blob: 03d9c82b0fe7911fd6e922f746249795ccb80d46 [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
628 * buffers for doing I/O to pages residing above @page. By default
629 * the block layer sets this to the highest numbered "low" memory page.
630 **/
631void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
632{
633 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
634
635 /*
636 * set appropriate bounce gfp mask -- unfortunately we don't have a
637 * full 4GB zone, so we have to resort to low memory for any bounces.
638 * ISA has its own < 16MB zone.
639 */
640 if (bounce_pfn < blk_max_low_pfn) {
641 BUG_ON(dma_addr < BLK_BOUNCE_ISA);
642 init_emergency_isa_pool();
643 q->bounce_gfp = GFP_NOIO | GFP_DMA;
644 } else
645 q->bounce_gfp = GFP_NOIO;
646
647 q->bounce_pfn = bounce_pfn;
648}
649
650EXPORT_SYMBOL(blk_queue_bounce_limit);
651
652/**
653 * blk_queue_max_sectors - set max sectors for a request for this queue
654 * @q: the request queue for the device
655 * @max_sectors: max sectors in the usual 512b unit
656 *
657 * Description:
658 * Enables a low level driver to set an upper limit on the size of
659 * received requests.
660 **/
Jens Axboe2cb2e142006-01-17 09:04:32 +0100661void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
664 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
665 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
666 }
667
Mike Christiedefd94b2005-12-05 02:37:06 -0600668 if (BLK_DEF_MAX_SECTORS > max_sectors)
669 q->max_hw_sectors = q->max_sectors = max_sectors;
670 else {
671 q->max_sectors = BLK_DEF_MAX_SECTORS;
672 q->max_hw_sectors = max_sectors;
673 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676EXPORT_SYMBOL(blk_queue_max_sectors);
677
678/**
679 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
680 * @q: the request queue for the device
681 * @max_segments: max number of segments
682 *
683 * Description:
684 * Enables a low level driver to set an upper limit on the number of
685 * physical data segments in a request. This would be the largest sized
686 * scatter list the driver could handle.
687 **/
688void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
689{
690 if (!max_segments) {
691 max_segments = 1;
692 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
693 }
694
695 q->max_phys_segments = max_segments;
696}
697
698EXPORT_SYMBOL(blk_queue_max_phys_segments);
699
700/**
701 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
702 * @q: the request queue for the device
703 * @max_segments: max number of segments
704 *
705 * Description:
706 * Enables a low level driver to set an upper limit on the number of
707 * hw data segments in a request. This would be the largest number of
708 * address/length pairs the host adapter can actually give as once
709 * to the device.
710 **/
711void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
712{
713 if (!max_segments) {
714 max_segments = 1;
715 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
716 }
717
718 q->max_hw_segments = max_segments;
719}
720
721EXPORT_SYMBOL(blk_queue_max_hw_segments);
722
723/**
724 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
725 * @q: the request queue for the device
726 * @max_size: max size of segment in bytes
727 *
728 * Description:
729 * Enables a low level driver to set an upper limit on the size of a
730 * coalesced segment
731 **/
732void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
733{
734 if (max_size < PAGE_CACHE_SIZE) {
735 max_size = PAGE_CACHE_SIZE;
736 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
737 }
738
739 q->max_segment_size = max_size;
740}
741
742EXPORT_SYMBOL(blk_queue_max_segment_size);
743
744/**
745 * blk_queue_hardsect_size - set hardware sector size for the queue
746 * @q: the request queue for the device
747 * @size: the hardware sector size, in bytes
748 *
749 * Description:
750 * This should typically be set to the lowest possible sector size
751 * that the hardware can operate on (possible without reverting to
752 * even internal read-modify-write operations). Usually the default
753 * of 512 covers most hardware.
754 **/
755void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
756{
757 q->hardsect_size = size;
758}
759
760EXPORT_SYMBOL(blk_queue_hardsect_size);
761
762/*
763 * Returns the minimum that is _not_ zero, unless both are zero.
764 */
765#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
766
767/**
768 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
769 * @t: the stacking driver (top)
770 * @b: the underlying device (bottom)
771 **/
772void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
773{
774 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600775 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
776 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
779 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
780 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
781 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
782}
783
784EXPORT_SYMBOL(blk_queue_stack_limits);
785
786/**
787 * blk_queue_segment_boundary - set boundary rules for segment merging
788 * @q: the request queue for the device
789 * @mask: the memory boundary mask
790 **/
791void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
792{
793 if (mask < PAGE_CACHE_SIZE - 1) {
794 mask = PAGE_CACHE_SIZE - 1;
795 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
796 }
797
798 q->seg_boundary_mask = mask;
799}
800
801EXPORT_SYMBOL(blk_queue_segment_boundary);
802
803/**
804 * blk_queue_dma_alignment - set dma length and memory alignment
805 * @q: the request queue for the device
806 * @mask: alignment mask
807 *
808 * description:
809 * set required memory and length aligment for direct dma transactions.
810 * this is used when buiding direct io requests for the queue.
811 *
812 **/
813void blk_queue_dma_alignment(request_queue_t *q, int mask)
814{
815 q->dma_alignment = mask;
816}
817
818EXPORT_SYMBOL(blk_queue_dma_alignment);
819
820/**
821 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 * @q: The request queue for the device
823 * @tag: The tag of the request
824 *
825 * Notes:
826 * Should be used when a device returns a tag and you want to match
827 * it with a request.
828 *
829 * no locks need be held.
830 **/
831struct request *blk_queue_find_tag(request_queue_t *q, int tag)
832{
833 struct blk_queue_tag *bqt = q->queue_tags;
834
Tejun Heoba025082005-08-05 13:28:11 -0700835 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 return NULL;
837
838 return bqt->tag_index[tag];
839}
840
841EXPORT_SYMBOL(blk_queue_find_tag);
842
843/**
844 * __blk_queue_free_tags - release tag maintenance info
845 * @q: the request queue for the device
846 *
847 * Notes:
848 * blk_cleanup_queue() will take care of calling this function, if tagging
849 * has been used. So there's no need to call this directly.
850 **/
851static void __blk_queue_free_tags(request_queue_t *q)
852{
853 struct blk_queue_tag *bqt = q->queue_tags;
854
855 if (!bqt)
856 return;
857
858 if (atomic_dec_and_test(&bqt->refcnt)) {
859 BUG_ON(bqt->busy);
860 BUG_ON(!list_empty(&bqt->busy_list));
861
862 kfree(bqt->tag_index);
863 bqt->tag_index = NULL;
864
865 kfree(bqt->tag_map);
866 bqt->tag_map = NULL;
867
868 kfree(bqt);
869 }
870
871 q->queue_tags = NULL;
872 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
873}
874
875/**
876 * blk_queue_free_tags - release tag maintenance info
877 * @q: the request queue for the device
878 *
879 * Notes:
880 * This is used to disabled tagged queuing to a device, yet leave
881 * queue in function.
882 **/
883void blk_queue_free_tags(request_queue_t *q)
884{
885 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
886}
887
888EXPORT_SYMBOL(blk_queue_free_tags);
889
890static int
891init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
892{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 struct request **tag_index;
894 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700895 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 if (depth > q->nr_requests * 2) {
898 depth = q->nr_requests * 2;
899 printk(KERN_ERR "%s: adjusted depth to %d\n",
900 __FUNCTION__, depth);
901 }
902
903 tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC);
904 if (!tag_index)
905 goto fail;
906
Tejun Heof7d37d02005-06-23 00:08:50 -0700907 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -0700908 tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (!tag_map)
910 goto fail;
911
912 memset(tag_index, 0, depth * sizeof(struct request *));
Tejun Heofa72b902005-06-23 00:08:49 -0700913 memset(tag_map, 0, nr_ulongs * sizeof(unsigned long));
Tejun Heoba025082005-08-05 13:28:11 -0700914 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 tags->tag_index = tag_index;
917 tags->tag_map = tag_map;
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 return 0;
920fail:
921 kfree(tag_index);
922 return -ENOMEM;
923}
924
925/**
926 * blk_queue_init_tags - initialize the queue tag info
927 * @q: the request queue for the device
928 * @depth: the maximum queue depth supported
929 * @tags: the tag to use
930 **/
931int blk_queue_init_tags(request_queue_t *q, int depth,
932 struct blk_queue_tag *tags)
933{
934 int rc;
935
936 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
937
938 if (!tags && !q->queue_tags) {
939 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
940 if (!tags)
941 goto fail;
942
943 if (init_tag_map(q, tags, depth))
944 goto fail;
945
946 INIT_LIST_HEAD(&tags->busy_list);
947 tags->busy = 0;
948 atomic_set(&tags->refcnt, 1);
949 } else if (q->queue_tags) {
950 if ((rc = blk_queue_resize_tags(q, depth)))
951 return rc;
952 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
953 return 0;
954 } else
955 atomic_inc(&tags->refcnt);
956
957 /*
958 * assign it, all done
959 */
960 q->queue_tags = tags;
961 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
962 return 0;
963fail:
964 kfree(tags);
965 return -ENOMEM;
966}
967
968EXPORT_SYMBOL(blk_queue_init_tags);
969
970/**
971 * blk_queue_resize_tags - change the queueing depth
972 * @q: the request queue for the device
973 * @new_depth: the new max command queueing depth
974 *
975 * Notes:
976 * Must be called with the queue lock held.
977 **/
978int blk_queue_resize_tags(request_queue_t *q, int new_depth)
979{
980 struct blk_queue_tag *bqt = q->queue_tags;
981 struct request **tag_index;
982 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700983 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985 if (!bqt)
986 return -ENXIO;
987
988 /*
Tejun Heoba025082005-08-05 13:28:11 -0700989 * if we already have large enough real_max_depth. just
990 * adjust max_depth. *NOTE* as requests with tag value
991 * between new_depth and real_max_depth can be in-flight, tag
992 * map can not be shrunk blindly here.
993 */
994 if (new_depth <= bqt->real_max_depth) {
995 bqt->max_depth = new_depth;
996 return 0;
997 }
998
999 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 * save the old state info, so we can copy it back
1001 */
1002 tag_index = bqt->tag_index;
1003 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001004 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 if (init_tag_map(q, bqt, new_depth))
1007 return -ENOMEM;
1008
1009 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001010 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001011 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
1013 kfree(tag_index);
1014 kfree(tag_map);
1015 return 0;
1016}
1017
1018EXPORT_SYMBOL(blk_queue_resize_tags);
1019
1020/**
1021 * blk_queue_end_tag - end tag operations for a request
1022 * @q: the request queue for the device
1023 * @rq: the request that has completed
1024 *
1025 * Description:
1026 * Typically called when end_that_request_first() returns 0, meaning
1027 * all transfers have been done for a request. It's important to call
1028 * this function before end_that_request_last(), as that will put the
1029 * request back on the free list thus corrupting the internal tag list.
1030 *
1031 * Notes:
1032 * queue lock must be held.
1033 **/
1034void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1035{
1036 struct blk_queue_tag *bqt = q->queue_tags;
1037 int tag = rq->tag;
1038
1039 BUG_ON(tag == -1);
1040
Tejun Heoba025082005-08-05 13:28:11 -07001041 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001042 /*
1043 * This can happen after tag depth has been reduced.
1044 * FIXME: how about a warning or info message here?
