blob: a4ff3271d4a88b984b3da89821cbc6a2fd65885a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
7 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
8 */
9
10/*
11 * This handles all read/write requests to block devices
12 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/kernel.h>
14#include <linux/module.h>
15#include <linux/backing-dev.h>
16#include <linux/bio.h>
17#include <linux/blkdev.h>
18#include <linux/highmem.h>
19#include <linux/mm.h>
20#include <linux/kernel_stat.h>
21#include <linux/string.h>
22#include <linux/init.h>
23#include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
24#include <linux/completion.h>
25#include <linux/slab.h>
26#include <linux/swap.h>
27#include <linux/writeback.h>
Jens Axboeff856ba2006-01-09 16:02:34 +010028#include <linux/interrupt.h>
29#include <linux/cpu.h>
Jens Axboe2056a782006-03-23 20:00:26 +010030#include <linux/blktrace_api.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32/*
33 * for max sense size
34 */
35#include <scsi/scsi_cmnd.h>
36
David Howells65f27f32006-11-22 14:55:48 +000037static void blk_unplug_work(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -070038static void blk_unplug_timeout(unsigned long data);
Adrian Bunk93d17d32005-06-25 14:59:10 -070039static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
Tejun Heo52d9e672006-01-06 09:49:58 +010040static void init_request_from_bio(struct request *req, struct bio *bio);
41static int __make_request(request_queue_t *q, struct bio *bio);
Jens Axboeb5deef92006-07-19 23:39:40 +020042static struct io_context *current_io_context(gfp_t gfp_flags, int node);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44/*
45 * For the allocated request tables
46 */
Christoph Lametere18b8902006-12-06 20:33:20 -080047static struct kmem_cache *request_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49/*
50 * For queue allocation
51 */
Christoph Lametere18b8902006-12-06 20:33:20 -080052static struct kmem_cache *requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * For io context allocations
56 */
Christoph Lametere18b8902006-12-06 20:33:20 -080057static struct kmem_cache *iocontext_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/*
60 * Controlling structure to kblockd
61 */
Jens Axboeff856ba2006-01-09 16:02:34 +010062static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64unsigned long blk_max_low_pfn, blk_max_pfn;
65
66EXPORT_SYMBOL(blk_max_low_pfn);
67EXPORT_SYMBOL(blk_max_pfn);
68
Jens Axboeff856ba2006-01-09 16:02:34 +010069static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071/* Amount of time in which a process may batch requests */
72#define BLK_BATCH_TIME (HZ/50UL)
73
74/* Number of requests a "batching" process may submit */
75#define BLK_BATCH_REQ 32
76
77/*
78 * Return the threshold (number of used requests) at which the queue is
79 * considered to be congested. It include a little hysteresis to keep the
80 * context switch rate down.
81 */
82static inline int queue_congestion_on_threshold(struct request_queue *q)
83{
84 return q->nr_congestion_on;
85}
86
87/*
88 * The threshold at which a queue is considered to be uncongested
89 */
90static inline int queue_congestion_off_threshold(struct request_queue *q)
91{
92 return q->nr_congestion_off;
93}
94
95static void blk_queue_congestion_threshold(struct request_queue *q)
96{
97 int nr;
98
99 nr = q->nr_requests - (q->nr_requests / 8) + 1;
100 if (nr > q->nr_requests)
101 nr = q->nr_requests;
102 q->nr_congestion_on = nr;
103
104 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
105 if (nr < 1)
106 nr = 1;
107 q->nr_congestion_off = nr;
108}
109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/**
111 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
112 * @bdev: device
113 *
114 * Locates the passed device's request queue and returns the address of its
115 * backing_dev_info
116 *
117 * Will return NULL if the request queue cannot be located.
118 */
119struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
120{
121 struct backing_dev_info *ret = NULL;
122 request_queue_t *q = bdev_get_queue(bdev);
123
124 if (q)
125 ret = &q->backing_dev_info;
126 return ret;
127}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128EXPORT_SYMBOL(blk_get_backing_dev_info);
129
130void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
131{
132 q->activity_fn = fn;
133 q->activity_data = data;
134}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135EXPORT_SYMBOL(blk_queue_activity_fn);
136
137/**
138 * blk_queue_prep_rq - set a prepare_request function for queue
139 * @q: queue
140 * @pfn: prepare_request function
141 *
142 * It's possible for a queue to register a prepare_request callback which
143 * is invoked before the request is handed to the request_fn. The goal of
144 * the function is to prepare a request for I/O, it can be used to build a
145 * cdb from the request data for instance.
146 *
147 */
148void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
149{
150 q->prep_rq_fn = pfn;
151}
152
153EXPORT_SYMBOL(blk_queue_prep_rq);
154
155/**
156 * blk_queue_merge_bvec - set a merge_bvec function for queue
157 * @q: queue
158 * @mbfn: merge_bvec_fn
159 *
160 * Usually queues have static limitations on the max sectors or segments that
161 * we can put in a request. Stacking drivers may have some settings that
162 * are dynamic, and thus we have to query the queue whether it is ok to
163 * add a new bio_vec to a bio at a given offset or not. If the block device
164 * has such limitations, it needs to register a merge_bvec_fn to control
165 * the size of bio's sent to it. Note that a block device *must* allow a
166 * single page to be added to an empty bio. The block device driver may want
167 * to use the bio_split() function to deal with these bio's. By default
168 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
169 * honored.
170 */
171void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
172{
173 q->merge_bvec_fn = mbfn;
174}
175
176EXPORT_SYMBOL(blk_queue_merge_bvec);
177
Jens Axboeff856ba2006-01-09 16:02:34 +0100178void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
179{
180 q->softirq_done_fn = fn;
181}
182
183EXPORT_SYMBOL(blk_queue_softirq_done);
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/**
186 * blk_queue_make_request - define an alternate make_request function for a device
187 * @q: the request queue for the device to be affected
188 * @mfn: the alternate make_request function
189 *
190 * Description:
191 * The normal way for &struct bios to be passed to a device
192 * driver is for them to be collected into requests on a request
193 * queue, and then to allow the device driver to select requests
194 * off that queue when it is ready. This works well for many block
195 * devices. However some block devices (typically virtual devices
196 * such as md or lvm) do not benefit from the processing on the
197 * request queue, and are served best by having the requests passed
198 * directly to them. This can be achieved by providing a function
199 * to blk_queue_make_request().
200 *
201 * Caveat:
202 * The driver that does this *must* be able to deal appropriately
203 * with buffers in "highmemory". This can be accomplished by either calling
204 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
205 * blk_queue_bounce() to create a buffer in normal memory.
206 **/
207void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
208{
209 /*
210 * set defaults
211 */
212 q->nr_requests = BLKDEV_MAX_RQ;
Stuart McLaren309c0a12005-09-06 15:17:47 -0700213 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
214 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 q->make_request_fn = mfn;
216 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
217 q->backing_dev_info.state = 0;
218 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
Mike Christiedefd94b2005-12-05 02:37:06 -0600219 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 blk_queue_hardsect_size(q, 512);
221 blk_queue_dma_alignment(q, 511);
222 blk_queue_congestion_threshold(q);
223 q->nr_batching = BLK_BATCH_REQ;
224
225 q->unplug_thresh = 4; /* hmm */
226 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
227 if (q->unplug_delay == 0)
228 q->unplug_delay = 1;
229
David Howells65f27f32006-11-22 14:55:48 +0000230 INIT_WORK(&q->unplug_work, blk_unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
232 q->unplug_timer.function = blk_unplug_timeout;
233 q->unplug_timer.data = (unsigned long)q;
234
235 /*
236 * by default assume old behaviour and bounce for any highmem page
237 */
238 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
239
240 blk_queue_activity_fn(q, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243EXPORT_SYMBOL(blk_queue_make_request);
244
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200245static void rq_init(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboeff856ba2006-01-09 16:02:34 +0100248 INIT_LIST_HEAD(&rq->donelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 rq->errors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 rq->bio = rq->biotail = NULL;
Jens Axboe2e662b62006-07-13 11:55:04 +0200252 INIT_HLIST_NODE(&rq->hash);
253 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe22e2c502005-06-27 10:55:12 +0200254 rq->ioprio = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 rq->buffer = NULL;
256 rq->ref_count = 1;
257 rq->q = q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 rq->special = NULL;
259 rq->data_len = 0;
260 rq->data = NULL;
Mike Christie df46b9a2005-06-20 14:04:44 +0200261 rq->nr_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 rq->sense = NULL;
263 rq->end_io = NULL;
264 rq->end_io_data = NULL;
Jens Axboeff856ba2006-01-09 16:02:34 +0100265 rq->completion_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
268/**
269 * blk_queue_ordered - does this queue support ordered writes
Tejun Heo797e7db2006-01-06 09:51:03 +0100270 * @q: the request queue
271 * @ordered: one of QUEUE_ORDERED_*
Jens Axboefddfdea2006-01-31 15:24:34 +0100272 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 *
274 * Description:
275 * For journalled file systems, doing ordered writes on a commit
276 * block instead of explicitly doing wait_on_buffer (which is bad
277 * for performance) can be a big win. Block drivers supporting this
278 * feature should call this function and indicate so.
279 *
280 **/
Tejun Heo797e7db2006-01-06 09:51:03 +0100281int blk_queue_ordered(request_queue_t *q, unsigned ordered,
282 prepare_flush_fn *prepare_flush_fn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
Tejun Heo797e7db2006-01-06 09:51:03 +0100284 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
285 prepare_flush_fn == NULL) {
286 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
287 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
Tejun Heo797e7db2006-01-06 09:51:03 +0100289
290 if (ordered != QUEUE_ORDERED_NONE &&
291 ordered != QUEUE_ORDERED_DRAIN &&
292 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
293 ordered != QUEUE_ORDERED_DRAIN_FUA &&
294 ordered != QUEUE_ORDERED_TAG &&
295 ordered != QUEUE_ORDERED_TAG_FLUSH &&
296 ordered != QUEUE_ORDERED_TAG_FUA) {
297 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
298 return -EINVAL;
299 }
300
Tetsuo Takata60481b12006-01-24 10:34:36 +0100301 q->ordered = ordered;
Tejun Heo797e7db2006-01-06 09:51:03 +0100302 q->next_ordered = ordered;
303 q->prepare_flush_fn = prepare_flush_fn;
304
305 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
308EXPORT_SYMBOL(blk_queue_ordered);
309
310/**
311 * blk_queue_issue_flush_fn - set function for issuing a flush
312 * @q: the request queue
313 * @iff: the function to be called issuing the flush
314 *
315 * Description:
316 * If a driver supports issuing a flush command, the support is notified
317 * to the block layer by defining it through this call.
318 *
319 **/
320void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
321{
322 q->issue_flush_fn = iff;
323}
324
325EXPORT_SYMBOL(blk_queue_issue_flush_fn);
326
327/*
328 * Cache flushing for ordered writes handling
329 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100330inline unsigned blk_ordered_cur_seq(request_queue_t *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Tejun Heo797e7db2006-01-06 09:51:03 +0100332 if (!q->ordseq)
333 return 0;
334 return 1 << ffz(q->ordseq);
335}
336
337unsigned blk_ordered_req_seq(struct request *rq)
338{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 request_queue_t *q = rq->q;
340
Tejun Heo797e7db2006-01-06 09:51:03 +0100341 BUG_ON(q->ordseq == 0);
Tejun Heo8922e162005-10-20 16:23:44 +0200342
Tejun Heo797e7db2006-01-06 09:51:03 +0100343 if (rq == &q->pre_flush_rq)
344 return QUEUE_ORDSEQ_PREFLUSH;
345 if (rq == &q->bar_rq)
346 return QUEUE_ORDSEQ_BAR;
347 if (rq == &q->post_flush_rq)
348 return QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Jens Axboe4aff5e22006-08-10 08:44:47 +0200350 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
351 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
Tejun Heo797e7db2006-01-06 09:51:03 +0100352 return QUEUE_ORDSEQ_DRAIN;
353 else
354 return QUEUE_ORDSEQ_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
Tejun Heo797e7db2006-01-06 09:51:03 +0100357void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Tejun Heo797e7db2006-01-06 09:51:03 +0100359 struct request *rq;
360 int uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
Tejun Heo797e7db2006-01-06 09:51:03 +0100362 if (error && !q->orderr)
363 q->orderr = error;
Tejun Heo8922e162005-10-20 16:23:44 +0200364
Tejun Heo797e7db2006-01-06 09:51:03 +0100365 BUG_ON(q->ordseq & seq);
366 q->ordseq |= seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Tejun Heo797e7db2006-01-06 09:51:03 +0100368 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
369 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
371 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100372 * Okay, sequence complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100374 rq = q->orig_bar_rq;
375 uptodate = q->orderr ? q->orderr : 1;
376
377 q->ordseq = 0;
378
379 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
380 end_that_request_last(rq, uptodate);
381}
382
383static void pre_flush_end_io(struct request *rq, int error)
384{
385 elv_completed_request(rq->q, rq);
386 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
387}
388
389static void bar_end_io(struct request *rq, int error)
390{
391 elv_completed_request(rq->q, rq);
392 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
393}
394
395static void post_flush_end_io(struct request *rq, int error)
396{
397 elv_completed_request(rq->q, rq);
398 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
399}
400
401static void queue_flush(request_queue_t *q, unsigned which)
402{
403 struct request *rq;
404 rq_end_io_fn *end_io;
405
406 if (which == QUEUE_ORDERED_PREFLUSH) {
407 rq = &q->pre_flush_rq;
408 end_io = pre_flush_end_io;
409 } else {
410 rq = &q->post_flush_rq;
411 end_io = post_flush_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 }
413
Jens Axboe4aff5e22006-08-10 08:44:47 +0200414 rq->cmd_flags = REQ_HARDBARRIER;
Tejun Heo797e7db2006-01-06 09:51:03 +0100415 rq_init(q, rq);
Tejun Heo797e7db2006-01-06 09:51:03 +0100416 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200417 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100418 rq->rq_disk = q->bar_rq.rq_disk;
Tejun Heo797e7db2006-01-06 09:51:03 +0100419 rq->end_io = end_io;
420 q->prepare_flush_fn(q, rq);
421
Tejun Heo30e96562006-02-08 01:01:31 -0800422 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100423}
424
425static inline struct request *start_ordered(request_queue_t *q,
426 struct request *rq)
427{
428 q->bi_size = 0;
429 q->orderr = 0;
430 q->ordered = q->next_ordered;
431 q->ordseq |= QUEUE_ORDSEQ_STARTED;
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100434 * Prep proxy barrier request.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 */
Tejun Heo797e7db2006-01-06 09:51:03 +0100436 blkdev_dequeue_request(rq);
437 q->orig_bar_rq = rq;
438 rq = &q->bar_rq;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200439 rq->cmd_flags = 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100440 rq_init(q, rq);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200441 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
442 rq->cmd_flags |= REQ_RW;
443 rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
Tejun Heo797e7db2006-01-06 09:51:03 +0100444 rq->elevator_private = NULL;
Jens Axboec00895a2006-09-30 20:29:12 +0200445 rq->elevator_private2 = NULL;
Tejun Heo797e7db2006-01-06 09:51:03 +0100446 init_request_from_bio(rq, q->orig_bar_rq->bio);
447 rq->end_io = bar_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Tejun Heo797e7db2006-01-06 09:51:03 +0100449 /*
450 * Queue ordered sequence. As we stack them at the head, we
451 * need to queue in reverse order. Note that we rely on that
452 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
453 * request gets inbetween ordered sequence.
454 */
455 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
456 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
457 else
458 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Tejun Heo30e96562006-02-08 01:01:31 -0800460 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
Tejun Heo797e7db2006-01-06 09:51:03 +0100461
462 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
463 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
464 rq = &q->pre_flush_rq;
465 } else
466 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
467
468 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
469 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
470 else
471 rq = NULL;
472
473 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
Tejun Heo797e7db2006-01-06 09:51:03 +0100476int blk_do_ordered(request_queue_t *q, struct request **rqp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800478 struct request *rq = *rqp;
Tejun Heo797e7db2006-01-06 09:51:03 +0100479 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Tejun Heo797e7db2006-01-06 09:51:03 +0100481 if (!q->ordseq) {
482 if (!is_barrier)
483 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Tejun Heo797e7db2006-01-06 09:51:03 +0100485 if (q->next_ordered != QUEUE_ORDERED_NONE) {
486 *rqp = start_ordered(q, rq);
487 return 1;
488 } else {
489 /*
490 * This can happen when the queue switches to
491 * ORDERED_NONE while this request is on it.
