Commit | Line | Data |
---|---|---|
86db1e29 JA |
1 | /* |
2 | * Functions related to barrier IO handling | |
3 | */ | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/bio.h> | |
7 | #include <linux/blkdev.h> | |
5a0e3ad6 | 8 | #include <linux/gfp.h> |
86db1e29 JA |
9 | |
10 | #include "blk.h" | |
11 | ||
12 | /** | |
13 | * blk_queue_ordered - does this queue support ordered writes | |
14 | * @q: the request queue | |
15 | * @ordered: one of QUEUE_ORDERED_* | |
16 | * @prepare_flush_fn: rq setup helper for cache flush ordered writes | |
17 | * | |
18 | * Description: | |
19 | * For journalled file systems, doing ordered writes on a commit | |
20 | * block instead of explicitly doing wait_on_buffer (which is bad | |
21 | * for performance) can be a big win. Block drivers supporting this | |
22 | * feature should call this function and indicate so. | |
23 | * | |
24 | **/ | |
25 | int blk_queue_ordered(struct request_queue *q, unsigned ordered, | |
26 | prepare_flush_fn *prepare_flush_fn) | |
27 | { | |
313e4299 TH |
28 | if (!prepare_flush_fn && (ordered & (QUEUE_ORDERED_DO_PREFLUSH | |
29 | QUEUE_ORDERED_DO_POSTFLUSH))) { | |
24c03d47 | 30 | printk(KERN_ERR "%s: prepare_flush_fn required\n", __func__); |
86db1e29 JA |
31 | return -EINVAL; |
32 | } | |
33 | ||
34 | if (ordered != QUEUE_ORDERED_NONE && | |
35 | ordered != QUEUE_ORDERED_DRAIN && | |
36 | ordered != QUEUE_ORDERED_DRAIN_FLUSH && | |
37 | ordered != QUEUE_ORDERED_DRAIN_FUA && | |
38 | ordered != QUEUE_ORDERED_TAG && | |
39 | ordered != QUEUE_ORDERED_TAG_FLUSH && | |
40 | ordered != QUEUE_ORDERED_TAG_FUA) { | |
41 | printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); | |
42 | return -EINVAL; | |
43 | } | |
44 | ||
45 | q->ordered = ordered; | |
46 | q->next_ordered = ordered; | |
47 | q->prepare_flush_fn = prepare_flush_fn; | |
48 | ||
49 | return 0; | |
50 | } | |
86db1e29 JA |
51 | EXPORT_SYMBOL(blk_queue_ordered); |
52 | ||
53 | /* | |
54 | * Cache flushing for ordered writes handling | |
55 | */ | |
6f6a036e | 56 | unsigned blk_ordered_cur_seq(struct request_queue *q) |
86db1e29 JA |
57 | { |
58 | if (!q->ordseq) | |
59 | return 0; | |
60 | return 1 << ffz(q->ordseq); | |
61 | } | |
62 | ||
63 | unsigned blk_ordered_req_seq(struct request *rq) | |
64 | { | |
65 | struct request_queue *q = rq->q; | |
66 | ||
67 | BUG_ON(q->ordseq == 0); | |
68 | ||
69 | if (rq == &q->pre_flush_rq) | |
70 | return QUEUE_ORDSEQ_PREFLUSH; | |
71 | if (rq == &q->bar_rq) | |
72 | return QUEUE_ORDSEQ_BAR; | |
73 | if (rq == &q->post_flush_rq) | |
74 | return QUEUE_ORDSEQ_POSTFLUSH; | |
75 | ||
76 | /* | |
77 | * !fs requests don't need to follow barrier ordering. Always | |
78 | * put them at the front. This fixes the following deadlock. | |
79 | * | |
80 | * http://thread.gmane.org/gmane.linux.kernel/537473 | |
81 | */ | |
82 | if (!blk_fs_request(rq)) | |
83 | return QUEUE_ORDSEQ_DRAIN; | |
84 | ||
85 | if ((rq->cmd_flags & REQ_ORDERED_COLOR) == | |
86 | (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR)) | |
87 | return QUEUE_ORDSEQ_DRAIN; | |
88 | else | |
89 | return QUEUE_ORDSEQ_DONE; | |
90 | } | |
91 | ||