1045 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 return;
1047
1048 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001049 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1050 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 return;
1052 }
1053
1054 list_del_init(&rq->queuelist);
1055 rq->flags &= ~REQ_QUEUED;
1056 rq->tag = -1;
1057
1058 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001059 printk(KERN_ERR "%s: tag %d is missing\n",
1060 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 bqt->tag_index[tag] = NULL;
1063 bqt->busy--;
1064}
1065
1066EXPORT_SYMBOL(blk_queue_end_tag);
1067
1068/**
1069 * blk_queue_start_tag - find a free tag and assign it
1070 * @q: the request queue for the device
1071 * @rq: the block request that needs tagging
1072 *
1073 * Description:
1074 * This can either be used as a stand-alone helper, or possibly be
1075 * assigned as the queue &prep_rq_fn (in which case &struct request
1076 * automagically gets a tag assigned). Note that this function
1077 * assumes that any type of request can be queued! if this is not
1078 * true for your device, you must check the request type before
1079 * calling this function. The request will also be removed from
1080 * the request queue, so it's the drivers responsibility to readd
1081 * it if it should need to be restarted for some reason.
1082 *
1083 * Notes:
1084 * queue lock must be held.
1085 **/
1086int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1087{
1088 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001089 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
1091 if (unlikely((rq->flags & REQ_QUEUED))) {
1092 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001093 "%s: request %p for device [%s] already tagged %d",
1094 __FUNCTION__, rq,
1095 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 BUG();
1097 }
1098
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001099 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1100 if (tag >= bqt->max_depth)
1101 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 __set_bit(tag, bqt->tag_map);
1104
1105 rq->flags |= REQ_QUEUED;
1106 rq->tag = tag;
1107 bqt->tag_index[tag] = rq;
1108 blkdev_dequeue_request(rq);
1109 list_add(&rq->queuelist, &bqt->busy_list);
1110 bqt->busy++;
1111 return 0;
1112}
1113
1114EXPORT_SYMBOL(blk_queue_start_tag);
1115
1116/**
1117 * blk_queue_invalidate_tags - invalidate all pending tags
1118 * @q: the request queue for the device
1119 *
1120 * Description:
1121 * Hardware conditions may dictate a need to stop all pending requests.
1122 * In this case, we will safely clear the block side of the tag queue and
1123 * readd all requests to the request queue in the right order.
1124 *
1125 * Notes:
1126 * queue lock must be held.
1127 **/
1128void blk_queue_invalidate_tags(request_queue_t *q)
1129{
1130 struct blk_queue_tag *bqt = q->queue_tags;
1131 struct list_head *tmp, *n;
1132 struct request *rq;
1133
1134 list_for_each_safe(tmp, n, &bqt->busy_list) {
1135 rq = list_entry_rq(tmp);
1136
1137 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001138 printk(KERN_ERR
1139 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 list_del_init(&rq->queuelist);
1141 rq->flags &= ~REQ_QUEUED;
1142 } else
1143 blk_queue_end_tag(q, rq);
1144
1145 rq->flags &= ~REQ_STARTED;
1146 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1147 }
1148}
1149
1150EXPORT_SYMBOL(blk_queue_invalidate_tags);
1151
Arjan van de Ven64100092006-01-06 09:46:02 +01001152static const char * const rq_flags[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 "REQ_RW",
1154 "REQ_FAILFAST",
Tejun Heo8922e162005-10-20 16:23:44 +02001155 "REQ_SORTED",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 "REQ_SOFTBARRIER",
1157 "REQ_HARDBARRIER",
Tejun Heo797e7db2006-01-06 09:51:03 +01001158 "REQ_FUA",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 "REQ_CMD",
1160 "REQ_NOMERGE",
1161 "REQ_STARTED",
1162 "REQ_DONTPREP",
1163 "REQ_QUEUED",
Tejun Heocb98fc82005-10-28 08:29:39 +02001164 "REQ_ELVPRIV",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 "REQ_PC",
1166 "REQ_BLOCK_PC",
1167 "REQ_SENSE",
1168 "REQ_FAILED",
1169 "REQ_QUIET",
1170 "REQ_SPECIAL",
1171 "REQ_DRIVE_CMD",
1172 "REQ_DRIVE_TASK",
1173 "REQ_DRIVE_TASKFILE",
1174 "REQ_PREEMPT",
1175 "REQ_PM_SUSPEND",
1176 "REQ_PM_RESUME",
1177 "REQ_PM_SHUTDOWN",
Tejun Heo797e7db2006-01-06 09:51:03 +01001178 "REQ_ORDERED_COLOR",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179};
1180
1181void blk_dump_rq_flags(struct request *rq, char *msg)
1182{
1183 int bit;
1184
1185 printk("%s: dev %s: flags = ", msg,
1186 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1187 bit = 0;
1188 do {
1189 if (rq->flags & (1 << bit))
1190 printk("%s ", rq_flags[bit]);
1191 bit++;
1192 } while (bit < __REQ_NR_BITS);
1193
1194 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1195 rq->nr_sectors,
1196 rq->current_nr_sectors);
1197 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1198
1199 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1200 printk("cdb: ");
1201 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1202 printk("%02x ", rq->cmd[bit]);
1203 printk("\n");
1204 }
1205}
1206
1207EXPORT_SYMBOL(blk_dump_rq_flags);
1208
1209void blk_recount_segments(request_queue_t *q, struct bio *bio)
1210{
1211 struct bio_vec *bv, *bvprv = NULL;
1212 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1213 int high, highprv = 1;
1214
1215 if (unlikely(!bio->bi_io_vec))
1216 return;
1217
1218 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1219 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1220 bio_for_each_segment(bv, bio, i) {
1221 /*
1222 * the trick here is making sure that a high page is never
1223 * considered part of another segment, since that might
1224 * change with the bounce page.
1225 */
1226 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1227 if (high || highprv)
1228 goto new_hw_segment;
1229 if (cluster) {
1230 if (seg_size + bv->bv_len > q->max_segment_size)
1231 goto new_segment;
1232 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1233 goto new_segment;
1234 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1235 goto new_segment;
1236 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1237 goto new_hw_segment;
1238
1239 seg_size += bv->bv_len;
1240 hw_seg_size += bv->bv_len;
1241 bvprv = bv;
1242 continue;
1243 }
1244new_segment:
1245 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1246 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1247 hw_seg_size += bv->bv_len;
1248 } else {
1249new_hw_segment:
1250 if (hw_seg_size > bio->bi_hw_front_size)
1251 bio->bi_hw_front_size = hw_seg_size;
1252 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1253 nr_hw_segs++;
1254 }
1255
1256 nr_phys_segs++;
1257 bvprv = bv;
1258 seg_size = bv->bv_len;
1259 highprv = high;
1260 }
1261 if (hw_seg_size > bio->bi_hw_back_size)
1262 bio->bi_hw_back_size = hw_seg_size;
1263 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1264 bio->bi_hw_front_size = hw_seg_size;
1265 bio->bi_phys_segments = nr_phys_segs;
1266 bio->bi_hw_segments = nr_hw_segs;
1267 bio->bi_flags |= (1 << BIO_SEG_VALID);
1268}
1269
1270
Adrian Bunk93d17d32005-06-25 14:59:10 -07001271static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 struct bio *nxt)
1273{
1274 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1275 return 0;
1276
1277 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1278 return 0;
1279 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1280 return 0;
1281
1282 /*
1283 * bio and nxt are contigous in memory, check if the queue allows
1284 * these two to be merged into one
1285 */
1286 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1287 return 1;
1288
1289 return 0;
1290}
1291
Adrian Bunk93d17d32005-06-25 14:59:10 -07001292static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 struct bio *nxt)
1294{
1295 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1296 blk_recount_segments(q, bio);
1297 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1298 blk_recount_segments(q, nxt);
1299 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1300 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1301 return 0;
1302 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1303 return 0;
1304
1305 return 1;
1306}
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308/*
1309 * map a request to scatterlist, return number of sg entries setup. Caller
1310 * must make sure sg can hold rq->nr_phys_segments entries
1311 */
1312int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1313{
1314 struct bio_vec *bvec, *bvprv;
1315 struct bio *bio;
1316 int nsegs, i, cluster;
1317
1318 nsegs = 0;
1319 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1320
1321 /*
1322 * for each bio in rq
1323 */
1324 bvprv = NULL;
1325 rq_for_each_bio(bio, rq) {
1326 /*
1327 * for each segment in bio
1328 */
1329 bio_for_each_segment(bvec, bio, i) {
1330 int nbytes = bvec->bv_len;
1331
1332 if (bvprv && cluster) {
1333 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1334 goto new_segment;
1335
1336 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1337 goto new_segment;
1338 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1339 goto new_segment;
1340
1341 sg[nsegs - 1].length += nbytes;
1342 } else {
1343new_segment:
1344 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1345 sg[nsegs].page = bvec->bv_page;
1346 sg[nsegs].length = nbytes;
1347 sg[nsegs].offset = bvec->bv_offset;
1348
1349 nsegs++;
1350 }
1351 bvprv = bvec;
1352 } /* segments in bio */
1353 } /* bios in rq */
1354
1355 return nsegs;
1356}
1357
1358EXPORT_SYMBOL(blk_rq_map_sg);
1359
1360/*
1361 * the standard queue merge functions, can be overridden with device
1362 * specific ones if so desired
1363 */
1364
1365static inline int ll_new_mergeable(request_queue_t *q,
1366 struct request *req,
1367 struct bio *bio)
1368{
1369 int nr_phys_segs = bio_phys_segments(q, bio);
1370
1371 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1372 req->flags |= REQ_NOMERGE;
1373 if (req == q->last_merge)
1374 q->last_merge = NULL;
1375 return 0;
1376 }
1377
1378 /*
1379 * A hw segment is just getting larger, bump just the phys
1380 * counter.
1381 */
1382 req->nr_phys_segments += nr_phys_segs;
1383 return 1;
1384}
1385
1386static inline int ll_new_hw_segment(request_queue_t *q,
1387 struct request *req,
1388 struct bio *bio)
1389{
1390 int nr_hw_segs = bio_hw_segments(q, bio);
1391 int nr_phys_segs = bio_phys_segments(q, bio);
1392
1393 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1394 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1395 req->flags |= REQ_NOMERGE;
1396 if (req == q->last_merge)
1397 q->last_merge = NULL;
1398 return 0;
1399 }
1400
1401 /*
1402 * This will form the start of a new hw segment. Bump both
1403 * counters.
1404 */
1405 req->nr_hw_segments += nr_hw_segs;
1406 req->nr_phys_segments += nr_phys_segs;
1407 return 1;
1408}
1409
1410static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1411 struct bio *bio)
1412{
Mike Christiedefd94b2005-12-05 02:37:06 -06001413 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 int len;
1415
Mike Christiedefd94b2005-12-05 02:37:06 -06001416 if (unlikely(blk_pc_request(req)))
1417 max_sectors = q->max_hw_sectors;
1418 else
1419 max_sectors = q->max_sectors;
1420
1421 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 req->flags |= REQ_NOMERGE;
1423 if (req == q->last_merge)
1424 q->last_merge = NULL;
1425 return 0;
1426 }
1427 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1428 blk_recount_segments(q, req->biotail);
1429 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1430 blk_recount_segments(q, bio);
1431 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1432 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1433 !BIOVEC_VIRT_OVERSIZE(len)) {
1434 int mergeable = ll_new_mergeable(q, req, bio);
1435
1436 if (mergeable) {
1437 if (req->nr_hw_segments == 1)
1438 req->bio->bi_hw_front_size = len;
1439 if (bio->bi_hw_segments == 1)
1440 bio->bi_hw_back_size = len;
1441 }
1442 return mergeable;
1443 }
1444
1445 return ll_new_hw_segment(q, req, bio);
1446}
1447
1448static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1449 struct bio *bio)
1450{
Mike Christiedefd94b2005-12-05 02:37:06 -06001451 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 int len;
1453
Mike Christiedefd94b2005-12-05 02:37:06 -06001454 if (unlikely(blk_pc_request(req)))
1455 max_sectors = q->max_hw_sectors;
1456 else
1457 max_sectors = q->max_sectors;
1458
1459
1460 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 req->flags |= REQ_NOMERGE;
1462 if (req == q->last_merge)
1463 q->last_merge = NULL;
1464 return 0;
1465 }
1466 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1467 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1468 blk_recount_segments(q, bio);
1469 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1470 blk_recount_segments(q, req->bio);
1471 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1472 !BIOVEC_VIRT_OVERSIZE(len)) {
1473 int mergeable = ll_new_mergeable(q, req, bio);
1474
1475 if (mergeable) {
1476 if (bio->bi_hw_segments == 1)
1477 bio->bi_hw_front_size = len;
1478 if (req->nr_hw_segments == 1)
1479 req->biotail->bi_hw_back_size = len;
1480 }
1481 return mergeable;
1482 }
1483
1484 return ll_new_hw_segment(q, req, bio);
1485}
1486
1487static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1488 struct request *next)
1489{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001490 int total_phys_segments;
1491 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
1493 /*
1494 * First check if the either of the requests are re-queued
1495 * requests. Can't merge them if they are.