492 */
493 blkdev_dequeue_request(rq);
494 end_that_request_first(rq, -EOPNOTSUPP,
495 rq->hard_nr_sectors);
496 end_that_request_last(rq, -EOPNOTSUPP);
497 *rqp = NULL;
498 return 0;
499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800502 /*
503 * Ordered sequence in progress
504 */
505
506 /* Special requests are not subject to ordering rules. */
507 if (!blk_fs_request(rq) &&
508 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
509 return 1;
510
Tejun Heo797e7db2006-01-06 09:51:03 +0100511 if (q->ordered & QUEUE_ORDERED_TAG) {
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800512 /* Ordered by tag. Blocking the next barrier is enough. */
Tejun Heo797e7db2006-01-06 09:51:03 +0100513 if (is_barrier && rq != &q->bar_rq)
514 *rqp = NULL;
Jens Axboe9a7a67a2006-02-04 23:27:38 -0800515 } else {
516 /* Ordered by draining. Wait for turn. */
517 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
518 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
519 *rqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 }
521
522 return 1;
523}
524
Tejun Heo797e7db2006-01-06 09:51:03 +0100525static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
Tejun Heo797e7db2006-01-06 09:51:03 +0100527 request_queue_t *q = bio->bi_private;
528 struct bio_vec *bvec;
529 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Tejun Heo797e7db2006-01-06 09:51:03 +0100531 /*
532 * This is dry run, restore bio_sector and size. We'll finish
533 * this request again with the original bi_end_io after an
534 * error occurs or post flush is complete.
535 */
536 q->bi_size += bytes;
537
538 if (bio->bi_size)
539 return 1;
540
541 /* Rewind bvec's */
542 bio->bi_idx = 0;
543 bio_for_each_segment(bvec, bio, i) {
544 bvec->bv_len += bvec->bv_offset;
545 bvec->bv_offset = 0;
546 }
547
548 /* Reset bio */
549 set_bit(BIO_UPTODATE, &bio->bi_flags);
550 bio->bi_size = q->bi_size;
551 bio->bi_sector -= (q->bi_size >> 9);
552 q->bi_size = 0;
553
554 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
Tejun Heo797e7db2006-01-06 09:51:03 +0100556
Jens Axboe1ea25ecb2006-07-18 22:24:11 +0200557static int ordered_bio_endio(struct request *rq, struct bio *bio,
558 unsigned int nbytes, int error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100559{
560 request_queue_t *q = rq->q;
561 bio_end_io_t *endio;
562 void *private;
563
564 if (&q->bar_rq != rq)
565 return 0;
566
567 /*
568 * Okay, this is the barrier request in progress, dry finish it.
569 */
570 if (error && !q->orderr)
571 q->orderr = error;
572
573 endio = bio->bi_end_io;
574 private = bio->bi_private;
575 bio->bi_end_io = flush_dry_bio_endio;
576 bio->bi_private = q;
577
578 bio_endio(bio, nbytes, error);
579
580 bio->bi_end_io = endio;
581 bio->bi_private = private;
582
583 return 1;
584}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586/**
587 * blk_queue_bounce_limit - set bounce buffer limit for queue
588 * @q: the request queue for the device
589 * @dma_addr: bus address limit
590 *
591 * Description:
592 * Different hardware can have different requirements as to what pages
593 * it can do I/O directly to. A low level driver can call
594 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
Andi Kleen5ee1af92006-03-08 17:57:26 -0800595 * buffers for doing I/O to pages residing above @page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 **/
597void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
598{
599 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800600 int dma = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Andi Kleen5ee1af92006-03-08 17:57:26 -0800602 q->bounce_gfp = GFP_NOIO;
603#if BITS_PER_LONG == 64
604 /* Assume anything <= 4GB can be handled by IOMMU.
605 Actually some IOMMUs can handle everything, but I don't
606 know of a way to test this here. */
Andi Kleen82697302006-06-21 14:48:09 +0200607 if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
Andi Kleen5ee1af92006-03-08 17:57:26 -0800608 dma = 1;
609 q->bounce_pfn = max_low_pfn;
610#else
611 if (bounce_pfn < blk_max_low_pfn)
612 dma = 1;
613 q->bounce_pfn = bounce_pfn;
614#endif
615 if (dma) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 init_emergency_isa_pool();
617 q->bounce_gfp = GFP_NOIO | GFP_DMA;
Andi Kleen5ee1af92006-03-08 17:57:26 -0800618 q->bounce_pfn = bounce_pfn;
619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622EXPORT_SYMBOL(blk_queue_bounce_limit);
623
624/**
625 * blk_queue_max_sectors - set max sectors for a request for this queue
626 * @q: the request queue for the device
627 * @max_sectors: max sectors in the usual 512b unit
628 *
629 * Description:
630 * Enables a low level driver to set an upper limit on the size of
631 * received requests.
632 **/
Jens Axboe2cb2e142006-01-17 09:04:32 +0100633void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
636 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
637 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
638 }
639
Mike Christiedefd94b2005-12-05 02:37:06 -0600640 if (BLK_DEF_MAX_SECTORS > max_sectors)
641 q->max_hw_sectors = q->max_sectors = max_sectors;
642 else {
643 q->max_sectors = BLK_DEF_MAX_SECTORS;
644 q->max_hw_sectors = max_sectors;
645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
648EXPORT_SYMBOL(blk_queue_max_sectors);
649
650/**
651 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
652 * @q: the request queue for the device
653 * @max_segments: max number of segments
654 *
655 * Description:
656 * Enables a low level driver to set an upper limit on the number of
657 * physical data segments in a request. This would be the largest sized
658 * scatter list the driver could handle.
659 **/
660void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
661{
662 if (!max_segments) {
663 max_segments = 1;
664 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
665 }
666
667 q->max_phys_segments = max_segments;
668}
669
670EXPORT_SYMBOL(blk_queue_max_phys_segments);
671
672/**
673 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
674 * @q: the request queue for the device
675 * @max_segments: max number of segments
676 *
677 * Description:
678 * Enables a low level driver to set an upper limit on the number of
679 * hw data segments in a request. This would be the largest number of
680 * address/length pairs the host adapter can actually give as once
681 * to the device.
682 **/
683void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
684{
685 if (!max_segments) {
686 max_segments = 1;
687 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
688 }
689
690 q->max_hw_segments = max_segments;
691}
692
693EXPORT_SYMBOL(blk_queue_max_hw_segments);
694
695/**
696 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
697 * @q: the request queue for the device
698 * @max_size: max size of segment in bytes
699 *
700 * Description:
701 * Enables a low level driver to set an upper limit on the size of a
702 * coalesced segment
703 **/
704void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
705{
706 if (max_size < PAGE_CACHE_SIZE) {
707 max_size = PAGE_CACHE_SIZE;
708 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
709 }
710
711 q->max_segment_size = max_size;
712}
713
714EXPORT_SYMBOL(blk_queue_max_segment_size);
715
716/**
717 * blk_queue_hardsect_size - set hardware sector size for the queue
718 * @q: the request queue for the device
719 * @size: the hardware sector size, in bytes
720 *
721 * Description:
722 * This should typically be set to the lowest possible sector size
723 * that the hardware can operate on (possible without reverting to
724 * even internal read-modify-write operations). Usually the default
725 * of 512 covers most hardware.
726 **/
727void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
728{
729 q->hardsect_size = size;
730}
731
732EXPORT_SYMBOL(blk_queue_hardsect_size);
733
734/*
735 * Returns the minimum that is _not_ zero, unless both are zero.
736 */
737#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
738
739/**
740 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
741 * @t: the stacking driver (top)
742 * @b: the underlying device (bottom)
743 **/
744void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
745{
746 /* zero is "infinity" */
Mike Christiedefd94b2005-12-05 02:37:06 -0600747 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
748 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
751 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
752 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
753 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
NeilBrown89e5c8b2006-03-27 01:18:02 -0800754 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
755 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756}
757
758EXPORT_SYMBOL(blk_queue_stack_limits);
759
760/**
761 * blk_queue_segment_boundary - set boundary rules for segment merging
762 * @q: the request queue for the device
763 * @mask: the memory boundary mask
764 **/
765void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
766{
767 if (mask < PAGE_CACHE_SIZE - 1) {
768 mask = PAGE_CACHE_SIZE - 1;
769 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
770 }
771
772 q->seg_boundary_mask = mask;
773}
774
775EXPORT_SYMBOL(blk_queue_segment_boundary);
776
777/**
778 * blk_queue_dma_alignment - set dma length and memory alignment
779 * @q: the request queue for the device
780 * @mask: alignment mask
781 *
782 * description:
783 * set required memory and length aligment for direct dma transactions.
784 * this is used when buiding direct io requests for the queue.
785 *
786 **/
787void blk_queue_dma_alignment(request_queue_t *q, int mask)
788{
789 q->dma_alignment = mask;
790}
791
792EXPORT_SYMBOL(blk_queue_dma_alignment);
793
794/**
795 * blk_queue_find_tag - find a request by its tag and queue
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 * @q: The request queue for the device
797 * @tag: The tag of the request
798 *
799 * Notes:
800 * Should be used when a device returns a tag and you want to match
801 * it with a request.
802 *
803 * no locks need be held.
804 **/
805struct request *blk_queue_find_tag(request_queue_t *q, int tag)
806{
David C Somayajuluf583f492006-10-04 08:27:25 +0200807 return blk_map_queue_find_tag(q->queue_tags, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
810EXPORT_SYMBOL(blk_queue_find_tag);
811
812/**
James Bottomley492dfb42006-08-30 15:48:45 -0400813 * __blk_free_tags - release a given set of tag maintenance info
814 * @bqt: the tag map to free
815 *
816 * Tries to free the specified @bqt@. Returns true if it was
817 * actually freed and false if there are still references using it
818 */
819static int __blk_free_tags(struct blk_queue_tag *bqt)
820{
821 int retval;
822
823 retval = atomic_dec_and_test(&bqt->refcnt);
824 if (retval) {
825 BUG_ON(bqt->busy);
826 BUG_ON(!list_empty(&bqt->busy_list));
827
828 kfree(bqt->tag_index);
829 bqt->tag_index = NULL;
830
831 kfree(bqt->tag_map);
832 bqt->tag_map = NULL;
833
834 kfree(bqt);
835
836 }
837
838 return retval;
839}
840
841/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 * __blk_queue_free_tags - release tag maintenance info
843 * @q: the request queue for the device
844 *
845 * Notes:
846 * blk_cleanup_queue() will take care of calling this function, if tagging
847 * has been used. So there's no need to call this directly.
848 **/
849static void __blk_queue_free_tags(request_queue_t *q)
850{
851 struct blk_queue_tag *bqt = q->queue_tags;
852
853 if (!bqt)
854 return;
855
James Bottomley492dfb42006-08-30 15:48:45 -0400856 __blk_free_tags(bqt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 q->queue_tags = NULL;
859 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
860}
861
James Bottomley492dfb42006-08-30 15:48:45 -0400862
863/**
864 * blk_free_tags - release a given set of tag maintenance info
865 * @bqt: the tag map to free
866 *
867 * For externally managed @bqt@ frees the map. Callers of this
868 * function must guarantee to have released all the queues that
869 * might have been using this tag map.
870 */
871void blk_free_tags(struct blk_queue_tag *bqt)
872{
873 if (unlikely(!__blk_free_tags(bqt)))
874 BUG();
875}
876EXPORT_SYMBOL(blk_free_tags);
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878/**
879 * blk_queue_free_tags - release tag maintenance info
880 * @q: the request queue for the device
881 *
882 * Notes:
883 * This is used to disabled tagged queuing to a device, yet leave
884 * queue in function.
885 **/
886void blk_queue_free_tags(request_queue_t *q)
887{
888 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
889}
890
891EXPORT_SYMBOL(blk_queue_free_tags);
892
893static int
894init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
895{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 struct request **tag_index;
897 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -0700898 int nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
James Bottomley492dfb42006-08-30 15:48:45 -0400900 if (q && depth > q->nr_requests * 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 depth = q->nr_requests * 2;
902 printk(KERN_ERR "%s: adjusted depth to %d\n",
903 __FUNCTION__, depth);
904 }
905
Jens Axboef68110f2006-03-08 13:31:44 +0100906 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (!tag_index)
908 goto fail;
909
Tejun Heof7d37d02005-06-23 00:08:50 -0700910 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
Jens Axboef68110f2006-03-08 13:31:44 +0100911 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (!tag_map)
913 goto fail;
914
Tejun Heoba025082005-08-05 13:28:11 -0700915 tags->real_max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 tags->max_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 tags->tag_index = tag_index;
918 tags->tag_map = tag_map;
919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return 0;
921fail:
922 kfree(tag_index);
923 return -ENOMEM;
924}
925
James Bottomley492dfb42006-08-30 15:48:45 -0400926static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
927 int depth)
928{
929 struct blk_queue_tag *tags;
930
931 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
932 if (!tags)
933 goto fail;
934
935 if (init_tag_map(q, tags, depth))
936 goto fail;
937
938 INIT_LIST_HEAD(&tags->busy_list);
939 tags->busy = 0;
940 atomic_set(&tags->refcnt, 1);
941 return tags;
942fail:
943 kfree(tags);
944 return NULL;
945}
946
947/**
948 * blk_init_tags - initialize the tag info for an external tag map
949 * @depth: the maximum queue depth supported
950 * @tags: the tag to use
951 **/
952struct blk_queue_tag *blk_init_tags(int depth)
953{
954 return __blk_queue_init_tags(NULL, depth);
955}
956EXPORT_SYMBOL(blk_init_tags);
957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958/**
959 * blk_queue_init_tags - initialize the queue tag info
960 * @q: the request queue for the device
961 * @depth: the maximum queue depth supported
962 * @tags: the tag to use
963 **/
964int blk_queue_init_tags(request_queue_t *q, int depth,
965 struct blk_queue_tag *tags)
966{
967 int rc;
968
969 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
970
971 if (!tags && !q->queue_tags) {
James Bottomley492dfb42006-08-30 15:48:45 -0400972 tags = __blk_queue_init_tags(q, depth);
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (!tags)
975 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 } else if (q->queue_tags) {
977 if ((rc = blk_queue_resize_tags(q, depth)))
978 return rc;
979 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
980 return 0;
981 } else
982 atomic_inc(&tags->refcnt);
983
984 /*
985 * assign it, all done
986 */
987 q->queue_tags = tags;
988 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
989 return 0;
990fail:
991 kfree(tags);
992 return -ENOMEM;
993}
994
995EXPORT_SYMBOL(blk_queue_init_tags);
996
997/**
998 * blk_queue_resize_tags - change the queueing depth
999 * @q: the request queue for the device
1000 * @new_depth: the new max command queueing depth
1001 *
1002 * Notes:
1003 * Must be called with the queue lock held.
1004 **/
1005int blk_queue_resize_tags(request_queue_t *q, int new_depth)
1006{
1007 struct blk_queue_tag *bqt = q->queue_tags;
1008 struct request **tag_index;
1009 unsigned long *tag_map;
Tejun Heofa72b902005-06-23 00:08:49 -07001010 int max_depth, nr_ulongs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 if (!bqt)
1013 return -ENXIO;
1014
1015 /*
Tejun Heoba025082005-08-05 13:28:11 -07001016 * if we already have large enough real_max_depth. just
1017 * adjust max_depth. *NOTE* as requests with tag value
1018 * between new_depth and real_max_depth can be in-flight, tag
1019 * map can not be shrunk blindly here.
1020 */
1021 if (new_depth <= bqt->real_max_depth) {
1022 bqt->max_depth = new_depth;
1023 return 0;
1024 }
1025
1026 /*
James Bottomley492dfb42006-08-30 15:48:45 -04001027 * Currently cannot replace a shared tag map with a new
1028 * one, so error out if this is the case
1029 */
1030 if (atomic_read(&bqt->refcnt) != 1)
1031 return -EBUSY;
1032
1033 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 * save the old state info, so we can copy it back
1035 */
1036 tag_index = bqt->tag_index;
1037 tag_map = bqt->tag_map;
Tejun Heoba025082005-08-05 13:28:11 -07001038 max_depth = bqt->real_max_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 if (init_tag_map(q, bqt, new_depth))
1041 return -ENOMEM;
1042
1043 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
Tejun Heof7d37d02005-06-23 00:08:50 -07001044 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
Tejun Heofa72b902005-06-23 00:08:49 -07001045 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
1047 kfree(tag_index);
1048 kfree(tag_map);
1049 return 0;
1050}
1051
1052EXPORT_SYMBOL(blk_queue_resize_tags);
1053
1054/**
1055 * blk_queue_end_tag - end tag operations for a request
1056 * @q: the request queue for the device
1057 * @rq: the request that has completed
1058 *
1059 * Description:
1060 * Typically called when end_that_request_first() returns 0, meaning
1061 * all transfers have been done for a request. It's important to call
1062 * this function before end_that_request_last(), as that will put the
1063 * request back on the free list thus corrupting the internal tag list.
1064 *
1065 * Notes:
1066 * queue lock must be held.
1067 **/
1068void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1069{
1070 struct blk_queue_tag *bqt = q->queue_tags;
1071 int tag = rq->tag;
1072
1073 BUG_ON(tag == -1);
1074
Tejun Heoba025082005-08-05 13:28:11 -07001075 if (unlikely(tag >= bqt->real_max_depth))
Tejun Heo040c9282005-06-23 00:08:51 -07001076 /*
1077 * This can happen after tag depth has been reduced.
1078 * FIXME: how about a warning or info message here?
