blob: 9b91bb70c5ed3c6d876e58eb3a7597c12023e859 [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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * For the allocated request tables
45 */
46static kmem_cache_t *request_cachep;
47
48/*
49 * For queue allocation
50 */
51static kmem_cache_t *requestq_cachep;
52
53/*
54 * For io context allocations
55 */
56static kmem_cache_t *iocontext_cachep;
57
58static wait_queue_head_t congestion_wqh[2] = {
59 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
61 };
62
63/*
64 * Controlling structure to kblockd
65 */
Jens Axboeff856ba2006-01-09 16:02:34 +010066static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68unsigned long blk_max_low_pfn, blk_max_pfn;
69
70EXPORT_SYMBOL(blk_max_low_pfn);
71EXPORT_SYMBOL(blk_max_pfn);
72
Jens Axboeff856ba2006-01-09 16:02:34 +010073static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/* Amount of time in which a process may batch requests */
76#define BLK_BATCH_TIME (HZ/50UL)
77
78/* Number of requests a "batching" process may submit */
79#define BLK_BATCH_REQ 32
80
81/*
82 * Return the threshold (number of used requests) at which the queue is
83 * considered to be congested. It include a little hysteresis to keep the
84 * context switch rate down.
85 */
86static inline int queue_congestion_on_threshold(struct request_queue *q)
87{
88 return q->nr_congestion_on;
89}
90
91/*
92 * The threshold at which a queue is considered to be uncongested
93 */
94static inline int queue_congestion_off_threshold(struct request_queue *q)
95{
96 return q->nr_congestion_off;
97}
98
99static void blk_queue_congestion_threshold(struct request_queue *q)
100{
101 int nr;
102
103 nr = q->nr_requests - (q->nr_requests / 8) + 1;
104 if (nr > q->nr_requests)
105 nr = q->nr_requests;
106 q->nr_congestion_on = nr;
107
108 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
109 if (nr < 1)
110 nr = 1;
111 q->nr_congestion_off = nr;
112}
113
114/*
115 * A queue has just exitted congestion. Note this in the global counter of
116 * congested queues, and wake up anyone who was waiting for requests to be
117 * put back.
118 */
119static void clear_queue_congested(request_queue_t *q, int rw)
120{
121 enum bdi_state bit;
122 wait_queue_head_t *wqh = &congestion_wqh[rw];
123
124 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
125 clear_bit(bit, &q->backing_dev_info.state);
126 smp_mb__after_clear_bit();
127 if (waitqueue_active(wqh))
128 wake_up(wqh);
129}
130
131/*
132 * A queue has just entered congestion. Flag that in the queue's VM-visible
133 * state flags and increment the global gounter of congested queues.
134 */
135static void set_queue_congested(request_queue_t *q, int rw)
136{
137 enum bdi_state bit;
138
139 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
140 set_bit(bit, &q->backing_dev_info.state);
141}
142
143/**
144 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
145 * @bdev: device
146 *
147 * Locates the passed device's request queue and returns the address of its
148 * backing_dev_info
149 *
150 * Will return NULL if the request queue cannot be located.
151 */
152struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
153{
154 struct backing_dev_info *ret = NULL;
155 request_queue_t *q = bdev_get_queue(bdev);
156
157 if (q)
158 ret = &q->backing_dev_info;
159 return ret;
160}
161
162EXPORT_SYMBOL(blk_get_backing_dev_info);
163
164void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
165{
166 q->activity_fn = fn;
167 q->activity_data = data;
168}
169
170EXPORT_SYMBOL(blk_queue_activity_fn);
171
172/**
173 * blk_queue_prep_rq - set a prepare_request function for queue
174 * @q: queue
175 * @pfn: prepare_request function
176 *
177 * It's possible for a queue to register a prepare_request callback which
178 * is invoked before the request is handed to the request_fn. The goal of
179 * the function is to prepare a request for I/O, it can be used to build a
180 * cdb from the request data for instance.
181 *
182 */
183void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
184{
185 q->prep_rq_fn = pfn;
186}
187
188EXPORT_SYMBOL(blk_queue_prep_rq);
189
190/**
191 * blk_queue_merge_bvec - set a merge_bvec function for queue
192 * @q: queue
193 * @mbfn: merge_bvec_fn
194 *
195 * Usually queues have static limitations on the max sectors or segments that
196 * we can put in a request. Stacking drivers may have some settings that
197 * are dynamic, and thus we have to query the queue whether it is ok to
198 * add a new bio_vec to a bio at a given offset or not. If the block device
199 * has such limitations, it needs to register a merge_bvec_fn to control
200 * the size of bio's sent to it. Note that a block device *must* allow a
201 * single page to be added to an empty bio. The block device driver may want
202 * to use the bio_split() function to deal with these bio's. By default
203 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
204 * honored.
205 */
206void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
207{
208 q->merge_bvec_fn = mbfn;
209}
210
211EXPORT_SYMBOL(blk_queue_merge_bvec);
212
Jens Axboeff856ba2006-01-09 16:02:34 +0100213void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
214{
215 q->softirq_done_fn = fn;
216}
217
218EXPORT_SYMBOL(blk_queue_softirq_done);
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220/**
221 * blk_queue_make_request - define an alternate make_request function for a device
222 * @q: the request queue for the device to be affected
223 * @mfn: the alternate make_request function
224 *
225 * Description:
226 * The normal way for &struct bios to be passed to a device
227 * driver is for them to be collected into requests on a request
228 * queue, and then to allow the device driver to select requests
229 * off that queue when it is ready. This works well for many block
230 * devices. However some block devices (typically virtual devices
231 * such as md or lvm) do not benefit from the processing on the
232 * request queue, and are served best by having the requests passed
233 * directly to them. This can be achieved by providing a function
234 * to blk_queue_make_request().
235 *
236 * Caveat:
237 * The driver that does this *must* be able to deal appropriately
238 * with buffers in "highmemory". This can be accomplished by either calling
239 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
240 * blk_queue_bounce() to create a buffer in normal memory.
241 **/
242void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
243{
244 /*
245 * set defaults
246 */
247 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700248 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
249 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 q->make_request_fn = mfn;
251 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
252 q->backing_dev_info.state = 0;
253 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600254 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 blk_queue_hardsect_size(q, 512);
256 blk_queue_dma_alignment(q, 511);
257 blk_queue_congestion_threshold(q);
258 q->nr_batching = BLK_BATCH_REQ;
259
260 q->unplug_thresh = 4; /* hmm */
261 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
262 if (q->unplug_delay == 0)
263 q->unplug_delay = 1;
264
265 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
266
267 q->unplug_timer.function = blk_unplug_timeout;
268 q->unplug_timer.data = (unsigned long)q;
269
270 /*
271 * by default assume old behaviour and bounce for any highmem page
272 */
273 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
274
275 blk_queue_activity_fn(q, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278EXPORT_SYMBOL(blk_queue_make_request);
279
280static inline void rq_init(request_queue_t *q, struct request *rq)
281{
282 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100283 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 rq->errors = 0;
286 rq->rq_status = RQ_ACTIVE;
287 rq->bio = rq->biotail = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +0200288 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 rq->buffer = NULL;
290 rq->ref_count = 1;
291 rq->q = q;
292 rq->waiting = NULL;
293 rq->special = NULL;
294 rq->data_len = 0;
295 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200296 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 rq->sense = NULL;
298 rq->end_io = NULL;
299 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100300 rq->completion_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
303/**
304 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100305 * @q: the request queue
306 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100307 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 *
309 * Description:
310 * For journalled file systems, doing ordered writes on a commit
311 * block instead of explicitly doing wait_on_buffer (which is bad
312 * for performance) can be a big win. Block drivers supporting this
313 * feature should call this function and indicate so.
314 *
315 **/
Tejun Heo797e7db2006-01-06 09:51:03 +0100316int blk_queue_ordered(request_queue_t *q, unsigned ordered,
317 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Tejun Heo797e7db2006-01-06 09:51:03 +0100319 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
320 prepare_flush_fn == NULL) {
321 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
322 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100324
325 if (ordered != QUEUE_ORDERED_NONE &&
326 ordered != QUEUE_ORDERED_DRAIN &&
327 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
328 ordered != QUEUE_ORDERED_DRAIN_FUA &&
329 ordered != QUEUE_ORDERED_TAG &&
330 ordered != QUEUE_ORDERED_TAG_FLUSH &&
331 ordered != QUEUE_ORDERED_TAG_FUA) {
332 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
333 return -EINVAL;
334 }
335
Tetsuo Takata60481b12006-01-24 10:34:36 +0100336 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100337 q->next_ordered = ordered;
338 q->prepare_flush_fn = prepare_flush_fn;
339
340 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
343EXPORT_SYMBOL(blk_queue_ordered);
344
345/**
346 * blk_queue_issue_flush_fn - set function for issuing a flush
347 * @q: the request queue
348 * @iff: the function to be called issuing the flush
349 *
350 * Description:
351 * If a driver supports issuing a flush command, the support is notified
352 * to the block layer by defining it through this call.
353 *
354 **/
355void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
356{
357 q->issue_flush_fn = iff;
358}
359
360EXPORT_SYMBOL(blk_queue_issue_flush_fn);
361
362/*
363 * Cache flushing for ordered writes handling
364 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100365inline unsigned blk_ordered_cur_seq(request_queue_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366{
Tejun Heo797e7db2006-01-06 09:51:03 +0100367 if (!q->ordseq)
368 return 0;
369 return 1 << ffz(q->ordseq);
370}
371
372unsigned blk_ordered_req_seq(struct request *rq)
373{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 request_queue_t *q = rq->q;
375
Tejun Heo797e7db2006-01-06 09:51:03 +0100376 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200377
Tejun Heo797e7db2006-01-06 09:51:03 +0100378 if (rq == &q->pre_flush_rq)
379 return QUEUE_ORDSEQ_PREFLUSH;
380 if (rq == &q->bar_rq)
381 return QUEUE_ORDSEQ_BAR;
382 if (rq == &q->post_flush_rq)
383 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Jens Axboe4aff5e22006-08-10 08:44:47 +0200385 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
386 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100387 return QUEUE_ORDSEQ_DRAIN;
388 else
389 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
Tejun Heo797e7db2006-01-06 09:51:03 +0100392void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Tejun Heo797e7db2006-01-06 09:51:03 +0100394 struct request *rq;
395 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Tejun Heo797e7db2006-01-06 09:51:03 +0100397 if (error && !q->orderr)
398 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200399
Tejun Heo797e7db2006-01-06 09:51:03 +0100400 BUG_ON(q->ordseq & seq);
401 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Tejun Heo797e7db2006-01-06 09:51:03 +0100403 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
404 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
406 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100407 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100409 rq = q->orig_bar_rq;
410 uptodate = q->orderr ? q->orderr : 1;
411
412 q->ordseq = 0;
413
414 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
415 end_that_request_last(rq, uptodate);
416}
417
418static void pre_flush_end_io(struct request *rq, int error)
419{
420 elv_completed_request(rq->q, rq);
421 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
422}
423
424static void bar_end_io(struct request *rq, int error)
425{
426 elv_completed_request(rq->q, rq);
427 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
428}
429
430static void post_flush_end_io(struct request *rq, int error)
431{
432 elv_completed_request(rq->q, rq);
433 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
434}
435
436static void queue_flush(request_queue_t *q, unsigned which)
437{
438 struct request *rq;
439 rq_end_io_fn *end_io;
440
441 if (which == QUEUE_ORDERED_PREFLUSH) {
442 rq = &q->pre_flush_rq;
443 end_io = pre_flush_end_io;
444 } else {
445 rq = &q->post_flush_rq;
446 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 }
448
Jens Axboe4aff5e22006-08-10 08:44:47 +0200449 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100450 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100451 rq->elevator_private = NULL;
452 rq->rq_disk = q->bar_rq.rq_disk;
453 rq->rl = NULL;
454 rq->end_io = end_io;
455 q->prepare_flush_fn(q, rq);
456
Tejun Heo30e96562006-02-08 01:01:31 -0800457 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100458}
459
460static inline struct request *start_ordered(request_queue_t *q,
461 struct request *rq)
462{
463 q->bi_size = 0;
464 q->orderr = 0;
465 q->ordered = q->next_ordered;
466 q->ordseq |= QUEUE_ORDSEQ_STARTED;
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100469 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100471 blkdev_dequeue_request(rq);
472 q->orig_bar_rq = rq;
473 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200474 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100475 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200476 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
477 rq->cmd_flags |= REQ_RW;
478 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100479 rq->elevator_private = NULL;
480 rq->rl = NULL;
481 init_request_from_bio(rq, q->orig_bar_rq->bio);
482 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Tejun Heo797e7db2006-01-06 09:51:03 +0100484 /*
485 * Queue ordered sequence. As we stack them at the head, we
486 * need to queue in reverse order. Note that we rely on that
487 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
488 * request gets inbetween ordered sequence.
489 */
490 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
491 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
492 else
493 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Tejun Heo30e96562006-02-08 01:01:31 -0800495 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100496
497 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
498 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
499 rq = &q->pre_flush_rq;
500 } else
501 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
502
503 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
504 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
505 else
506 rq = NULL;
507
508 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509}
510
Tejun Heo797e7db2006-01-06 09:51:03 +0100511int blk_do_ordered(request_queue_t *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800513 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100514 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Tejun Heo797e7db2006-01-06 09:51:03 +0100516 if (!q->ordseq) {
517 if (!is_barrier)
518 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Tejun Heo797e7db2006-01-06 09:51:03 +0100520 if (q->next_ordered != QUEUE_ORDERED_NONE) {
521 *rqp = start_ordered(q, rq);
522 return 1;
523 } else {
524 /*
525 * This can happen when the queue switches to
526 * ORDERED_NONE while this request is on it.