8f11b3e9 | 92 | bool blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error) |
86db1e29 JA |
93 | { |
94 | struct request *rq; | |
95 | ||
96 | if (error && !q->orderr) | |
97 | q->orderr = error; | |
98 | ||
99 | BUG_ON(q->ordseq & seq); | |
100 | q->ordseq |= seq; | |
101 | ||
102 | if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) | |
8f11b3e9 | 103 | return false; |
86db1e29 JA |
104 | |
105 | /* | |
106 | * Okay, sequence complete. | |
107 | */ | |
108 | q->ordseq = 0; | |
109 | rq = q->orig_bar_rq; | |
40cbbb78 | 110 | __blk_end_request_all(rq, q->orderr); |
8f11b3e9 | 111 | return true; |
86db1e29 JA |
112 | } |
113 | ||
114 | static void pre_flush_end_io(struct request *rq, int error) | |
115 | { | |
116 | elv_completed_request(rq->q, rq); | |
117 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); | |
118 | } | |
119 | ||
120 | static void bar_end_io(struct request *rq, int error) | |
121 | { | |
122 | elv_completed_request(rq->q, rq); | |
123 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); | |
124 | } | |
125 | ||
126 | static void post_flush_end_io(struct request *rq, int error) | |
127 | { | |
128 | elv_completed_request(rq->q, rq); | |
129 | blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); | |
130 | } | |
131 | ||
132 | static void queue_flush(struct request_queue *q, unsigned which) | |
133 | { | |
134 | struct request *rq; | |
135 | rq_end_io_fn *end_io; | |
136 | ||
313e4299 | 137 | if (which == QUEUE_ORDERED_DO_PREFLUSH) { |
86db1e29 JA |
138 | rq = &q->pre_flush_rq; |
139 | end_io = pre_flush_end_io; | |
140 | } else { | |
141 | rq = &q->post_flush_rq; | |
142 | end_io = post_flush_end_io; | |
143 | } | |
144 | ||
2a4aa30c | 145 | blk_rq_init(q, rq); |
1afb20f3 | 146 | rq->cmd_flags = REQ_HARDBARRIER; |
86db1e29 JA |
147 | rq->rq_disk = q->bar_rq.rq_disk; |
148 | rq->end_io = end_io; | |
149 | q->prepare_flush_fn(q, rq); | |
150 | ||
151 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); | |
152 | } | |
153 | ||
8f11b3e9 | 154 | static inline bool start_ordered(struct request_queue *q, struct request **rqp) |
86db1e29 | 155 | { |
8f11b3e9 TH |
156 | struct request *rq = *rqp; |
157 | unsigned skip = 0; | |
158 | ||
86db1e29 JA |
159 | q->orderr = 0; |
160 | q->ordered = q->next_ordered; | |
161 | q->ordseq |= QUEUE_ORDSEQ_STARTED; | |
162 | ||
58eea927 TH |
163 | /* |
164 | * For an empty barrier, there's no actual BAR request, which | |
165 | * in turn makes POSTFLUSH unnecessary. Mask them off. | |
166 | */ | |
5b93629b | 167 | if (!blk_rq_sectors(rq)) { |
58eea927 TH |
168 | q->ordered &= ~(QUEUE_ORDERED_DO_BAR | |
169 | QUEUE_ORDERED_DO_POSTFLUSH); | |
a185eb4b TH |
170 | /* |
171 | * Empty barrier on a write-through device w/ ordered | |
172 | * tag has no command to issue and without any command | |
173 | * to issue, ordering by tag can't be used. Drain | |
174 | * instead. | |
175 | */ | |
176 | if ((q->ordered & QUEUE_ORDERED_BY_TAG) && | |
177 | !(q->ordered & QUEUE_ORDERED_DO_PREFLUSH)) { | |
178 | q->ordered &= ~QUEUE_ORDERED_BY_TAG; | |
179 | q->ordered |= QUEUE_ORDERED_BY_DRAIN; | |
180 | } | |
181 | } | |
58eea927 | 182 | |
f671620e | 183 | /* stash away the original request */ |
