blob: 6c793b196aa918da946d7378667f8a08dd7c4005 [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 */
13#include <linux/config.h>
14#include <linux/kernel.h>
15#include <linux/module.h>
16#include <linux/backing-dev.h>
17#include <linux/bio.h>
18#include <linux/blkdev.h>
19#include <linux/highmem.h>
20#include <linux/mm.h>
21#include <linux/kernel_stat.h>
22#include <linux/string.h>
23#include <linux/init.h>
24#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
25#include <linux/completion.h>
26#include <linux/slab.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
Jens Axboeff856ba2006-01-09 16:02:34 +010029#include <linux/interrupt.h>
30#include <linux/cpu.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
Tejun Heo797e7db2006-01-06 09:51:03 +0100385 if ((rq->flags & REQ_ORDERED_COLOR) ==
386 (q->orig_bar_rq->flags & REQ_ORDERED_COLOR))
387 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
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 rq_init(q, rq);
450 rq->flags = REQ_HARDBARRIER;
451 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;
474 rq_init(q, rq);
475 rq->flags = bio_data_dir(q->orig_bar_rq->bio);
476 rq->flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
477 rq->elevator_private = NULL;
478 rq->rl = NULL;
479 init_request_from_bio(rq, q->orig_bar_rq->bio);
480 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Tejun Heo797e7db2006-01-06 09:51:03 +0100482 /*
483 * Queue ordered sequence. As we stack them at the head, we
484 * need to queue in reverse order. Note that we rely on that
485 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
486 * request gets inbetween ordered sequence.
487 */
488 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
489 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
490 else
491 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Tejun Heo30e96562006-02-08 01:01:31 -0800493 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100494
495 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
496 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
497 rq = &q->pre_flush_rq;
498 } else
499 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
500
501 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
502 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
503 else
504 rq = NULL;
505
506 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
Tejun Heo797e7db2006-01-06 09:51:03 +0100509int blk_do_ordered(request_queue_t *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800511 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100512 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Tejun Heo797e7db2006-01-06 09:51:03 +0100514 if (!q->ordseq) {
515 if (!is_barrier)
516 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Tejun Heo797e7db2006-01-06 09:51:03 +0100518 if (q->next_ordered != QUEUE_ORDERED_NONE) {
519 *rqp = start_ordered(q, rq);
520 return 1;
521 } else {
522 /*
523 * This can happen when the queue switches to
524 * ORDERED_NONE while this request is on it.
525 */
526 blkdev_dequeue_request(rq);
527 end_that_request_first(rq, -EOPNOTSUPP,
528 rq->hard_nr_sectors);
529 end_that_request_last(rq, -EOPNOTSUPP);
530 *rqp = NULL;
531 return 0;
532 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800535 /*
536 * Ordered sequence in progress
537 */
538
539 /* Special requests are not subject to ordering rules. */
540 if (!blk_fs_request(rq) &&
541 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
542 return 1;
543
Tejun Heo797e7db2006-01-06 09:51:03 +0100544 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800545 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100546 if (is_barrier && rq != &q->bar_rq)
547 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800548 } else {
549 /* Ordered by draining. Wait for turn. */
550 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
551 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
552 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554
555 return 1;
556}
557
Tejun Heo797e7db2006-01-06 09:51:03 +0100558static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559{
Tejun Heo797e7db2006-01-06 09:51:03 +0100560 request_queue_t *q = bio->bi_private;
561 struct bio_vec *bvec;
562 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Tejun Heo797e7db2006-01-06 09:51:03 +0100564 /*
565 * This is dry run, restore bio_sector and size. We'll finish
566 * this request again with the original bi_end_io after an
567 * error occurs or post flush is complete.
568 */
569 q->bi_size += bytes;
570
571 if (bio->bi_size)
572 return 1;
573
574 /* Rewind bvec's */
575 bio->bi_idx = 0;
576 bio_for_each_segment(bvec, bio, i) {
577 bvec->bv_len += bvec->bv_offset;
578 bvec->bv_offset = 0;
579 }
580
581 /* Reset bio */
582 set_bit(BIO_UPTODATE, &bio->bi_flags);
583 bio->bi_size = q->bi_size;
584 bio->bi_sector -= (q->bi_size >> 9);
585 q->bi_size = 0;
586
587 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
Tejun Heo797e7db2006-01-06 09:51:03 +0100589
590static inline int ordered_bio_endio(struct request *rq, struct bio *bio,
591 unsigned int nbytes, int error)
592{
593 request_queue_t *q = rq->q;
594 bio_end_io_t *endio;
595 void *private;
596
597 if (&q->bar_rq != rq)
598 return 0;
599
600 /*
601 * Okay, this is the barrier request in progress, dry finish it.
602 */
603 if (error && !q->orderr)
604 q->orderr = error;
605
606 endio = bio->bi_end_io;
607 private = bio->bi_private;
608 bio->bi_end_io = flush_dry_bio_endio;
609 bio->bi_private = q;
610
611 bio_endio(bio, nbytes, error);
612
613 bio->bi_end_io = endio;
614 bio->bi_private = private;
615
616 return 1;
617}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619/**
620 * blk_queue_bounce_limit - set bounce buffer limit for queue
621 * @q: the request queue for the device
622 * @dma_addr: bus address limit
623 *
624 * Description:
625 * Different hardware can have different requirements as to what pages
626 * it can do I/O directly to. A low level driver can call
627 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800628 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 **/
630void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
631{
632 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800633 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Andi Kleen5ee1af92006-03-08 17:57:26 -0800635 q->bounce_gfp = GFP_NOIO;
636#if BITS_PER_LONG == 64
637 /* Assume anything <= 4GB can be handled by IOMMU.
638 Actually some IOMMUs can handle everything, but I don't
639 know of a way to test this here. */
640 if (bounce_pfn < (0xffffffff>>PAGE_SHIFT))
641 dma = 1;
642 q->bounce_pfn = max_low_pfn;
643#else
644 if (bounce_pfn < blk_max_low_pfn)
645 dma = 1;
646 q->bounce_pfn = bounce_pfn;
647#endif
648 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 init_emergency_isa_pool();
650 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800651 q->bounce_pfn = bounce_pfn;
652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
655EXPORT_SYMBOL(blk_queue_bounce_limit);
656
657/**
658 * blk_queue_max_sectors - set max sectors for a request for this queue
659 * @q: the request queue for the device
660 * @max_sectors: max sectors in the usual 512b unit
661 *
662 * Description:
663 * Enables a low level driver to set an upper limit on the size of
664 * received requests.
665 **/
Jens Axboe2cb2e142006-01-17 09:04:32 +0100666void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
668 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
669 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
670 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
671 }
672
Mike Christiedefd94b2005-12-05 02:37:06 -0600673 if (BLK_DEF_MAX_SECTORS > max_sectors)
674 q->max_hw_sectors = q->max_sectors = max_sectors;
675 else {
676 q->max_sectors = BLK_DEF_MAX_SECTORS;
677 q->max_hw_sectors = max_sectors;
678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
680
681EXPORT_SYMBOL(blk_queue_max_sectors);
682
683/**
684 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
685 * @q: the request queue for the device
686 * @max_segments: max number of segments
687 *
688 * Description:
689 * Enables a low level driver to set an upper limit on the number of
690 * physical data segments in a request. This would be the largest sized
691 * scatter list the driver could handle.
692 **/
693void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
694{
695 if (!max_segments) {
696 max_segments = 1;
697 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
698 }
699
700 q->max_phys_segments = max_segments;
701}
702
703EXPORT_SYMBOL(blk_queue_max_phys_segments);
704
705/**
706 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
707 * @q: the request queue for the device
708 * @max_segments: max number of segments
709 *
710 * Description:
711 * Enables a low level driver to set an upper limit on the number of
712 * hw data segments in a request. This would be the largest number of
713 * address/length pairs the host adapter can actually give as once
714 * to the device.
715 **/
716void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
717{
718 if (!max_segments) {
719 max_segments = 1;
720 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
721 }
722
723 q->max_hw_segments = max_segments;
724}
725
726EXPORT_SYMBOL(blk_queue_max_hw_segments);
727
728/**
729 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
730 * @q: the request queue for the device
731 * @max_size: max size of segment in bytes
732 *
733 * Description:
734 * Enables a low level driver to set an upper limit on the size of a
735 * coalesced segment
736 **/
737void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
738{
739 if (max_size < PAGE_CACHE_SIZE) {
740 max_size = PAGE_CACHE_SIZE;
741 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
742 }
743
744 q->max_segment_size = max_size;
745}
746
747EXPORT_SYMBOL(blk_queue_max_segment_size);
748
749/**
750 * blk_queue_hardsect_size - set hardware sector size for the queue
751 * @q: the request queue for the device
752 * @size: the hardware sector size, in bytes
753 *
754 * Description:
755 * This should typically be set to the lowest possible sector size
756 * that the hardware can operate on (possible without reverting to
757 * even internal read-modify-write operations). Usually the default
758 * of 512 covers most hardware.
759 **/
760void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
761{
762 q->hardsect_size = size;
763}
764
765EXPORT_SYMBOL(blk_queue_hardsect_size);
766
767/*
768 * Returns the minimum that is _not_ zero, unless both are zero.
769 */
770#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
771
772/**
773 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
774 * @t: the stacking driver (top)
775 * @b: the underlying device (bottom)
776 **/
777void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
778{
779 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600780 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
781 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
784 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
785 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
786 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
787}
788
789EXPORT_SYMBOL(blk_queue_stack_limits);
790
791/**
792 * blk_queue_segment_boundary - set boundary rules for segment merging
793 * @q: the request queue for the device
794 * @mask: the memory boundary mask
795 **/
796void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
797{
798 if (mask < PAGE_CACHE_SIZE - 1) {
799 mask = PAGE_CACHE_SIZE - 1;
800 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
801 }
802
803 q->seg_boundary_mask = mask;
804}
805
806EXPORT_SYMBOL(blk_queue_segment_boundary);
807
808/**
809 * blk_queue_dma_alignment - set dma length and memory alignment
810 * @q: the request queue for the device
811 * @mask: alignment mask
812 *
813 * description:
814 * set required memory and length aligment for direct dma transactions.
815 * this is used when buiding direct io requests for the queue.
816 *
817 **/
818void blk_queue_dma_alignment(request_queue_t *q, int mask)
819{
820 q->dma_alignment = mask;
821}
822
823EXPORT_SYMBOL(blk_queue_dma_alignment);
824
825/**
826 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 * @q: The request queue for the device
828 * @tag: The tag of the request
829 *
830 * Notes:
831 * Should be used when a device returns a tag and you want to match
832 * it with a request.
833 *
834 * no locks need be held.
835 **/
836struct request *blk_queue_find_tag(request_queue_t *q, int tag)
837{
838 struct blk_queue_tag *bqt = q->queue_tags;
839
Tejun Heoba025082005-08-05 13:28:11 -0700840 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 return NULL;
842
843 return bqt->tag_index[tag];
844}
845
846EXPORT_SYMBOL(blk_queue_find_tag);
847
848/**
849 * __blk_queue_free_tags - release tag maintenance info
850 * @q: the request queue for the device
851 *
852 * Notes:
853 * blk_cleanup_queue() will take care of calling this function, if tagging
854 * has been used. So there's no need to call this directly.
855 **/
856static void __blk_queue_free_tags(request_queue_t *q)
857{
858 struct blk_queue_tag *bqt = q->queue_tags;
859
860 if (!bqt)
861 return;
862
863 if (atomic_dec_and_test(&bqt->refcnt)) {
864 BUG_ON(bqt->busy);
865 BUG_ON(!list_empty(&bqt->busy_list));
866
867 kfree(bqt->tag_index);
868 bqt->tag_index = NULL;
869
870 kfree(bqt->tag_map);
871 bqt->tag_map = NULL;
872
873 kfree(bqt);
874 }
875
876 q->queue_tags = NULL;
877 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
878}
879
880/**
881 * blk_queue_free_tags - release tag maintenance info
882 * @q: the request queue for the device
883 *
884 * Notes:
885 * This is used to disabled tagged queuing to a device, yet leave
886 * queue in function.
