blob: 83425fb3c8dba6e2b62122aaa6a02a25d2436ab4 [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>
Jens Axboeff856ba2006-01-09 16:02:34 +010028#include <linux/interrupt.h>
29#include <linux/cpu.h>
Jens Axboe2056a782006-03-23 20:00:26 +010030#include <linux/blktrace_api.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32/*
33 * for max sense size
34 */
35#include <scsi/scsi_cmnd.h>
36
37static void blk_unplug_work(void *data);
38static void blk_unplug_timeout(unsigned long data);
Adrian Bunk93d17d32005-06-25 14:59:10 -070039static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
Tejun Heo52d9e672006-01-06 09:49:58 +010040static void init_request_from_bio(struct request *req, struct bio *bio);
41static int __make_request(request_queue_t *q, struct bio *bio);
Jens Axboeb5deef92006-07-19 23:39:40 +020042static struct io_context *current_io_context(gfp_t gfp_flags, int node);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44/*
45 * For the allocated request tables
46 */
47static kmem_cache_t *request_cachep;
48
49/*
50 * For queue allocation
51 */
52static kmem_cache_t *requestq_cachep;
53
54/*
55 * For io context allocations
56 */
57static kmem_cache_t *iocontext_cachep;
58
59static wait_queue_head_t congestion_wqh[2] = {
60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
61 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
62 };
63
64/*
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
115/*
116 * A queue has just exitted congestion. Note this in the global counter of
117 * congested queues, and wake up anyone who was waiting for requests to be
118 * put back.
119 */
120static void clear_queue_congested(request_queue_t *q, int rw)
121{
122 enum bdi_state bit;
123 wait_queue_head_t *wqh = &congestion_wqh[rw];
124
125 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
126 clear_bit(bit, &q->backing_dev_info.state);
127 smp_mb__after_clear_bit();
128 if (waitqueue_active(wqh))
129 wake_up(wqh);
130}
131
132/*
133 * A queue has just entered congestion. Flag that in the queue's VM-visible
134 * state flags and increment the global gounter of congested queues.
135 */
136static void set_queue_congested(request_queue_t *q, int rw)
137{
138 enum bdi_state bit;
139
140 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
141 set_bit(bit, &q->backing_dev_info.state);
142}
143
144/**
145 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
146 * @bdev: device
147 *
148 * Locates the passed device's request queue and returns the address of its
149 * backing_dev_info
150 *
151 * Will return NULL if the request queue cannot be located.
152 */
153struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
154{
155 struct backing_dev_info *ret = NULL;
156 request_queue_t *q = bdev_get_queue(bdev);
157
158 if (q)
159 ret = &q->backing_dev_info;
160 return ret;
161}
162
163EXPORT_SYMBOL(blk_get_backing_dev_info);
164
165void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
166{
167 q->activity_fn = fn;
168 q->activity_data = data;
169}
170
171EXPORT_SYMBOL(blk_queue_activity_fn);
172
173/**
174 * blk_queue_prep_rq - set a prepare_request function for queue
175 * @q: queue
176 * @pfn: prepare_request function
177 *
178 * It's possible for a queue to register a prepare_request callback which
179 * is invoked before the request is handed to the request_fn. The goal of
180 * the function is to prepare a request for I/O, it can be used to build a
181 * cdb from the request data for instance.
182 *
183 */
184void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
185{
186 q->prep_rq_fn = pfn;
187}
188
189EXPORT_SYMBOL(blk_queue_prep_rq);
190
191/**
192 * blk_queue_merge_bvec - set a merge_bvec function for queue
193 * @q: queue
194 * @mbfn: merge_bvec_fn
195 *
196 * Usually queues have static limitations on the max sectors or segments that
197 * we can put in a request. Stacking drivers may have some settings that
198 * are dynamic, and thus we have to query the queue whether it is ok to
199 * add a new bio_vec to a bio at a given offset or not. If the block device
200 * has such limitations, it needs to register a merge_bvec_fn to control
201 * the size of bio's sent to it. Note that a block device *must* allow a
202 * single page to be added to an empty bio. The block device driver may want
203 * to use the bio_split() function to deal with these bio's. By default
204 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
205 * honored.
206 */
207void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
208{
209 q->merge_bvec_fn = mbfn;
210}
211
212EXPORT_SYMBOL(blk_queue_merge_bvec);
213
Jens Axboeff856ba2006-01-09 16:02:34 +0100214void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
215{
216 q->softirq_done_fn = fn;
217}
218
219EXPORT_SYMBOL(blk_queue_softirq_done);
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221/**
222 * blk_queue_make_request - define an alternate make_request function for a device
223 * @q: the request queue for the device to be affected
224 * @mfn: the alternate make_request function
225 *
226 * Description:
227 * The normal way for &struct bios to be passed to a device
228 * driver is for them to be collected into requests on a request
229 * queue, and then to allow the device driver to select requests
230 * off that queue when it is ready. This works well for many block
231 * devices. However some block devices (typically virtual devices
232 * such as md or lvm) do not benefit from the processing on the
233 * request queue, and are served best by having the requests passed
234 * directly to them. This can be achieved by providing a function
235 * to blk_queue_make_request().
236 *
237 * Caveat:
238 * The driver that does this *must* be able to deal appropriately
239 * with buffers in "highmemory". This can be accomplished by either calling
240 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
241 * blk_queue_bounce() to create a buffer in normal memory.
242 **/
243void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
244{
245 /*
246 * set defaults
247 */
248 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700249 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
250 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 q->make_request_fn = mfn;
252 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
253 q->backing_dev_info.state = 0;
254 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600255 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 blk_queue_hardsect_size(q, 512);
257 blk_queue_dma_alignment(q, 511);
258 blk_queue_congestion_threshold(q);
259 q->nr_batching = BLK_BATCH_REQ;
260
261 q->unplug_thresh = 4; /* hmm */
262 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
263 if (q->unplug_delay == 0)
264 q->unplug_delay = 1;
265
266 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
267
268 q->unplug_timer.function = blk_unplug_timeout;
269 q->unplug_timer.data = (unsigned long)q;
270
271 /*
272 * by default assume old behaviour and bounce for any highmem page
273 */
274 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
275
276 blk_queue_activity_fn(q, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
279EXPORT_SYMBOL(blk_queue_make_request);
280
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200281static void rq_init(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
283 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100284 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200288 INIT_HLIST_NODE(&rq->hash);
289 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200290 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 rq->buffer = NULL;
292 rq->ref_count = 1;
293 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 rq->special = NULL;
295 rq->data_len = 0;
296 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200297 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 rq->sense = NULL;
299 rq->end_io = NULL;
300 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100301 rq->completion_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
304/**
305 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100306 * @q: the request queue
307 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100308 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 *
310 * Description:
311 * For journalled file systems, doing ordered writes on a commit
312 * block instead of explicitly doing wait_on_buffer (which is bad
313 * for performance) can be a big win. Block drivers supporting this
314 * feature should call this function and indicate so.
315 *
316 **/
Tejun Heo797e7db2006-01-06 09:51:03 +0100317int blk_queue_ordered(request_queue_t *q, unsigned ordered,
318 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Tejun Heo797e7db2006-01-06 09:51:03 +0100320 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
321 prepare_flush_fn == NULL) {
322 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
323 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100325
326 if (ordered != QUEUE_ORDERED_NONE &&
327 ordered != QUEUE_ORDERED_DRAIN &&
328 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
329 ordered != QUEUE_ORDERED_DRAIN_FUA &&
330 ordered != QUEUE_ORDERED_TAG &&
331 ordered != QUEUE_ORDERED_TAG_FLUSH &&
332 ordered != QUEUE_ORDERED_TAG_FUA) {
333 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
334 return -EINVAL;
335 }
336
Tetsuo Takata60481b12006-01-24 10:34:36 +0100337 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100338 q->next_ordered = ordered;
339 q->prepare_flush_fn = prepare_flush_fn;
340
341 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
344EXPORT_SYMBOL(blk_queue_ordered);
345
346/**
347 * blk_queue_issue_flush_fn - set function for issuing a flush
348 * @q: the request queue
349 * @iff: the function to be called issuing the flush
350 *
351 * Description:
352 * If a driver supports issuing a flush command, the support is notified
353 * to the block layer by defining it through this call.
354 *
355 **/
356void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
357{
358 q->issue_flush_fn = iff;
359}
360
361EXPORT_SYMBOL(blk_queue_issue_flush_fn);
362
363/*
364 * Cache flushing for ordered writes handling
365 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100366inline unsigned blk_ordered_cur_seq(request_queue_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Tejun Heo797e7db2006-01-06 09:51:03 +0100368 if (!q->ordseq)
369 return 0;
370 return 1 << ffz(q->ordseq);
371}
372
373unsigned blk_ordered_req_seq(struct request *rq)
374{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 request_queue_t *q = rq->q;
376
Tejun Heo797e7db2006-01-06 09:51:03 +0100377 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200378
Tejun Heo797e7db2006-01-06 09:51:03 +0100379 if (rq == &q->pre_flush_rq)
380 return QUEUE_ORDSEQ_PREFLUSH;
381 if (rq == &q->bar_rq)
382 return QUEUE_ORDSEQ_BAR;
383 if (rq == &q->post_flush_rq)
384 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Jens Axboe4aff5e22006-08-10 08:44:47 +0200386 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
387 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100388 return QUEUE_ORDSEQ_DRAIN;
389 else
390 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Tejun Heo797e7db2006-01-06 09:51:03 +0100393void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Tejun Heo797e7db2006-01-06 09:51:03 +0100395 struct request *rq;
396 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Tejun Heo797e7db2006-01-06 09:51:03 +0100398 if (error && !q->orderr)
399 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200400
Tejun Heo797e7db2006-01-06 09:51:03 +0100401 BUG_ON(q->ordseq & seq);
402 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Tejun Heo797e7db2006-01-06 09:51:03 +0100404 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
405 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100408 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100410 rq = q->orig_bar_rq;
411 uptodate = q->orderr ? q->orderr : 1;
412
413 q->ordseq = 0;
414
415 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
416 end_that_request_last(rq, uptodate);
417}
418
419static void pre_flush_end_io(struct request *rq, int error)
420{
421 elv_completed_request(rq->q, rq);
422 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
423}
424
425static void bar_end_io(struct request *rq, int error)
426{
427 elv_completed_request(rq->q, rq);
428 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
429}
430
431static void post_flush_end_io(struct request *rq, int error)
432{
433 elv_completed_request(rq->q, rq);
434 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
435}
436
437static void queue_flush(request_queue_t *q, unsigned which)
438{
439 struct request *rq;
440 rq_end_io_fn *end_io;
441
442 if (which == QUEUE_ORDERED_PREFLUSH) {
443 rq = &q->pre_flush_rq;
444 end_io = pre_flush_end_io;
445 } else {
446 rq = &q->post_flush_rq;
447 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
449
Jens Axboe4aff5e22006-08-10 08:44:47 +0200450 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100451 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100452 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200453 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100454 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100455 rq->end_io = end_io;
456 q->prepare_flush_fn(q, rq);
457
Tejun Heo30e96562006-02-08 01:01:31 -0800458 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100459}
460
461static inline struct request *start_ordered(request_queue_t *q,
462 struct request *rq)
463{
464 q->bi_size = 0;
465 q->orderr = 0;
466 q->ordered = q->next_ordered;
467 q->ordseq |= QUEUE_ORDSEQ_STARTED;
468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100470 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100472 blkdev_dequeue_request(rq);
473 q->orig_bar_rq = rq;
474 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200475 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100476 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200477 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
478 rq->cmd_flags |= REQ_RW;
479 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100480 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200481 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100482 init_request_from_bio(rq, q->orig_bar_rq->bio);
483 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Tejun Heo797e7db2006-01-06 09:51:03 +0100485 /*
486 * Queue ordered sequence. As we stack them at the head, we
487 * need to queue in reverse order. Note that we rely on that
488 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
489 * request gets inbetween ordered sequence.
490 */
491 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
492 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
493 else
494 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Tejun Heo30e96562006-02-08 01:01:31 -0800496 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100497
498 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
499 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
500 rq = &q->pre_flush_rq;
501 } else
502 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
503
504 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
505 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
506 else
507 rq = NULL;
508
509 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
Tejun Heo797e7db2006-01-06 09:51:03 +0100512int blk_do_ordered(request_queue_t *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800514 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100515 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Tejun Heo797e7db2006-01-06 09:51:03 +0100517 if (!q->ordseq) {
518 if (!is_barrier)
519 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Tejun Heo797e7db2006-01-06 09:51:03 +0100521 if (q->next_ordered != QUEUE_ORDERED_NONE) {
522 *rqp = start_ordered(q, rq);
523 return 1;
524 } else {
525 /*
526 * This can happen when the queue switches to
527 * ORDERED_NONE while this request is on it.
528 */
529 blkdev_dequeue_request(rq);
530 end_that_request_first(rq, -EOPNOTSUPP,
531 rq->hard_nr_sectors);
532 end_that_request_last(rq, -EOPNOTSUPP);
533 *rqp = NULL;
534 return 0;
535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800538 /*
539 * Ordered sequence in progress
540 */
541
542 /* Special requests are not subject to ordering rules. */
543 if (!blk_fs_request(rq) &&
544 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
545 return 1;
546
Tejun Heo797e7db2006-01-06 09:51:03 +0100547 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800548 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100549 if (is_barrier && rq != &q->bar_rq)
550 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800551 } else {
552 /* Ordered by draining. Wait for turn. */
553 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
554 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
555 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
557
558 return 1;
559}
560
Tejun Heo797e7db2006-01-06 09:51:03 +0100561static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
Tejun Heo797e7db2006-01-06 09:51:03 +0100563 request_queue_t *q = bio->bi_private;
564 struct bio_vec *bvec;
565 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Tejun Heo797e7db2006-01-06 09:51:03 +0100567 /*
568 * This is dry run, restore bio_sector and size. We'll finish
569 * this request again with the original bi_end_io after an
570 * error occurs or post flush is complete.