527 */
528 blkdev_dequeue_request(rq);
529 end_that_request_first(rq, -EOPNOTSUPP,
530 rq->hard_nr_sectors);
531 end_that_request_last(rq, -EOPNOTSUPP);
532 *rqp = NULL;
533 return 0;
534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800537 /*
538 * Ordered sequence in progress
539 */
540
541 /* Special requests are not subject to ordering rules. */
542 if (!blk_fs_request(rq) &&
543 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
544 return 1;
545
Tejun Heo797e7db2006-01-06 09:51:03 +0100546 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800547 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100548 if (is_barrier && rq != &q->bar_rq)
549 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800550 } else {
551 /* Ordered by draining. Wait for turn. */
552 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
553 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
554 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 }
556
557 return 1;
558}
559
Tejun Heo797e7db2006-01-06 09:51:03 +0100560static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
Tejun Heo797e7db2006-01-06 09:51:03 +0100562 request_queue_t *q = bio->bi_private;
563 struct bio_vec *bvec;
564 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Tejun Heo797e7db2006-01-06 09:51:03 +0100566 /*
567 * This is dry run, restore bio_sector and size. We'll finish
568 * this request again with the original bi_end_io after an
569 * error occurs or post flush is complete.
570 */
571 q->bi_size += bytes;
572
573 if (bio->bi_size)
574 return 1;
575
576 /* Rewind bvec's */
577 bio->bi_idx = 0;
578 bio_for_each_segment(bvec, bio, i) {
579 bvec->bv_len += bvec->bv_offset;
580 bvec->bv_offset = 0;
581 }
582
583 /* Reset bio */
584 set_bit(BIO_UPTODATE, &bio->bi_flags);
585 bio->bi_size = q->bi_size;
586 bio->bi_sector -= (q->bi_size >> 9);
587 q->bi_size = 0;
588
589 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}
Tejun Heo797e7db2006-01-06 09:51:03 +0100591
592static inline int ordered_bio_endio(struct request *rq, struct bio *bio,
593 unsigned int nbytes, int error)
594{
595 request_queue_t *q = rq->q;
596 bio_end_io_t *endio;
597 void *private;
598
599 if (&q->bar_rq != rq)
600 return 0;
601
602 /*
603 * Okay, this is the barrier request in progress, dry finish it.
604 */
605 if (error && !q->orderr)
606 q->orderr = error;
607
608 endio = bio->bi_end_io;
609 private = bio->bi_private;
610 bio->bi_end_io = flush_dry_bio_endio;
611 bio->bi_private = q;
612
613 bio_endio(bio, nbytes, error);
614
615 bio->bi_end_io = endio;
616 bio->bi_private = private;
617
618 return 1;
619}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621/**
622 * blk_queue_bounce_limit - set bounce buffer limit for queue
623 * @q: the request queue for the device
624 * @dma_addr: bus address limit
625 *
626 * Description:
627 * Different hardware can have different requirements as to what pages
628 * it can do I/O directly to. A low level driver can call
629 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800630 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 **/
632void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
633{
634 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800635 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Andi Kleen5ee1af92006-03-08 17:57:26 -0800637 q->bounce_gfp = GFP_NOIO;
638#if BITS_PER_LONG == 64
639 /* Assume anything <= 4GB can be handled by IOMMU.
640 Actually some IOMMUs can handle everything, but I don't
641 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200642 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800643 dma = 1;
644 q->bounce_pfn = max_low_pfn;
645#else
646 if (bounce_pfn < blk_max_low_pfn)
647 dma = 1;
648 q->bounce_pfn = bounce_pfn;
649#endif
650 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 init_emergency_isa_pool();
652 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800653 q->bounce_pfn = bounce_pfn;
654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655}
656
657EXPORT_SYMBOL(blk_queue_bounce_limit);
658
659/**
660 * blk_queue_max_sectors - set max sectors for a request for this queue
661 * @q: the request queue for the device
662 * @max_sectors: max sectors in the usual 512b unit
663 *
664 * Description:
665 * Enables a low level driver to set an upper limit on the size of
666 * received requests.
667 **/
Jens Axboe2cb2e142006-01-17 09:04:32 +0100668void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
671 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
672 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
673 }
674
Mike Christiedefd94b2005-12-05 02:37:06 -0600675 if (BLK_DEF_MAX_SECTORS > max_sectors)
676 q->max_hw_sectors = q->max_sectors = max_sectors;
677 else {
678 q->max_sectors = BLK_DEF_MAX_SECTORS;
679 q->max_hw_sectors = max_sectors;
680 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
683EXPORT_SYMBOL(blk_queue_max_sectors);
684
685/**
686 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
687 * @q: the request queue for the device
688 * @max_segments: max number of segments
689 *
690 * Description:
691 * Enables a low level driver to set an upper limit on the number of
692 * physical data segments in a request. This would be the largest sized
693 * scatter list the driver could handle.
694 **/
695void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
696{
697 if (!max_segments) {
698 max_segments = 1;
699 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
700 }
701
702 q->max_phys_segments = max_segments;
703}
704
705EXPORT_SYMBOL(blk_queue_max_phys_segments);
706
707/**
708 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
709 * @q: the request queue for the device
710 * @max_segments: max number of segments
711 *
712 * Description:
713 * Enables a low level driver to set an upper limit on the number of
714 * hw data segments in a request. This would be the largest number of
715 * address/length pairs the host adapter can actually give as once
716 * to the device.
717 **/
718void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
719{
720 if (!max_segments) {
721 max_segments = 1;
722 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
723 }
724
725 q->max_hw_segments = max_segments;
726}
727
728EXPORT_SYMBOL(blk_queue_max_hw_segments);
729
730/**
731 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
732 * @q: the request queue for the device
733 * @max_size: max size of segment in bytes
734 *
735 * Description:
736 * Enables a low level driver to set an upper limit on the size of a
737 * coalesced segment
738 **/
739void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
740{
741 if (max_size < PAGE_CACHE_SIZE) {
742 max_size = PAGE_CACHE_SIZE;
743 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
744 }
745
746 q->max_segment_size = max_size;
747}
748
749EXPORT_SYMBOL(blk_queue_max_segment_size);
750
751/**
752 * blk_queue_hardsect_size - set hardware sector size for the queue
753 * @q: the request queue for the device
754 * @size: the hardware sector size, in bytes
755 *
756 * Description:
757 * This should typically be set to the lowest possible sector size
758 * that the hardware can operate on (possible without reverting to
759 * even internal read-modify-write operations). Usually the default
760 * of 512 covers most hardware.
761 **/
762void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
763{
764 q->hardsect_size = size;
765}
766
767EXPORT_SYMBOL(blk_queue_hardsect_size);
768
769/*
770 * Returns the minimum that is _not_ zero, unless both are zero.
771 */
772#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
773
774/**
775 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
776 * @t: the stacking driver (top)
777 * @b: the underlying device (bottom)
778 **/
779void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
780{
781 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600782 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
783 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
785 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
786 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
787 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
788 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800789 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
790 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
793EXPORT_SYMBOL(blk_queue_stack_limits);
794
795/**
796 * blk_queue_segment_boundary - set boundary rules for segment merging
797 * @q: the request queue for the device
798 * @mask: the memory boundary mask
799 **/
800void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
801{
802 if (mask < PAGE_CACHE_SIZE - 1) {
803 mask = PAGE_CACHE_SIZE - 1;
804 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
805 }
806
807 q->seg_boundary_mask = mask;
808}
809
810EXPORT_SYMBOL(blk_queue_segment_boundary);
811
812/**
813 * blk_queue_dma_alignment - set dma length and memory alignment
814 * @q: the request queue for the device
815 * @mask: alignment mask
816 *
817 * description:
818 * set required memory and length aligment for direct dma transactions.
819 * this is used when buiding direct io requests for the queue.
820 *
821 **/
822void blk_queue_dma_alignment(request_queue_t *q, int mask)
823{
824 q->dma_alignment = mask;
825}
826
827EXPORT_SYMBOL(blk_queue_dma_alignment);
828
829/**
830 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 * @q: The request queue for the device
832 * @tag: The tag of the request
833 *
834 * Notes:
835 * Should be used when a device returns a tag and you want to match
836 * it with a request.
837 *
838 * no locks need be held.
839 **/
840struct request *blk_queue_find_tag(request_queue_t *q, int tag)
841{
842 struct blk_queue_tag *bqt = q->queue_tags;
843
Tejun Heoba025082005-08-05 13:28:11 -0700844 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 return NULL;
846
847 return bqt->tag_index[tag];
848}
849
850EXPORT_SYMBOL(blk_queue_find_tag);
851
852/**
James Bottomley492dfb42006-08-30 15:48:45 -0400853 * __blk_free_tags - release a given set of tag maintenance info
854 * @bqt: the tag map to free
855 *
856 * Tries to free the specified @bqt@. Returns true if it was
857 * actually freed and false if there are still references using it
858 */
859static int __blk_free_tags(struct blk_queue_tag *bqt)
860{
861 int retval;
862
863 retval = atomic_dec_and_test(&bqt->refcnt);
864 if (retval) {
865 BUG_ON(bqt->busy);
866 BUG_ON(!list_empty(&bqt->busy_list));
867
868 kfree(bqt->tag_index);
869 bqt->tag_index = NULL;
870
871 kfree(bqt->tag_map);
872 bqt->tag_map = NULL;
873
874 kfree(bqt);
875
876 }
877
878 return retval;
879}
880
881/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 * __blk_queue_free_tags - release tag maintenance info
883 * @q: the request queue for the device
884 *
885 * Notes:
886 * blk_cleanup_queue() will take care of calling this function, if tagging
887 * has been used. So there's no need to call this directly.
888 **/
889static void __blk_queue_free_tags(request_queue_t *q)
890{
891 struct blk_queue_tag *bqt = q->queue_tags;
892
893 if (!bqt)
894 return;
895
James Bottomley492dfb42006-08-30 15:48:45 -0400896 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 q->queue_tags = NULL;
899 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
900}
901
James Bottomley492dfb42006-08-30 15:48:45 -0400902
903/**
904 * blk_free_tags - release a given set of tag maintenance info
905 * @bqt: the tag map to free
906 *
907 * For externally managed @bqt@ frees the map. Callers of this
908 * function must guarantee to have released all the queues that
909 * might have been using this tag map.
910 */
911void blk_free_tags(struct blk_queue_tag *bqt)
912{
913 if (unlikely(!__blk_free_tags(bqt)))
914 BUG();
915}
916EXPORT_SYMBOL(blk_free_tags);
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918/**
919 * blk_queue_free_tags - release tag maintenance info
920 * @q: the request queue for the device
921 *
922 * Notes:
923 * This is used to disabled tagged queuing to a device, yet leave
924 * queue in function.
925 **/
926void blk_queue_free_tags(request_queue_t *q)
927{
928 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
929}
930
931EXPORT_SYMBOL(blk_queue_free_tags);
932
933static int
934init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
935{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 struct request **tag_index;
937 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700938 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
James Bottomley492dfb42006-08-30 15:48:45 -0400940 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 depth = q->nr_requests * 2;
942 printk(KERN_ERR "%s: adjusted depth to %d\n",
943 __FUNCTION__, depth);
944 }
945
Jens Axboef68110f2006-03-08 13:31:44 +0100946 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (!tag_index)
948 goto fail;
949
Tejun Heof7d37d02005-06-23 00:08:50 -0700950 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100951 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (!tag_map)
953 goto fail;
954
Tejun Heoba025082005-08-05 13:28:11 -0700955 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 tags->tag_index = tag_index;
958 tags->tag_map = tag_map;
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 return 0;
961fail:
962 kfree(tag_index);
963 return -ENOMEM;
964}
965
James Bottomley492dfb42006-08-30 15:48:45 -0400966static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
967 int depth)
968{
969 struct blk_queue_tag *tags;
970
971 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
972 if (!tags)
973 goto fail;
974
975 if (init_tag_map(q, tags, depth))
976 goto fail;
977
978 INIT_LIST_HEAD(&tags->busy_list);
979 tags->busy = 0;
980 atomic_set(&tags->refcnt, 1);
981 return tags;
982fail:
983 kfree(tags);
984 return NULL;
985}
986
987/**
988 * blk_init_tags - initialize the tag info for an external tag map
989 * @depth: the maximum queue depth supported
990 * @tags: the tag to use
991 **/
992struct blk_queue_tag *blk_init_tags(int depth)
993{
994 return __blk_queue_init_tags(NULL, depth);
995}
996EXPORT_SYMBOL(blk_init_tags);
997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998/**
999 * blk_queue_init_tags - initialize the queue tag info
1000 * @q: the request queue for the device
1001 * @depth: the maximum queue depth supported
1002 * @tags: the tag to use
1003 **/
1004int blk_queue_init_tags(request_queue_t *q, int depth,
1005 struct blk_queue_tag *tags)
1006{
1007 int rc;
1008
1009 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
1010
1011 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -04001012 tags = __blk_queue_init_tags(q, depth);
1013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (!tags)
1015 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 } else if (q->queue_tags) {
1017 if ((rc = blk_queue_resize_tags(q, depth)))
1018 return rc;
1019 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
1020 return 0;
1021 } else
1022 atomic_inc(&tags->refcnt);
1023
1024 /*
1025 * assign it, all done
1026 */
1027 q->queue_tags = tags;
1028 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
1029 return 0;
1030fail:
1031 kfree(tags);
1032 return -ENOMEM;
1033}
1034
1035EXPORT_SYMBOL(blk_queue_init_tags);
1036
1037/**
1038 * blk_queue_resize_tags - change the queueing depth
1039 * @q: the request queue for the device
1040 * @new_depth: the new max command queueing depth
1041 *
1042 * Notes:
1043 * Must be called with the queue lock held.
1044 **/
1045int blk_queue_resize_tags(request_queue_t *q, int new_depth)
1046{
1047 struct blk_queue_tag *bqt = q->queue_tags;
1048 struct request **tag_index;
1049 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -07001050 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 if (!bqt)
1053 return -ENXIO;
1054
1055 /*
Tejun Heoba025082005-08-05 13:28:11 -07001056 * if we already have large enough real_max_depth. just
1057 * adjust max_depth. *NOTE* as requests with tag value
1058 * between new_depth and real_max_depth can be in-flight, tag
1059 * map can not be shrunk blindly here.