1496 */
1497 if (req->special || next->special)
1498 return 0;
1499
1500 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001501 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 */
1503 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1504 return 0;
1505
1506 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1507 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1508 total_phys_segments--;
1509
1510 if (total_phys_segments > q->max_phys_segments)
1511 return 0;
1512
1513 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1514 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1515 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1516 /*
1517 * propagate the combined length to the end of the requests
1518 */
1519 if (req->nr_hw_segments == 1)
1520 req->bio->bi_hw_front_size = len;
1521 if (next->nr_hw_segments == 1)
1522 next->biotail->bi_hw_back_size = len;
1523 total_hw_segments--;
1524 }
1525
1526 if (total_hw_segments > q->max_hw_segments)
1527 return 0;
1528
1529 /* Merge is OK... */
1530 req->nr_phys_segments = total_phys_segments;
1531 req->nr_hw_segments = total_hw_segments;
1532 return 1;
1533}
1534
1535/*
1536 * "plug" the device if there are no outstanding requests: this will
1537 * force the transfer to start only after we have put all the requests
1538 * on the list.
1539 *
1540 * This is called with interrupts off and no requests on the queue and
1541 * with the queue lock held.
1542 */
1543void blk_plug_device(request_queue_t *q)
1544{
1545 WARN_ON(!irqs_disabled());
1546
1547 /*
1548 * don't plug a stopped queue, it must be paired with blk_start_queue()
1549 * which will restart the queueing
1550 */
1551 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))
1552 return;
1553
1554 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1555 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1556}
1557
1558EXPORT_SYMBOL(blk_plug_device);
1559
1560/*
1561 * remove the queue from the plugged list, if present. called with
1562 * queue lock held and interrupts disabled.
1563 */
1564int blk_remove_plug(request_queue_t *q)
1565{
1566 WARN_ON(!irqs_disabled());
1567
1568 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1569 return 0;
1570
1571 del_timer(&q->unplug_timer);
1572 return 1;
1573}
1574
1575EXPORT_SYMBOL(blk_remove_plug);
1576
1577/*
1578 * remove the plug and let it rip..
1579 */
1580void __generic_unplug_device(request_queue_t *q)
1581{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001582 if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 return;
1584
1585 if (!blk_remove_plug(q))
1586 return;
1587
Jens Axboe22e2c502005-06-27 10:55:12 +02001588 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589}
1590EXPORT_SYMBOL(__generic_unplug_device);
1591
1592/**
1593 * generic_unplug_device - fire a request queue
1594 * @q: The &request_queue_t in question
1595 *
1596 * Description:
1597 * Linux uses plugging to build bigger requests queues before letting
1598 * the device have at them. If a queue is plugged, the I/O scheduler
1599 * is still adding and merging requests on the queue. Once the queue
1600 * gets unplugged, the request_fn defined for the queue is invoked and
1601 * transfers started.
1602 **/
1603void generic_unplug_device(request_queue_t *q)
1604{
1605 spin_lock_irq(q->queue_lock);
1606 __generic_unplug_device(q);
1607 spin_unlock_irq(q->queue_lock);
1608}
1609EXPORT_SYMBOL(generic_unplug_device);
1610
1611static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1612 struct page *page)
1613{
1614 request_queue_t *q = bdi->unplug_io_data;
1615
1616 /*
1617 * devices don't necessarily have an ->unplug_fn defined
1618 */
1619 if (q->unplug_fn)
1620 q->unplug_fn(q);
1621}
1622
1623static void blk_unplug_work(void *data)
1624{
1625 request_queue_t *q = data;
1626
1627 q->unplug_fn(q);
1628}
1629
1630static void blk_unplug_timeout(unsigned long data)
1631{
1632 request_queue_t *q = (request_queue_t *)data;
1633
1634 kblockd_schedule_work(&q->unplug_work);
1635}
1636
1637/**
1638 * blk_start_queue - restart a previously stopped queue
1639 * @q: The &request_queue_t in question
1640 *
1641 * Description:
1642 * blk_start_queue() will clear the stop flag on the queue, and call
1643 * the request_fn for the queue if it was in a stopped state when
1644 * entered. Also see blk_stop_queue(). Queue lock must be held.
1645 **/
1646void blk_start_queue(request_queue_t *q)
1647{
1648 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1649
1650 /*
1651 * one level of recursion is ok and is much faster than kicking
1652 * the unplug handling
1653 */
1654 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1655 q->request_fn(q);
1656 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1657 } else {
1658 blk_plug_device(q);
1659 kblockd_schedule_work(&q->unplug_work);
1660 }
1661}
1662
1663EXPORT_SYMBOL(blk_start_queue);
1664
1665/**
1666 * blk_stop_queue - stop a queue
1667 * @q: The &request_queue_t in question
1668 *
1669 * Description:
1670 * The Linux block layer assumes that a block driver will consume all
1671 * entries on the request queue when the request_fn strategy is called.
1672 * Often this will not happen, because of hardware limitations (queue
1673 * depth settings). If a device driver gets a 'queue full' response,
1674 * or if it simply chooses not to queue more I/O at one point, it can
1675 * call this function to prevent the request_fn from being called until
1676 * the driver has signalled it's ready to go again. This happens by calling
1677 * blk_start_queue() to restart queue operations. Queue lock must be held.
1678 **/
1679void blk_stop_queue(request_queue_t *q)
1680{
1681 blk_remove_plug(q);
1682 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1683}
1684EXPORT_SYMBOL(blk_stop_queue);
1685
1686/**
1687 * blk_sync_queue - cancel any pending callbacks on a queue
1688 * @q: the queue
1689 *
1690 * Description:
1691 * The block layer may perform asynchronous callback activity
1692 * on a queue, such as calling the unplug function after a timeout.
1693 * A block device may call blk_sync_queue to ensure that any
1694 * such activity is cancelled, thus allowing it to release resources
1695 * the the callbacks might use. The caller must already have made sure
1696 * that its ->make_request_fn will not re-add plugging prior to calling
1697 * this function.
1698 *
1699 */
1700void blk_sync_queue(struct request_queue *q)
1701{
1702 del_timer_sync(&q->unplug_timer);
1703 kblockd_flush();
1704}
1705EXPORT_SYMBOL(blk_sync_queue);
1706
1707/**
1708 * blk_run_queue - run a single device queue
1709 * @q: The queue to run
1710 */
1711void blk_run_queue(struct request_queue *q)
1712{
1713 unsigned long flags;
1714
1715 spin_lock_irqsave(q->queue_lock, flags);
1716 blk_remove_plug(q);
Ken Chena2997382005-04-16 15:25:43 -07001717 if (!elv_queue_empty(q))
1718 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 spin_unlock_irqrestore(q->queue_lock, flags);
1720}
1721EXPORT_SYMBOL(blk_run_queue);
1722
1723/**
1724 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1725 * @q: the request queue to be released
1726 *
1727 * Description:
1728 * blk_cleanup_queue is the pair to blk_init_queue() or
1729 * blk_queue_make_request(). It should be called when a request queue is
1730 * being released; typically when a block device is being de-registered.
1731 * Currently, its primary task it to free all the &struct request
1732 * structures that were allocated to the queue and the queue itself.
1733 *
1734 * Caveat:
1735 * Hopefully the low level driver will have finished any
1736 * outstanding requests first...
1737 **/
1738void blk_cleanup_queue(request_queue_t * q)
1739{
1740 struct request_list *rl = &q->rq;
1741
1742 if (!atomic_dec_and_test(&q->refcnt))
1743 return;
1744
1745 if (q->elevator)
1746 elevator_exit(q->elevator);
1747
1748 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
1759EXPORT_SYMBOL(blk_cleanup_queue);
1760
1761static int blk_init_free_list(request_queue_t *q)
1762{
1763 struct request_list *rl = &q->rq;
1764
1765 rl->count[READ] = rl->count[WRITE] = 0;
1766 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001767 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 init_waitqueue_head(&rl->wait[READ]);
1769 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Christoph Lameter19460892005-06-23 00:08:19 -07001771 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1772 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 if (!rl->rq_pool)
1775 return -ENOMEM;
1776
1777 return 0;
1778}
1779
Al Viro8267e262005-10-21 03:20:53 -04001780request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
Christoph Lameter19460892005-06-23 00:08:19 -07001782 return blk_alloc_queue_node(gfp_mask, -1);
1783}
1784EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Al Viro8267e262005-10-21 03:20:53 -04001786request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001787{
1788 request_queue_t *q;
1789
1790 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 if (!q)
1792 return NULL;
1793
1794 memset(q, 0, sizeof(*q));
1795 init_timer(&q->unplug_timer);
1796 atomic_set(&q->refcnt, 1);
1797
1798 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1799 q->backing_dev_info.unplug_io_data = q;
1800
1801 return q;
1802}
Christoph Lameter19460892005-06-23 00:08:19 -07001803EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
1805/**
1806 * blk_init_queue - prepare a request queue for use with a block device
1807 * @rfn: The function to be called to process requests that have been
1808 * placed on the queue.
1809 * @lock: Request queue spin lock
1810 *
1811 * Description:
1812 * If a block device wishes to use the standard request handling procedures,
1813 * which sorts requests and coalesces adjacent requests, then it must
1814 * call blk_init_queue(). The function @rfn will be called when there
1815 * are requests on the queue that need to be processed. If the device
1816 * supports plugging, then @rfn may not be called immediately when requests
1817 * are available on the queue, but may be called at some time later instead.
1818 * Plugged queues are generally unplugged when a buffer belonging to one
1819 * of the requests on the queue is needed, or due to memory pressure.
1820 *
1821 * @rfn is not required, or even expected, to remove all requests off the
1822 * queue, but only as many as it can handle at a time. If it does leave
1823 * requests on the queue, it is responsible for arranging that the requests
1824 * get dealt with eventually.
1825 *
1826 * The queue spin lock must be held while manipulating the requests on the
1827 * request queue.
1828 *
1829 * Function returns a pointer to the initialized request queue, or NULL if
1830 * it didn't succeed.
1831 *
1832 * Note:
1833 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1834 * when the block device is deactivated (such as at module unload).
