Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
1f965b19 | 3 | * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
1da177e4 | 8 | #include "dm-bio-list.h" |
06386bbf | 9 | #include "dm-bio-record.h" |
1da177e4 | 10 | |
1da177e4 LT |
11 | #include <linux/init.h> |
12 | #include <linux/mempool.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/slab.h> | |
1da177e4 | 16 | #include <linux/workqueue.h> |
1f965b19 | 17 | #include <linux/device-mapper.h> |
a765e20e AK |
18 | #include <linux/dm-io.h> |
19 | #include <linux/dm-dirty-log.h> | |
20 | #include <linux/dm-kcopyd.h> | |
1f965b19 | 21 | #include <linux/dm-region-hash.h> |
1da177e4 | 22 | |
72d94861 | 23 | #define DM_MSG_PREFIX "raid1" |
1f965b19 HM |
24 | |
25 | #define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */ | |
88be163a | 26 | #define DM_IO_PAGES 64 |
1f965b19 | 27 | #define DM_KCOPYD_PAGES 64 |
72d94861 | 28 | |
a8e6afa2 | 29 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
f44db678 | 30 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
a8e6afa2 | 31 | |
33184048 | 32 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
1da177e4 | 33 | |
e4c8b3ba NB |
34 | /*----------------------------------------------------------------- |
35 | * Mirror set structures. | |
36 | *---------------------------------------------------------------*/ | |
72f4b314 JB |
37 | enum dm_raid1_error { |
38 | DM_RAID1_WRITE_ERROR, | |
39 | DM_RAID1_SYNC_ERROR, | |
40 | DM_RAID1_READ_ERROR | |
41 | }; | |
42 | ||
e4c8b3ba | 43 | struct mirror { |
aa5617c5 | 44 | struct mirror_set *ms; |
e4c8b3ba | 45 | atomic_t error_count; |
39ed7adb | 46 | unsigned long error_type; |
e4c8b3ba NB |
47 | struct dm_dev *dev; |
48 | sector_t offset; | |
49 | }; | |
50 | ||
51 | struct mirror_set { | |
52 | struct dm_target *ti; | |
53 | struct list_head list; | |
1f965b19 | 54 | |
a8e6afa2 | 55 | uint64_t features; |
e4c8b3ba | 56 | |
72f4b314 | 57 | spinlock_t lock; /* protects the lists */ |
e4c8b3ba NB |
58 | struct bio_list reads; |
59 | struct bio_list writes; | |
72f4b314 | 60 | struct bio_list failures; |
e4c8b3ba | 61 | |
1f965b19 HM |
62 | struct dm_region_hash *rh; |
63 | struct dm_kcopyd_client *kcopyd_client; | |
88be163a | 64 | struct dm_io_client *io_client; |
06386bbf | 65 | mempool_t *read_record_pool; |
88be163a | 66 | |
e4c8b3ba NB |
67 | /* recovery */ |
68 | region_t nr_regions; | |
69 | int in_sync; | |
fc1ff958 | 70 | int log_failure; |
b80aa7a0 | 71 | atomic_t suspend; |
e4c8b3ba | 72 | |
72f4b314 | 73 | atomic_t default_mirror; /* Default mirror */ |
e4c8b3ba | 74 | |
6ad36fe2 HS |
75 | struct workqueue_struct *kmirrord_wq; |
76 | struct work_struct kmirrord_work; | |
a2aebe03 MP |
77 | struct timer_list timer; |
78 | unsigned long timer_pending; | |
79 | ||
72f4b314 | 80 | struct work_struct trigger_event; |
6ad36fe2 | 81 | |
1f965b19 | 82 | unsigned nr_mirrors; |
e4c8b3ba NB |
83 | struct mirror mirror[0]; |
84 | }; | |
85 | ||
1f965b19 | 86 | static void wakeup_mirrord(void *context) |
1da177e4 | 87 | { |
1f965b19 | 88 | struct mirror_set *ms = context; |
1da177e4 | 89 | |
6ad36fe2 HS |
90 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); |
91 | } | |
92 | ||
a2aebe03 MP |
93 | static void delayed_wake_fn(unsigned long data) |
94 | { | |
95 | struct mirror_set *ms = (struct mirror_set *) data; | |
96 | ||
97 | clear_bit(0, &ms->timer_pending); | |