887 **/
888void blk_queue_free_tags(request_queue_t *q)
889{
890 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
891}
892
893EXPORT_SYMBOL(blk_queue_free_tags);
894
895static int
896init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
897{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 struct request **tag_index;
899 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700900 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 if (depth > q->nr_requests * 2) {
903 depth = q->nr_requests * 2;
904 printk(KERN_ERR "%s: adjusted depth to %d\n",
905 __FUNCTION__, depth);
906 }
907
908 tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
909 if (!tag_index)
910 goto fail;
911
Tejun Heof7d37d02005-06-23 00:08:50 -0700912 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700913 tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (!tag_map)
915 goto fail;
916
917 memset(tag_index, 0, depth * sizeof(struct request *));
Tejun Heofa72b902005-06-23 00:08:49 -0700918 memset(tag_map, 0, nr_ulongs * sizeof(unsigned long));
Tejun Heoba025082005-08-05 13:28:11 -0700919 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 tags->tag_index = tag_index;
922 tags->tag_map = tag_map;
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 return 0;
925fail:
926 kfree(tag_index);
927 return -ENOMEM;
928}
929
930/**
931 * blk_queue_init_tags - initialize the queue tag info
932 * @q: the request queue for the device
933 * @depth: the maximum queue depth supported
934 * @tags: the tag to use
935 **/
936int blk_queue_init_tags(request_queue_t *q, int depth,
937 struct blk_queue_tag *tags)
938{
939 int rc;
940
941 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
942
943 if (!tags && !q->queue_tags) {
944 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
945 if (!tags)
946 goto fail;
947
948 if (init_tag_map(q, tags, depth))
949 goto fail;
950
951 INIT_LIST_HEAD(&tags->busy_list);
952 tags->busy = 0;
953 atomic_set(&tags->refcnt, 1);
954 } else if (q->queue_tags) {
955 if ((rc = blk_queue_resize_tags(q, depth)))
956 return rc;
957 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
958 return 0;
959 } else
960 atomic_inc(&tags->refcnt);
961
962 /*
963 * assign it, all done
964 */
965 q->queue_tags = tags;
966 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
967 return 0;
968fail:
969 kfree(tags);
970 return -ENOMEM;
971}
972
973EXPORT_SYMBOL(blk_queue_init_tags);
974
975/**
976 * blk_queue_resize_tags - change the queueing depth
977 * @q: the request queue for the device
978 * @new_depth: the new max command queueing depth
979 *
980 * Notes:
981 * Must be called with the queue lock held.
982 **/
983int blk_queue_resize_tags(request_queue_t *q, int new_depth)
984{
985 struct blk_queue_tag *bqt = q->queue_tags;
986 struct request **tag_index;
987 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700988 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 if (!bqt)
991 return -ENXIO;
992
993 /*
Tejun Heoba025082005-08-05 13:28:11 -0700994 * if we already have large enough real_max_depth. just
995 * adjust max_depth. *NOTE* as requests with tag value
996 * between new_depth and real_max_depth can be in-flight, tag
997 * map can not be shrunk blindly here.
998 */
999 if (new_depth <= bqt->real_max_depth) {
1000 bqt->max_depth = new_depth;
1001 return 0;
1002 }
1003
1004 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 * save the old state info, so we can copy it back
1006 */
1007 tag_index = bqt->tag_index;
1008 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001009 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 if (init_tag_map(q, bqt, new_depth))
1012 return -ENOMEM;
1013
1014 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001015 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001016 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 kfree(tag_index);
1019 kfree(tag_map);
1020 return 0;
1021}
1022
1023EXPORT_SYMBOL(blk_queue_resize_tags);
1024
1025/**
1026 * blk_queue_end_tag - end tag operations for a request
1027 * @q: the request queue for the device
1028 * @rq: the request that has completed
1029 *
1030 * Description:
1031 * Typically called when end_that_request_first() returns 0, meaning
1032 * all transfers have been done for a request. It's important to call
1033 * this function before end_that_request_last(), as that will put the
1034 * request back on the free list thus corrupting the internal tag list.
1035 *
1036 * Notes:
1037 * queue lock must be held.
1038 **/
1039void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1040{
1041 struct blk_queue_tag *bqt = q->queue_tags;
1042 int tag = rq->tag;
1043
1044 BUG_ON(tag == -1);
1045
Tejun Heoba025082005-08-05 13:28:11 -07001046 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001047 /*
1048 * This can happen after tag depth has been reduced.
1049 * FIXME: how about a warning or info message here?
1050 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return;
1052
1053 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001054 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1055 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 return;
1057 }
1058
1059 list_del_init(&rq->queuelist);
1060 rq->flags &= ~REQ_QUEUED;
1061 rq->tag = -1;
1062
1063 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001064 printk(KERN_ERR "%s: tag %d is missing\n",
1065 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067 bqt->tag_index[tag] = NULL;
1068 bqt->busy--;
1069}
1070
1071EXPORT_SYMBOL(blk_queue_end_tag);
1072
1073/**
1074 * blk_queue_start_tag - find a free tag and assign it
1075 * @q: the request queue for the device
1076 * @rq: the block request that needs tagging
1077 *
1078 * Description:
1079 * This can either be used as a stand-alone helper, or possibly be
1080 * assigned as the queue &prep_rq_fn (in which case &struct request
1081 * automagically gets a tag assigned). Note that this function
1082 * assumes that any type of request can be queued! if this is not
1083 * true for your device, you must check the request type before
1084 * calling this function. The request will also be removed from
1085 * the request queue, so it's the drivers responsibility to readd
1086 * it if it should need to be restarted for some reason.
1087 *
1088 * Notes:
1089 * queue lock must be held.
1090 **/
1091int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1092{
1093 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001094 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 if (unlikely((rq->flags & REQ_QUEUED))) {
1097 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001098 "%s: request %p for device [%s] already tagged %d",
1099 __FUNCTION__, rq,
1100 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 BUG();
1102 }
1103
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001104 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1105 if (tag >= bqt->max_depth)
1106 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 __set_bit(tag, bqt->tag_map);
1109
1110 rq->flags |= REQ_QUEUED;
1111 rq->tag = tag;
1112 bqt->tag_index[tag] = rq;
1113 blkdev_dequeue_request(rq);
1114 list_add(&rq->queuelist, &bqt->busy_list);
1115 bqt->busy++;
1116 return 0;
1117}
1118
1119EXPORT_SYMBOL(blk_queue_start_tag);
1120
1121/**
1122 * blk_queue_invalidate_tags - invalidate all pending tags
1123 * @q: the request queue for the device
1124 *
1125 * Description:
1126 * Hardware conditions may dictate a need to stop all pending requests.
1127 * In this case, we will safely clear the block side of the tag queue and
1128 * readd all requests to the request queue in the right order.
1129 *
1130 * Notes:
1131 * queue lock must be held.
1132 **/
1133void blk_queue_invalidate_tags(request_queue_t *q)
1134{
1135 struct blk_queue_tag *bqt = q->queue_tags;
1136 struct list_head *tmp, *n;
1137 struct request *rq;
1138
1139 list_for_each_safe(tmp, n, &bqt->busy_list) {
1140 rq = list_entry_rq(tmp);
1141
1142 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001143 printk(KERN_ERR
1144 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 list_del_init(&rq->queuelist);
1146 rq->flags &= ~REQ_QUEUED;
1147 } else
1148 blk_queue_end_tag(q, rq);
1149
1150 rq->flags &= ~REQ_STARTED;
1151 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1152 }
1153}
1154
1155EXPORT_SYMBOL(blk_queue_invalidate_tags);
1156
Arjan van de Ven64100092006-01-06 09:46:02 +01001157static const char * const rq_flags[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 "REQ_RW",
1159 "REQ_FAILFAST",
Tejun Heo8922e162005-10-20 16:23:44 +02001160 "REQ_SORTED",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 "REQ_SOFTBARRIER",
1162 "REQ_HARDBARRIER",
Tejun Heo797e7db2006-01-06 09:51:03 +01001163 "REQ_FUA",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 "REQ_CMD",
1165 "REQ_NOMERGE",
1166 "REQ_STARTED",
1167 "REQ_DONTPREP",
1168 "REQ_QUEUED",
Tejun Heocb98fc82005-10-28 08:29:39 +02001169 "REQ_ELVPRIV",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 "REQ_PC",
1171 "REQ_BLOCK_PC",
1172 "REQ_SENSE",
1173 "REQ_FAILED",
1174 "REQ_QUIET",
1175 "REQ_SPECIAL",
1176 "REQ_DRIVE_CMD",
1177 "REQ_DRIVE_TASK",
1178 "REQ_DRIVE_TASKFILE",
1179 "REQ_PREEMPT",
1180 "REQ_PM_SUSPEND",
1181 "REQ_PM_RESUME",
1182 "REQ_PM_SHUTDOWN",
Tejun Heo797e7db2006-01-06 09:51:03 +01001183 "REQ_ORDERED_COLOR",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184};
1185
1186void blk_dump_rq_flags(struct request *rq, char *msg)
1187{
1188 int bit;
1189
1190 printk("%s: dev %s: flags = ", msg,
1191 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1192 bit = 0;
1193 do {
1194 if (rq->flags & (1 << bit))
1195 printk("%s ", rq_flags[bit]);
1196 bit++;
1197 } while (bit < __REQ_NR_BITS);
1198
1199 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1200 rq->nr_sectors,
1201 rq->current_nr_sectors);
1202 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1203
1204 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1205 printk("cdb: ");
1206 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1207 printk("%02x ", rq->cmd[bit]);
1208 printk("\n");
1209 }
1210}
1211
1212EXPORT_SYMBOL(blk_dump_rq_flags);
1213
1214void blk_recount_segments(request_queue_t *q, struct bio *bio)
1215{
1216 struct bio_vec *bv, *bvprv = NULL;
1217 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1218 int high, highprv = 1;
1219
1220 if (unlikely(!bio->bi_io_vec))
1221 return;
1222
1223 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1224 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1225 bio_for_each_segment(bv, bio, i) {
1226 /*
1227 * the trick here is making sure that a high page is never
1228 * considered part of another segment, since that might
1229 * change with the bounce page.
1230 */
1231 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1232 if (high || highprv)
1233 goto new_hw_segment;
1234 if (cluster) {
1235 if (seg_size + bv->bv_len > q->max_segment_size)
1236 goto new_segment;
1237 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1238 goto new_segment;
1239 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1240 goto new_segment;
1241 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1242 goto new_hw_segment;
1243
1244 seg_size += bv->bv_len;
1245 hw_seg_size += bv->bv_len;
1246 bvprv = bv;
1247 continue;
1248 }
1249new_segment:
1250 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1251 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1252 hw_seg_size += bv->bv_len;
1253 } else {
1254new_hw_segment:
1255 if (hw_seg_size > bio->bi_hw_front_size)
1256 bio->bi_hw_front_size = hw_seg_size;
1257 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1258 nr_hw_segs++;
1259 }
1260
1261 nr_phys_segs++;
1262 bvprv = bv;
1263 seg_size = bv->bv_len;
1264 highprv = high;
1265 }
1266 if (hw_seg_size > bio->bi_hw_back_size)
1267 bio->bi_hw_back_size = hw_seg_size;
1268 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1269 bio->bi_hw_front_size = hw_seg_size;
1270 bio->bi_phys_segments = nr_phys_segs;
1271 bio->bi_hw_segments = nr_hw_segs;
1272 bio->bi_flags |= (1 << BIO_SEG_VALID);
1273}
1274
1275
Adrian Bunk93d17d32005-06-25 14:59:10 -07001276static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 struct bio *nxt)
1278{
1279 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1280 return 0;
1281
1282 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1283 return 0;
1284 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1285 return 0;
1286
1287 /*
1288 * bio and nxt are contigous in memory, check if the queue allows
1289 * these two to be merged into one
1290 */
1291 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1292 return 1;
1293
1294 return 0;
1295}
1296
Adrian Bunk93d17d32005-06-25 14:59:10 -07001297static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 struct bio *nxt)
1299{
1300 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1301 blk_recount_segments(q, bio);
1302 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1303 blk_recount_segments(q, nxt);
1304 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1305 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1306 return 0;
1307 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1308 return 0;
1309
1310 return 1;
1311}
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313/*
1314 * map a request to scatterlist, return number of sg entries setup. Caller
1315 * must make sure sg can hold rq->nr_phys_segments entries
1316 */
1317int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1318{
1319 struct bio_vec *bvec, *bvprv;
1320 struct bio *bio;
1321 int nsegs, i, cluster;
1322
1323 nsegs = 0;
1324 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1325
1326 /*
1327 * for each bio in rq
1328 */
1329 bvprv = NULL;
1330 rq_for_each_bio(bio, rq) {
1331 /*
1332 * for each segment in bio
1333 */
1334 bio_for_each_segment(bvec, bio, i) {
1335 int nbytes = bvec->bv_len;
1336
1337 if (bvprv && cluster) {
1338 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1339 goto new_segment;
1340
1341 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1342 goto new_segment;
1343 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1344 goto new_segment;
1345
1346 sg[nsegs - 1].length += nbytes;
1347 } else {
1348new_segment:
1349 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1350 sg[nsegs].page = bvec->bv_page;
1351 sg[nsegs].length = nbytes;
1352 sg[nsegs].offset = bvec->bv_offset;
1353
1354 nsegs++;
1355 }
1356 bvprv = bvec;
1357 } /* segments in bio */
1358 } /* bios in rq */
1359
1360 return nsegs;
1361}
1362
1363EXPORT_SYMBOL(blk_rq_map_sg);
1364
1365/*
1366 * the standard queue merge functions, can be overridden with device
1367 * specific ones if so desired
1368 */
1369
1370static inline int ll_new_mergeable(request_queue_t *q,
1371 struct request *req,
1372 struct bio *bio)
1373{
1374 int nr_phys_segs = bio_phys_segments(q, bio);
1375
1376 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1377 req->flags |= REQ_NOMERGE;
1378 if (req == q->last_merge)
1379 q->last_merge = NULL;
1380 return 0;
1381 }
1382
1383 /*
1384 * A hw segment is just getting larger, bump just the phys
1385 * counter.