1079 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return;
1081
1082 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
Tejun Heo040c9282005-06-23 00:08:51 -07001083 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1084 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 return;
1086 }
1087
1088 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001089 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 rq->tag = -1;
1091
1092 if (unlikely(bqt->tag_index[tag] == NULL))
Tejun Heo040c9282005-06-23 00:08:51 -07001093 printk(KERN_ERR "%s: tag %d is missing\n",
1094 __FUNCTION__, tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 bqt->tag_index[tag] = NULL;
1097 bqt->busy--;
1098}
1099
1100EXPORT_SYMBOL(blk_queue_end_tag);
1101
1102/**
1103 * blk_queue_start_tag - find a free tag and assign it
1104 * @q: the request queue for the device
1105 * @rq: the block request that needs tagging
1106 *
1107 * Description:
1108 * This can either be used as a stand-alone helper, or possibly be
1109 * assigned as the queue &prep_rq_fn (in which case &struct request
1110 * automagically gets a tag assigned). Note that this function
1111 * assumes that any type of request can be queued! if this is not
1112 * true for your device, you must check the request type before
1113 * calling this function. The request will also be removed from
1114 * the request queue, so it's the drivers responsibility to readd
1115 * it if it should need to be restarted for some reason.
1116 *
1117 * Notes:
1118 * queue lock must be held.
1119 **/
1120int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1121{
1122 struct blk_queue_tag *bqt = q->queue_tags;
Tejun Heo2bf0fda2005-06-23 00:08:48 -07001123 int tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Jens Axboe4aff5e22006-08-10 08:44:47 +02001125 if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 printk(KERN_ERR
Tejun Heo040c9282005-06-23 00:08:51 -07001127 "%s: request %p for device [%s] already tagged %d",
1128 __FUNCTION__, rq,
1129 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 BUG();
1131 }
1132
Jens Axboe059af492006-09-21 20:37:22 +02001133 /*
1134 * Protect against shared tag maps, as we may not have exclusive
1135 * access to the tag map.
1136 */
1137 do {
1138 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1139 if (tag >= bqt->max_depth)
1140 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Jens Axboe059af492006-09-21 20:37:22 +02001142 } while (test_and_set_bit(tag, bqt->tag_map));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Jens Axboe4aff5e22006-08-10 08:44:47 +02001144 rq->cmd_flags |= REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 rq->tag = tag;
1146 bqt->tag_index[tag] = rq;
1147 blkdev_dequeue_request(rq);
1148 list_add(&rq->queuelist, &bqt->busy_list);
1149 bqt->busy++;
1150 return 0;
1151}
1152
1153EXPORT_SYMBOL(blk_queue_start_tag);
1154
1155/**
1156 * blk_queue_invalidate_tags - invalidate all pending tags
1157 * @q: the request queue for the device
1158 *
1159 * Description:
1160 * Hardware conditions may dictate a need to stop all pending requests.
1161 * In this case, we will safely clear the block side of the tag queue and
1162 * readd all requests to the request queue in the right order.
1163 *
1164 * Notes:
1165 * queue lock must be held.
1166 **/
1167void blk_queue_invalidate_tags(request_queue_t *q)
1168{
1169 struct blk_queue_tag *bqt = q->queue_tags;
1170 struct list_head *tmp, *n;
1171 struct request *rq;
1172
1173 list_for_each_safe(tmp, n, &bqt->busy_list) {
1174 rq = list_entry_rq(tmp);
1175
1176 if (rq->tag == -1) {
Tejun Heo040c9282005-06-23 00:08:51 -07001177 printk(KERN_ERR
1178 "%s: bad tag found on list\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 list_del_init(&rq->queuelist);
Jens Axboe4aff5e22006-08-10 08:44:47 +02001180 rq->cmd_flags &= ~REQ_QUEUED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 } else
1182 blk_queue_end_tag(q, rq);
1183
Jens Axboe4aff5e22006-08-10 08:44:47 +02001184 rq->cmd_flags &= ~REQ_STARTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1186 }
1187}
1188
1189EXPORT_SYMBOL(blk_queue_invalidate_tags);
1190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191void blk_dump_rq_flags(struct request *rq, char *msg)
1192{
1193 int bit;
1194
Jens Axboe4aff5e22006-08-10 08:44:47 +02001195 printk("%s: dev %s: type=%x, flags=%x\n", msg,
1196 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
1197 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1200 rq->nr_sectors,
1201 rq->current_nr_sectors);
1202 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1203
Jens Axboe4aff5e22006-08-10 08:44:47 +02001204 if (blk_pc_request(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 printk("cdb: ");
1206 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1207 printk("%02x ", rq->cmd[bit]);
1208 printk("\n");
1209 }
1210}
1211
1212EXPORT_SYMBOL(blk_dump_rq_flags);
1213
1214void blk_recount_segments(request_queue_t *q, struct bio *bio)
1215{
1216 struct bio_vec *bv, *bvprv = NULL;
1217 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1218 int high, highprv = 1;
1219
1220 if (unlikely(!bio->bi_io_vec))
1221 return;
1222
1223 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1224 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1225 bio_for_each_segment(bv, bio, i) {
1226 /*
1227 * the trick here is making sure that a high page is never
1228 * considered part of another segment, since that might
1229 * change with the bounce page.
1230 */
1231 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1232 if (high || highprv)
1233 goto new_hw_segment;
1234 if (cluster) {
1235 if (seg_size + bv->bv_len > q->max_segment_size)
1236 goto new_segment;
1237 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1238 goto new_segment;
1239 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1240 goto new_segment;
1241 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1242 goto new_hw_segment;
1243
1244 seg_size += bv->bv_len;
1245 hw_seg_size += bv->bv_len;
1246 bvprv = bv;
1247 continue;
1248 }
1249new_segment:
1250 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1251 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1252 hw_seg_size += bv->bv_len;
1253 } else {
1254new_hw_segment:
1255 if (hw_seg_size > bio->bi_hw_front_size)
1256 bio->bi_hw_front_size = hw_seg_size;
1257 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1258 nr_hw_segs++;
1259 }
1260
1261 nr_phys_segs++;
1262 bvprv = bv;
1263 seg_size = bv->bv_len;
1264 highprv = high;
1265 }
1266 if (hw_seg_size > bio->bi_hw_back_size)
1267 bio->bi_hw_back_size = hw_seg_size;
1268 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1269 bio->bi_hw_front_size = hw_seg_size;
1270 bio->bi_phys_segments = nr_phys_segs;
1271 bio->bi_hw_segments = nr_hw_segs;
1272 bio->bi_flags |= (1 << BIO_SEG_VALID);
1273}
1274
1275
Adrian Bunk93d17d32005-06-25 14:59:10 -07001276static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 struct bio *nxt)
1278{
1279 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1280 return 0;
1281
1282 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1283 return 0;
1284 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1285 return 0;
1286
1287 /*
1288 * bio and nxt are contigous in memory, check if the queue allows
1289 * these two to be merged into one
1290 */
1291 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1292 return 1;
1293
1294 return 0;
1295}
1296
Adrian Bunk93d17d32005-06-25 14:59:10 -07001297static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 struct bio *nxt)
1299{
1300 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1301 blk_recount_segments(q, bio);
1302 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1303 blk_recount_segments(q, nxt);
1304 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1305 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1306 return 0;
1307 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1308 return 0;
1309
1310 return 1;
1311}
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313/*
1314 * map a request to scatterlist, return number of sg entries setup. Caller
1315 * must make sure sg can hold rq->nr_phys_segments entries
1316 */
1317int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1318{
1319 struct bio_vec *bvec, *bvprv;
1320 struct bio *bio;
1321 int nsegs, i, cluster;
1322
1323 nsegs = 0;
1324 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1325
1326 /*
1327 * for each bio in rq
1328 */
1329 bvprv = NULL;
1330 rq_for_each_bio(bio, rq) {
1331 /*
1332 * for each segment in bio
1333 */
1334 bio_for_each_segment(bvec, bio, i) {
1335 int nbytes = bvec->bv_len;
1336
1337 if (bvprv && cluster) {
1338 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1339 goto new_segment;
1340
1341 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1342 goto new_segment;
1343 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1344 goto new_segment;
1345
1346 sg[nsegs - 1].length += nbytes;
1347 } else {
1348new_segment:
1349 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1350 sg[nsegs].page = bvec->bv_page;
1351 sg[nsegs].length = nbytes;
1352 sg[nsegs].offset = bvec->bv_offset;
1353
1354 nsegs++;
1355 }
1356 bvprv = bvec;
1357 } /* segments in bio */
1358 } /* bios in rq */
1359
1360 return nsegs;
1361}
1362
1363EXPORT_SYMBOL(blk_rq_map_sg);
1364
1365/*
1366 * the standard queue merge functions, can be overridden with device
1367 * specific ones if so desired
1368 */
1369
1370static inline int ll_new_mergeable(request_queue_t *q,
1371 struct request *req,
1372 struct bio *bio)
1373{
1374 int nr_phys_segs = bio_phys_segments(q, bio);
1375
1376 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001377 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 if (req == q->last_merge)
1379 q->last_merge = NULL;
1380 return 0;
1381 }
1382
1383 /*
1384 * A hw segment is just getting larger, bump just the phys
1385 * counter.
1386 */
1387 req->nr_phys_segments += nr_phys_segs;
1388 return 1;
1389}
1390
1391static inline int ll_new_hw_segment(request_queue_t *q,
1392 struct request *req,
1393 struct bio *bio)
1394{
1395 int nr_hw_segs = bio_hw_segments(q, bio);
1396 int nr_phys_segs = bio_phys_segments(q, bio);
1397
1398 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1399 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001400 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (req == q->last_merge)
1402 q->last_merge = NULL;
1403 return 0;
1404 }
1405
1406 /*
1407 * This will form the start of a new hw segment. Bump both
1408 * counters.
1409 */
1410 req->nr_hw_segments += nr_hw_segs;
1411 req->nr_phys_segments += nr_phys_segs;
1412 return 1;
1413}
1414
1415static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1416 struct bio *bio)
1417{
Mike Christiedefd94b2005-12-05 02:37:06 -06001418 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 int len;
1420
Mike Christiedefd94b2005-12-05 02:37:06 -06001421 if (unlikely(blk_pc_request(req)))
1422 max_sectors = q->max_hw_sectors;
1423 else
1424 max_sectors = q->max_sectors;
1425
1426 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001427 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 if (req == q->last_merge)
1429 q->last_merge = NULL;
1430 return 0;
1431 }
1432 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1433 blk_recount_segments(q, req->biotail);
1434 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1435 blk_recount_segments(q, bio);
1436 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1437 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1438 !BIOVEC_VIRT_OVERSIZE(len)) {
1439 int mergeable = ll_new_mergeable(q, req, bio);
1440
1441 if (mergeable) {
1442 if (req->nr_hw_segments == 1)
1443 req->bio->bi_hw_front_size = len;
1444 if (bio->bi_hw_segments == 1)
1445 bio->bi_hw_back_size = len;
1446 }
1447 return mergeable;
1448 }
1449
1450 return ll_new_hw_segment(q, req, bio);
1451}
1452
1453static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1454 struct bio *bio)
1455{
Mike Christiedefd94b2005-12-05 02:37:06 -06001456 unsigned short max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 int len;
1458
Mike Christiedefd94b2005-12-05 02:37:06 -06001459 if (unlikely(blk_pc_request(req)))
1460 max_sectors = q->max_hw_sectors;
1461 else
1462 max_sectors = q->max_sectors;
1463
1464
1465 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02001466 req->cmd_flags |= REQ_NOMERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 if (req == q->last_merge)
1468 q->last_merge = NULL;
1469 return 0;
1470 }
1471 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1472 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1473 blk_recount_segments(q, bio);
1474 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1475 blk_recount_segments(q, req->bio);
1476 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1477 !BIOVEC_VIRT_OVERSIZE(len)) {
1478 int mergeable = ll_new_mergeable(q, req, bio);
1479
1480 if (mergeable) {
1481 if (bio->bi_hw_segments == 1)
1482 bio->bi_hw_front_size = len;
1483 if (req->nr_hw_segments == 1)
1484 req->biotail->bi_hw_back_size = len;
1485 }
1486 return mergeable;
1487 }
1488
1489 return ll_new_hw_segment(q, req, bio);
1490}
1491
1492static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1493 struct request *next)
1494{
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001495 int total_phys_segments;
1496 int total_hw_segments;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498 /*
1499 * First check if the either of the requests are re-queued
1500 * requests. Can't merge them if they are.
1501 */
1502 if (req->special || next->special)
1503 return 0;
1504
1505 /*
Nikita Danilovdfa1a552005-06-25 14:59:20 -07001506 * Will it become too large?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 */
1508 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1509 return 0;
1510
1511 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1512 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1513 total_phys_segments--;
1514
1515 if (total_phys_segments > q->max_phys_segments)
1516 return 0;
1517
1518 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1519 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1520 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1521 /*
1522 * propagate the combined length to the end of the requests
1523 */
1524 if (req->nr_hw_segments == 1)
1525 req->bio->bi_hw_front_size = len;
1526 if (next->nr_hw_segments == 1)
1527 next->biotail->bi_hw_back_size = len;
1528 total_hw_segments--;
1529 }
1530
1531 if (total_hw_segments > q->max_hw_segments)
1532 return 0;
1533
1534 /* Merge is OK... */
1535 req->nr_phys_segments = total_phys_segments;
1536 req->nr_hw_segments = total_hw_segments;
1537 return 1;
1538}
1539
1540/*
1541 * "plug" the device if there are no outstanding requests: this will
1542 * force the transfer to start only after we have put all the requests
1543 * on the list.
1544 *
1545 * This is called with interrupts off and no requests on the queue and
1546 * with the queue lock held.
1547 */
1548void blk_plug_device(request_queue_t *q)
1549{
1550 WARN_ON(!irqs_disabled());
1551
1552 /*
1553 * don't plug a stopped queue, it must be paired with blk_start_queue()
1554 * which will restart the queueing
1555 */
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001556 if (blk_queue_stopped(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 return;
1558
Jens Axboe2056a782006-03-23 20:00:26 +01001559 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
Jens Axboe2056a782006-03-23 20:00:26 +01001561 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
1565EXPORT_SYMBOL(blk_plug_device);
1566
1567/*
1568 * remove the queue from the plugged list, if present. called with
1569 * queue lock held and interrupts disabled.
1570 */
1571int blk_remove_plug(request_queue_t *q)
1572{
1573 WARN_ON(!irqs_disabled());
1574
1575 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1576 return 0;
1577
1578 del_timer(&q->unplug_timer);
1579 return 1;
1580}
1581
1582EXPORT_SYMBOL(blk_remove_plug);
1583
1584/*
1585 * remove the plug and let it rip..
1586 */
1587void __generic_unplug_device(request_queue_t *q)
1588{
Coywolf Qi Hunt7daac492006-04-19 10:14:49 +02001589 if (unlikely(blk_queue_stopped(q)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 return;
1591
1592 if (!blk_remove_plug(q))
1593 return;
1594
Jens Axboe22e2c502005-06-27 10:55:12 +02001595 q->request_fn(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596}
1597EXPORT_SYMBOL(__generic_unplug_device);
1598
1599/**
1600 * generic_unplug_device - fire a request queue
1601 * @q: The &request_queue_t in question
1602 *
1603 * Description:
1604 * Linux uses plugging to build bigger requests queues before letting
1605 * the device have at them. If a queue is plugged, the I/O scheduler
1606 * is still adding and merging requests on the queue. Once the queue
1607 * gets unplugged, the request_fn defined for the queue is invoked and
1608 * transfers started.
1609 **/
1610void generic_unplug_device(request_queue_t *q)
1611{
1612 spin_lock_irq(q->queue_lock);
1613 __generic_unplug_device(q);
1614 spin_unlock_irq(q->queue_lock);
1615}
1616EXPORT_SYMBOL(generic_unplug_device);
1617
1618static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1619 struct page *page)
1620{
1621 request_queue_t *q = bdi->unplug_io_data;
1622
1623 /*
1624 * devices don't necessarily have an ->unplug_fn defined
1625 */
Jens Axboe2056a782006-03-23 20:00:26 +01001626 if (q->unplug_fn) {
1627 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1628 q->rq.count[READ] + q->rq.count[WRITE]);
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 q->unplug_fn(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632}
1633
David Howells65f27f32006-11-22 14:55:48 +00001634static void blk_unplug_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635{
David Howells65f27f32006-11-22 14:55:48 +00001636 request_queue_t *q = container_of(work, request_queue_t, unplug_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637
Jens Axboe2056a782006-03-23 20:00:26 +01001638 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1639 q->rq.count[READ] + q->rq.count[WRITE]);
1640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 q->unplug_fn(q);
1642}
1643
1644static void blk_unplug_timeout(unsigned long data)
1645{
1646 request_queue_t *q = (request_queue_t *)data;
1647
Jens Axboe2056a782006-03-23 20:00:26 +01001648 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1649 q->rq.count[READ] + q->rq.count[WRITE]);
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 kblockd_schedule_work(&q->unplug_work);
1652}
1653
1654/**
1655 * blk_start_queue - restart a previously stopped queue
1656 * @q: The &request_queue_t in question
1657 *
1658 * Description:
1659 * blk_start_queue() will clear the stop flag on the queue, and call
1660 * the request_fn for the queue if it was in a stopped state when
1661 * entered. Also see blk_stop_queue(). Queue lock must be held.