1f965b19 | 98 | wakeup_mirrord(ms); |
a2aebe03 MP |
99 | } |
100 | ||
101 | static void delayed_wake(struct mirror_set *ms) | |
102 | { | |
103 | if (test_and_set_bit(0, &ms->timer_pending)) | |
104 | return; | |
105 | ||
106 | ms->timer.expires = jiffies + HZ / 5; | |
107 | ms->timer.data = (unsigned long) ms; | |
108 | ms->timer.function = delayed_wake_fn; | |
109 | add_timer(&ms->timer); | |
110 | } | |
111 | ||
1f965b19 | 112 | static void wakeup_all_recovery_waiters(void *context) |
1da177e4 | 113 | { |
1f965b19 | 114 | wake_up_all(&_kmirrord_recovery_stopped); |
1da177e4 LT |
115 | } |
116 | ||
1f965b19 | 117 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) |
1da177e4 LT |
118 | { |
119 | unsigned long flags; | |
1da177e4 | 120 | int should_wake = 0; |
1f965b19 | 121 | struct bio_list *bl; |
1da177e4 | 122 | |
1f965b19 HM |
123 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; |
124 | spin_lock_irqsave(&ms->lock, flags); | |
125 | should_wake = !(bl->head); | |
126 | bio_list_add(bl, bio); | |
127 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 LT |
128 | |
129 | if (should_wake) | |
1f965b19 | 130 | wakeup_mirrord(ms); |
1da177e4 LT |
131 | } |
132 | ||
1f965b19 | 133 | static void dispatch_bios(void *context, struct bio_list *bio_list) |
1da177e4 | 134 | { |
1f965b19 HM |
135 | struct mirror_set *ms = context; |
136 | struct bio *bio; | |
1da177e4 | 137 | |
1f965b19 HM |
138 | while ((bio = bio_list_pop(bio_list))) |
139 | queue_bio(ms, bio, WRITE); | |
1da177e4 LT |
140 | } |
141 | ||
06386bbf JB |
142 | #define MIN_READ_RECORDS 20 |
143 | struct dm_raid1_read_record { | |
144 | struct mirror *m; | |
145 | struct dm_bio_details details; | |
146 | }; | |
147 | ||
1da177e4 LT |
148 | /* |
149 | * Every mirror should look like this one. | |
150 | */ | |
151 | #define DEFAULT_MIRROR 0 | |
152 | ||
153 | /* | |
06386bbf JB |
154 | * This is yucky. We squirrel the mirror struct away inside |
155 | * bi_next for read/write buffers. This is safe since the bh | |
1da177e4 LT |
156 | * doesn't get submitted to the lower levels of block layer. |
157 | */ | |
06386bbf | 158 | static struct mirror *bio_get_m(struct bio *bio) |
1da177e4 | 159 | { |
06386bbf | 160 | return (struct mirror *) bio->bi_next; |
1da177e4 LT |
161 | } |
162 | ||
06386bbf | 163 | static void bio_set_m(struct bio *bio, struct mirror *m) |
1da177e4 | 164 | { |
06386bbf | 165 | bio->bi_next = (struct bio *) m; |
1da177e4 LT |
166 | } |
167 | ||
72f4b314 JB |
168 | static struct mirror *get_default_mirror(struct mirror_set *ms) |
169 | { | |
170 | return &ms->mirror[atomic_read(&ms->default_mirror)]; | |
171 | } | |
172 | ||
173 | static void set_default_mirror(struct mirror *m) | |
174 | { | |
175 | struct mirror_set *ms = m->ms; | |
176 | struct mirror *m0 = &(ms->mirror[0]); | |
177 | ||
178 | atomic_set(&ms->default_mirror, m - m0); | |
179 | } | |
180 | ||
181 | /* fail_mirror | |
182 | * @m: mirror device to fail | |
183 | * @error_type: one of the enum's, DM_RAID1_*_ERROR | |
184 | * | |
185 | * If errors are being handled, record the type of | |
186 | * error encountered for this device. If this type | |
187 | * of error has already been recorded, we can return; | |
188 | * otherwise, we must signal userspace by triggering | |
189 | * an event. Additionally, if the device is the | |
190 | * primary device, we must choose a new primary, but | |