571 */
572 q->bi_size += bytes;
573
574 if (bio->bi_size)
575 return 1;
576
577 /* Rewind bvec's */
578 bio->bi_idx = 0;
579 bio_for_each_segment(bvec, bio, i) {
580 bvec->bv_len += bvec->bv_offset;
581 bvec->bv_offset = 0;
582 }
583
584 /* Reset bio */
585 set_bit(BIO_UPTODATE, &bio->bi_flags);
586 bio->bi_size = q->bi_size;
587 bio->bi_sector -= (q->bi_size >> 9);
588 q->bi_size = 0;
589
590 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
Tejun Heo797e7db2006-01-06 09:51:03 +0100592
Jens Axboe1ea25ec2006-07-18 22:24:11 +0200593static int ordered_bio_endio(struct request *rq, struct bio *bio,
594 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100595{
596 request_queue_t *q = rq->q;
597 bio_end_io_t *endio;
598 void *private;
599
600 if (&q->bar_rq != rq)
601 return 0;
602
603 /*
604 * Okay, this is the barrier request in progress, dry finish it.
605 */
606 if (error && !q->orderr)
607 q->orderr = error;
608
609 endio = bio->bi_end_io;
610 private = bio->bi_private;
611 bio->bi_end_io = flush_dry_bio_endio;
612 bio->bi_private = q;
613
614 bio_endio(bio, nbytes, error);
615
616 bio->bi_end_io = endio;
617 bio->bi_private = private;
618
619 return 1;
620}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622/**
623 * blk_queue_bounce_limit - set bounce buffer limit for queue
624 * @q: the request queue for the device
625 * @dma_addr: bus address limit
626 *
627 * Description:
628 * Different hardware can have different requirements as to what pages
629 * it can do I/O directly to. A low level driver can call
630 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800631 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 **/
633void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
634{
635 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800636 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Andi Kleen5ee1af92006-03-08 17:57:26 -0800638 q->bounce_gfp = GFP_NOIO;
639#if BITS_PER_LONG == 64
640 /* Assume anything <= 4GB can be handled by IOMMU.
641 Actually some IOMMUs can handle everything, but I don't
642 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200643 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800644 dma = 1;
645 q->bounce_pfn = max_low_pfn;
646#else
647 if (bounce_pfn < blk_max_low_pfn)
648 dma = 1;
649 q->bounce_pfn = bounce_pfn;
650#endif
651 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 init_emergency_isa_pool();
653 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800654 q->bounce_pfn = bounce_pfn;
655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656}
657
658EXPORT_SYMBOL(blk_queue_bounce_limit);
659
660/**
661 * blk_queue_max_sectors - set max sectors for a request for this queue
662 * @q: the request queue for the device
663 * @max_sectors: max sectors in the usual 512b unit
664 *
665 * Description:
666 * Enables a low level driver to set an upper limit on the size of
667 * received requests.
668 **/
Jens Axboe2cb2e142006-01-17 09:04:32 +0100669void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
671 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
672 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
673 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
674 }
675
Mike Christiedefd94b2005-12-05 02:37:06 -0600676 if (BLK_DEF_MAX_SECTORS > max_sectors)
677 q->max_hw_sectors = q->max_sectors = max_sectors;
678 else {
679 q->max_sectors = BLK_DEF_MAX_SECTORS;
680 q->max_hw_sectors = max_sectors;
681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
684EXPORT_SYMBOL(blk_queue_max_sectors);
685
686/**
687 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
688 * @q: the request queue for the device
689 * @max_segments: max number of segments
690 *
691 * Description:
692 * Enables a low level driver to set an upper limit on the number of
693 * physical data segments in a request. This would be the largest sized
694 * scatter list the driver could handle.
695 **/
696void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
697{
698 if (!max_segments) {
699 max_segments = 1;
700 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
701 }
702
703 q->max_phys_segments = max_segments;
704}
705
706EXPORT_SYMBOL(blk_queue_max_phys_segments);
707
708/**
709 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
710 * @q: the request queue for the device
711 * @max_segments: max number of segments
712 *
713 * Description:
714 * Enables a low level driver to set an upper limit on the number of
715 * hw data segments in a request. This would be the largest number of
716 * address/length pairs the host adapter can actually give as once
717 * to the device.
718 **/
719void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
720{
721 if (!max_segments) {
722 max_segments = 1;
723 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
724 }
725
726 q->max_hw_segments = max_segments;
727}
728
729EXPORT_SYMBOL(blk_queue_max_hw_segments);
730
731/**
732 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
733 * @q: the request queue for the device
734 * @max_size: max size of segment in bytes
735 *
736 * Description:
737 * Enables a low level driver to set an upper limit on the size of a
738 * coalesced segment
739 **/
740void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
741{
742 if (max_size < PAGE_CACHE_SIZE) {
743 max_size = PAGE_CACHE_SIZE;
744 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
745 }
746
747 q->max_segment_size = max_size;
748}
749
750EXPORT_SYMBOL(blk_queue_max_segment_size);
751
752/**
753 * blk_queue_hardsect_size - set hardware sector size for the queue
754 * @q: the request queue for the device
755 * @size: the hardware sector size, in bytes
756 *
757 * Description:
758 * This should typically be set to the lowest possible sector size
759 * that the hardware can operate on (possible without reverting to
760 * even internal read-modify-write operations). Usually the default
761 * of 512 covers most hardware.
762 **/
763void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
764{
765 q->hardsect_size = size;
766}
767
768EXPORT_SYMBOL(blk_queue_hardsect_size);
769
770/*
771 * Returns the minimum that is _not_ zero, unless both are zero.
772 */
773#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
774
775/**
776 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
777 * @t: the stacking driver (top)
778 * @b: the underlying device (bottom)
779 **/
780void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
781{
782 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600783 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
784 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
786 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
787 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
788 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
789 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800790 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
791 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792}
793
794EXPORT_SYMBOL(blk_queue_stack_limits);
795
796/**
797 * blk_queue_segment_boundary - set boundary rules for segment merging
798 * @q: the request queue for the device
799 * @mask: the memory boundary mask
800 **/
801void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
802{
803 if (mask < PAGE_CACHE_SIZE - 1) {
804 mask = PAGE_CACHE_SIZE - 1;
805 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
806 }
807
808 q->seg_boundary_mask = mask;
809}
810
811EXPORT_SYMBOL(blk_queue_segment_boundary);
812
813/**
814 * blk_queue_dma_alignment - set dma length and memory alignment
815 * @q: the request queue for the device
816 * @mask: alignment mask
817 *
818 * description:
819 * set required memory and length aligment for direct dma transactions.
820 * this is used when buiding direct io requests for the queue.
821 *
822 **/
823void blk_queue_dma_alignment(request_queue_t *q, int mask)
824{
825 q->dma_alignment = mask;
826}
827
828EXPORT_SYMBOL(blk_queue_dma_alignment);
829
830/**
831 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 * @q: The request queue for the device
833 * @tag: The tag of the request
834 *
835 * Notes:
836 * Should be used when a device returns a tag and you want to match
837 * it with a request.
838 *
839 * no locks need be held.
840 **/
841struct request *blk_queue_find_tag(request_queue_t *q, int tag)
842{
843 struct blk_queue_tag *bqt = q->queue_tags;
844
Tejun Heoba025082005-08-05 13:28:11 -0700845 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return NULL;
847
848 return bqt->tag_index[tag];
849}
850
851EXPORT_SYMBOL(blk_queue_find_tag);
852
853/**
James Bottomley492dfb42006-08-30 15:48:45 -0400854 * __blk_free_tags - release a given set of tag maintenance info
855 * @bqt: the tag map to free
856 *
857 * Tries to free the specified @bqt@. Returns true if it was
858 * actually freed and false if there are still references using it
859 */
860static int __blk_free_tags(struct blk_queue_tag *bqt)
861{
862 int retval;
863
864 retval = atomic_dec_and_test(&bqt->refcnt);
865 if (retval) {
866 BUG_ON(bqt->busy);
867 BUG_ON(!list_empty(&bqt->busy_list));
868
869 kfree(bqt->tag_index);
870 bqt->tag_index = NULL;
871
872 kfree(bqt->tag_map);
873 bqt->tag_map = NULL;
874
875 kfree(bqt);
876
877 }
878
879 return retval;
880}
881
882/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 * __blk_queue_free_tags - release tag maintenance info
884 * @q: the request queue for the device
885 *
886 * Notes:
887 * blk_cleanup_queue() will take care of calling this function, if tagging
888 * has been used. So there's no need to call this directly.
889 **/
890static void __blk_queue_free_tags(request_queue_t *q)
891{
892 struct blk_queue_tag *bqt = q->queue_tags;
893
894 if (!bqt)
895 return;
896
James Bottomley492dfb42006-08-30 15:48:45 -0400897 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 q->queue_tags = NULL;
900 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
901}
902
James Bottomley492dfb42006-08-30 15:48:45 -0400903
904/**
905 * blk_free_tags - release a given set of tag maintenance info
906 * @bqt: the tag map to free
907 *
908 * For externally managed @bqt@ frees the map. Callers of this
909 * function must guarantee to have released all the queues that
910 * might have been using this tag map.
911 */
912void blk_free_tags(struct blk_queue_tag *bqt)
913{
914 if (unlikely(!__blk_free_tags(bqt)))
915 BUG();
916}
917EXPORT_SYMBOL(blk_free_tags);
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919/**
920 * blk_queue_free_tags - release tag maintenance info
921 * @q: the request queue for the device
922 *
923 * Notes:
924 * This is used to disabled tagged queuing to a device, yet leave
925 * queue in function.
926 **/
927void blk_queue_free_tags(request_queue_t *q)
928{
929 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
930}
931
932EXPORT_SYMBOL(blk_queue_free_tags);
933
934static int
935init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
936{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 struct request **tag_index;
938 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700939 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
James Bottomley492dfb42006-08-30 15:48:45 -0400941 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 depth = q->nr_requests * 2;
943 printk(KERN_ERR "%s: adjusted depth to %d\n",
944 __FUNCTION__, depth);
945 }
946
Jens Axboef68110f2006-03-08 13:31:44 +0100947 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (!tag_index)
949 goto fail;
950
Tejun Heof7d37d02005-06-23 00:08:50 -0700951 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100952 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (!tag_map)
954 goto fail;
955
Tejun Heoba025082005-08-05 13:28:11 -0700956 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 tags->tag_index = tag_index;
959 tags->tag_map = tag_map;
960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 return 0;
962fail:
963 kfree(tag_index);
964 return -ENOMEM;
965}
966
James Bottomley492dfb42006-08-30 15:48:45 -0400967static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
968 int depth)
969{
970 struct blk_queue_tag *tags;
971
972 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
973 if (!tags)
974 goto fail;
975
976 if (init_tag_map(q, tags, depth))
977 goto fail;
978
979 INIT_LIST_HEAD(&tags->busy_list);
980 tags->busy = 0;
981 atomic_set(&tags->refcnt, 1);
982 return tags;
983fail:
984 kfree(tags);
985 return NULL;
986}
987
988/**
989 * blk_init_tags - initialize the tag info for an external tag map
990 * @depth: the maximum queue depth supported
991 * @tags: the tag to use
992 **/
993struct blk_queue_tag *blk_init_tags(int depth)
994{
995 return __blk_queue_init_tags(NULL, depth);
996}
997EXPORT_SYMBOL(blk_init_tags);
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/**
1000 * blk_queue_init_tags - initialize the queue tag info
1001 * @q: the request queue for the device
1002 * @depth: the maximum queue depth supported
1003 * @tags: the tag to use
1004 **/
1005int blk_queue_init_tags(request_queue_t *q, int depth,
1006 struct blk_queue_tag *tags)
1007{
1008 int rc;
1009
1010 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
1011
1012 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -04001013 tags = __blk_queue_init_tags(q, depth);
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (!tags)
1016 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 } else if (q->queue_tags) {
1018 if ((rc = blk_queue_resize_tags(q, depth)))
1019 return rc;
1020 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
1021 return 0;
1022 } else
1023 atomic_inc(&tags->refcnt);
1024
1025 /*
1026 * assign it, all done
1027 */
1028 q->queue_tags = tags;
1029 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
1030 return 0;
1031fail:
1032 kfree(tags);
1033 return -ENOMEM;
1034}
1035
1036EXPORT_SYMBOL(blk_queue_init_tags);
1037
1038/**
1039 * blk_queue_resize_tags - change the queueing depth
1040 * @q: the request queue for the device
1041 * @new_depth: the new max command queueing depth
1042 *
1043 * Notes:
1044 * Must be called with the queue lock held.
1045 **/
1046int blk_queue_resize_tags(request_queue_t *q, int new_depth)
1047{
1048 struct blk_queue_tag *bqt = q->queue_tags;
1049 struct request **tag_index;
1050 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -07001051 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 if (!bqt)
1054 return -ENXIO;
1055
1056 /*
Tejun Heoba025082005-08-05 13:28:11 -07001057 * if we already have large enough real_max_depth. just
1058 * adjust max_depth. *NOTE* as requests with tag value
1059 * between new_depth and real_max_depth can be in-flight, tag
1060 * map can not be shrunk blindly here.
