nr_phys_segs++;
1276 bvprv = bv;
1277 seg_size = bv->bv_len;
1278 highprv = high;
1279 }
NeilBrown9dfa5282007-08-16 13:27:52 +02001280
1281 if (nr_hw_segs == 1 &&
1282 hw_seg_size > rq->bio->bi_hw_front_size)
1283 rq->bio->bi_hw_front_size = hw_seg_size;
1284 if (hw_seg_size > rq->biotail->bi_hw_back_size)
1285 rq->biotail->bi_hw_back_size = hw_seg_size;
1286 rq->nr_phys_segments = nr_phys_segs;
1287 rq->nr_hw_segments = nr_hw_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Jens Axboe165125e2007-07-24 09:28:11 +02001290static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 struct bio *nxt)
1292{
1293 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1294 return 0;
1295
1296 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1297 return 0;
1298 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1299 return 0;
1300
1301 /*
1302 * bio and nxt are contigous in memory, check if the queue allows
1303 * these two to be merged into one
1304 */
1305 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1306 return 1;
1307
1308 return 0;
1309}
1310
Jens Axboe165125e2007-07-24 09:28:11 +02001311static int blk_hw_contig_segment(struct request_queue *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 struct bio *nxt)
1313{
1314 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1315 blk_recount_segments(q, bio);
1316 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1317 blk_recount_segments(q, nxt);
1318 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
Jens Axboe32eef962007-06-19 09:09:27 +02001319 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_back_size + nxt->bi_hw_front_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 return 0;
Jens Axboe32eef962007-06-19 09:09:27 +02001321 if (bio->bi_hw_back_size + nxt->bi_hw_front_size > q->max_segment_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 return 0;
1323
1324 return 1;
1325}
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327/*
1328 * map a request to scatterlist, return number of sg entries setup. Caller
1329 * must make sure sg can hold rq->nr_phys_segments entries
1330 */
Jens Axboe165125e2007-07-24 09:28:11 +02001331int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1332 struct scatterlist *sg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 struct bio_vec *bvec, *bvprv;
NeilBrown5705f702007-09-25 12:35:59 +02001335 struct req_iterator iter;
1336 int nsegs, cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 nsegs = 0;
1339 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1340
1341 /*
1342 * for each bio in rq
1343 */
1344 bvprv = NULL;
NeilBrown5705f702007-09-25 12:35:59 +02001345 rq_for_each_segment(bvec, rq, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +02001346 int nbytes = bvec->bv_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Jens Axboe6c92e692007-08-16 13:43:12 +02001348 if (bvprv && cluster) {
1349 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1350 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Jens Axboe6c92e692007-08-16 13:43:12 +02001352 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1353 goto new_segment;
1354 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1355 goto new_segment;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Jens Axboe6c92e692007-08-16 13:43:12 +02001357 sg[nsegs - 1].length += nbytes;
1358 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359new_segment:
Jens Axboe6c92e692007-08-16 13:43:12 +02001360 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1361 sg[nsegs].page = bvec->bv_page;
1362 sg[nsegs].length = nbytes;
1363 sg[nsegs].offset = bvec->bv_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Jens Axboe6c92e692007-08-16 13:43:12 +02001365 nsegs++;
1366 }
1367 bvprv = bvec;
NeilBrown5705f702007-09-25 12:35:59 +02001368 } /* segments in rq */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 return nsegs;
1371}
1372
1373EXPORT_SYMBOL(blk_rq_map_sg);
1374
1375/*
1376 * the standard queue merge functions, can be overridden with device
1377 * specific ones if so desired
1378 */
1379
Jens Axboe165125e2007-07-24 09:28:11 +02001380static inline int ll_new_mergeable(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 struct request *req,
1382 struct bio *bio)
1383{
1384 int nr_phys_segs = bio_phys_segments(q, bio);
1385
1386 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001387 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (req == q->last_merge)
1389 q->last_merge = NULL;
1390 return 0;
1391 }
1392
1393 /*
1394 * A hw segment is just getting larger, bump just the phys
1395 * counter.
1396 */
1397 req->nr_phys_segments += nr_phys_segs;
1398 return 1;
1399}
1400
Jens Axboe165125e2007-07-24 09:28:11 +02001401static inline int ll_new_hw_segment(struct request_queue *q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 struct request *req,
1403 struct bio *bio)
1404{
1405 int nr_hw_segs = bio_hw_segments(q, bio);
1406 int nr_phys_segs = bio_phys_segments(q, bio);
1407
1408 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1409 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001410 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (req == q->last_merge)
1412 q->last_merge = NULL;
1413 return 0;
1414 }
1415
1416 /*
1417 * This will form the start of a new hw segment. Bump both
1418 * counters.
1419 */
1420 req->nr_hw_segments += nr_hw_segs;
1421 req->nr_phys_segments += nr_phys_segs;
1422 return 1;
1423}
1424
NeilBrown3001ca72007-08-16 13:31:27 +02001425static int ll_back_merge_fn(struct request_queue *q, struct request *req,
1426 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Mike Christiedefd94b2005-12-05 02:37:06 -06001428 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 int len;
1430
Mike Christiedefd94b2005-12-05 02:37:06 -06001431 if (unlikely(blk_pc_request(req)))
1432 max_sectors = q->max_hw_sectors;
1433 else
1434 max_sectors = q->max_sectors;
1435
1436 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001437 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 if (req == q->last_merge)
1439 q->last_merge = NULL;
1440 return 0;
1441 }
1442 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1443 blk_recount_segments(q, req->biotail);
1444 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1445 blk_recount_segments(q, bio);
1446 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1447 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1448 !BIOVEC_VIRT_OVERSIZE(len)) {
1449 int mergeable = ll_new_mergeable(q, req, bio);
1450
1451 if (mergeable) {
1452 if (req->nr_hw_segments == 1)
1453 req->bio->bi_hw_front_size = len;
1454 if (bio->bi_hw_segments == 1)
1455 bio->bi_hw_back_size = len;
1456 }
1457 return mergeable;
1458 }
1459
1460 return ll_new_hw_segment(q, req, bio);
1461}
1462
Jens Axboe165125e2007-07-24 09:28:11 +02001463static int ll_front_merge_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 struct bio *bio)
1465{
Mike Christiedefd94b2005-12-05 02:37:06 -06001466 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 int len;
1468
Mike Christiedefd94b2005-12-05 02:37:06 -06001469 if (unlikely(blk_pc_request(req)))
1470 max_sectors = q->max_hw_sectors;
1471 else
1472 max_sectors = q->max_sectors;
1473
1474
1475 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001476 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (req == q->last_merge)
1478 q->last_merge = NULL;
1479 return 0;
1480 }
1481 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1482 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1483 blk_recount_segments(q, bio);
1484 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1485 blk_recount_segments(q, req->bio);
1486 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1487 !BIOVEC_VIRT_OVERSIZE(len)) {
1488 int mergeable = ll_new_mergeable(q, req, bio);
1489
1490 if (mergeable) {
1491 if (bio->bi_hw_segments == 1)
1492 bio->bi_hw_front_size = len;
1493 if (req->nr_hw_segments == 1)
1494 req->biotail->bi_hw_back_size = len;
1495 }
1496 return mergeable;
1497 }
1498
1499 return ll_new_hw_segment(q, req, bio);
1500}
1501
Jens Axboe165125e2007-07-24 09:28:11 +02001502static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 struct request *next)
1504{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001505 int total_phys_segments;
1506 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
1508 /*
1509 * First check if the either of the requests are re-queued
1510 * requests. Can't merge them if they are.
1511 */
1512 if (req->special || next->special)
1513 return 0;
1514
1515 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001516 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 */
1518 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1519 return 0;
1520
1521 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1522 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1523 total_phys_segments--;
1524
1525 if (total_phys_segments > q->max_phys_segments)
1526 return 0;
1527
1528 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1529 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1530 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1531 /*
1532 * propagate the combined length to the end of the requests
1533 */
1534 if (req->nr_hw_segments == 1)
1535 req->bio->bi_hw_front_size = len;
1536 if (next->nr_hw_segments == 1)
1537 next->biotail->bi_hw_back_size = len;
1538 total_hw_segments--;
1539 }
1540
1541 if (total_hw_segments > q->max_hw_segments)
1542 return 0;
1543
1544 /* Merge is OK... */
1545 req->nr_phys_segments = total_phys_segments;
1546 req->nr_hw_segments = total_hw_segments;
1547 return 1;
1548}
1549
1550/*
1551 * "plug" the device if there are no outstanding requests: this will
1552 * force the transfer to start only after we have put all the requests
1553 * on the list.
1554 *
1555 * This is called with interrupts off and no requests on the queue and
1556 * with the queue lock held.
1557 */
Jens Axboe165125e2007-07-24 09:28:11 +02001558void blk_plug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
1560 WARN_ON(!irqs_disabled());
1561
1562 /*
1563 * don't plug a stopped queue, it must be paired with blk_start_queue()
1564 * which will restart the queueing
1565 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001566 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 return;
1568
Jens Axboe2056a782006-03-23 20:00:26 +01001569 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001571 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573}
1574
1575EXPORT_SYMBOL(blk_plug_device);
1576
1577/*
1578 * remove the queue from the plugged list, if present. called with
1579 * queue lock held and interrupts disabled.
1580 */
Jens Axboe165125e2007-07-24 09:28:11 +02001581int blk_remove_plug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
1583 WARN_ON(!irqs_disabled());
1584
1585 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1586 return 0;
1587
1588 del_timer(&q->unplug_timer);
1589 return 1;
1590}
1591
1592EXPORT_SYMBOL(blk_remove_plug);
1593
1594/*
1595 * remove the plug and let it rip..
1596 */
Jens Axboe165125e2007-07-24 09:28:11 +02001597void __generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001599 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 return;
1601
1602 if (!blk_remove_plug(q))
1603 return;
1604
Jens Axboe22e2c502005-06-27 10:55:12 +02001605 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606}
1607EXPORT_SYMBOL(__generic_unplug_device);
1608
1609/**
1610 * generic_unplug_device - fire a request queue
Jens Axboe165125e2007-07-24 09:28:11 +02001611 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 *
1613 * Description:
1614 * Linux uses plugging to build bigger requests queues before letting
1615 * the device have at them. If a queue is plugged, the I/O scheduler
1616 * is still adding and merging requests on the queue. Once the queue
1617 * gets unplugged, the request_fn defined for the queue is invoked and
1618 * transfers started.