1060 */
1061 if (new_depth <= bqt->real_max_depth) {
1062 bqt->max_depth = new_depth;
1063 return 0;
1064 }
1065
1066 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001067 * Currently cannot replace a shared tag map with a new
1068 * one, so error out if this is the case
1069 */
1070 if (atomic_read(&bqt->refcnt) != 1)
1071 return -EBUSY;
1072
1073 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 * save the old state info, so we can copy it back
1075 */
1076 tag_index = bqt->tag_index;
1077 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001078 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
1080 if (init_tag_map(q, bqt, new_depth))
1081 return -ENOMEM;
1082
1083 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001084 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001085 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 kfree(tag_index);
1088 kfree(tag_map);
1089 return 0;
1090}
1091
1092EXPORT_SYMBOL(blk_queue_resize_tags);
1093
1094/**
1095 * blk_queue_end_tag - end tag operations for a request
1096 * @q: the request queue for the device
1097 * @rq: the request that has completed
1098 *
1099 * Description:
1100 * Typically called when end_that_request_first() returns 0, meaning
1101 * all transfers have been done for a request. It's important to call
1102 * this function before end_that_request_last(), as that will put the
1103 * request back on the free list thus corrupting the internal tag list.
1104 *
1105 * Notes:
1106 * queue lock must be held.
1107 **/
1108void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1109{
1110 struct blk_queue_tag *bqt = q->queue_tags;
1111 int tag = rq->tag;
1112
1113 BUG_ON(tag == -1);
1114
Tejun Heoba025082005-08-05 13:28:11 -07001115 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001116 /*
1117 * This can happen after tag depth has been reduced.
1118 * FIXME: how about a warning or info message here?
1119 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return;
1121
1122 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001123 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1124 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return;
1126 }
1127
1128 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001129 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 rq->tag = -1;
1131
1132 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001133 printk(KERN_ERR "%s: tag %d is missing\n",
1134 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 bqt->tag_index[tag] = NULL;
1137 bqt->busy--;
1138}
1139
1140EXPORT_SYMBOL(blk_queue_end_tag);
1141
1142/**
1143 * blk_queue_start_tag - find a free tag and assign it
1144 * @q: the request queue for the device
1145 * @rq: the block request that needs tagging
1146 *
1147 * Description:
1148 * This can either be used as a stand-alone helper, or possibly be
1149 * assigned as the queue &prep_rq_fn (in which case &struct request
1150 * automagically gets a tag assigned). Note that this function
1151 * assumes that any type of request can be queued! if this is not
1152 * true for your device, you must check the request type before
1153 * calling this function. The request will also be removed from
1154 * the request queue, so it's the drivers responsibility to readd
1155 * it if it should need to be restarted for some reason.
1156 *
1157 * Notes:
1158 * queue lock must be held.
1159 **/
1160int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1161{
1162 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001163 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Jens Axboe4aff5e22006-08-10 08:44:47 +02001165 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001167 "%s: request %p for device [%s] already tagged %d",
1168 __FUNCTION__, rq,
1169 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 BUG();
1171 }
1172
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001173 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1174 if (tag >= bqt->max_depth)
1175 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 __set_bit(tag, bqt->tag_map);
1178
Jens Axboe4aff5e22006-08-10 08:44:47 +02001179 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 rq->tag = tag;
1181 bqt->tag_index[tag] = rq;
1182 blkdev_dequeue_request(rq);
1183 list_add(&rq->queuelist, &bqt->busy_list);
1184 bqt->busy++;
1185 return 0;
1186}
1187
1188EXPORT_SYMBOL(blk_queue_start_tag);
1189
1190/**
1191 * blk_queue_invalidate_tags - invalidate all pending tags
1192 * @q: the request queue for the device
1193 *
1194 * Description:
1195 * Hardware conditions may dictate a need to stop all pending requests.
1196 * In this case, we will safely clear the block side of the tag queue and
1197 * readd all requests to the request queue in the right order.
1198 *
1199 * Notes:
1200 * queue lock must be held.
1201 **/
1202void blk_queue_invalidate_tags(request_queue_t *q)
1203{
1204 struct blk_queue_tag *bqt = q->queue_tags;
1205 struct list_head *tmp, *n;
1206 struct request *rq;
1207
1208 list_for_each_safe(tmp, n, &bqt->busy_list) {
1209 rq = list_entry_rq(tmp);
1210
1211 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001212 printk(KERN_ERR
1213 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001215 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 } else
1217 blk_queue_end_tag(q, rq);
1218
Jens Axboe4aff5e22006-08-10 08:44:47 +02001219 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1221 }
1222}
1223
1224EXPORT_SYMBOL(blk_queue_invalidate_tags);
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226void blk_dump_rq_flags(struct request *rq, char *msg)
1227{
1228 int bit;
1229
Jens Axboe4aff5e22006-08-10 08:44:47 +02001230 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1231 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1232 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1235 rq->nr_sectors,
1236 rq->current_nr_sectors);
1237 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1238
Jens Axboe4aff5e22006-08-10 08:44:47 +02001239 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 printk("cdb: ");
1241 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1242 printk("%02x ", rq->cmd[bit]);
1243 printk("\n");
1244 }
1245}
1246
1247EXPORT_SYMBOL(blk_dump_rq_flags);
1248
1249void blk_recount_segments(request_queue_t *q, struct bio *bio)
1250{
1251 struct bio_vec *bv, *bvprv = NULL;
1252 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1253 int high, highprv = 1;
1254
1255 if (unlikely(!bio->bi_io_vec))
1256 return;
1257
1258 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1259 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1260 bio_for_each_segment(bv, bio, i) {
1261 /*
1262 * the trick here is making sure that a high page is never
1263 * considered part of another segment, since that might
1264 * change with the bounce page.
1265 */
1266 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1267 if (high || highprv)
1268 goto new_hw_segment;
1269 if (cluster) {
1270 if (seg_size + bv->bv_len > q->max_segment_size)
1271 goto new_segment;
1272 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1273 goto new_segment;
1274 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1275 goto new_segment;
1276 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1277 goto new_hw_segment;
1278
1279 seg_size += bv->bv_len;
1280 hw_seg_size += bv->bv_len;
1281 bvprv = bv;
1282 continue;
1283 }
1284new_segment:
1285 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1286 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1287 hw_seg_size += bv->bv_len;
1288 } else {
1289new_hw_segment:
1290 if (hw_seg_size > bio->bi_hw_front_size)
1291 bio->bi_hw_front_size = hw_seg_size;
1292 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1293 nr_hw_segs++;
1294 }
1295
1296 nr_phys_segs++;
1297 bvprv = bv;
1298 seg_size = bv->bv_len;
1299 highprv = high;
1300 }
1301 if (hw_seg_size > bio->bi_hw_back_size)
1302 bio->bi_hw_back_size = hw_seg_size;
1303 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1304 bio->bi_hw_front_size = hw_seg_size;
1305 bio->bi_phys_segments = nr_phys_segs;
1306 bio->bi_hw_segments = nr_hw_segs;
1307 bio->bi_flags |= (1 << BIO_SEG_VALID);
1308}
1309
1310
Adrian Bunk93d17d32005-06-25 14:59:10 -07001311static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 struct bio *nxt)
1313{
1314 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1315 return 0;
1316
1317 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1318 return 0;
1319 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1320 return 0;
1321
1322 /*
1323 * bio and nxt are contigous in memory, check if the queue allows
1324 * these two to be merged into one
1325 */
1326 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1327 return 1;
1328
1329 return 0;
1330}
1331
Adrian Bunk93d17d32005-06-25 14:59:10 -07001332static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 struct bio *nxt)
1334{
1335 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1336 blk_recount_segments(q, bio);
1337 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1338 blk_recount_segments(q, nxt);
1339 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1340 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1341 return 0;
1342 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1343 return 0;
1344
1345 return 1;
1346}
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348/*
1349 * map a request to scatterlist, return number of sg entries setup. Caller
1350 * must make sure sg can hold rq->nr_phys_segments entries
1351 */
1352int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1353{
1354 struct bio_vec *bvec, *bvprv;
1355 struct bio *bio;
1356 int nsegs, i, cluster;
1357
1358 nsegs = 0;
1359 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1360
1361 /*
1362 * for each bio in rq
1363 */
1364 bvprv = NULL;
1365 rq_for_each_bio(bio, rq) {
1366 /*
1367 * for each segment in bio
1368 */
1369 bio_for_each_segment(bvec, bio, i) {
1370 int nbytes = bvec->bv_len;
1371
1372 if (bvprv && cluster) {
1373 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1374 goto new_segment;
1375
1376 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1377 goto new_segment;
1378 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1379 goto new_segment;
1380
1381 sg[nsegs - 1].length += nbytes;
1382 } else {
1383new_segment:
1384 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1385 sg[nsegs].page = bvec->bv_page;
1386 sg[nsegs].length = nbytes;
1387 sg[nsegs].offset = bvec->bv_offset;
1388
1389 nsegs++;
1390 }
1391 bvprv = bvec;
1392 } /* segments in bio */
1393 } /* bios in rq */
1394
1395 return nsegs;
1396}
1397
1398EXPORT_SYMBOL(blk_rq_map_sg);
1399
1400/*
1401 * the standard queue merge functions, can be overridden with device
1402 * specific ones if so desired
1403 */
1404
1405static inline int ll_new_mergeable(request_queue_t *q,
1406 struct request *req,
1407 struct bio *bio)
1408{
1409 int nr_phys_segs = bio_phys_segments(q, bio);
1410
1411 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001412 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (req == q->last_merge)
1414 q->last_merge = NULL;
1415 return 0;
1416 }
1417
1418 /*
1419 * A hw segment is just getting larger, bump just the phys
1420 * counter.
1421 */
1422 req->nr_phys_segments += nr_phys_segs;
1423 return 1;
1424}
1425
1426static inline int ll_new_hw_segment(request_queue_t *q,
1427 struct request *req,
1428 struct bio *bio)
1429{
1430 int nr_hw_segs = bio_hw_segments(q, bio);
1431 int nr_phys_segs = bio_phys_segments(q, bio);
1432
1433 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1434 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001435 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (req == q->last_merge)
1437 q->last_merge = NULL;
1438 return 0;
1439 }
1440
1441 /*
1442 * This will form the start of a new hw segment. Bump both
1443 * counters.
1444 */
1445 req->nr_hw_segments += nr_hw_segs;
1446 req->nr_phys_segments += nr_phys_segs;
1447 return 1;
1448}
1449
1450static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1451 struct bio *bio)
1452{
Mike Christiedefd94b2005-12-05 02:37:06 -06001453 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 int len;
1455
Mike Christiedefd94b2005-12-05 02:37:06 -06001456 if (unlikely(blk_pc_request(req)))
1457 max_sectors = q->max_hw_sectors;
1458 else
1459 max_sectors = q->max_sectors;
1460
1461 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001462 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 if (req == q->last_merge)
1464 q->last_merge = NULL;
1465 return 0;
1466 }
1467 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1468 blk_recount_segments(q, req->biotail);
1469 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1470 blk_recount_segments(q, bio);
1471 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1472 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1473 !BIOVEC_VIRT_OVERSIZE(len)) {
1474 int mergeable = ll_new_mergeable(q, req, bio);
1475
1476 if (mergeable) {
1477 if (req->nr_hw_segments == 1)
1478 req->bio->bi_hw_front_size = len;
1479 if (bio->bi_hw_segments == 1)
1480 bio->bi_hw_back_size = len;
1481 }
1482 return mergeable;
1483 }
1484
1485 return ll_new_hw_segment(q, req, bio);
1486}
1487
1488static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1489 struct bio *bio)
1490{
Mike Christiedefd94b2005-12-05 02:37:06 -06001491 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 int len;
1493
Mike Christiedefd94b2005-12-05 02:37:06 -06001494 if (unlikely(blk_pc_request(req)))
1495 max_sectors = q->max_hw_sectors;
1496 else
1497 max_sectors = q->max_sectors;
1498
1499
1500 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001501 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (req == q->last_merge)
1503 q->last_merge = NULL;
1504 return 0;
1505 }
1506 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1507 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1508 blk_recount_segments(q, bio);
1509 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1510 blk_recount_segments(q, req->bio);
1511 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1512 !BIOVEC_VIRT_OVERSIZE(len)) {
1513 int mergeable = ll_new_mergeable(q, req, bio);
1514
1515 if (mergeable) {
1516 if (bio->bi_hw_segments == 1)
1517 bio->bi_hw_front_size = len;
1518 if (req->nr_hw_segments == 1)
1519 req->biotail->bi_hw_back_size = len;
1520 }
1521 return mergeable;
1522 }
1523
1524 return ll_new_hw_segment(q, req, bio);
1525}
1526
1527static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1528 struct request *next)
1529{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001530 int total_phys_segments;
1531 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533 /*
1534 * First check if the either of the requests are re-queued
1535 * requests. Can't merge them if they are.
1536 */
1537 if (req->special || next->special)
1538 return 0;
1539
1540 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001541 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 */
1543 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1544 return 0;
1545
1546 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1547 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1548 total_phys_segments--;
1549
1550 if (total_phys_segments > q->max_phys_segments)
1551 return 0;
1552
1553 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1554 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1555 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1556 /*
1557 * propagate the combined length to the end of the requests
1558 */
1559 if (req->nr_hw_segments == 1)
1560 req->bio->bi_hw_front_size = len;
1561 if (next->nr_hw_segments == 1)
1562 next->biotail->bi_hw_back_size = len;
1563 total_hw_segments--;
1564 }
1565
1566 if (total_hw_segments > q->max_hw_segments)
1567 return 0;
1568
1569 /* Merge is OK... */
1570 req->nr_phys_segments = total_phys_segments;
1571 req->nr_hw_segments = total_hw_segments;
1572 return 1;
1573}
1574
1575/*
1576 * "plug" the device if there are no outstanding requests: this will
1577 * force the transfer to start only after we have put all the requests
1578 * on the list.