1386 */
1387 req->nr_phys_segments += nr_phys_segs;
1388 return 1;
1389}
1390
1391static inline int ll_new_hw_segment(request_queue_t *q,
1392 struct request *req,
1393 struct bio *bio)
1394{
1395 int nr_hw_segs = bio_hw_segments(q, bio);
1396 int nr_phys_segs = bio_phys_segments(q, bio);
1397
1398 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1399 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1400 req->flags |= REQ_NOMERGE;
1401 if (req == q->last_merge)
1402 q->last_merge = NULL;
1403 return 0;
1404 }
1405
1406 /*
1407 * This will form the start of a new hw segment. Bump both
1408 * counters.
1409 */
1410 req->nr_hw_segments += nr_hw_segs;
1411 req->nr_phys_segments += nr_phys_segs;
1412 return 1;
1413}
1414
1415static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1416 struct bio *bio)
1417{
Mike Christiedefd94b2005-12-05 02:37:06 -06001418 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 int len;
1420
Mike Christiedefd94b2005-12-05 02:37:06 -06001421 if (unlikely(blk_pc_request(req)))
1422 max_sectors = q->max_hw_sectors;
1423 else
1424 max_sectors = q->max_sectors;
1425
1426 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 req->flags |= REQ_NOMERGE;
1428 if (req == q->last_merge)
1429 q->last_merge = NULL;
1430 return 0;
1431 }
1432 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1433 blk_recount_segments(q, req->biotail);
1434 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1435 blk_recount_segments(q, bio);
1436 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1437 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1438 !BIOVEC_VIRT_OVERSIZE(len)) {
1439 int mergeable = ll_new_mergeable(q, req, bio);
1440
1441 if (mergeable) {
1442 if (req->nr_hw_segments == 1)
1443 req->bio->bi_hw_front_size = len;
1444 if (bio->bi_hw_segments == 1)
1445 bio->bi_hw_back_size = len;
1446 }
1447 return mergeable;
1448 }
1449
1450 return ll_new_hw_segment(q, req, bio);
1451}
1452
1453static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1454 struct bio *bio)
1455{
Mike Christiedefd94b2005-12-05 02:37:06 -06001456 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 int len;
1458
Mike Christiedefd94b2005-12-05 02:37:06 -06001459 if (unlikely(blk_pc_request(req)))
1460 max_sectors = q->max_hw_sectors;
1461 else
1462 max_sectors = q->max_sectors;
1463
1464
1465 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 req->flags |= REQ_NOMERGE;
1467 if (req == q->last_merge)
1468 q->last_merge = NULL;
1469 return 0;
1470 }
1471 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1472 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1473 blk_recount_segments(q, bio);
1474 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1475 blk_recount_segments(q, req->bio);
1476 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1477 !BIOVEC_VIRT_OVERSIZE(len)) {
1478 int mergeable = ll_new_mergeable(q, req, bio);
1479
1480 if (mergeable) {
1481 if (bio->bi_hw_segments == 1)
1482 bio->bi_hw_front_size = len;
1483 if (req->nr_hw_segments == 1)
1484 req->biotail->bi_hw_back_size = len;
1485 }
1486 return mergeable;
1487 }
1488
1489 return ll_new_hw_segment(q, req, bio);
1490}
1491
1492static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1493 struct request *next)
1494{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001495 int total_phys_segments;
1496 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 /*
1499 * First check if the either of the requests are re-queued
1500 * requests. Can't merge them if they are.
1501 */
1502 if (req->special || next->special)
1503 return 0;
1504
1505 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001506 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 */
1508 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1509 return 0;
1510
1511 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1512 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1513 total_phys_segments--;
1514
1515 if (total_phys_segments > q->max_phys_segments)
1516 return 0;
1517
1518 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1519 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1520 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1521 /*
1522 * propagate the combined length to the end of the requests
1523 */
1524 if (req->nr_hw_segments == 1)
1525 req->bio->bi_hw_front_size = len;
1526 if (next->nr_hw_segments == 1)
1527 next->biotail->bi_hw_back_size = len;
1528 total_hw_segments--;
1529 }
1530
1531 if (total_hw_segments > q->max_hw_segments)
1532 return 0;
1533
1534 /* Merge is OK... */
1535 req->nr_phys_segments = total_phys_segments;
1536 req->nr_hw_segments = total_hw_segments;
1537 return 1;
1538}
1539
1540/*
1541 * "plug" the device if there are no outstanding requests: this will
1542 * force the transfer to start only after we have put all the requests
1543 * on the list.
1544 *
1545 * This is called with interrupts off and no requests on the queue and
1546 * with the queue lock held.
1547 */
1548void blk_plug_device(request_queue_t *q)
1549{
1550 WARN_ON(!irqs_disabled());
1551
1552 /*
1553 * don't plug a stopped queue, it must be paired with blk_start_queue()
1554 * which will restart the queueing
1555 */
1556 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1557 return;
1558
1559 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1560 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1561}
1562
1563EXPORT_SYMBOL(blk_plug_device);
1564
1565/*
1566 * remove the queue from the plugged list, if present. called with
1567 * queue lock held and interrupts disabled.
1568 */
1569int blk_remove_plug(request_queue_t *q)
1570{
1571 WARN_ON(!irqs_disabled());
1572
1573 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1574 return 0;
1575
1576 del_timer(&q->unplug_timer);
1577 return 1;
1578}
1579
1580EXPORT_SYMBOL(blk_remove_plug);
1581
1582/*
1583 * remove the plug and let it rip..
1584 */
1585void __generic_unplug_device(request_queue_t *q)
1586{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001587 if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 return;
1589
1590 if (!blk_remove_plug(q))
1591 return;
1592
Jens Axboe22e2c502005-06-27 10:55:12 +02001593 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594}
1595EXPORT_SYMBOL(__generic_unplug_device);
1596
1597/**
1598 * generic_unplug_device - fire a request queue
1599 * @q: The &request_queue_t in question
1600 *
1601 * Description:
1602 * Linux uses plugging to build bigger requests queues before letting
1603 * the device have at them. If a queue is plugged, the I/O scheduler
1604 * is still adding and merging requests on the queue. Once the queue
1605 * gets unplugged, the request_fn defined for the queue is invoked and
1606 * transfers started.
1607 **/
1608void generic_unplug_device(request_queue_t *q)
1609{
1610 spin_lock_irq(q->queue_lock);
1611 __generic_unplug_device(q);
1612 spin_unlock_irq(q->queue_lock);
1613}
1614EXPORT_SYMBOL(generic_unplug_device);
1615
1616static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1617 struct page *page)
1618{
1619 request_queue_t *q = bdi->unplug_io_data;
1620
1621 /*
1622 * devices don't necessarily have an ->unplug_fn defined
1623 */
1624 if (q->unplug_fn)
1625 q->unplug_fn(q);
1626}
1627
1628static void blk_unplug_work(void *data)
1629{
1630 request_queue_t *q = data;
1631
1632 q->unplug_fn(q);
1633}
1634
1635static void blk_unplug_timeout(unsigned long data)
1636{
1637 request_queue_t *q = (request_queue_t *)data;
1638
1639 kblockd_schedule_work(&q->unplug_work);
1640}
1641
1642/**
1643 * blk_start_queue - restart a previously stopped queue
1644 * @q: The &request_queue_t in question
1645 *
1646 * Description:
1647 * blk_start_queue() will clear the stop flag on the queue, and call
1648 * the request_fn for the queue if it was in a stopped state when
1649 * entered. Also see blk_stop_queue(). Queue lock must be held.
1650 **/
1651void blk_start_queue(request_queue_t *q)
1652{
1653 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1654
1655 /*
1656 * one level of recursion is ok and is much faster than kicking
1657 * the unplug handling
1658 */
1659 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1660 q->request_fn(q);
1661 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1662 } else {
1663 blk_plug_device(q);
1664 kblockd_schedule_work(&q->unplug_work);
1665 }
1666}
1667
1668EXPORT_SYMBOL(blk_start_queue);
1669
1670/**
1671 * blk_stop_queue - stop a queue
1672 * @q: The &request_queue_t in question
1673 *
1674 * Description:
1675 * The Linux block layer assumes that a block driver will consume all
1676 * entries on the request queue when the request_fn strategy is called.
1677 * Often this will not happen, because of hardware limitations (queue
1678 * depth settings). If a device driver gets a 'queue full' response,
1679 * or if it simply chooses not to queue more I/O at one point, it can
1680 * call this function to prevent the request_fn from being called until
1681 * the driver has signalled it's ready to go again. This happens by calling
1682 * blk_start_queue() to restart queue operations. Queue lock must be held.
1683 **/
1684void blk_stop_queue(request_queue_t *q)
1685{
1686 blk_remove_plug(q);
1687 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1688}
1689EXPORT_SYMBOL(blk_stop_queue);
1690
1691/**
1692 * blk_sync_queue - cancel any pending callbacks on a queue
1693 * @q: the queue
1694 *
1695 * Description:
1696 * The block layer may perform asynchronous callback activity
1697 * on a queue, such as calling the unplug function after a timeout.
1698 * A block device may call blk_sync_queue to ensure that any
1699 * such activity is cancelled, thus allowing it to release resources
1700 * the the callbacks might use. The caller must already have made sure
1701 * that its ->make_request_fn will not re-add plugging prior to calling
1702 * this function.
1703 *
1704 */
1705void blk_sync_queue(struct request_queue *q)
1706{
1707 del_timer_sync(&q->unplug_timer);
1708 kblockd_flush();
1709}
1710EXPORT_SYMBOL(blk_sync_queue);
1711
1712/**
1713 * blk_run_queue - run a single device queue
1714 * @q: The queue to run
1715 */
1716void blk_run_queue(struct request_queue *q)
1717{
1718 unsigned long flags;
1719
1720 spin_lock_irqsave(q->queue_lock, flags);
1721 blk_remove_plug(q);
Ken Chena2997382005-04-16 15:25:43 -07001722 if (!elv_queue_empty(q))
1723 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 spin_unlock_irqrestore(q->queue_lock, flags);
1725}
1726EXPORT_SYMBOL(blk_run_queue);
1727
1728/**
1729 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1730 * @q: the request queue to be released
1731 *
1732 * Description:
1733 * blk_cleanup_queue is the pair to blk_init_queue() or
1734 * blk_queue_make_request(). It should be called when a request queue is
1735 * being released; typically when a block device is being de-registered.
1736 * Currently, its primary task it to free all the &struct request
1737 * structures that were allocated to the queue and the queue itself.
1738 *
1739 * Caveat:
1740 * Hopefully the low level driver will have finished any
1741 * outstanding requests first...