1619 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001620void generic_unplug_device(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621{
1622 spin_lock_irq(q->queue_lock);
1623 __generic_unplug_device(q);
1624 spin_unlock_irq(q->queue_lock);
1625}
1626EXPORT_SYMBOL(generic_unplug_device);
1627
1628static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1629 struct page *page)
1630{
Jens Axboe165125e2007-07-24 09:28:11 +02001631 struct request_queue *q = bdi->unplug_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633 /*
1634 * devices don't necessarily have an ->unplug_fn defined
1635 */
Jens Axboe2056a782006-03-23 20:00:26 +01001636 if (q->unplug_fn) {
1637 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1638 q->rq.count[READ] + q->rq.count[WRITE]);
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
David Howells65f27f32006-11-22 14:55:48 +00001644static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645{
Jens Axboe165125e2007-07-24 09:28:11 +02001646 struct request_queue *q =
1647 container_of(work, struct request_queue, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Jens Axboe2056a782006-03-23 20:00:26 +01001649 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1650 q->rq.count[READ] + q->rq.count[WRITE]);
1651
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 q->unplug_fn(q);
1653}
1654
1655static void blk_unplug_timeout(unsigned long data)
1656{
Jens Axboe165125e2007-07-24 09:28:11 +02001657 struct request_queue *q = (struct request_queue *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
Jens Axboe2056a782006-03-23 20:00:26 +01001659 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1660 q->rq.count[READ] + q->rq.count[WRITE]);
1661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 kblockd_schedule_work(&q->unplug_work);
1663}
1664
1665/**
1666 * blk_start_queue - restart a previously stopped queue
Jens Axboe165125e2007-07-24 09:28:11 +02001667 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 *
1669 * Description:
1670 * blk_start_queue() will clear the stop flag on the queue, and call
1671 * the request_fn for the queue if it was in a stopped state when
1672 * entered. Also see blk_stop_queue(). Queue lock must be held.
1673 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001674void blk_start_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001676 WARN_ON(!irqs_disabled());
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1679
1680 /*
1681 * one level of recursion is ok and is much faster than kicking
1682 * the unplug handling
1683 */
1684 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1685 q->request_fn(q);
1686 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1687 } else {
1688 blk_plug_device(q);
1689 kblockd_schedule_work(&q->unplug_work);
1690 }
1691}
1692
1693EXPORT_SYMBOL(blk_start_queue);
1694
1695/**
1696 * blk_stop_queue - stop a queue
Jens Axboe165125e2007-07-24 09:28:11 +02001697 * @q: The &struct request_queue in question
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 *
1699 * Description:
1700 * The Linux block layer assumes that a block driver will consume all
1701 * entries on the request queue when the request_fn strategy is called.
1702 * Often this will not happen, because of hardware limitations (queue
1703 * depth settings). If a device driver gets a 'queue full' response,
1704 * or if it simply chooses not to queue more I/O at one point, it can
1705 * call this function to prevent the request_fn from being called until
1706 * the driver has signalled it's ready to go again. This happens by calling
1707 * blk_start_queue() to restart queue operations. Queue lock must be held.
1708 **/
Jens Axboe165125e2007-07-24 09:28:11 +02001709void blk_stop_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710{
1711 blk_remove_plug(q);
1712 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1713}
1714EXPORT_SYMBOL(blk_stop_queue);
1715
1716/**
1717 * blk_sync_queue - cancel any pending callbacks on a queue
1718 * @q: the queue
1719 *
1720 * Description:
1721 * The block layer may perform asynchronous callback activity
1722 * on a queue, such as calling the unplug function after a timeout.
1723 * A block device may call blk_sync_queue to ensure that any
1724 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +02001725 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 * that its ->make_request_fn will not re-add plugging prior to calling
1727 * this function.
1728 *
1729 */
1730void blk_sync_queue(struct request_queue *q)
1731{
1732 del_timer_sync(&q->unplug_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733}
1734EXPORT_SYMBOL(blk_sync_queue);
1735
1736/**
1737 * blk_run_queue - run a single device queue
1738 * @q: The queue to run
1739 */
1740void blk_run_queue(struct request_queue *q)
1741{
1742 unsigned long flags;
1743
1744 spin_lock_irqsave(q->queue_lock, flags);
1745 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001746
1747 /*
1748 * Only recurse once to avoid overrunning the stack, let the unplug
1749 * handling reinvoke the handler shortly if we already got there.
1750 */
1751 if (!elv_queue_empty(q)) {
1752 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1753 q->request_fn(q);
1754 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1755 } else {
1756 blk_plug_device(q);
1757 kblockd_schedule_work(&q->unplug_work);
1758 }
1759 }
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 spin_unlock_irqrestore(q->queue_lock, flags);
1762}
1763EXPORT_SYMBOL(blk_run_queue);
1764
1765/**
Jens Axboe165125e2007-07-24 09:28:11 +02001766 * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001767 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 *
1769 * Description:
1770 * blk_cleanup_queue is the pair to blk_init_queue() or
1771 * blk_queue_make_request(). It should be called when a request queue is
1772 * being released; typically when a block device is being de-registered.
1773 * Currently, its primary task it to free all the &struct request
1774 * structures that were allocated to the queue and the queue itself.
1775 *
1776 * Caveat:
1777 * Hopefully the low level driver will have finished any
1778 * outstanding requests first...
1779 **/
Al Viro483f4af2006-03-18 18:34:37 -05001780static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
Jens Axboe165125e2007-07-24 09:28:11 +02001782 struct request_queue *q =
1783 container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 struct request_list *rl = &q->rq;
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 blk_sync_queue(q);
1787
1788 if (rl->rq_pool)
1789 mempool_destroy(rl->rq_pool);
1790
1791 if (q->queue_tags)
1792 __blk_queue_free_tags(q);
1793
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001794 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 kmem_cache_free(requestq_cachep, q);
1797}
1798
Jens Axboe165125e2007-07-24 09:28:11 +02001799void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -05001800{
1801 kobject_put(&q->kobj);
1802}
1803EXPORT_SYMBOL(blk_put_queue);
1804
Jens Axboe165125e2007-07-24 09:28:11 +02001805void blk_cleanup_queue(struct request_queue * q)
Al Viro483f4af2006-03-18 18:34:37 -05001806{
1807 mutex_lock(&q->sysfs_lock);
1808 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1809 mutex_unlock(&q->sysfs_lock);
1810
1811 if (q->elevator)
1812 elevator_exit(q->elevator);
1813
1814 blk_put_queue(q);
1815}
1816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817EXPORT_SYMBOL(blk_cleanup_queue);
1818
Jens Axboe165125e2007-07-24 09:28:11 +02001819static int blk_init_free_list(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820{
1821 struct request_list *rl = &q->rq;
1822
1823 rl->count[READ] = rl->count[WRITE] = 0;
1824 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001825 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 init_waitqueue_head(&rl->wait[READ]);
1827 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
Christoph Lameter19460892005-06-23 00:08:19 -07001829 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1830 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
1832 if (!rl->rq_pool)
1833 return -ENOMEM;
1834
1835 return 0;
1836}
1837
Jens Axboe165125e2007-07-24 09:28:11 +02001838struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
Christoph Lameter19460892005-06-23 00:08:19 -07001840 return blk_alloc_queue_node(gfp_mask, -1);
1841}
1842EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
Al Viro483f4af2006-03-18 18:34:37 -05001844static struct kobj_type queue_ktype;
1845
Jens Axboe165125e2007-07-24 09:28:11 +02001846struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001847{
Jens Axboe165125e2007-07-24 09:28:11 +02001848 struct request_queue *q;
Christoph Lameter19460892005-06-23 00:08:19 -07001849
Christoph Lameter94f60302007-07-17 04:03:29 -07001850 q = kmem_cache_alloc_node(requestq_cachep,
1851 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (!q)
1853 return NULL;
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001856
1857 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1858 q->kobj.ktype = &queue_ktype;
1859 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
1861 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1862 q->backing_dev_info.unplug_io_data = q;
1863
Al Viro483f4af2006-03-18 18:34:37 -05001864 mutex_init(&q->sysfs_lock);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 return q;
1867}
Christoph Lameter19460892005-06-23 00:08:19 -07001868EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
1870/**
1871 * blk_init_queue - prepare a request queue for use with a block device
1872 * @rfn: The function to be called to process requests that have been
1873 * placed on the queue.
1874 * @lock: Request queue spin lock
1875 *
1876 * Description:
1877 * If a block device wishes to use the standard request handling procedures,
1878 * which sorts requests and coalesces adjacent requests, then it must
1879 * call blk_init_queue(). The function @rfn will be called when there
1880 * are requests on the queue that need to be processed. If the device
1881 * supports plugging, then @rfn may not be called immediately when requests
1882 * are available on the queue, but may be called at some time later instead.
1883 * Plugged queues are generally unplugged when a buffer belonging to one
1884 * of the requests on the queue is needed, or due to memory pressure.
1885 *
1886 * @rfn is not required, or even expected, to remove all requests off the
1887 * queue, but only as many as it can handle at a time. If it does leave
1888 * requests on the queue, it is responsible for arranging that the requests
1889 * get dealt with eventually.
1890 *
1891 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001892 * request queue; this lock will be taken also from interrupt context, so irq
1893 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 *
1895 * Function returns a pointer to the initialized request queue, or NULL if
1896 * it didn't succeed.
1897 *
1898 * Note:
1899 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1900 * when the block device is deactivated (such as at module unload).
1901 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001902
Jens Axboe165125e2007-07-24 09:28:11 +02001903struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904{
Christoph Lameter19460892005-06-23 00:08:19 -07001905 return blk_init_queue_node(rfn, lock, -1);
1906}
1907EXPORT_SYMBOL(blk_init_queue);
1908
Jens Axboe165125e2007-07-24 09:28:11 +02001909struct request_queue *
Christoph Lameter19460892005-06-23 00:08:19 -07001910blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1911{
Jens Axboe165125e2007-07-24 09:28:11 +02001912 struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
1914 if (!q)
1915 return NULL;
1916
Christoph Lameter19460892005-06-23 00:08:19 -07001917 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001918 if (blk_init_free_list(q)) {
1919 kmem_cache_free(requestq_cachep, q);
1920 return NULL;
1921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
152587d2005-04-12 16:22:06 -05001923 /*
1924 * if caller didn't supply a lock, they get per-queue locking with
1925 * our embedded lock
1926 */
1927 if (!lock) {
1928 spin_lock_init(&q->__queue_lock);
1929 lock = &q->__queue_lock;
1930 }
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 q->request_fn = rfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 q->prep_rq_fn = NULL;
1934 q->unplug_fn = generic_unplug_device;
1935 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1936 q->queue_lock = lock;
1937
1938 blk_queue_segment_boundary(q, 0xffffffff);
1939
1940 blk_queue_make_request(q, __make_request);
1941 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1942
1943 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1944 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1945
Alan Stern44ec9542007-02-20 11:01:57 -05001946 q->sg_reserved_size = INT_MAX;
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 /*
1949 * all done
1950 */
1951 if (!elevator_init(q, NULL)) {
1952 blk_queue_congestion_threshold(q);
1953 return q;
1954 }
1955
Al Viro8669aaf2006-03-18 13:50:00 -05001956 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 return NULL;
1958}
Christoph Lameter19460892005-06-23 00:08:19 -07001959EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
Jens Axboe165125e2007-07-24 09:28:11 +02001961int blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001963 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001964 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 return 0;
1966 }
1967
1968 return 1;
1969}
1970
1971EXPORT_SYMBOL(blk_get_queue);
1972
Jens Axboe165125e2007-07-24 09:28:11 +02001973static inline void blk_free_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001975 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001976 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 mempool_free(rq, q->rq.rq_pool);
1978}
1979
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001980static struct request *
Jens Axboe165125e2007-07-24 09:28:11 +02001981blk_alloc_request(struct request_queue *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
1983 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1984
1985 if (!rq)
1986 return NULL;
1987
1988 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001989 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 * see bio.h and blkdev.h
1991 */
Jens Axboe49171e52006-08-10 08:59:11 +02001992 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Tejun Heocb98fc82005-10-28 08:29:39 +02001994 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001995 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001996 mempool_free(rq, q->rq.rq_pool);
1997 return NULL;
1998 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02001999 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
Tejun Heocb98fc82005-10-28 08:29:39 +02002002 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003}
2004
2005/*
2006 * ioc_batching returns true if the ioc is a valid batching request and
2007 * should be given priority access to a request.