1579 *
1580 * This is called with interrupts off and no requests on the queue and
1581 * with the queue lock held.
1582 */
1583void blk_plug_device(request_queue_t *q)
1584{
1585 WARN_ON(!irqs_disabled());
1586
1587 /*
1588 * don't plug a stopped queue, it must be paired with blk_start_queue()
1589 * which will restart the queueing
1590 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001591 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return;
1593
Jens Axboe2056a782006-03-23 20:00:26 +01001594 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001596 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598}
1599
1600EXPORT_SYMBOL(blk_plug_device);
1601
1602/*
1603 * remove the queue from the plugged list, if present. called with
1604 * queue lock held and interrupts disabled.
1605 */
1606int blk_remove_plug(request_queue_t *q)
1607{
1608 WARN_ON(!irqs_disabled());
1609
1610 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1611 return 0;
1612
1613 del_timer(&q->unplug_timer);
1614 return 1;
1615}
1616
1617EXPORT_SYMBOL(blk_remove_plug);
1618
1619/*
1620 * remove the plug and let it rip..
1621 */
1622void __generic_unplug_device(request_queue_t *q)
1623{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001624 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 return;
1626
1627 if (!blk_remove_plug(q))
1628 return;
1629
Jens Axboe22e2c502005-06-27 10:55:12 +02001630 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631}
1632EXPORT_SYMBOL(__generic_unplug_device);
1633
1634/**
1635 * generic_unplug_device - fire a request queue
1636 * @q: The &request_queue_t in question
1637 *
1638 * Description:
1639 * Linux uses plugging to build bigger requests queues before letting
1640 * the device have at them. If a queue is plugged, the I/O scheduler
1641 * is still adding and merging requests on the queue. Once the queue
1642 * gets unplugged, the request_fn defined for the queue is invoked and
1643 * transfers started.
1644 **/
1645void generic_unplug_device(request_queue_t *q)
1646{
1647 spin_lock_irq(q->queue_lock);
1648 __generic_unplug_device(q);
1649 spin_unlock_irq(q->queue_lock);
1650}
1651EXPORT_SYMBOL(generic_unplug_device);
1652
1653static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1654 struct page *page)
1655{
1656 request_queue_t *q = bdi->unplug_io_data;
1657
1658 /*
1659 * devices don't necessarily have an ->unplug_fn defined
1660 */
Jens Axboe2056a782006-03-23 20:00:26 +01001661 if (q->unplug_fn) {
1662 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1663 q->rq.count[READ] + q->rq.count[WRITE]);
1664
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667}
1668
1669static void blk_unplug_work(void *data)
1670{
1671 request_queue_t *q = data;
1672
Jens Axboe2056a782006-03-23 20:00:26 +01001673 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1674 q->rq.count[READ] + q->rq.count[WRITE]);
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 q->unplug_fn(q);
1677}
1678
1679static void blk_unplug_timeout(unsigned long data)
1680{
1681 request_queue_t *q = (request_queue_t *)data;
1682
Jens Axboe2056a782006-03-23 20:00:26 +01001683 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1684 q->rq.count[READ] + q->rq.count[WRITE]);
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 kblockd_schedule_work(&q->unplug_work);
1687}
1688
1689/**
1690 * blk_start_queue - restart a previously stopped queue
1691 * @q: The &request_queue_t in question
1692 *
1693 * Description:
1694 * blk_start_queue() will clear the stop flag on the queue, and call
1695 * the request_fn for the queue if it was in a stopped state when
1696 * entered. Also see blk_stop_queue(). Queue lock must be held.
1697 **/
1698void blk_start_queue(request_queue_t *q)
1699{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001700 WARN_ON(!irqs_disabled());
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1703
1704 /*
1705 * one level of recursion is ok and is much faster than kicking
1706 * the unplug handling
1707 */
1708 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1709 q->request_fn(q);
1710 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1711 } else {
1712 blk_plug_device(q);
1713 kblockd_schedule_work(&q->unplug_work);
1714 }
1715}
1716
1717EXPORT_SYMBOL(blk_start_queue);
1718
1719/**
1720 * blk_stop_queue - stop a queue
1721 * @q: The &request_queue_t in question
1722 *
1723 * Description:
1724 * The Linux block layer assumes that a block driver will consume all
1725 * entries on the request queue when the request_fn strategy is called.
1726 * Often this will not happen, because of hardware limitations (queue
1727 * depth settings). If a device driver gets a 'queue full' response,
1728 * or if it simply chooses not to queue more I/O at one point, it can
1729 * call this function to prevent the request_fn from being called until
1730 * the driver has signalled it's ready to go again. This happens by calling
1731 * blk_start_queue() to restart queue operations. Queue lock must be held.
1732 **/
1733void blk_stop_queue(request_queue_t *q)
1734{
1735 blk_remove_plug(q);
1736 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1737}
1738EXPORT_SYMBOL(blk_stop_queue);
1739
1740/**
1741 * blk_sync_queue - cancel any pending callbacks on a queue
1742 * @q: the queue
1743 *
1744 * Description:
1745 * The block layer may perform asynchronous callback activity
1746 * on a queue, such as calling the unplug function after a timeout.
1747 * A block device may call blk_sync_queue to ensure that any
1748 * such activity is cancelled, thus allowing it to release resources
1749 * the the callbacks might use. The caller must already have made sure
1750 * that its ->make_request_fn will not re-add plugging prior to calling
1751 * this function.
1752 *
1753 */
1754void blk_sync_queue(struct request_queue *q)
1755{
1756 del_timer_sync(&q->unplug_timer);
1757 kblockd_flush();
1758}
1759EXPORT_SYMBOL(blk_sync_queue);
1760
1761/**
1762 * blk_run_queue - run a single device queue
1763 * @q: The queue to run
1764 */
1765void blk_run_queue(struct request_queue *q)
1766{
1767 unsigned long flags;
1768
1769 spin_lock_irqsave(q->queue_lock, flags);
1770 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001771
1772 /*
1773 * Only recurse once to avoid overrunning the stack, let the unplug
1774 * handling reinvoke the handler shortly if we already got there.
1775 */
1776 if (!elv_queue_empty(q)) {
1777 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1778 q->request_fn(q);
1779 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1780 } else {
1781 blk_plug_device(q);
1782 kblockd_schedule_work(&q->unplug_work);
1783 }
1784 }
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 spin_unlock_irqrestore(q->queue_lock, flags);
1787}
1788EXPORT_SYMBOL(blk_run_queue);
1789
1790/**
1791 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001792 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 *
1794 * Description:
1795 * blk_cleanup_queue is the pair to blk_init_queue() or
1796 * blk_queue_make_request(). It should be called when a request queue is
1797 * being released; typically when a block device is being de-registered.
1798 * Currently, its primary task it to free all the &struct request
1799 * structures that were allocated to the queue and the queue itself.
1800 *
1801 * Caveat:
1802 * Hopefully the low level driver will have finished any
1803 * outstanding requests first...
1804 **/
Al Viro483f4af2006-03-18 18:34:37 -05001805static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806{
Al Viro483f4af2006-03-18 18:34:37 -05001807 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 struct request_list *rl = &q->rq;
1809
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 blk_sync_queue(q);
1811
1812 if (rl->rq_pool)
1813 mempool_destroy(rl->rq_pool);
1814
1815 if (q->queue_tags)
1816 __blk_queue_free_tags(q);
1817
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001818 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 kmem_cache_free(requestq_cachep, q);
1821}
1822
Al Viro483f4af2006-03-18 18:34:37 -05001823void blk_put_queue(request_queue_t *q)
1824{
1825 kobject_put(&q->kobj);
1826}
1827EXPORT_SYMBOL(blk_put_queue);
1828
1829void blk_cleanup_queue(request_queue_t * q)
1830{
1831 mutex_lock(&q->sysfs_lock);
1832 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1833 mutex_unlock(&q->sysfs_lock);
1834
1835 if (q->elevator)
1836 elevator_exit(q->elevator);
1837
1838 blk_put_queue(q);
1839}
1840
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841EXPORT_SYMBOL(blk_cleanup_queue);
1842
1843static int blk_init_free_list(request_queue_t *q)
1844{
1845 struct request_list *rl = &q->rq;
1846
1847 rl->count[READ] = rl->count[WRITE] = 0;
1848 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001849 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 init_waitqueue_head(&rl->wait[READ]);
1851 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Christoph Lameter19460892005-06-23 00:08:19 -07001853 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1854 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856 if (!rl->rq_pool)
1857 return -ENOMEM;
1858
1859 return 0;
1860}
1861
Al Viro8267e262005-10-21 03:20:53 -04001862request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863{
Christoph Lameter19460892005-06-23 00:08:19 -07001864 return blk_alloc_queue_node(gfp_mask, -1);
1865}
1866EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
Al Viro483f4af2006-03-18 18:34:37 -05001868static struct kobj_type queue_ktype;
1869
Al Viro8267e262005-10-21 03:20:53 -04001870request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001871{
1872 request_queue_t *q;
1873
1874 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (!q)
1876 return NULL;
1877
1878 memset(q, 0, sizeof(*q));
1879 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001880
1881 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1882 q->kobj.ktype = &queue_ktype;
1883 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
1885 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1886 q->backing_dev_info.unplug_io_data = q;
1887
Al Viro483f4af2006-03-18 18:34:37 -05001888 mutex_init(&q->sysfs_lock);
1889
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 return q;
1891}
Christoph Lameter19460892005-06-23 00:08:19 -07001892EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894/**
1895 * blk_init_queue - prepare a request queue for use with a block device
1896 * @rfn: The function to be called to process requests that have been
1897 * placed on the queue.
1898 * @lock: Request queue spin lock
1899 *
1900 * Description:
1901 * If a block device wishes to use the standard request handling procedures,
1902 * which sorts requests and coalesces adjacent requests, then it must
1903 * call blk_init_queue(). The function @rfn will be called when there
1904 * are requests on the queue that need to be processed. If the device
1905 * supports plugging, then @rfn may not be called immediately when requests
1906 * are available on the queue, but may be called at some time later instead.
1907 * Plugged queues are generally unplugged when a buffer belonging to one
1908 * of the requests on the queue is needed, or due to memory pressure.
1909 *
1910 * @rfn is not required, or even expected, to remove all requests off the
1911 * queue, but only as many as it can handle at a time. If it does leave
1912 * requests on the queue, it is responsible for arranging that the requests
1913 * get dealt with eventually.
1914 *
1915 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001916 * request queue; this lock will be taken also from interrupt context, so irq
1917 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 *
1919 * Function returns a pointer to the initialized request queue, or NULL if
1920 * it didn't succeed.
1921 *
1922 * Note:
1923 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1924 * when the block device is deactivated (such as at module unload).
1925 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1928{
Christoph Lameter19460892005-06-23 00:08:19 -07001929 return blk_init_queue_node(rfn, lock, -1);
1930}
1931EXPORT_SYMBOL(blk_init_queue);
1932
1933request_queue_t *
1934blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1935{
1936 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
1938 if (!q)
1939 return NULL;
1940
Christoph Lameter19460892005-06-23 00:08:19 -07001941 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001942 if (blk_init_free_list(q)) {
1943 kmem_cache_free(requestq_cachep, q);
1944 return NULL;
1945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
152587d2005-04-12 16:22:06 -05001947 /*
1948 * if caller didn't supply a lock, they get per-queue locking with
1949 * our embedded lock
1950 */
1951 if (!lock) {
1952 spin_lock_init(&q->__queue_lock);
1953 lock = &q->__queue_lock;
1954 }
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 q->request_fn = rfn;
1957 q->back_merge_fn = ll_back_merge_fn;
1958 q->front_merge_fn = ll_front_merge_fn;
1959 q->merge_requests_fn = ll_merge_requests_fn;
1960 q->prep_rq_fn = NULL;
1961 q->unplug_fn = generic_unplug_device;
1962 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1963 q->queue_lock = lock;
1964
1965 blk_queue_segment_boundary(q, 0xffffffff);
1966
1967 blk_queue_make_request(q, __make_request);
1968 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1969
1970 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1971 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1972
1973 /*
1974 * all done
1975 */
1976 if (!elevator_init(q, NULL)) {
1977 blk_queue_congestion_threshold(q);
1978 return q;
1979 }
1980
Al Viro8669aaf2006-03-18 13:50:00 -05001981 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 return NULL;
1983}
Christoph Lameter19460892005-06-23 00:08:19 -07001984EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986int blk_get_queue(request_queue_t *q)
1987{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001988 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001989 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 return 0;
1991 }
1992
1993 return 1;
1994}
1995
1996EXPORT_SYMBOL(blk_get_queue);
1997
1998static inline void blk_free_request(request_queue_t *q, struct request *rq)
1999{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002000 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02002001 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 mempool_free(rq, q->rq.rq_pool);
2003}
2004
Jens Axboe22e2c502005-06-27 10:55:12 +02002005static inline struct request *
Tejun Heocb98fc82005-10-28 08:29:39 +02002006blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
Linus Torvalds5dd96242005-10-28 08:56:34 -07002007 int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008{
2009 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
2010
2011 if (!rq)
2012 return NULL;
2013
2014 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02002015 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 * see bio.h and blkdev.h
2017 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002018 rq->cmd_flags = rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Tejun Heocb98fc82005-10-28 08:29:39 +02002020 if (priv) {
2021 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
2022 mempool_free(rq, q->rq.rq_pool);
2023 return NULL;
2024 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02002025 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02002026 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Tejun Heocb98fc82005-10-28 08:29:39 +02002028 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030
2031/*
2032 * ioc_batching returns true if the ioc is a valid batching request and
2033 * should be given priority access to a request.
2034 */
2035static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
2036{
2037 if (!ioc)
2038 return 0;
2039
2040 /*
2041 * Make sure the process is able to allocate at least 1 request
2042 * even if the batch times out, otherwise we could theoretically
2043 * lose wakeups.