1835 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1838{
Christoph Lameter19460892005-06-23 00:08:19 -07001839 return blk_init_queue_node(rfn, lock, -1);
1840}
1841EXPORT_SYMBOL(blk_init_queue);
1842
1843request_queue_t *
1844blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1845{
1846 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 if (!q)
1849 return NULL;
1850
Christoph Lameter19460892005-06-23 00:08:19 -07001851 q->node = node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (blk_init_free_list(q))
1853 goto out_init;
1854
152587d2005-04-12 16:22:06 -05001855 /*
1856 * if caller didn't supply a lock, they get per-queue locking with
1857 * our embedded lock
1858 */
1859 if (!lock) {
1860 spin_lock_init(&q->__queue_lock);
1861 lock = &q->__queue_lock;
1862 }
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 q->request_fn = rfn;
1865 q->back_merge_fn = ll_back_merge_fn;
1866 q->front_merge_fn = ll_front_merge_fn;
1867 q->merge_requests_fn = ll_merge_requests_fn;
1868 q->prep_rq_fn = NULL;
1869 q->unplug_fn = generic_unplug_device;
1870 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1871 q->queue_lock = lock;
1872
1873 blk_queue_segment_boundary(q, 0xffffffff);
1874
1875 blk_queue_make_request(q, __make_request);
1876 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1877
1878 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1879 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1880
1881 /*
1882 * all done
1883 */
1884 if (!elevator_init(q, NULL)) {
1885 blk_queue_congestion_threshold(q);
1886 return q;
1887 }
1888
1889 blk_cleanup_queue(q);
1890out_init:
1891 kmem_cache_free(requestq_cachep, q);
1892 return NULL;
1893}
Christoph Lameter19460892005-06-23 00:08:19 -07001894EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896int blk_get_queue(request_queue_t *q)
1897{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001898 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 atomic_inc(&q->refcnt);
1900 return 0;
1901 }
1902
1903 return 1;
1904}
1905
1906EXPORT_SYMBOL(blk_get_queue);
1907
1908static inline void blk_free_request(request_queue_t *q, struct request *rq)
1909{
Tejun Heocb98fc82005-10-28 08:29:39 +02001910 if (rq->flags & REQ_ELVPRIV)
1911 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 mempool_free(rq, q->rq.rq_pool);
1913}
1914
Jens Axboe22e2c502005-06-27 10:55:12 +02001915static inline struct request *
Tejun Heocb98fc82005-10-28 08:29:39 +02001916blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
Linus Torvalds5dd96242005-10-28 08:56:34 -07001917 int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918{
1919 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1920
1921 if (!rq)
1922 return NULL;
1923
1924 /*
1925 * first three bits are identical in rq->flags and bio->bi_rw,
1926 * see bio.h and blkdev.h
1927 */
1928 rq->flags = rw;
1929
Tejun Heocb98fc82005-10-28 08:29:39 +02001930 if (priv) {
1931 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
1932 mempool_free(rq, q->rq.rq_pool);
1933 return NULL;
1934 }
1935 rq->flags |= REQ_ELVPRIV;
1936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Tejun Heocb98fc82005-10-28 08:29:39 +02001938 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939}
1940
1941/*
1942 * ioc_batching returns true if the ioc is a valid batching request and
1943 * should be given priority access to a request.
1944 */
1945static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1946{
1947 if (!ioc)
1948 return 0;
1949
1950 /*
1951 * Make sure the process is able to allocate at least 1 request
1952 * even if the batch times out, otherwise we could theoretically
1953 * lose wakeups.
1954 */
1955 return ioc->nr_batch_requests == q->nr_batching ||
1956 (ioc->nr_batch_requests > 0
1957 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1958}
1959
1960/*
1961 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1962 * will cause the process to be a "batcher" on all queues in the system. This
1963 * is the behaviour we want though - once it gets a wakeup it should be given
1964 * a nice run.
1965 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07001966static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
1968 if (!ioc || ioc_batching(q, ioc))
1969 return;
1970
1971 ioc->nr_batch_requests = q->nr_batching;
1972 ioc->last_waited = jiffies;
1973}
1974
1975static void __freed_request(request_queue_t *q, int rw)
1976{
1977 struct request_list *rl = &q->rq;
1978
1979 if (rl->count[rw] < queue_congestion_off_threshold(q))
1980 clear_queue_congested(q, rw);
1981
1982 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (waitqueue_active(&rl->wait[rw]))
1984 wake_up(&rl->wait[rw]);
1985
1986 blk_clear_queue_full(q, rw);
1987 }
1988}
1989
1990/*
1991 * A request has just been released. Account for it, update the full and
1992 * congestion status, wake up any waiters. Called under q->queue_lock.
1993 */
Tejun Heocb98fc82005-10-28 08:29:39 +02001994static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995{
1996 struct request_list *rl = &q->rq;
1997
1998 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02001999 if (priv)
2000 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001
2002 __freed_request(q, rw);
2003
2004 if (unlikely(rl->starved[rw ^ 1]))
2005 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006}
2007
2008#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2009/*
Nick Piggind6344532005-06-28 20:45:14 -07002010 * Get a free request, queue_lock must be held.
2011 * Returns NULL on failure, with queue_lock held.
2012 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002014static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04002015 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016{
2017 struct request *rq = NULL;
2018 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002019 struct io_context *ioc = NULL;
2020 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002022 may_queue = elv_may_queue(q, rw, bio);
2023 if (may_queue == ELV_MQUEUE_NO)
2024 goto rq_starved;
2025
2026 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2027 if (rl->count[rw]+1 >= q->nr_requests) {
2028 ioc = current_io_context(GFP_ATOMIC);
2029 /*
2030 * The queue will fill after this allocation, so set
2031 * it as full, and mark this process as "batching".
2032 * This process will be allowed to complete a batch of
2033 * requests, others will be blocked.
2034 */
2035 if (!blk_queue_full(q, rw)) {
2036 ioc_set_batching(q, ioc);
2037 blk_set_queue_full(q, rw);
2038 } else {
2039 if (may_queue != ELV_MQUEUE_MUST
2040 && !ioc_batching(q, ioc)) {
2041 /*
2042 * The queue is full and the allocating
2043 * process is not a "batcher", and not
2044 * exempted by the IO scheduler
2045 */
2046 goto out;
2047 }
2048 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002050 set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
2052
Jens Axboe082cf692005-06-28 16:35:11 +02002053 /*
2054 * Only allow batching queuers to allocate up to 50% over the defined
2055 * limit of requests, otherwise we could have thousands of requests
2056 * allocated with any setting of ->nr_requests
2057 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002058 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002059 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 rl->count[rw]++;
2062 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002063
Jens Axboe64521d12005-10-28 08:30:39 +02002064 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002065 if (priv)
2066 rl->elvpriv++;
2067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 spin_unlock_irq(q->queue_lock);
2069
Tejun Heocb98fc82005-10-28 08:29:39 +02002070 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002071 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 /*
2073 * Allocation failed presumably due to memory. Undo anything
2074 * we might have messed up.
2075 *
2076 * Allocating task should really be put onto the front of the
2077 * wait queue, but this is pretty rare.
2078 */
2079 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002080 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
2082 /*
2083 * in the very unlikely event that allocation failed and no
2084 * requests for this direction was pending, mark us starved
2085 * so that freeing of a request in the other direction will
2086 * notice us. another possible fix would be to split the
2087 * rq mempool into READ and WRITE
2088 */
2089rq_starved:
2090 if (unlikely(rl->count[rw] == 0))
2091 rl->starved[rw] = 1;
2092
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 goto out;
2094 }
2095
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002096 /*
2097 * ioc may be NULL here, and ioc_batching will be false. That's
2098 * OK, if the queue is under the request limit then requests need
2099 * not count toward the nr_batch_requests limit. There will always
2100 * be some limit enforced by BLK_BATCH_TIME.
2101 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (ioc_batching(q, ioc))
2103 ioc->nr_batch_requests--;
2104
2105 rq_init(q, rq);
2106 rq->rl = rl;
2107out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 return rq;
2109}
2110
2111/*
2112 * No available requests for this queue, unplug the device and wait for some
2113 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002114 *
2115 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002117static struct request *get_request_wait(request_queue_t *q, int rw,
2118 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 struct request *rq;
2121
Nick Piggin450991b2005-06-28 20:45:13 -07002122 rq = get_request(q, rw, bio, GFP_NOIO);
2123 while (!rq) {
2124 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 struct request_list *rl = &q->rq;
2126
2127 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2128 TASK_UNINTERRUPTIBLE);
2129
Jens Axboe22e2c502005-06-27 10:55:12 +02002130 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132 if (!rq) {
2133 struct io_context *ioc;
2134
Nick Piggind6344532005-06-28 20:45:14 -07002135 __generic_unplug_device(q);
2136 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 io_schedule();
2138
2139 /*
2140 * After sleeping, we become a "batching" process and
2141 * will be able to allocate at least one request, and
2142 * up to a big batch of them for a small period time.
2143 * See ioc_batching, ioc_set_batching
2144 */
Nick Pigginfb3cc432005-06-28 20:45:15 -07002145 ioc = current_io_context(GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002147
2148 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 }
2150 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
2153 return rq;
2154}
2155
Al Viro8267e262005-10-21 03:20:53 -04002156struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157{
2158 struct request *rq;
2159
2160 BUG_ON(rw != READ && rw != WRITE);
2161
Nick Piggind6344532005-06-28 20:45:14 -07002162 spin_lock_irq(q->queue_lock);
2163 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002164 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002165 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002166 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002167 if (!rq)
2168 spin_unlock_irq(q->queue_lock);
2169 }
2170 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
2172 return rq;
2173}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174EXPORT_SYMBOL(blk_get_request);
2175
2176/**
2177 * blk_requeue_request - put a request back on queue
2178 * @q: request queue where request should be inserted
2179 * @rq: request to be inserted
2180 *
2181 * Description:
2182 * Drivers often keep queueing requests until the hardware cannot accept
2183 * more, when that condition happens we need to put the request back
2184 * on the queue. Must be called with queue lock held.
2185 */
2186void blk_requeue_request(request_queue_t *q, struct request *rq)
2187{
2188 if (blk_rq_tagged(rq))
2189 blk_queue_end_tag(q, rq);
2190
2191 elv_requeue_request(q, rq);
2192}
2193
2194EXPORT_SYMBOL(blk_requeue_request);
2195
2196/**
2197 * blk_insert_request - insert a special request in to a request queue
2198 * @q: request queue where request should be inserted
2199 * @rq: request to be inserted
2200 * @at_head: insert request at head or tail of queue
2201 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 *
2203 * Description:
2204 * Many block devices need to execute commands asynchronously, so they don't
2205 * block the whole kernel from preemption during request execution. This is
2206 * accomplished normally by inserting aritficial requests tagged as
2207 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2208 * scheduled for actual execution by the request queue.
2209 *
2210 * We have the option of inserting the head or the tail of the queue.
2211 * Typically we use the tail for new ioctls and so forth. We use the head
2212 * of the queue for things like a QUEUE_FULL message from a device, or a
2213 * host that is unable to accept a particular command.
2214 */
2215void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002216 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217{
Tejun Heo 867d1192005-04-24 02:06:05 -05002218 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 unsigned long flags;
2220
2221 /*
2222 * tell I/O scheduler that this isn't a regular read/write (ie it
2223 * must not attempt merges on this) and that it acts as a soft
2224 * barrier
2225 */
2226 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2227
2228 rq->special = data;
2229
2230 spin_lock_irqsave(q->queue_lock, flags);
2231
2232 /*
2233 * If command is tagged, release the tag
2234 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002235 if (blk_rq_tagged(rq))
2236 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Tejun Heo 867d1192005-04-24 02:06:05 -05002238 drive_stat_acct(rq, rq->nr_sectors, 1);
2239 __elv_add_request(q, rq, where, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 if (blk_queue_plugged(q))
2242 __generic_unplug_device(q);
2243 else
2244 q->request_fn(q);
2245 spin_unlock_irqrestore(q->queue_lock, flags);
2246}
2247
2248EXPORT_SYMBOL(blk_insert_request);
2249
2250/**
2251 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2252 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002253 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 * @ubuf: the user buffer
2255 * @len: length of user data
2256 *
2257 * Description:
2258 * Data will be mapped directly for zero copy io, if possible. Otherwise
2259 * a kernel bounce buffer is used.