2008 */
Jens Axboe165125e2007-07-24 09:28:11 +02002009static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010{
2011 if (!ioc)
2012 return 0;
2013
2014 /*
2015 * Make sure the process is able to allocate at least 1 request
2016 * even if the batch times out, otherwise we could theoretically
2017 * lose wakeups.
2018 */
2019 return ioc->nr_batch_requests == q->nr_batching ||
2020 (ioc->nr_batch_requests > 0
2021 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2022}
2023
2024/*
2025 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2026 * will cause the process to be a "batcher" on all queues in the system. This
2027 * is the behaviour we want though - once it gets a wakeup it should be given
2028 * a nice run.
2029 */
Jens Axboe165125e2007-07-24 09:28:11 +02002030static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031{
2032 if (!ioc || ioc_batching(q, ioc))
2033 return;
2034
2035 ioc->nr_batch_requests = q->nr_batching;
2036 ioc->last_waited = jiffies;
2037}
2038
Jens Axboe165125e2007-07-24 09:28:11 +02002039static void __freed_request(struct request_queue *q, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040{
2041 struct request_list *rl = &q->rq;
2042
2043 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002044 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 if (waitqueue_active(&rl->wait[rw]))
2048 wake_up(&rl->wait[rw]);
2049
2050 blk_clear_queue_full(q, rw);
2051 }
2052}
2053
2054/*
2055 * A request has just been released. Account for it, update the full and
2056 * congestion status, wake up any waiters. Called under q->queue_lock.
2057 */
Jens Axboe165125e2007-07-24 09:28:11 +02002058static void freed_request(struct request_queue *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059{
2060 struct request_list *rl = &q->rq;
2061
2062 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002063 if (priv)
2064 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066 __freed_request(q, rw);
2067
2068 if (unlikely(rl->starved[rw ^ 1]))
2069 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
2072#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2073/*
Nick Piggind6344532005-06-28 20:45:14 -07002074 * Get a free request, queue_lock must be held.
2075 * Returns NULL on failure, with queue_lock held.
2076 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 */
Jens Axboe165125e2007-07-24 09:28:11 +02002078static struct request *get_request(struct request_queue *q, int rw_flags,
Jens Axboe7749a8d2006-12-13 13:02:26 +01002079 struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080{
2081 struct request *rq = NULL;
2082 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002083 struct io_context *ioc = NULL;
Jens Axboe7749a8d2006-12-13 13:02:26 +01002084 const int rw = rw_flags & 0x01;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002085 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Jens Axboe7749a8d2006-12-13 13:02:26 +01002087 may_queue = elv_may_queue(q, rw_flags);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002088 if (may_queue == ELV_MQUEUE_NO)
2089 goto rq_starved;
2090
2091 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2092 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002093 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002094 /*
2095 * The queue will fill after this allocation, so set
2096 * it as full, and mark this process as "batching".
2097 * This process will be allowed to complete a batch of
2098 * requests, others will be blocked.
2099 */
2100 if (!blk_queue_full(q, rw)) {
2101 ioc_set_batching(q, ioc);
2102 blk_set_queue_full(q, rw);
2103 } else {
2104 if (may_queue != ELV_MQUEUE_MUST
2105 && !ioc_batching(q, ioc)) {
2106 /*
2107 * The queue is full and the allocating
2108 * process is not a "batcher", and not
2109 * exempted by the IO scheduler
2110 */
2111 goto out;
2112 }
2113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002115 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
2117
Jens Axboe082cf692005-06-28 16:35:11 +02002118 /*
2119 * Only allow batching queuers to allocate up to 50% over the defined
2120 * limit of requests, otherwise we could have thousands of requests
2121 * allocated with any setting of ->nr_requests
2122 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002123 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002124 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 rl->count[rw]++;
2127 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002128
Jens Axboe64521d12005-10-28 08:30:39 +02002129 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002130 if (priv)
2131 rl->elvpriv++;
2132
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 spin_unlock_irq(q->queue_lock);
2134
Jens Axboe7749a8d2006-12-13 13:02:26 +01002135 rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002136 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 /*
2138 * Allocation failed presumably due to memory. Undo anything
2139 * we might have messed up.
2140 *
2141 * Allocating task should really be put onto the front of the
2142 * wait queue, but this is pretty rare.
2143 */
2144 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002145 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
2147 /*
2148 * in the very unlikely event that allocation failed and no
2149 * requests for this direction was pending, mark us starved
2150 * so that freeing of a request in the other direction will
2151 * notice us. another possible fix would be to split the
2152 * rq mempool into READ and WRITE
2153 */
2154rq_starved:
2155 if (unlikely(rl->count[rw] == 0))
2156 rl->starved[rw] = 1;
2157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 goto out;
2159 }
2160
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002161 /*
2162 * ioc may be NULL here, and ioc_batching will be false. That's
2163 * OK, if the queue is under the request limit then requests need
2164 * not count toward the nr_batch_requests limit. There will always
2165 * be some limit enforced by BLK_BATCH_TIME.
2166 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 if (ioc_batching(q, ioc))
2168 ioc->nr_batch_requests--;
2169
2170 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002171
2172 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 return rq;
2175}
2176
2177/*
2178 * No available requests for this queue, unplug the device and wait for some
2179 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002180 *
2181 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 */
Jens Axboe165125e2007-07-24 09:28:11 +02002183static struct request *get_request_wait(struct request_queue *q, int rw_flags,
Jens Axboe22e2c502005-06-27 10:55:12 +02002184 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185{
Jens Axboe7749a8d2006-12-13 13:02:26 +01002186 const int rw = rw_flags & 0x01;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 struct request *rq;
2188
Jens Axboe7749a8d2006-12-13 13:02:26 +01002189 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Nick Piggin450991b2005-06-28 20:45:13 -07002190 while (!rq) {
2191 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 struct request_list *rl = &q->rq;
2193
2194 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2195 TASK_UNINTERRUPTIBLE);
2196
Jens Axboe7749a8d2006-12-13 13:02:26 +01002197 rq = get_request(q, rw_flags, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199 if (!rq) {
2200 struct io_context *ioc;
2201
Jens Axboe2056a782006-03-23 20:00:26 +01002202 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2203
Nick Piggind6344532005-06-28 20:45:14 -07002204 __generic_unplug_device(q);
2205 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 io_schedule();
2207
2208 /*
2209 * After sleeping, we become a "batching" process and
2210 * will be able to allocate at least one request, and
2211 * up to a big batch of them for a small period time.
2212 * See ioc_batching, ioc_set_batching
2213 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002214 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002216
2217 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 }
2219 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222 return rq;
2223}
2224
Jens Axboe165125e2007-07-24 09:28:11 +02002225struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
2227 struct request *rq;
2228
2229 BUG_ON(rw != READ && rw != WRITE);
2230
Nick Piggind6344532005-06-28 20:45:14 -07002231 spin_lock_irq(q->queue_lock);
2232 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002233 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002234 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002235 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002236 if (!rq)
2237 spin_unlock_irq(q->queue_lock);
2238 }
2239 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 return rq;
2242}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243EXPORT_SYMBOL(blk_get_request);
2244
2245/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002246 * blk_start_queueing - initiate dispatch of requests to device
2247 * @q: request queue to kick into gear
2248 *
2249 * This is basically a helper to remove the need to know whether a queue
2250 * is plugged or not if someone just wants to initiate dispatch of requests
2251 * for this queue.
2252 *
2253 * The queue lock must be held with interrupts disabled.
2254 */
Jens Axboe165125e2007-07-24 09:28:11 +02002255void blk_start_queueing(struct request_queue *q)
Jens Axboedc72ef42006-07-20 14:54:05 +02002256{
2257 if (!blk_queue_plugged(q))
2258 q->request_fn(q);
2259 else
2260 __generic_unplug_device(q);
2261}
2262EXPORT_SYMBOL(blk_start_queueing);
2263
2264/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 * blk_requeue_request - put a request back on queue
2266 * @q: request queue where request should be inserted
2267 * @rq: request to be inserted
2268 *
2269 * Description:
2270 * Drivers often keep queueing requests until the hardware cannot accept
2271 * more, when that condition happens we need to put the request back
2272 * on the queue. Must be called with queue lock held.
2273 */
Jens Axboe165125e2007-07-24 09:28:11 +02002274void blk_requeue_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
Jens Axboe2056a782006-03-23 20:00:26 +01002276 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 if (blk_rq_tagged(rq))
2279 blk_queue_end_tag(q, rq);
2280
2281 elv_requeue_request(q, rq);
2282}
2283
2284EXPORT_SYMBOL(blk_requeue_request);
2285
2286/**
2287 * blk_insert_request - insert a special request in to a request queue
2288 * @q: request queue where request should be inserted
2289 * @rq: request to be inserted
2290 * @at_head: insert request at head or tail of queue
2291 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 *
2293 * Description:
2294 * Many block devices need to execute commands asynchronously, so they don't
2295 * block the whole kernel from preemption during request execution. This is
2296 * accomplished normally by inserting aritficial requests tagged as
2297 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2298 * scheduled for actual execution by the request queue.
2299 *
2300 * We have the option of inserting the head or the tail of the queue.
2301 * Typically we use the tail for new ioctls and so forth. We use the head
2302 * of the queue for things like a QUEUE_FULL message from a device, or a
2303 * host that is unable to accept a particular command.