2044 */
2045 return ioc->nr_batch_requests == q->nr_batching ||
2046 (ioc->nr_batch_requests > 0
2047 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2048}
2049
2050/*
2051 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2052 * will cause the process to be a "batcher" on all queues in the system. This
2053 * is the behaviour we want though - once it gets a wakeup it should be given
2054 * a nice run.
2055 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07002056static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057{
2058 if (!ioc || ioc_batching(q, ioc))
2059 return;
2060
2061 ioc->nr_batch_requests = q->nr_batching;
2062 ioc->last_waited = jiffies;
2063}
2064
2065static void __freed_request(request_queue_t *q, int rw)
2066{
2067 struct request_list *rl = &q->rq;
2068
2069 if (rl->count[rw] < queue_congestion_off_threshold(q))
2070 clear_queue_congested(q, rw);
2071
2072 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 if (waitqueue_active(&rl->wait[rw]))
2074 wake_up(&rl->wait[rw]);
2075
2076 blk_clear_queue_full(q, rw);
2077 }
2078}
2079
2080/*
2081 * A request has just been released. Account for it, update the full and
2082 * congestion status, wake up any waiters. Called under q->queue_lock.
2083 */
Tejun Heocb98fc82005-10-28 08:29:39 +02002084static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085{
2086 struct request_list *rl = &q->rq;
2087
2088 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002089 if (priv)
2090 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
2092 __freed_request(q, rw);
2093
2094 if (unlikely(rl->starved[rw ^ 1]))
2095 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096}
2097
2098#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2099/*
Nick Piggind6344532005-06-28 20:45:14 -07002100 * Get a free request, queue_lock must be held.
2101 * Returns NULL on failure, with queue_lock held.
2102 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002104static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04002105 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
2107 struct request *rq = NULL;
2108 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002109 struct io_context *ioc = NULL;
2110 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002112 may_queue = elv_may_queue(q, rw, bio);
2113 if (may_queue == ELV_MQUEUE_NO)
2114 goto rq_starved;
2115
2116 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2117 if (rl->count[rw]+1 >= q->nr_requests) {
2118 ioc = current_io_context(GFP_ATOMIC);
2119 /*
2120 * The queue will fill after this allocation, so set
2121 * it as full, and mark this process as "batching".
2122 * This process will be allowed to complete a batch of
2123 * requests, others will be blocked.
2124 */
2125 if (!blk_queue_full(q, rw)) {
2126 ioc_set_batching(q, ioc);
2127 blk_set_queue_full(q, rw);
2128 } else {
2129 if (may_queue != ELV_MQUEUE_MUST
2130 && !ioc_batching(q, ioc)) {
2131 /*
2132 * The queue is full and the allocating
2133 * process is not a "batcher", and not
2134 * exempted by the IO scheduler
2135 */
2136 goto out;
2137 }
2138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 }
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002140 set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 }
2142
Jens Axboe082cf692005-06-28 16:35:11 +02002143 /*
2144 * Only allow batching queuers to allocate up to 50% over the defined
2145 * limit of requests, otherwise we could have thousands of requests
2146 * allocated with any setting of ->nr_requests
2147 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002148 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002149 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002150
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 rl->count[rw]++;
2152 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002153
Jens Axboe64521d12005-10-28 08:30:39 +02002154 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002155 if (priv)
2156 rl->elvpriv++;
2157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 spin_unlock_irq(q->queue_lock);
2159
Tejun Heocb98fc82005-10-28 08:29:39 +02002160 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002161 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 /*
2163 * Allocation failed presumably due to memory. Undo anything
2164 * we might have messed up.
2165 *
2166 * Allocating task should really be put onto the front of the
2167 * wait queue, but this is pretty rare.
2168 */
2169 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002170 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172 /*
2173 * in the very unlikely event that allocation failed and no
2174 * requests for this direction was pending, mark us starved
2175 * so that freeing of a request in the other direction will
2176 * notice us. another possible fix would be to split the
2177 * rq mempool into READ and WRITE
2178 */
2179rq_starved:
2180 if (unlikely(rl->count[rw] == 0))
2181 rl->starved[rw] = 1;
2182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 goto out;
2184 }
2185
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002186 /*
2187 * ioc may be NULL here, and ioc_batching will be false. That's
2188 * OK, if the queue is under the request limit then requests need
2189 * not count toward the nr_batch_requests limit. There will always
2190 * be some limit enforced by BLK_BATCH_TIME.
2191 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 if (ioc_batching(q, ioc))
2193 ioc->nr_batch_requests--;
2194
2195 rq_init(q, rq);
2196 rq->rl = rl;
Jens Axboe2056a782006-03-23 20:00:26 +01002197
2198 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 return rq;
2201}
2202
2203/*
2204 * No available requests for this queue, unplug the device and wait for some
2205 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002206 *
2207 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002209static struct request *get_request_wait(request_queue_t *q, int rw,
2210 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 struct request *rq;
2213
Nick Piggin450991b2005-06-28 20:45:13 -07002214 rq = get_request(q, rw, bio, GFP_NOIO);
2215 while (!rq) {
2216 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 struct request_list *rl = &q->rq;
2218
2219 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2220 TASK_UNINTERRUPTIBLE);
2221
Jens Axboe22e2c502005-06-27 10:55:12 +02002222 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
2224 if (!rq) {
2225 struct io_context *ioc;
2226
Jens Axboe2056a782006-03-23 20:00:26 +01002227 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2228
Nick Piggind6344532005-06-28 20:45:14 -07002229 __generic_unplug_device(q);
2230 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 io_schedule();
2232
2233 /*
2234 * After sleeping, we become a "batching" process and
2235 * will be able to allocate at least one request, and
2236 * up to a big batch of them for a small period time.
2237 * See ioc_batching, ioc_set_batching
2238 */
Nick Pigginfb3cc432005-06-28 20:45:15 -07002239 ioc = current_io_context(GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002241
2242 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 }
2244 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
2247 return rq;
2248}
2249
Al Viro8267e262005-10-21 03:20:53 -04002250struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251{
2252 struct request *rq;
2253
2254 BUG_ON(rw != READ && rw != WRITE);
2255
Nick Piggind6344532005-06-28 20:45:14 -07002256 spin_lock_irq(q->queue_lock);
2257 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002258 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002259 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002260 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002261 if (!rq)
2262 spin_unlock_irq(q->queue_lock);
2263 }
2264 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 return rq;
2267}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268EXPORT_SYMBOL(blk_get_request);
2269
2270/**
2271 * blk_requeue_request - put a request back on queue
2272 * @q: request queue where request should be inserted
2273 * @rq: request to be inserted
2274 *
2275 * Description:
2276 * Drivers often keep queueing requests until the hardware cannot accept
2277 * more, when that condition happens we need to put the request back
2278 * on the queue. Must be called with queue lock held.
2279 */
2280void blk_requeue_request(request_queue_t *q, struct request *rq)
2281{
Jens Axboe2056a782006-03-23 20:00:26 +01002282 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2283
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 if (blk_rq_tagged(rq))
2285 blk_queue_end_tag(q, rq);
2286
2287 elv_requeue_request(q, rq);
2288}
2289
2290EXPORT_SYMBOL(blk_requeue_request);
2291
2292/**
2293 * blk_insert_request - insert a special request in to a request queue
2294 * @q: request queue where request should be inserted
2295 * @rq: request to be inserted
2296 * @at_head: insert request at head or tail of queue
2297 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 *
2299 * Description:
2300 * Many block devices need to execute commands asynchronously, so they don't
2301 * block the whole kernel from preemption during request execution. This is
2302 * accomplished normally by inserting aritficial requests tagged as
2303 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2304 * scheduled for actual execution by the request queue.
2305 *
2306 * We have the option of inserting the head or the tail of the queue.
2307 * Typically we use the tail for new ioctls and so forth. We use the head
2308 * of the queue for things like a QUEUE_FULL message from a device, or a
2309 * host that is unable to accept a particular command.
2310 */
2311void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002312 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313{
Tejun Heo 867d1192005-04-24 02:06:05 -05002314 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 unsigned long flags;
2316
2317 /*
2318 * tell I/O scheduler that this isn't a regular read/write (ie it
2319 * must not attempt merges on this) and that it acts as a soft
2320 * barrier
2321 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002322 rq->cmd_type = REQ_TYPE_SPECIAL;
2323 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
2325 rq->special = data;
2326
2327 spin_lock_irqsave(q->queue_lock, flags);
2328
2329 /*
2330 * If command is tagged, release the tag
2331 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002332 if (blk_rq_tagged(rq))
2333 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
Tejun Heo 867d1192005-04-24 02:06:05 -05002335 drive_stat_acct(rq, rq->nr_sectors, 1);
2336 __elv_add_request(q, rq, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 if (blk_queue_plugged(q))
2339 __generic_unplug_device(q);
2340 else
2341 q->request_fn(q);
2342 spin_unlock_irqrestore(q->queue_lock, flags);
2343}
2344
2345EXPORT_SYMBOL(blk_insert_request);
2346
2347/**
2348 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2349 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002350 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 * @ubuf: the user buffer
2352 * @len: length of user data
2353 *
2354 * Description:
2355 * Data will be mapped directly for zero copy io, if possible. Otherwise
2356 * a kernel bounce buffer is used.
2357 *
2358 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2359 * still in process context.
2360 *
2361 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2362 * before being submitted to the device, as pages mapped may be out of
2363 * reach. It's the callers responsibility to make sure this happens. The
2364 * original bio must be passed back in to blk_rq_unmap_user() for proper
2365 * unmapping.
2366 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002367int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2368 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369{
2370 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002372 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Mike Christiedefd94b2005-12-05 02:37:06 -06002374 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002375 return -EINVAL;
2376 if (!len || !ubuf)
2377 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
Jens Axboedd1cab92005-06-20 14:06:01 +02002379 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
2381 /*
2382 * if alignment requirement is satisfied, map in user pages for
2383 * direct dma. else, set up kernel bounce buffers
2384 */
2385 uaddr = (unsigned long) ubuf;
2386 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002387 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002389 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
2391 if (!IS_ERR(bio)) {
2392 rq->bio = rq->biotail = bio;
2393 blk_rq_bio_prep(q, rq, bio);
2394
2395 rq->buffer = rq->data = NULL;
2396 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002397 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 }
2399
2400 /*
2401 * bio is the err-ptr
2402 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002403 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404}
2405
2406EXPORT_SYMBOL(blk_rq_map_user);
2407
2408/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002409 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2410 * @q: request queue where request should be inserted
2411 * @rq: request to map data to
2412 * @iov: pointer to the iovec
2413 * @iov_count: number of elements in the iovec
2414 *
2415 * Description:
2416 * Data will be mapped directly for zero copy io, if possible. Otherwise
2417 * a kernel bounce buffer is used.
2418 *
2419 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2420 * still in process context.
2421 *
2422 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2423 * before being submitted to the device, as pages mapped may be out of
2424 * reach. It's the callers responsibility to make sure this happens. The
2425 * original bio must be passed back in to blk_rq_unmap_user() for proper
2426 * unmapping.
2427 */
2428int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2429 struct sg_iovec *iov, int iov_count)
2430{
2431 struct bio *bio;
2432
2433 if (!iov || iov_count <= 0)
2434 return -EINVAL;
2435
2436 /* we don't allow misaligned data like bio_map_user() does. If the
2437 * user is using sg, they're expected to know the alignment constraints
2438 * and respect them accordingly */
2439 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2440 if (IS_ERR(bio))
2441 return PTR_ERR(bio);
2442
2443 rq->bio = rq->biotail = bio;
2444 blk_rq_bio_prep(q, rq, bio);
2445 rq->buffer = rq->data = NULL;
2446 rq->data_len = bio->bi_size;
2447 return 0;
2448}
2449
2450EXPORT_SYMBOL(blk_rq_map_user_iov);
2451
2452/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002454 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 * @ulen: length of user buffer
2456 *
2457 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002458 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002460int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461{
2462 int ret = 0;
2463
2464 if (bio) {
2465 if (bio_flagged(bio, BIO_USER_MAPPED))
2466 bio_unmap_user(bio);
2467 else
2468 ret = bio_uncopy_user(bio);
2469 }
2470
Jens Axboedd1cab92005-06-20 14:06:01 +02002471 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472}
2473
2474EXPORT_SYMBOL(blk_rq_unmap_user);
2475
2476/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002477 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2478 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002479 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002480 * @kbuf: the kernel buffer
2481 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002482 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002483 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002484int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002485 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002486{
Mike Christie df46b9a2005-06-20 14:04:44 +02002487 struct bio *bio;
2488
Mike Christiedefd94b2005-12-05 02:37:06 -06002489 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002490 return -EINVAL;
2491 if (!len || !kbuf)
2492 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002493
2494 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002495 if (IS_ERR(bio))
2496 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002497
Jens Axboedd1cab92005-06-20 14:06:01 +02002498 if (rq_data_dir(rq) == WRITE)
2499 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002500
Jens Axboedd1cab92005-06-20 14:06:01 +02002501 rq->bio = rq->biotail = bio;
2502 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002503
Jens Axboedd1cab92005-06-20 14:06:01 +02002504 rq->buffer = rq->data = NULL;
2505 rq->data_len = len;
2506 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002507}
2508
2509EXPORT_SYMBOL(blk_rq_map_kern);
2510
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002511/**
2512 * blk_execute_rq_nowait - insert a request into queue for execution
2513 * @q: queue to insert the request in
2514 * @bd_disk: matching gendisk
2515 * @rq: request to insert
2516 * @at_head: insert request at head or tail of queue
2517 * @done: I/O completion handler
2518 *
2519 * Description:
2520 * Insert a fully prepared request at the back of the io scheduler queue
2521 * for execution. Don't wait for completion.