1662 **/
1663void blk_start_queue(request_queue_t *q)
1664{
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001665 WARN_ON(!irqs_disabled());
1666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1668
1669 /*
1670 * one level of recursion is ok and is much faster than kicking
1671 * the unplug handling
1672 */
1673 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1674 q->request_fn(q);
1675 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1676 } else {
1677 blk_plug_device(q);
1678 kblockd_schedule_work(&q->unplug_work);
1679 }
1680}
1681
1682EXPORT_SYMBOL(blk_start_queue);
1683
1684/**
1685 * blk_stop_queue - stop a queue
1686 * @q: The &request_queue_t in question
1687 *
1688 * Description:
1689 * The Linux block layer assumes that a block driver will consume all
1690 * entries on the request queue when the request_fn strategy is called.
1691 * Often this will not happen, because of hardware limitations (queue
1692 * depth settings). If a device driver gets a 'queue full' response,
1693 * or if it simply chooses not to queue more I/O at one point, it can
1694 * call this function to prevent the request_fn from being called until
1695 * the driver has signalled it's ready to go again. This happens by calling
1696 * blk_start_queue() to restart queue operations. Queue lock must be held.
1697 **/
1698void blk_stop_queue(request_queue_t *q)
1699{
1700 blk_remove_plug(q);
1701 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1702}
1703EXPORT_SYMBOL(blk_stop_queue);
1704
1705/**
1706 * blk_sync_queue - cancel any pending callbacks on a queue
1707 * @q: the queue
1708 *
1709 * Description:
1710 * The block layer may perform asynchronous callback activity
1711 * on a queue, such as calling the unplug function after a timeout.
1712 * A block device may call blk_sync_queue to ensure that any
1713 * such activity is cancelled, thus allowing it to release resources
1714 * the the callbacks might use. The caller must already have made sure
1715 * that its ->make_request_fn will not re-add plugging prior to calling
1716 * this function.
1717 *
1718 */
1719void blk_sync_queue(struct request_queue *q)
1720{
1721 del_timer_sync(&q->unplug_timer);
1722 kblockd_flush();
1723}
1724EXPORT_SYMBOL(blk_sync_queue);
1725
1726/**
1727 * blk_run_queue - run a single device queue
1728 * @q: The queue to run
1729 */
1730void blk_run_queue(struct request_queue *q)
1731{
1732 unsigned long flags;
1733
1734 spin_lock_irqsave(q->queue_lock, flags);
1735 blk_remove_plug(q);
Jens Axboedac07ec2006-05-11 08:20:16 +02001736
1737 /*
1738 * Only recurse once to avoid overrunning the stack, let the unplug
1739 * handling reinvoke the handler shortly if we already got there.
1740 */
1741 if (!elv_queue_empty(q)) {
1742 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1743 q->request_fn(q);
1744 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1745 } else {
1746 blk_plug_device(q);
1747 kblockd_schedule_work(&q->unplug_work);
1748 }
1749 }
1750
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 spin_unlock_irqrestore(q->queue_lock, flags);
1752}
1753EXPORT_SYMBOL(blk_run_queue);
1754
1755/**
1756 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
Martin Waitza5802902006-04-02 13:59:55 +02001757 * @kobj: the kobj belonging of the request queue to be released
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 *
1759 * Description:
1760 * blk_cleanup_queue is the pair to blk_init_queue() or
1761 * blk_queue_make_request(). It should be called when a request queue is
1762 * being released; typically when a block device is being de-registered.
1763 * Currently, its primary task it to free all the &struct request
1764 * structures that were allocated to the queue and the queue itself.
1765 *
1766 * Caveat:
1767 * Hopefully the low level driver will have finished any
1768 * outstanding requests first...
1769 **/
Al Viro483f4af2006-03-18 18:34:37 -05001770static void blk_release_queue(struct kobject *kobj)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771{
Al Viro483f4af2006-03-18 18:34:37 -05001772 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 struct request_list *rl = &q->rq;
1774
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 blk_sync_queue(q);
1776
1777 if (rl->rq_pool)
1778 mempool_destroy(rl->rq_pool);
1779
1780 if (q->queue_tags)
1781 __blk_queue_free_tags(q);
1782
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -07001783 blk_trace_shutdown(q);
Jens Axboe2056a782006-03-23 20:00:26 +01001784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 kmem_cache_free(requestq_cachep, q);
1786}
1787
Al Viro483f4af2006-03-18 18:34:37 -05001788void blk_put_queue(request_queue_t *q)
1789{
1790 kobject_put(&q->kobj);
1791}
1792EXPORT_SYMBOL(blk_put_queue);
1793
1794void blk_cleanup_queue(request_queue_t * q)
1795{
1796 mutex_lock(&q->sysfs_lock);
1797 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1798 mutex_unlock(&q->sysfs_lock);
1799
1800 if (q->elevator)
1801 elevator_exit(q->elevator);
1802
1803 blk_put_queue(q);
1804}
1805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806EXPORT_SYMBOL(blk_cleanup_queue);
1807
1808static int blk_init_free_list(request_queue_t *q)
1809{
1810 struct request_list *rl = &q->rq;
1811
1812 rl->count[READ] = rl->count[WRITE] = 0;
1813 rl->starved[READ] = rl->starved[WRITE] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02001814 rl->elvpriv = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 init_waitqueue_head(&rl->wait[READ]);
1816 init_waitqueue_head(&rl->wait[WRITE]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Christoph Lameter19460892005-06-23 00:08:19 -07001818 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1819 mempool_free_slab, request_cachep, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 if (!rl->rq_pool)
1822 return -ENOMEM;
1823
1824 return 0;
1825}
1826
Al Viro8267e262005-10-21 03:20:53 -04001827request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828{
Christoph Lameter19460892005-06-23 00:08:19 -07001829 return blk_alloc_queue_node(gfp_mask, -1);
1830}
1831EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Al Viro483f4af2006-03-18 18:34:37 -05001833static struct kobj_type queue_ktype;
1834
Al Viro8267e262005-10-21 03:20:53 -04001835request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -07001836{
1837 request_queue_t *q;
1838
1839 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 if (!q)
1841 return NULL;
1842
1843 memset(q, 0, sizeof(*q));
1844 init_timer(&q->unplug_timer);
Al Viro483f4af2006-03-18 18:34:37 -05001845
1846 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1847 q->kobj.ktype = &queue_ktype;
1848 kobject_init(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
1850 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1851 q->backing_dev_info.unplug_io_data = q;
1852
Al Viro483f4af2006-03-18 18:34:37 -05001853 mutex_init(&q->sysfs_lock);
1854
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 return q;
1856}
Christoph Lameter19460892005-06-23 00:08:19 -07001857EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859/**
1860 * blk_init_queue - prepare a request queue for use with a block device
1861 * @rfn: The function to be called to process requests that have been
1862 * placed on the queue.
1863 * @lock: Request queue spin lock
1864 *
1865 * Description:
1866 * If a block device wishes to use the standard request handling procedures,
1867 * which sorts requests and coalesces adjacent requests, then it must
1868 * call blk_init_queue(). The function @rfn will be called when there
1869 * are requests on the queue that need to be processed. If the device
1870 * supports plugging, then @rfn may not be called immediately when requests
1871 * are available on the queue, but may be called at some time later instead.
1872 * Plugged queues are generally unplugged when a buffer belonging to one
1873 * of the requests on the queue is needed, or due to memory pressure.
1874 *
1875 * @rfn is not required, or even expected, to remove all requests off the
1876 * queue, but only as many as it can handle at a time. If it does leave
1877 * requests on the queue, it is responsible for arranging that the requests
1878 * get dealt with eventually.
1879 *
1880 * The queue spin lock must be held while manipulating the requests on the
Paolo 'Blaisorblade' Giarrussoa038e252006-06-05 12:09:01 +02001881 * request queue; this lock will be taken also from interrupt context, so irq
1882 * disabling is needed for it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 *
1884 * Function returns a pointer to the initialized request queue, or NULL if
1885 * it didn't succeed.
1886 *
1887 * Note:
1888 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1889 * when the block device is deactivated (such as at module unload).
1890 **/
Christoph Lameter19460892005-06-23 00:08:19 -07001891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1893{
Christoph Lameter19460892005-06-23 00:08:19 -07001894 return blk_init_queue_node(rfn, lock, -1);
1895}
1896EXPORT_SYMBOL(blk_init_queue);
1897
1898request_queue_t *
1899blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1900{
1901 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
1903 if (!q)
1904 return NULL;
1905
Christoph Lameter19460892005-06-23 00:08:19 -07001906 q->node = node_id;
Al Viro8669aaf2006-03-18 13:50:00 -05001907 if (blk_init_free_list(q)) {
1908 kmem_cache_free(requestq_cachep, q);
1909 return NULL;
1910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
152587d2005-04-12 16:22:06 -05001912 /*
1913 * if caller didn't supply a lock, they get per-queue locking with
1914 * our embedded lock
1915 */
1916 if (!lock) {
1917 spin_lock_init(&q->__queue_lock);
1918 lock = &q->__queue_lock;
1919 }
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 q->request_fn = rfn;
1922 q->back_merge_fn = ll_back_merge_fn;
1923 q->front_merge_fn = ll_front_merge_fn;
1924 q->merge_requests_fn = ll_merge_requests_fn;
1925 q->prep_rq_fn = NULL;
1926 q->unplug_fn = generic_unplug_device;
1927 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1928 q->queue_lock = lock;
1929
1930 blk_queue_segment_boundary(q, 0xffffffff);
1931
1932 blk_queue_make_request(q, __make_request);
1933 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1934
1935 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1936 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1937
1938 /*
1939 * all done
1940 */
1941 if (!elevator_init(q, NULL)) {
1942 blk_queue_congestion_threshold(q);
1943 return q;
1944 }
1945
Al Viro8669aaf2006-03-18 13:50:00 -05001946 blk_put_queue(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 return NULL;
1948}
Christoph Lameter19460892005-06-23 00:08:19 -07001949EXPORT_SYMBOL(blk_init_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951int blk_get_queue(request_queue_t *q)
1952{
Nick Pigginfde6ad22005-06-23 00:08:53 -07001953 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
Al Viro483f4af2006-03-18 18:34:37 -05001954 kobject_get(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return 0;
1956 }
1957
1958 return 1;
1959}
1960
1961EXPORT_SYMBOL(blk_get_queue);
1962
1963static inline void blk_free_request(request_queue_t *q, struct request *rq)
1964{
Jens Axboe4aff5e22006-08-10 08:44:47 +02001965 if (rq->cmd_flags & REQ_ELVPRIV)
Tejun Heocb98fc82005-10-28 08:29:39 +02001966 elv_put_request(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 mempool_free(rq, q->rq.rq_pool);
1968}
1969
Jens Axboe1ea25ecb2006-07-18 22:24:11 +02001970static struct request *
Jens Axboecb78b282006-07-28 09:32:57 +02001971blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972{
1973 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1974
1975 if (!rq)
1976 return NULL;
1977
1978 /*
Jens Axboe4aff5e22006-08-10 08:44:47 +02001979 * first three bits are identical in rq->cmd_flags and bio->bi_rw,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 * see bio.h and blkdev.h
1981 */
Jens Axboe49171e52006-08-10 08:59:11 +02001982 rq->cmd_flags = rw | REQ_ALLOCED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Tejun Heocb98fc82005-10-28 08:29:39 +02001984 if (priv) {
Jens Axboecb78b282006-07-28 09:32:57 +02001985 if (unlikely(elv_set_request(q, rq, gfp_mask))) {
Tejun Heocb98fc82005-10-28 08:29:39 +02001986 mempool_free(rq, q->rq.rq_pool);
1987 return NULL;
1988 }
Jens Axboe4aff5e22006-08-10 08:44:47 +02001989 rq->cmd_flags |= REQ_ELVPRIV;
Tejun Heocb98fc82005-10-28 08:29:39 +02001990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Tejun Heocb98fc82005-10-28 08:29:39 +02001992 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993}
1994
1995/*
1996 * ioc_batching returns true if the ioc is a valid batching request and
1997 * should be given priority access to a request.
1998 */
1999static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
2000{
2001 if (!ioc)
2002 return 0;
2003
2004 /*
2005 * Make sure the process is able to allocate at least 1 request
2006 * even if the batch times out, otherwise we could theoretically
2007 * lose wakeups.
2008 */
2009 return ioc->nr_batch_requests == q->nr_batching ||
2010 (ioc->nr_batch_requests > 0
2011 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2012}
2013
2014/*
2015 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2016 * will cause the process to be a "batcher" on all queues in the system. This
2017 * is the behaviour we want though - once it gets a wakeup it should be given
2018 * a nice run.
2019 */
Adrian Bunk93d17d32005-06-25 14:59:10 -07002020static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021{
2022 if (!ioc || ioc_batching(q, ioc))
2023 return;
2024
2025 ioc->nr_batch_requests = q->nr_batching;
2026 ioc->last_waited = jiffies;
2027}
2028
2029static void __freed_request(request_queue_t *q, int rw)
2030{
2031 struct request_list *rl = &q->rq;
2032
2033 if (rl->count[rw] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07002034 blk_clear_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 if (rl->count[rw] + 1 <= q->nr_requests) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (waitqueue_active(&rl->wait[rw]))
2038 wake_up(&rl->wait[rw]);
2039
2040 blk_clear_queue_full(q, rw);
2041 }
2042}
2043
2044/*
2045 * A request has just been released. Account for it, update the full and
2046 * congestion status, wake up any waiters. Called under q->queue_lock.
2047 */
Tejun Heocb98fc82005-10-28 08:29:39 +02002048static void freed_request(request_queue_t *q, int rw, int priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049{
2050 struct request_list *rl = &q->rq;
2051
2052 rl->count[rw]--;
Tejun Heocb98fc82005-10-28 08:29:39 +02002053 if (priv)
2054 rl->elvpriv--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055
2056 __freed_request(q, rw);
2057
2058 if (unlikely(rl->starved[rw ^ 1]))
2059 __freed_request(q, rw ^ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060}
2061
2062#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2063/*
Nick Piggind6344532005-06-28 20:45:14 -07002064 * Get a free request, queue_lock must be held.
2065 * Returns NULL on failure, with queue_lock held.
2066 * Returns !NULL on success, with queue_lock *not held*.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002068static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -04002069 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
2071 struct request *rq = NULL;
2072 struct request_list *rl = &q->rq;
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002073 struct io_context *ioc = NULL;
2074 int may_queue, priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
Jens Axboecb78b282006-07-28 09:32:57 +02002076 may_queue = elv_may_queue(q, rw);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002077 if (may_queue == ELV_MQUEUE_NO)
2078 goto rq_starved;
2079
2080 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2081 if (rl->count[rw]+1 >= q->nr_requests) {
Jens Axboeb5deef92006-07-19 23:39:40 +02002082 ioc = current_io_context(GFP_ATOMIC, q->node);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002083 /*
2084 * The queue will fill after this allocation, so set
2085 * it as full, and mark this process as "batching".
2086 * This process will be allowed to complete a batch of
2087 * requests, others will be blocked.
2088 */
2089 if (!blk_queue_full(q, rw)) {
2090 ioc_set_batching(q, ioc);
2091 blk_set_queue_full(q, rw);
2092 } else {
2093 if (may_queue != ELV_MQUEUE_MUST
2094 && !ioc_batching(q, ioc)) {
2095 /*
2096 * The queue is full and the allocating
2097 * process is not a "batcher", and not
2098 * exempted by the IO scheduler
2099 */
2100 goto out;
2101 }
2102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
Thomas Maier79e2de42006-10-19 23:28:15 -07002104 blk_set_queue_congested(q, rw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 }
2106
Jens Axboe082cf692005-06-28 16:35:11 +02002107 /*
2108 * Only allow batching queuers to allocate up to 50% over the defined
2109 * limit of requests, otherwise we could have thousands of requests
2110 * allocated with any setting of ->nr_requests
2111 */
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002112 if (rl->count[rw] >= (3 * q->nr_requests / 2))
Jens Axboe082cf692005-06-28 16:35:11 +02002113 goto out;
Hugh Dickinsfd782a42005-06-29 15:15:40 +01002114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 rl->count[rw]++;
2116 rl->starved[rw] = 0;
Tejun Heocb98fc82005-10-28 08:29:39 +02002117
Jens Axboe64521d12005-10-28 08:30:39 +02002118 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +02002119 if (priv)
2120 rl->elvpriv++;
2121
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 spin_unlock_irq(q->queue_lock);
2123
Jens Axboecb78b282006-07-28 09:32:57 +02002124 rq = blk_alloc_request(q, rw, priv, gfp_mask);
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002125 if (unlikely(!rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 /*
2127 * Allocation failed presumably due to memory. Undo anything
2128 * we might have messed up.
2129 *
2130 * Allocating task should really be put onto the front of the
2131 * wait queue, but this is pretty rare.
2132 */
2133 spin_lock_irq(q->queue_lock);
Tejun Heocb98fc82005-10-28 08:29:39 +02002134 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
2136 /*
2137 * in the very unlikely event that allocation failed and no
2138 * requests for this direction was pending, mark us starved
2139 * so that freeing of a request in the other direction will
2140 * notice us. another possible fix would be to split the
2141 * rq mempool into READ and WRITE
2142 */
2143rq_starved:
2144 if (unlikely(rl->count[rw] == 0))
2145 rl->starved[rw] = 1;
2146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 goto out;
2148 }
2149
Jens Axboe88ee5ef2005-11-12 11:09:12 +01002150 /*
2151 * ioc may be NULL here, and ioc_batching will be false. That's
2152 * OK, if the queue is under the request limit then requests need
2153 * not count toward the nr_batch_requests limit. There will always
2154 * be some limit enforced by BLK_BATCH_TIME.