2304 */
Jens Axboe165125e2007-07-24 09:28:11 +02002305void blk_insert_request(struct request_queue *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002306 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307{
Tejun Heo 867d1192005-04-24 02:06:05 -05002308 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 unsigned long flags;
2310
2311 /*
2312 * tell I/O scheduler that this isn't a regular read/write (ie it
2313 * must not attempt merges on this) and that it acts as a soft
2314 * barrier
2315 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002316 rq->cmd_type = REQ_TYPE_SPECIAL;
2317 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319 rq->special = data;
2320
2321 spin_lock_irqsave(q->queue_lock, flags);
2322
2323 /*
2324 * If command is tagged, release the tag
2325 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002326 if (blk_rq_tagged(rq))
2327 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Tejun Heo 867d1192005-04-24 02:06:05 -05002329 drive_stat_acct(rq, rq->nr_sectors, 1);
2330 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002331 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 spin_unlock_irqrestore(q->queue_lock, flags);
2333}
2334
2335EXPORT_SYMBOL(blk_insert_request);
2336
Mike Christie0e75f902006-12-01 10:40:55 +01002337static int __blk_rq_unmap_user(struct bio *bio)
2338{
2339 int ret = 0;
2340
2341 if (bio) {
2342 if (bio_flagged(bio, BIO_USER_MAPPED))
2343 bio_unmap_user(bio);
2344 else
2345 ret = bio_uncopy_user(bio);
2346 }
2347
2348 return ret;
2349}
2350
NeilBrown3001ca72007-08-16 13:31:27 +02002351int blk_rq_append_bio(struct request_queue *q, struct request *rq,
2352 struct bio *bio)
2353{
2354 if (!rq->bio)
2355 blk_rq_bio_prep(q, rq, bio);
2356 else if (!ll_back_merge_fn(q, rq, bio))
2357 return -EINVAL;
2358 else {
2359 rq->biotail->bi_next = bio;
2360 rq->biotail = bio;
2361
2362 rq->data_len += bio->bi_size;
2363 }
2364 return 0;
2365}
2366EXPORT_SYMBOL(blk_rq_append_bio);
2367
Jens Axboe165125e2007-07-24 09:28:11 +02002368static int __blk_rq_map_user(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002369 void __user *ubuf, unsigned int len)
2370{
2371 unsigned long uaddr;
2372 struct bio *bio, *orig_bio;
2373 int reading, ret;
2374
2375 reading = rq_data_dir(rq) == READ;
2376
2377 /*
2378 * if alignment requirement is satisfied, map in user pages for
2379 * direct dma. else, set up kernel bounce buffers
2380 */
2381 uaddr = (unsigned long) ubuf;
2382 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2383 bio = bio_map_user(q, NULL, uaddr, len, reading);
2384 else
2385 bio = bio_copy_user(q, uaddr, len, reading);
2386
Jens Axboe29852592006-12-19 08:27:31 +01002387 if (IS_ERR(bio))
Mike Christie0e75f902006-12-01 10:40:55 +01002388 return PTR_ERR(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002389
2390 orig_bio = bio;
2391 blk_queue_bounce(q, &bio);
Jens Axboe29852592006-12-19 08:27:31 +01002392
Mike Christie0e75f902006-12-01 10:40:55 +01002393 /*
2394 * We link the bounce buffer in and could have to traverse it
2395 * later so we have to get a ref to prevent it from being freed
2396 */
2397 bio_get(bio);
2398
NeilBrown3001ca72007-08-16 13:31:27 +02002399 ret = blk_rq_append_bio(q, rq, bio);
2400 if (!ret)
2401 return bio->bi_size;
Mike Christie0e75f902006-12-01 10:40:55 +01002402
Mike Christie0e75f902006-12-01 10:40:55 +01002403 /* if it was boucned we must call the end io function */
NeilBrown6712ecf2007-09-27 12:47:43 +02002404 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002405 __blk_rq_unmap_user(orig_bio);
2406 bio_put(bio);
2407 return ret;
2408}
2409
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410/**
2411 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2412 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002413 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 * @ubuf: the user buffer
2415 * @len: length of user data
2416 *
2417 * Description:
2418 * Data will be mapped directly for zero copy io, if possible. Otherwise
2419 * a kernel bounce buffer is used.
2420 *
2421 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2422 * still in process context.
2423 *
2424 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2425 * before being submitted to the device, as pages mapped may be out of
2426 * reach. It's the callers responsibility to make sure this happens. The
2427 * original bio must be passed back in to blk_rq_unmap_user() for proper
2428 * unmapping.
2429 */
Jens Axboe165125e2007-07-24 09:28:11 +02002430int blk_rq_map_user(struct request_queue *q, struct request *rq,
2431 void __user *ubuf, unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432{
Mike Christie0e75f902006-12-01 10:40:55 +01002433 unsigned long bytes_read = 0;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002434 struct bio *bio = NULL;
Mike Christie0e75f902006-12-01 10:40:55 +01002435 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Mike Christiedefd94b2005-12-05 02:37:06 -06002437 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002438 return -EINVAL;
2439 if (!len || !ubuf)
2440 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Mike Christie0e75f902006-12-01 10:40:55 +01002442 while (bytes_read != len) {
2443 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Mike Christie0e75f902006-12-01 10:40:55 +01002445 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2446 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2447 >> PAGE_SHIFT;
2448 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Mike Christie0e75f902006-12-01 10:40:55 +01002450 /*
2451 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2452 * pages. If this happens we just lower the requested
2453 * mapping len by a page so that we can fit
2454 */
2455 if (end - start > BIO_MAX_PAGES)
2456 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
Mike Christie0e75f902006-12-01 10:40:55 +01002458 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2459 if (ret < 0)
2460 goto unmap_rq;
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002461 if (!bio)
2462 bio = rq->bio;
Mike Christie0e75f902006-12-01 10:40:55 +01002463 bytes_read += ret;
2464 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 }
2466
Mike Christie0e75f902006-12-01 10:40:55 +01002467 rq->buffer = rq->data = NULL;
2468 return 0;
2469unmap_rq:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002470 blk_rq_unmap_user(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002471 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472}
2473
2474EXPORT_SYMBOL(blk_rq_map_user);
2475
2476/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002477 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2478 * @q: request queue where request should be inserted
2479 * @rq: request to map data to
2480 * @iov: pointer to the iovec
2481 * @iov_count: number of elements in the iovec
Randy Dunlapaf9997e2006-12-22 01:06:52 -08002482 * @len: I/O byte count
James Bottomley f1970ba2005-06-20 14:06:52 +02002483 *
2484 * Description:
2485 * Data will be mapped directly for zero copy io, if possible. Otherwise
2486 * a kernel bounce buffer is used.
2487 *
2488 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2489 * still in process context.
2490 *
2491 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2492 * before being submitted to the device, as pages mapped may be out of
2493 * reach. It's the callers responsibility to make sure this happens. The
2494 * original bio must be passed back in to blk_rq_unmap_user() for proper
2495 * unmapping.
2496 */
Jens Axboe165125e2007-07-24 09:28:11 +02002497int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002498 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002499{
2500 struct bio *bio;
2501
2502 if (!iov || iov_count <= 0)
2503 return -EINVAL;
2504
2505 /* we don't allow misaligned data like bio_map_user() does. If the
2506 * user is using sg, they're expected to know the alignment constraints
2507 * and respect them accordingly */
2508 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2509 if (IS_ERR(bio))
2510 return PTR_ERR(bio);
2511
Mike Christie0e75f902006-12-01 10:40:55 +01002512 if (bio->bi_size != len) {
NeilBrown6712ecf2007-09-27 12:47:43 +02002513 bio_endio(bio, 0);
Mike Christie0e75f902006-12-01 10:40:55 +01002514 bio_unmap_user(bio);
2515 return -EINVAL;
2516 }
2517
2518 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002519 blk_rq_bio_prep(q, rq, bio);
2520 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002521 return 0;
2522}
2523
2524EXPORT_SYMBOL(blk_rq_map_user_iov);
2525
2526/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 * blk_rq_unmap_user - unmap a request with user data
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002528 * @bio: start of bio list
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 *
2530 * Description:
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002531 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
2532 * supply the original rq->bio from the blk_rq_map_user() return, since
2533 * the io completion may have changed rq->bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 */
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002535int blk_rq_unmap_user(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536{
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002537 struct bio *mapped_bio;
Jens Axboe48785bb2006-12-19 11:07:59 +01002538 int ret = 0, ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002540 while (bio) {
2541 mapped_bio = bio;
2542 if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
Mike Christie0e75f902006-12-01 10:40:55 +01002543 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Jens Axboe48785bb2006-12-19 11:07:59 +01002545 ret2 = __blk_rq_unmap_user(mapped_bio);
2546 if (ret2 && !ret)
2547 ret = ret2;
2548
Jens Axboe8e5cfc42006-12-19 11:12:46 +01002549 mapped_bio = bio;
2550 bio = bio->bi_next;
2551 bio_put(mapped_bio);
Mike Christie0e75f902006-12-01 10:40:55 +01002552 }
Jens Axboe48785bb2006-12-19 11:07:59 +01002553
2554 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555}
2556
2557EXPORT_SYMBOL(blk_rq_unmap_user);
2558
2559/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002560 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2561 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002562 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002563 * @kbuf: the kernel buffer
2564 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002565 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002566 */
Jens Axboe165125e2007-07-24 09:28:11 +02002567int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002568 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002569{
Mike Christie df46b9a2005-06-20 14:04:44 +02002570 struct bio *bio;
2571
Mike Christiedefd94b2005-12-05 02:37:06 -06002572 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002573 return -EINVAL;
2574 if (!len || !kbuf)
2575 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002576
2577 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002578 if (IS_ERR(bio))
2579 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002580
Jens Axboedd1cab92005-06-20 14:06:01 +02002581 if (rq_data_dir(rq) == WRITE)
2582 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002583
Jens Axboedd1cab92005-06-20 14:06:01 +02002584 blk_rq_bio_prep(q, rq, bio);
Mike Christie821de3a2007-05-08 19:12:23 +02002585 blk_queue_bounce(q, &rq->bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002586 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002587 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002588}
2589
2590EXPORT_SYMBOL(blk_rq_map_kern);
2591
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002592/**
2593 * blk_execute_rq_nowait - insert a request into queue for execution
2594 * @q: queue to insert the request in
2595 * @bd_disk: matching gendisk
2596 * @rq: request to insert
2597 * @at_head: insert request at head or tail of queue
2598 * @done: I/O completion handler
2599 *
2600 * Description:
2601 * Insert a fully prepared request at the back of the io scheduler queue
2602 * for execution. Don't wait for completion.
2603 */
Jens Axboe165125e2007-07-24 09:28:11 +02002604void blk_execute_rq_nowait(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley f1970ba2005-06-20 14:06:52 +02002605 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002606 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002607{
2608 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2609
2610 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002611 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002612 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002613 WARN_ON(irqs_disabled());
2614 spin_lock_irq(q->queue_lock);
2615 __elv_add_request(q, rq, where, 1);
2616 __generic_unplug_device(q);
2617 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002618}
Mike Christie6e39b69e2005-11-11 05:30:24 -06002619EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2620
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621/**
2622 * blk_execute_rq - insert a request into queue for execution
2623 * @q: queue to insert the request in
2624 * @bd_disk: matching gendisk
2625 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002626 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 *
2628 * Description:
2629 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002630 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 */
Jens Axboe165125e2007-07-24 09:28:11 +02002632int blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002633 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002635 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 char sense[SCSI_SENSE_BUFFERSIZE];
2637 int err = 0;
2638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 /*
2640 * we need an extra reference to the request, so we can look at
2641 * it after io completion
2642 */
2643 rq->ref_count++;
2644
2645 if (!rq->sense) {
2646 memset(sense, 0, sizeof(sense));
2647 rq->sense = sense;
2648 rq->sense_len = 0;
2649 }
2650
Jens Axboec00895a2006-09-30 20:29:12 +02002651 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002652 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654
2655 if (rq->errors)
2656 err = -EIO;
2657
2658 return err;
2659}
2660
2661EXPORT_SYMBOL(blk_execute_rq);
2662
2663/**
2664 * blkdev_issue_flush - queue a flush
2665 * @bdev: blockdev to issue flush for
2666 * @error_sector: error sector
2667 *
2668 * Description:
2669 * Issue a flush for the block device in question. Caller can supply
2670 * room for storing the error offset in case of a flush error, if they
2671 * wish to. Caller must run wait_for_completion() on its own.