2260 *
2261 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2262 * still in process context.
2263 *
2264 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2265 * before being submitted to the device, as pages mapped may be out of
2266 * reach. It's the callers responsibility to make sure this happens. The
2267 * original bio must be passed back in to blk_rq_unmap_user() for proper
2268 * unmapping.
2269 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002270int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2271 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272{
2273 unsigned long uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 struct bio *bio;
Jens Axboedd1cab92005-06-20 14:06:01 +02002275 int reading;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Mike Christiedefd94b2005-12-05 02:37:06 -06002277 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002278 return -EINVAL;
2279 if (!len || !ubuf)
2280 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
Jens Axboedd1cab92005-06-20 14:06:01 +02002282 reading = rq_data_dir(rq) == READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284 /*
2285 * if alignment requirement is satisfied, map in user pages for
2286 * direct dma. else, set up kernel bounce buffers
2287 */
2288 uaddr = (unsigned long) ubuf;
2289 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
Jens Axboedd1cab92005-06-20 14:06:01 +02002290 bio = bio_map_user(q, NULL, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 else
Jens Axboedd1cab92005-06-20 14:06:01 +02002292 bio = bio_copy_user(q, uaddr, len, reading);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
2294 if (!IS_ERR(bio)) {
2295 rq->bio = rq->biotail = bio;
2296 blk_rq_bio_prep(q, rq, bio);
2297
2298 rq->buffer = rq->data = NULL;
2299 rq->data_len = len;
Jens Axboedd1cab92005-06-20 14:06:01 +02002300 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 }
2302
2303 /*
2304 * bio is the err-ptr
2305 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002306 return PTR_ERR(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
2308
2309EXPORT_SYMBOL(blk_rq_map_user);
2310
2311/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002312 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2313 * @q: request queue where request should be inserted
2314 * @rq: request to map data to
2315 * @iov: pointer to the iovec
2316 * @iov_count: number of elements in the iovec
2317 *
2318 * Description:
2319 * Data will be mapped directly for zero copy io, if possible. Otherwise
2320 * a kernel bounce buffer is used.
2321 *
2322 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2323 * still in process context.
2324 *
2325 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2326 * before being submitted to the device, as pages mapped may be out of
2327 * reach. It's the callers responsibility to make sure this happens. The
2328 * original bio must be passed back in to blk_rq_unmap_user() for proper
2329 * unmapping.
2330 */
2331int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2332 struct sg_iovec *iov, int iov_count)
2333{
2334 struct bio *bio;
2335
2336 if (!iov || iov_count <= 0)
2337 return -EINVAL;
2338
2339 /* we don't allow misaligned data like bio_map_user() does. If the
2340 * user is using sg, they're expected to know the alignment constraints
2341 * and respect them accordingly */
2342 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2343 if (IS_ERR(bio))
2344 return PTR_ERR(bio);
2345
2346 rq->bio = rq->biotail = bio;
2347 blk_rq_bio_prep(q, rq, bio);
2348 rq->buffer = rq->data = NULL;
2349 rq->data_len = bio->bi_size;
2350 return 0;
2351}
2352
2353EXPORT_SYMBOL(blk_rq_map_user_iov);
2354
2355/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 * blk_rq_unmap_user - unmap a request with user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002357 * @bio: bio to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 * @ulen: length of user buffer
2359 *
2360 * Description:
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002361 * Unmap a bio previously mapped by blk_rq_map_user().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002363int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364{
2365 int ret = 0;
2366
2367 if (bio) {
2368 if (bio_flagged(bio, BIO_USER_MAPPED))
2369 bio_unmap_user(bio);
2370 else
2371 ret = bio_uncopy_user(bio);
2372 }
2373
Jens Axboedd1cab92005-06-20 14:06:01 +02002374 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375}
2376
2377EXPORT_SYMBOL(blk_rq_unmap_user);
2378
2379/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002380 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2381 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002382 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002383 * @kbuf: the kernel buffer
2384 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002385 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002386 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002387int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002388 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002389{
Mike Christie df46b9a2005-06-20 14:04:44 +02002390 struct bio *bio;
2391
Mike Christiedefd94b2005-12-05 02:37:06 -06002392 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002393 return -EINVAL;
2394 if (!len || !kbuf)
2395 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002396
2397 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002398 if (IS_ERR(bio))
2399 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002400
Jens Axboedd1cab92005-06-20 14:06:01 +02002401 if (rq_data_dir(rq) == WRITE)
2402 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002403
Jens Axboedd1cab92005-06-20 14:06:01 +02002404 rq->bio = rq->biotail = bio;
2405 blk_rq_bio_prep(q, rq, bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002406
Jens Axboedd1cab92005-06-20 14:06:01 +02002407 rq->buffer = rq->data = NULL;
2408 rq->data_len = len;
2409 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002410}
2411
2412EXPORT_SYMBOL(blk_rq_map_kern);
2413
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002414/**
2415 * blk_execute_rq_nowait - insert a request into queue for execution
2416 * @q: queue to insert the request in
2417 * @bd_disk: matching gendisk
2418 * @rq: request to insert
2419 * @at_head: insert request at head or tail of queue
2420 * @done: I/O completion handler
2421 *
2422 * Description:
2423 * Insert a fully prepared request at the back of the io scheduler queue
2424 * for execution. Don't wait for completion.
2425 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002426void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2427 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002428 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002429{
2430 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2431
2432 rq->rq_disk = bd_disk;
2433 rq->flags |= REQ_NOMERGE;
2434 rq->end_io = done;
2435 elv_add_request(q, rq, where, 1);
2436 generic_unplug_device(q);
2437}
2438
Mike Christie6e39b692005-11-11 05:30:24 -06002439EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2440
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441/**
2442 * blk_execute_rq - insert a request into queue for execution
2443 * @q: queue to insert the request in
2444 * @bd_disk: matching gendisk
2445 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002446 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 *
2448 * Description:
2449 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002450 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 */
2452int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002453 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
2455 DECLARE_COMPLETION(wait);
2456 char sense[SCSI_SENSE_BUFFERSIZE];
2457 int err = 0;
2458
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 /*
2460 * we need an extra reference to the request, so we can look at
2461 * it after io completion
2462 */
2463 rq->ref_count++;
2464
2465 if (!rq->sense) {
2466 memset(sense, 0, sizeof(sense));
2467 rq->sense = sense;
2468 rq->sense_len = 0;
2469 }
2470
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 rq->waiting = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002472 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 wait_for_completion(&wait);
2474 rq->waiting = NULL;
2475
2476 if (rq->errors)
2477 err = -EIO;
2478
2479 return err;
2480}
2481
2482EXPORT_SYMBOL(blk_execute_rq);
2483
2484/**
2485 * blkdev_issue_flush - queue a flush
2486 * @bdev: blockdev to issue flush for
2487 * @error_sector: error sector
2488 *
2489 * Description:
2490 * Issue a flush for the block device in question. Caller can supply
2491 * room for storing the error offset in case of a flush error, if they
2492 * wish to. Caller must run wait_for_completion() on its own.
2493 */
2494int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2495{
2496 request_queue_t *q;
2497
2498 if (bdev->bd_disk == NULL)
2499 return -ENXIO;
2500
2501 q = bdev_get_queue(bdev);
2502 if (!q)
2503 return -ENXIO;
2504 if (!q->issue_flush_fn)
2505 return -EOPNOTSUPP;
2506
2507 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2508}
2509
2510EXPORT_SYMBOL(blkdev_issue_flush);
2511
Adrian Bunk93d17d32005-06-25 14:59:10 -07002512static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513{
2514 int rw = rq_data_dir(rq);
2515
2516 if (!blk_fs_request(rq) || !rq->rq_disk)
2517 return;
2518
Jens Axboed72d9042005-11-01 08:35:42 +01002519 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002520 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002521 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 disk_round_stats(rq->rq_disk);
2523 rq->rq_disk->in_flight++;
2524 }
2525}
2526
2527/*
2528 * add-request adds a request to the linked list.
2529 * queue lock is held and interrupts disabled, as we muck with the
2530 * request queue list.
2531 */
2532static inline void add_request(request_queue_t * q, struct request * req)
2533{
2534 drive_stat_acct(req, req->nr_sectors, 1);
2535
2536 if (q->activity_fn)
2537 q->activity_fn(q->activity_data, rq_data_dir(req));
2538
2539 /*
2540 * elevator indicated where it wants this request to be
2541 * inserted at elevator_merge time
2542 */
2543 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2544}
2545
2546/*
2547 * disk_round_stats() - Round off the performance stats on a struct
2548 * disk_stats.
2549 *
2550 * The average IO queue length and utilisation statistics are maintained
2551 * by observing the current state of the queue length and the amount of
2552 * time it has been in this state for.
2553 *
2554 * Normally, that accounting is done on IO completion, but that can result
2555 * in more than a second's worth of IO being accounted for within any one
2556 * second, leading to >100% utilisation. To deal with that, we call this
2557 * function to do a round-off before returning the results when reading
2558 * /proc/diskstats. This accounts immediately for all queue usage up to
2559 * the current jiffies and restarts the counters again.
2560 */
2561void disk_round_stats(struct gendisk *disk)
2562{
2563 unsigned long now = jiffies;
2564
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002565 if (now == disk->stamp)
2566 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002568 if (disk->in_flight) {
2569 __disk_stat_add(disk, time_in_queue,
2570 disk->in_flight * (now - disk->stamp));
2571 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574}
2575
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002576EXPORT_SYMBOL_GPL(disk_round_stats);
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578/*
2579 * queue lock must be held
2580 */
Mike Christie6e39b692005-11-11 05:30:24 -06002581void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582{
2583 struct request_list *rl = req->rl;
2584
2585 if (unlikely(!q))
2586 return;
2587 if (unlikely(--req->ref_count))
2588 return;
2589
Tejun Heo8922e162005-10-20 16:23:44 +02002590 elv_completed_request(q, req);
2591
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 req->rq_status = RQ_INACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 req->rl = NULL;
2594
2595 /*
2596 * Request may not have originated from ll_rw_blk. if not,
2597 * it didn't come out of our reserved rq pools
2598 */
2599 if (rl) {
2600 int rw = rq_data_dir(req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002601 int priv = req->flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 BUG_ON(!list_empty(&req->queuelist));
2604
2605 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002606 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 }
2608}
2609
Mike Christie6e39b692005-11-11 05:30:24 -06002610EXPORT_SYMBOL_GPL(__blk_put_request);
2611
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612void blk_put_request(struct request *req)
2613{
Tejun Heo8922e162005-10-20 16:23:44 +02002614 unsigned long flags;
2615 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Tejun Heo8922e162005-10-20 16:23:44 +02002617 /*
2618 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2619 * following if (q) test.
2620 */
2621 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 spin_lock_irqsave(q->queue_lock, flags);
2623 __blk_put_request(q, req);
2624 spin_unlock_irqrestore(q->queue_lock, flags);
2625 }
2626}
2627
2628EXPORT_SYMBOL(blk_put_request);
2629
2630/**
2631 * blk_end_sync_rq - executes a completion event on a request
2632 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002633 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002635void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636{
2637 struct completion *waiting = rq->waiting;
2638
2639 rq->waiting = NULL;
2640 __blk_put_request(rq->q, rq);
2641
2642 /*
2643 * complete last, if this is a stack request the process (and thus
2644 * the rq pointer) could be invalid right after this complete()
2645 */
2646 complete(waiting);
2647}
2648EXPORT_SYMBOL(blk_end_sync_rq);
2649
2650/**
2651 * blk_congestion_wait - wait for a queue to become uncongested
2652 * @rw: READ or WRITE
2653 * @timeout: timeout in jiffies
2654 *
2655 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2656 * If no queues are congested then just wait for the next request to be
2657 * returned.