1061 */
1062 if (new_depth <= bqt->real_max_depth) {
1063 bqt->max_depth = new_depth;
1064 return 0;
1065 }
1066
1067 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001068 * Currently cannot replace a shared tag map with a new
1069 * one, so error out if this is the case
1070 */
1071 if (atomic_read(&bqt->refcnt) != 1)
1072 return -EBUSY;
1073
1074 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 * save the old state info, so we can copy it back
1076 */
1077 tag_index = bqt->tag_index;
1078 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001079 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 if (init_tag_map(q, bqt, new_depth))
1082 return -ENOMEM;
1083
1084 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001085 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001086 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 kfree(tag_index);
1089 kfree(tag_map);
1090 return 0;
1091}
1092
1093EXPORT_SYMBOL(blk_queue_resize_tags);
1094
1095/**
1096 * blk_queue_end_tag - end tag operations for a request
1097 * @q: the request queue for the device
1098 * @rq: the request that has completed
1099 *
1100 * Description:
1101 * Typically called when end_that_request_first() returns 0, meaning
1102 * all transfers have been done for a request. It's important to call
1103 * this function before end_that_request_last(), as that will put the
1104 * request back on the free list thus corrupting the internal tag list.
1105 *
1106 * Notes:
1107 * queue lock must be held.
1108 **/
1109void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1110{
1111 struct blk_queue_tag *bqt = q->queue_tags;
1112 int tag = rq->tag;
1113
1114 BUG_ON(tag == -1);
1115
Tejun Heoba025082005-08-05 13:28:11 -07001116 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001117 /*
1118 * This can happen after tag depth has been reduced.
1119 * FIXME: how about a warning or info message here?
1120 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 return;
1122
1123 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001124 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1125 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 return;
1127 }
1128
1129 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001130 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 rq->tag = -1;
1132
1133 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001134 printk(KERN_ERR "%s: tag %d is missing\n",
1135 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 bqt->tag_index[tag] = NULL;
1138 bqt->busy--;
1139}
1140
1141EXPORT_SYMBOL(blk_queue_end_tag);
1142
1143/**
1144 * blk_queue_start_tag - find a free tag and assign it
1145 * @q: the request queue for the device
1146 * @rq: the block request that needs tagging
1147 *
1148 * Description:
1149 * This can either be used as a stand-alone helper, or possibly be
1150 * assigned as the queue &prep_rq_fn (in which case &struct request
1151 * automagically gets a tag assigned). Note that this function
1152 * assumes that any type of request can be queued! if this is not
1153 * true for your device, you must check the request type before
1154 * calling this function. The request will also be removed from
1155 * the request queue, so it's the drivers responsibility to readd
1156 * it if it should need to be restarted for some reason.
1157 *
1158 * Notes:
1159 * queue lock must be held.
1160 **/
1161int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1162{
1163 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001164 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
Jens Axboe4aff5e22006-08-10 08:44:47 +02001166 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001168 "%s: request %p for device [%s] already tagged %d",
1169 __FUNCTION__, rq,
1170 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 BUG();
1172 }
1173
Jens Axboe059af492006-09-21 20:37:22 +02001174 /*
1175 * Protect against shared tag maps, as we may not have exclusive
1176 * access to the tag map.
1177 */
1178 do {
1179 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1180 if (tag >= bqt->max_depth)
1181 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Jens Axboe059af492006-09-21 20:37:22 +02001183 } while (test_and_set_bit(tag, bqt->tag_map));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
Jens Axboe4aff5e22006-08-10 08:44:47 +02001185 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 rq->tag = tag;
1187 bqt->tag_index[tag] = rq;
1188 blkdev_dequeue_request(rq);
1189 list_add(&rq->queuelist, &bqt->busy_list);
1190 bqt->busy++;
1191 return 0;
1192}
1193
1194EXPORT_SYMBOL(blk_queue_start_tag);
1195
1196/**
1197 * blk_queue_invalidate_tags - invalidate all pending tags
1198 * @q: the request queue for the device
1199 *
1200 * Description:
1201 * Hardware conditions may dictate a need to stop all pending requests.
1202 * In this case, we will safely clear the block side of the tag queue and
1203 * readd all requests to the request queue in the right order.
1204 *
1205 * Notes:
1206 * queue lock must be held.
1207 **/
1208void blk_queue_invalidate_tags(request_queue_t *q)
1209{
1210 struct blk_queue_tag *bqt = q->queue_tags;
1211 struct list_head *tmp, *n;
1212 struct request *rq;
1213
1214 list_for_each_safe(tmp, n, &bqt->busy_list) {
1215 rq = list_entry_rq(tmp);
1216
1217 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001218 printk(KERN_ERR
1219 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001221 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 } else
1223 blk_queue_end_tag(q, rq);
1224
Jens Axboe4aff5e22006-08-10 08:44:47 +02001225 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1227 }
1228}
1229
1230EXPORT_SYMBOL(blk_queue_invalidate_tags);
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232void blk_dump_rq_flags(struct request *rq, char *msg)
1233{
1234 int bit;
1235
Jens Axboe4aff5e22006-08-10 08:44:47 +02001236 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1237 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1238 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1241 rq->nr_sectors,
1242 rq->current_nr_sectors);
1243 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1244
Jens Axboe4aff5e22006-08-10 08:44:47 +02001245 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 printk("cdb: ");
1247 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1248 printk("%02x ", rq->cmd[bit]);
1249 printk("\n");
1250 }
1251}
1252
1253EXPORT_SYMBOL(blk_dump_rq_flags);
1254
1255void blk_recount_segments(request_queue_t *q, struct bio *bio)
1256{
1257 struct bio_vec *bv, *bvprv = NULL;
1258 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1259 int high, highprv = 1;
1260
1261 if (unlikely(!bio->bi_io_vec))
1262 return;
1263
1264 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1265 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1266 bio_for_each_segment(bv, bio, i) {
1267 /*
1268 * the trick here is making sure that a high page is never
1269 * considered part of another segment, since that might
1270 * change with the bounce page.
1271 */
1272 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1273 if (high || highprv)
1274 goto new_hw_segment;
1275 if (cluster) {
1276 if (seg_size + bv->bv_len > q->max_segment_size)
1277 goto new_segment;
1278 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1279 goto new_segment;
1280 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1281 goto new_segment;
1282 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1283 goto new_hw_segment;
1284
1285 seg_size += bv->bv_len;
1286 hw_seg_size += bv->bv_len;
1287 bvprv = bv;
1288 continue;
1289 }
1290new_segment:
1291 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1292 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1293 hw_seg_size += bv->bv_len;
1294 } else {
1295new_hw_segment:
1296 if (hw_seg_size > bio->bi_hw_front_size)
1297 bio->bi_hw_front_size = hw_seg_size;
1298 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1299 nr_hw_segs++;
1300 }
1301
1302 nr_phys_segs++;
1303 bvprv = bv;
1304 seg_size = bv->bv_len;
1305 highprv = high;
1306 }
1307 if (hw_seg_size > bio->bi_hw_back_size)
1308 bio->bi_hw_back_size = hw_seg_size;
1309 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1310 bio->bi_hw_front_size = hw_seg_size;
1311 bio->bi_phys_segments = nr_phys_segs;
1312 bio->bi_hw_segments = nr_hw_segs;
1313 bio->bi_flags |= (1 << BIO_SEG_VALID);
1314}
1315
1316
Adrian Bunk93d17d32005-06-25 14:59:10 -07001317static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 struct bio *nxt)
1319{
1320 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1321 return 0;
1322
1323 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1324 return 0;
1325 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1326 return 0;
1327
1328 /*
1329 * bio and nxt are contigous in memory, check if the queue allows
1330 * these two to be merged into one
1331 */
1332 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1333 return 1;
1334
1335 return 0;
1336}
1337
Adrian Bunk93d17d32005-06-25 14:59:10 -07001338static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 struct bio *nxt)
1340{
1341 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1342 blk_recount_segments(q, bio);
1343 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1344 blk_recount_segments(q, nxt);
1345 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1346 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1347 return 0;
1348 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1349 return 0;
1350
1351 return 1;
1352}
1353
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354/*
1355 * map a request to scatterlist, return number of sg entries setup. Caller
1356 * must make sure sg can hold rq->nr_phys_segments entries
1357 */
1358int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1359{
1360 struct bio_vec *bvec, *bvprv;
1361 struct bio *bio;
1362 int nsegs, i, cluster;
1363
1364 nsegs = 0;
1365 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1366
1367 /*
1368 * for each bio in rq
1369 */
1370 bvprv = NULL;
1371 rq_for_each_bio(bio, rq) {
1372 /*
1373 * for each segment in bio
1374 */
1375 bio_for_each_segment(bvec, bio, i) {
1376 int nbytes = bvec->bv_len;
1377
1378 if (bvprv && cluster) {
1379 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1380 goto new_segment;
1381
1382 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1383 goto new_segment;
1384 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1385 goto new_segment;
1386
1387 sg[nsegs - 1].length += nbytes;
1388 } else {
1389new_segment:
1390 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1391 sg[nsegs].page = bvec->bv_page;
1392 sg[nsegs].length = nbytes;
1393 sg[nsegs].offset = bvec->bv_offset;
1394
1395 nsegs++;
1396 }
1397 bvprv = bvec;
1398 } /* segments in bio */
1399 } /* bios in rq */
1400
1401 return nsegs;
1402}
1403
1404EXPORT_SYMBOL(blk_rq_map_sg);
1405
1406/*
1407 * the standard queue merge functions, can be overridden with device
1408 * specific ones if so desired
1409 */
1410
1411static inline int ll_new_mergeable(request_queue_t *q,
1412 struct request *req,
1413 struct bio *bio)
1414{
1415 int nr_phys_segs = bio_phys_segments(q, bio);
1416
1417 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001418 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (req == q->last_merge)
1420 q->last_merge = NULL;
1421 return 0;
1422 }
1423
1424 /*
1425 * A hw segment is just getting larger, bump just the phys
1426 * counter.
1427 */
1428 req->nr_phys_segments += nr_phys_segs;
1429 return 1;
1430}
1431
1432static inline int ll_new_hw_segment(request_queue_t *q,
1433 struct request *req,
1434 struct bio *bio)
1435{
1436 int nr_hw_segs = bio_hw_segments(q, bio);
1437 int nr_phys_segs = bio_phys_segments(q, bio);
1438
1439 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1440 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001441 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 if (req == q->last_merge)
1443 q->last_merge = NULL;
1444 return 0;
1445 }
1446
1447 /*
1448 * This will form the start of a new hw segment. Bump both
1449 * counters.
1450 */
1451 req->nr_hw_segments += nr_hw_segs;
1452 req->nr_phys_segments += nr_phys_segs;
1453 return 1;
1454}
1455
1456static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1457 struct bio *bio)
1458{
Mike Christiedefd94b2005-12-05 02:37:06 -06001459 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 int len;
1461
Mike Christiedefd94b2005-12-05 02:37:06 -06001462 if (unlikely(blk_pc_request(req)))
1463 max_sectors = q->max_hw_sectors;
1464 else
1465 max_sectors = q->max_sectors;
1466
1467 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001468 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (req == q->last_merge)
1470 q->last_merge = NULL;
1471 return 0;
1472 }
1473 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1474 blk_recount_segments(q, req->biotail);
1475 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1476 blk_recount_segments(q, bio);
1477 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1478 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1479 !BIOVEC_VIRT_OVERSIZE(len)) {
1480 int mergeable = ll_new_mergeable(q, req, bio);
1481
1482 if (mergeable) {
1483 if (req->nr_hw_segments == 1)
1484 req->bio->bi_hw_front_size = len;
1485 if (bio->bi_hw_segments == 1)
1486 bio->bi_hw_back_size = len;
1487 }
1488 return mergeable;
1489 }
1490
1491 return ll_new_hw_segment(q, req, bio);
1492}
1493
1494static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1495 struct bio *bio)
1496{
Mike Christiedefd94b2005-12-05 02:37:06 -06001497 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 int len;
1499
Mike Christiedefd94b2005-12-05 02:37:06 -06001500 if (unlikely(blk_pc_request(req)))
1501 max_sectors = q->max_hw_sectors;
1502 else
1503 max_sectors = q->max_sectors;
1504
1505
1506 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001507 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (req == q->last_merge)
1509 q->last_merge = NULL;
1510 return 0;
1511 }
1512 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1513 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1514 blk_recount_segments(q, bio);
1515 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1516 blk_recount_segments(q, req->bio);
1517 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1518 !BIOVEC_VIRT_OVERSIZE(len)) {
1519 int mergeable = ll_new_mergeable(q, req, bio);
1520
1521 if (mergeable) {
1522 if (bio->bi_hw_segments == 1)
1523 bio->bi_hw_front_size = len;
1524 if (req->nr_hw_segments == 1)
1525 req->biotail->bi_hw_back_size = len;
1526 }
1527 return mergeable;
1528 }
1529
1530 return ll_new_hw_segment(q, req, bio);
1531}
1532
1533static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1534 struct request *next)
1535{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001536 int total_phys_segments;
1537 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539 /*
1540 * First check if the either of the requests are re-queued
1541 * requests. Can't merge them if they are.
1542 */
1543 if (req->special || next->special)
1544 return 0;
1545
1546 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001547 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 */
1549 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1550 return 0;
1551
1552 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1553 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1554 total_phys_segments--;
1555
1556 if (total_phys_segments > q->max_phys_segments)
1557 return 0;
1558
1559 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1560 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1561 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1562 /*
1563 * propagate the combined length to the end of the requests
1564 */
1565 if (req->nr_hw_segments == 1)
1566 req->bio->bi_hw_front_size = len;
1567 if (next->nr_hw_segments == 1)
1568 next->biotail->bi_hw_back_size = len;
1569 total_hw_segments--;
1570 }
1571
1572 if (total_hw_segments > q->max_hw_segments)
1573 return 0;
1574
1575 /* Merge is OK... */
1576 req->nr_phys_segments = total_phys_segments;
1577 req->nr_hw_segments = total_hw_segments;
1578 return 1;
1579}
1580
1581/*
1582 * "plug" the device if there are no outstanding requests: this will
1583 * force the transfer to start only after we have put all the requests
1584 * on the list.