2672 */
2673int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2674{
Jens Axboe165125e2007-07-24 09:28:11 +02002675 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
2677 if (bdev->bd_disk == NULL)
2678 return -ENXIO;
2679
2680 q = bdev_get_queue(bdev);
2681 if (!q)
2682 return -ENXIO;
2683 if (!q->issue_flush_fn)
2684 return -EOPNOTSUPP;
2685
2686 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2687}
2688
2689EXPORT_SYMBOL(blkdev_issue_flush);
2690
Adrian Bunk93d17d32005-06-25 14:59:10 -07002691static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692{
2693 int rw = rq_data_dir(rq);
2694
2695 if (!blk_fs_request(rq) || !rq->rq_disk)
2696 return;
2697
Jens Axboed72d9042005-11-01 08:35:42 +01002698 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002699 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002700 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 disk_round_stats(rq->rq_disk);
2702 rq->rq_disk->in_flight++;
2703 }
2704}
2705
2706/*
2707 * add-request adds a request to the linked list.
2708 * queue lock is held and interrupts disabled, as we muck with the
2709 * request queue list.
2710 */
Jens Axboe165125e2007-07-24 09:28:11 +02002711static inline void add_request(struct request_queue * q, struct request * req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712{
2713 drive_stat_acct(req, req->nr_sectors, 1);
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 /*
2716 * elevator indicated where it wants this request to be
2717 * inserted at elevator_merge time
2718 */
2719 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2720}
2721
2722/*
2723 * disk_round_stats() - Round off the performance stats on a struct
2724 * disk_stats.
2725 *
2726 * The average IO queue length and utilisation statistics are maintained
2727 * by observing the current state of the queue length and the amount of
2728 * time it has been in this state for.
2729 *
2730 * Normally, that accounting is done on IO completion, but that can result
2731 * in more than a second's worth of IO being accounted for within any one
2732 * second, leading to >100% utilisation. To deal with that, we call this
2733 * function to do a round-off before returning the results when reading
2734 * /proc/diskstats. This accounts immediately for all queue usage up to
2735 * the current jiffies and restarts the counters again.
2736 */
2737void disk_round_stats(struct gendisk *disk)
2738{
2739 unsigned long now = jiffies;
2740
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002741 if (now == disk->stamp)
2742 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002744 if (disk->in_flight) {
2745 __disk_stat_add(disk, time_in_queue,
2746 disk->in_flight * (now - disk->stamp));
2747 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750}
2751
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002752EXPORT_SYMBOL_GPL(disk_round_stats);
2753
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754/*
2755 * queue lock must be held
2756 */
Jens Axboe165125e2007-07-24 09:28:11 +02002757void __blk_put_request(struct request_queue *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 if (unlikely(!q))
2760 return;
2761 if (unlikely(--req->ref_count))
2762 return;
2763
Tejun Heo8922e162005-10-20 16:23:44 +02002764 elv_completed_request(q, req);
2765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766 /*
2767 * Request may not have originated from ll_rw_blk. if not,
2768 * it didn't come out of our reserved rq pools
2769 */
Jens Axboe49171e52006-08-10 08:59:11 +02002770 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002772 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002775 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
2777 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002778 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 }
2780}
2781
Mike Christie6e39b69e2005-11-11 05:30:24 -06002782EXPORT_SYMBOL_GPL(__blk_put_request);
2783
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784void blk_put_request(struct request *req)
2785{
Tejun Heo8922e162005-10-20 16:23:44 +02002786 unsigned long flags;
Jens Axboe165125e2007-07-24 09:28:11 +02002787 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Tejun Heo8922e162005-10-20 16:23:44 +02002789 /*
2790 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2791 * following if (q) test.
2792 */
2793 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 spin_lock_irqsave(q->queue_lock, flags);
2795 __blk_put_request(q, req);
2796 spin_unlock_irqrestore(q->queue_lock, flags);
2797 }
2798}
2799
2800EXPORT_SYMBOL(blk_put_request);
2801
2802/**
2803 * blk_end_sync_rq - executes a completion event on a request
2804 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002805 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002807void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808{
Jens Axboec00895a2006-09-30 20:29:12 +02002809 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
Jens Axboec00895a2006-09-30 20:29:12 +02002811 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 __blk_put_request(rq->q, rq);
2813
2814 /*
2815 * complete last, if this is a stack request the process (and thus
2816 * the rq pointer) could be invalid right after this complete()
2817 */
2818 complete(waiting);
2819}
2820EXPORT_SYMBOL(blk_end_sync_rq);
2821
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822/*
2823 * Has to be called with the request spinlock acquired
2824 */
Jens Axboe165125e2007-07-24 09:28:11 +02002825static int attempt_merge(struct request_queue *q, struct request *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 struct request *next)
2827{
2828 if (!rq_mergeable(req) || !rq_mergeable(next))
2829 return 0;
2830
2831 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002832 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 */
2834 if (req->sector + req->nr_sectors != next->sector)
2835 return 0;
2836
2837 if (rq_data_dir(req) != rq_data_dir(next)
2838 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002839 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 return 0;
2841
2842 /*
2843 * If we are allowed to merge, then append bio list
2844 * from next to rq and release next. merge_requests_fn
2845 * will have updated segment counts, update sector
2846 * counts here.
2847 */
Jens Axboe1aa4f242006-12-19 08:33:11 +01002848 if (!ll_merge_requests_fn(q, req, next))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 return 0;
2850
2851 /*
2852 * At this point we have either done a back merge
2853 * or front merge. We need the smaller start_time of
2854 * the merged requests to be the current request
2855 * for accounting purposes.
2856 */
2857 if (time_after(req->start_time, next->start_time))
2858 req->start_time = next->start_time;
2859
2860 req->biotail->bi_next = next->bio;
2861 req->biotail = next->biotail;
2862
2863 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2864
2865 elv_merge_requests(q, req, next);
2866
2867 if (req->rq_disk) {
2868 disk_round_stats(req->rq_disk);
2869 req->rq_disk->in_flight--;
2870 }
2871
Jens Axboe22e2c502005-06-27 10:55:12 +02002872 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2873
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 __blk_put_request(q, next);
2875 return 1;
2876}
2877
Jens Axboe165125e2007-07-24 09:28:11 +02002878static inline int attempt_back_merge(struct request_queue *q,
2879 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880{
2881 struct request *next = elv_latter_request(q, rq);
2882
2883 if (next)
2884 return attempt_merge(q, rq, next);
2885
2886 return 0;
2887}
2888
Jens Axboe165125e2007-07-24 09:28:11 +02002889static inline int attempt_front_merge(struct request_queue *q,
2890 struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891{
2892 struct request *prev = elv_former_request(q, rq);
2893
2894 if (prev)
2895 return attempt_merge(q, prev, rq);
2896
2897 return 0;
2898}
2899
Tejun Heo52d9e672006-01-06 09:49:58 +01002900static void init_request_from_bio(struct request *req, struct bio *bio)
2901{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002902 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002903
2904 /*
2905 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2906 */
2907 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002908 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002909
2910 /*
2911 * REQ_BARRIER implies no merging, but lets make it explicit
2912 */
2913 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002914 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002915
Jens Axboeb31dc662006-06-13 08:26:10 +02002916 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002917 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002918 if (bio_rw_meta(bio))
2919 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002920
Tejun Heo52d9e672006-01-06 09:49:58 +01002921 req->errors = 0;
2922 req->hard_sector = req->sector = bio->bi_sector;
Tejun Heo52d9e672006-01-06 09:49:58 +01002923 req->ioprio = bio_prio(bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002924 req->start_time = jiffies;
NeilBrownbc1c56f2007-08-16 13:31:30 +02002925 blk_rq_bio_prep(req->q, req, bio);
Tejun Heo52d9e672006-01-06 09:49:58 +01002926}
2927
Jens Axboe165125e2007-07-24 09:28:11 +02002928static int __make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929{
Nick Piggin450991b2005-06-28 20:45:13 -07002930 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002931 int el_ret, nr_sectors, barrier, err;
2932 const unsigned short prio = bio_prio(bio);
2933 const int sync = bio_sync(bio);
Jens Axboe7749a8d2006-12-13 13:02:26 +01002934 int rw_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937
2938 /*
2939 * low level driver can indicate that it wants pages above a
2940 * certain limit bounced to low memory (ie for highmem, or even
2941 * ISA dma in theory)
2942 */
2943 blk_queue_bounce(q, &bio);
2944
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002946 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 err = -EOPNOTSUPP;
2948 goto end_io;
2949 }
2950
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 spin_lock_irq(q->queue_lock);
2952
Nick Piggin450991b2005-06-28 20:45:13 -07002953 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 goto get_rq;
2955
2956 el_ret = elv_merge(q, &req, bio);
2957 switch (el_ret) {
2958 case ELEVATOR_BACK_MERGE:
2959 BUG_ON(!rq_mergeable(req));
2960
Jens Axboe1aa4f242006-12-19 08:33:11 +01002961 if (!ll_back_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 break;
2963
Jens Axboe2056a782006-03-23 20:00:26 +01002964 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2965
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 req->biotail->bi_next = bio;
2967 req->biotail = bio;
2968 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002969 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 drive_stat_acct(req, nr_sectors, 0);
2971 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002972 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 goto out;
2974
2975 case ELEVATOR_FRONT_MERGE:
2976 BUG_ON(!rq_mergeable(req));
2977
Jens Axboe1aa4f242006-12-19 08:33:11 +01002978 if (!ll_front_merge_fn(q, req, bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 break;
2980
Jens Axboe2056a782006-03-23 20:00:26 +01002981 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2982
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 bio->bi_next = req->bio;
2984 req->bio = bio;
2985
2986 /*
2987 * may not be valid. if the low level driver said
2988 * it didn't need a bounce buffer then it better
2989 * not touch req->buffer either...
2990 */
2991 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02002992 req->current_nr_sectors = bio_cur_sectors(bio);
2993 req->hard_cur_sectors = req->current_nr_sectors;
2994 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002996 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 drive_stat_acct(req, nr_sectors, 0);
2998 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002999 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 goto out;
3001
Nick Piggin450991b2005-06-28 20:45:13 -07003002 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 default:
Nick Piggin450991b2005-06-28 20:45:13 -07003004 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 }
3006
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07003008 /*
Jens Axboe7749a8d2006-12-13 13:02:26 +01003009 * This sync check and mask will be re-done in init_request_from_bio(),
3010 * but we need to set it earlier to expose the sync flag to the
3011 * rq allocator and io schedulers.