2658 */
2659long blk_congestion_wait(int rw, long timeout)
2660{
2661 long ret;
2662 DEFINE_WAIT(wait);
2663 wait_queue_head_t *wqh = &congestion_wqh[rw];
2664
2665 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2666 ret = io_schedule_timeout(timeout);
2667 finish_wait(wqh, &wait);
2668 return ret;
2669}
2670
2671EXPORT_SYMBOL(blk_congestion_wait);
2672
2673/*
2674 * Has to be called with the request spinlock acquired
2675 */
2676static int attempt_merge(request_queue_t *q, struct request *req,
2677 struct request *next)
2678{
2679 if (!rq_mergeable(req) || !rq_mergeable(next))
2680 return 0;
2681
2682 /*
2683 * not contigious
2684 */
2685 if (req->sector + req->nr_sectors != next->sector)
2686 return 0;
2687
2688 if (rq_data_dir(req) != rq_data_dir(next)
2689 || req->rq_disk != next->rq_disk
2690 || next->waiting || next->special)
2691 return 0;
2692
2693 /*
2694 * If we are allowed to merge, then append bio list
2695 * from next to rq and release next. merge_requests_fn
2696 * will have updated segment counts, update sector
2697 * counts here.
2698 */
2699 if (!q->merge_requests_fn(q, req, next))
2700 return 0;
2701
2702 /*
2703 * At this point we have either done a back merge
2704 * or front merge. We need the smaller start_time of
2705 * the merged requests to be the current request
2706 * for accounting purposes.
2707 */
2708 if (time_after(req->start_time, next->start_time))
2709 req->start_time = next->start_time;
2710
2711 req->biotail->bi_next = next->bio;
2712 req->biotail = next->biotail;
2713
2714 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2715
2716 elv_merge_requests(q, req, next);
2717
2718 if (req->rq_disk) {
2719 disk_round_stats(req->rq_disk);
2720 req->rq_disk->in_flight--;
2721 }
2722
Jens Axboe22e2c502005-06-27 10:55:12 +02002723 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 __blk_put_request(q, next);
2726 return 1;
2727}
2728
2729static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2730{
2731 struct request *next = elv_latter_request(q, rq);
2732
2733 if (next)
2734 return attempt_merge(q, rq, next);
2735
2736 return 0;
2737}
2738
2739static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2740{
2741 struct request *prev = elv_former_request(q, rq);
2742
2743 if (prev)
2744 return attempt_merge(q, prev, rq);
2745
2746 return 0;
2747}
2748
Tejun Heo52d9e672006-01-06 09:49:58 +01002749static void init_request_from_bio(struct request *req, struct bio *bio)
2750{
2751 req->flags |= REQ_CMD;
2752
2753 /*
2754 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2755 */
2756 if (bio_rw_ahead(bio) || bio_failfast(bio))
2757 req->flags |= REQ_FAILFAST;
2758
2759 /*
2760 * REQ_BARRIER implies no merging, but lets make it explicit
2761 */
2762 if (unlikely(bio_barrier(bio)))
2763 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2764
2765 req->errors = 0;
2766 req->hard_sector = req->sector = bio->bi_sector;
2767 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2768 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2769 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2770 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2771 req->buffer = bio_data(bio); /* see ->buffer comment above */
2772 req->waiting = NULL;
2773 req->bio = req->biotail = bio;
2774 req->ioprio = bio_prio(bio);
2775 req->rq_disk = bio->bi_bdev->bd_disk;
2776 req->start_time = jiffies;
2777}
2778
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779static int __make_request(request_queue_t *q, struct bio *bio)
2780{
Nick Piggin450991b2005-06-28 20:45:13 -07002781 struct request *req;
Jens Axboe4a534f92005-04-16 15:25:40 -07002782 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
Jens Axboe22e2c502005-06-27 10:55:12 +02002783 unsigned short prio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 sector_t sector;
2785
2786 sector = bio->bi_sector;
2787 nr_sectors = bio_sectors(bio);
2788 cur_nr_sectors = bio_cur_sectors(bio);
Jens Axboe22e2c502005-06-27 10:55:12 +02002789 prio = bio_prio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790
2791 rw = bio_data_dir(bio);
Jens Axboe4a534f92005-04-16 15:25:40 -07002792 sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793
2794 /*
2795 * low level driver can indicate that it wants pages above a
2796 * certain limit bounced to low memory (ie for highmem, or even
2797 * ISA dma in theory)
2798 */
2799 blk_queue_bounce(q, &bio);
2800
2801 spin_lock_prefetch(q->queue_lock);
2802
2803 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002804 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 err = -EOPNOTSUPP;
2806 goto end_io;
2807 }
2808
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 spin_lock_irq(q->queue_lock);
2810
Nick Piggin450991b2005-06-28 20:45:13 -07002811 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 goto get_rq;
2813
2814 el_ret = elv_merge(q, &req, bio);
2815 switch (el_ret) {
2816 case ELEVATOR_BACK_MERGE:
2817 BUG_ON(!rq_mergeable(req));
2818
2819 if (!q->back_merge_fn(q, req, bio))
2820 break;
2821
2822 req->biotail->bi_next = bio;
2823 req->biotail = bio;
2824 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002825 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 drive_stat_acct(req, nr_sectors, 0);
2827 if (!attempt_back_merge(q, req))
2828 elv_merged_request(q, req);
2829 goto out;
2830
2831 case ELEVATOR_FRONT_MERGE:
2832 BUG_ON(!rq_mergeable(req));
2833
2834 if (!q->front_merge_fn(q, req, bio))
2835 break;
2836
2837 bio->bi_next = req->bio;
2838 req->bio = bio;
2839
2840 /*
2841 * may not be valid. if the low level driver said
2842 * it didn't need a bounce buffer then it better
2843 * not touch req->buffer either...
2844 */
2845 req->buffer = bio_data(bio);
2846 req->current_nr_sectors = cur_nr_sectors;
2847 req->hard_cur_sectors = cur_nr_sectors;
2848 req->sector = req->hard_sector = sector;
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_front_merge(q, req))
2853 elv_merged_request(q, req);
2854 goto out;
2855
Nick Piggin450991b2005-06-28 20:45:13 -07002856 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002858 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 }
2860
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002862 /*
2863 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002864 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002865 */
Nick Piggin450991b2005-06-28 20:45:13 -07002866 req = get_request_wait(q, rw, bio);
Nick Piggind6344532005-06-28 20:45:14 -07002867
Nick Piggin450991b2005-06-28 20:45:13 -07002868 /*
2869 * After dropping the lock and possibly sleeping here, our request
2870 * may now be mergeable after it had proven unmergeable (above).
2871 * We don't worry about that case for efficiency. It won't happen
2872 * often, and the elevators are able to handle it.
2873 */
Tejun Heo52d9e672006-01-06 09:49:58 +01002874 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
Nick Piggin450991b2005-06-28 20:45:13 -07002876 spin_lock_irq(q->queue_lock);
2877 if (elv_queue_empty(q))
2878 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 add_request(q, req);
2880out:
Jens Axboe4a534f92005-04-16 15:25:40 -07002881 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 __generic_unplug_device(q);
2883
2884 spin_unlock_irq(q->queue_lock);
2885 return 0;
2886
2887end_io:
2888 bio_endio(bio, nr_sectors << 9, err);
2889 return 0;
2890}
2891
2892/*
2893 * If bio->bi_dev is a partition, remap the location
2894 */
2895static inline void blk_partition_remap(struct bio *bio)
2896{
2897 struct block_device *bdev = bio->bi_bdev;
2898
2899 if (bdev != bdev->bd_contains) {
2900 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01002901 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
Jens Axboea3623572005-11-01 09:26:16 +01002903 p->sectors[rw] += bio_sectors(bio);
2904 p->ios[rw]++;
2905
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 bio->bi_sector += p->start_sect;
2907 bio->bi_bdev = bdev->bd_contains;
2908 }
2909}
2910
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911static void handle_bad_sector(struct bio *bio)
2912{
2913 char b[BDEVNAME_SIZE];
2914
2915 printk(KERN_INFO "attempt to access beyond end of device\n");
2916 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2917 bdevname(bio->bi_bdev, b),
2918 bio->bi_rw,
2919 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2920 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2921
2922 set_bit(BIO_EOF, &bio->bi_flags);
2923}
2924
2925/**
2926 * generic_make_request: hand a buffer to its device driver for I/O
2927 * @bio: The bio describing the location in memory and on the device.
2928 *
2929 * generic_make_request() is used to make I/O requests of block
2930 * devices. It is passed a &struct bio, which describes the I/O that needs
2931 * to be done.
2932 *
2933 * generic_make_request() does not return any status. The
2934 * success/failure status of the request, along with notification of
2935 * completion, is delivered asynchronously through the bio->bi_end_io
2936 * function described (one day) else where.
2937 *
2938 * The caller of generic_make_request must make sure that bi_io_vec
2939 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2940 * set to describe the device address, and the
2941 * bi_end_io and optionally bi_private are set to describe how
2942 * completion notification should be signaled.
2943 *
2944 * generic_make_request and the drivers it calls may use bi_next if this
2945 * bio happens to be merged with someone else, and may change bi_dev and
2946 * bi_sector for remaps as it sees fit. So the values of these fields
2947 * should NOT be depended on after the call to generic_make_request.
2948 */
2949void generic_make_request(struct bio *bio)
2950{
2951 request_queue_t *q;
2952 sector_t maxsector;
2953 int ret, nr_sectors = bio_sectors(bio);
2954
2955 might_sleep();
2956 /* Test device or partition size, when known. */
2957 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
2958 if (maxsector) {
2959 sector_t sector = bio->bi_sector;
2960
2961 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2962 /*
2963 * This may well happen - the kernel calls bread()
2964 * without checking the size of the device, e.g., when
2965 * mounting a device.
2966 */
2967 handle_bad_sector(bio);
2968 goto end_io;
2969 }
2970 }
2971
2972 /*
2973 * Resolve the mapping until finished. (drivers are
2974 * still free to implement/resolve their own stacking
2975 * by explicitly returning 0)
2976 *
2977 * NOTE: we don't repeat the blk_size check for each new device.
2978 * Stacking drivers are expected to know what they are doing.
2979 */
2980 do {
2981 char b[BDEVNAME_SIZE];
2982
2983 q = bdev_get_queue(bio->bi_bdev);
2984 if (!q) {
2985 printk(KERN_ERR
2986 "generic_make_request: Trying to access "
2987 "nonexistent block-device %s (%Lu)\n",
2988 bdevname(bio->bi_bdev, b),
2989 (long long) bio->bi_sector);
2990end_io:
2991 bio_endio(bio, bio->bi_size, -EIO);
2992 break;
2993 }
2994
2995 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
2996 printk("bio too big device %s (%u > %u)\n",
2997 bdevname(bio->bi_bdev, b),
2998 bio_sectors(bio),
2999 q->max_hw_sectors);
3000 goto end_io;
3001 }
3002
Nick Pigginfde6ad22005-06-23 00:08:53 -07003003 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 goto end_io;
3005
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 /*
3007 * If this device has partitions, remap block n
3008 * of partition p to block n+start(p) of the disk.
3009 */
3010 blk_partition_remap(bio);
3011
3012 ret = q->make_request_fn(q, bio);
3013 } while (ret);
3014}
3015
3016EXPORT_SYMBOL(generic_make_request);
3017
3018/**
3019 * submit_bio: submit a bio to the block device layer for I/O
3020 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3021 * @bio: The &struct bio which describes the I/O
3022 *
3023 * submit_bio() is very similar in purpose to generic_make_request(), and
3024 * uses that function to do most of the work. Both are fairly rough
3025 * interfaces, @bio must be presetup and ready for I/O.