1742 **/
Al Viro483f4af2006-03-18 18:34:37 -05001743static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
Al Viro483f4af2006-03-18 18:34:37 -05001745 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 struct request_list *rl = &q->rq;
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 blk_sync_queue(q);
1749
1750 if (rl->rq_pool)
1751 mempool_destroy(rl->rq_pool);
1752
1753 if (q->queue_tags)
1754 __blk_queue_free_tags(q);
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 kmem_cache_free(requestq_cachep, q);
1757}
1758
Al Viro483f4af2006-03-18 18:34:37 -05001759void blk_put_queue(request_queue_t *q)
1760{
1761 kobject_put(&q->kobj);
1762}
1763EXPORT_SYMBOL(blk_put_queue);
1764
1765void blk_cleanup_queue(request_queue_t * q)
1766{
1767 mutex_lock(&q->sysfs_lock);
1768 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1769 mutex_unlock(&q->sysfs_lock);
1770
1771 if (q->elevator)
1772 elevator_exit(q->elevator);
1773
1774 blk_put_queue(q);
1775}
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777EXPORT_SYMBOL(blk_cleanup_queue);
1778
1779static int blk_init_free_list(request_queue_t *q)
1780{
1781 struct request_list *rl = &q->rq;
1782
1783 rl->count[READ] = rl->count[WRITE] = 0;
1784 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001785 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 init_waitqueue_head(&rl->wait[READ]);
1787 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Christoph Lameter19460892005-06-23 00:08:19 -07001789 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1790 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
1792 if (!rl->rq_pool)
1793 return -ENOMEM;
1794
1795 return 0;
1796}
1797
Al Viro8267e262005-10-21 03:20:53 -04001798request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799{
Christoph Lameter19460892005-06-23 00:08:19 -07001800 return blk_alloc_queue_node(gfp_mask, -1);
1801}
1802EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Al Viro483f4af2006-03-18 18:34:37 -05001804static struct kobj_type queue_ktype;
1805
Al Viro8267e262005-10-21 03:20:53 -04001806request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001807{
1808 request_queue_t *q;
1809
1810 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (!q)
1812 return NULL;
1813
1814 memset(q, 0, sizeof(*q));
1815 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001816
1817 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1818 q->kobj.ktype = &queue_ktype;
1819 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1822 q->backing_dev_info.unplug_io_data = q;
1823
Al Viro483f4af2006-03-18 18:34:37 -05001824 mutex_init(&q->sysfs_lock);
1825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 return q;
1827}
Christoph Lameter19460892005-06-23 00:08:19 -07001828EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830/**
1831 * blk_init_queue - prepare a request queue for use with a block device
1832 * @rfn: The function to be called to process requests that have been
1833 * placed on the queue.
1834 * @lock: Request queue spin lock
1835 *
1836 * Description:
1837 * If a block device wishes to use the standard request handling procedures,
1838 * which sorts requests and coalesces adjacent requests, then it must
1839 * call blk_init_queue(). The function @rfn will be called when there
1840 * are requests on the queue that need to be processed. If the device
1841 * supports plugging, then @rfn may not be called immediately when requests
1842 * are available on the queue, but may be called at some time later instead.
1843 * Plugged queues are generally unplugged when a buffer belonging to one
1844 * of the requests on the queue is needed, or due to memory pressure.
1845 *
1846 * @rfn is not required, or even expected, to remove all requests off the
1847 * queue, but only as many as it can handle at a time. If it does leave
1848 * requests on the queue, it is responsible for arranging that the requests
1849 * get dealt with eventually.
1850 *
1851 * The queue spin lock must be held while manipulating the requests on the
1852 * request queue.
1853 *
1854 * Function returns a pointer to the initialized request queue, or NULL if
1855 * it didn't succeed.
1856 *
1857 * Note:
1858 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1859 * when the block device is deactivated (such as at module unload).
1860 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1863{
Christoph Lameter19460892005-06-23 00:08:19 -07001864 return blk_init_queue_node(rfn, lock, -1);
1865}
1866EXPORT_SYMBOL(blk_init_queue);
1867
1868request_queue_t *
1869blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1870{
1871 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872
1873 if (!q)
1874 return NULL;
1875
Christoph Lameter19460892005-06-23 00:08:19 -07001876 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001877 if (blk_init_free_list(q)) {
1878 kmem_cache_free(requestq_cachep, q);
1879 return NULL;
1880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
152587d2005-04-12 16:22:06 -05001882 /*
1883 * if caller didn't supply a lock, they get per-queue locking with
1884 * our embedded lock
1885 */
1886 if (!lock) {
1887 spin_lock_init(&q->__queue_lock);
1888 lock = &q->__queue_lock;
1889 }
1890
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 q->request_fn = rfn;
1892 q->back_merge_fn = ll_back_merge_fn;
1893 q->front_merge_fn = ll_front_merge_fn;
1894 q->merge_requests_fn = ll_merge_requests_fn;
1895 q->prep_rq_fn = NULL;
1896 q->unplug_fn = generic_unplug_device;
1897 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1898 q->queue_lock = lock;
1899
1900 blk_queue_segment_boundary(q, 0xffffffff);
1901
1902 blk_queue_make_request(q, __make_request);
1903 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1904
1905 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1906 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1907
1908 /*
1909 * all done
1910 */
1911 if (!elevator_init(q, NULL)) {
1912 blk_queue_congestion_threshold(q);
1913 return q;
1914 }
1915
Al Viro8669aaf2006-03-18 13:50:00 -05001916 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 return NULL;
1918}
Christoph Lameter19460892005-06-23 00:08:19 -07001919EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921int blk_get_queue(request_queue_t *q)
1922{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001923 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001924 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 return 0;
1926 }
1927
1928 return 1;
1929}
1930
1931EXPORT_SYMBOL(blk_get_queue);
1932
1933static inline void blk_free_request(request_queue_t *q, struct request *rq)
1934{
Tejun Heocb98fc82005-10-28 08:29:39 +02001935 if (rq->flags & REQ_ELVPRIV)
1936 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 mempool_free(rq, q->rq.rq_pool);
1938}
1939
Jens Axboe22e2c502005-06-27 10:55:12 +02001940static inline struct request *
Tejun Heocb98fc82005-10-28 08:29:39 +02001941blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
Linus Torvalds5dd96242005-10-28 08:56:34 -07001942 int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943{
1944 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1945
1946 if (!rq)
1947 return NULL;
1948
1949 /*
1950 * first three bits are identical in rq->flags and bio->bi_rw,
1951 * see bio.h and blkdev.h
1952 */
1953 rq->flags = rw;
1954
Tejun Heocb98fc82005-10-28 08:29:39 +02001955 if (priv) {
1956 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
1957 mempool_free(rq, q->rq.rq_pool);
1958 return NULL;
1959 }
1960 rq->flags |= REQ_ELVPRIV;
1961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
Tejun Heocb98fc82005-10-28 08:29:39 +02001963 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964}
1965
1966/*
1967 * ioc_batching returns true if the ioc is a valid batching request and
1968 * should be given priority access to a request.
1969 */
1970static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1971{
1972 if (!ioc)
1973 return 0;
1974
1975 /*
1976 * Make sure the process is able to allocate at least 1 request
1977 * even if the batch times out, otherwise we could theoretically
1978 * lose wakeups.
1979 */
1980 return ioc->nr_batch_requests == q->nr_batching ||
1981 (ioc->nr_batch_requests > 0
1982 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1983}
1984
1985/*
1986 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1987 * will cause the process to be a "batcher" on all queues in the system. This
1988 * is the behaviour we want though - once it gets a wakeup it should be given
1989 * a nice run.
1990 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07001991static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992{
1993 if (!ioc || ioc_batching(q, ioc))
1994 return;
1995
1996 ioc->nr_batch_requests = q->nr_batching;
1997 ioc->last_waited = jiffies;
1998}
1999
2000static void __freed_request(request_queue_t *q, int rw)
2001{
2002 struct request_list *rl = &q->rq;
2003
2004 if (rl->count[rw] < queue_congestion_off_threshold(q))
2005 clear_queue_congested(q, rw);
2006
2007 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 if (waitqueue_active(&rl->wait[rw]))
2009 wake_up(&rl->wait[rw]);
2010
2011 blk_clear_queue_full(q, rw);
2012 }
2013}
2014
2015/*
2016 * A request has just been released. Account for it, update the full and
2017 * congestion status, wake up any waiters. Called under q->queue_lock.
2018 */
Tejun Heocb98fc82005-10-28 08:29:39 +02002019static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020{
2021 struct request_list *rl = &q->rq;
2022
2023 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002024 if (priv)
2025 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027 __freed_request(q, rw);
2028
2029 if (unlikely(rl->starved[rw ^ 1]))
2030 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031}
2032
2033#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2034/*
Nick Piggind6344532005-06-28 20:45:14 -07002035 * Get a free request, queue_lock must be held.
2036 * Returns NULL on failure, with queue_lock held.
2037 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002039static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04002040 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
2042 struct request *rq = NULL;
2043 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002044 struct io_context *ioc = NULL;
2045 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002047 may_queue = elv_may_queue(q, rw, bio);
2048 if (may_queue == ELV_MQUEUE_NO)
2049 goto rq_starved;
2050
2051 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2052 if (rl->count[rw]+1 >= q->nr_requests) {
2053 ioc = current_io_context(GFP_ATOMIC);
2054 /*
2055 * The queue will fill after this allocation, so set
2056 * it as full, and mark this process as "batching".
2057 * This process will be allowed to complete a batch of
2058 * requests, others will be blocked.
2059 */
2060 if (!blk_queue_full(q, rw)) {
2061 ioc_set_batching(q, ioc);
2062 blk_set_queue_full(q, rw);
2063 } else {
2064 if (may_queue != ELV_MQUEUE_MUST
2065 && !ioc_batching(q, ioc)) {
2066 /*
2067 * The queue is full and the allocating
2068 * process is not a "batcher", and not
2069 * exempted by the IO scheduler
2070 */
2071 goto out;
2072 }
2073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 }
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002075 set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 }
2077
Jens Axboe082cf692005-06-28 16:35:11 +02002078 /*
2079 * Only allow batching queuers to allocate up to 50% over the defined
2080 * limit of requests, otherwise we could have thousands of requests
2081 * allocated with any setting of ->nr_requests
2082 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002083 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002084 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 rl->count[rw]++;
2087 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002088
Jens Axboe64521d12005-10-28 08:30:39 +02002089 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002090 if (priv)
2091 rl->elvpriv++;
2092
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 spin_unlock_irq(q->queue_lock);
2094
Tejun Heocb98fc82005-10-28 08:29:39 +02002095 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002096 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 /*
2098 * Allocation failed presumably due to memory. Undo anything
2099 * we might have messed up.
2100 *
2101 * Allocating task should really be put onto the front of the
2102 * wait queue, but this is pretty rare.
2103 */
2104 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002105 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
2107 /*
2108 * in the very unlikely event that allocation failed and no
2109 * requests for this direction was pending, mark us starved
2110 * so that freeing of a request in the other direction will
2111 * notice us. another possible fix would be to split the
2112 * rq mempool into READ and WRITE
2113 */
2114rq_starved:
2115 if (unlikely(rl->count[rw] == 0))
2116 rl->starved[rw] = 1;
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 goto out;
2119 }
2120
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002121 /*
2122 * ioc may be NULL here, and ioc_batching will be false. That's
2123 * OK, if the queue is under the request limit then requests need
2124 * not count toward the nr_batch_requests limit. There will always
2125 * be some limit enforced by BLK_BATCH_TIME.
2126 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 if (ioc_batching(q, ioc))
2128 ioc->nr_batch_requests--;
2129
2130 rq_init(q, rq);
2131 rq->rl = rl;
2132out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 return rq;
2134}
2135
2136/*
2137 * No available requests for this queue, unplug the device and wait for some
2138 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002139 *
2140 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002142static struct request *get_request_wait(request_queue_t *q, int rw,
2143 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 struct request *rq;
2146
Nick Piggin450991b2005-06-28 20:45:13 -07002147 rq = get_request(q, rw, bio, GFP_NOIO);
2148 while (!rq) {
2149 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 struct request_list *rl = &q->rq;
2151
2152 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2153 TASK_UNINTERRUPTIBLE);
2154
Jens Axboe22e2c502005-06-27 10:55:12 +02002155 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
2157 if (!rq) {
2158 struct io_context *ioc;
2159
Nick Piggind6344532005-06-28 20:45:14 -07002160 __generic_unplug_device(q);
2161 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 io_schedule();
2163
2164 /*
2165 * After sleeping, we become a "batching" process and
2166 * will be able to allocate at least one request, and
2167 * up to a big batch of them for a small period time.