1585 *
1586 * This is called with interrupts off and no requests on the queue and
1587 * with the queue lock held.
1588 */
1589void blk_plug_device(request_queue_t *q)
1590{
1591 WARN_ON(!irqs_disabled());
1592
1593 /*
1594 * don't plug a stopped queue, it must be paired with blk_start_queue()
1595 * which will restart the queueing
1596 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001597 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return;
1599
Jens Axboe2056a782006-03-23 20:00:26 +01001600 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001602 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
1605
1606EXPORT_SYMBOL(blk_plug_device);
1607
1608/*
1609 * remove the queue from the plugged list, if present. called with
1610 * queue lock held and interrupts disabled.
1611 */
1612int blk_remove_plug(request_queue_t *q)
1613{
1614 WARN_ON(!irqs_disabled());
1615
1616 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1617 return 0;
1618
1619 del_timer(&q->unplug_timer);
1620 return 1;
1621}
1622
1623EXPORT_SYMBOL(blk_remove_plug);
1624
1625/*
1626 * remove the plug and let it rip..
1627 */
1628void __generic_unplug_device(request_queue_t *q)
1629{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001630 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 return;
1632
1633 if (!blk_remove_plug(q))
1634 return;
1635
Jens Axboe22e2c502005-06-27 10:55:12 +02001636 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637}
1638EXPORT_SYMBOL(__generic_unplug_device);
1639
1640/**
1641 * generic_unplug_device - fire a request queue
1642 * @q: The &request_queue_t in question
1643 *
1644 * Description:
1645 * Linux uses plugging to build bigger requests queues before letting
1646 * the device have at them. If a queue is plugged, the I/O scheduler
1647 * is still adding and merging requests on the queue. Once the queue
1648 * gets unplugged, the request_fn defined for the queue is invoked and
1649 * transfers started.
1650 **/
1651void generic_unplug_device(request_queue_t *q)
1652{
1653 spin_lock_irq(q->queue_lock);
1654 __generic_unplug_device(q);
1655 spin_unlock_irq(q->queue_lock);
1656}
1657EXPORT_SYMBOL(generic_unplug_device);
1658
1659static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1660 struct page *page)
1661{
1662 request_queue_t *q = bdi->unplug_io_data;
1663
1664 /*
1665 * devices don't necessarily have an ->unplug_fn defined
1666 */
Jens Axboe2056a782006-03-23 20:00:26 +01001667 if (q->unplug_fn) {
1668 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1669 q->rq.count[READ] + q->rq.count[WRITE]);
1670
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673}
1674
1675static void blk_unplug_work(void *data)
1676{
1677 request_queue_t *q = data;
1678
Jens Axboe2056a782006-03-23 20:00:26 +01001679 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1680 q->rq.count[READ] + q->rq.count[WRITE]);
1681
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 q->unplug_fn(q);
1683}
1684
1685static void blk_unplug_timeout(unsigned long data)
1686{
1687 request_queue_t *q = (request_queue_t *)data;
1688
Jens Axboe2056a782006-03-23 20:00:26 +01001689 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1690 q->rq.count[READ] + q->rq.count[WRITE]);
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 kblockd_schedule_work(&q->unplug_work);
1693}
1694
1695/**
1696 * blk_start_queue - restart a previously stopped queue
1697 * @q: The &request_queue_t in question
1698 *
1699 * Description:
1700 * blk_start_queue() will clear the stop flag on the queue, and call
1701 * the request_fn for the queue if it was in a stopped state when
1702 * entered. Also see blk_stop_queue(). Queue lock must be held.
1703 **/
1704void blk_start_queue(request_queue_t *q)
1705{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001706 WARN_ON(!irqs_disabled());
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1709
1710 /*
1711 * one level of recursion is ok and is much faster than kicking
1712 * the unplug handling
1713 */
1714 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1715 q->request_fn(q);
1716 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1717 } else {
1718 blk_plug_device(q);
1719 kblockd_schedule_work(&q->unplug_work);
1720 }
1721}
1722
1723EXPORT_SYMBOL(blk_start_queue);
1724
1725/**
1726 * blk_stop_queue - stop a queue
1727 * @q: The &request_queue_t in question
1728 *
1729 * Description:
1730 * The Linux block layer assumes that a block driver will consume all
1731 * entries on the request queue when the request_fn strategy is called.
1732 * Often this will not happen, because of hardware limitations (queue
1733 * depth settings). If a device driver gets a 'queue full' response,
1734 * or if it simply chooses not to queue more I/O at one point, it can
1735 * call this function to prevent the request_fn from being called until
1736 * the driver has signalled it's ready to go again. This happens by calling
1737 * blk_start_queue() to restart queue operations. Queue lock must be held.
1738 **/
1739void blk_stop_queue(request_queue_t *q)
1740{
1741 blk_remove_plug(q);
1742 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1743}
1744EXPORT_SYMBOL(blk_stop_queue);
1745
1746/**
1747 * blk_sync_queue - cancel any pending callbacks on a queue
1748 * @q: the queue
1749 *
1750 * Description:
1751 * The block layer may perform asynchronous callback activity
1752 * on a queue, such as calling the unplug function after a timeout.
1753 * A block device may call blk_sync_queue to ensure that any
1754 * such activity is cancelled, thus allowing it to release resources
1755 * the the callbacks might use. The caller must already have made sure
1756 * that its ->make_request_fn will not re-add plugging prior to calling
1757 * this function.
1758 *
1759 */
1760void blk_sync_queue(struct request_queue *q)
1761{
1762 del_timer_sync(&q->unplug_timer);
1763 kblockd_flush();
1764}
1765EXPORT_SYMBOL(blk_sync_queue);
1766
1767/**
1768 * blk_run_queue - run a single device queue
1769 * @q: The queue to run
1770 */
1771void blk_run_queue(struct request_queue *q)
1772{
1773 unsigned long flags;
1774
1775 spin_lock_irqsave(q->queue_lock, flags);
1776 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001777
1778 /*
1779 * Only recurse once to avoid overrunning the stack, let the unplug
1780 * handling reinvoke the handler shortly if we already got there.
1781 */
1782 if (!elv_queue_empty(q)) {
1783 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1784 q->request_fn(q);
1785 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1786 } else {
1787 blk_plug_device(q);
1788 kblockd_schedule_work(&q->unplug_work);
1789 }
1790 }
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 spin_unlock_irqrestore(q->queue_lock, flags);
1793}
1794EXPORT_SYMBOL(blk_run_queue);
1795
1796/**
1797 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001798 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 *
1800 * Description:
1801 * blk_cleanup_queue is the pair to blk_init_queue() or
1802 * blk_queue_make_request(). It should be called when a request queue is
1803 * being released; typically when a block device is being de-registered.
1804 * Currently, its primary task it to free all the &struct request
1805 * structures that were allocated to the queue and the queue itself.
1806 *
1807 * Caveat:
1808 * Hopefully the low level driver will have finished any
1809 * outstanding requests first...
1810 **/
Al Viro483f4af2006-03-18 18:34:37 -05001811static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
Al Viro483f4af2006-03-18 18:34:37 -05001813 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 struct request_list *rl = &q->rq;
1815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 blk_sync_queue(q);
1817
1818 if (rl->rq_pool)
1819 mempool_destroy(rl->rq_pool);
1820
1821 if (q->queue_tags)
1822 __blk_queue_free_tags(q);
1823
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001824 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 kmem_cache_free(requestq_cachep, q);
1827}
1828
Al Viro483f4af2006-03-18 18:34:37 -05001829void blk_put_queue(request_queue_t *q)
1830{
1831 kobject_put(&q->kobj);
1832}
1833EXPORT_SYMBOL(blk_put_queue);
1834
1835void blk_cleanup_queue(request_queue_t * q)
1836{
1837 mutex_lock(&q->sysfs_lock);
1838 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1839 mutex_unlock(&q->sysfs_lock);
1840
1841 if (q->elevator)
1842 elevator_exit(q->elevator);
1843
1844 blk_put_queue(q);
1845}
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847EXPORT_SYMBOL(blk_cleanup_queue);
1848
1849static int blk_init_free_list(request_queue_t *q)
1850{
1851 struct request_list *rl = &q->rq;
1852
1853 rl->count[READ] = rl->count[WRITE] = 0;
1854 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001855 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 init_waitqueue_head(&rl->wait[READ]);
1857 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Christoph Lameter19460892005-06-23 00:08:19 -07001859 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1860 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862 if (!rl->rq_pool)
1863 return -ENOMEM;
1864
1865 return 0;
1866}
1867
Al Viro8267e262005-10-21 03:20:53 -04001868request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869{
Christoph Lameter19460892005-06-23 00:08:19 -07001870 return blk_alloc_queue_node(gfp_mask, -1);
1871}
1872EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Al Viro483f4af2006-03-18 18:34:37 -05001874static struct kobj_type queue_ktype;
1875
Al Viro8267e262005-10-21 03:20:53 -04001876request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001877{
1878 request_queue_t *q;
1879
1880 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (!q)
1882 return NULL;
1883
1884 memset(q, 0, sizeof(*q));
1885 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001886
1887 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1888 q->kobj.ktype = &queue_ktype;
1889 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
1891 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1892 q->backing_dev_info.unplug_io_data = q;
1893
Al Viro483f4af2006-03-18 18:34:37 -05001894 mutex_init(&q->sysfs_lock);
1895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 return q;
1897}
Christoph Lameter19460892005-06-23 00:08:19 -07001898EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
1900/**
1901 * blk_init_queue - prepare a request queue for use with a block device
1902 * @rfn: The function to be called to process requests that have been
1903 * placed on the queue.
1904 * @lock: Request queue spin lock
1905 *
1906 * Description:
1907 * If a block device wishes to use the standard request handling procedures,
1908 * which sorts requests and coalesces adjacent requests, then it must
1909 * call blk_init_queue(). The function @rfn will be called when there
1910 * are requests on the queue that need to be processed. If the device
1911 * supports plugging, then @rfn may not be called immediately when requests
1912 * are available on the queue, but may be called at some time later instead.
1913 * Plugged queues are generally unplugged when a buffer belonging to one
1914 * of the requests on the queue is needed, or due to memory pressure.
1915 *
1916 * @rfn is not required, or even expected, to remove all requests off the
1917 * queue, but only as many as it can handle at a time. If it does leave
1918 * requests on the queue, it is responsible for arranging that the requests
1919 * get dealt with eventually.
1920 *
1921 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001922 * request queue; this lock will be taken also from interrupt context, so irq
1923 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 *
1925 * Function returns a pointer to the initialized request queue, or NULL if
1926 * it didn't succeed.
1927 *
1928 * Note:
1929 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1930 * when the block device is deactivated (such as at module unload).
1931 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001932
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1934{
Christoph Lameter19460892005-06-23 00:08:19 -07001935 return blk_init_queue_node(rfn, lock, -1);
1936}
1937EXPORT_SYMBOL(blk_init_queue);
1938
1939request_queue_t *
1940blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1941{
1942 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 if (!q)
1945 return NULL;
1946
Christoph Lameter19460892005-06-23 00:08:19 -07001947 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001948 if (blk_init_free_list(q)) {
1949 kmem_cache_free(requestq_cachep, q);
1950 return NULL;
1951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
152587d2005-04-12 16:22:06 -05001953 /*
1954 * if caller didn't supply a lock, they get per-queue locking with
1955 * our embedded lock
1956 */
1957 if (!lock) {
1958 spin_lock_init(&q->__queue_lock);
1959 lock = &q->__queue_lock;
1960 }
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 q->request_fn = rfn;
1963 q->back_merge_fn = ll_back_merge_fn;
1964 q->front_merge_fn = ll_front_merge_fn;
1965 q->merge_requests_fn = ll_merge_requests_fn;
1966 q->prep_rq_fn = NULL;
1967 q->unplug_fn = generic_unplug_device;
1968 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1969 q->queue_lock = lock;
1970
1971 blk_queue_segment_boundary(q, 0xffffffff);
1972
1973 blk_queue_make_request(q, __make_request);
1974 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1975
1976 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1977 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1978
1979 /*
1980 * all done
1981 */
1982 if (!elevator_init(q, NULL)) {
1983 blk_queue_congestion_threshold(q);
1984 return q;
1985 }
1986
Al Viro8669aaf2006-03-18 13:50:00 -05001987 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 return NULL;
1989}
Christoph Lameter19460892005-06-23 00:08:19 -07001990EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992int blk_get_queue(request_queue_t *q)
1993{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001994 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001995 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 return 0;
1997 }
1998
1999 return 1;
2000}
2001
2002EXPORT_SYMBOL(blk_get_queue);
2003
2004static inline void blk_free_request(request_queue_t *q, struct request *rq)
2005{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002006 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02002007 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 mempool_free(rq, q->rq.rq_pool);
2009}
2010
Jens Axboe1ea25ec2006-07-18 22:24:11 +02002011static struct request *
Jens Axboecb78b282006-07-28 09:32:57 +02002012blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013{
2014 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
2015
2016 if (!rq)
2017 return NULL;
2018
2019 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02002020 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 * see bio.h and blkdev.h
2022 */
Jens Axboe49171e52006-08-10 08:59:11 +02002023 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
Tejun Heocb98fc82005-10-28 08:29:39 +02002025 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02002026 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02002027 mempool_free(rq, q->rq.rq_pool);
2028 return NULL;
2029 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02002030 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
Tejun Heocb98fc82005-10-28 08:29:39 +02002033 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
2036/*
2037 * ioc_batching returns true if the ioc is a valid batching request and
2038 * should be given priority access to a request.
2039 */
2040static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
2041{
2042 if (!ioc)
2043 return 0;
2044
2045 /*
2046 * Make sure the process is able to allocate at least 1 request
2047 * even if the batch times out, otherwise we could theoretically
2048 * lose wakeups.