2155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 if (ioc_batching(q, ioc))
2157 ioc->nr_batch_requests--;
2158
2159 rq_init(q, rq);
Jens Axboe2056a782006-03-23 20:00:26 +01002160
2161 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 return rq;
2164}
2165
2166/*
2167 * No available requests for this queue, unplug the device and wait for some
2168 * requests to become available.
Nick Piggind6344532005-06-28 20:45:14 -07002169 *
2170 * Called with q->queue_lock held, and returns with it unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 */
Jens Axboe22e2c502005-06-27 10:55:12 +02002172static struct request *get_request_wait(request_queue_t *q, int rw,
2173 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 struct request *rq;
2176
Nick Piggin450991b2005-06-28 20:45:13 -07002177 rq = get_request(q, rw, bio, GFP_NOIO);
2178 while (!rq) {
2179 DEFINE_WAIT(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 struct request_list *rl = &q->rq;
2181
2182 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2183 TASK_UNINTERRUPTIBLE);
2184
Jens Axboe22e2c502005-06-27 10:55:12 +02002185 rq = get_request(q, rw, bio, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 if (!rq) {
2188 struct io_context *ioc;
2189
Jens Axboe2056a782006-03-23 20:00:26 +01002190 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2191
Nick Piggind6344532005-06-28 20:45:14 -07002192 __generic_unplug_device(q);
2193 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 io_schedule();
2195
2196 /*
2197 * After sleeping, we become a "batching" process and
2198 * will be able to allocate at least one request, and
2199 * up to a big batch of them for a small period time.
2200 * See ioc_batching, ioc_set_batching
2201 */
Jens Axboeb5deef92006-07-19 23:39:40 +02002202 ioc = current_io_context(GFP_NOIO, q->node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 ioc_set_batching(q, ioc);
Nick Piggind6344532005-06-28 20:45:14 -07002204
2205 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 }
2207 finish_wait(&rl->wait[rw], &wait);
Nick Piggin450991b2005-06-28 20:45:13 -07002208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
2210 return rq;
2211}
2212
Al Viro8267e262005-10-21 03:20:53 -04002213struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214{
2215 struct request *rq;
2216
2217 BUG_ON(rw != READ && rw != WRITE);
2218
Nick Piggind6344532005-06-28 20:45:14 -07002219 spin_lock_irq(q->queue_lock);
2220 if (gfp_mask & __GFP_WAIT) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002221 rq = get_request_wait(q, rw, NULL);
Nick Piggind6344532005-06-28 20:45:14 -07002222 } else {
Jens Axboe22e2c502005-06-27 10:55:12 +02002223 rq = get_request(q, rw, NULL, gfp_mask);
Nick Piggind6344532005-06-28 20:45:14 -07002224 if (!rq)
2225 spin_unlock_irq(q->queue_lock);
2226 }
2227 /* q->queue_lock is unlocked at this point */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 return rq;
2230}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231EXPORT_SYMBOL(blk_get_request);
2232
2233/**
Jens Axboedc72ef42006-07-20 14:54:05 +02002234 * blk_start_queueing - initiate dispatch of requests to device
2235 * @q: request queue to kick into gear
2236 *
2237 * This is basically a helper to remove the need to know whether a queue
2238 * is plugged or not if someone just wants to initiate dispatch of requests
2239 * for this queue.
2240 *
2241 * The queue lock must be held with interrupts disabled.
2242 */
2243void blk_start_queueing(request_queue_t *q)
2244{
2245 if (!blk_queue_plugged(q))
2246 q->request_fn(q);
2247 else
2248 __generic_unplug_device(q);
2249}
2250EXPORT_SYMBOL(blk_start_queueing);
2251
2252/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 * blk_requeue_request - put a request back on queue
2254 * @q: request queue where request should be inserted
2255 * @rq: request to be inserted
2256 *
2257 * Description:
2258 * Drivers often keep queueing requests until the hardware cannot accept
2259 * more, when that condition happens we need to put the request back
2260 * on the queue. Must be called with queue lock held.
2261 */
2262void blk_requeue_request(request_queue_t *q, struct request *rq)
2263{
Jens Axboe2056a782006-03-23 20:00:26 +01002264 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 if (blk_rq_tagged(rq))
2267 blk_queue_end_tag(q, rq);
2268
2269 elv_requeue_request(q, rq);
2270}
2271
2272EXPORT_SYMBOL(blk_requeue_request);
2273
2274/**
2275 * blk_insert_request - insert a special request in to a request queue
2276 * @q: request queue where request should be inserted
2277 * @rq: request to be inserted
2278 * @at_head: insert request at head or tail of queue
2279 * @data: private data
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 *
2281 * Description:
2282 * Many block devices need to execute commands asynchronously, so they don't
2283 * block the whole kernel from preemption during request execution. This is
2284 * accomplished normally by inserting aritficial requests tagged as
2285 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2286 * scheduled for actual execution by the request queue.
2287 *
2288 * We have the option of inserting the head or the tail of the queue.
2289 * Typically we use the tail for new ioctls and so forth. We use the head
2290 * of the queue for things like a QUEUE_FULL message from a device, or a
2291 * host that is unable to accept a particular command.
2292 */
2293void blk_insert_request(request_queue_t *q, struct request *rq,
Tejun Heo 867d1192005-04-24 02:06:05 -05002294 int at_head, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295{
Tejun Heo 867d1192005-04-24 02:06:05 -05002296 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 unsigned long flags;
2298
2299 /*
2300 * tell I/O scheduler that this isn't a regular read/write (ie it
2301 * must not attempt merges on this) and that it acts as a soft
2302 * barrier
2303 */
Jens Axboe4aff5e22006-08-10 08:44:47 +02002304 rq->cmd_type = REQ_TYPE_SPECIAL;
2305 rq->cmd_flags |= REQ_SOFTBARRIER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
2307 rq->special = data;
2308
2309 spin_lock_irqsave(q->queue_lock, flags);
2310
2311 /*
2312 * If command is tagged, release the tag
2313 */
Tejun Heo 867d1192005-04-24 02:06:05 -05002314 if (blk_rq_tagged(rq))
2315 blk_queue_end_tag(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316
Tejun Heo 867d1192005-04-24 02:06:05 -05002317 drive_stat_acct(rq, rq->nr_sectors, 1);
2318 __elv_add_request(q, rq, where, 0);
Jens Axboedc72ef42006-07-20 14:54:05 +02002319 blk_start_queueing(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 spin_unlock_irqrestore(q->queue_lock, flags);
2321}
2322
2323EXPORT_SYMBOL(blk_insert_request);
2324
Mike Christie0e75f902006-12-01 10:40:55 +01002325static int __blk_rq_unmap_user(struct bio *bio)
2326{
2327 int ret = 0;
2328
2329 if (bio) {
2330 if (bio_flagged(bio, BIO_USER_MAPPED))
2331 bio_unmap_user(bio);
2332 else
2333 ret = bio_uncopy_user(bio);
2334 }
2335
2336 return ret;
2337}
2338
2339static int __blk_rq_map_user(request_queue_t *q, struct request *rq,
2340 void __user *ubuf, unsigned int len)
2341{
2342 unsigned long uaddr;
2343 struct bio *bio, *orig_bio;
2344 int reading, ret;
2345
2346 reading = rq_data_dir(rq) == READ;
2347
2348 /*
2349 * if alignment requirement is satisfied, map in user pages for
2350 * direct dma. else, set up kernel bounce buffers
2351 */
2352 uaddr = (unsigned long) ubuf;
2353 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2354 bio = bio_map_user(q, NULL, uaddr, len, reading);
2355 else
2356 bio = bio_copy_user(q, uaddr, len, reading);
2357
2358 if (IS_ERR(bio)) {
2359 return PTR_ERR(bio);
2360 }
2361
2362 orig_bio = bio;
2363 blk_queue_bounce(q, &bio);
2364 /*
2365 * We link the bounce buffer in and could have to traverse it
2366 * later so we have to get a ref to prevent it from being freed
2367 */
2368 bio_get(bio);
2369
2370 /*
2371 * for most (all? don't know of any) queues we could
2372 * skip grabbing the queue lock here. only drivers with
2373 * funky private ->back_merge_fn() function could be
2374 * problematic.
2375 */
2376 spin_lock_irq(q->queue_lock);
2377 if (!rq->bio)
2378 blk_rq_bio_prep(q, rq, bio);
2379 else if (!q->back_merge_fn(q, rq, bio)) {
2380 ret = -EINVAL;
2381 spin_unlock_irq(q->queue_lock);
2382 goto unmap_bio;
2383 } else {
2384 rq->biotail->bi_next = bio;
2385 rq->biotail = bio;
2386
2387 rq->nr_sectors += bio_sectors(bio);
2388 rq->hard_nr_sectors = rq->nr_sectors;
2389 rq->data_len += bio->bi_size;
2390 }
2391 spin_unlock_irq(q->queue_lock);
2392
2393 return bio->bi_size;
2394
2395unmap_bio:
2396 /* if it was boucned we must call the end io function */
2397 bio_endio(bio, bio->bi_size, 0);
2398 __blk_rq_unmap_user(orig_bio);
2399 bio_put(bio);
2400 return ret;
2401}
2402
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403/**
2404 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2405 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002406 * @rq: request structure to fill
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 * @ubuf: the user buffer
2408 * @len: length of user data
2409 *
2410 * Description:
2411 * Data will be mapped directly for zero copy io, if possible. Otherwise
2412 * a kernel bounce buffer is used.
2413 *
2414 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2415 * still in process context.
2416 *
2417 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2418 * before being submitted to the device, as pages mapped may be out of
2419 * reach. It's the callers responsibility to make sure this happens. The
2420 * original bio must be passed back in to blk_rq_unmap_user() for proper
2421 * unmapping.
2422 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002423int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
Mike Christie0e75f902006-12-01 10:40:55 +01002424 unsigned long len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425{
Mike Christie0e75f902006-12-01 10:40:55 +01002426 unsigned long bytes_read = 0;
2427 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Mike Christiedefd94b2005-12-05 02:37:06 -06002429 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002430 return -EINVAL;
2431 if (!len || !ubuf)
2432 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Mike Christie0e75f902006-12-01 10:40:55 +01002434 while (bytes_read != len) {
2435 unsigned long map_len, end, start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Mike Christie0e75f902006-12-01 10:40:55 +01002437 map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
2438 end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
2439 >> PAGE_SHIFT;
2440 start = (unsigned long)ubuf >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Mike Christie0e75f902006-12-01 10:40:55 +01002442 /*
2443 * A bad offset could cause us to require BIO_MAX_PAGES + 1
2444 * pages. If this happens we just lower the requested
2445 * mapping len by a page so that we can fit
2446 */
2447 if (end - start > BIO_MAX_PAGES)
2448 map_len -= PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Mike Christie0e75f902006-12-01 10:40:55 +01002450 ret = __blk_rq_map_user(q, rq, ubuf, map_len);
2451 if (ret < 0)
2452 goto unmap_rq;
2453 bytes_read += ret;
2454 ubuf += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 }
2456
Mike Christie0e75f902006-12-01 10:40:55 +01002457 rq->buffer = rq->data = NULL;
2458 return 0;
2459unmap_rq:
2460 blk_rq_unmap_user(rq);
2461 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462}
2463
2464EXPORT_SYMBOL(blk_rq_map_user);
2465
2466/**
James Bottomley f1970ba2005-06-20 14:06:52 +02002467 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2468 * @q: request queue where request should be inserted
2469 * @rq: request to map data to
2470 * @iov: pointer to the iovec
2471 * @iov_count: number of elements in the iovec
2472 *
2473 * Description:
2474 * Data will be mapped directly for zero copy io, if possible. Otherwise
2475 * a kernel bounce buffer is used.
2476 *
2477 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2478 * still in process context.
2479 *
2480 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2481 * before being submitted to the device, as pages mapped may be out of
2482 * reach. It's the callers responsibility to make sure this happens. The
2483 * original bio must be passed back in to blk_rq_unmap_user() for proper
2484 * unmapping.
2485 */
2486int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
Mike Christie0e75f902006-12-01 10:40:55 +01002487 struct sg_iovec *iov, int iov_count, unsigned int len)
James Bottomley f1970ba2005-06-20 14:06:52 +02002488{
2489 struct bio *bio;
2490
2491 if (!iov || iov_count <= 0)
2492 return -EINVAL;
2493
2494 /* we don't allow misaligned data like bio_map_user() does. If the
2495 * user is using sg, they're expected to know the alignment constraints
2496 * and respect them accordingly */
2497 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2498 if (IS_ERR(bio))
2499 return PTR_ERR(bio);
2500
Mike Christie0e75f902006-12-01 10:40:55 +01002501 if (bio->bi_size != len) {
2502 bio_endio(bio, bio->bi_size, 0);
2503 bio_unmap_user(bio);
2504 return -EINVAL;
2505 }
2506
2507 bio_get(bio);
James Bottomley f1970ba2005-06-20 14:06:52 +02002508 blk_rq_bio_prep(q, rq, bio);
2509 rq->buffer = rq->data = NULL;
James Bottomley f1970ba2005-06-20 14:06:52 +02002510 return 0;
2511}
2512
2513EXPORT_SYMBOL(blk_rq_map_user_iov);
2514
2515/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 * blk_rq_unmap_user - unmap a request with user data
Mike Christie0e75f902006-12-01 10:40:55 +01002517 * @rq: rq to be unmapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 *
2519 * Description:
Mike Christie0e75f902006-12-01 10:40:55 +01002520 * Unmap a rq previously mapped by blk_rq_map_user().
2521 * rq->bio must be set to the original head of the request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
Mike Christie0e75f902006-12-01 10:40:55 +01002523int blk_rq_unmap_user(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
Mike Christie0e75f902006-12-01 10:40:55 +01002525 struct bio *bio, *mapped_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Mike Christie0e75f902006-12-01 10:40:55 +01002527 while ((bio = rq->bio)) {
2528 if (bio_flagged(bio, BIO_BOUNCED))
2529 mapped_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 else
Mike Christie0e75f902006-12-01 10:40:55 +01002531 mapped_bio = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Mike Christie0e75f902006-12-01 10:40:55 +01002533 __blk_rq_unmap_user(mapped_bio);
2534 rq->bio = bio->bi_next;
2535 bio_put(bio);
2536 }
Jens Axboedd1cab92005-06-20 14:06:01 +02002537 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538}
2539
2540EXPORT_SYMBOL(blk_rq_unmap_user);
2541
2542/**
Mike Christie df46b9a2005-06-20 14:04:44 +02002543 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2544 * @q: request queue where request should be inserted
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002545 * @rq: request to fill
Mike Christie df46b9a2005-06-20 14:04:44 +02002546 * @kbuf: the kernel buffer
2547 * @len: length of user data
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002548 * @gfp_mask: memory allocation flags
Mike Christie df46b9a2005-06-20 14:04:44 +02002549 */
Jens Axboedd1cab92005-06-20 14:06:01 +02002550int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
Al Viro8267e262005-10-21 03:20:53 -04002551 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02002552{
Mike Christie df46b9a2005-06-20 14:04:44 +02002553 struct bio *bio;
2554
Mike Christiedefd94b2005-12-05 02:37:06 -06002555 if (len > (q->max_hw_sectors << 9))
Jens Axboedd1cab92005-06-20 14:06:01 +02002556 return -EINVAL;
2557 if (!len || !kbuf)
2558 return -EINVAL;
Mike Christie df46b9a2005-06-20 14:04:44 +02002559
2560 bio = bio_map_kern(q, kbuf, len, gfp_mask);
Jens Axboedd1cab92005-06-20 14:06:01 +02002561 if (IS_ERR(bio))
2562 return PTR_ERR(bio);
Mike Christie df46b9a2005-06-20 14:04:44 +02002563
Jens Axboedd1cab92005-06-20 14:06:01 +02002564 if (rq_data_dir(rq) == WRITE)
2565 bio->bi_rw |= (1 << BIO_RW);
Mike Christie df46b9a2005-06-20 14:04:44 +02002566
Jens Axboedd1cab92005-06-20 14:06:01 +02002567 blk_rq_bio_prep(q, rq, bio);
Jens Axboedd1cab92005-06-20 14:06:01 +02002568 rq->buffer = rq->data = NULL;
Jens Axboedd1cab92005-06-20 14:06:01 +02002569 return 0;
Mike Christie df46b9a2005-06-20 14:04:44 +02002570}
2571
2572EXPORT_SYMBOL(blk_rq_map_kern);
2573
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002574/**
2575 * blk_execute_rq_nowait - insert a request into queue for execution
2576 * @q: queue to insert the request in
2577 * @bd_disk: matching gendisk
2578 * @rq: request to insert
2579 * @at_head: insert request at head or tail of queue
2580 * @done: I/O completion handler
2581 *
2582 * Description:
2583 * Insert a fully prepared request at the back of the io scheduler queue
2584 * for execution. Don't wait for completion.