2522 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002523void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2524 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002525 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002526{
2527 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2528
2529 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002530 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002531 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002532 WARN_ON(irqs_disabled());
2533 spin_lock_irq(q->queue_lock);
2534 __elv_add_request(q, rq, where, 1);
2535 __generic_unplug_device(q);
2536 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002537}
Mike Christie6e39b692005-11-11 05:30:24 -06002538EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2539
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540/**
2541 * blk_execute_rq - insert a request into queue for execution
2542 * @q: queue to insert the request in
2543 * @bd_disk: matching gendisk
2544 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002545 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 *
2547 * Description:
2548 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002549 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 */
2551int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002552 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002554 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 char sense[SCSI_SENSE_BUFFERSIZE];
2556 int err = 0;
2557
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 /*
2559 * we need an extra reference to the request, so we can look at
2560 * it after io completion
2561 */
2562 rq->ref_count++;
2563
2564 if (!rq->sense) {
2565 memset(sense, 0, sizeof(sense));
2566 rq->sense = sense;
2567 rq->sense_len = 0;
2568 }
2569
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 rq->waiting = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002571 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 wait_for_completion(&wait);
2573 rq->waiting = NULL;
2574
2575 if (rq->errors)
2576 err = -EIO;
2577
2578 return err;
2579}
2580
2581EXPORT_SYMBOL(blk_execute_rq);
2582
2583/**
2584 * blkdev_issue_flush - queue a flush
2585 * @bdev: blockdev to issue flush for
2586 * @error_sector: error sector
2587 *
2588 * Description:
2589 * Issue a flush for the block device in question. Caller can supply
2590 * room for storing the error offset in case of a flush error, if they
2591 * wish to. Caller must run wait_for_completion() on its own.
2592 */
2593int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2594{
2595 request_queue_t *q;
2596
2597 if (bdev->bd_disk == NULL)
2598 return -ENXIO;
2599
2600 q = bdev_get_queue(bdev);
2601 if (!q)
2602 return -ENXIO;
2603 if (!q->issue_flush_fn)
2604 return -EOPNOTSUPP;
2605
2606 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2607}
2608
2609EXPORT_SYMBOL(blkdev_issue_flush);
2610
Adrian Bunk93d17d32005-06-25 14:59:10 -07002611static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612{
2613 int rw = rq_data_dir(rq);
2614
2615 if (!blk_fs_request(rq) || !rq->rq_disk)
2616 return;
2617
Jens Axboed72d9042005-11-01 08:35:42 +01002618 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002619 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002620 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 disk_round_stats(rq->rq_disk);
2622 rq->rq_disk->in_flight++;
2623 }
2624}
2625
2626/*
2627 * add-request adds a request to the linked list.
2628 * queue lock is held and interrupts disabled, as we muck with the
2629 * request queue list.
2630 */
2631static inline void add_request(request_queue_t * q, struct request * req)
2632{
2633 drive_stat_acct(req, req->nr_sectors, 1);
2634
2635 if (q->activity_fn)
2636 q->activity_fn(q->activity_data, rq_data_dir(req));
2637
2638 /*
2639 * elevator indicated where it wants this request to be
2640 * inserted at elevator_merge time
2641 */
2642 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2643}
2644
2645/*
2646 * disk_round_stats() - Round off the performance stats on a struct
2647 * disk_stats.
2648 *
2649 * The average IO queue length and utilisation statistics are maintained
2650 * by observing the current state of the queue length and the amount of
2651 * time it has been in this state for.
2652 *
2653 * Normally, that accounting is done on IO completion, but that can result
2654 * in more than a second's worth of IO being accounted for within any one
2655 * second, leading to >100% utilisation. To deal with that, we call this
2656 * function to do a round-off before returning the results when reading
2657 * /proc/diskstats. This accounts immediately for all queue usage up to
2658 * the current jiffies and restarts the counters again.
2659 */
2660void disk_round_stats(struct gendisk *disk)
2661{
2662 unsigned long now = jiffies;
2663
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002664 if (now == disk->stamp)
2665 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002667 if (disk->in_flight) {
2668 __disk_stat_add(disk, time_in_queue,
2669 disk->in_flight * (now - disk->stamp));
2670 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673}
2674
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002675EXPORT_SYMBOL_GPL(disk_round_stats);
2676
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677/*
2678 * queue lock must be held
2679 */
Mike Christie6e39b692005-11-11 05:30:24 -06002680void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681{
2682 struct request_list *rl = req->rl;
2683
2684 if (unlikely(!q))
2685 return;
2686 if (unlikely(--req->ref_count))
2687 return;
2688
Tejun Heo8922e162005-10-20 16:23:44 +02002689 elv_completed_request(q, req);
2690
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 req->rq_status = RQ_INACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 req->rl = NULL;
2693
2694 /*
2695 * Request may not have originated from ll_rw_blk. if not,
2696 * it didn't come out of our reserved rq pools
2697 */
2698 if (rl) {
2699 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002700 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 BUG_ON(!list_empty(&req->queuelist));
2703
2704 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002705 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 }
2707}
2708
Mike Christie6e39b692005-11-11 05:30:24 -06002709EXPORT_SYMBOL_GPL(__blk_put_request);
2710
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711void blk_put_request(struct request *req)
2712{
Tejun Heo8922e162005-10-20 16:23:44 +02002713 unsigned long flags;
2714 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Tejun Heo8922e162005-10-20 16:23:44 +02002716 /*
2717 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2718 * following if (q) test.
2719 */
2720 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 spin_lock_irqsave(q->queue_lock, flags);
2722 __blk_put_request(q, req);
2723 spin_unlock_irqrestore(q->queue_lock, flags);
2724 }
2725}
2726
2727EXPORT_SYMBOL(blk_put_request);
2728
2729/**
2730 * blk_end_sync_rq - executes a completion event on a request
2731 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002732 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002734void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735{
2736 struct completion *waiting = rq->waiting;
2737
2738 rq->waiting = NULL;
2739 __blk_put_request(rq->q, rq);
2740
2741 /*
2742 * complete last, if this is a stack request the process (and thus
2743 * the rq pointer) could be invalid right after this complete()
2744 */
2745 complete(waiting);
2746}
2747EXPORT_SYMBOL(blk_end_sync_rq);
2748
2749/**
2750 * blk_congestion_wait - wait for a queue to become uncongested
2751 * @rw: READ or WRITE
2752 * @timeout: timeout in jiffies
2753 *
2754 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2755 * If no queues are congested then just wait for the next request to be
2756 * returned.
2757 */
2758long blk_congestion_wait(int rw, long timeout)
2759{
2760 long ret;
2761 DEFINE_WAIT(wait);
2762 wait_queue_head_t *wqh = &congestion_wqh[rw];
2763
2764 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2765 ret = io_schedule_timeout(timeout);
2766 finish_wait(wqh, &wait);
2767 return ret;
2768}
2769
2770EXPORT_SYMBOL(blk_congestion_wait);
2771
Trond Myklebust275a0822006-08-22 20:06:24 -04002772/**
2773 * blk_congestion_end - wake up sleepers on a congestion queue
2774 * @rw: READ or WRITE
2775 */
2776void blk_congestion_end(int rw)
2777{
2778 wait_queue_head_t *wqh = &congestion_wqh[rw];
2779
2780 if (waitqueue_active(wqh))
2781 wake_up(wqh);
2782}
2783
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784/*
2785 * Has to be called with the request spinlock acquired
2786 */
2787static int attempt_merge(request_queue_t *q, struct request *req,
2788 struct request *next)
2789{
2790 if (!rq_mergeable(req) || !rq_mergeable(next))
2791 return 0;
2792
2793 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002794 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 */
2796 if (req->sector + req->nr_sectors != next->sector)
2797 return 0;
2798
2799 if (rq_data_dir(req) != rq_data_dir(next)
2800 || req->rq_disk != next->rq_disk
2801 || next->waiting || next->special)
2802 return 0;
2803
2804 /*
2805 * If we are allowed to merge, then append bio list
2806 * from next to rq and release next. merge_requests_fn
2807 * will have updated segment counts, update sector
2808 * counts here.
2809 */
2810 if (!q->merge_requests_fn(q, req, next))
2811 return 0;
2812
2813 /*
2814 * At this point we have either done a back merge
2815 * or front merge. We need the smaller start_time of
2816 * the merged requests to be the current request
2817 * for accounting purposes.
2818 */
2819 if (time_after(req->start_time, next->start_time))
2820 req->start_time = next->start_time;
2821
2822 req->biotail->bi_next = next->bio;
2823 req->biotail = next->biotail;
2824
2825 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2826
2827 elv_merge_requests(q, req, next);
2828
2829 if (req->rq_disk) {
2830 disk_round_stats(req->rq_disk);
2831 req->rq_disk->in_flight--;
2832 }
2833
Jens Axboe22e2c502005-06-27 10:55:12 +02002834 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2835
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 __blk_put_request(q, next);
2837 return 1;
2838}
2839
2840static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2841{
2842 struct request *next = elv_latter_request(q, rq);
2843
2844 if (next)
2845 return attempt_merge(q, rq, next);
2846
2847 return 0;
2848}
2849
2850static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2851{
2852 struct request *prev = elv_former_request(q, rq);
2853
2854 if (prev)
2855 return attempt_merge(q, prev, rq);
2856
2857 return 0;
2858}
2859
Tejun Heo52d9e672006-01-06 09:49:58 +01002860static void init_request_from_bio(struct request *req, struct bio *bio)
2861{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002862 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002863
2864 /*
2865 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2866 */
2867 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002868 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002869
2870 /*
2871 * REQ_BARRIER implies no merging, but lets make it explicit
2872 */
2873 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002874 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002875
Jens Axboeb31dc662006-06-13 08:26:10 +02002876 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002877 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboeb31dc662006-06-13 08:26:10 +02002878
Tejun Heo52d9e672006-01-06 09:49:58 +01002879 req->errors = 0;
2880 req->hard_sector = req->sector = bio->bi_sector;
2881 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2882 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2883 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2884 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2885 req->buffer = bio_data(bio); /* see ->buffer comment above */
2886 req->waiting = NULL;
2887 req->bio = req->biotail = bio;
2888 req->ioprio = bio_prio(bio);
2889 req->rq_disk = bio->bi_bdev->bd_disk;
2890 req->start_time = jiffies;
2891}
2892
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893static int __make_request(request_queue_t *q, struct bio *bio)
2894{
Nick Piggin450991b2005-06-28 20:45:13 -07002895 struct request *req;
Jens Axboe4a534f92005-04-16 15:25:40 -07002896 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
Jens Axboe22e2c502005-06-27 10:55:12 +02002897 unsigned short prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 sector_t sector;
2899
2900 sector = bio->bi_sector;
2901 nr_sectors = bio_sectors(bio);
2902 cur_nr_sectors = bio_cur_sectors(bio);
Jens Axboe22e2c502005-06-27 10:55:12 +02002903 prio = bio_prio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904
2905 rw = bio_data_dir(bio);
Jens Axboe4a534f92005-04-16 15:25:40 -07002906 sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907
2908 /*
2909 * low level driver can indicate that it wants pages above a
2910 * certain limit bounced to low memory (ie for highmem, or even
2911 * ISA dma in theory)
2912 */
2913 blk_queue_bounce(q, &bio);
2914
2915 spin_lock_prefetch(q->queue_lock);
2916
2917 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002918 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 err = -EOPNOTSUPP;
2920 goto end_io;
2921 }
2922
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 spin_lock_irq(q->queue_lock);
2924
Nick Piggin450991b2005-06-28 20:45:13 -07002925 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 goto get_rq;
2927
2928 el_ret = elv_merge(q, &req, bio);
2929 switch (el_ret) {
2930 case ELEVATOR_BACK_MERGE:
2931 BUG_ON(!rq_mergeable(req));
2932
2933 if (!q->back_merge_fn(q, req, bio))
2934 break;
2935
Jens Axboe2056a782006-03-23 20:00:26 +01002936 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2937
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 req->biotail->bi_next = bio;
2939 req->biotail = bio;
2940 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002941 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 drive_stat_acct(req, nr_sectors, 0);
2943 if (!attempt_back_merge(q, req))
2944 elv_merged_request(q, req);
2945 goto out;
2946
2947 case ELEVATOR_FRONT_MERGE:
2948 BUG_ON(!rq_mergeable(req));
2949
2950 if (!q->front_merge_fn(q, req, bio))
2951 break;
2952
Jens Axboe2056a782006-03-23 20:00:26 +01002953 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2954
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 bio->bi_next = req->bio;
2956 req->bio = bio;
2957
2958 /*
2959 * may not be valid. if the low level driver said
2960 * it didn't need a bounce buffer then it better
2961 * not touch req->buffer either...
2962 */
2963 req->buffer = bio_data(bio);
2964 req->current_nr_sectors = cur_nr_sectors;
2965 req->hard_cur_sectors = cur_nr_sectors;
2966 req->sector = req->hard_sector = sector;
2967 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002968 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 drive_stat_acct(req, nr_sectors, 0);
2970 if (!attempt_front_merge(q, req))
2971 elv_merged_request(q, req);
2972 goto out;
2973
Nick Piggin450991b2005-06-28 20:45:13 -07002974 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002976 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 }
2978
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002980 /*
2981 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002982 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002983 */
Nick Piggin450991b2005-06-28 20:45:13 -07002984 req = get_request_wait(q, rw, bio);
Nick Piggind6344532005-06-28 20:45:14 -07002985
Nick Piggin450991b2005-06-28 20:45:13 -07002986 /*
2987 * After dropping the lock and possibly sleeping here, our request
2988 * may now be mergeable after it had proven unmergeable (above).
2989 * We don't worry about that case for efficiency. It won't happen
2990 * often, and the elevators are able to handle it.