2168 * See ioc_batching, ioc_set_batching
2169 */
Nick Pigginfb3cc432005-06-28 20:45:15 -07002170 ioc = current_io_context(GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002172
2173 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 }
2175 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
2178 return rq;
2179}
2180
Al Viro8267e262005-10-21 03:20:53 -04002181struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182{
2183 struct request *rq;
2184
2185 BUG_ON(rw != READ && rw != WRITE);
2186
Nick Piggind6344532005-06-28 20:45:14 -07002187 spin_lock_irq(q->queue_lock);
2188 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002189 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002190 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002191 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002192 if (!rq)
2193 spin_unlock_irq(q->queue_lock);
2194 }
2195 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
2197 return rq;
2198}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199EXPORT_SYMBOL(blk_get_request);
2200
2201/**
2202 * blk_requeue_request - put a request back on queue
2203 * @q: request queue where request should be inserted
2204 * @rq: request to be inserted
2205 *
2206 * Description:
2207 * Drivers often keep queueing requests until the hardware cannot accept
2208 * more, when that condition happens we need to put the request back
2209 * on the queue. Must be called with queue lock held.
2210 */
2211void blk_requeue_request(request_queue_t *q, struct request *rq)
2212{
2213 if (blk_rq_tagged(rq))
2214 blk_queue_end_tag(q, rq);
2215
2216 elv_requeue_request(q, rq);
2217}
2218
2219EXPORT_SYMBOL(blk_requeue_request);
2220
2221/**
2222 * blk_insert_request - insert a special request in to a request queue
2223 * @q: request queue where request should be inserted
2224 * @rq: request to be inserted
2225 * @at_head: insert request at head or tail of queue
2226 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 *
2228 * Description:
2229 * Many block devices need to execute commands asynchronously, so they don't
2230 * block the whole kernel from preemption during request execution. This is
2231 * accomplished normally by inserting aritficial requests tagged as
2232 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2233 * scheduled for actual execution by the request queue.
2234 *
2235 * We have the option of inserting the head or the tail of the queue.
2236 * Typically we use the tail for new ioctls and so forth. We use the head
2237 * of the queue for things like a QUEUE_FULL message from a device, or a
2238 * host that is unable to accept a particular command.
2239 */
2240void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002241 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242{
Tejun Heo 867d1192005-04-24 02:06:05 -05002243 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 unsigned long flags;
2245
2246 /*
2247 * tell I/O scheduler that this isn't a regular read/write (ie it
2248 * must not attempt merges on this) and that it acts as a soft
2249 * barrier
2250 */
2251 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2252
2253 rq->special = data;
2254
2255 spin_lock_irqsave(q->queue_lock, flags);
2256
2257 /*
2258 * If command is tagged, release the tag
2259 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002260 if (blk_rq_tagged(rq))
2261 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262
Tejun Heo 867d1192005-04-24 02:06:05 -05002263 drive_stat_acct(rq, rq->nr_sectors, 1);
2264 __elv_add_request(q, rq, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 if (blk_queue_plugged(q))
2267 __generic_unplug_device(q);
2268 else
2269 q->request_fn(q);
2270 spin_unlock_irqrestore(q->queue_lock, flags);
2271}
2272
2273EXPORT_SYMBOL(blk_insert_request);
2274
2275/**
2276 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2277 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002278 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 * @ubuf: the user buffer
2280 * @len: length of user data
2281 *
2282 * Description:
2283 * Data will be mapped directly for zero copy io, if possible. Otherwise
2284 * a kernel bounce buffer is used.
2285 *
2286 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2287 * still in process context.
2288 *
2289 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2290 * before being submitted to the device, as pages mapped may be out of
2291 * reach. It's the callers responsibility to make sure this happens. The
2292 * original bio must be passed back in to blk_rq_unmap_user() for proper
2293 * unmapping.
2294 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002295int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2296 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297{
2298 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002300 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Mike Christiedefd94b2005-12-05 02:37:06 -06002302 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002303 return -EINVAL;
2304 if (!len || !ubuf)
2305 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Jens Axboedd1cab92005-06-20 14:06:01 +02002307 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309 /*
2310 * if alignment requirement is satisfied, map in user pages for
2311 * direct dma. else, set up kernel bounce buffers
2312 */
2313 uaddr = (unsigned long) ubuf;
2314 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002315 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002317 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319 if (!IS_ERR(bio)) {
2320 rq->bio = rq->biotail = bio;
2321 blk_rq_bio_prep(q, rq, bio);
2322
2323 rq->buffer = rq->data = NULL;
2324 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002325 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
2327
2328 /*
2329 * bio is the err-ptr
2330 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002331 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332}
2333
2334EXPORT_SYMBOL(blk_rq_map_user);
2335
2336/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002337 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2338 * @q: request queue where request should be inserted
2339 * @rq: request to map data to
2340 * @iov: pointer to the iovec
2341 * @iov_count: number of elements in the iovec
2342 *
2343 * Description:
2344 * Data will be mapped directly for zero copy io, if possible. Otherwise
2345 * a kernel bounce buffer is used.
2346 *
2347 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2348 * still in process context.
2349 *
2350 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2351 * before being submitted to the device, as pages mapped may be out of
2352 * reach. It's the callers responsibility to make sure this happens. The
2353 * original bio must be passed back in to blk_rq_unmap_user() for proper
2354 * unmapping.
2355 */
2356int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2357 struct sg_iovec *iov, int iov_count)
2358{
2359 struct bio *bio;
2360
2361 if (!iov || iov_count <= 0)
2362 return -EINVAL;
2363
2364 /* we don't allow misaligned data like bio_map_user() does. If the
2365 * user is using sg, they're expected to know the alignment constraints
2366 * and respect them accordingly */
2367 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2368 if (IS_ERR(bio))
2369 return PTR_ERR(bio);
2370
2371 rq->bio = rq->biotail = bio;
2372 blk_rq_bio_prep(q, rq, bio);
2373 rq->buffer = rq->data = NULL;
2374 rq->data_len = bio->bi_size;
2375 return 0;
2376}
2377
2378EXPORT_SYMBOL(blk_rq_map_user_iov);
2379
2380/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002382 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 * @ulen: length of user buffer
2384 *
2385 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002386 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002388int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
2390 int ret = 0;
2391
2392 if (bio) {
2393 if (bio_flagged(bio, BIO_USER_MAPPED))
2394 bio_unmap_user(bio);
2395 else
2396 ret = bio_uncopy_user(bio);
2397 }
2398
Jens Axboedd1cab92005-06-20 14:06:01 +02002399 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400}
2401
2402EXPORT_SYMBOL(blk_rq_unmap_user);
2403
2404/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002405 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2406 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002407 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002408 * @kbuf: the kernel buffer
2409 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002410 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002411 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002412int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002413 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002414{
Mike Christie df46b9a2005-06-20 14:04:44 +02002415 struct bio *bio;
2416
Mike Christiedefd94b2005-12-05 02:37:06 -06002417 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002418 return -EINVAL;
2419 if (!len || !kbuf)
2420 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002421
2422 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002423 if (IS_ERR(bio))
2424 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002425
Jens Axboedd1cab92005-06-20 14:06:01 +02002426 if (rq_data_dir(rq) == WRITE)
2427 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002428
Jens Axboedd1cab92005-06-20 14:06:01 +02002429 rq->bio = rq->biotail = bio;
2430 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002431
Jens Axboedd1cab92005-06-20 14:06:01 +02002432 rq->buffer = rq->data = NULL;
2433 rq->data_len = len;
2434 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002435}
2436
2437EXPORT_SYMBOL(blk_rq_map_kern);
2438
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002439/**
2440 * blk_execute_rq_nowait - insert a request into queue for execution
2441 * @q: queue to insert the request in
2442 * @bd_disk: matching gendisk
2443 * @rq: request to insert
2444 * @at_head: insert request at head or tail of queue
2445 * @done: I/O completion handler
2446 *
2447 * Description:
2448 * Insert a fully prepared request at the back of the io scheduler queue
2449 * for execution. Don't wait for completion.
2450 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002451void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2452 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002453 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002454{
2455 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2456
2457 rq->rq_disk = bd_disk;
2458 rq->flags |= REQ_NOMERGE;
2459 rq->end_io = done;
2460 elv_add_request(q, rq, where, 1);
2461 generic_unplug_device(q);
2462}
2463
Mike Christie6e39b692005-11-11 05:30:24 -06002464EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2465
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466/**
2467 * blk_execute_rq - insert a request into queue for execution
2468 * @q: queue to insert the request in
2469 * @bd_disk: matching gendisk
2470 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002471 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 *
2473 * Description:
2474 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002475 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 */
2477int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002478 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
2480 DECLARE_COMPLETION(wait);
2481 char sense[SCSI_SENSE_BUFFERSIZE];
2482 int err = 0;
2483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 /*
2485 * we need an extra reference to the request, so we can look at
2486 * it after io completion
2487 */
2488 rq->ref_count++;
2489
2490 if (!rq->sense) {
2491 memset(sense, 0, sizeof(sense));
2492 rq->sense = sense;
2493 rq->sense_len = 0;
2494 }
2495
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 rq->waiting = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002497 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 wait_for_completion(&wait);
2499 rq->waiting = NULL;
2500
2501 if (rq->errors)
2502 err = -EIO;
2503
2504 return err;
2505}
2506
2507EXPORT_SYMBOL(blk_execute_rq);
2508
2509/**
2510 * blkdev_issue_flush - queue a flush
2511 * @bdev: blockdev to issue flush for
2512 * @error_sector: error sector
2513 *
2514 * Description:
2515 * Issue a flush for the block device in question. Caller can supply
2516 * room for storing the error offset in case of a flush error, if they
2517 * wish to. Caller must run wait_for_completion() on its own.
2518 */
2519int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2520{
2521 request_queue_t *q;
2522
2523 if (bdev->bd_disk == NULL)
2524 return -ENXIO;
2525
2526 q = bdev_get_queue(bdev);
2527 if (!q)
2528 return -ENXIO;
2529 if (!q->issue_flush_fn)
2530 return -EOPNOTSUPP;
2531
2532 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2533}
2534
2535EXPORT_SYMBOL(blkdev_issue_flush);
2536
Adrian Bunk93d17d32005-06-25 14:59:10 -07002537static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538{
2539 int rw = rq_data_dir(rq);
2540
2541 if (!blk_fs_request(rq) || !rq->rq_disk)
2542 return;
2543
Jens Axboed72d9042005-11-01 08:35:42 +01002544 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002545 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002546 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 disk_round_stats(rq->rq_disk);
2548 rq->rq_disk->in_flight++;
2549 }
2550}
2551
2552/*
2553 * add-request adds a request to the linked list.
2554 * queue lock is held and interrupts disabled, as we muck with the
2555 * request queue list.
2556 */
2557static inline void add_request(request_queue_t * q, struct request * req)
2558{
2559 drive_stat_acct(req, req->nr_sectors, 1);
2560
2561 if (q->activity_fn)
2562 q->activity_fn(q->activity_data, rq_data_dir(req));
2563
2564 /*
2565 * elevator indicated where it wants this request to be
2566 * inserted at elevator_merge time
2567 */
2568 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2569}
2570
2571/*
2572 * disk_round_stats() - Round off the performance stats on a struct
2573 * disk_stats.
2574 *
2575 * The average IO queue length and utilisation statistics are maintained
2576 * by observing the current state of the queue length and the amount of
2577 * time it has been in this state for.
2578 *
2579 * Normally, that accounting is done on IO completion, but that can result
2580 * in more than a second's worth of IO being accounted for within any one
2581 * second, leading to >100% utilisation. To deal with that, we call this
2582 * function to do a round-off before returning the results when reading
2583 * /proc/diskstats. This accounts immediately for all queue usage up to
2584 * the current jiffies and restarts the counters again.