3026 *
3027 */
3028void submit_bio(int rw, struct bio *bio)
3029{
3030 int count = bio_sectors(bio);
3031
3032 BIO_BUG_ON(!bio->bi_size);
3033 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003034 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 if (rw & WRITE)
3036 mod_page_state(pgpgout, count);
3037 else
3038 mod_page_state(pgpgin, count);
3039
3040 if (unlikely(block_dump)) {
3041 char b[BDEVNAME_SIZE];
3042 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3043 current->comm, current->pid,
3044 (rw & WRITE) ? "WRITE" : "READ",
3045 (unsigned long long)bio->bi_sector,
3046 bdevname(bio->bi_bdev,b));
3047 }
3048
3049 generic_make_request(bio);
3050}
3051
3052EXPORT_SYMBOL(submit_bio);
3053
Adrian Bunk93d17d32005-06-25 14:59:10 -07003054static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055{
3056 struct bio *bio, *prevbio = NULL;
3057 int nr_phys_segs, nr_hw_segs;
3058 unsigned int phys_size, hw_size;
3059 request_queue_t *q = rq->q;
3060
3061 if (!rq->bio)
3062 return;
3063
3064 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3065 rq_for_each_bio(bio, rq) {
3066 /* Force bio hw/phys segs to be recalculated. */
3067 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3068
3069 nr_phys_segs += bio_phys_segments(q, bio);
3070 nr_hw_segs += bio_hw_segments(q, bio);
3071 if (prevbio) {
3072 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3073 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3074
3075 if (blk_phys_contig_segment(q, prevbio, bio) &&
3076 pseg <= q->max_segment_size) {
3077 nr_phys_segs--;
3078 phys_size += prevbio->bi_size + bio->bi_size;
3079 } else
3080 phys_size = 0;
3081
3082 if (blk_hw_contig_segment(q, prevbio, bio) &&
3083 hseg <= q->max_segment_size) {
3084 nr_hw_segs--;
3085 hw_size += prevbio->bi_size + bio->bi_size;
3086 } else
3087 hw_size = 0;
3088 }
3089 prevbio = bio;
3090 }
3091
3092 rq->nr_phys_segments = nr_phys_segs;
3093 rq->nr_hw_segments = nr_hw_segs;
3094}
3095
Adrian Bunk93d17d32005-06-25 14:59:10 -07003096static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097{
3098 if (blk_fs_request(rq)) {
3099 rq->hard_sector += nsect;
3100 rq->hard_nr_sectors -= nsect;
3101
3102 /*
3103 * Move the I/O submission pointers ahead if required.
3104 */
3105 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3106 (rq->sector <= rq->hard_sector)) {
3107 rq->sector = rq->hard_sector;
3108 rq->nr_sectors = rq->hard_nr_sectors;
3109 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3110 rq->current_nr_sectors = rq->hard_cur_sectors;
3111 rq->buffer = bio_data(rq->bio);
3112 }
3113
3114 /*
3115 * if total number of sectors is less than the first segment
3116 * size, something has gone terribly wrong
3117 */
3118 if (rq->nr_sectors < rq->current_nr_sectors) {
3119 printk("blk: request botched\n");
3120 rq->nr_sectors = rq->current_nr_sectors;
3121 }
3122 }
3123}
3124
3125static int __end_that_request_first(struct request *req, int uptodate,
3126 int nr_bytes)
3127{
3128 int total_bytes, bio_nbytes, error, next_idx = 0;
3129 struct bio *bio;
3130
3131 /*
3132 * extend uptodate bool to allow < 0 value to be direct io error
3133 */
3134 error = 0;
3135 if (end_io_error(uptodate))
3136 error = !uptodate ? -EIO : uptodate;
3137
3138 /*
3139 * for a REQ_BLOCK_PC request, we want to carry any eventual
3140 * sense key with us all the way through
3141 */
3142 if (!blk_pc_request(req))
3143 req->errors = 0;
3144
3145 if (!uptodate) {
3146 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3147 printk("end_request: I/O error, dev %s, sector %llu\n",
3148 req->rq_disk ? req->rq_disk->disk_name : "?",
3149 (unsigned long long)req->sector);
3150 }
3151
Jens Axboed72d9042005-11-01 08:35:42 +01003152 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003153 const int rw = rq_data_dir(req);
3154
Jens Axboe53e86062006-01-17 11:09:27 +01003155 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003156 }
3157
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 total_bytes = bio_nbytes = 0;
3159 while ((bio = req->bio) != NULL) {
3160 int nbytes;
3161
3162 if (nr_bytes >= bio->bi_size) {
3163 req->bio = bio->bi_next;
3164 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003165 if (!ordered_bio_endio(req, bio, nbytes, error))
3166 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 next_idx = 0;
3168 bio_nbytes = 0;
3169 } else {
3170 int idx = bio->bi_idx + next_idx;
3171
3172 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3173 blk_dump_rq_flags(req, "__end_that");
3174 printk("%s: bio idx %d >= vcnt %d\n",
3175 __FUNCTION__,
3176 bio->bi_idx, bio->bi_vcnt);
3177 break;
3178 }
3179
3180 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3181 BIO_BUG_ON(nbytes > bio->bi_size);
3182
3183 /*
3184 * not a complete bvec done
3185 */
3186 if (unlikely(nbytes > nr_bytes)) {
3187 bio_nbytes += nr_bytes;
3188 total_bytes += nr_bytes;
3189 break;
3190 }
3191
3192 /*
3193 * advance to the next vector
3194 */
3195 next_idx++;
3196 bio_nbytes += nbytes;
3197 }
3198
3199 total_bytes += nbytes;
3200 nr_bytes -= nbytes;
3201
3202 if ((bio = req->bio)) {
3203 /*
3204 * end more in this run, or just return 'not-done'
3205 */
3206 if (unlikely(nr_bytes <= 0))
3207 break;
3208 }
3209 }
3210
3211 /*
3212 * completely done
3213 */
3214 if (!req->bio)
3215 return 0;
3216
3217 /*
3218 * if the request wasn't completed, update state
3219 */
3220 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003221 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3222 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 bio->bi_idx += next_idx;
3224 bio_iovec(bio)->bv_offset += nr_bytes;
3225 bio_iovec(bio)->bv_len -= nr_bytes;
3226 }
3227
3228 blk_recalc_rq_sectors(req, total_bytes >> 9);
3229 blk_recalc_rq_segments(req);
3230 return 1;
3231}
3232
3233/**
3234 * end_that_request_first - end I/O on a request
3235 * @req: the request being processed
3236 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3237 * @nr_sectors: number of sectors to end I/O on
3238 *
3239 * Description:
3240 * Ends I/O on a number of sectors attached to @req, and sets it up
3241 * for the next range of segments (if any) in the cluster.
3242 *
3243 * Return:
3244 * 0 - we are done with this request, call end_that_request_last()
3245 * 1 - still buffers pending for this request
3246 **/
3247int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3248{
3249 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3250}
3251
3252EXPORT_SYMBOL(end_that_request_first);
3253
3254/**
3255 * end_that_request_chunk - end I/O on a request
3256 * @req: the request being processed
3257 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3258 * @nr_bytes: number of bytes to complete
3259 *
3260 * Description:
3261 * Ends I/O on a number of bytes attached to @req, and sets it up
3262 * for the next range of segments (if any). Like end_that_request_first(),
3263 * but deals with bytes instead of sectors.
3264 *
3265 * Return:
3266 * 0 - we are done with this request, call end_that_request_last()
3267 * 1 - still buffers pending for this request
3268 **/
3269int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3270{
3271 return __end_that_request_first(req, uptodate, nr_bytes);
3272}
3273
3274EXPORT_SYMBOL(end_that_request_chunk);
3275
3276/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003277 * splice the completion data to a local structure and hand off to
3278 * process_completion_queue() to complete the requests
3279 */
3280static void blk_done_softirq(struct softirq_action *h)
3281{
3282 struct list_head *cpu_list;
3283 LIST_HEAD(local_list);
3284
3285 local_irq_disable();
3286 cpu_list = &__get_cpu_var(blk_cpu_done);
3287 list_splice_init(cpu_list, &local_list);
3288 local_irq_enable();
3289
3290 while (!list_empty(&local_list)) {
3291 struct request *rq = list_entry(local_list.next, struct request, donelist);
3292
3293 list_del_init(&rq->donelist);
3294 rq->q->softirq_done_fn(rq);
3295 }
3296}
3297
3298#ifdef CONFIG_HOTPLUG_CPU
3299
3300static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3301 void *hcpu)
3302{
3303 /*
3304 * If a CPU goes away, splice its entries to the current CPU
3305 * and trigger a run of the softirq
3306 */
3307 if (action == CPU_DEAD) {
3308 int cpu = (unsigned long) hcpu;
3309
3310 local_irq_disable();
3311 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3312 &__get_cpu_var(blk_cpu_done));
3313 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3314 local_irq_enable();
3315 }
3316
3317 return NOTIFY_OK;
3318}
3319
3320
3321static struct notifier_block __devinitdata blk_cpu_notifier = {
3322 .notifier_call = blk_cpu_notify,
3323};
3324
3325#endif /* CONFIG_HOTPLUG_CPU */
3326
3327/**
3328 * blk_complete_request - end I/O on a request
3329 * @req: the request being processed
3330 *
3331 * Description:
3332 * Ends all I/O on a request. It does not handle partial completions,
3333 * unless the driver actually implements this in its completionc callback
3334 * through requeueing. Theh actual completion happens out-of-order,
3335 * through a softirq handler. The user must have registered a completion
3336 * callback through blk_queue_softirq_done().