2991 */
Tejun Heo52d9e672006-01-06 09:49:58 +01002992 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
Nick Piggin450991b2005-06-28 20:45:13 -07002994 spin_lock_irq(q->queue_lock);
2995 if (elv_queue_empty(q))
2996 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 add_request(q, req);
2998out:
Jens Axboe4a534f92005-04-16 15:25:40 -07002999 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 __generic_unplug_device(q);
3001
3002 spin_unlock_irq(q->queue_lock);
3003 return 0;
3004
3005end_io:
3006 bio_endio(bio, nr_sectors << 9, err);
3007 return 0;
3008}
3009
3010/*
3011 * If bio->bi_dev is a partition, remap the location
3012 */
3013static inline void blk_partition_remap(struct bio *bio)
3014{
3015 struct block_device *bdev = bio->bi_bdev;
3016
3017 if (bdev != bdev->bd_contains) {
3018 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003019 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
Jens Axboea3623572005-11-01 09:26:16 +01003021 p->sectors[rw] += bio_sectors(bio);
3022 p->ios[rw]++;
3023
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 bio->bi_sector += p->start_sect;
3025 bio->bi_bdev = bdev->bd_contains;
3026 }
3027}
3028
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029static void handle_bad_sector(struct bio *bio)
3030{
3031 char b[BDEVNAME_SIZE];
3032
3033 printk(KERN_INFO "attempt to access beyond end of device\n");
3034 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3035 bdevname(bio->bi_bdev, b),
3036 bio->bi_rw,
3037 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3038 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3039
3040 set_bit(BIO_EOF, &bio->bi_flags);
3041}
3042
3043/**
3044 * generic_make_request: hand a buffer to its device driver for I/O
3045 * @bio: The bio describing the location in memory and on the device.
3046 *
3047 * generic_make_request() is used to make I/O requests of block
3048 * devices. It is passed a &struct bio, which describes the I/O that needs
3049 * to be done.
3050 *
3051 * generic_make_request() does not return any status. The
3052 * success/failure status of the request, along with notification of
3053 * completion, is delivered asynchronously through the bio->bi_end_io
3054 * function described (one day) else where.
3055 *
3056 * The caller of generic_make_request must make sure that bi_io_vec
3057 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3058 * set to describe the device address, and the
3059 * bi_end_io and optionally bi_private are set to describe how
3060 * completion notification should be signaled.
3061 *
3062 * generic_make_request and the drivers it calls may use bi_next if this
3063 * bio happens to be merged with someone else, and may change bi_dev and
3064 * bi_sector for remaps as it sees fit. So the values of these fields
3065 * should NOT be depended on after the call to generic_make_request.
3066 */
3067void generic_make_request(struct bio *bio)
3068{
3069 request_queue_t *q;
3070 sector_t maxsector;
3071 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003072 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073
3074 might_sleep();
3075 /* Test device or partition size, when known. */
3076 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3077 if (maxsector) {
3078 sector_t sector = bio->bi_sector;
3079
3080 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3081 /*
3082 * This may well happen - the kernel calls bread()
3083 * without checking the size of the device, e.g., when
3084 * mounting a device.
3085 */
3086 handle_bad_sector(bio);
3087 goto end_io;
3088 }
3089 }
3090
3091 /*
3092 * Resolve the mapping until finished. (drivers are
3093 * still free to implement/resolve their own stacking
3094 * by explicitly returning 0)
3095 *
3096 * NOTE: we don't repeat the blk_size check for each new device.
3097 * Stacking drivers are expected to know what they are doing.
3098 */
Jens Axboe2056a782006-03-23 20:00:26 +01003099 maxsector = -1;
3100 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 do {
3102 char b[BDEVNAME_SIZE];
3103
3104 q = bdev_get_queue(bio->bi_bdev);
3105 if (!q) {
3106 printk(KERN_ERR
3107 "generic_make_request: Trying to access "
3108 "nonexistent block-device %s (%Lu)\n",
3109 bdevname(bio->bi_bdev, b),
3110 (long long) bio->bi_sector);
3111end_io:
3112 bio_endio(bio, bio->bi_size, -EIO);
3113 break;
3114 }
3115
3116 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3117 printk("bio too big device %s (%u > %u)\n",
3118 bdevname(bio->bi_bdev, b),
3119 bio_sectors(bio),
3120 q->max_hw_sectors);
3121 goto end_io;
3122 }
3123
Nick Pigginfde6ad22005-06-23 00:08:53 -07003124 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 goto end_io;
3126
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 /*
3128 * If this device has partitions, remap block n
3129 * of partition p to block n+start(p) of the disk.
3130 */
3131 blk_partition_remap(bio);
3132
Jens Axboe2056a782006-03-23 20:00:26 +01003133 if (maxsector != -1)
3134 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
3135 maxsector);
3136
3137 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3138
3139 maxsector = bio->bi_sector;
3140 old_dev = bio->bi_bdev->bd_dev;
3141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 ret = q->make_request_fn(q, bio);
3143 } while (ret);
3144}
3145
3146EXPORT_SYMBOL(generic_make_request);
3147
3148/**
3149 * submit_bio: submit a bio to the block device layer for I/O
3150 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3151 * @bio: The &struct bio which describes the I/O
3152 *
3153 * submit_bio() is very similar in purpose to generic_make_request(), and
3154 * uses that function to do most of the work. Both are fairly rough
3155 * interfaces, @bio must be presetup and ready for I/O.
3156 *
3157 */
3158void submit_bio(int rw, struct bio *bio)
3159{
3160 int count = bio_sectors(bio);
3161
3162 BIO_BUG_ON(!bio->bi_size);
3163 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003164 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 if (rw & WRITE)
Christoph Lameterf8891e52006-06-30 01:55:45 -07003166 count_vm_events(PGPGOUT, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 else
Christoph Lameterf8891e52006-06-30 01:55:45 -07003168 count_vm_events(PGPGIN, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169
3170 if (unlikely(block_dump)) {
3171 char b[BDEVNAME_SIZE];
3172 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3173 current->comm, current->pid,
3174 (rw & WRITE) ? "WRITE" : "READ",
3175 (unsigned long long)bio->bi_sector,
3176 bdevname(bio->bi_bdev,b));
3177 }
3178
3179 generic_make_request(bio);
3180}
3181
3182EXPORT_SYMBOL(submit_bio);
3183
Adrian Bunk93d17d32005-06-25 14:59:10 -07003184static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
3186 struct bio *bio, *prevbio = NULL;
3187 int nr_phys_segs, nr_hw_segs;
3188 unsigned int phys_size, hw_size;
3189 request_queue_t *q = rq->q;
3190
3191 if (!rq->bio)
3192 return;
3193
3194 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3195 rq_for_each_bio(bio, rq) {
3196 /* Force bio hw/phys segs to be recalculated. */
3197 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3198
3199 nr_phys_segs += bio_phys_segments(q, bio);
3200 nr_hw_segs += bio_hw_segments(q, bio);
3201 if (prevbio) {
3202 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3203 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3204
3205 if (blk_phys_contig_segment(q, prevbio, bio) &&
3206 pseg <= q->max_segment_size) {
3207 nr_phys_segs--;
3208 phys_size += prevbio->bi_size + bio->bi_size;
3209 } else
3210 phys_size = 0;
3211
3212 if (blk_hw_contig_segment(q, prevbio, bio) &&
3213 hseg <= q->max_segment_size) {
3214 nr_hw_segs--;
3215 hw_size += prevbio->bi_size + bio->bi_size;
3216 } else
3217 hw_size = 0;
3218 }
3219 prevbio = bio;
3220 }
3221
3222 rq->nr_phys_segments = nr_phys_segs;
3223 rq->nr_hw_segments = nr_hw_segs;
3224}
3225
Adrian Bunk93d17d32005-06-25 14:59:10 -07003226static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227{
3228 if (blk_fs_request(rq)) {
3229 rq->hard_sector += nsect;
3230 rq->hard_nr_sectors -= nsect;
3231
3232 /*
3233 * Move the I/O submission pointers ahead if required.
3234 */
3235 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3236 (rq->sector <= rq->hard_sector)) {
3237 rq->sector = rq->hard_sector;
3238 rq->nr_sectors = rq->hard_nr_sectors;
3239 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3240 rq->current_nr_sectors = rq->hard_cur_sectors;
3241 rq->buffer = bio_data(rq->bio);
3242 }
3243
3244 /*
3245 * if total number of sectors is less than the first segment
3246 * size, something has gone terribly wrong
3247 */
3248 if (rq->nr_sectors < rq->current_nr_sectors) {
3249 printk("blk: request botched\n");
3250 rq->nr_sectors = rq->current_nr_sectors;
3251 }
3252 }
3253}
3254
3255static int __end_that_request_first(struct request *req, int uptodate,
3256 int nr_bytes)
3257{
3258 int total_bytes, bio_nbytes, error, next_idx = 0;
3259 struct bio *bio;
3260
Jens Axboe2056a782006-03-23 20:00:26 +01003261 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3262
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 /*
3264 * extend uptodate bool to allow < 0 value to be direct io error
3265 */
3266 error = 0;
3267 if (end_io_error(uptodate))
3268 error = !uptodate ? -EIO : uptodate;
3269
3270 /*
3271 * for a REQ_BLOCK_PC request, we want to carry any eventual
3272 * sense key with us all the way through
3273 */
3274 if (!blk_pc_request(req))
3275 req->errors = 0;
3276
3277 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003278 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 printk("end_request: I/O error, dev %s, sector %llu\n",
3280 req->rq_disk ? req->rq_disk->disk_name : "?",
3281 (unsigned long long)req->sector);
3282 }
3283
Jens Axboed72d9042005-11-01 08:35:42 +01003284 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003285 const int rw = rq_data_dir(req);
3286
Jens Axboe53e86062006-01-17 11:09:27 +01003287 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003288 }
3289
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 total_bytes = bio_nbytes = 0;
3291 while ((bio = req->bio) != NULL) {
3292 int nbytes;
3293
3294 if (nr_bytes >= bio->bi_size) {
3295 req->bio = bio->bi_next;
3296 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003297 if (!ordered_bio_endio(req, bio, nbytes, error))
3298 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 next_idx = 0;
3300 bio_nbytes = 0;
3301 } else {
3302 int idx = bio->bi_idx + next_idx;
3303
3304 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3305 blk_dump_rq_flags(req, "__end_that");
3306 printk("%s: bio idx %d >= vcnt %d\n",
3307 __FUNCTION__,
3308 bio->bi_idx, bio->bi_vcnt);
3309 break;
3310 }
3311
3312 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3313 BIO_BUG_ON(nbytes > bio->bi_size);
3314
3315 /*
3316 * not a complete bvec done
3317 */
3318 if (unlikely(nbytes > nr_bytes)) {
3319 bio_nbytes += nr_bytes;
3320 total_bytes += nr_bytes;
3321 break;
3322 }
3323
3324 /*
3325 * advance to the next vector
3326 */
3327 next_idx++;
3328 bio_nbytes += nbytes;
3329 }
3330
3331 total_bytes += nbytes;
3332 nr_bytes -= nbytes;
3333
3334 if ((bio = req->bio)) {
3335 /*
3336 * end more in this run, or just return 'not-done'
3337 */
3338 if (unlikely(nr_bytes <= 0))
3339 break;
3340 }
3341 }
3342
3343 /*
3344 * completely done
3345 */
3346 if (!req->bio)
3347 return 0;
3348
3349 /*
3350 * if the request wasn't completed, update state
3351 */
3352 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003353 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3354 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 bio->bi_idx += next_idx;
3356 bio_iovec(bio)->bv_offset += nr_bytes;
3357 bio_iovec(bio)->bv_len -= nr_bytes;
3358 }
3359
3360 blk_recalc_rq_sectors(req, total_bytes >> 9);
3361 blk_recalc_rq_segments(req);
3362 return 1;
3363}
3364
3365/**
3366 * end_that_request_first - end I/O on a request
3367 * @req: the request being processed
3368 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3369 * @nr_sectors: number of sectors to end I/O on
3370 *
3371 * Description:
3372 * Ends I/O on a number of sectors attached to @req, and sets it up
3373 * for the next range of segments (if any) in the cluster.
3374 *
3375 * Return:
3376 * 0 - we are done with this request, call end_that_request_last()
3377 * 1 - still buffers pending for this request
3378 **/
3379int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3380{
3381 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3382}
3383
3384EXPORT_SYMBOL(end_that_request_first);
3385
3386/**
3387 * end_that_request_chunk - end I/O on a request
3388 * @req: the request being processed
3389 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3390 * @nr_bytes: number of bytes to complete
3391 *
3392 * Description:
3393 * Ends I/O on a number of bytes attached to @req, and sets it up
3394 * for the next range of segments (if any). Like end_that_request_first(),
3395 * but deals with bytes instead of sectors.
3396 *
3397 * Return:
3398 * 0 - we are done with this request, call end_that_request_last()
3399 * 1 - still buffers pending for this request
3400 **/
3401int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3402{
3403 return __end_that_request_first(req, uptodate, nr_bytes);
3404}
3405
3406EXPORT_SYMBOL(end_that_request_chunk);
3407
3408/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003409 * splice the completion data to a local structure and hand off to
3410 * process_completion_queue() to complete the requests
3411 */
3412static void blk_done_softirq(struct softirq_action *h)
3413{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003414 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003415
3416 local_irq_disable();
3417 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003418 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003419 local_irq_enable();
3420
3421 while (!list_empty(&local_list)) {
3422 struct request *rq = list_entry(local_list.next, struct request, donelist);
3423
3424 list_del_init(&rq->donelist);
3425 rq->q->softirq_done_fn(rq);
3426 }
3427}
3428
3429#ifdef CONFIG_HOTPLUG_CPU
3430
3431static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3432 void *hcpu)
3433{
3434 /*
3435 * If a CPU goes away, splice its entries to the current CPU
3436 * and trigger a run of the softirq
3437 */
3438 if (action == CPU_DEAD) {
3439 int cpu = (unsigned long) hcpu;
3440
3441 local_irq_disable();
3442 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3443 &__get_cpu_var(blk_cpu_done));
3444 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3445 local_irq_enable();
3446 }
3447
3448 return NOTIFY_OK;
3449}
3450
3451
Chandra Seetharaman054cc8a2006-06-27 02:54:07 -07003452static struct notifier_block __devinitdata blk_cpu_notifier = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003453 .notifier_call = blk_cpu_notify,
3454};
3455
3456#endif /* CONFIG_HOTPLUG_CPU */
3457
3458/**
3459 * blk_complete_request - end I/O on a request
3460 * @req: the request being processed
3461 *
3462 * Description:
3463 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003464 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003465 * through requeueing. Theh actual completion happens out-of-order,
3466 * through a softirq handler. The user must have registered a completion
3467 * callback through blk_queue_softirq_done().