2585 */
James Bottomley f1970ba2005-06-20 14:06:52 +02002586void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2587 struct request *rq, int at_head,
Tejun Heo8ffdc652006-01-06 09:49:03 +01002588 rq_end_io_fn *done)
James Bottomley f1970ba2005-06-20 14:06:52 +02002589{
2590 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2591
2592 rq->rq_disk = bd_disk;
Jens Axboe4aff5e22006-08-10 08:44:47 +02002593 rq->cmd_flags |= REQ_NOMERGE;
James Bottomley f1970ba2005-06-20 14:06:52 +02002594 rq->end_io = done;
Andrew Morton4c5d0bb2006-03-22 08:08:01 +01002595 WARN_ON(irqs_disabled());
2596 spin_lock_irq(q->queue_lock);
2597 __elv_add_request(q, rq, where, 1);
2598 __generic_unplug_device(q);
2599 spin_unlock_irq(q->queue_lock);
James Bottomley f1970ba2005-06-20 14:06:52 +02002600}
Mike Christie6e39b69e2005-11-11 05:30:24 -06002601EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603/**
2604 * blk_execute_rq - insert a request into queue for execution
2605 * @q: queue to insert the request in
2606 * @bd_disk: matching gendisk
2607 * @rq: request to insert
James Bottomley 994ca9a2005-06-20 14:11:09 +02002608 * @at_head: insert request at head or tail of queue
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 *
2610 * Description:
2611 * Insert a fully prepared request at the back of the io scheduler queue
Christoph Hellwig 73747ae2005-06-20 14:21:01 +02002612 * for execution and wait for completion.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 */
2614int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
James Bottomley 994ca9a2005-06-20 14:11:09 +02002615 struct request *rq, int at_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616{
Ingo Molnar60be6b92006-07-03 00:25:26 -07002617 DECLARE_COMPLETION_ONSTACK(wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 char sense[SCSI_SENSE_BUFFERSIZE];
2619 int err = 0;
2620
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 /*
2622 * we need an extra reference to the request, so we can look at
2623 * it after io completion
2624 */
2625 rq->ref_count++;
2626
2627 if (!rq->sense) {
2628 memset(sense, 0, sizeof(sense));
2629 rq->sense = sense;
2630 rq->sense_len = 0;
2631 }
2632
Jens Axboec00895a2006-09-30 20:29:12 +02002633 rq->end_io_data = &wait;
James Bottomley 994ca9a2005-06-20 14:11:09 +02002634 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 wait_for_completion(&wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
2637 if (rq->errors)
2638 err = -EIO;
2639
2640 return err;
2641}
2642
2643EXPORT_SYMBOL(blk_execute_rq);
2644
2645/**
2646 * blkdev_issue_flush - queue a flush
2647 * @bdev: blockdev to issue flush for
2648 * @error_sector: error sector
2649 *
2650 * Description:
2651 * Issue a flush for the block device in question. Caller can supply
2652 * room for storing the error offset in case of a flush error, if they
2653 * wish to. Caller must run wait_for_completion() on its own.
2654 */
2655int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2656{
2657 request_queue_t *q;
2658
2659 if (bdev->bd_disk == NULL)
2660 return -ENXIO;
2661
2662 q = bdev_get_queue(bdev);
2663 if (!q)
2664 return -ENXIO;
2665 if (!q->issue_flush_fn)
2666 return -EOPNOTSUPP;
2667
2668 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2669}
2670
2671EXPORT_SYMBOL(blkdev_issue_flush);
2672
Adrian Bunk93d17d32005-06-25 14:59:10 -07002673static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674{
2675 int rw = rq_data_dir(rq);
2676
2677 if (!blk_fs_request(rq) || !rq->rq_disk)
2678 return;
2679
Jens Axboed72d9042005-11-01 08:35:42 +01002680 if (!new_io) {
Jens Axboea3623572005-11-01 09:26:16 +01002681 __disk_stat_inc(rq->rq_disk, merges[rw]);
Jens Axboed72d9042005-11-01 08:35:42 +01002682 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 disk_round_stats(rq->rq_disk);
2684 rq->rq_disk->in_flight++;
2685 }
2686}
2687
2688/*
2689 * add-request adds a request to the linked list.
2690 * queue lock is held and interrupts disabled, as we muck with the
2691 * request queue list.
2692 */
2693static inline void add_request(request_queue_t * q, struct request * req)
2694{
2695 drive_stat_acct(req, req->nr_sectors, 1);
2696
2697 if (q->activity_fn)
2698 q->activity_fn(q->activity_data, rq_data_dir(req));
2699
2700 /*
2701 * elevator indicated where it wants this request to be
2702 * inserted at elevator_merge time
2703 */
2704 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2705}
2706
2707/*
2708 * disk_round_stats() - Round off the performance stats on a struct
2709 * disk_stats.
2710 *
2711 * The average IO queue length and utilisation statistics are maintained
2712 * by observing the current state of the queue length and the amount of
2713 * time it has been in this state for.
2714 *
2715 * Normally, that accounting is done on IO completion, but that can result
2716 * in more than a second's worth of IO being accounted for within any one
2717 * second, leading to >100% utilisation. To deal with that, we call this
2718 * function to do a round-off before returning the results when reading
2719 * /proc/diskstats. This accounts immediately for all queue usage up to
2720 * the current jiffies and restarts the counters again.
2721 */
2722void disk_round_stats(struct gendisk *disk)
2723{
2724 unsigned long now = jiffies;
2725
Chen, Kenneth Wb2982642005-10-13 21:49:29 +02002726 if (now == disk->stamp)
2727 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Chen, Kenneth W20e5c812005-10-13 21:48:42 +02002729 if (disk->in_flight) {
2730 __disk_stat_add(disk, time_in_queue,
2731 disk->in_flight * (now - disk->stamp));
2732 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2733 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 disk->stamp = now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735}
2736
Jun'ichi "Nick" Nomura3eaf8402006-02-01 03:04:53 -08002737EXPORT_SYMBOL_GPL(disk_round_stats);
2738
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739/*
2740 * queue lock must be held
2741 */
Mike Christie6e39b69e2005-11-11 05:30:24 -06002742void __blk_put_request(request_queue_t *q, struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 if (unlikely(!q))
2745 return;
2746 if (unlikely(--req->ref_count))
2747 return;
2748
Tejun Heo8922e162005-10-20 16:23:44 +02002749 elv_completed_request(q, req);
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 /*
2752 * Request may not have originated from ll_rw_blk. if not,
2753 * it didn't come out of our reserved rq pools
2754 */
Jens Axboe49171e52006-08-10 08:59:11 +02002755 if (req->cmd_flags & REQ_ALLOCED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 int rw = rq_data_dir(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +02002757 int priv = req->cmd_flags & REQ_ELVPRIV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 BUG_ON(!list_empty(&req->queuelist));
Jens Axboe98170642006-07-28 09:23:08 +02002760 BUG_ON(!hlist_unhashed(&req->hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
2762 blk_free_request(q, req);
Tejun Heocb98fc82005-10-28 08:29:39 +02002763 freed_request(q, rw, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 }
2765}
2766
Mike Christie6e39b69e2005-11-11 05:30:24 -06002767EXPORT_SYMBOL_GPL(__blk_put_request);
2768
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769void blk_put_request(struct request *req)
2770{
Tejun Heo8922e162005-10-20 16:23:44 +02002771 unsigned long flags;
2772 request_queue_t *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Tejun Heo8922e162005-10-20 16:23:44 +02002774 /*
2775 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2776 * following if (q) test.
2777 */
2778 if (q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 spin_lock_irqsave(q->queue_lock, flags);
2780 __blk_put_request(q, req);
2781 spin_unlock_irqrestore(q->queue_lock, flags);
2782 }
2783}
2784
2785EXPORT_SYMBOL(blk_put_request);
2786
2787/**
2788 * blk_end_sync_rq - executes a completion event on a request
2789 * @rq: request to complete
Jens Axboefddfdea2006-01-31 15:24:34 +01002790 * @error: end io status of the request
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01002792void blk_end_sync_rq(struct request *rq, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
Jens Axboec00895a2006-09-30 20:29:12 +02002794 struct completion *waiting = rq->end_io_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795
Jens Axboec00895a2006-09-30 20:29:12 +02002796 rq->end_io_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 __blk_put_request(rq->q, rq);
2798
2799 /*
2800 * complete last, if this is a stack request the process (and thus
2801 * the rq pointer) could be invalid right after this complete()
2802 */
2803 complete(waiting);
2804}
2805EXPORT_SYMBOL(blk_end_sync_rq);
2806
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807/*
2808 * Has to be called with the request spinlock acquired
2809 */
2810static int attempt_merge(request_queue_t *q, struct request *req,
2811 struct request *next)
2812{
2813 if (!rq_mergeable(req) || !rq_mergeable(next))
2814 return 0;
2815
2816 /*
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02002817 * not contiguous
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 */
2819 if (req->sector + req->nr_sectors != next->sector)
2820 return 0;
2821
2822 if (rq_data_dir(req) != rq_data_dir(next)
2823 || req->rq_disk != next->rq_disk
Jens Axboec00895a2006-09-30 20:29:12 +02002824 || next->special)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 return 0;
2826
2827 /*
2828 * If we are allowed to merge, then append bio list
2829 * from next to rq and release next. merge_requests_fn
2830 * will have updated segment counts, update sector
2831 * counts here.
2832 */
2833 if (!q->merge_requests_fn(q, req, next))
2834 return 0;
2835
2836 /*
2837 * At this point we have either done a back merge
2838 * or front merge. We need the smaller start_time of
2839 * the merged requests to be the current request
2840 * for accounting purposes.
2841 */
2842 if (time_after(req->start_time, next->start_time))
2843 req->start_time = next->start_time;
2844
2845 req->biotail->bi_next = next->bio;
2846 req->biotail = next->biotail;
2847
2848 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2849
2850 elv_merge_requests(q, req, next);
2851
2852 if (req->rq_disk) {
2853 disk_round_stats(req->rq_disk);
2854 req->rq_disk->in_flight--;
2855 }
2856
Jens Axboe22e2c502005-06-27 10:55:12 +02002857 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2858
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 __blk_put_request(q, next);
2860 return 1;
2861}
2862
2863static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2864{
2865 struct request *next = elv_latter_request(q, rq);
2866
2867 if (next)
2868 return attempt_merge(q, rq, next);
2869
2870 return 0;
2871}
2872
2873static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2874{
2875 struct request *prev = elv_former_request(q, rq);
2876
2877 if (prev)
2878 return attempt_merge(q, prev, rq);
2879
2880 return 0;
2881}
2882
Tejun Heo52d9e672006-01-06 09:49:58 +01002883static void init_request_from_bio(struct request *req, struct bio *bio)
2884{
Jens Axboe4aff5e22006-08-10 08:44:47 +02002885 req->cmd_type = REQ_TYPE_FS;
Tejun Heo52d9e672006-01-06 09:49:58 +01002886
2887 /*
2888 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2889 */
2890 if (bio_rw_ahead(bio) || bio_failfast(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002891 req->cmd_flags |= REQ_FAILFAST;
Tejun Heo52d9e672006-01-06 09:49:58 +01002892
2893 /*
2894 * REQ_BARRIER implies no merging, but lets make it explicit
2895 */
2896 if (unlikely(bio_barrier(bio)))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002897 req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
Tejun Heo52d9e672006-01-06 09:49:58 +01002898
Jens Axboeb31dc662006-06-13 08:26:10 +02002899 if (bio_sync(bio))
Jens Axboe4aff5e22006-08-10 08:44:47 +02002900 req->cmd_flags |= REQ_RW_SYNC;
Jens Axboe5404bc72006-08-10 09:01:02 +02002901 if (bio_rw_meta(bio))
2902 req->cmd_flags |= REQ_RW_META;
Jens Axboeb31dc662006-06-13 08:26:10 +02002903
Tejun Heo52d9e672006-01-06 09:49:58 +01002904 req->errors = 0;
2905 req->hard_sector = req->sector = bio->bi_sector;
2906 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2907 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2908 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2909 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2910 req->buffer = bio_data(bio); /* see ->buffer comment above */
Tejun Heo52d9e672006-01-06 09:49:58 +01002911 req->bio = req->biotail = bio;
2912 req->ioprio = bio_prio(bio);
2913 req->rq_disk = bio->bi_bdev->bd_disk;
2914 req->start_time = jiffies;
2915}
2916
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917static int __make_request(request_queue_t *q, struct bio *bio)
2918{
Nick Piggin450991b2005-06-28 20:45:13 -07002919 struct request *req;
Jens Axboe51da90f2006-07-18 04:14:45 +02002920 int el_ret, nr_sectors, barrier, err;
2921 const unsigned short prio = bio_prio(bio);
2922 const int sync = bio_sync(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 nr_sectors = bio_sectors(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925
2926 /*
2927 * low level driver can indicate that it wants pages above a
2928 * certain limit bounced to low memory (ie for highmem, or even
2929 * ISA dma in theory)
2930 */
2931 blk_queue_bounce(q, &bio);
2932
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 barrier = bio_barrier(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +01002934 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 err = -EOPNOTSUPP;
2936 goto end_io;
2937 }
2938
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 spin_lock_irq(q->queue_lock);
2940
Nick Piggin450991b2005-06-28 20:45:13 -07002941 if (unlikely(barrier) || elv_queue_empty(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 goto get_rq;
2943
2944 el_ret = elv_merge(q, &req, bio);
2945 switch (el_ret) {
2946 case ELEVATOR_BACK_MERGE:
2947 BUG_ON(!rq_mergeable(req));
2948
2949 if (!q->back_merge_fn(q, req, bio))
2950 break;
2951
Jens Axboe2056a782006-03-23 20:00:26 +01002952 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2953
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 req->biotail->bi_next = bio;
2955 req->biotail = bio;
2956 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002957 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 drive_stat_acct(req, nr_sectors, 0);
2959 if (!attempt_back_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002960 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 goto out;
2962
2963 case ELEVATOR_FRONT_MERGE:
2964 BUG_ON(!rq_mergeable(req));
2965
2966 if (!q->front_merge_fn(q, req, bio))
2967 break;
2968
Jens Axboe2056a782006-03-23 20:00:26 +01002969 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2970
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 bio->bi_next = req->bio;
2972 req->bio = bio;
2973
2974 /*
2975 * may not be valid. if the low level driver said
2976 * it didn't need a bounce buffer then it better
2977 * not touch req->buffer either...
2978 */
2979 req->buffer = bio_data(bio);
Jens Axboe51da90f2006-07-18 04:14:45 +02002980 req->current_nr_sectors = bio_cur_sectors(bio);
2981 req->hard_cur_sectors = req->current_nr_sectors;
2982 req->sector = req->hard_sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
Jens Axboe22e2c502005-06-27 10:55:12 +02002984 req->ioprio = ioprio_best(req->ioprio, prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 drive_stat_acct(req, nr_sectors, 0);
2986 if (!attempt_front_merge(q, req))
Jens Axboe2e662b62006-07-13 11:55:04 +02002987 elv_merged_request(q, req, el_ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 goto out;
2989
Nick Piggin450991b2005-06-28 20:45:13 -07002990 /* ELV_NO_MERGE: elevator says don't/can't merge. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 default:
Nick Piggin450991b2005-06-28 20:45:13 -07002992 ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 }
2994
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995get_rq:
Nick Piggin450991b2005-06-28 20:45:13 -07002996 /*
2997 * Grab a free request. This is might sleep but can not fail.
Nick Piggind6344532005-06-28 20:45:14 -07002998 * Returns with the queue unlocked.
Nick Piggin450991b2005-06-28 20:45:13 -07002999 */
Jens Axboe51da90f2006-07-18 04:14:45 +02003000 req = get_request_wait(q, bio_data_dir(bio), bio);
Nick Piggind6344532005-06-28 20:45:14 -07003001
Nick Piggin450991b2005-06-28 20:45:13 -07003002 /*
3003 * After dropping the lock and possibly sleeping here, our request
3004 * may now be mergeable after it had proven unmergeable (above).
3005 * We don't worry about that case for efficiency. It won't happen
3006 * often, and the elevators are able to handle it.
3007 */
Tejun Heo52d9e672006-01-06 09:49:58 +01003008 init_request_from_bio(req, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
Nick Piggin450991b2005-06-28 20:45:13 -07003010 spin_lock_irq(q->queue_lock);
3011 if (elv_queue_empty(q))
3012 blk_plug_device(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 add_request(q, req);
3014out:
Jens Axboe4a534f92005-04-16 15:25:40 -07003015 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 __generic_unplug_device(q);
3017
3018 spin_unlock_irq(q->queue_lock);
3019 return 0;
3020
3021end_io:
3022 bio_endio(bio, nr_sectors << 9, err);
3023 return 0;
3024}
3025
3026/*
3027 * If bio->bi_dev is a partition, remap the location
3028 */
3029static inline void blk_partition_remap(struct bio *bio)
3030{
3031 struct block_device *bdev = bio->bi_bdev;
3032
3033 if (bdev != bdev->bd_contains) {
3034 struct hd_struct *p = bdev->bd_part;
Jens Axboea3623572005-11-01 09:26:16 +01003035 const int rw = bio_data_dir(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
Jens Axboea3623572005-11-01 09:26:16 +01003037 p->sectors[rw] += bio_sectors(bio);
3038 p->ios[rw]++;
3039
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 bio->bi_sector += p->start_sect;
3041 bio->bi_bdev = bdev->bd_contains;
3042 }
3043}
3044
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045static void handle_bad_sector(struct bio *bio)
3046{
3047 char b[BDEVNAME_SIZE];
3048
3049 printk(KERN_INFO "attempt to access beyond end of device\n");
3050 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
3051 bdevname(bio->bi_bdev, b),
3052 bio->bi_rw,
3053 (unsigned long long)bio->bi_sector + bio_sectors(bio),
3054 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
3055
3056 set_bit(BIO_EOF, &bio->bi_flags);
3057}
3058
3059/**
3060 * generic_make_request: hand a buffer to its device driver for I/O
3061 * @bio: The bio describing the location in memory and on the device.