2049 */
2050 return ioc->nr_batch_requests == q->nr_batching ||
2051 (ioc->nr_batch_requests > 0
2052 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2053}
2054
2055/*
2056 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2057 * will cause the process to be a "batcher" on all queues in the system. This
2058 * is the behaviour we want though - once it gets a wakeup it should be given
2059 * a nice run.
2060 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07002061static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
2063 if (!ioc || ioc_batching(q, ioc))
2064 return;
2065
2066 ioc->nr_batch_requests = q->nr_batching;
2067 ioc->last_waited = jiffies;
2068}
2069
2070static void __freed_request(request_queue_t *q, int rw)
2071{
2072 struct request_list *rl = &q->rq;
2073
2074 if (rl->count[rw] < queue_congestion_off_threshold(q))
2075 clear_queue_congested(q, rw);
2076
2077 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (waitqueue_active(&rl->wait[rw]))
2079 wake_up(&rl->wait[rw]);
2080
2081 blk_clear_queue_full(q, rw);
2082 }
2083}
2084
2085/*
2086 * A request has just been released. Account for it, update the full and
2087 * congestion status, wake up any waiters. Called under q->queue_lock.
2088 */
Tejun Heocb98fc82005-10-28 08:29:39 +02002089static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090{
2091 struct request_list *rl = &q->rq;
2092
2093 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002094 if (priv)
2095 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
2097 __freed_request(q, rw);
2098
2099 if (unlikely(rl->starved[rw ^ 1]))
2100 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101}
2102
2103#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2104/*
Nick Piggind6344532005-06-28 20:45:14 -07002105 * Get a free request, queue_lock must be held.
2106 * Returns NULL on failure, with queue_lock held.
2107 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002109static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04002110 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
2112 struct request *rq = NULL;
2113 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002114 struct io_context *ioc = NULL;
2115 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Jens Axboecb78b282006-07-28 09:32:57 +02002117 may_queue = elv_may_queue(q, rw);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002118 if (may_queue == ELV_MQUEUE_NO)
2119 goto rq_starved;
2120
2121 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2122 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002123 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002124 /*
2125 * The queue will fill after this allocation, so set
2126 * it as full, and mark this process as "batching".
2127 * This process will be allowed to complete a batch of
2128 * requests, others will be blocked.
2129 */
2130 if (!blk_queue_full(q, rw)) {
2131 ioc_set_batching(q, ioc);
2132 blk_set_queue_full(q, rw);
2133 } else {
2134 if (may_queue != ELV_MQUEUE_MUST
2135 && !ioc_batching(q, ioc)) {
2136 /*
2137 * The queue is full and the allocating
2138 * process is not a "batcher", and not
2139 * exempted by the IO scheduler
2140 */
2141 goto out;
2142 }
2143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 }
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002145 set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 }
2147
Jens Axboe082cf692005-06-28 16:35:11 +02002148 /*
2149 * Only allow batching queuers to allocate up to 50% over the defined
2150 * limit of requests, otherwise we could have thousands of requests
2151 * allocated with any setting of ->nr_requests
2152 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002153 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002154 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002155
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 rl->count[rw]++;
2157 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002158
Jens Axboe64521d12005-10-28 08:30:39 +02002159 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002160 if (priv)
2161 rl->elvpriv++;
2162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 spin_unlock_irq(q->queue_lock);
2164
Jens Axboecb78b282006-07-28 09:32:57 +02002165 rq = blk_alloc_request(q, rw, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002166 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 /*
2168 * Allocation failed presumably due to memory. Undo anything
2169 * we might have messed up.
2170 *
2171 * Allocating task should really be put onto the front of the
2172 * wait queue, but this is pretty rare.
2173 */
2174 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002175 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
2177 /*
2178 * in the very unlikely event that allocation failed and no
2179 * requests for this direction was pending, mark us starved
2180 * so that freeing of a request in the other direction will
2181 * notice us. another possible fix would be to split the
2182 * rq mempool into READ and WRITE
2183 */
2184rq_starved:
2185 if (unlikely(rl->count[rw] == 0))
2186 rl->starved[rw] = 1;
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 goto out;
2189 }
2190
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002191 /*
2192 * ioc may be NULL here, and ioc_batching will be false. That's
2193 * OK, if the queue is under the request limit then requests need
2194 * not count toward the nr_batch_requests limit. There will always
2195 * be some limit enforced by BLK_BATCH_TIME.
2196 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 if (ioc_batching(q, ioc))
2198 ioc->nr_batch_requests--;
2199
2200 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002201
2202 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return rq;
2205}
2206
2207/*
2208 * No available requests for this queue, unplug the device and wait for some
2209 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002210 *
2211 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002213static struct request *get_request_wait(request_queue_t *q, int rw,
2214 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 struct request *rq;
2217
Nick Piggin450991b2005-06-28 20:45:13 -07002218 rq = get_request(q, rw, bio, GFP_NOIO);
2219 while (!rq) {
2220 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 struct request_list *rl = &q->rq;
2222
2223 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2224 TASK_UNINTERRUPTIBLE);
2225
Jens Axboe22e2c502005-06-27 10:55:12 +02002226 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
2228 if (!rq) {
2229 struct io_context *ioc;
2230
Jens Axboe2056a782006-03-23 20:00:26 +01002231 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2232
Nick Piggind6344532005-06-28 20:45:14 -07002233 __generic_unplug_device(q);
2234 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 io_schedule();
2236
2237 /*
2238 * After sleeping, we become a "batching" process and
2239 * will be able to allocate at least one request, and
2240 * up to a big batch of them for a small period time.
2241 * See ioc_batching, ioc_set_batching
2242 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002243 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002245
2246 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 }
2248 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
2251 return rq;
2252}
2253
Al Viro8267e262005-10-21 03:20:53 -04002254struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
2256 struct request *rq;
2257
2258 BUG_ON(rw != READ && rw != WRITE);
2259
Nick Piggind6344532005-06-28 20:45:14 -07002260 spin_lock_irq(q->queue_lock);
2261 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002262 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002263 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002264 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002265 if (!rq)
2266 spin_unlock_irq(q->queue_lock);
2267 }
2268 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
2270 return rq;
2271}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272EXPORT_SYMBOL(blk_get_request);
2273
2274/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002275 * blk_start_queueing - initiate dispatch of requests to device
2276 * @q: request queue to kick into gear
2277 *
2278 * This is basically a helper to remove the need to know whether a queue
2279 * is plugged or not if someone just wants to initiate dispatch of requests
2280 * for this queue.
2281 *
2282 * The queue lock must be held with interrupts disabled.
2283 */
2284void blk_start_queueing(request_queue_t *q)
2285{
2286 if (!blk_queue_plugged(q))
2287 q->request_fn(q);
2288 else
2289 __generic_unplug_device(q);
2290}
2291EXPORT_SYMBOL(blk_start_queueing);
2292
2293/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 * blk_requeue_request - put a request back on queue
2295 * @q: request queue where request should be inserted
2296 * @rq: request to be inserted
2297 *
2298 * Description:
2299 * Drivers often keep queueing requests until the hardware cannot accept
2300 * more, when that condition happens we need to put the request back
2301 * on the queue. Must be called with queue lock held.
2302 */
2303void blk_requeue_request(request_queue_t *q, struct request *rq)
2304{
Jens Axboe2056a782006-03-23 20:00:26 +01002305 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2306
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 if (blk_rq_tagged(rq))
2308 blk_queue_end_tag(q, rq);
2309
2310 elv_requeue_request(q, rq);
2311}
2312
2313EXPORT_SYMBOL(blk_requeue_request);
2314
2315/**
2316 * blk_insert_request - insert a special request in to a request queue
2317 * @q: request queue where request should be inserted
2318 * @rq: request to be inserted
2319 * @at_head: insert request at head or tail of queue
2320 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 *
2322 * Description:
2323 * Many block devices need to execute commands asynchronously, so they don't
2324 * block the whole kernel from preemption during request execution. This is
2325 * accomplished normally by inserting aritficial requests tagged as
2326 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2327 * scheduled for actual execution by the request queue.
2328 *
2329 * We have the option of inserting the head or the tail of the queue.
2330 * Typically we use the tail for new ioctls and so forth. We use the head
2331 * of the queue for things like a QUEUE_FULL message from a device, or a
2332 * host that is unable to accept a particular command.
2333 */
2334void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002335 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336{
Tejun Heo 867d1192005-04-24 02:06:05 -05002337 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 unsigned long flags;
2339
2340 /*
2341 * tell I/O scheduler that this isn't a regular read/write (ie it
2342 * must not attempt merges on this) and that it acts as a soft
2343 * barrier
2344 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002345 rq->cmd_type = REQ_TYPE_SPECIAL;
2346 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
2348 rq->special = data;
2349
2350 spin_lock_irqsave(q->queue_lock, flags);
2351
2352 /*
2353 * If command is tagged, release the tag
2354 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002355 if (blk_rq_tagged(rq))
2356 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
Tejun Heo 867d1192005-04-24 02:06:05 -05002358 drive_stat_acct(rq, rq->nr_sectors, 1);
2359 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002360 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 spin_unlock_irqrestore(q->queue_lock, flags);
2362}
2363
2364EXPORT_SYMBOL(blk_insert_request);
2365
2366/**
2367 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2368 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002369 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 * @ubuf: the user buffer
2371 * @len: length of user data
2372 *
2373 * Description:
2374 * Data will be mapped directly for zero copy io, if possible. Otherwise
2375 * a kernel bounce buffer is used.
2376 *
2377 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2378 * still in process context.
2379 *
2380 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2381 * before being submitted to the device, as pages mapped may be out of
2382 * reach. It's the callers responsibility to make sure this happens. The
2383 * original bio must be passed back in to blk_rq_unmap_user() for proper
2384 * unmapping.
2385 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002386int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2387 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388{
2389 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002391 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Mike Christiedefd94b2005-12-05 02:37:06 -06002393 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002394 return -EINVAL;
2395 if (!len || !ubuf)
2396 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397
Jens Axboedd1cab92005-06-20 14:06:01 +02002398 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
2400 /*
2401 * if alignment requirement is satisfied, map in user pages for
2402 * direct dma. else, set up kernel bounce buffers
2403 */
2404 uaddr = (unsigned long) ubuf;
2405 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002406 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002408 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
2410 if (!IS_ERR(bio)) {
2411 rq->bio = rq->biotail = bio;
2412 blk_rq_bio_prep(q, rq, bio);
2413
2414 rq->buffer = rq->data = NULL;
2415 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002416 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 }
2418
2419 /*
2420 * bio is the err-ptr
2421 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002422 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
2424
2425EXPORT_SYMBOL(blk_rq_map_user);
2426
2427/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002428 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2429 * @q: request queue where request should be inserted
2430 * @rq: request to map data to
2431 * @iov: pointer to the iovec
2432 * @iov_count: number of elements in the iovec
2433 *
2434 * Description:
2435 * Data will be mapped directly for zero copy io, if possible. Otherwise
2436 * a kernel bounce buffer is used.
2437 *
2438 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2439 * still in process context.
2440 *
2441 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2442 * before being submitted to the device, as pages mapped may be out of
2443 * reach. It's the callers responsibility to make sure this happens. The
2444 * original bio must be passed back in to blk_rq_unmap_user() for proper
2445 * unmapping.
2446 */
2447int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2448 struct sg_iovec *iov, int iov_count)
2449{
2450 struct bio *bio;
2451
2452 if (!iov || iov_count <= 0)
2453 return -EINVAL;
2454
2455 /* we don't allow misaligned data like bio_map_user() does. If the
2456 * user is using sg, they're expected to know the alignment constraints
2457 * and respect them accordingly */
2458 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2459 if (IS_ERR(bio))
2460 return PTR_ERR(bio);
2461
2462 rq->bio = rq->biotail = bio;
2463 blk_rq_bio_prep(q, rq, bio);
2464 rq->buffer = rq->data = NULL;
2465 rq->data_len = bio->bi_size;
2466 return 0;
2467}
2468
2469EXPORT_SYMBOL(blk_rq_map_user_iov);
2470
2471/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002473 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 * @ulen: length of user buffer
2475 *
2476 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002477 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002479int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
2481 int ret = 0;
2482
2483 if (bio) {
2484 if (bio_flagged(bio, BIO_USER_MAPPED))
2485 bio_unmap_user(bio);
2486 else
2487 ret = bio_uncopy_user(bio);
2488 }
2489
Jens Axboedd1cab92005-06-20 14:06:01 +02002490 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491}
2492
2493EXPORT_SYMBOL(blk_rq_unmap_user);
2494
2495/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002496 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2497 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002498 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002499 * @kbuf: the kernel buffer
2500 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002501 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002502 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002503int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002504 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002505{
Mike Christie df46b9a2005-06-20 14:04:44 +02002506 struct bio *bio;
2507
Mike Christiedefd94b2005-12-05 02:37:06 -06002508 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002509 return -EINVAL;
2510 if (!len || !kbuf)
2511 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002512
2513 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002514 if (IS_ERR(bio))
2515 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002516
Jens Axboedd1cab92005-06-20 14:06:01 +02002517 if (rq_data_dir(rq) == WRITE)
2518 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002519
Jens Axboedd1cab92005-06-20 14:06:01 +02002520 rq->bio = rq->biotail = bio;
2521 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002522
Jens Axboedd1cab92005-06-20 14:06:01 +02002523 rq->buffer = rq->data = NULL;
2524 rq->data_len = len;
2525 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002526}
2527
2528EXPORT_SYMBOL(blk_rq_map_kern);
2529
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002530/**
2531 * blk_execute_rq_nowait - insert a request into queue for execution
2532 * @q: queue to insert the request in
2533 * @bd_disk: matching gendisk
2534 * @rq: request to insert
2535 * @at_head: insert request at head or tail of queue
2536 * @done: I/O completion handler
2537 *
2538 * Description:
2539 * Insert a fully prepared request at the back of the io scheduler queue
2540 * for execution. Don't wait for completion.