191 | * only if the mirror is in-sync. | |
192 | * | |
193 | * This function must not block. | |
194 | */ | |
195 | static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) | |
196 | { | |
197 | struct mirror_set *ms = m->ms; | |
198 | struct mirror *new; | |
199 | ||
72f4b314 JB |
200 | /* |
201 | * error_count is used for nothing more than a | |
202 | * simple way to tell if a device has encountered | |
203 | * errors. | |
204 | */ | |
205 | atomic_inc(&m->error_count); | |
206 | ||
207 | if (test_and_set_bit(error_type, &m->error_type)) | |
208 | return; | |
209 | ||
d460c65a JB |
210 | if (!errors_handled(ms)) |
211 | return; | |
212 | ||
72f4b314 JB |
213 | if (m != get_default_mirror(ms)) |
214 | goto out; | |
215 | ||
216 | if (!ms->in_sync) { | |
217 | /* | |
218 | * Better to issue requests to same failing device | |
219 | * than to risk returning corrupt data. | |
220 | */ | |
221 | DMERR("Primary mirror (%s) failed while out-of-sync: " | |
222 | "Reads may fail.", m->dev->name); | |
223 | goto out; | |
224 | } | |
225 | ||
226 | for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++) | |
227 | if (!atomic_read(&new->error_count)) { | |
228 | set_default_mirror(new); | |
229 | break; | |
230 | } | |
231 | ||
232 | if (unlikely(new == ms->mirror + ms->nr_mirrors)) | |
233 | DMWARN("All sides of mirror have failed."); | |
234 | ||
235 | out: | |
236 | schedule_work(&ms->trigger_event); | |
237 | } | |
238 | ||
1da177e4 LT |
239 | /*----------------------------------------------------------------- |
240 | * Recovery. | |
241 | * | |
242 | * When a mirror is first activated we may find that some regions | |
243 | * are in the no-sync state. We have to recover these by | |
244 | * recopying from the default mirror to all the others. | |
245 | *---------------------------------------------------------------*/ | |
4cdc1d1f | 246 | static void recovery_complete(int read_err, unsigned long write_err, |
1da177e4 LT |
247 | void *context) |
248 | { | |
1f965b19 HM |
249 | struct dm_region *reg = context; |
250 | struct mirror_set *ms = dm_rh_region_context(reg); | |
8f0205b7 | 251 | int m, bit = 0; |
1da177e4 | 252 | |
8f0205b7 | 253 | if (read_err) { |
f44db678 JB |
254 | /* Read error means the failure of default mirror. */ |
255 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
8f0205b7 JB |
256 | fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); |
257 | } | |
f44db678 | 258 | |
8f0205b7 | 259 | if (write_err) { |
4cdc1d1f | 260 | DMERR_LIMIT("Write error during recovery (error = 0x%lx)", |
f44db678 | 261 | write_err); |
8f0205b7 JB |
262 | /* |
263 | * Bits correspond to devices (excluding default mirror). | |
264 | * The default mirror cannot change during recovery. | |
265 | */ | |
266 | for (m = 0; m < ms->nr_mirrors; m++) { | |
267 | if (&ms->mirror[m] == get_default_mirror(ms)) | |
268 | continue; | |
269 | if (test_bit(bit, &write_err)) | |
270 | fail_mirror(ms->mirror + m, | |
271 | DM_RAID1_SYNC_ERROR); | |
272 | bit++; | |
273 | } | |
274 | } | |
f44db678 | 275 | |
1f965b19 | 276 | dm_rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
277 | } |
278 | ||
1f965b19 | 279 | static int recover(struct mirror_set *ms, struct dm_region *reg) |
1da177e4 LT |
280 | { |
281 | int r; | |
1f965b19 | 282 | unsigned i; |
eb69aca5 | 283 | struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; |
1da177e4 LT |