2585 */
2586void disk_round_stats(struct gendisk *disk)
2587{
2588 unsigned long now = jiffies;
2589
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002590 if (now == disk->stamp)
2591 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002593 if (disk->in_flight) {
2594 __disk_stat_add(disk, time_in_queue,
2595 disk->in_flight * (now - disk->stamp));
2596 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599}
2600
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002601EXPORT_SYMBOL_GPL(disk_round_stats);
2602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603/*
2604 * queue lock must be held
2605 */
Mike Christie6e39b692005-11-11 05:30:24 -06002606void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
2608 struct request_list *rl = req->rl;
2609
2610 if (unlikely(!q))
2611 return;
2612 if (unlikely(--req->ref_count))
2613 return;
2614
Tejun Heo8922e162005-10-20 16:23:44 +02002615 elv_completed_request(q, req);
2616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 req->rq_status = RQ_INACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 req->rl = NULL;
2619
2620 /*
2621 * Request may not have originated from ll_rw_blk. if not,
2622 * it didn't come out of our reserved rq pools
2623 */
2624 if (rl) {
2625 int rw = rq_data_dir(req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002626 int priv = req->flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 BUG_ON(!list_empty(&req->queuelist));
2629
2630 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002631 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 }
2633}
2634
Mike Christie6e39b692005-11-11 05:30:24 -06002635EXPORT_SYMBOL_GPL(__blk_put_request);
2636
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637void blk_put_request(struct request *req)
2638{
Tejun Heo8922e162005-10-20 16:23:44 +02002639 unsigned long flags;
2640 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
Tejun Heo8922e162005-10-20 16:23:44 +02002642 /*
2643 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2644 * following if (q) test.
2645 */
2646 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 spin_lock_irqsave(q->queue_lock, flags);
2648 __blk_put_request(q, req);
2649 spin_unlock_irqrestore(q->queue_lock, flags);
2650 }
2651}
2652
2653EXPORT_SYMBOL(blk_put_request);
2654
2655/**
2656 * blk_end_sync_rq - executes a completion event on a request
2657 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002658 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002660void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661{
2662 struct completion *waiting = rq->waiting;
2663
2664 rq->waiting = NULL;
2665 __blk_put_request(rq->q, rq);
2666
2667 /*
2668 * complete last, if this is a stack request the process (and thus
2669 * the rq pointer) could be invalid right after this complete()
2670 */
2671 complete(waiting);
2672}
2673EXPORT_SYMBOL(blk_end_sync_rq);
2674
2675/**
2676 * blk_congestion_wait - wait for a queue to become uncongested
2677 * @rw: READ or WRITE
2678 * @timeout: timeout in jiffies
2679 *
2680 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2681 * If no queues are congested then just wait for the next request to be
2682 * returned.
2683 */
2684long blk_congestion_wait(int rw, long timeout)
2685{
2686 long ret;
2687 DEFINE_WAIT(wait);
2688 wait_queue_head_t *wqh = &congestion_wqh[rw];
2689
2690 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2691 ret = io_schedule_timeout(timeout);
2692 finish_wait(wqh, &wait);
2693 return ret;
2694}
2695
2696EXPORT_SYMBOL(blk_congestion_wait);
2697
2698/*
2699 * Has to be called with the request spinlock acquired
2700 */
2701static int attempt_merge(request_queue_t *q, struct request *req,
2702 struct request *next)
2703{
2704 if (!rq_mergeable(req) || !rq_mergeable(next))
2705 return 0;
2706
2707 /*
2708 * not contigious
2709 */
2710 if (req->sector + req->nr_sectors != next->sector)
2711 return 0;
2712
2713 if (rq_data_dir(req) != rq_data_dir(next)
2714 || req->rq_disk != next->rq_disk
2715 || next->waiting || next->special)
2716 return 0;
2717
2718 /*
2719 * If we are allowed to merge, then append bio list
2720 * from next to rq and release next. merge_requests_fn
2721 * will have updated segment counts, update sector
2722 * counts here.
2723 */
2724 if (!q->merge_requests_fn(q, req, next))
2725 return 0;
2726
2727 /*
2728 * At this point we have either done a back merge
2729 * or front merge. We need the smaller start_time of
2730 * the merged requests to be the current request
2731 * for accounting purposes.
2732 */
2733 if (time_after(req->start_time, next->start_time))
2734 req->start_time = next->start_time;
2735
2736 req->biotail->bi_next = next->bio;
2737 req->biotail = next->biotail;
2738
2739 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2740
2741 elv_merge_requests(q, req, next);
2742
2743 if (req->rq_disk) {
2744 disk_round_stats(req->rq_disk);
2745 req->rq_disk->in_flight--;
2746 }
2747
Jens Axboe22e2c502005-06-27 10:55:12 +02002748 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2749
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 __blk_put_request(q, next);
2751 return 1;
2752}
2753
2754static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2755{
2756 struct request *next = elv_latter_request(q, rq);
2757
2758 if (next)
2759 return attempt_merge(q, rq, next);
2760
2761 return 0;
2762}
2763
2764static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2765{
2766 struct request *prev = elv_former_request(q, rq);
2767
2768 if (prev)
2769 return attempt_merge(q, prev, rq);
2770
2771 return 0;
2772}
2773
Tejun Heo52d9e672006-01-06 09:49:58 +01002774static void init_request_from_bio(struct request *req, struct bio *bio)
2775{
2776 req->flags |= REQ_CMD;
2777
2778 /*
2779 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2780 */
2781 if (bio_rw_ahead(bio) || bio_failfast(bio))
2782 req->flags |= REQ_FAILFAST;
2783
2784 /*
2785 * REQ_BARRIER implies no merging, but lets make it explicit
2786 */
2787 if (unlikely(bio_barrier(bio)))
2788 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2789
2790 req->errors = 0;
2791 req->hard_sector = req->sector = bio->bi_sector;
2792 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2793 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2794 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2795 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2796 req->buffer = bio_data(bio); /* see ->buffer comment above */
2797 req->waiting = NULL;
2798 req->bio = req->biotail = bio;
2799 req->ioprio = bio_prio(bio);
2800 req->rq_disk = bio->bi_bdev->bd_disk;
2801 req->start_time = jiffies;
2802}
2803
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804static int __make_request(request_queue_t *q, struct bio *bio)
2805{
Nick Piggin450991b2005-06-28 20:45:13 -07002806 struct request *req;
Jens Axboe4a534f92005-04-16 15:25:40 -07002807 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
Jens Axboe22e2c502005-06-27 10:55:12 +02002808 unsigned short prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 sector_t sector;
2810
2811 sector = bio->bi_sector;
2812 nr_sectors = bio_sectors(bio);
2813 cur_nr_sectors = bio_cur_sectors(bio);
Jens Axboe22e2c502005-06-27 10:55:12 +02002814 prio = bio_prio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816 rw = bio_data_dir(bio);
Jens Axboe4a534f92005-04-16 15:25:40 -07002817 sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
2819 /*
2820 * low level driver can indicate that it wants pages above a
2821 * certain limit bounced to low memory (ie for highmem, or even
2822 * ISA dma in theory)
2823 */
2824 blk_queue_bounce(q, &bio);
2825
2826 spin_lock_prefetch(q->queue_lock);
2827
2828 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002829 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 err = -EOPNOTSUPP;
2831 goto end_io;
2832 }
2833
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 spin_lock_irq(q->queue_lock);
2835
Nick Piggin450991b2005-06-28 20:45:13 -07002836 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 goto get_rq;
2838
2839 el_ret = elv_merge(q, &req, bio);
2840 switch (el_ret) {
2841 case ELEVATOR_BACK_MERGE:
2842 BUG_ON(!rq_mergeable(req));
2843
2844 if (!q->back_merge_fn(q, req, bio))
2845 break;
2846
2847 req->biotail->bi_next = bio;
2848 req->biotail = bio;
2849 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002850 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 drive_stat_acct(req, nr_sectors, 0);
2852 if (!attempt_back_merge(q, req))
2853 elv_merged_request(q, req);
2854 goto out;
2855
2856 case ELEVATOR_FRONT_MERGE:
2857 BUG_ON(!rq_mergeable(req));
2858
2859 if (!q->front_merge_fn(q, req, bio))
2860 break;
2861
2862 bio->bi_next = req->bio;
2863 req->bio = bio;
2864
2865 /*
2866 * may not be valid. if the low level driver said
2867 * it didn't need a bounce buffer then it better
2868 * not touch req->buffer either...
2869 */
2870 req->buffer = bio_data(bio);
2871 req->current_nr_sectors = cur_nr_sectors;
2872 req->hard_cur_sectors = cur_nr_sectors;
2873 req->sector = req->hard_sector = sector;
2874 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002875 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 drive_stat_acct(req, nr_sectors, 0);
2877 if (!attempt_front_merge(q, req))
2878 elv_merged_request(q, req);
2879 goto out;
2880
Nick Piggin450991b2005-06-28 20:45:13 -07002881 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002883 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 }
2885
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002887 /*
2888 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002889 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002890 */
Nick Piggin450991b2005-06-28 20:45:13 -07002891 req = get_request_wait(q, rw, bio);
Nick Piggind6344532005-06-28 20:45:14 -07002892
Nick Piggin450991b2005-06-28 20:45:13 -07002893 /*
2894 * After dropping the lock and possibly sleeping here, our request
2895 * may now be mergeable after it had proven unmergeable (above).
2896 * We don't worry about that case for efficiency. It won't happen
2897 * often, and the elevators are able to handle it.
2898 */
Tejun Heo52d9e672006-01-06 09:49:58 +01002899 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
Nick Piggin450991b2005-06-28 20:45:13 -07002901 spin_lock_irq(q->queue_lock);
2902 if (elv_queue_empty(q))
2903 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 add_request(q, req);
2905out:
Jens Axboe4a534f92005-04-16 15:25:40 -07002906 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 __generic_unplug_device(q);
2908
2909 spin_unlock_irq(q->queue_lock);
2910 return 0;
2911
2912end_io:
2913 bio_endio(bio, nr_sectors << 9, err);
2914 return 0;
2915}
2916
2917/*
2918 * If bio->bi_dev is a partition, remap the location
2919 */
2920static inline void blk_partition_remap(struct bio *bio)
2921{
2922 struct block_device *bdev = bio->bi_bdev;
2923
2924 if (bdev != bdev->bd_contains) {
2925 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01002926 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
Jens Axboea3623572005-11-01 09:26:16 +01002928 p->sectors[rw] += bio_sectors(bio);
2929 p->ios[rw]++;
2930
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 bio->bi_sector += p->start_sect;
2932 bio->bi_bdev = bdev->bd_contains;
2933 }
2934}
2935
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936static void handle_bad_sector(struct bio *bio)
2937{
2938 char b[BDEVNAME_SIZE];
2939
2940 printk(KERN_INFO "attempt to access beyond end of device\n");
2941 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2942 bdevname(bio->bi_bdev, b),
2943 bio->bi_rw,
2944 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2945 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2946
2947 set_bit(BIO_EOF, &bio->bi_flags);
2948}
2949
2950/**
2951 * generic_make_request: hand a buffer to its device driver for I/O
2952 * @bio: The bio describing the location in memory and on the device.
2953 *
2954 * generic_make_request() is used to make I/O requests of block
2955 * devices. It is passed a &struct bio, which describes the I/O that needs
2956 * to be done.
2957 *
2958 * generic_make_request() does not return any status. The
2959 * success/failure status of the request, along with notification of
2960 * completion, is delivered asynchronously through the bio->bi_end_io
2961 * function described (one day) else where.
2962 *
2963 * The caller of generic_make_request must make sure that bi_io_vec
2964 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2965 * set to describe the device address, and the
2966 * bi_end_io and optionally bi_private are set to describe how
2967 * completion notification should be signaled.
2968 *
2969 * generic_make_request and the drivers it calls may use bi_next if this
2970 * bio happens to be merged with someone else, and may change bi_dev and
2971 * bi_sector for remaps as it sees fit. So the values of these fields
2972 * should NOT be depended on after the call to generic_make_request.
2973 */
2974void generic_make_request(struct bio *bio)
2975{
2976 request_queue_t *q;
2977 sector_t maxsector;
2978 int ret, nr_sectors = bio_sectors(bio);
2979
2980 might_sleep();
2981 /* Test device or partition size, when known. */
2982 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2983 if (maxsector) {
2984 sector_t sector = bio->bi_sector;
2985
2986 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2987 /*
2988 * This may well happen - the kernel calls bread()
2989 * without checking the size of the device, e.g., when
2990 * mounting a device.
2991 */
2992 handle_bad_sector(bio);
2993 goto end_io;
2994 }
2995 }
2996
2997 /*
2998 * Resolve the mapping until finished. (drivers are
2999 * still free to implement/resolve their own stacking
3000 * by explicitly returning 0)
3001 *
3002 * NOTE: we don't repeat the blk_size check for each new device.
3003 * Stacking drivers are expected to know what they are doing.