2541 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002542void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2543 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002544 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002545{
2546 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2547
2548 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002549 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002550 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002551 WARN_ON(irqs_disabled());
2552 spin_lock_irq(q->queue_lock);
2553 __elv_add_request(q, rq, where, 1);
2554 __generic_unplug_device(q);
2555 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002556}
Mike Christie6e39b692005-11-11 05:30:24 -06002557EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559/**
2560 * blk_execute_rq - insert a request into queue for execution
2561 * @q: queue to insert the request in
2562 * @bd_disk: matching gendisk
2563 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002564 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 *
2566 * Description:
2567 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002568 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 */
2570int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002571 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002573 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 char sense[SCSI_SENSE_BUFFERSIZE];
2575 int err = 0;
2576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 /*
2578 * we need an extra reference to the request, so we can look at
2579 * it after io completion
2580 */
2581 rq->ref_count++;
2582
2583 if (!rq->sense) {
2584 memset(sense, 0, sizeof(sense));
2585 rq->sense = sense;
2586 rq->sense_len = 0;
2587 }
2588
Jens Axboec00895a2006-09-30 20:29:12 +02002589 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002590 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
2593 if (rq->errors)
2594 err = -EIO;
2595
2596 return err;
2597}
2598
2599EXPORT_SYMBOL(blk_execute_rq);
2600
2601/**
2602 * blkdev_issue_flush - queue a flush
2603 * @bdev: blockdev to issue flush for
2604 * @error_sector: error sector
2605 *
2606 * Description:
2607 * Issue a flush for the block device in question. Caller can supply
2608 * room for storing the error offset in case of a flush error, if they
2609 * wish to. Caller must run wait_for_completion() on its own.
2610 */
2611int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2612{
2613 request_queue_t *q;
2614
2615 if (bdev->bd_disk == NULL)
2616 return -ENXIO;
2617
2618 q = bdev_get_queue(bdev);
2619 if (!q)
2620 return -ENXIO;
2621 if (!q->issue_flush_fn)
2622 return -EOPNOTSUPP;
2623
2624 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2625}
2626
2627EXPORT_SYMBOL(blkdev_issue_flush);
2628
Adrian Bunk93d17d32005-06-25 14:59:10 -07002629static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630{
2631 int rw = rq_data_dir(rq);
2632
2633 if (!blk_fs_request(rq) || !rq->rq_disk)
2634 return;
2635
Jens Axboed72d9042005-11-01 08:35:42 +01002636 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002637 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002638 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 disk_round_stats(rq->rq_disk);
2640 rq->rq_disk->in_flight++;
2641 }
2642}
2643
2644/*
2645 * add-request adds a request to the linked list.
2646 * queue lock is held and interrupts disabled, as we muck with the
2647 * request queue list.
2648 */
2649static inline void add_request(request_queue_t * q, struct request * req)
2650{
2651 drive_stat_acct(req, req->nr_sectors, 1);
2652
2653 if (q->activity_fn)
2654 q->activity_fn(q->activity_data, rq_data_dir(req));
2655
2656 /*
2657 * elevator indicated where it wants this request to be
2658 * inserted at elevator_merge time
2659 */
2660 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2661}
2662
2663/*
2664 * disk_round_stats() - Round off the performance stats on a struct
2665 * disk_stats.
2666 *
2667 * The average IO queue length and utilisation statistics are maintained
2668 * by observing the current state of the queue length and the amount of
2669 * time it has been in this state for.
2670 *
2671 * Normally, that accounting is done on IO completion, but that can result
2672 * in more than a second's worth of IO being accounted for within any one
2673 * second, leading to >100% utilisation. To deal with that, we call this
2674 * function to do a round-off before returning the results when reading
2675 * /proc/diskstats. This accounts immediately for all queue usage up to
2676 * the current jiffies and restarts the counters again.
2677 */
2678void disk_round_stats(struct gendisk *disk)
2679{
2680 unsigned long now = jiffies;
2681
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002682 if (now == disk->stamp)
2683 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002685 if (disk->in_flight) {
2686 __disk_stat_add(disk, time_in_queue,
2687 disk->in_flight * (now - disk->stamp));
2688 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691}
2692
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002693EXPORT_SYMBOL_GPL(disk_round_stats);
2694
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695/*
2696 * queue lock must be held
2697 */
Mike Christie6e39b692005-11-11 05:30:24 -06002698void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 if (unlikely(!q))
2701 return;
2702 if (unlikely(--req->ref_count))
2703 return;
2704
Tejun Heo8922e162005-10-20 16:23:44 +02002705 elv_completed_request(q, req);
2706
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 /*
2708 * Request may not have originated from ll_rw_blk. if not,
2709 * it didn't come out of our reserved rq pools
2710 */
Jens Axboe49171e52006-08-10 08:59:11 +02002711 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002713 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002716 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
2718 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002719 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 }
2721}
2722
Mike Christie6e39b692005-11-11 05:30:24 -06002723EXPORT_SYMBOL_GPL(__blk_put_request);
2724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725void blk_put_request(struct request *req)
2726{
Tejun Heo8922e162005-10-20 16:23:44 +02002727 unsigned long flags;
2728 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
Tejun Heo8922e162005-10-20 16:23:44 +02002730 /*
2731 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2732 * following if (q) test.
2733 */
2734 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 spin_lock_irqsave(q->queue_lock, flags);
2736 __blk_put_request(q, req);
2737 spin_unlock_irqrestore(q->queue_lock, flags);
2738 }
2739}
2740
2741EXPORT_SYMBOL(blk_put_request);
2742
2743/**
2744 * blk_end_sync_rq - executes a completion event on a request
2745 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002746 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002748void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749{
Jens Axboec00895a2006-09-30 20:29:12 +02002750 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
Jens Axboec00895a2006-09-30 20:29:12 +02002752 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 __blk_put_request(rq->q, rq);
2754
2755 /*
2756 * complete last, if this is a stack request the process (and thus
2757 * the rq pointer) could be invalid right after this complete()
2758 */
2759 complete(waiting);
2760}
2761EXPORT_SYMBOL(blk_end_sync_rq);
2762
2763/**
2764 * blk_congestion_wait - wait for a queue to become uncongested
2765 * @rw: READ or WRITE
2766 * @timeout: timeout in jiffies
2767 *
2768 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2769 * If no queues are congested then just wait for the next request to be
2770 * returned.
2771 */
2772long blk_congestion_wait(int rw, long timeout)
2773{
2774 long ret;
2775 DEFINE_WAIT(wait);
2776 wait_queue_head_t *wqh = &congestion_wqh[rw];
2777
2778 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2779 ret = io_schedule_timeout(timeout);
2780 finish_wait(wqh, &wait);
2781 return ret;
2782}
2783
2784EXPORT_SYMBOL(blk_congestion_wait);
2785
Trond Myklebust275a0822006-08-22 20:06:24 -04002786/**
2787 * blk_congestion_end - wake up sleepers on a congestion queue
2788 * @rw: READ or WRITE
2789 */
2790void blk_congestion_end(int rw)
2791{
2792 wait_queue_head_t *wqh = &congestion_wqh[rw];
2793
2794 if (waitqueue_active(wqh))
2795 wake_up(wqh);
2796}
2797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798/*
2799 * Has to be called with the request spinlock acquired
2800 */
2801static int attempt_merge(request_queue_t *q, struct request *req,
2802 struct request *next)
2803{
2804 if (!rq_mergeable(req) || !rq_mergeable(next))
2805 return 0;
2806
2807 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002808 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 */
2810 if (req->sector + req->nr_sectors != next->sector)
2811 return 0;
2812
2813 if (rq_data_dir(req) != rq_data_dir(next)
2814 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002815 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 return 0;
2817
2818 /*
2819 * If we are allowed to merge, then append bio list
2820 * from next to rq and release next. merge_requests_fn
2821 * will have updated segment counts, update sector
2822 * counts here.
2823 */
2824 if (!q->merge_requests_fn(q, req, next))
2825 return 0;
2826
2827 /*
2828 * At this point we have either done a back merge
2829 * or front merge. We need the smaller start_time of
2830 * the merged requests to be the current request
2831 * for accounting purposes.
2832 */
2833 if (time_after(req->start_time, next->start_time))
2834 req->start_time = next->start_time;
2835
2836 req->biotail->bi_next = next->bio;
2837 req->biotail = next->biotail;
2838
2839 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2840
2841 elv_merge_requests(q, req, next);
2842
2843 if (req->rq_disk) {
2844 disk_round_stats(req->rq_disk);
2845 req->rq_disk->in_flight--;
2846 }
2847
Jens Axboe22e2c502005-06-27 10:55:12 +02002848 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2849
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 __blk_put_request(q, next);
2851 return 1;
2852}
2853
2854static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2855{
2856 struct request *next = elv_latter_request(q, rq);
2857
2858 if (next)
2859 return attempt_merge(q, rq, next);
2860
2861 return 0;
2862}
2863
2864static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2865{
2866 struct request *prev = elv_former_request(q, rq);
2867
2868 if (prev)
2869 return attempt_merge(q, prev, rq);
2870
2871 return 0;
2872}
2873
Tejun Heo52d9e672006-01-06 09:49:58 +01002874static void init_request_from_bio(struct request *req, struct bio *bio)
2875{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002876 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002877
2878 /*
2879 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2880 */
2881 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002882 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002883
2884 /*
2885 * REQ_BARRIER implies no merging, but lets make it explicit
2886 */
2887 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002888 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002889
Jens Axboeb31dc662006-06-13 08:26:10 +02002890 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002891 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002892 if (bio_rw_meta(bio))
2893 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002894
Tejun Heo52d9e672006-01-06 09:49:58 +01002895 req->errors = 0;
2896 req->hard_sector = req->sector = bio->bi_sector;
2897 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2898 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2899 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2900 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2901 req->buffer = bio_data(bio); /* see ->buffer comment above */
Tejun Heo52d9e672006-01-06 09:49:58 +01002902 req->bio = req->biotail = bio;
2903 req->ioprio = bio_prio(bio);
2904 req->rq_disk = bio->bi_bdev->bd_disk;
2905 req->start_time = jiffies;
2906}
2907
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908static int __make_request(request_queue_t *q, struct bio *bio)
2909{
Nick Piggin450991b2005-06-28 20:45:13 -07002910 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002911 int el_ret, nr_sectors, barrier, err;
2912 const unsigned short prio = bio_prio(bio);
2913 const int sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
2917 /*
2918 * low level driver can indicate that it wants pages above a
2919 * certain limit bounced to low memory (ie for highmem, or even
2920 * ISA dma in theory)
2921 */
2922 blk_queue_bounce(q, &bio);
2923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002925 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 err = -EOPNOTSUPP;
2927 goto end_io;
2928 }
2929
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 spin_lock_irq(q->queue_lock);
2931
Nick Piggin450991b2005-06-28 20:45:13 -07002932 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 goto get_rq;
2934
2935 el_ret = elv_merge(q, &req, bio);
2936 switch (el_ret) {
2937 case ELEVATOR_BACK_MERGE:
2938 BUG_ON(!rq_mergeable(req));
2939
2940 if (!q->back_merge_fn(q, req, bio))
2941 break;
2942
Jens Axboe2056a782006-03-23 20:00:26 +01002943 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2944
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 req->biotail->bi_next = bio;
2946 req->biotail = bio;
2947 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002948 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 drive_stat_acct(req, nr_sectors, 0);
2950 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002951 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 goto out;
2953
2954 case ELEVATOR_FRONT_MERGE:
2955 BUG_ON(!rq_mergeable(req));
2956
2957 if (!q->front_merge_fn(q, req, bio))
2958 break;
2959
Jens Axboe2056a782006-03-23 20:00:26 +01002960 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2961
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 bio->bi_next = req->bio;
2963 req->bio = bio;
2964
2965 /*
2966 * may not be valid. if the low level driver said
2967 * it didn't need a bounce buffer then it better
2968 * not touch req->buffer either...
2969 */
2970 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02002971 req->current_nr_sectors = bio_cur_sectors(bio);
2972 req->hard_cur_sectors = req->current_nr_sectors;
2973 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002975 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 drive_stat_acct(req, nr_sectors, 0);
2977 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002978 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 goto out;
2980
Nick Piggin450991b2005-06-28 20:45:13 -07002981 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002983 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 }
2985
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002987 /*
2988 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002989 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002990 */
Jens Axboe51da90f2006-07-18 04:14:45 +02002991 req = get_request_wait(q, bio_data_dir(bio), bio);
Nick Piggind6344532005-06-28 20:45:14 -07002992
Nick Piggin450991b2005-06-28 20:45:13 -07002993 /*
2994 * After dropping the lock and possibly sleeping here, our request
2995 * may now be mergeable after it had proven unmergeable (above).
2996 * We don't worry about that case for efficiency. It won't happen
2997 * often, and the elevators are able to handle it.