284 | struct mirror *m; |
285 | unsigned long flags = 0; | |
1f965b19 HM |
286 | region_t key = dm_rh_get_region_key(reg); |
287 | sector_t region_size = dm_rh_get_region_size(ms->rh); | |
1da177e4 LT |
288 | |
289 | /* fill in the source */ | |
72f4b314 | 290 | m = get_default_mirror(ms); |
1da177e4 | 291 | from.bdev = m->dev->bdev; |
1f965b19 HM |
292 | from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
293 | if (key == (ms->nr_regions - 1)) { | |
1da177e4 LT |
294 | /* |
295 | * The final region may be smaller than | |
296 | * region_size. | |
297 | */ | |
1f965b19 | 298 | from.count = ms->ti->len & (region_size - 1); |
1da177e4 | 299 | if (!from.count) |
1f965b19 | 300 | from.count = region_size; |
1da177e4 | 301 | } else |
1f965b19 | 302 | from.count = region_size; |
1da177e4 LT |
303 | |
304 | /* fill in the destinations */ | |
305 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
72f4b314 | 306 | if (&ms->mirror[i] == get_default_mirror(ms)) |
1da177e4 LT |
307 | continue; |
308 | ||
309 | m = ms->mirror + i; | |
310 | dest->bdev = m->dev->bdev; | |
1f965b19 | 311 | dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
1da177e4 LT |
312 | dest->count = from.count; |
313 | dest++; | |
314 | } | |
315 | ||
316 | /* hand to kcopyd */ | |
f7c83e2e JB |
317 | if (!errors_handled(ms)) |
318 | set_bit(DM_KCOPYD_IGNORE_ERROR, &flags); | |
319 | ||
eb69aca5 HM |
320 | r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, |
321 | flags, recovery_complete, reg); | |
1da177e4 LT |
322 | |
323 | return r; | |
324 | } | |
325 | ||
326 | static void do_recovery(struct mirror_set *ms) | |
327 | { | |
1f965b19 HM |
328 | struct dm_region *reg; |
329 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
1da177e4 | 330 | int r; |
1da177e4 LT |
331 | |
332 | /* | |
333 | * Start quiescing some regions. | |
334 | */ | |
1f965b19 | 335 | dm_rh_recovery_prepare(ms->rh); |
1da177e4 LT |
336 | |
337 | /* | |
338 | * Copy any already quiesced regions. | |
339 | */ | |
1f965b19 | 340 | while ((reg = dm_rh_recovery_start(ms->rh))) { |
1da177e4 LT |
341 | r = recover(ms, reg); |
342 | if (r) | |
1f965b19 | 343 | dm_rh_recovery_end(reg, 0); |
1da177e4 LT |
344 | } |
345 | ||
346 | /* | |
347 | * Update the in sync flag. | |
348 | */ | |
349 | if (!ms->in_sync && | |
350 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
351 | /* the sync is complete */ | |
352 | dm_table_event(ms->ti->table); | |
353 | ms->in_sync = 1; | |
354 | } | |
355 | } | |
356 | ||
357 | /*----------------------------------------------------------------- | |
358 | * Reads | |
359 | *---------------------------------------------------------------*/ | |
360 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
361 | { | |
06386bbf JB |
362 | struct mirror *m = get_default_mirror(ms); |
363 | ||
364 | do { | |
365 | if (likely(!atomic_read(&m->error_count))) | |
366 | return m; | |
367 | ||
368 | if (m-- == ms->mirror) | |
369 | m += ms->nr_mirrors; | |
370 | } while (m != get_default_mirror(ms)); | |
371 | ||
372 | return NULL; | |
373 | } | |
374 | ||
375 | static int default_ok(struct mirror *m) | |
376 | { | |
377 | struct mirror *default_mirror = get_default_mirror(m->ms); | |
378 | ||
379 | return !atomic_read(&default_mirror->error_count); | |
380 | } | |
381 | ||
382 | static int mirror_available(struct mirror_set *ms, struct bio *bio) | |
383 | { | |
1f965b19 HM |