3004 */
3005 do {
3006 char b[BDEVNAME_SIZE];
3007
3008 q = bdev_get_queue(bio->bi_bdev);
3009 if (!q) {
3010 printk(KERN_ERR
3011 "generic_make_request: Trying to access "
3012 "nonexistent block-device %s (%Lu)\n",
3013 bdevname(bio->bi_bdev, b),
3014 (long long) bio->bi_sector);
3015end_io:
3016 bio_endio(bio, bio->bi_size, -EIO);
3017 break;
3018 }
3019
3020 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3021 printk("bio too big device %s (%u > %u)\n",
3022 bdevname(bio->bi_bdev, b),
3023 bio_sectors(bio),
3024 q->max_hw_sectors);
3025 goto end_io;
3026 }
3027
Nick Pigginfde6ad22005-06-23 00:08:53 -07003028 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 goto end_io;
3030
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 /*
3032 * If this device has partitions, remap block n
3033 * of partition p to block n+start(p) of the disk.
3034 */
3035 blk_partition_remap(bio);
3036
3037 ret = q->make_request_fn(q, bio);
3038 } while (ret);
3039}
3040
3041EXPORT_SYMBOL(generic_make_request);
3042
3043/**
3044 * submit_bio: submit a bio to the block device layer for I/O
3045 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3046 * @bio: The &struct bio which describes the I/O
3047 *
3048 * submit_bio() is very similar in purpose to generic_make_request(), and
3049 * uses that function to do most of the work. Both are fairly rough
3050 * interfaces, @bio must be presetup and ready for I/O.
3051 *
3052 */
3053void submit_bio(int rw, struct bio *bio)
3054{
3055 int count = bio_sectors(bio);
3056
3057 BIO_BUG_ON(!bio->bi_size);
3058 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003059 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 if (rw & WRITE)
3061 mod_page_state(pgpgout, count);
3062 else
3063 mod_page_state(pgpgin, count);
3064
3065 if (unlikely(block_dump)) {
3066 char b[BDEVNAME_SIZE];
3067 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3068 current->comm, current->pid,
3069 (rw & WRITE) ? "WRITE" : "READ",
3070 (unsigned long long)bio->bi_sector,
3071 bdevname(bio->bi_bdev,b));
3072 }
3073
3074 generic_make_request(bio);
3075}
3076
3077EXPORT_SYMBOL(submit_bio);
3078
Adrian Bunk93d17d32005-06-25 14:59:10 -07003079static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080{
3081 struct bio *bio, *prevbio = NULL;
3082 int nr_phys_segs, nr_hw_segs;
3083 unsigned int phys_size, hw_size;
3084 request_queue_t *q = rq->q;
3085
3086 if (!rq->bio)
3087 return;
3088
3089 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3090 rq_for_each_bio(bio, rq) {
3091 /* Force bio hw/phys segs to be recalculated. */
3092 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3093
3094 nr_phys_segs += bio_phys_segments(q, bio);
3095 nr_hw_segs += bio_hw_segments(q, bio);
3096 if (prevbio) {
3097 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3098 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3099
3100 if (blk_phys_contig_segment(q, prevbio, bio) &&
3101 pseg <= q->max_segment_size) {
3102 nr_phys_segs--;
3103 phys_size += prevbio->bi_size + bio->bi_size;
3104 } else
3105 phys_size = 0;
3106
3107 if (blk_hw_contig_segment(q, prevbio, bio) &&
3108 hseg <= q->max_segment_size) {
3109 nr_hw_segs--;
3110 hw_size += prevbio->bi_size + bio->bi_size;
3111 } else
3112 hw_size = 0;
3113 }
3114 prevbio = bio;
3115 }
3116
3117 rq->nr_phys_segments = nr_phys_segs;
3118 rq->nr_hw_segments = nr_hw_segs;
3119}
3120
Adrian Bunk93d17d32005-06-25 14:59:10 -07003121static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122{
3123 if (blk_fs_request(rq)) {
3124 rq->hard_sector += nsect;
3125 rq->hard_nr_sectors -= nsect;
3126
3127 /*
3128 * Move the I/O submission pointers ahead if required.
3129 */
3130 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3131 (rq->sector <= rq->hard_sector)) {
3132 rq->sector = rq->hard_sector;
3133 rq->nr_sectors = rq->hard_nr_sectors;
3134 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3135 rq->current_nr_sectors = rq->hard_cur_sectors;
3136 rq->buffer = bio_data(rq->bio);
3137 }
3138
3139 /*
3140 * if total number of sectors is less than the first segment
3141 * size, something has gone terribly wrong
3142 */
3143 if (rq->nr_sectors < rq->current_nr_sectors) {
3144 printk("blk: request botched\n");
3145 rq->nr_sectors = rq->current_nr_sectors;
3146 }
3147 }
3148}
3149
3150static int __end_that_request_first(struct request *req, int uptodate,
3151 int nr_bytes)
3152{
3153 int total_bytes, bio_nbytes, error, next_idx = 0;
3154 struct bio *bio;
3155
3156 /*
3157 * extend uptodate bool to allow < 0 value to be direct io error
3158 */
3159 error = 0;
3160 if (end_io_error(uptodate))
3161 error = !uptodate ? -EIO : uptodate;
3162
3163 /*
3164 * for a REQ_BLOCK_PC request, we want to carry any eventual
3165 * sense key with us all the way through
3166 */
3167 if (!blk_pc_request(req))
3168 req->errors = 0;
3169
3170 if (!uptodate) {
3171 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3172 printk("end_request: I/O error, dev %s, sector %llu\n",
3173 req->rq_disk ? req->rq_disk->disk_name : "?",
3174 (unsigned long long)req->sector);
3175 }
3176
Jens Axboed72d9042005-11-01 08:35:42 +01003177 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003178 const int rw = rq_data_dir(req);
3179
Jens Axboe53e86062006-01-17 11:09:27 +01003180 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003181 }
3182
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183 total_bytes = bio_nbytes = 0;
3184 while ((bio = req->bio) != NULL) {
3185 int nbytes;
3186
3187 if (nr_bytes >= bio->bi_size) {
3188 req->bio = bio->bi_next;
3189 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003190 if (!ordered_bio_endio(req, bio, nbytes, error))
3191 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 next_idx = 0;
3193 bio_nbytes = 0;
3194 } else {
3195 int idx = bio->bi_idx + next_idx;
3196
3197 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3198 blk_dump_rq_flags(req, "__end_that");
3199 printk("%s: bio idx %d >= vcnt %d\n",
3200 __FUNCTION__,
3201 bio->bi_idx, bio->bi_vcnt);
3202 break;
3203 }
3204
3205 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3206 BIO_BUG_ON(nbytes > bio->bi_size);
3207
3208 /*
3209 * not a complete bvec done
3210 */
3211 if (unlikely(nbytes > nr_bytes)) {
3212 bio_nbytes += nr_bytes;
3213 total_bytes += nr_bytes;
3214 break;
3215 }
3216
3217 /*
3218 * advance to the next vector
3219 */
3220 next_idx++;
3221 bio_nbytes += nbytes;
3222 }
3223
3224 total_bytes += nbytes;
3225 nr_bytes -= nbytes;
3226
3227 if ((bio = req->bio)) {
3228 /*
3229 * end more in this run, or just return 'not-done'
3230 */
3231 if (unlikely(nr_bytes <= 0))
3232 break;
3233 }
3234 }
3235
3236 /*
3237 * completely done
3238 */
3239 if (!req->bio)
3240 return 0;
3241
3242 /*
3243 * if the request wasn't completed, update state
3244 */
3245 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003246 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3247 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 bio->bi_idx += next_idx;
3249 bio_iovec(bio)->bv_offset += nr_bytes;
3250 bio_iovec(bio)->bv_len -= nr_bytes;
3251 }
3252
3253 blk_recalc_rq_sectors(req, total_bytes >> 9);
3254 blk_recalc_rq_segments(req);
3255 return 1;
3256}
3257
3258/**
3259 * end_that_request_first - end I/O on a request
3260 * @req: the request being processed
3261 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3262 * @nr_sectors: number of sectors to end I/O on
3263 *
3264 * Description:
3265 * Ends I/O on a number of sectors attached to @req, and sets it up
3266 * for the next range of segments (if any) in the cluster.
3267 *
3268 * Return:
3269 * 0 - we are done with this request, call end_that_request_last()
3270 * 1 - still buffers pending for this request
3271 **/
3272int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3273{
3274 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3275}
3276
3277EXPORT_SYMBOL(end_that_request_first);
3278
3279/**
3280 * end_that_request_chunk - end I/O on a request
3281 * @req: the request being processed
3282 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3283 * @nr_bytes: number of bytes to complete
3284 *
3285 * Description:
3286 * Ends I/O on a number of bytes attached to @req, and sets it up
3287 * for the next range of segments (if any). Like end_that_request_first(),
3288 * but deals with bytes instead of sectors.
3289 *
3290 * Return:
3291 * 0 - we are done with this request, call end_that_request_last()
3292 * 1 - still buffers pending for this request
3293 **/
3294int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3295{
3296 return __end_that_request_first(req, uptodate, nr_bytes);
3297}
3298
3299EXPORT_SYMBOL(end_that_request_chunk);
3300
3301/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003302 * splice the completion data to a local structure and hand off to
3303 * process_completion_queue() to complete the requests
3304 */
3305static void blk_done_softirq(struct softirq_action *h)
3306{
3307 struct list_head *cpu_list;
3308 LIST_HEAD(local_list);
3309
3310 local_irq_disable();
3311 cpu_list = &__get_cpu_var(blk_cpu_done);
3312 list_splice_init(cpu_list, &local_list);
3313 local_irq_enable();
3314
3315 while (!list_empty(&local_list)) {
3316 struct request *rq = list_entry(local_list.next, struct request, donelist);
3317
3318 list_del_init(&rq->donelist);
3319 rq->q->softirq_done_fn(rq);
3320 }
3321}
3322
3323#ifdef CONFIG_HOTPLUG_CPU
3324
3325static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3326 void *hcpu)
3327{
3328 /*
3329 * If a CPU goes away, splice its entries to the current CPU
3330 * and trigger a run of the softirq
3331 */
3332 if (action == CPU_DEAD) {
3333 int cpu = (unsigned long) hcpu;
3334
3335 local_irq_disable();
3336 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3337 &__get_cpu_var(blk_cpu_done));
3338 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3339 local_irq_enable();
3340 }
3341
3342 return NOTIFY_OK;
3343}
3344
3345
3346static struct notifier_block __devinitdata blk_cpu_notifier = {
3347 .notifier_call = blk_cpu_notify,
3348};
3349
3350#endif /* CONFIG_HOTPLUG_CPU */
3351
3352/**
3353 * blk_complete_request - end I/O on a request
3354 * @req: the request being processed
3355 *
3356 * Description:
3357 * Ends all I/O on a request. It does not handle partial completions,
3358 * unless the driver actually implements this in its completionc callback
3359 * through requeueing. Theh actual completion happens out-of-order,
3360 * through a softirq handler. The user must have registered a completion
3361 * callback through blk_queue_softirq_done().