3337 **/
3338
3339void blk_complete_request(struct request *req)
3340{
3341 struct list_head *cpu_list;
3342 unsigned long flags;
3343
3344 BUG_ON(!req->q->softirq_done_fn);
3345
3346 local_irq_save(flags);
3347
3348 cpu_list = &__get_cpu_var(blk_cpu_done);
3349 list_add_tail(&req->donelist, cpu_list);
3350 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3351
3352 local_irq_restore(flags);
3353}
3354
3355EXPORT_SYMBOL(blk_complete_request);
3356
3357/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 * queue lock must be held
3359 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003360void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361{
3362 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003363 int error;
3364
3365 /*
3366 * extend uptodate bool to allow < 0 value to be direct io error
3367 */
3368 error = 0;
3369 if (end_io_error(uptodate))
3370 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371
3372 if (unlikely(laptop_mode) && blk_fs_request(req))
3373 laptop_io_completion();
3374
3375 if (disk && blk_fs_request(req)) {
3376 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003377 const int rw = rq_data_dir(req);
3378
3379 __disk_stat_inc(disk, ios[rw]);
3380 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 disk_round_stats(disk);
3382 disk->in_flight--;
3383 }
3384 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003385 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 else
3387 __blk_put_request(req->q, req);
3388}
3389
3390EXPORT_SYMBOL(end_that_request_last);
3391
3392void end_request(struct request *req, int uptodate)
3393{
3394 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3395 add_disk_randomness(req->rq_disk);
3396 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003397 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 }
3399}
3400
3401EXPORT_SYMBOL(end_request);
3402
3403void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3404{
3405 /* first three bits are identical in rq->flags and bio->bi_rw */
3406 rq->flags |= (bio->bi_rw & 7);
3407
3408 rq->nr_phys_segments = bio_phys_segments(q, bio);
3409 rq->nr_hw_segments = bio_hw_segments(q, bio);
3410 rq->current_nr_sectors = bio_cur_sectors(bio);
3411 rq->hard_cur_sectors = rq->current_nr_sectors;
3412 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3413 rq->buffer = bio_data(bio);
3414
3415 rq->bio = rq->biotail = bio;
3416}
3417
3418EXPORT_SYMBOL(blk_rq_bio_prep);
3419
3420int kblockd_schedule_work(struct work_struct *work)
3421{
3422 return queue_work(kblockd_workqueue, work);
3423}
3424
3425EXPORT_SYMBOL(kblockd_schedule_work);
3426
3427void kblockd_flush(void)
3428{
3429 flush_workqueue(kblockd_workqueue);
3430}
3431EXPORT_SYMBOL(kblockd_flush);
3432
3433int __init blk_dev_init(void)
3434{
Jens Axboeff856ba2006-01-09 16:02:34 +01003435 int i;
3436
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 kblockd_workqueue = create_workqueue("kblockd");
3438 if (!kblockd_workqueue)
3439 panic("Failed to create kblockd\n");
3440
3441 request_cachep = kmem_cache_create("blkdev_requests",
3442 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3443
3444 requestq_cachep = kmem_cache_create("blkdev_queue",
3445 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3446
3447 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3448 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3449
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -08003450 for_each_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003451 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3452
3453 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
3454#ifdef CONFIG_HOTPLUG_CPU
3455 register_cpu_notifier(&blk_cpu_notifier);
3456#endif
3457
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 blk_max_low_pfn = max_low_pfn;
3459 blk_max_pfn = max_pfn;
3460
3461 return 0;
3462}
3463
3464/*
3465 * IO Context helper functions
3466 */
3467void put_io_context(struct io_context *ioc)
3468{
3469 if (ioc == NULL)
3470 return;
3471
3472 BUG_ON(atomic_read(&ioc->refcount) == 0);
3473
3474 if (atomic_dec_and_test(&ioc->refcount)) {
3475 if (ioc->aic && ioc->aic->dtor)
3476 ioc->aic->dtor(ioc->aic);
3477 if (ioc->cic && ioc->cic->dtor)
3478 ioc->cic->dtor(ioc->cic);
3479
3480 kmem_cache_free(iocontext_cachep, ioc);
3481 }
3482}
3483EXPORT_SYMBOL(put_io_context);
3484
3485/* Called by the exitting task */
3486void exit_io_context(void)
3487{
3488 unsigned long flags;
3489 struct io_context *ioc;
3490
3491 local_irq_save(flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02003492 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 ioc = current->io_context;
3494 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003495 ioc->task = NULL;
3496 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 local_irq_restore(flags);
3498
3499 if (ioc->aic && ioc->aic->exit)
3500 ioc->aic->exit(ioc->aic);
3501 if (ioc->cic && ioc->cic->exit)
3502 ioc->cic->exit(ioc->cic);
3503
3504 put_io_context(ioc);
3505}
3506
3507/*
3508 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003509 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003511 * This returned IO context doesn't have a specifically elevated refcount,
3512 * but since the current task itself holds a reference, the context can be
3513 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 */
Al Viro8267e262005-10-21 03:20:53 -04003515struct io_context *current_io_context(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516{
3517 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518 struct io_context *ret;
3519
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003521 if (likely(ret))
3522 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523
3524 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3525 if (ret) {
3526 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003527 ret->task = current;
3528 ret->set_ioprio = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529 ret->last_waited = jiffies; /* doesn't matter... */
3530 ret->nr_batch_requests = 0; /* because this is 0 */
3531 ret->aic = NULL;
3532 ret->cic = NULL;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003533 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 }
3535
3536 return ret;
3537}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003538EXPORT_SYMBOL(current_io_context);
3539
3540/*
3541 * If the current task has no IO context then create one and initialise it.
3542 * If it does have a context, take a ref on it.
3543 *
3544 * This is always called in the context of the task which submitted the I/O.
3545 */
Al Viro8267e262005-10-21 03:20:53 -04003546struct io_context *get_io_context(gfp_t gfp_flags)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003547{
3548 struct io_context *ret;
3549 ret = current_io_context(gfp_flags);
3550 if (likely(ret))
3551 atomic_inc(&ret->refcount);
3552 return ret;
3553}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554EXPORT_SYMBOL(get_io_context);
3555
3556void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3557{
3558 struct io_context *src = *psrc;
3559 struct io_context *dst = *pdst;
3560
3561 if (src) {
3562 BUG_ON(atomic_read(&src->refcount) == 0);
3563 atomic_inc(&src->refcount);
3564 put_io_context(dst);
3565 *pdst = src;
3566 }
3567}
3568EXPORT_SYMBOL(copy_io_context);
3569
3570void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3571{
3572 struct io_context *temp;
3573 temp = *ioc1;
3574 *ioc1 = *ioc2;
3575 *ioc2 = temp;
3576}
3577EXPORT_SYMBOL(swap_io_context);
3578
3579/*
3580 * sysfs parts below
3581 */
3582struct queue_sysfs_entry {
3583 struct attribute attr;
3584 ssize_t (*show)(struct request_queue *, char *);
3585 ssize_t (*store)(struct request_queue *, const char *, size_t);
3586};
3587
3588static ssize_t
3589queue_var_show(unsigned int var, char *page)
3590{
3591 return sprintf(page, "%d\n", var);
3592}
3593
3594static ssize_t
3595queue_var_store(unsigned long *var, const char *page, size_t count)
3596{
3597 char *p = (char *) page;
3598
3599 *var = simple_strtoul(p, &p, 10);
3600 return count;
3601}
3602
3603static ssize_t queue_requests_show(struct request_queue *q, char *page)
3604{
3605 return queue_var_show(q->nr_requests, (page));
3606}
3607
3608static ssize_t
3609queue_requests_store(struct request_queue *q, const char *page, size_t count)
3610{
3611 struct request_list *rl = &q->rq;
3612
3613 int ret = queue_var_store(&q->nr_requests, page, count);
3614 if (q->nr_requests < BLKDEV_MIN_RQ)
3615 q->nr_requests = BLKDEV_MIN_RQ;
3616 blk_queue_congestion_threshold(q);
3617
3618 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3619 set_queue_congested(q, READ);
3620 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3621 clear_queue_congested(q, READ);
3622
3623 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3624 set_queue_congested(q, WRITE);
3625 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3626 clear_queue_congested(q, WRITE);
3627
3628 if (rl->count[READ] >= q->nr_requests) {
3629 blk_set_queue_full(q, READ);
3630 } else if (rl->count[READ]+1 <= q->nr_requests) {
3631 blk_clear_queue_full(q, READ);
3632 wake_up(&rl->wait[READ]);
3633 }
3634
3635 if (rl->count[WRITE] >= q->nr_requests) {
3636 blk_set_queue_full(q, WRITE);
3637 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3638 blk_clear_queue_full(q, WRITE);
3639 wake_up(&rl->wait[WRITE]);
3640 }
3641 return ret;
3642}
3643
3644static ssize_t queue_ra_show(struct request_queue *q, char *page)
3645{
3646 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3647
3648 return queue_var_show(ra_kb, (page));
3649}
3650
3651static ssize_t
3652queue_ra_store(struct request_queue *q, const char *page, size_t count)
3653{
3654 unsigned long ra_kb;
3655 ssize_t ret = queue_var_store(&ra_kb, page, count);
3656
3657 spin_lock_irq(q->queue_lock);
3658 if (ra_kb > (q->max_sectors >> 1))
3659 ra_kb = (q->max_sectors >> 1);
3660
3661 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3662 spin_unlock_irq(q->queue_lock);
3663
3664 return ret;
3665}
3666
3667static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3668{
3669 int max_sectors_kb = q->max_sectors >> 1;
3670
3671 return queue_var_show(max_sectors_kb, (page));
3672}
3673
3674static ssize_t
3675queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3676{
3677 unsigned long max_sectors_kb,
3678 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3679 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3680 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3681 int ra_kb;
3682
3683 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3684 return -EINVAL;
3685 /*
3686 * Take the queue lock to update the readahead and max_sectors
3687 * values synchronously:
3688 */
3689 spin_lock_irq(q->queue_lock);
3690 /*
3691 * Trim readahead window as well, if necessary:
3692 */
3693 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3694 if (ra_kb > max_sectors_kb)
3695 q->backing_dev_info.ra_pages =
3696 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3697
3698 q->max_sectors = max_sectors_kb << 1;
3699 spin_unlock_irq(q->queue_lock);
3700
3701 return ret;
3702}
3703
3704static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3705{
3706 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3707
3708 return queue_var_show(max_hw_sectors_kb, (page));
3709}
3710
3711
3712static struct queue_sysfs_entry queue_requests_entry = {
3713 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3714 .show = queue_requests_show,
3715 .store = queue_requests_store,
3716};
3717
3718static struct queue_sysfs_entry queue_ra_entry = {
3719 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3720 .show = queue_ra_show,
3721 .store = queue_ra_store,
3722};
3723
3724static struct queue_sysfs_entry queue_max_sectors_entry = {
3725 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3726 .show = queue_max_sectors_show,
3727 .store = queue_max_sectors_store,
3728};
3729
3730static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3731 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3732 .show = queue_max_hw_sectors_show,
3733};
3734
3735static struct queue_sysfs_entry queue_iosched_entry = {
3736 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3737 .show = elv_iosched_show,
3738 .store = elv_iosched_store,
3739};
3740
3741static struct attribute *default_attrs[] = {
3742 &queue_requests_entry.attr,
3743 &queue_ra_entry.attr,
3744 &queue_max_hw_sectors_entry.attr,
3745 &queue_max_sectors_entry.attr,
3746 &queue_iosched_entry.attr,
3747 NULL,
3748};
3749
3750#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3751
3752static ssize_t
3753queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3754{
3755 struct queue_sysfs_entry *entry = to_queue(attr);
3756 struct request_queue *q;
3757
3758 q = container_of(kobj, struct request_queue, kobj);
3759 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003760 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761
3762 return entry->show(q, page);
3763}
3764
3765static ssize_t
3766queue_attr_store(struct kobject *kobj, struct attribute *attr,
3767 const char *page, size_t length)
3768{
3769 struct queue_sysfs_entry *entry = to_queue(attr);
3770 struct request_queue *q;
3771
3772 q = container_of(kobj, struct request_queue, kobj);
3773 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003774 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775
3776 return entry->store(q, page, length);
3777}
3778
3779static struct sysfs_ops queue_sysfs_ops = {
3780 .show = queue_attr_show,
3781 .store = queue_attr_store,
3782};
3783
Adrian Bunk93d17d32005-06-25 14:59:10 -07003784static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 .sysfs_ops = &queue_sysfs_ops,
3786 .default_attrs = default_attrs,
3787};
3788
3789int blk_register_queue(struct gendisk *disk)
3790{
3791 int ret;
3792
3793 request_queue_t *q = disk->queue;
3794
3795 if (!q || !q->request_fn)
3796 return -ENXIO;
3797
3798 q->kobj.parent = kobject_get(&disk->kobj);
3799 if (!q->kobj.parent)
3800 return -EBUSY;
3801
3802 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
3803 q->kobj.ktype = &queue_ktype;
3804
3805 ret = kobject_register(&q->kobj);
3806 if (ret < 0)
3807 return ret;
3808
3809 ret = elv_register_queue(q);
3810 if (ret) {
3811 kobject_unregister(&q->kobj);
3812 return ret;
3813 }
3814
3815 return 0;
3816}
3817
3818void blk_unregister_queue(struct gendisk *disk)
3819{
3820 request_queue_t *q = disk->queue;
3821
3822 if (q && q->request_fn) {
3823 elv_unregister_queue(q);
3824
3825 kobject_unregister(&q->kobj);
3826 kobject_put(&disk->kobj);
3827 }
3828}