2998 */
Tejun Heo52d9e672006-01-06 09:49:58 +01002999 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000
Nick Piggin450991b2005-06-28 20:45:13 -07003001 spin_lock_irq(q->queue_lock);
3002 if (elv_queue_empty(q))
3003 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 add_request(q, req);
3005out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003006 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 __generic_unplug_device(q);
3008
3009 spin_unlock_irq(q->queue_lock);
3010 return 0;
3011
3012end_io:
3013 bio_endio(bio, nr_sectors << 9, err);
3014 return 0;
3015}
3016
3017/*
3018 * If bio->bi_dev is a partition, remap the location
3019 */
3020static inline void blk_partition_remap(struct bio *bio)
3021{
3022 struct block_device *bdev = bio->bi_bdev;
3023
3024 if (bdev != bdev->bd_contains) {
3025 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003026 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
Jens Axboea3623572005-11-01 09:26:16 +01003028 p->sectors[rw] += bio_sectors(bio);
3029 p->ios[rw]++;
3030
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 bio->bi_sector += p->start_sect;
3032 bio->bi_bdev = bdev->bd_contains;
3033 }
3034}
3035
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036static void handle_bad_sector(struct bio *bio)
3037{
3038 char b[BDEVNAME_SIZE];
3039
3040 printk(KERN_INFO "attempt to access beyond end of device\n");
3041 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3042 bdevname(bio->bi_bdev, b),
3043 bio->bi_rw,
3044 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3045 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3046
3047 set_bit(BIO_EOF, &bio->bi_flags);
3048}
3049
3050/**
3051 * generic_make_request: hand a buffer to its device driver for I/O
3052 * @bio: The bio describing the location in memory and on the device.
3053 *
3054 * generic_make_request() is used to make I/O requests of block
3055 * devices. It is passed a &struct bio, which describes the I/O that needs
3056 * to be done.
3057 *
3058 * generic_make_request() does not return any status. The
3059 * success/failure status of the request, along with notification of
3060 * completion, is delivered asynchronously through the bio->bi_end_io
3061 * function described (one day) else where.
3062 *
3063 * The caller of generic_make_request must make sure that bi_io_vec
3064 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3065 * set to describe the device address, and the
3066 * bi_end_io and optionally bi_private are set to describe how
3067 * completion notification should be signaled.
3068 *
3069 * generic_make_request and the drivers it calls may use bi_next if this
3070 * bio happens to be merged with someone else, and may change bi_dev and
3071 * bi_sector for remaps as it sees fit. So the values of these fields
3072 * should NOT be depended on after the call to generic_make_request.
3073 */
3074void generic_make_request(struct bio *bio)
3075{
3076 request_queue_t *q;
3077 sector_t maxsector;
3078 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003079 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
3081 might_sleep();
3082 /* Test device or partition size, when known. */
3083 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3084 if (maxsector) {
3085 sector_t sector = bio->bi_sector;
3086
3087 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3088 /*
3089 * This may well happen - the kernel calls bread()
3090 * without checking the size of the device, e.g., when
3091 * mounting a device.
3092 */
3093 handle_bad_sector(bio);
3094 goto end_io;
3095 }
3096 }
3097
3098 /*
3099 * Resolve the mapping until finished. (drivers are
3100 * still free to implement/resolve their own stacking
3101 * by explicitly returning 0)
3102 *
3103 * NOTE: we don't repeat the blk_size check for each new device.
3104 * Stacking drivers are expected to know what they are doing.
3105 */
Jens Axboe2056a782006-03-23 20:00:26 +01003106 maxsector = -1;
3107 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 do {
3109 char b[BDEVNAME_SIZE];
3110
3111 q = bdev_get_queue(bio->bi_bdev);
3112 if (!q) {
3113 printk(KERN_ERR
3114 "generic_make_request: Trying to access "
3115 "nonexistent block-device %s (%Lu)\n",
3116 bdevname(bio->bi_bdev, b),
3117 (long long) bio->bi_sector);
3118end_io:
3119 bio_endio(bio, bio->bi_size, -EIO);
3120 break;
3121 }
3122
3123 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3124 printk("bio too big device %s (%u > %u)\n",
3125 bdevname(bio->bi_bdev, b),
3126 bio_sectors(bio),
3127 q->max_hw_sectors);
3128 goto end_io;
3129 }
3130
Nick Pigginfde6ad22005-06-23 00:08:53 -07003131 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 goto end_io;
3133
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 /*
3135 * If this device has partitions, remap block n
3136 * of partition p to block n+start(p) of the disk.
3137 */
3138 blk_partition_remap(bio);
3139
Jens Axboe2056a782006-03-23 20:00:26 +01003140 if (maxsector != -1)
3141 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
3142 maxsector);
3143
3144 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3145
3146 maxsector = bio->bi_sector;
3147 old_dev = bio->bi_bdev->bd_dev;
3148
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 ret = q->make_request_fn(q, bio);
3150 } while (ret);
3151}
3152
3153EXPORT_SYMBOL(generic_make_request);
3154
3155/**
3156 * submit_bio: submit a bio to the block device layer for I/O
3157 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3158 * @bio: The &struct bio which describes the I/O
3159 *
3160 * submit_bio() is very similar in purpose to generic_make_request(), and
3161 * uses that function to do most of the work. Both are fairly rough
3162 * interfaces, @bio must be presetup and ready for I/O.
3163 *
3164 */
3165void submit_bio(int rw, struct bio *bio)
3166{
3167 int count = bio_sectors(bio);
3168
3169 BIO_BUG_ON(!bio->bi_size);
3170 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003171 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 if (rw & WRITE)
Christoph Lameterf8891e52006-06-30 01:55:45 -07003173 count_vm_events(PGPGOUT, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174 else
Christoph Lameterf8891e52006-06-30 01:55:45 -07003175 count_vm_events(PGPGIN, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176
3177 if (unlikely(block_dump)) {
3178 char b[BDEVNAME_SIZE];
3179 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3180 current->comm, current->pid,
3181 (rw & WRITE) ? "WRITE" : "READ",
3182 (unsigned long long)bio->bi_sector,
3183 bdevname(bio->bi_bdev,b));
3184 }
3185
3186 generic_make_request(bio);
3187}
3188
3189EXPORT_SYMBOL(submit_bio);
3190
Adrian Bunk93d17d32005-06-25 14:59:10 -07003191static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192{
3193 struct bio *bio, *prevbio = NULL;
3194 int nr_phys_segs, nr_hw_segs;
3195 unsigned int phys_size, hw_size;
3196 request_queue_t *q = rq->q;
3197
3198 if (!rq->bio)
3199 return;
3200
3201 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3202 rq_for_each_bio(bio, rq) {
3203 /* Force bio hw/phys segs to be recalculated. */
3204 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3205
3206 nr_phys_segs += bio_phys_segments(q, bio);
3207 nr_hw_segs += bio_hw_segments(q, bio);
3208 if (prevbio) {
3209 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3210 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3211
3212 if (blk_phys_contig_segment(q, prevbio, bio) &&
3213 pseg <= q->max_segment_size) {
3214 nr_phys_segs--;
3215 phys_size += prevbio->bi_size + bio->bi_size;
3216 } else
3217 phys_size = 0;
3218
3219 if (blk_hw_contig_segment(q, prevbio, bio) &&
3220 hseg <= q->max_segment_size) {
3221 nr_hw_segs--;
3222 hw_size += prevbio->bi_size + bio->bi_size;
3223 } else
3224 hw_size = 0;
3225 }
3226 prevbio = bio;
3227 }
3228
3229 rq->nr_phys_segments = nr_phys_segs;
3230 rq->nr_hw_segments = nr_hw_segs;
3231}
3232
Adrian Bunk93d17d32005-06-25 14:59:10 -07003233static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
3235 if (blk_fs_request(rq)) {
3236 rq->hard_sector += nsect;
3237 rq->hard_nr_sectors -= nsect;
3238
3239 /*
3240 * Move the I/O submission pointers ahead if required.
3241 */
3242 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3243 (rq->sector <= rq->hard_sector)) {
3244 rq->sector = rq->hard_sector;
3245 rq->nr_sectors = rq->hard_nr_sectors;
3246 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3247 rq->current_nr_sectors = rq->hard_cur_sectors;
3248 rq->buffer = bio_data(rq->bio);
3249 }
3250
3251 /*
3252 * if total number of sectors is less than the first segment
3253 * size, something has gone terribly wrong
3254 */
3255 if (rq->nr_sectors < rq->current_nr_sectors) {
3256 printk("blk: request botched\n");
3257 rq->nr_sectors = rq->current_nr_sectors;
3258 }
3259 }
3260}
3261
3262static int __end_that_request_first(struct request *req, int uptodate,
3263 int nr_bytes)
3264{
3265 int total_bytes, bio_nbytes, error, next_idx = 0;
3266 struct bio *bio;
3267
Jens Axboe2056a782006-03-23 20:00:26 +01003268 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3269
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 /*
3271 * extend uptodate bool to allow < 0 value to be direct io error
3272 */
3273 error = 0;
3274 if (end_io_error(uptodate))
3275 error = !uptodate ? -EIO : uptodate;
3276
3277 /*
3278 * for a REQ_BLOCK_PC request, we want to carry any eventual
3279 * sense key with us all the way through
3280 */
3281 if (!blk_pc_request(req))
3282 req->errors = 0;
3283
3284 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003285 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 printk("end_request: I/O error, dev %s, sector %llu\n",
3287 req->rq_disk ? req->rq_disk->disk_name : "?",
3288 (unsigned long long)req->sector);
3289 }
3290
Jens Axboed72d9042005-11-01 08:35:42 +01003291 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003292 const int rw = rq_data_dir(req);
3293
Jens Axboe53e86062006-01-17 11:09:27 +01003294 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003295 }
3296
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 total_bytes = bio_nbytes = 0;
3298 while ((bio = req->bio) != NULL) {
3299 int nbytes;
3300
3301 if (nr_bytes >= bio->bi_size) {
3302 req->bio = bio->bi_next;
3303 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003304 if (!ordered_bio_endio(req, bio, nbytes, error))
3305 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 next_idx = 0;
3307 bio_nbytes = 0;
3308 } else {
3309 int idx = bio->bi_idx + next_idx;
3310
3311 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3312 blk_dump_rq_flags(req, "__end_that");
3313 printk("%s: bio idx %d >= vcnt %d\n",
3314 __FUNCTION__,
3315 bio->bi_idx, bio->bi_vcnt);
3316 break;
3317 }
3318
3319 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3320 BIO_BUG_ON(nbytes > bio->bi_size);
3321
3322 /*
3323 * not a complete bvec done
3324 */
3325 if (unlikely(nbytes > nr_bytes)) {
3326 bio_nbytes += nr_bytes;
3327 total_bytes += nr_bytes;
3328 break;
3329 }
3330
3331 /*
3332 * advance to the next vector
3333 */
3334 next_idx++;
3335 bio_nbytes += nbytes;
3336 }
3337
3338 total_bytes += nbytes;
3339 nr_bytes -= nbytes;
3340
3341 if ((bio = req->bio)) {
3342 /*
3343 * end more in this run, or just return 'not-done'
3344 */
3345 if (unlikely(nr_bytes <= 0))
3346 break;
3347 }
3348 }
3349
3350 /*
3351 * completely done
3352 */
3353 if (!req->bio)
3354 return 0;
3355
3356 /*
3357 * if the request wasn't completed, update state
3358 */
3359 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003360 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3361 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 bio->bi_idx += next_idx;
3363 bio_iovec(bio)->bv_offset += nr_bytes;
3364 bio_iovec(bio)->bv_len -= nr_bytes;
3365 }
3366
3367 blk_recalc_rq_sectors(req, total_bytes >> 9);
3368 blk_recalc_rq_segments(req);
3369 return 1;
3370}
3371
3372/**
3373 * end_that_request_first - end I/O on a request
3374 * @req: the request being processed
3375 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3376 * @nr_sectors: number of sectors to end I/O on
3377 *
3378 * Description:
3379 * Ends I/O on a number of sectors attached to @req, and sets it up
3380 * for the next range of segments (if any) in the cluster.
3381 *
3382 * Return:
3383 * 0 - we are done with this request, call end_that_request_last()
3384 * 1 - still buffers pending for this request
3385 **/
3386int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3387{
3388 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3389}
3390
3391EXPORT_SYMBOL(end_that_request_first);
3392
3393/**
3394 * end_that_request_chunk - end I/O on a request
3395 * @req: the request being processed
3396 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3397 * @nr_bytes: number of bytes to complete
3398 *
3399 * Description:
3400 * Ends I/O on a number of bytes attached to @req, and sets it up
3401 * for the next range of segments (if any). Like end_that_request_first(),
3402 * but deals with bytes instead of sectors.
3403 *
3404 * Return:
3405 * 0 - we are done with this request, call end_that_request_last()
3406 * 1 - still buffers pending for this request
3407 **/
3408int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3409{
3410 return __end_that_request_first(req, uptodate, nr_bytes);
3411}
3412
3413EXPORT_SYMBOL(end_that_request_chunk);
3414
3415/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003416 * splice the completion data to a local structure and hand off to
3417 * process_completion_queue() to complete the requests
3418 */
3419static void blk_done_softirq(struct softirq_action *h)
3420{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003421 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003422
3423 local_irq_disable();
3424 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003425 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003426 local_irq_enable();
3427
3428 while (!list_empty(&local_list)) {
3429 struct request *rq = list_entry(local_list.next, struct request, donelist);
3430
3431 list_del_init(&rq->donelist);
3432 rq->q->softirq_done_fn(rq);
3433 }
3434}
3435
3436#ifdef CONFIG_HOTPLUG_CPU
3437
3438static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3439 void *hcpu)
3440{
3441 /*
3442 * If a CPU goes away, splice its entries to the current CPU
3443 * and trigger a run of the softirq
3444 */
3445 if (action == CPU_DEAD) {
3446 int cpu = (unsigned long) hcpu;
3447
3448 local_irq_disable();
3449 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3450 &__get_cpu_var(blk_cpu_done));
3451 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3452 local_irq_enable();
3453 }
3454
3455 return NOTIFY_OK;
3456}
3457
3458
Chandra Seetharaman054cc8a2006-06-27 02:54:07 -07003459static struct notifier_block __devinitdata blk_cpu_notifier = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003460 .notifier_call = blk_cpu_notify,
3461};
3462
3463#endif /* CONFIG_HOTPLUG_CPU */
3464
3465/**
3466 * blk_complete_request - end I/O on a request
3467 * @req: the request being processed
3468 *
3469 * Description:
3470 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003471 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003472 * through requeueing. Theh actual completion happens out-of-order,
3473 * through a softirq handler. The user must have registered a completion
3474 * callback through blk_queue_softirq_done().