3062 *
3063 * generic_make_request() is used to make I/O requests of block
3064 * devices. It is passed a &struct bio, which describes the I/O that needs
3065 * to be done.
3066 *
3067 * generic_make_request() does not return any status. The
3068 * success/failure status of the request, along with notification of
3069 * completion, is delivered asynchronously through the bio->bi_end_io
3070 * function described (one day) else where.
3071 *
3072 * The caller of generic_make_request must make sure that bi_io_vec
3073 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3074 * set to describe the device address, and the
3075 * bi_end_io and optionally bi_private are set to describe how
3076 * completion notification should be signaled.
3077 *
3078 * generic_make_request and the drivers it calls may use bi_next if this
3079 * bio happens to be merged with someone else, and may change bi_dev and
3080 * bi_sector for remaps as it sees fit. So the values of these fields
3081 * should NOT be depended on after the call to generic_make_request.
3082 */
3083void generic_make_request(struct bio *bio)
3084{
3085 request_queue_t *q;
3086 sector_t maxsector;
NeilBrown5ddfe962006-10-30 22:07:21 -08003087 sector_t old_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 int ret, nr_sectors = bio_sectors(bio);
Jens Axboe2056a782006-03-23 20:00:26 +01003089 dev_t old_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090
3091 might_sleep();
3092 /* Test device or partition size, when known. */
3093 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3094 if (maxsector) {
3095 sector_t sector = bio->bi_sector;
3096
3097 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3098 /*
3099 * This may well happen - the kernel calls bread()
3100 * without checking the size of the device, e.g., when
3101 * mounting a device.
3102 */
3103 handle_bad_sector(bio);
3104 goto end_io;
3105 }
3106 }
3107
3108 /*
3109 * Resolve the mapping until finished. (drivers are
3110 * still free to implement/resolve their own stacking
3111 * by explicitly returning 0)
3112 *
3113 * NOTE: we don't repeat the blk_size check for each new device.
3114 * Stacking drivers are expected to know what they are doing.
3115 */
NeilBrown5ddfe962006-10-30 22:07:21 -08003116 old_sector = -1;
Jens Axboe2056a782006-03-23 20:00:26 +01003117 old_dev = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 do {
3119 char b[BDEVNAME_SIZE];
3120
3121 q = bdev_get_queue(bio->bi_bdev);
3122 if (!q) {
3123 printk(KERN_ERR
3124 "generic_make_request: Trying to access "
3125 "nonexistent block-device %s (%Lu)\n",
3126 bdevname(bio->bi_bdev, b),
3127 (long long) bio->bi_sector);
3128end_io:
3129 bio_endio(bio, bio->bi_size, -EIO);
3130 break;
3131 }
3132
3133 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3134 printk("bio too big device %s (%u > %u)\n",
3135 bdevname(bio->bi_bdev, b),
3136 bio_sectors(bio),
3137 q->max_hw_sectors);
3138 goto end_io;
3139 }
3140
Nick Pigginfde6ad22005-06-23 00:08:53 -07003141 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 goto end_io;
3143
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 /*
3145 * If this device has partitions, remap block n
3146 * of partition p to block n+start(p) of the disk.
3147 */
3148 blk_partition_remap(bio);
3149
NeilBrown5ddfe962006-10-30 22:07:21 -08003150 if (old_sector != -1)
Jens Axboe2056a782006-03-23 20:00:26 +01003151 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
NeilBrown5ddfe962006-10-30 22:07:21 -08003152 old_sector);
Jens Axboe2056a782006-03-23 20:00:26 +01003153
3154 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3155
NeilBrown5ddfe962006-10-30 22:07:21 -08003156 old_sector = bio->bi_sector;
Jens Axboe2056a782006-03-23 20:00:26 +01003157 old_dev = bio->bi_bdev->bd_dev;
3158
NeilBrown5ddfe962006-10-30 22:07:21 -08003159 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3160 if (maxsector) {
3161 sector_t sector = bio->bi_sector;
3162
Andrew Mortondf66b852006-11-02 22:06:56 -08003163 if (maxsector < nr_sectors ||
3164 maxsector - nr_sectors < sector) {
NeilBrown5ddfe962006-10-30 22:07:21 -08003165 /*
Andrew Mortondf66b852006-11-02 22:06:56 -08003166 * This may well happen - partitions are not
3167 * checked to make sure they are within the size
3168 * of the whole device.
NeilBrown5ddfe962006-10-30 22:07:21 -08003169 */
3170 handle_bad_sector(bio);
3171 goto end_io;
3172 }
3173 }
3174
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 ret = q->make_request_fn(q, bio);
3176 } while (ret);
3177}
3178
3179EXPORT_SYMBOL(generic_make_request);
3180
3181/**
3182 * submit_bio: submit a bio to the block device layer for I/O
3183 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3184 * @bio: The &struct bio which describes the I/O
3185 *
3186 * submit_bio() is very similar in purpose to generic_make_request(), and
3187 * uses that function to do most of the work. Both are fairly rough
3188 * interfaces, @bio must be presetup and ready for I/O.
3189 *
3190 */
3191void submit_bio(int rw, struct bio *bio)
3192{
3193 int count = bio_sectors(bio);
3194
3195 BIO_BUG_ON(!bio->bi_size);
3196 BIO_BUG_ON(!bio->bi_io_vec);
Jens Axboe22e2c502005-06-27 10:55:12 +02003197 bio->bi_rw |= rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 if (rw & WRITE)
Christoph Lameterf8891e52006-06-30 01:55:45 -07003199 count_vm_events(PGPGOUT, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 else
Christoph Lameterf8891e52006-06-30 01:55:45 -07003201 count_vm_events(PGPGIN, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202
3203 if (unlikely(block_dump)) {
3204 char b[BDEVNAME_SIZE];
3205 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3206 current->comm, current->pid,
3207 (rw & WRITE) ? "WRITE" : "READ",
3208 (unsigned long long)bio->bi_sector,
3209 bdevname(bio->bi_bdev,b));
3210 }
3211
3212 generic_make_request(bio);
3213}
3214
3215EXPORT_SYMBOL(submit_bio);
3216
Adrian Bunk93d17d32005-06-25 14:59:10 -07003217static void blk_recalc_rq_segments(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218{
3219 struct bio *bio, *prevbio = NULL;
3220 int nr_phys_segs, nr_hw_segs;
3221 unsigned int phys_size, hw_size;
3222 request_queue_t *q = rq->q;
3223
3224 if (!rq->bio)
3225 return;
3226
3227 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3228 rq_for_each_bio(bio, rq) {
3229 /* Force bio hw/phys segs to be recalculated. */
3230 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3231
3232 nr_phys_segs += bio_phys_segments(q, bio);
3233 nr_hw_segs += bio_hw_segments(q, bio);
3234 if (prevbio) {
3235 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3236 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3237
3238 if (blk_phys_contig_segment(q, prevbio, bio) &&
3239 pseg <= q->max_segment_size) {
3240 nr_phys_segs--;
3241 phys_size += prevbio->bi_size + bio->bi_size;
3242 } else
3243 phys_size = 0;
3244
3245 if (blk_hw_contig_segment(q, prevbio, bio) &&
3246 hseg <= q->max_segment_size) {
3247 nr_hw_segs--;
3248 hw_size += prevbio->bi_size + bio->bi_size;
3249 } else
3250 hw_size = 0;
3251 }
3252 prevbio = bio;
3253 }
3254
3255 rq->nr_phys_segments = nr_phys_segs;
3256 rq->nr_hw_segments = nr_hw_segs;
3257}
3258
Adrian Bunk93d17d32005-06-25 14:59:10 -07003259static void blk_recalc_rq_sectors(struct request *rq, int nsect)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260{
3261 if (blk_fs_request(rq)) {
3262 rq->hard_sector += nsect;
3263 rq->hard_nr_sectors -= nsect;
3264
3265 /*
3266 * Move the I/O submission pointers ahead if required.
3267 */
3268 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3269 (rq->sector <= rq->hard_sector)) {
3270 rq->sector = rq->hard_sector;
3271 rq->nr_sectors = rq->hard_nr_sectors;
3272 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3273 rq->current_nr_sectors = rq->hard_cur_sectors;
3274 rq->buffer = bio_data(rq->bio);
3275 }
3276
3277 /*
3278 * if total number of sectors is less than the first segment
3279 * size, something has gone terribly wrong
3280 */
3281 if (rq->nr_sectors < rq->current_nr_sectors) {
3282 printk("blk: request botched\n");
3283 rq->nr_sectors = rq->current_nr_sectors;
3284 }
3285 }
3286}
3287
3288static int __end_that_request_first(struct request *req, int uptodate,
3289 int nr_bytes)
3290{
3291 int total_bytes, bio_nbytes, error, next_idx = 0;
3292 struct bio *bio;
3293
Jens Axboe2056a782006-03-23 20:00:26 +01003294 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3295
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 /*
3297 * extend uptodate bool to allow < 0 value to be direct io error
3298 */
3299 error = 0;
3300 if (end_io_error(uptodate))
3301 error = !uptodate ? -EIO : uptodate;
3302
3303 /*
3304 * for a REQ_BLOCK_PC request, we want to carry any eventual
3305 * sense key with us all the way through
3306 */
3307 if (!blk_pc_request(req))
3308 req->errors = 0;
3309
3310 if (!uptodate) {
Jens Axboe4aff5e22006-08-10 08:44:47 +02003311 if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 printk("end_request: I/O error, dev %s, sector %llu\n",
3313 req->rq_disk ? req->rq_disk->disk_name : "?",
3314 (unsigned long long)req->sector);
3315 }
3316
Jens Axboed72d9042005-11-01 08:35:42 +01003317 if (blk_fs_request(req) && req->rq_disk) {
Jens Axboea3623572005-11-01 09:26:16 +01003318 const int rw = rq_data_dir(req);
3319
Jens Axboe53e86062006-01-17 11:09:27 +01003320 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
Jens Axboed72d9042005-11-01 08:35:42 +01003321 }
3322
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 total_bytes = bio_nbytes = 0;
3324 while ((bio = req->bio) != NULL) {
3325 int nbytes;
3326
3327 if (nr_bytes >= bio->bi_size) {
3328 req->bio = bio->bi_next;
3329 nbytes = bio->bi_size;
Tejun Heo797e7db2006-01-06 09:51:03 +01003330 if (!ordered_bio_endio(req, bio, nbytes, error))
3331 bio_endio(bio, nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 next_idx = 0;
3333 bio_nbytes = 0;
3334 } else {
3335 int idx = bio->bi_idx + next_idx;
3336
3337 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3338 blk_dump_rq_flags(req, "__end_that");
3339 printk("%s: bio idx %d >= vcnt %d\n",
3340 __FUNCTION__,
3341 bio->bi_idx, bio->bi_vcnt);
3342 break;
3343 }
3344
3345 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3346 BIO_BUG_ON(nbytes > bio->bi_size);
3347
3348 /*
3349 * not a complete bvec done
3350 */
3351 if (unlikely(nbytes > nr_bytes)) {
3352 bio_nbytes += nr_bytes;
3353 total_bytes += nr_bytes;
3354 break;
3355 }
3356
3357 /*
3358 * advance to the next vector
3359 */
3360 next_idx++;
3361 bio_nbytes += nbytes;
3362 }
3363
3364 total_bytes += nbytes;
3365 nr_bytes -= nbytes;
3366
3367 if ((bio = req->bio)) {
3368 /*
3369 * end more in this run, or just return 'not-done'
3370 */
3371 if (unlikely(nr_bytes <= 0))
3372 break;
3373 }
3374 }
3375
3376 /*
3377 * completely done
3378 */
3379 if (!req->bio)
3380 return 0;
3381
3382 /*
3383 * if the request wasn't completed, update state
3384 */
3385 if (bio_nbytes) {
Tejun Heo797e7db2006-01-06 09:51:03 +01003386 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3387 bio_endio(bio, bio_nbytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 bio->bi_idx += next_idx;
3389 bio_iovec(bio)->bv_offset += nr_bytes;
3390 bio_iovec(bio)->bv_len -= nr_bytes;
3391 }
3392
3393 blk_recalc_rq_sectors(req, total_bytes >> 9);
3394 blk_recalc_rq_segments(req);
3395 return 1;
3396}
3397
3398/**
3399 * end_that_request_first - end I/O on a request
3400 * @req: the request being processed
3401 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3402 * @nr_sectors: number of sectors to end I/O on
3403 *
3404 * Description:
3405 * Ends I/O on a number of sectors attached to @req, and sets it up
3406 * for the next range of segments (if any) in the cluster.
3407 *
3408 * Return:
3409 * 0 - we are done with this request, call end_that_request_last()
3410 * 1 - still buffers pending for this request
3411 **/
3412int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3413{
3414 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3415}
3416
3417EXPORT_SYMBOL(end_that_request_first);
3418
3419/**
3420 * end_that_request_chunk - end I/O on a request
3421 * @req: the request being processed
3422 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3423 * @nr_bytes: number of bytes to complete
3424 *
3425 * Description:
3426 * Ends I/O on a number of bytes attached to @req, and sets it up
3427 * for the next range of segments (if any). Like end_that_request_first(),
3428 * but deals with bytes instead of sectors.
3429 *
3430 * Return:
3431 * 0 - we are done with this request, call end_that_request_last()
3432 * 1 - still buffers pending for this request
3433 **/
3434int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3435{
3436 return __end_that_request_first(req, uptodate, nr_bytes);
3437}
3438
3439EXPORT_SYMBOL(end_that_request_chunk);
3440
3441/*
Jens Axboeff856ba2006-01-09 16:02:34 +01003442 * splice the completion data to a local structure and hand off to
3443 * process_completion_queue() to complete the requests
3444 */
3445static void blk_done_softirq(struct softirq_action *h)
3446{
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003447 struct list_head *cpu_list, local_list;
Jens Axboeff856ba2006-01-09 16:02:34 +01003448
3449 local_irq_disable();
3450 cpu_list = &__get_cpu_var(blk_cpu_done);
Oleg Nesterov626ab0e2006-06-23 02:05:55 -07003451 list_replace_init(cpu_list, &local_list);
Jens Axboeff856ba2006-01-09 16:02:34 +01003452 local_irq_enable();
3453
3454 while (!list_empty(&local_list)) {
3455 struct request *rq = list_entry(local_list.next, struct request, donelist);
3456
3457 list_del_init(&rq->donelist);
3458 rq->q->softirq_done_fn(rq);
3459 }
3460}
3461
3462#ifdef CONFIG_HOTPLUG_CPU
3463
3464static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3465 void *hcpu)
3466{
3467 /*
3468 * If a CPU goes away, splice its entries to the current CPU
3469 * and trigger a run of the softirq
3470 */
3471 if (action == CPU_DEAD) {
3472 int cpu = (unsigned long) hcpu;
3473
3474 local_irq_disable();
3475 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3476 &__get_cpu_var(blk_cpu_done));
3477 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3478 local_irq_enable();
3479 }
3480
3481 return NOTIFY_OK;
3482}
3483
3484
Chandra Seetharaman054cc8a2006-06-27 02:54:07 -07003485static struct notifier_block __devinitdata blk_cpu_notifier = {
Jens Axboeff856ba2006-01-09 16:02:34 +01003486 .notifier_call = blk_cpu_notify,
3487};
3488
3489#endif /* CONFIG_HOTPLUG_CPU */
3490
3491/**
3492 * blk_complete_request - end I/O on a request
3493 * @req: the request being processed
3494 *
3495 * Description:
3496 * Ends all I/O on a request. It does not handle partial completions,
Andreas Mohrd6e05ed2006-06-26 18:35:02 +02003497 * unless the driver actually implements this in its completion callback
Jens Axboeff856ba2006-01-09 16:02:34 +01003498 * through requeueing. Theh actual completion happens out-of-order,
3499 * through a softirq handler. The user must have registered a completion
3500 * callback through blk_queue_softirq_done().