384 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
385 | region_t region = dm_rh_bio_to_region(ms->rh, bio); | |
06386bbf | 386 | |
1f965b19 | 387 | if (log->type->in_sync(log, region, 0)) |
06386bbf JB |
388 | return choose_mirror(ms, bio->bi_sector) ? 1 : 0; |
389 | ||
390 | return 0; | |
1da177e4 LT |
391 | } |
392 | ||
393 | /* | |
394 | * remap a buffer to a particular mirror. | |
395 | */ | |
06386bbf JB |
396 | static sector_t map_sector(struct mirror *m, struct bio *bio) |
397 | { | |
398 | return m->offset + (bio->bi_sector - m->ms->ti->begin); | |
399 | } | |
400 | ||
401 | static void map_bio(struct mirror *m, struct bio *bio) | |
1da177e4 LT |
402 | { |
403 | bio->bi_bdev = m->dev->bdev; | |
06386bbf JB |
404 | bio->bi_sector = map_sector(m, bio); |
405 | } | |
406 | ||
22a1ceb1 | 407 | static void map_region(struct dm_io_region *io, struct mirror *m, |
06386bbf JB |
408 | struct bio *bio) |
409 | { | |
410 | io->bdev = m->dev->bdev; | |
411 | io->sector = map_sector(m, bio); | |
412 | io->count = bio->bi_size >> 9; | |
413 | } | |
414 | ||
415 | /*----------------------------------------------------------------- | |
416 | * Reads | |
417 | *---------------------------------------------------------------*/ | |
418 | static void read_callback(unsigned long error, void *context) | |
419 | { | |
420 | struct bio *bio = context; | |
421 | struct mirror *m; | |
422 | ||
423 | m = bio_get_m(bio); | |
424 | bio_set_m(bio, NULL); | |
425 | ||
426 | if (likely(!error)) { | |
427 | bio_endio(bio, 0); | |
428 | return; | |
429 | } | |
430 | ||
431 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
432 | ||
433 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
434 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
435 | "Trying alternative device.", | |
436 | m->dev->name); | |
437 | queue_bio(m->ms, bio, bio_rw(bio)); | |
438 | return; | |
439 | } | |
440 | ||
441 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
442 | m->dev->name); | |
443 | bio_endio(bio, -EIO); | |
444 | } | |
445 | ||
446 | /* Asynchronous read. */ | |
447 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
448 | { | |
22a1ceb1 | 449 | struct dm_io_region io; |
06386bbf JB |
450 | struct dm_io_request io_req = { |
451 | .bi_rw = READ, | |
452 | .mem.type = DM_IO_BVEC, | |
453 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
454 | .notify.fn = read_callback, | |
455 | .notify.context = bio, | |
456 | .client = m->ms->io_client, | |
457 | }; | |
458 | ||
459 | map_region(&io, m, bio); | |
460 | bio_set_m(bio, m); | |
1f965b19 HM |
461 | BUG_ON(dm_io(&io_req, 1, &io, NULL)); |
462 | } | |
463 | ||
464 | static inline int region_in_sync(struct mirror_set *ms, region_t region, | |
465 | int may_block) | |
466 | { | |
467 | int state = dm_rh_get_state(ms->rh, region, may_block); | |
468 | return state == DM_RH_CLEAN || state == DM_RH_DIRTY; | |
1da177e4 LT |
469 | } |
470 | ||
471 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
472 | { | |
473 | region_t region; | |
474 | struct bio *bio; | |
475 | struct mirror *m; | |
476 | ||
477 | while ((bio = bio_list_pop(reads))) { | |
1f965b19 | 478 | region = dm_rh_bio_to_region(ms->rh, bio); |
06386bbf | 479 | m = get_default_mirror(ms); |
1da177e4 LT |
480 | |
481 | /* | |
482 | * We can only read balance if the region is in sync. | |
483 | */ | |
1f965b19 | 484 | if (likely(region_in_sync(ms, region, 1))) |