9934c8c0 | 184 | blk_dequeue_request(rq); |
86db1e29 | 185 | q->orig_bar_rq = rq; |
f671620e | 186 | rq = NULL; |
86db1e29 JA |
187 | |
188 | /* | |
189 | * Queue ordered sequence. As we stack them at the head, we | |
190 | * need to queue in reverse order. Note that we rely on that | |
191 | * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs | |
58eea927 | 192 | * request gets inbetween ordered sequence. |
86db1e29 | 193 | */ |
58eea927 | 194 | if (q->ordered & QUEUE_ORDERED_DO_POSTFLUSH) { |
313e4299 | 195 | queue_flush(q, QUEUE_ORDERED_DO_POSTFLUSH); |
f671620e TH |
196 | rq = &q->post_flush_rq; |
197 | } else | |
8f11b3e9 | 198 | skip |= QUEUE_ORDSEQ_POSTFLUSH; |
86db1e29 | 199 | |
f671620e TH |
200 | if (q->ordered & QUEUE_ORDERED_DO_BAR) { |
201 | rq = &q->bar_rq; | |
202 | ||
203 | /* initialize proxy request and queue it */ | |
204 | blk_rq_init(q, rq); | |
205 | if (bio_data_dir(q->orig_bar_rq->bio) == WRITE) | |
206 | rq->cmd_flags |= REQ_RW; | |
207 | if (q->ordered & QUEUE_ORDERED_DO_FUA) | |
208 | rq->cmd_flags |= REQ_FUA; | |
209 | init_request_from_bio(rq, q->orig_bar_rq->bio); | |
210 | rq->end_io = bar_end_io; | |
211 | ||
212 | elv_insert(q, rq, ELEVATOR_INSERT_FRONT); | |
213 | } else | |
8f11b3e9 | 214 | skip |= QUEUE_ORDSEQ_BAR; |
86db1e29 | 215 | |
313e4299 TH |
216 | if (q->ordered & QUEUE_ORDERED_DO_PREFLUSH) { |
217 | queue_flush(q, QUEUE_ORDERED_DO_PREFLUSH); | |
86db1e29 JA |
218 | rq = &q->pre_flush_rq; |
219 | } else | |
8f11b3e9 | 220 | skip |= QUEUE_ORDSEQ_PREFLUSH; |
86db1e29 | 221 | |
0a7ae2ff | 222 | if ((q->ordered & QUEUE_ORDERED_BY_DRAIN) && queue_in_flight(q)) |
86db1e29 | 223 | rq = NULL; |
f671620e | 224 | else |
8f11b3e9 | 225 | skip |= QUEUE_ORDSEQ_DRAIN; |
86db1e29 | 226 | |
8f11b3e9 TH |
227 | *rqp = rq; |
228 | ||
229 | /* | |
230 | * Complete skipped sequences. If whole sequence is complete, | |
231 | * return false to tell elevator that this request is gone. | |
232 | */ | |
233 | return !blk_ordered_complete_seq(q, skip, 0); | |
86db1e29 JA |
234 | } |
235 | ||
8f11b3e9 | 236 | bool blk_do_ordered(struct request_queue *q, struct request **rqp) |
86db1e29 JA |
237 | { |
238 | struct request *rq = *rqp; | |
239 | const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); | |
240 | ||
241 | if (!q->ordseq) { | |
242 | if (!is_barrier) | |
8f11b3e9 | 243 | return true; |
86db1e29 | 244 | |
8f11b3e9 TH |
245 | if (q->next_ordered != QUEUE_ORDERED_NONE) |
246 | return start_ordered(q, rqp); | |
247 | else { | |
86db1e29 | 248 | /* |
a7384677 TH |
249 | * Queue ordering not supported. Terminate |
250 | * with prejudice. | |
86db1e29 | 251 | */ |
9934c8c0 | 252 | blk_dequeue_request(rq); |
40cbbb78 | 253 | __blk_end_request_all(rq, -EOPNOTSUPP); |
86db1e29 | 254 | *rqp = NULL; |
8f11b3e9 | 255 | return false; |
86db1e29 JA |
256 | } |
257 | } | |
258 | ||
259 | /* | |
260 | * Ordered sequence in progress | |
261 | */ | |
262 | ||
263 | /* Special requests are not subject to ordering rules. */ | |
264 | if (!blk_fs_request(rq) && | |
265 | rq != &q->pre_flush_rq && rq != &q->post_flush_rq) | |
8f11b3e9 | 266 | return true; |
86db1e29 | 267 | |