3012 */
3013 rw_flags = bio_data_dir(bio);
3014 if (sync)
3015 rw_flags |= REQ_RW_SYNC;
3016
3017 /*
Nick Piggin450991b2005-06-28 20:45:13 -07003018 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07003019 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07003020 */
Jens Axboe7749a8d2006-12-13 13:02:26 +01003021 req = get_request_wait(q, rw_flags, bio);
Nick Piggind6344532005-06-28 20:45:14 -07003022
Nick Piggin450991b2005-06-28 20:45:13 -07003023 /*
3024 * After dropping the lock and possibly sleeping here, our request
3025 * may now be mergeable after it had proven unmergeable (above).
3026 * We don't worry about that case for efficiency. It won't happen
3027 * often, and the elevators are able to handle it.
3028 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003029 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
Nick Piggin450991b2005-06-28 20:45:13 -07003031 spin_lock_irq(q->queue_lock);
3032 if (elv_queue_empty(q))
3033 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 add_request(q, req);
3035out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003036 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037 __generic_unplug_device(q);
3038
3039 spin_unlock_irq(q->queue_lock);
3040 return 0;
3041
3042end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003043 bio_endio(bio, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 return 0;
3045}
3046
3047/*
3048 * If bio->bi_dev is a partition, remap the location
3049 */
3050static inline void blk_partition_remap(struct bio *bio)
3051{
3052 struct block_device *bdev = bio->bi_bdev;
3053
3054 if (bdev != bdev->bd_contains) {
3055 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003056 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Jens Axboea3623572005-11-01 09:26:16 +01003058 p->sectors[rw] += bio_sectors(bio);
3059 p->ios[rw]++;
3060
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 bio->bi_sector += p->start_sect;
3062 bio->bi_bdev = bdev->bd_contains;
Alan D. Brunellec7149d62007-08-07 15:30:23 +02003063
3064 blk_add_trace_remap(bdev_get_queue(bio->bi_bdev), bio,
3065 bdev->bd_dev, bio->bi_sector,
3066 bio->bi_sector - p->start_sect);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 }
3068}
3069
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070static void handle_bad_sector(struct bio *bio)
3071{
3072 char b[BDEVNAME_SIZE];
3073
3074 printk(KERN_INFO "attempt to access beyond end of device\n");
3075 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3076 bdevname(bio->bi_bdev, b),
3077 bio->bi_rw,
3078 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3079 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3080
3081 set_bit(BIO_EOF, &bio->bi_flags);
3082}
3083
Akinobu Mitac17bb492006-12-08 02:39:46 -08003084#ifdef CONFIG_FAIL_MAKE_REQUEST
3085
3086static DECLARE_FAULT_ATTR(fail_make_request);
3087
3088static int __init setup_fail_make_request(char *str)
3089{
3090 return setup_fault_attr(&fail_make_request, str);
3091}
3092__setup("fail_make_request=", setup_fail_make_request);
3093
3094static int should_fail_request(struct bio *bio)
3095{
3096 if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
3097 (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
3098 return should_fail(&fail_make_request, bio->bi_size);
3099
3100 return 0;
3101}
3102
3103static int __init fail_make_request_debugfs(void)
3104{
3105 return init_fault_attr_dentries(&fail_make_request,
3106 "fail_make_request");
3107}
3108
3109late_initcall(fail_make_request_debugfs);
3110
3111#else /* CONFIG_FAIL_MAKE_REQUEST */
3112
3113static inline int should_fail_request(struct bio *bio)
3114{
3115 return 0;
3116}
3117
3118#endif /* CONFIG_FAIL_MAKE_REQUEST */
3119
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120/**
3121 * generic_make_request: hand a buffer to its device driver for I/O
3122 * @bio: The bio describing the location in memory and on the device.
3123 *
3124 * generic_make_request() is used to make I/O requests of block
3125 * devices. It is passed a &struct bio, which describes the I/O that needs
3126 * to be done.
3127 *
3128 * generic_make_request() does not return any status. The
3129 * success/failure status of the request, along with notification of
3130 * completion, is delivered asynchronously through the bio->bi_end_io
3131 * function described (one day) else where.
3132 *
3133 * The caller of generic_make_request must make sure that bi_io_vec
3134 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3135 * set to describe the device address, and the
3136 * bi_end_io and optionally bi_private are set to describe how
3137 * completion notification should be signaled.
3138 *
3139 * generic_make_request and the drivers it calls may use bi_next if this
3140 * bio happens to be merged with someone else, and may change bi_dev and
3141 * bi_sector for remaps as it sees fit. So the values of these fields
3142 * should NOT be depended on after the call to generic_make_request.
3143 */
Neil Brownd89d8792007-05-01 09:53:42 +02003144static inline void __generic_make_request(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145{
Jens Axboe165125e2007-07-24 09:28:11 +02003146 struct request_queue *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 sector_t maxsector;
NeilBrown5ddfe962006-10-30 22:07:21 -08003148 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003150 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151
3152 might_sleep();
3153 /* Test device or partition size, when known. */
3154 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3155 if (maxsector) {
3156 sector_t sector = bio->bi_sector;
3157
3158 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3159 /*
3160 * This may well happen - the kernel calls bread()
3161 * without checking the size of the device, e.g., when
3162 * mounting a device.
3163 */
3164 handle_bad_sector(bio);
3165 goto end_io;
3166 }
3167 }
3168
3169 /*
3170 * Resolve the mapping until finished. (drivers are
3171 * still free to implement/resolve their own stacking
3172 * by explicitly returning 0)
3173 *
3174 * NOTE: we don't repeat the blk_size check for each new device.
3175 * Stacking drivers are expected to know what they are doing.
3176 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003177 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003178 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 do {
3180 char b[BDEVNAME_SIZE];
3181
3182 q = bdev_get_queue(bio->bi_bdev);
3183 if (!q) {
3184 printk(KERN_ERR
3185 "generic_make_request: Trying to access "
3186 "nonexistent block-device %s (%Lu)\n",
3187 bdevname(bio->bi_bdev, b),
3188 (long long) bio->bi_sector);
3189end_io:
NeilBrown6712ecf2007-09-27 12:47:43 +02003190 bio_endio(bio, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 break;
3192 }
3193
3194 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3195 printk("bio too big device %s (%u > %u)\n",
3196 bdevname(bio->bi_bdev, b),
3197 bio_sectors(bio),
3198 q->max_hw_sectors);
3199 goto end_io;
3200 }
3201
Nick Pigginfde6ad22005-06-23 00:08:53 -07003202 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 goto end_io;
3204
Akinobu Mitac17bb492006-12-08 02:39:46 -08003205 if (should_fail_request(bio))
3206 goto end_io;
3207
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 /*
3209 * If this device has partitions, remap block n
3210 * of partition p to block n+start(p) of the disk.
3211 */
3212 blk_partition_remap(bio);
3213
NeilBrown5ddfe962006-10-30 22:07:21 -08003214 if (old_sector != -1)
Jens Axboe2056a782006-03-23 20:00:26 +01003215 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003216 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003217
3218 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3219
NeilBrown5ddfe962006-10-30 22:07:21 -08003220 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003221 old_dev = bio->bi_bdev->bd_dev;
3222
NeilBrown5ddfe962006-10-30 22:07:21 -08003223 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3224 if (maxsector) {
3225 sector_t sector = bio->bi_sector;
3226
Andrew Mortondf66b852006-11-02 22:06:56 -08003227 if (maxsector < nr_sectors ||
3228 maxsector - nr_sectors < sector) {
NeilBrown5ddfe962006-10-30 22:07:21 -08003229 /*
Andrew Mortondf66b852006-11-02 22:06:56 -08003230 * This may well happen - partitions are not
3231 * checked to make sure they are within the size
3232 * of the whole device.
NeilBrown5ddfe962006-10-30 22:07:21 -08003233 */
3234 handle_bad_sector(bio);
3235 goto end_io;
3236 }
3237 }
3238
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 ret = q->make_request_fn(q, bio);
3240 } while (ret);
3241}
3242
Neil Brownd89d8792007-05-01 09:53:42 +02003243/*
3244 * We only want one ->make_request_fn to be active at a time,
3245 * else stack usage with stacked devices could be a problem.
3246 * So use current->bio_{list,tail} to keep a list of requests
3247 * submited by a make_request_fn function.
3248 * current->bio_tail is also used as a flag to say if
3249 * generic_make_request is currently active in this task or not.
3250 * If it is NULL, then no make_request is active. If it is non-NULL,
3251 * then a make_request is active, and new requests should be added
3252 * at the tail
3253 */
3254void generic_make_request(struct bio *bio)
3255{
3256 if (current->bio_tail) {
3257 /* make_request is active */
3258 *(current->bio_tail) = bio;
3259 bio->bi_next = NULL;
3260 current->bio_tail = &bio->bi_next;
3261 return;
3262 }
3263 /* following loop may be a bit non-obvious, and so deserves some
3264 * explanation.
3265 * Before entering the loop, bio->bi_next is NULL (as all callers
3266 * ensure that) so we have a list with a single bio.
3267 * We pretend that we have just taken it off a longer list, so
3268 * we assign bio_list to the next (which is NULL) and bio_tail
3269 * to &bio_list, thus initialising the bio_list of new bios to be
3270 * added. __generic_make_request may indeed add some more bios
3271 * through a recursive call to generic_make_request. If it
3272 * did, we find a non-NULL value in bio_list and re-enter the loop
3273 * from the top. In this case we really did just take the bio
3274 * of the top of the list (no pretending) and so fixup bio_list and
3275 * bio_tail or bi_next, and call into __generic_make_request again.
3276 *
3277 * The loop was structured like this to make only one call to
3278 * __generic_make_request (which is important as it is large and
3279 * inlined) and to keep the structure simple.
3280 */
3281 BUG_ON(bio->bi_next);
3282 do {
3283 current->bio_list = bio->bi_next;
3284 if (bio->bi_next == NULL)
3285 current->bio_tail = &current->bio_list;
3286 else
3287 bio->bi_next = NULL;
3288 __generic_make_request(bio);
3289 bio = current->bio_list;
3290 } while (bio);
3291 current->bio_tail = NULL; /* deactivate */
3292}
3293
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294EXPORT_SYMBOL(generic_make_request);
3295
3296/**
3297 * submit_bio: submit a bio to the block device layer for I/O
3298 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3299 * @bio: The &struct bio which describes the I/O
3300 *
3301 * submit_bio() is very similar in purpose to generic_make_request(), and
3302 * uses that function to do most of the work. Both are fairly rough
3303 * interfaces, @bio must be presetup and ready for I/O.
3304 *
3305 */
3306void submit_bio(int rw, struct bio *bio)
3307{
3308 int count = bio_sectors(bio);
3309
3310 BIO_BUG_ON(!bio->bi_size);
3311 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003312 bio->bi_rw |= rw;
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003313 if (rw & WRITE) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07003314 count_vm_events(PGPGOUT, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003315 } else {
3316 task_io_account_read(bio->bi_size);
Christoph Lameterf8891e52006-06-30 01:55:45 -07003317 count_vm_events(PGPGIN, count);
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -08003318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319
3320 if (unlikely(block_dump)) {
3321 char b[BDEVNAME_SIZE];
3322 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3323 current->comm, current->pid,
3324 (rw & WRITE) ? "WRITE" : "READ",
3325 (unsigned long long)bio->bi_sector,
3326 bdevname(bio->bi_bdev,b));
3327 }
3328
3329 generic_make_request(bio);
3330}
3331
3332EXPORT_SYMBOL(submit_bio);
3333
Adrian Bunk93d17d32005-06-25 14:59:10 -07003334static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335{
3336 if (blk_fs_request(rq)) {
3337 rq->hard_sector += nsect;
3338 rq->hard_nr_sectors -= nsect;
3339
3340 /*
3341 * Move the I/O submission pointers ahead if required.