3362 **/
3363
3364void blk_complete_request(struct request *req)
3365{
3366 struct list_head *cpu_list;
3367 unsigned long flags;
3368
3369 BUG_ON(!req->q->softirq_done_fn);
3370
3371 local_irq_save(flags);
3372
3373 cpu_list = &__get_cpu_var(blk_cpu_done);
3374 list_add_tail(&req->donelist, cpu_list);
3375 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3376
3377 local_irq_restore(flags);
3378}
3379
3380EXPORT_SYMBOL(blk_complete_request);
3381
3382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 * queue lock must be held
3384 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003385void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386{
3387 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003388 int error;
3389
3390 /*
3391 * extend uptodate bool to allow < 0 value to be direct io error
3392 */
3393 error = 0;
3394 if (end_io_error(uptodate))
3395 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396
3397 if (unlikely(laptop_mode) && blk_fs_request(req))
3398 laptop_io_completion();
3399
3400 if (disk && blk_fs_request(req)) {
3401 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003402 const int rw = rq_data_dir(req);
3403
3404 __disk_stat_inc(disk, ios[rw]);
3405 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 disk_round_stats(disk);
3407 disk->in_flight--;
3408 }
3409 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003410 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 else
3412 __blk_put_request(req->q, req);
3413}
3414
3415EXPORT_SYMBOL(end_that_request_last);
3416
3417void end_request(struct request *req, int uptodate)
3418{
3419 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3420 add_disk_randomness(req->rq_disk);
3421 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003422 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 }
3424}
3425
3426EXPORT_SYMBOL(end_request);
3427
3428void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3429{
3430 /* first three bits are identical in rq->flags and bio->bi_rw */
3431 rq->flags |= (bio->bi_rw & 7);
3432
3433 rq->nr_phys_segments = bio_phys_segments(q, bio);
3434 rq->nr_hw_segments = bio_hw_segments(q, bio);
3435 rq->current_nr_sectors = bio_cur_sectors(bio);
3436 rq->hard_cur_sectors = rq->current_nr_sectors;
3437 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3438 rq->buffer = bio_data(bio);
3439
3440 rq->bio = rq->biotail = bio;
3441}
3442
3443EXPORT_SYMBOL(blk_rq_bio_prep);
3444
3445int kblockd_schedule_work(struct work_struct *work)
3446{
3447 return queue_work(kblockd_workqueue, work);
3448}
3449
3450EXPORT_SYMBOL(kblockd_schedule_work);
3451
3452void kblockd_flush(void)
3453{
3454 flush_workqueue(kblockd_workqueue);
3455}
3456EXPORT_SYMBOL(kblockd_flush);
3457
3458int __init blk_dev_init(void)
3459{
Jens Axboeff856ba2006-01-09 16:02:34 +01003460 int i;
3461
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462 kblockd_workqueue = create_workqueue("kblockd");
3463 if (!kblockd_workqueue)
3464 panic("Failed to create kblockd\n");
3465
3466 request_cachep = kmem_cache_create("blkdev_requests",
3467 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3468
3469 requestq_cachep = kmem_cache_create("blkdev_queue",
3470 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3471
3472 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3473 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3474
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003475 for_each_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003476 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3477
3478 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
3479#ifdef CONFIG_HOTPLUG_CPU
3480 register_cpu_notifier(&blk_cpu_notifier);
3481#endif
3482
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483 blk_max_low_pfn = max_low_pfn;
3484 blk_max_pfn = max_pfn;
3485
3486 return 0;
3487}
3488
3489/*
3490 * IO Context helper functions
3491 */
3492void put_io_context(struct io_context *ioc)
3493{
3494 if (ioc == NULL)
3495 return;
3496
3497 BUG_ON(atomic_read(&ioc->refcount) == 0);
3498
3499 if (atomic_dec_and_test(&ioc->refcount)) {
Al Viro334e94d2006-03-18 15:05:53 -05003500 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 if (ioc->aic && ioc->aic->dtor)
3502 ioc->aic->dtor(ioc->aic);
3503 if (ioc->cic && ioc->cic->dtor)
3504 ioc->cic->dtor(ioc->cic);
Al Viro334e94d2006-03-18 15:05:53 -05003505 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506
3507 kmem_cache_free(iocontext_cachep, ioc);
3508 }
3509}
3510EXPORT_SYMBOL(put_io_context);
3511
3512/* Called by the exitting task */
3513void exit_io_context(void)
3514{
3515 unsigned long flags;
3516 struct io_context *ioc;
3517
3518 local_irq_save(flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02003519 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 ioc = current->io_context;
3521 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003522 ioc->task = NULL;
3523 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 local_irq_restore(flags);
3525
3526 if (ioc->aic && ioc->aic->exit)
3527 ioc->aic->exit(ioc->aic);
3528 if (ioc->cic && ioc->cic->exit)
3529 ioc->cic->exit(ioc->cic);
3530
3531 put_io_context(ioc);
3532}
3533
3534/*
3535 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003536 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003538 * This returned IO context doesn't have a specifically elevated refcount,
3539 * but since the current task itself holds a reference, the context can be
3540 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 */
Al Viro8267e262005-10-21 03:20:53 -04003542struct io_context *current_io_context(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543{
3544 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 struct io_context *ret;
3546
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003548 if (likely(ret))
3549 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550
3551 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3552 if (ret) {
3553 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003554 ret->task = current;
3555 ret->set_ioprio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 ret->last_waited = jiffies; /* doesn't matter... */
3557 ret->nr_batch_requests = 0; /* because this is 0 */
3558 ret->aic = NULL;
3559 ret->cic = NULL;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003560 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561 }
3562
3563 return ret;
3564}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003565EXPORT_SYMBOL(current_io_context);
3566
3567/*
3568 * If the current task has no IO context then create one and initialise it.
3569 * If it does have a context, take a ref on it.
3570 *
3571 * This is always called in the context of the task which submitted the I/O.
3572 */
Al Viro8267e262005-10-21 03:20:53 -04003573struct io_context *get_io_context(gfp_t gfp_flags)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003574{
3575 struct io_context *ret;
3576 ret = current_io_context(gfp_flags);
3577 if (likely(ret))
3578 atomic_inc(&ret->refcount);
3579 return ret;
3580}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581EXPORT_SYMBOL(get_io_context);
3582
3583void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3584{
3585 struct io_context *src = *psrc;
3586 struct io_context *dst = *pdst;
3587
3588 if (src) {
3589 BUG_ON(atomic_read(&src->refcount) == 0);
3590 atomic_inc(&src->refcount);
3591 put_io_context(dst);
3592 *pdst = src;
3593 }
3594}
3595EXPORT_SYMBOL(copy_io_context);
3596
3597void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3598{
3599 struct io_context *temp;
3600 temp = *ioc1;
3601 *ioc1 = *ioc2;
3602 *ioc2 = temp;
3603}
3604EXPORT_SYMBOL(swap_io_context);
3605
3606/*
3607 * sysfs parts below
3608 */
3609struct queue_sysfs_entry {
3610 struct attribute attr;
3611 ssize_t (*show)(struct request_queue *, char *);
3612 ssize_t (*store)(struct request_queue *, const char *, size_t);
3613};
3614
3615static ssize_t
3616queue_var_show(unsigned int var, char *page)
3617{
3618 return sprintf(page, "%d\n", var);
3619}
3620
3621static ssize_t
3622queue_var_store(unsigned long *var, const char *page, size_t count)
3623{
3624 char *p = (char *) page;
3625
3626 *var = simple_strtoul(p, &p, 10);
3627 return count;
3628}
3629
3630static ssize_t queue_requests_show(struct request_queue *q, char *page)
3631{
3632 return queue_var_show(q->nr_requests, (page));
3633}
3634
3635static ssize_t
3636queue_requests_store(struct request_queue *q, const char *page, size_t count)
3637{
3638 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003639 unsigned long nr;
3640 int ret = queue_var_store(&nr, page, count);
3641 if (nr < BLKDEV_MIN_RQ)
3642 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643
Al Viroc981ff92006-03-18 13:51:29 -05003644 spin_lock_irq(q->queue_lock);
3645 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 blk_queue_congestion_threshold(q);
3647
3648 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3649 set_queue_congested(q, READ);
3650 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3651 clear_queue_congested(q, READ);
3652
3653 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3654 set_queue_congested(q, WRITE);
3655 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3656 clear_queue_congested(q, WRITE);
3657
3658 if (rl->count[READ] >= q->nr_requests) {
3659 blk_set_queue_full(q, READ);
3660 } else if (rl->count[READ]+1 <= q->nr_requests) {
3661 blk_clear_queue_full(q, READ);
3662 wake_up(&rl->wait[READ]);
3663 }
3664
3665 if (rl->count[WRITE] >= q->nr_requests) {
3666 blk_set_queue_full(q, WRITE);
3667 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3668 blk_clear_queue_full(q, WRITE);
3669 wake_up(&rl->wait[WRITE]);
3670 }
Al Viroc981ff92006-03-18 13:51:29 -05003671 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 return ret;
3673}
3674
3675static ssize_t queue_ra_show(struct request_queue *q, char *page)
3676{
3677 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3678
3679 return queue_var_show(ra_kb, (page));
3680}
3681
3682static ssize_t
3683queue_ra_store(struct request_queue *q, const char *page, size_t count)
3684{
3685 unsigned long ra_kb;
3686 ssize_t ret = queue_var_store(&ra_kb, page, count);
3687
3688 spin_lock_irq(q->queue_lock);
3689 if (ra_kb > (q->max_sectors >> 1))
3690 ra_kb = (q->max_sectors >> 1);
3691
3692 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3693 spin_unlock_irq(q->queue_lock);
3694
3695 return ret;
3696}
3697
3698static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3699{
3700 int max_sectors_kb = q->max_sectors >> 1;
3701
3702 return queue_var_show(max_sectors_kb, (page));
3703}
3704
3705static ssize_t
3706queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3707{
3708 unsigned long max_sectors_kb,
3709 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3710 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3711 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3712 int ra_kb;
3713
3714 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3715 return -EINVAL;
3716 /*
3717 * Take the queue lock to update the readahead and max_sectors
3718 * values synchronously:
3719 */
3720 spin_lock_irq(q->queue_lock);
3721 /*
3722 * Trim readahead window as well, if necessary:
3723 */
3724 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3725 if (ra_kb > max_sectors_kb)
3726 q->backing_dev_info.ra_pages =
3727 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3728
3729 q->max_sectors = max_sectors_kb << 1;
3730 spin_unlock_irq(q->queue_lock);
3731
3732 return ret;
3733}
3734
3735static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3736{
3737 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3738
3739 return queue_var_show(max_hw_sectors_kb, (page));
3740}
3741
3742
3743static struct queue_sysfs_entry queue_requests_entry = {
3744 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3745 .show = queue_requests_show,
3746 .store = queue_requests_store,
3747};
3748
3749static struct queue_sysfs_entry queue_ra_entry = {
3750 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3751 .show = queue_ra_show,
3752 .store = queue_ra_store,
3753};
3754
3755static struct queue_sysfs_entry queue_max_sectors_entry = {
3756 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3757 .show = queue_max_sectors_show,
3758 .store = queue_max_sectors_store,
3759};
3760
3761static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3762 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3763 .show = queue_max_hw_sectors_show,
3764};
3765
3766static struct queue_sysfs_entry queue_iosched_entry = {
3767 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3768 .show = elv_iosched_show,
3769 .store = elv_iosched_store,
3770};
3771
3772static struct attribute *default_attrs[] = {
3773 &queue_requests_entry.attr,
3774 &queue_ra_entry.attr,
3775 &queue_max_hw_sectors_entry.attr,
3776 &queue_max_sectors_entry.attr,
3777 &queue_iosched_entry.attr,
3778 NULL,
3779};
3780
3781#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3782
3783static ssize_t
3784queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3785{
3786 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003787 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3788 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003791 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003792 mutex_lock(&q->sysfs_lock);
3793 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3794 mutex_unlock(&q->sysfs_lock);
3795 return -ENOENT;
3796 }
3797 res = entry->show(q, page);
3798 mutex_unlock(&q->sysfs_lock);
3799 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800}
3801
3802static ssize_t
3803queue_attr_store(struct kobject *kobj, struct attribute *attr,
3804 const char *page, size_t length)
3805{
3806 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003807 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808
Al Viro483f4af2006-03-18 18:34:37 -05003809 ssize_t res;
3810
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003812 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003813 mutex_lock(&q->sysfs_lock);
3814 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3815 mutex_unlock(&q->sysfs_lock);
3816 return -ENOENT;
3817 }
3818 res = entry->store(q, page, length);
3819 mutex_unlock(&q->sysfs_lock);
3820 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821}
3822
3823static struct sysfs_ops queue_sysfs_ops = {
3824 .show = queue_attr_show,
3825 .store = queue_attr_store,
3826};
3827
Adrian Bunk93d17d32005-06-25 14:59:10 -07003828static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 .sysfs_ops = &queue_sysfs_ops,
3830 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05003831 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003832};
3833
3834int blk_register_queue(struct gendisk *disk)
3835{
3836 int ret;
3837
3838 request_queue_t *q = disk->queue;
3839
3840 if (!q || !q->request_fn)
3841 return -ENXIO;
3842
3843 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844
Al Viro483f4af2006-03-18 18:34:37 -05003845 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 if (ret < 0)
3847 return ret;
3848
Al Viro483f4af2006-03-18 18:34:37 -05003849 kobject_uevent(&q->kobj, KOBJ_ADD);
3850
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851 ret = elv_register_queue(q);
3852 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05003853 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3854 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 return ret;
3856 }
3857
3858 return 0;
3859}
3860
3861void blk_unregister_queue(struct gendisk *disk)
3862{
3863 request_queue_t *q = disk->queue;
3864
3865 if (q && q->request_fn) {
3866 elv_unregister_queue(q);
3867
Al Viro483f4af2006-03-18 18:34:37 -05003868 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3869 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 kobject_put(&disk->kobj);
3871 }
3872}