313e4299 | 268 | if (q->ordered & QUEUE_ORDERED_BY_TAG) { |
86db1e29 JA |
269 | /* Ordered by tag. Blocking the next barrier is enough. */ |
270 | if (is_barrier && rq != &q->bar_rq) | |
271 | *rqp = NULL; | |
272 | } else { | |
273 | /* Ordered by draining. Wait for turn. */ | |
274 | WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); | |
275 | if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) | |
276 | *rqp = NULL; | |
277 | } | |
278 | ||
8f11b3e9 | 279 | return true; |
86db1e29 JA |
280 | } |
281 | ||
282 | static void bio_end_empty_barrier(struct bio *bio, int err) | |
283 | { | |
cc66b451 JA |
284 | if (err) { |
285 | if (err == -EOPNOTSUPP) | |
286 | set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); | |
86db1e29 | 287 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
cc66b451 | 288 | } |
f17e232e DM |
289 | if (bio->bi_private) |
290 | complete(bio->bi_private); | |
291 | bio_put(bio); | |
86db1e29 JA |
292 | } |
293 | ||
294 | /** | |
295 | * blkdev_issue_flush - queue a flush | |
296 | * @bdev: blockdev to issue flush for | |
fbd9b09a | 297 | * @gfp_mask: memory allocation flags (for bio_alloc) |
86db1e29 | 298 | * @error_sector: error sector |
fbd9b09a | 299 | * @flags: BLKDEV_IFL_* flags to control behaviour |
86db1e29 JA |
300 | * |
301 | * Description: | |
302 | * Issue a flush for the block device in question. Caller can supply | |
303 | * room for storing the error offset in case of a flush error, if they | |
f17e232e DM |
304 | * wish to. If WAIT flag is not passed then caller may check only what |
305 | * request was pushed in some internal queue for later handling. | |
86db1e29 | 306 | */ |
fbd9b09a DM |
307 | int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask, |
308 | sector_t *error_sector, unsigned long flags) | |
86db1e29 JA |
309 | { |
310 | DECLARE_COMPLETION_ONSTACK(wait); | |
311 | struct request_queue *q; | |
312 | struct bio *bio; | |
fbd9b09a | 313 | int ret = 0; |
86db1e29 JA |
314 | |
315 | if (bdev->bd_disk == NULL) | |
316 | return -ENXIO; | |
317 | ||
318 | q = bdev_get_queue(bdev); | |
319 | if (!q) | |
320 | return -ENXIO; | |
321 | ||
fbd9b09a | 322 | bio = bio_alloc(gfp_mask, 0); |
86db1e29 | 323 | bio->bi_end_io = bio_end_empty_barrier; |
86db1e29 | 324 | bio->bi_bdev = bdev; |
f17e232e DM |
325 | if (test_bit(BLKDEV_WAIT, &flags)) |
326 | bio->bi_private = &wait; | |
86db1e29 | 327 | |
f17e232e DM |
328 | bio_get(bio); |
329 | submit_bio(WRITE_BARRIER, bio); | |
330 | if (test_bit(BLKDEV_WAIT, &flags)) { | |
331 | wait_for_completion(&wait); | |
332 | /* | |
333 | * The driver must store the error location in ->bi_sector, if | |
334 | * it supports it. For non-stacked drivers, this should be | |
335 | * copied from blk_rq_pos(rq). | |
336 | */ | |
337 | if (error_sector) | |
338 | *error_sector = bio->bi_sector; | |
339 | } | |
86db1e29 | 340 | |
cc66b451 JA |
341 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) |
342 | ret = -EOPNOTSUPP; | |
343 | else if (!bio_flagged(bio, BIO_UPTODATE)) | |
86db1e29 JA |
344 | ret = -EIO; |
345 | ||
346 | bio_put(bio); | |
347 | return ret; | |
348 | } | |
86db1e29 | 349 | EXPORT_SYMBOL(blkdev_issue_flush); |