3342 */
3343 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3344 (rq->sector <= rq->hard_sector)) {
3345 rq->sector = rq->hard_sector;
3346 rq->nr_sectors = rq->hard_nr_sectors;
3347 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3348 rq->current_nr_sectors = rq->hard_cur_sectors;
3349 rq->buffer = bio_data(rq->bio);
3350 }
3351
3352 /*
3353 * if total number of sectors is less than the first segment
3354 * size, something has gone terribly wrong
3355 */
3356 if (rq->nr_sectors < rq->current_nr_sectors) {
3357 printk("blk: request botched\n");
3358 rq->nr_sectors = rq->current_nr_sectors;
3359 }
3360 }
3361}
3362
3363static int __end_that_request_first(struct request *req, int uptodate,
3364 int nr_bytes)
3365{
3366 int total_bytes, bio_nbytes, error, next_idx = 0;
3367 struct bio *bio;
3368
Jens Axboe2056a782006-03-23 20:00:26 +01003369 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3370
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 /*
3372 * extend uptodate bool to allow < 0 value to be direct io error
3373 */
3374 error = 0;
3375 if (end_io_error(uptodate))
3376 error = !uptodate ? -EIO : uptodate;
3377
3378 /*
3379 * for a REQ_BLOCK_PC request, we want to carry any eventual
3380 * sense key with us all the way through
3381 */
3382 if (!blk_pc_request(req))
3383 req->errors = 0;
3384
3385 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003386 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 printk("end_request: I/O error, dev %s, sector %llu\n",
3388 req->rq_disk ? req->rq_disk->disk_name : "?",
3389 (unsigned long long)req->sector);
3390 }
3391
Jens Axboed72d9042005-11-01 08:35:42 +01003392 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003393 const int rw = rq_data_dir(req);
3394
Jens Axboe53e86062006-01-17 11:09:27 +01003395 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003396 }
3397
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 total_bytes = bio_nbytes = 0;
3399 while ((bio = req->bio) != NULL) {
3400 int nbytes;
3401
3402 if (nr_bytes >= bio->bi_size) {
3403 req->bio = bio->bi_next;
3404 nbytes = bio->bi_size;
NeilBrown5bb23a62007-09-27 12:46:13 +02003405 req_bio_endio(req, bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 next_idx = 0;
3407 bio_nbytes = 0;
3408 } else {
3409 int idx = bio->bi_idx + next_idx;
3410
3411 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3412 blk_dump_rq_flags(req, "__end_that");
3413 printk("%s: bio idx %d >= vcnt %d\n",
3414 __FUNCTION__,
3415 bio->bi_idx, bio->bi_vcnt);
3416 break;
3417 }
3418
3419 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3420 BIO_BUG_ON(nbytes > bio->bi_size);
3421
3422 /*
3423 * not a complete bvec done
3424 */
3425 if (unlikely(nbytes > nr_bytes)) {
3426 bio_nbytes += nr_bytes;
3427 total_bytes += nr_bytes;
3428 break;
3429 }
3430
3431 /*
3432 * advance to the next vector
3433 */
3434 next_idx++;
3435 bio_nbytes += nbytes;
3436 }
3437
3438 total_bytes += nbytes;
3439 nr_bytes -= nbytes;
3440
3441 if ((bio = req->bio)) {
3442 /*
3443 * end more in this run, or just return 'not-done'
3444 */
3445 if (unlikely(nr_bytes <= 0))
3446 break;
3447 }
3448 }
3449
3450 /*
3451 * completely done
3452 */
3453 if (!req->bio)
3454 return 0;
3455
3456 /*
3457 * if the request wasn't completed, update state
3458 */
3459 if (bio_nbytes) {
NeilBrown5bb23a62007-09-27 12:46:13 +02003460 req_bio_endio(req, bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 bio->bi_idx += next_idx;
3462 bio_iovec(bio)->bv_offset += nr_bytes;
3463 bio_iovec(bio)->bv_len -= nr_bytes;
3464 }
3465
3466 blk_recalc_rq_sectors(req, total_bytes >> 9);
3467 blk_recalc_rq_segments(req);
3468 return 1;
3469}
3470
3471/**
3472 * end_that_request_first - end I/O on a request
3473 * @req: the request being processed
3474 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3475 * @nr_sectors: number of sectors to end I/O on
3476 *
3477 * Description:
3478 * Ends I/O on a number of sectors attached to @req, and sets it up
3479 * for the next range of segments (if any) in the cluster.
3480 *
3481 * Return:
3482 * 0 - we are done with this request, call end_that_request_last()
3483 * 1 - still buffers pending for this request
3484 **/
3485int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3486{
3487 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3488}
3489
3490EXPORT_SYMBOL(end_that_request_first);
3491
3492/**
3493 * end_that_request_chunk - end I/O on a request
3494 * @req: the request being processed
3495 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3496 * @nr_bytes: number of bytes to complete
3497 *
3498 * Description:
3499 * Ends I/O on a number of bytes attached to @req, and sets it up
3500 * for the next range of segments (if any). Like end_that_request_first(),
3501 * but deals with bytes instead of sectors.
3502 *
3503 * Return:
3504 * 0 - we are done with this request, call end_that_request_last()
3505 * 1 - still buffers pending for this request
3506 **/
3507int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3508{
3509 return __end_that_request_first(req, uptodate, nr_bytes);
3510}
3511
3512EXPORT_SYMBOL(end_that_request_chunk);
3513
3514/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003515 * splice the completion data to a local structure and hand off to
3516 * process_completion_queue() to complete the requests
3517 */
3518static void blk_done_softirq(struct softirq_action *h)
3519{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003520 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003521
3522 local_irq_disable();
3523 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003524 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003525 local_irq_enable();
3526
3527 while (!list_empty(&local_list)) {
3528 struct request *rq = list_entry(local_list.next, struct request, donelist);
3529
3530 list_del_init(&rq->donelist);
3531 rq->q->softirq_done_fn(rq);
3532 }
3533}
3534
Satyam Sharmadb47d472007-08-23 09:29:40 +02003535static int __cpuinit blk_cpu_notify(struct notifier_block *self, unsigned long action,
Jens Axboeff856ba2006-01-09 16:02:34 +01003536 void *hcpu)
3537{
3538 /*
3539 * If a CPU goes away, splice its entries to the current CPU
3540 * and trigger a run of the softirq
3541 */
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003542 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
Jens Axboeff856ba2006-01-09 16:02:34 +01003543 int cpu = (unsigned long) hcpu;
3544
3545 local_irq_disable();
3546 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3547 &__get_cpu_var(blk_cpu_done));
3548 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3549 local_irq_enable();
3550 }
3551
3552 return NOTIFY_OK;
3553}
3554
3555
Satyam Sharmadb47d472007-08-23 09:29:40 +02003556static struct notifier_block blk_cpu_notifier __cpuinitdata = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003557 .notifier_call = blk_cpu_notify,
3558};
3559
Jens Axboeff856ba2006-01-09 16:02:34 +01003560/**
3561 * blk_complete_request - end I/O on a request
3562 * @req: the request being processed
3563 *
3564 * Description:
3565 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003566 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003567 * through requeueing. Theh actual completion happens out-of-order,
3568 * through a softirq handler. The user must have registered a completion
3569 * callback through blk_queue_softirq_done().
3570 **/
3571
3572void blk_complete_request(struct request *req)
3573{
3574 struct list_head *cpu_list;
3575 unsigned long flags;
3576
3577 BUG_ON(!req->q->softirq_done_fn);
3578
3579 local_irq_save(flags);
3580
3581 cpu_list = &__get_cpu_var(blk_cpu_done);
3582 list_add_tail(&req->donelist, cpu_list);
3583 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3584
3585 local_irq_restore(flags);
3586}
3587
3588EXPORT_SYMBOL(blk_complete_request);
3589
3590/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 * queue lock must be held
3592 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003593void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594{
3595 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003596 int error;
3597
3598 /*
3599 * extend uptodate bool to allow < 0 value to be direct io error
3600 */
3601 error = 0;
3602 if (end_io_error(uptodate))
3603 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604
3605 if (unlikely(laptop_mode) && blk_fs_request(req))
3606 laptop_io_completion();
3607
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003608 /*
3609 * Account IO completion. bar_rq isn't accounted as a normal
3610 * IO on queueing nor completion. Accounting the containing
3611 * request is enough.