3468 **/
3469
3470void blk_complete_request(struct request *req)
3471{
3472 struct list_head *cpu_list;
3473 unsigned long flags;
3474
3475 BUG_ON(!req->q->softirq_done_fn);
3476
3477 local_irq_save(flags);
3478
3479 cpu_list = &__get_cpu_var(blk_cpu_done);
3480 list_add_tail(&req->donelist, cpu_list);
3481 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3482
3483 local_irq_restore(flags);
3484}
3485
3486EXPORT_SYMBOL(blk_complete_request);
3487
3488/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 * queue lock must be held
3490 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003491void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492{
3493 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003494 int error;
3495
3496 /*
3497 * extend uptodate bool to allow < 0 value to be direct io error
3498 */
3499 error = 0;
3500 if (end_io_error(uptodate))
3501 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502
3503 if (unlikely(laptop_mode) && blk_fs_request(req))
3504 laptop_io_completion();
3505
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003506 /*
3507 * Account IO completion. bar_rq isn't accounted as a normal
3508 * IO on queueing nor completion. Accounting the containing
3509 * request is enough.
3510 */
3511 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003513 const int rw = rq_data_dir(req);
3514
3515 __disk_stat_inc(disk, ios[rw]);
3516 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 disk_round_stats(disk);
3518 disk->in_flight--;
3519 }
3520 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003521 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 else
3523 __blk_put_request(req->q, req);
3524}
3525
3526EXPORT_SYMBOL(end_that_request_last);
3527
3528void end_request(struct request *req, int uptodate)
3529{
3530 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3531 add_disk_randomness(req->rq_disk);
3532 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003533 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 }
3535}
3536
3537EXPORT_SYMBOL(end_request);
3538
3539void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3540{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003541 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3542 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543
3544 rq->nr_phys_segments = bio_phys_segments(q, bio);
3545 rq->nr_hw_segments = bio_hw_segments(q, bio);
3546 rq->current_nr_sectors = bio_cur_sectors(bio);
3547 rq->hard_cur_sectors = rq->current_nr_sectors;
3548 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3549 rq->buffer = bio_data(bio);
3550
3551 rq->bio = rq->biotail = bio;
3552}
3553
3554EXPORT_SYMBOL(blk_rq_bio_prep);
3555
3556int kblockd_schedule_work(struct work_struct *work)
3557{
3558 return queue_work(kblockd_workqueue, work);
3559}
3560
3561EXPORT_SYMBOL(kblockd_schedule_work);
3562
3563void kblockd_flush(void)
3564{
3565 flush_workqueue(kblockd_workqueue);
3566}
3567EXPORT_SYMBOL(kblockd_flush);
3568
3569int __init blk_dev_init(void)
3570{
Jens Axboeff856ba2006-01-09 16:02:34 +01003571 int i;
3572
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 kblockd_workqueue = create_workqueue("kblockd");
3574 if (!kblockd_workqueue)
3575 panic("Failed to create kblockd\n");
3576
3577 request_cachep = kmem_cache_create("blkdev_requests",
3578 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3579
3580 requestq_cachep = kmem_cache_create("blkdev_queue",
3581 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3582
3583 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3584 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3585
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003586 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003587 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3588
3589 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003590 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003591
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 blk_max_low_pfn = max_low_pfn;
3593 blk_max_pfn = max_pfn;
3594
3595 return 0;
3596}
3597
3598/*
3599 * IO Context helper functions
3600 */
3601void put_io_context(struct io_context *ioc)
3602{
3603 if (ioc == NULL)
3604 return;
3605
3606 BUG_ON(atomic_read(&ioc->refcount) == 0);
3607
3608 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003609 struct cfq_io_context *cic;
3610
Al Viro334e94d2006-03-18 15:05:53 -05003611 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 if (ioc->aic && ioc->aic->dtor)
3613 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003614 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003615 struct rb_node *n = rb_first(&ioc->cic_root);
3616
3617 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003618 cic->dtor(ioc);
3619 }
Al Viro334e94d2006-03-18 15:05:53 -05003620 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621
3622 kmem_cache_free(iocontext_cachep, ioc);
3623 }
3624}
3625EXPORT_SYMBOL(put_io_context);
3626
3627/* Called by the exitting task */
3628void exit_io_context(void)
3629{
3630 unsigned long flags;
3631 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003632 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633
3634 local_irq_save(flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02003635 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 ioc = current->io_context;
3637 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003638 ioc->task = NULL;
3639 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640 local_irq_restore(flags);
3641
3642 if (ioc->aic && ioc->aic->exit)
3643 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003644 if (ioc->cic_root.rb_node != NULL) {
3645 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3646 cic->exit(ioc);
3647 }
3648
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 put_io_context(ioc);
3650}
3651
3652/*
3653 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003654 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003656 * This returned IO context doesn't have a specifically elevated refcount,
3657 * but since the current task itself holds a reference, the context can be
3658 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003659 */
Al Viro8267e262005-10-21 03:20:53 -04003660struct io_context *current_io_context(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661{
3662 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 struct io_context *ret;
3664
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003666 if (likely(ret))
3667 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668
3669 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3670 if (ret) {
3671 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003672 ret->task = current;
3673 ret->set_ioprio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 ret->last_waited = jiffies; /* doesn't matter... */
3675 ret->nr_batch_requests = 0; /* because this is 0 */
3676 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003677 ret->cic_root.rb_node = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003678 /* make sure set_task_ioprio() sees the settings above */
3679 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003680 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 }
3682
3683 return ret;
3684}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003685EXPORT_SYMBOL(current_io_context);
3686
3687/*
3688 * If the current task has no IO context then create one and initialise it.
3689 * If it does have a context, take a ref on it.
3690 *
3691 * This is always called in the context of the task which submitted the I/O.
3692 */
Al Viro8267e262005-10-21 03:20:53 -04003693struct io_context *get_io_context(gfp_t gfp_flags)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003694{
3695 struct io_context *ret;
3696 ret = current_io_context(gfp_flags);
3697 if (likely(ret))
3698 atomic_inc(&ret->refcount);
3699 return ret;
3700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701EXPORT_SYMBOL(get_io_context);
3702
3703void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3704{
3705 struct io_context *src = *psrc;
3706 struct io_context *dst = *pdst;
3707
3708 if (src) {
3709 BUG_ON(atomic_read(&src->refcount) == 0);
3710 atomic_inc(&src->refcount);
3711 put_io_context(dst);
3712 *pdst = src;
3713 }
3714}
3715EXPORT_SYMBOL(copy_io_context);
3716
3717void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3718{
3719 struct io_context *temp;
3720 temp = *ioc1;
3721 *ioc1 = *ioc2;
3722 *ioc2 = temp;
3723}
3724EXPORT_SYMBOL(swap_io_context);
3725
3726/*
3727 * sysfs parts below
3728 */
3729struct queue_sysfs_entry {
3730 struct attribute attr;
3731 ssize_t (*show)(struct request_queue *, char *);
3732 ssize_t (*store)(struct request_queue *, const char *, size_t);
3733};
3734
3735static ssize_t
3736queue_var_show(unsigned int var, char *page)
3737{
3738 return sprintf(page, "%d\n", var);
3739}
3740
3741static ssize_t
3742queue_var_store(unsigned long *var, const char *page, size_t count)
3743{
3744 char *p = (char *) page;
3745
3746 *var = simple_strtoul(p, &p, 10);
3747 return count;
3748}
3749
3750static ssize_t queue_requests_show(struct request_queue *q, char *page)
3751{
3752 return queue_var_show(q->nr_requests, (page));
3753}
3754
3755static ssize_t
3756queue_requests_store(struct request_queue *q, const char *page, size_t count)
3757{
3758 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003759 unsigned long nr;
3760 int ret = queue_var_store(&nr, page, count);
3761 if (nr < BLKDEV_MIN_RQ)
3762 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763
Al Viroc981ff92006-03-18 13:51:29 -05003764 spin_lock_irq(q->queue_lock);
3765 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003766 blk_queue_congestion_threshold(q);
3767
3768 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3769 set_queue_congested(q, READ);
3770 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3771 clear_queue_congested(q, READ);
3772
3773 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3774 set_queue_congested(q, WRITE);
3775 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3776 clear_queue_congested(q, WRITE);
3777
3778 if (rl->count[READ] >= q->nr_requests) {
3779 blk_set_queue_full(q, READ);
3780 } else if (rl->count[READ]+1 <= q->nr_requests) {
3781 blk_clear_queue_full(q, READ);
3782 wake_up(&rl->wait[READ]);
3783 }
3784
3785 if (rl->count[WRITE] >= q->nr_requests) {
3786 blk_set_queue_full(q, WRITE);
3787 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3788 blk_clear_queue_full(q, WRITE);
3789 wake_up(&rl->wait[WRITE]);
3790 }
Al Viroc981ff92006-03-18 13:51:29 -05003791 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 return ret;
3793}
3794
3795static ssize_t queue_ra_show(struct request_queue *q, char *page)
3796{
3797 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3798
3799 return queue_var_show(ra_kb, (page));
3800}
3801
3802static ssize_t
3803queue_ra_store(struct request_queue *q, const char *page, size_t count)
3804{
3805 unsigned long ra_kb;
3806 ssize_t ret = queue_var_store(&ra_kb, page, count);
3807
3808 spin_lock_irq(q->queue_lock);
3809 if (ra_kb > (q->max_sectors >> 1))
3810 ra_kb = (q->max_sectors >> 1);
3811
3812 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3813 spin_unlock_irq(q->queue_lock);
3814
3815 return ret;
3816}
3817
3818static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3819{
3820 int max_sectors_kb = q->max_sectors >> 1;
3821
3822 return queue_var_show(max_sectors_kb, (page));
3823}
3824
3825static ssize_t
3826queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3827{
3828 unsigned long max_sectors_kb,
3829 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3830 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3831 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3832 int ra_kb;
3833
3834 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3835 return -EINVAL;
3836 /*
3837 * Take the queue lock to update the readahead and max_sectors
3838 * values synchronously:
3839 */
3840 spin_lock_irq(q->queue_lock);
3841 /*
3842 * Trim readahead window as well, if necessary:
3843 */
3844 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3845 if (ra_kb > max_sectors_kb)
3846 q->backing_dev_info.ra_pages =
3847 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3848
3849 q->max_sectors = max_sectors_kb << 1;
3850 spin_unlock_irq(q->queue_lock);
3851
3852 return ret;
3853}
3854
3855static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3856{
3857 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3858
3859 return queue_var_show(max_hw_sectors_kb, (page));
3860}
3861
3862
3863static struct queue_sysfs_entry queue_requests_entry = {
3864 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3865 .show = queue_requests_show,
3866 .store = queue_requests_store,
3867};
3868
3869static struct queue_sysfs_entry queue_ra_entry = {
3870 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3871 .show = queue_ra_show,
3872 .store = queue_ra_store,
3873};
3874
3875static struct queue_sysfs_entry queue_max_sectors_entry = {
3876 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3877 .show = queue_max_sectors_show,
3878 .store = queue_max_sectors_store,
3879};
3880
3881static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3882 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3883 .show = queue_max_hw_sectors_show,
3884};
3885
3886static struct queue_sysfs_entry queue_iosched_entry = {
3887 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3888 .show = elv_iosched_show,
3889 .store = elv_iosched_store,
3890};
3891
3892static struct attribute *default_attrs[] = {
3893 &queue_requests_entry.attr,
3894 &queue_ra_entry.attr,
3895 &queue_max_hw_sectors_entry.attr,
3896 &queue_max_sectors_entry.attr,
3897 &queue_iosched_entry.attr,
3898 NULL,
3899};
3900
3901#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3902
3903static ssize_t
3904queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3905{
3906 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003907 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3908 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003911 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003912 mutex_lock(&q->sysfs_lock);
3913 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3914 mutex_unlock(&q->sysfs_lock);
3915 return -ENOENT;
3916 }
3917 res = entry->show(q, page);
3918 mutex_unlock(&q->sysfs_lock);
3919 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920}
3921
3922static ssize_t
3923queue_attr_store(struct kobject *kobj, struct attribute *attr,
3924 const char *page, size_t length)
3925{
3926 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003927 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928
Al Viro483f4af2006-03-18 18:34:37 -05003929 ssize_t res;
3930
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003932 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003933 mutex_lock(&q->sysfs_lock);
3934 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3935 mutex_unlock(&q->sysfs_lock);
3936 return -ENOENT;
3937 }
3938 res = entry->store(q, page, length);
3939 mutex_unlock(&q->sysfs_lock);
3940 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941}
3942
3943static struct sysfs_ops queue_sysfs_ops = {
3944 .show = queue_attr_show,
3945 .store = queue_attr_store,
3946};
3947
Adrian Bunk93d17d32005-06-25 14:59:10 -07003948static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 .sysfs_ops = &queue_sysfs_ops,
3950 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05003951 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952};
3953
3954int blk_register_queue(struct gendisk *disk)
3955{
3956 int ret;
3957
3958 request_queue_t *q = disk->queue;
3959
3960 if (!q || !q->request_fn)
3961 return -ENXIO;
3962
3963 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964
Al Viro483f4af2006-03-18 18:34:37 -05003965 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 if (ret < 0)
3967 return ret;
3968
Al Viro483f4af2006-03-18 18:34:37 -05003969 kobject_uevent(&q->kobj, KOBJ_ADD);
3970
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971 ret = elv_register_queue(q);
3972 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05003973 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3974 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975 return ret;
3976 }
3977
3978 return 0;
3979}
3980
3981void blk_unregister_queue(struct gendisk *disk)
3982{
3983 request_queue_t *q = disk->queue;
3984
3985 if (q && q->request_fn) {
3986 elv_unregister_queue(q);
3987
Al Viro483f4af2006-03-18 18:34:37 -05003988 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3989 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990 kobject_put(&disk->kobj);
3991 }
3992}