3475 **/
3476
3477void blk_complete_request(struct request *req)
3478{
3479 struct list_head *cpu_list;
3480 unsigned long flags;
3481
3482 BUG_ON(!req->q->softirq_done_fn);
3483
3484 local_irq_save(flags);
3485
3486 cpu_list = &__get_cpu_var(blk_cpu_done);
3487 list_add_tail(&req->donelist, cpu_list);
3488 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3489
3490 local_irq_restore(flags);
3491}
3492
3493EXPORT_SYMBOL(blk_complete_request);
3494
3495/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 * queue lock must be held
3497 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003498void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499{
3500 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003501 int error;
3502
3503 /*
3504 * extend uptodate bool to allow < 0 value to be direct io error
3505 */
3506 error = 0;
3507 if (end_io_error(uptodate))
3508 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509
3510 if (unlikely(laptop_mode) && blk_fs_request(req))
3511 laptop_io_completion();
3512
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003513 /*
3514 * Account IO completion. bar_rq isn't accounted as a normal
3515 * IO on queueing nor completion. Accounting the containing
3516 * request is enough.
3517 */
3518 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003520 const int rw = rq_data_dir(req);
3521
3522 __disk_stat_inc(disk, ios[rw]);
3523 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 disk_round_stats(disk);
3525 disk->in_flight--;
3526 }
3527 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003528 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 else
3530 __blk_put_request(req->q, req);
3531}
3532
3533EXPORT_SYMBOL(end_that_request_last);
3534
3535void end_request(struct request *req, int uptodate)
3536{
3537 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3538 add_disk_randomness(req->rq_disk);
3539 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003540 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 }
3542}
3543
3544EXPORT_SYMBOL(end_request);
3545
3546void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3547{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003548 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3549 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550
3551 rq->nr_phys_segments = bio_phys_segments(q, bio);
3552 rq->nr_hw_segments = bio_hw_segments(q, bio);
3553 rq->current_nr_sectors = bio_cur_sectors(bio);
3554 rq->hard_cur_sectors = rq->current_nr_sectors;
3555 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3556 rq->buffer = bio_data(bio);
3557
3558 rq->bio = rq->biotail = bio;
3559}
3560
3561EXPORT_SYMBOL(blk_rq_bio_prep);
3562
3563int kblockd_schedule_work(struct work_struct *work)
3564{
3565 return queue_work(kblockd_workqueue, work);
3566}
3567
3568EXPORT_SYMBOL(kblockd_schedule_work);
3569
3570void kblockd_flush(void)
3571{
3572 flush_workqueue(kblockd_workqueue);
3573}
3574EXPORT_SYMBOL(kblockd_flush);
3575
3576int __init blk_dev_init(void)
3577{
Jens Axboeff856ba2006-01-09 16:02:34 +01003578 int i;
3579
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 kblockd_workqueue = create_workqueue("kblockd");
3581 if (!kblockd_workqueue)
3582 panic("Failed to create kblockd\n");
3583
3584 request_cachep = kmem_cache_create("blkdev_requests",
3585 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3586
3587 requestq_cachep = kmem_cache_create("blkdev_queue",
3588 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3589
3590 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3591 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3592
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003593 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003594 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3595
3596 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003597 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003598
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 blk_max_low_pfn = max_low_pfn;
3600 blk_max_pfn = max_pfn;
3601
3602 return 0;
3603}
3604
3605/*
3606 * IO Context helper functions
3607 */
3608void put_io_context(struct io_context *ioc)
3609{
3610 if (ioc == NULL)
3611 return;
3612
3613 BUG_ON(atomic_read(&ioc->refcount) == 0);
3614
3615 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003616 struct cfq_io_context *cic;
3617
Al Viro334e94d2006-03-18 15:05:53 -05003618 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 if (ioc->aic && ioc->aic->dtor)
3620 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003621 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003622 struct rb_node *n = rb_first(&ioc->cic_root);
3623
3624 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003625 cic->dtor(ioc);
3626 }
Al Viro334e94d2006-03-18 15:05:53 -05003627 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628
3629 kmem_cache_free(iocontext_cachep, ioc);
3630 }
3631}
3632EXPORT_SYMBOL(put_io_context);
3633
3634/* Called by the exitting task */
3635void exit_io_context(void)
3636{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003638 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639
Jens Axboe22e2c502005-06-27 10:55:12 +02003640 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 ioc = current->io_context;
3642 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003643 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644
Oleg Nesterov25034d72006-08-29 09:15:14 +02003645 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 if (ioc->aic && ioc->aic->exit)
3647 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003648 if (ioc->cic_root.rb_node != NULL) {
3649 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3650 cic->exit(ioc);
3651 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003652
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 put_io_context(ioc);
3654}
3655
3656/*
3657 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003658 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003660 * This returned IO context doesn't have a specifically elevated refcount,
3661 * but since the current task itself holds a reference, the context can be
3662 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003664static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665{
3666 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 struct io_context *ret;
3668
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003670 if (likely(ret))
3671 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672
Jens Axboeb5deef92006-07-19 23:39:40 +02003673 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 if (ret) {
3675 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003676 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003677 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 ret->last_waited = jiffies; /* doesn't matter... */
3679 ret->nr_batch_requests = 0; /* because this is 0 */
3680 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003681 ret->cic_root.rb_node = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003682 /* make sure set_task_ioprio() sees the settings above */
3683 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003684 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 }
3686
3687 return ret;
3688}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003689EXPORT_SYMBOL(current_io_context);
3690
3691/*
3692 * If the current task has no IO context then create one and initialise it.
3693 * If it does have a context, take a ref on it.
3694 *
3695 * This is always called in the context of the task which submitted the I/O.
3696 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003697struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003698{
3699 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003700 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003701 if (likely(ret))
3702 atomic_inc(&ret->refcount);
3703 return ret;
3704}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705EXPORT_SYMBOL(get_io_context);
3706
3707void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3708{
3709 struct io_context *src = *psrc;
3710 struct io_context *dst = *pdst;
3711
3712 if (src) {
3713 BUG_ON(atomic_read(&src->refcount) == 0);
3714 atomic_inc(&src->refcount);
3715 put_io_context(dst);
3716 *pdst = src;
3717 }
3718}
3719EXPORT_SYMBOL(copy_io_context);
3720
3721void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3722{
3723 struct io_context *temp;
3724 temp = *ioc1;
3725 *ioc1 = *ioc2;
3726 *ioc2 = temp;
3727}
3728EXPORT_SYMBOL(swap_io_context);
3729
3730/*
3731 * sysfs parts below
3732 */
3733struct queue_sysfs_entry {
3734 struct attribute attr;
3735 ssize_t (*show)(struct request_queue *, char *);
3736 ssize_t (*store)(struct request_queue *, const char *, size_t);
3737};
3738
3739static ssize_t
3740queue_var_show(unsigned int var, char *page)
3741{
3742 return sprintf(page, "%d\n", var);
3743}
3744
3745static ssize_t
3746queue_var_store(unsigned long *var, const char *page, size_t count)
3747{
3748 char *p = (char *) page;
3749
3750 *var = simple_strtoul(p, &p, 10);
3751 return count;
3752}
3753
3754static ssize_t queue_requests_show(struct request_queue *q, char *page)
3755{
3756 return queue_var_show(q->nr_requests, (page));
3757}
3758
3759static ssize_t
3760queue_requests_store(struct request_queue *q, const char *page, size_t count)
3761{
3762 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003763 unsigned long nr;
3764 int ret = queue_var_store(&nr, page, count);
3765 if (nr < BLKDEV_MIN_RQ)
3766 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767
Al Viroc981ff92006-03-18 13:51:29 -05003768 spin_lock_irq(q->queue_lock);
3769 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 blk_queue_congestion_threshold(q);
3771
3772 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3773 set_queue_congested(q, READ);
3774 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3775 clear_queue_congested(q, READ);
3776
3777 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3778 set_queue_congested(q, WRITE);
3779 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3780 clear_queue_congested(q, WRITE);
3781
3782 if (rl->count[READ] >= q->nr_requests) {
3783 blk_set_queue_full(q, READ);
3784 } else if (rl->count[READ]+1 <= q->nr_requests) {
3785 blk_clear_queue_full(q, READ);
3786 wake_up(&rl->wait[READ]);
3787 }
3788
3789 if (rl->count[WRITE] >= q->nr_requests) {
3790 blk_set_queue_full(q, WRITE);
3791 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3792 blk_clear_queue_full(q, WRITE);
3793 wake_up(&rl->wait[WRITE]);
3794 }
Al Viroc981ff92006-03-18 13:51:29 -05003795 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796 return ret;
3797}
3798
3799static ssize_t queue_ra_show(struct request_queue *q, char *page)
3800{
3801 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3802
3803 return queue_var_show(ra_kb, (page));
3804}
3805
3806static ssize_t
3807queue_ra_store(struct request_queue *q, const char *page, size_t count)
3808{
3809 unsigned long ra_kb;
3810 ssize_t ret = queue_var_store(&ra_kb, page, count);
3811
3812 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3814 spin_unlock_irq(q->queue_lock);
3815
3816 return ret;
3817}
3818
3819static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3820{
3821 int max_sectors_kb = q->max_sectors >> 1;
3822
3823 return queue_var_show(max_sectors_kb, (page));
3824}
3825
3826static ssize_t
3827queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3828{
3829 unsigned long max_sectors_kb,
3830 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3831 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3832 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3833 int ra_kb;
3834
3835 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3836 return -EINVAL;
3837 /*
3838 * Take the queue lock to update the readahead and max_sectors
3839 * values synchronously:
3840 */
3841 spin_lock_irq(q->queue_lock);
3842 /*
3843 * Trim readahead window as well, if necessary:
3844 */
3845 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3846 if (ra_kb > max_sectors_kb)
3847 q->backing_dev_info.ra_pages =
3848 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3849
3850 q->max_sectors = max_sectors_kb << 1;
3851 spin_unlock_irq(q->queue_lock);
3852
3853 return ret;
3854}
3855
3856static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3857{
3858 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3859
3860 return queue_var_show(max_hw_sectors_kb, (page));
3861}
3862
3863
3864static struct queue_sysfs_entry queue_requests_entry = {
3865 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3866 .show = queue_requests_show,
3867 .store = queue_requests_store,
3868};
3869
3870static struct queue_sysfs_entry queue_ra_entry = {
3871 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3872 .show = queue_ra_show,
3873 .store = queue_ra_store,
3874};
3875
3876static struct queue_sysfs_entry queue_max_sectors_entry = {
3877 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3878 .show = queue_max_sectors_show,
3879 .store = queue_max_sectors_store,
3880};
3881
3882static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3883 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3884 .show = queue_max_hw_sectors_show,
3885};
3886
3887static struct queue_sysfs_entry queue_iosched_entry = {
3888 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3889 .show = elv_iosched_show,
3890 .store = elv_iosched_store,
3891};
3892
3893static struct attribute *default_attrs[] = {
3894 &queue_requests_entry.attr,
3895 &queue_ra_entry.attr,
3896 &queue_max_hw_sectors_entry.attr,
3897 &queue_max_sectors_entry.attr,
3898 &queue_iosched_entry.attr,
3899 NULL,
3900};
3901
3902#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3903
3904static ssize_t
3905queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3906{
3907 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003908 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3909 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003912 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003913 mutex_lock(&q->sysfs_lock);
3914 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3915 mutex_unlock(&q->sysfs_lock);
3916 return -ENOENT;
3917 }
3918 res = entry->show(q, page);
3919 mutex_unlock(&q->sysfs_lock);
3920 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921}
3922
3923static ssize_t
3924queue_attr_store(struct kobject *kobj, struct attribute *attr,
3925 const char *page, size_t length)
3926{
3927 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003928 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003929
Al Viro483f4af2006-03-18 18:34:37 -05003930 ssize_t res;
3931
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003933 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003934 mutex_lock(&q->sysfs_lock);
3935 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3936 mutex_unlock(&q->sysfs_lock);
3937 return -ENOENT;
3938 }
3939 res = entry->store(q, page, length);
3940 mutex_unlock(&q->sysfs_lock);
3941 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942}
3943
3944static struct sysfs_ops queue_sysfs_ops = {
3945 .show = queue_attr_show,
3946 .store = queue_attr_store,
3947};
3948
Adrian Bunk93d17d32005-06-25 14:59:10 -07003949static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 .sysfs_ops = &queue_sysfs_ops,
3951 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05003952 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003953};
3954
3955int blk_register_queue(struct gendisk *disk)
3956{
3957 int ret;
3958
3959 request_queue_t *q = disk->queue;
3960
3961 if (!q || !q->request_fn)
3962 return -ENXIO;
3963
3964 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965
Al Viro483f4af2006-03-18 18:34:37 -05003966 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967 if (ret < 0)
3968 return ret;
3969
Al Viro483f4af2006-03-18 18:34:37 -05003970 kobject_uevent(&q->kobj, KOBJ_ADD);
3971
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 ret = elv_register_queue(q);
3973 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05003974 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3975 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 return ret;
3977 }
3978
3979 return 0;
3980}
3981
3982void blk_unregister_queue(struct gendisk *disk)
3983{
3984 request_queue_t *q = disk->queue;
3985
3986 if (q && q->request_fn) {
3987 elv_unregister_queue(q);
3988
Al Viro483f4af2006-03-18 18:34:37 -05003989 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3990 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 kobject_put(&disk->kobj);
3992 }
3993}