3612 */
3613 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003615 const int rw = rq_data_dir(req);
3616
3617 __disk_stat_inc(disk, ios[rw]);
3618 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 disk_round_stats(disk);
3620 disk->in_flight--;
3621 }
3622 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003623 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 else
3625 __blk_put_request(req->q, req);
3626}
3627
3628EXPORT_SYMBOL(end_that_request_last);
3629
3630void end_request(struct request *req, int uptodate)
3631{
3632 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3633 add_disk_randomness(req->rq_disk);
3634 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003635 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 }
3637}
3638
3639EXPORT_SYMBOL(end_request);
3640
NeilBrown66846572007-08-16 13:31:28 +02003641static void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3642 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003644 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3645 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646
3647 rq->nr_phys_segments = bio_phys_segments(q, bio);
3648 rq->nr_hw_segments = bio_hw_segments(q, bio);
3649 rq->current_nr_sectors = bio_cur_sectors(bio);
3650 rq->hard_cur_sectors = rq->current_nr_sectors;
3651 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3652 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003653 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654
3655 rq->bio = rq->biotail = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656
NeilBrown66846572007-08-16 13:31:28 +02003657 if (bio->bi_bdev)
3658 rq->rq_disk = bio->bi_bdev->bd_disk;
3659}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660
3661int kblockd_schedule_work(struct work_struct *work)
3662{
3663 return queue_work(kblockd_workqueue, work);
3664}
3665
3666EXPORT_SYMBOL(kblockd_schedule_work);
3667
Andrew Morton19a75d82007-05-09 02:33:56 -07003668void kblockd_flush_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669{
Oleg Nesterov28e53bd2007-05-09 02:34:22 -07003670 cancel_work_sync(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671}
Andrew Morton19a75d82007-05-09 02:33:56 -07003672EXPORT_SYMBOL(kblockd_flush_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673
3674int __init blk_dev_init(void)
3675{
Jens Axboeff856ba2006-01-09 16:02:34 +01003676 int i;
3677
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 kblockd_workqueue = create_workqueue("kblockd");
3679 if (!kblockd_workqueue)
3680 panic("Failed to create kblockd\n");
3681
3682 request_cachep = kmem_cache_create("blkdev_requests",
Paul Mundt20c2df82007-07-20 10:11:58 +09003683 sizeof(struct request), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003684
3685 requestq_cachep = kmem_cache_create("blkdev_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02003686 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687
3688 iocontext_cachep = kmem_cache_create("blkdev_ioc",
Paul Mundt20c2df82007-07-20 10:11:58 +09003689 sizeof(struct io_context), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003691 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003692 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3693
3694 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003695 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003696
Vasily Tarasovf772b3d2007-03-27 08:52:47 +02003697 blk_max_low_pfn = max_low_pfn - 1;
3698 blk_max_pfn = max_pfn - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699
3700 return 0;
3701}
3702
3703/*
3704 * IO Context helper functions
3705 */
3706void put_io_context(struct io_context *ioc)
3707{
3708 if (ioc == NULL)
3709 return;
3710
3711 BUG_ON(atomic_read(&ioc->refcount) == 0);
3712
3713 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003714 struct cfq_io_context *cic;
3715
Al Viro334e94d2006-03-18 15:05:53 -05003716 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003717 if (ioc->aic && ioc->aic->dtor)
3718 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003719 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003720 struct rb_node *n = rb_first(&ioc->cic_root);
3721
3722 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003723 cic->dtor(ioc);
3724 }
Al Viro334e94d2006-03-18 15:05:53 -05003725 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726
3727 kmem_cache_free(iocontext_cachep, ioc);
3728 }
3729}
3730EXPORT_SYMBOL(put_io_context);
3731
3732/* Called by the exitting task */
3733void exit_io_context(void)
3734{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003736 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737
Jens Axboe22e2c502005-06-27 10:55:12 +02003738 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 ioc = current->io_context;
3740 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003741 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742
Oleg Nesterov25034d72006-08-29 09:15:14 +02003743 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003744 if (ioc->aic && ioc->aic->exit)
3745 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003746 if (ioc->cic_root.rb_node != NULL) {
3747 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3748 cic->exit(ioc);
3749 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003750
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751 put_io_context(ioc);
3752}
3753
3754/*
3755 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003756 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003758 * This returned IO context doesn't have a specifically elevated refcount,
3759 * but since the current task itself holds a reference, the context can be
3760 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003762static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763{
3764 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 struct io_context *ret;
3766
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003768 if (likely(ret))
3769 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770
Jens Axboeb5deef92006-07-19 23:39:40 +02003771 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772 if (ret) {
3773 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003774 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003775 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776 ret->last_waited = jiffies; /* doesn't matter... */
3777 ret->nr_batch_requests = 0; /* because this is 0 */
3778 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003779 ret->cic_root.rb_node = NULL;
Jens Axboe4e521c22007-04-24 21:17:33 +02003780 ret->ioc_data = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003781 /* make sure set_task_ioprio() sees the settings above */
3782 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003783 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 }
3785
3786 return ret;
3787}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003788
3789/*
3790 * If the current task has no IO context then create one and initialise it.
3791 * If it does have a context, take a ref on it.
3792 *
3793 * This is always called in the context of the task which submitted the I/O.
3794 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003795struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003796{
3797 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003798 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003799 if (likely(ret))
3800 atomic_inc(&ret->refcount);
3801 return ret;
3802}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803EXPORT_SYMBOL(get_io_context);
3804
3805void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3806{
3807 struct io_context *src = *psrc;
3808 struct io_context *dst = *pdst;
3809
3810 if (src) {
3811 BUG_ON(atomic_read(&src->refcount) == 0);
3812 atomic_inc(&src->refcount);
3813 put_io_context(dst);
3814 *pdst = src;
3815 }
3816}
3817EXPORT_SYMBOL(copy_io_context);
3818
3819void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3820{
3821 struct io_context *temp;
3822 temp = *ioc1;
3823 *ioc1 = *ioc2;
3824 *ioc2 = temp;
3825}
3826EXPORT_SYMBOL(swap_io_context);
3827
3828/*
3829 * sysfs parts below
3830 */
3831struct queue_sysfs_entry {
3832 struct attribute attr;
3833 ssize_t (*show)(struct request_queue *, char *);
3834 ssize_t (*store)(struct request_queue *, const char *, size_t);
3835};
3836
3837static ssize_t
3838queue_var_show(unsigned int var, char *page)
3839{
3840 return sprintf(page, "%d\n", var);
3841}
3842
3843static ssize_t
3844queue_var_store(unsigned long *var, const char *page, size_t count)
3845{
3846 char *p = (char *) page;
3847
3848 *var = simple_strtoul(p, &p, 10);
3849 return count;
3850}
3851
3852static ssize_t queue_requests_show(struct request_queue *q, char *page)
3853{
3854 return queue_var_show(q->nr_requests, (page));
3855}
3856
3857static ssize_t
3858queue_requests_store(struct request_queue *q, const char *page, size_t count)
3859{
3860 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003861 unsigned long nr;
3862 int ret = queue_var_store(&nr, page, count);
3863 if (nr < BLKDEV_MIN_RQ)
3864 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865
Al Viroc981ff92006-03-18 13:51:29 -05003866 spin_lock_irq(q->queue_lock);
3867 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868 blk_queue_congestion_threshold(q);
3869
3870 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003871 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003873 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874
3875 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003876 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003878 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879
3880 if (rl->count[READ] >= q->nr_requests) {
3881 blk_set_queue_full(q, READ);
3882 } else if (rl->count[READ]+1 <= q->nr_requests) {
3883 blk_clear_queue_full(q, READ);
3884 wake_up(&rl->wait[READ]);
3885 }
3886
3887 if (rl->count[WRITE] >= q->nr_requests) {
3888 blk_set_queue_full(q, WRITE);
3889 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3890 blk_clear_queue_full(q, WRITE);
3891 wake_up(&rl->wait[WRITE]);
3892 }
Al Viroc981ff92006-03-18 13:51:29 -05003893 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894 return ret;
3895}
3896
3897static ssize_t queue_ra_show(struct request_queue *q, char *page)
3898{
3899 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3900
3901 return queue_var_show(ra_kb, (page));
3902}
3903
3904static ssize_t
3905queue_ra_store(struct request_queue *q, const char *page, size_t count)
3906{
3907 unsigned long ra_kb;
3908 ssize_t ret = queue_var_store(&ra_kb, page, count);
3909
3910 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3912 spin_unlock_irq(q->queue_lock);
3913
3914 return ret;
3915}
3916
3917static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3918{
3919 int max_sectors_kb = q->max_sectors >> 1;
3920
3921 return queue_var_show(max_sectors_kb, (page));
3922}
3923
3924static ssize_t
3925queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3926{
3927 unsigned long max_sectors_kb,
3928 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3929 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3930 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3931 int ra_kb;
3932
3933 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3934 return -EINVAL;
3935 /*
3936 * Take the queue lock to update the readahead and max_sectors
3937 * values synchronously:
3938 */
3939 spin_lock_irq(q->queue_lock);
3940 /*
3941 * Trim readahead window as well, if necessary:
3942 */
3943 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3944 if (ra_kb > max_sectors_kb)
3945 q->backing_dev_info.ra_pages =
3946 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3947
3948 q->max_sectors = max_sectors_kb << 1;
3949 spin_unlock_irq(q->queue_lock);
3950
3951 return ret;
3952}
3953
3954static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3955{
3956 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3957
3958 return queue_var_show(max_hw_sectors_kb, (page));
3959}
3960
3961
3962static struct queue_sysfs_entry queue_requests_entry = {
3963 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3964 .show = queue_requests_show,
3965 .store = queue_requests_store,
3966};
3967
3968static struct queue_sysfs_entry queue_ra_entry = {
3969 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3970 .show = queue_ra_show,
3971 .store = queue_ra_store,
3972};
3973
3974static struct queue_sysfs_entry queue_max_sectors_entry = {
3975 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3976 .show = queue_max_sectors_show,
3977 .store = queue_max_sectors_store,
3978};
3979
3980static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3981 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3982 .show = queue_max_hw_sectors_show,
3983};
3984
3985static struct queue_sysfs_entry queue_iosched_entry = {
3986 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3987 .show = elv_iosched_show,
3988 .store = elv_iosched_store,
3989};
3990
3991static struct attribute *default_attrs[] = {
3992 &queue_requests_entry.attr,
3993 &queue_ra_entry.attr,
3994 &queue_max_hw_sectors_entry.attr,
3995 &queue_max_sectors_entry.attr,
3996 &queue_iosched_entry.attr,
3997 NULL,
3998};
3999
4000#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
4001
4002static ssize_t
4003queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4004{
4005 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004006 struct request_queue *q =
4007 container_of(kobj, struct request_queue, kobj);
Al Viro483f4af2006-03-18 18:34:37 -05004008 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009
Linus Torvalds1da177e2005-04-16 15:20:36 -07004010 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004011 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004012 mutex_lock(&q->sysfs_lock);
4013 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4014 mutex_unlock(&q->sysfs_lock);
4015 return -ENOENT;
4016 }
4017 res = entry->show(q, page);
4018 mutex_unlock(&q->sysfs_lock);
4019 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020}
4021
4022static ssize_t
4023queue_attr_store(struct kobject *kobj, struct attribute *attr,
4024 const char *page, size_t length)
4025{
4026 struct queue_sysfs_entry *entry = to_queue(attr);
Jens Axboe165125e2007-07-24 09:28:11 +02004027 struct request_queue *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028
Al Viro483f4af2006-03-18 18:34:37 -05004029 ssize_t res;
4030
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05004032 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05004033 mutex_lock(&q->sysfs_lock);
4034 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
4035 mutex_unlock(&q->sysfs_lock);
4036 return -ENOENT;
4037 }
4038 res = entry->store(q, page, length);
4039 mutex_unlock(&q->sysfs_lock);
4040 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004041}
4042
4043static struct sysfs_ops queue_sysfs_ops = {
4044 .show = queue_attr_show,
4045 .store = queue_attr_store,
4046};
4047
Adrian Bunk93d17d32005-06-25 14:59:10 -07004048static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049 .sysfs_ops = &queue_sysfs_ops,
4050 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05004051 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052};
4053
4054int blk_register_queue(struct gendisk *disk)
4055{
4056 int ret;
4057
Jens Axboe165125e2007-07-24 09:28:11 +02004058 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004059
4060 if (!q || !q->request_fn)
4061 return -ENXIO;
4062
4063 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004064
Al Viro483f4af2006-03-18 18:34:37 -05004065 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 if (ret < 0)
4067 return ret;
4068
Al Viro483f4af2006-03-18 18:34:37 -05004069 kobject_uevent(&q->kobj, KOBJ_ADD);
4070
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071 ret = elv_register_queue(q);
4072 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004073 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4074 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 return ret;
4076 }
4077
4078 return 0;
4079}
4080
4081void blk_unregister_queue(struct gendisk *disk)
4082{
Jens Axboe165125e2007-07-24 09:28:11 +02004083 struct request_queue *q = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084
4085 if (q && q->request_fn) {
4086 elv_unregister_queue(q);
4087
Al Viro483f4af2006-03-18 18:34:37 -05004088 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4089 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 kobject_put(&disk->kobj);
4091 }
4092}