3501 **/
3502
3503void blk_complete_request(struct request *req)
3504{
3505 struct list_head *cpu_list;
3506 unsigned long flags;
3507
3508 BUG_ON(!req->q->softirq_done_fn);
3509
3510 local_irq_save(flags);
3511
3512 cpu_list = &__get_cpu_var(blk_cpu_done);
3513 list_add_tail(&req->donelist, cpu_list);
3514 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3515
3516 local_irq_restore(flags);
3517}
3518
3519EXPORT_SYMBOL(blk_complete_request);
3520
3521/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 * queue lock must be held
3523 */
Tejun Heo8ffdc652006-01-06 09:49:03 +01003524void end_that_request_last(struct request *req, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525{
3526 struct gendisk *disk = req->rq_disk;
Tejun Heo8ffdc652006-01-06 09:49:03 +01003527 int error;
3528
3529 /*
3530 * extend uptodate bool to allow < 0 value to be direct io error
3531 */
3532 error = 0;
3533 if (end_io_error(uptodate))
3534 error = !uptodate ? -EIO : uptodate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535
3536 if (unlikely(laptop_mode) && blk_fs_request(req))
3537 laptop_io_completion();
3538
Jens Axboefd0ff8a2006-05-23 11:23:49 +02003539 /*
3540 * Account IO completion. bar_rq isn't accounted as a normal
3541 * IO on queueing nor completion. Accounting the containing
3542 * request is enough.
3543 */
3544 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 unsigned long duration = jiffies - req->start_time;
Jens Axboea3623572005-11-01 09:26:16 +01003546 const int rw = rq_data_dir(req);
3547
3548 __disk_stat_inc(disk, ios[rw]);
3549 __disk_stat_add(disk, ticks[rw], duration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 disk_round_stats(disk);
3551 disk->in_flight--;
3552 }
3553 if (req->end_io)
Tejun Heo8ffdc652006-01-06 09:49:03 +01003554 req->end_io(req, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 else
3556 __blk_put_request(req->q, req);
3557}
3558
3559EXPORT_SYMBOL(end_that_request_last);
3560
3561void end_request(struct request *req, int uptodate)
3562{
3563 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3564 add_disk_randomness(req->rq_disk);
3565 blkdev_dequeue_request(req);
Tejun Heo8ffdc652006-01-06 09:49:03 +01003566 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 }
3568}
3569
3570EXPORT_SYMBOL(end_request);
3571
3572void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3573{
Jens Axboe4aff5e22006-08-10 08:44:47 +02003574 /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
3575 rq->cmd_flags |= (bio->bi_rw & 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576
3577 rq->nr_phys_segments = bio_phys_segments(q, bio);
3578 rq->nr_hw_segments = bio_hw_segments(q, bio);
3579 rq->current_nr_sectors = bio_cur_sectors(bio);
3580 rq->hard_cur_sectors = rq->current_nr_sectors;
3581 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3582 rq->buffer = bio_data(bio);
Mike Christie0e75f902006-12-01 10:40:55 +01003583 rq->data_len = bio->bi_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584
3585 rq->bio = rq->biotail = bio;
3586}
3587
3588EXPORT_SYMBOL(blk_rq_bio_prep);
3589
3590int kblockd_schedule_work(struct work_struct *work)
3591{
3592 return queue_work(kblockd_workqueue, work);
3593}
3594
3595EXPORT_SYMBOL(kblockd_schedule_work);
3596
3597void kblockd_flush(void)
3598{
3599 flush_workqueue(kblockd_workqueue);
3600}
3601EXPORT_SYMBOL(kblockd_flush);
3602
3603int __init blk_dev_init(void)
3604{
Jens Axboeff856ba2006-01-09 16:02:34 +01003605 int i;
3606
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607 kblockd_workqueue = create_workqueue("kblockd");
3608 if (!kblockd_workqueue)
3609 panic("Failed to create kblockd\n");
3610
3611 request_cachep = kmem_cache_create("blkdev_requests",
3612 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3613
3614 requestq_cachep = kmem_cache_create("blkdev_queue",
3615 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3616
3617 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3618 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3619
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -08003620 for_each_possible_cpu(i)
Jens Axboeff856ba2006-01-09 16:02:34 +01003621 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3622
3623 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
Chandra Seetharaman5a67e4c2006-06-27 02:54:11 -07003624 register_hotcpu_notifier(&blk_cpu_notifier);
Jens Axboeff856ba2006-01-09 16:02:34 +01003625
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 blk_max_low_pfn = max_low_pfn;
3627 blk_max_pfn = max_pfn;
3628
3629 return 0;
3630}
3631
3632/*
3633 * IO Context helper functions
3634 */
3635void put_io_context(struct io_context *ioc)
3636{
3637 if (ioc == NULL)
3638 return;
3639
3640 BUG_ON(atomic_read(&ioc->refcount) == 0);
3641
3642 if (atomic_dec_and_test(&ioc->refcount)) {
Jens Axboee2d74ac2006-03-28 08:59:01 +02003643 struct cfq_io_context *cic;
3644
Al Viro334e94d2006-03-18 15:05:53 -05003645 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 if (ioc->aic && ioc->aic->dtor)
3647 ioc->aic->dtor(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003648 if (ioc->cic_root.rb_node != NULL) {
Jens Axboe7143dd42006-03-28 09:00:28 +02003649 struct rb_node *n = rb_first(&ioc->cic_root);
3650
3651 cic = rb_entry(n, struct cfq_io_context, rb_node);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003652 cic->dtor(ioc);
3653 }
Al Viro334e94d2006-03-18 15:05:53 -05003654 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655
3656 kmem_cache_free(iocontext_cachep, ioc);
3657 }
3658}
3659EXPORT_SYMBOL(put_io_context);
3660
3661/* Called by the exitting task */
3662void exit_io_context(void)
3663{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 struct io_context *ioc;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003665 struct cfq_io_context *cic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666
Jens Axboe22e2c502005-06-27 10:55:12 +02003667 task_lock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 ioc = current->io_context;
3669 current->io_context = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02003670 task_unlock(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671
Oleg Nesterov25034d72006-08-29 09:15:14 +02003672 ioc->task = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673 if (ioc->aic && ioc->aic->exit)
3674 ioc->aic->exit(ioc->aic);
Jens Axboee2d74ac2006-03-28 08:59:01 +02003675 if (ioc->cic_root.rb_node != NULL) {
3676 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3677 cic->exit(ioc);
3678 }
Oleg Nesterov25034d72006-08-29 09:15:14 +02003679
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 put_io_context(ioc);
3681}
3682
3683/*
3684 * If the current task has no IO context then create one and initialise it.
Nick Pigginfb3cc432005-06-28 20:45:15 -07003685 * Otherwise, return its existing IO context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686 *
Nick Pigginfb3cc432005-06-28 20:45:15 -07003687 * This returned IO context doesn't have a specifically elevated refcount,
3688 * but since the current task itself holds a reference, the context can be
3689 * used in general code, so long as it stays within `current` context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003691static struct io_context *current_io_context(gfp_t gfp_flags, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692{
3693 struct task_struct *tsk = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 struct io_context *ret;
3695
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 ret = tsk->io_context;
Nick Pigginfb3cc432005-06-28 20:45:15 -07003697 if (likely(ret))
3698 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699
Jens Axboeb5deef92006-07-19 23:39:40 +02003700 ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 if (ret) {
3702 atomic_set(&ret->refcount, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02003703 ret->task = current;
Jens Axboefc463792006-08-29 09:05:44 +02003704 ret->ioprio_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705 ret->last_waited = jiffies; /* doesn't matter... */
3706 ret->nr_batch_requests = 0; /* because this is 0 */
3707 ret->aic = NULL;
Jens Axboee2d74ac2006-03-28 08:59:01 +02003708 ret->cic_root.rb_node = NULL;
Oleg Nesterov9f83e452006-08-21 08:34:15 +02003709 /* make sure set_task_ioprio() sees the settings above */
3710 smp_wmb();
Nick Pigginfb3cc432005-06-28 20:45:15 -07003711 tsk->io_context = ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 }
3713
3714 return ret;
3715}
Nick Pigginfb3cc432005-06-28 20:45:15 -07003716EXPORT_SYMBOL(current_io_context);
3717
3718/*
3719 * If the current task has no IO context then create one and initialise it.
3720 * If it does have a context, take a ref on it.
3721 *
3722 * This is always called in the context of the task which submitted the I/O.
3723 */
Jens Axboeb5deef92006-07-19 23:39:40 +02003724struct io_context *get_io_context(gfp_t gfp_flags, int node)
Nick Pigginfb3cc432005-06-28 20:45:15 -07003725{
3726 struct io_context *ret;
Jens Axboeb5deef92006-07-19 23:39:40 +02003727 ret = current_io_context(gfp_flags, node);
Nick Pigginfb3cc432005-06-28 20:45:15 -07003728 if (likely(ret))
3729 atomic_inc(&ret->refcount);
3730 return ret;
3731}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003732EXPORT_SYMBOL(get_io_context);
3733
3734void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3735{
3736 struct io_context *src = *psrc;
3737 struct io_context *dst = *pdst;
3738
3739 if (src) {
3740 BUG_ON(atomic_read(&src->refcount) == 0);
3741 atomic_inc(&src->refcount);
3742 put_io_context(dst);
3743 *pdst = src;
3744 }
3745}
3746EXPORT_SYMBOL(copy_io_context);
3747
3748void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3749{
3750 struct io_context *temp;
3751 temp = *ioc1;
3752 *ioc1 = *ioc2;
3753 *ioc2 = temp;
3754}
3755EXPORT_SYMBOL(swap_io_context);
3756
3757/*
3758 * sysfs parts below
3759 */
3760struct queue_sysfs_entry {
3761 struct attribute attr;
3762 ssize_t (*show)(struct request_queue *, char *);
3763 ssize_t (*store)(struct request_queue *, const char *, size_t);
3764};
3765
3766static ssize_t
3767queue_var_show(unsigned int var, char *page)
3768{
3769 return sprintf(page, "%d\n", var);
3770}
3771
3772static ssize_t
3773queue_var_store(unsigned long *var, const char *page, size_t count)
3774{
3775 char *p = (char *) page;
3776
3777 *var = simple_strtoul(p, &p, 10);
3778 return count;
3779}
3780
3781static ssize_t queue_requests_show(struct request_queue *q, char *page)
3782{
3783 return queue_var_show(q->nr_requests, (page));
3784}
3785
3786static ssize_t
3787queue_requests_store(struct request_queue *q, const char *page, size_t count)
3788{
3789 struct request_list *rl = &q->rq;
Al Viroc981ff92006-03-18 13:51:29 -05003790 unsigned long nr;
3791 int ret = queue_var_store(&nr, page, count);
3792 if (nr < BLKDEV_MIN_RQ)
3793 nr = BLKDEV_MIN_RQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794
Al Viroc981ff92006-03-18 13:51:29 -05003795 spin_lock_irq(q->queue_lock);
3796 q->nr_requests = nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 blk_queue_congestion_threshold(q);
3798
3799 if (rl->count[READ] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003800 blk_set_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 else if (rl->count[READ] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003802 blk_clear_queue_congested(q, READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803
3804 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003805 blk_set_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
Thomas Maier79e2de42006-10-19 23:28:15 -07003807 blk_clear_queue_congested(q, WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808
3809 if (rl->count[READ] >= q->nr_requests) {
3810 blk_set_queue_full(q, READ);
3811 } else if (rl->count[READ]+1 <= q->nr_requests) {
3812 blk_clear_queue_full(q, READ);
3813 wake_up(&rl->wait[READ]);
3814 }
3815
3816 if (rl->count[WRITE] >= q->nr_requests) {
3817 blk_set_queue_full(q, WRITE);
3818 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3819 blk_clear_queue_full(q, WRITE);
3820 wake_up(&rl->wait[WRITE]);
3821 }
Al Viroc981ff92006-03-18 13:51:29 -05003822 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 return ret;
3824}
3825
3826static ssize_t queue_ra_show(struct request_queue *q, char *page)
3827{
3828 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3829
3830 return queue_var_show(ra_kb, (page));
3831}
3832
3833static ssize_t
3834queue_ra_store(struct request_queue *q, const char *page, size_t count)
3835{
3836 unsigned long ra_kb;
3837 ssize_t ret = queue_var_store(&ra_kb, page, count);
3838
3839 spin_lock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3841 spin_unlock_irq(q->queue_lock);
3842
3843 return ret;
3844}
3845
3846static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3847{
3848 int max_sectors_kb = q->max_sectors >> 1;
3849
3850 return queue_var_show(max_sectors_kb, (page));
3851}
3852
3853static ssize_t
3854queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3855{
3856 unsigned long max_sectors_kb,
3857 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3858 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3859 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3860 int ra_kb;
3861
3862 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3863 return -EINVAL;
3864 /*
3865 * Take the queue lock to update the readahead and max_sectors
3866 * values synchronously:
3867 */
3868 spin_lock_irq(q->queue_lock);
3869 /*
3870 * Trim readahead window as well, if necessary:
3871 */
3872 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3873 if (ra_kb > max_sectors_kb)
3874 q->backing_dev_info.ra_pages =
3875 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3876
3877 q->max_sectors = max_sectors_kb << 1;
3878 spin_unlock_irq(q->queue_lock);
3879
3880 return ret;
3881}
3882
3883static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3884{
3885 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3886
3887 return queue_var_show(max_hw_sectors_kb, (page));
3888}
3889
3890
3891static struct queue_sysfs_entry queue_requests_entry = {
3892 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3893 .show = queue_requests_show,
3894 .store = queue_requests_store,
3895};
3896
3897static struct queue_sysfs_entry queue_ra_entry = {
3898 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3899 .show = queue_ra_show,
3900 .store = queue_ra_store,
3901};
3902
3903static struct queue_sysfs_entry queue_max_sectors_entry = {
3904 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3905 .show = queue_max_sectors_show,
3906 .store = queue_max_sectors_store,
3907};
3908
3909static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3910 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3911 .show = queue_max_hw_sectors_show,
3912};
3913
3914static struct queue_sysfs_entry queue_iosched_entry = {
3915 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3916 .show = elv_iosched_show,
3917 .store = elv_iosched_store,
3918};
3919
3920static struct attribute *default_attrs[] = {
3921 &queue_requests_entry.attr,
3922 &queue_ra_entry.attr,
3923 &queue_max_hw_sectors_entry.attr,
3924 &queue_max_sectors_entry.attr,
3925 &queue_iosched_entry.attr,
3926 NULL,
3927};
3928
3929#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3930
3931static ssize_t
3932queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3933{
3934 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003935 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3936 ssize_t res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937
Linus Torvalds1da177e2005-04-16 15:20:36 -07003938 if (!entry->show)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003939 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003940 mutex_lock(&q->sysfs_lock);
3941 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3942 mutex_unlock(&q->sysfs_lock);
3943 return -ENOENT;
3944 }
3945 res = entry->show(q, page);
3946 mutex_unlock(&q->sysfs_lock);
3947 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948}
3949
3950static ssize_t
3951queue_attr_store(struct kobject *kobj, struct attribute *attr,
3952 const char *page, size_t length)
3953{
3954 struct queue_sysfs_entry *entry = to_queue(attr);
Al Viro483f4af2006-03-18 18:34:37 -05003955 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003956
Al Viro483f4af2006-03-18 18:34:37 -05003957 ssize_t res;
3958
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 if (!entry->store)
Dmitry Torokhov6c1852a2005-04-29 01:26:06 -05003960 return -EIO;
Al Viro483f4af2006-03-18 18:34:37 -05003961 mutex_lock(&q->sysfs_lock);
3962 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3963 mutex_unlock(&q->sysfs_lock);
3964 return -ENOENT;
3965 }
3966 res = entry->store(q, page, length);
3967 mutex_unlock(&q->sysfs_lock);
3968 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003969}
3970
3971static struct sysfs_ops queue_sysfs_ops = {
3972 .show = queue_attr_show,
3973 .store = queue_attr_store,
3974};
3975
Adrian Bunk93d17d32005-06-25 14:59:10 -07003976static struct kobj_type queue_ktype = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 .sysfs_ops = &queue_sysfs_ops,
3978 .default_attrs = default_attrs,
Al Viro483f4af2006-03-18 18:34:37 -05003979 .release = blk_release_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980};
3981
3982int blk_register_queue(struct gendisk *disk)
3983{
3984 int ret;
3985
3986 request_queue_t *q = disk->queue;
3987
3988 if (!q || !q->request_fn)
3989 return -ENXIO;
3990
3991 q->kobj.parent = kobject_get(&disk->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992
Al Viro483f4af2006-03-18 18:34:37 -05003993 ret = kobject_add(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 if (ret < 0)
3995 return ret;
3996
Al Viro483f4af2006-03-18 18:34:37 -05003997 kobject_uevent(&q->kobj, KOBJ_ADD);
3998
Linus Torvalds1da177e2005-04-16 15:20:36 -07003999 ret = elv_register_queue(q);
4000 if (ret) {
Al Viro483f4af2006-03-18 18:34:37 -05004001 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4002 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003 return ret;
4004 }
4005
4006 return 0;
4007}
4008
4009void blk_unregister_queue(struct gendisk *disk)
4010{
4011 request_queue_t *q = disk->queue;
4012
4013 if (q && q->request_fn) {
4014 elv_unregister_queue(q);
4015
Al Viro483f4af2006-03-18 18:34:37 -05004016 kobject_uevent(&q->kobj, KOBJ_REMOVE);
4017 kobject_del(&q->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004018 kobject_put(&disk->kobj);
4019 }
4020}