1da177e4 | 485 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
486 | else if (m && atomic_read(&m->error_count)) |
487 | m = NULL; | |
1da177e4 | 488 | |
06386bbf JB |
489 | if (likely(m)) |
490 | read_async_bio(m, bio); | |
491 | else | |
492 | bio_endio(bio, -EIO); | |
1da177e4 LT |
493 | } |
494 | } | |
495 | ||
496 | /*----------------------------------------------------------------- | |
497 | * Writes. | |
498 | * | |
499 | * We do different things with the write io depending on the | |
500 | * state of the region that it's in: | |
501 | * | |
502 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
503 | * RECOVERING: delay the io until recovery completes | |
504 | * NOSYNC: increment pending, just write to the default mirror | |
505 | *---------------------------------------------------------------*/ | |
72f4b314 | 506 | |
72f4b314 | 507 | |
1da177e4 LT |
508 | static void write_callback(unsigned long error, void *context) |
509 | { | |
72f4b314 | 510 | unsigned i, ret = 0; |
1da177e4 LT |
511 | struct bio *bio = (struct bio *) context; |
512 | struct mirror_set *ms; | |
72f4b314 JB |
513 | int uptodate = 0; |
514 | int should_wake = 0; | |
515 | unsigned long flags; | |
1da177e4 | 516 | |
06386bbf JB |
517 | ms = bio_get_m(bio)->ms; |
518 | bio_set_m(bio, NULL); | |
1da177e4 LT |
519 | |
520 | /* | |
521 | * NOTE: We don't decrement the pending count here, | |
522 | * instead it is done by the targets endio function. | |
523 | * This way we handle both writes to SYNC and NOSYNC | |
524 | * regions with the same code. | |
525 | */ | |
72f4b314 JB |
526 | if (likely(!error)) |
527 | goto out; | |
1da177e4 | 528 | |
72f4b314 JB |
529 | for (i = 0; i < ms->nr_mirrors; i++) |
530 | if (test_bit(i, &error)) | |
531 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
532 | else | |
533 | uptodate = 1; | |
534 | ||
535 | if (unlikely(!uptodate)) { | |
536 | DMERR("All replicated volumes dead, failing I/O"); | |
537 | /* None of the writes succeeded, fail the I/O. */ | |
538 | ret = -EIO; | |
539 | } else if (errors_handled(ms)) { | |
1da177e4 | 540 | /* |
72f4b314 JB |
541 | * Need to raise event. Since raising |
542 | * events can block, we need to do it in | |
543 | * the main thread. | |
1da177e4 | 544 | */ |
72f4b314 JB |
545 | spin_lock_irqsave(&ms->lock, flags); |
546 | if (!ms->failures.head) | |
547 | should_wake = 1; | |
548 | bio_list_add(&ms->failures, bio); | |
549 | spin_unlock_irqrestore(&ms->lock, flags); | |
550 | if (should_wake) | |
1f965b19 | 551 | wakeup_mirrord(ms); |
72f4b314 | 552 | return; |
1da177e4 | 553 | } |
72f4b314 JB |
554 | out: |
555 | bio_endio(bio, ret); | |
1da177e4 LT |
556 | } |
557 | ||
558 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
559 | { | |
560 | unsigned int i; | |
22a1ceb1 | 561 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 562 | struct mirror *m; |
88be163a MB |
563 | struct dm_io_request io_req = { |
564 | .bi_rw = WRITE, | |
565 | .mem.type = DM_IO_BVEC, | |
566 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
567 | .notify.fn = write_callback, | |
568 | .notify.context = bio, | |
569 | .client = ms->io_client, | |
570 | }; | |
1da177e4 | 571 | |
06386bbf JB |
572 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
573 | map_region(dest++, m, bio); | |
1da177e4 | 574 | |
06386bbf JB |
575 | /* |
576 | * Use default mirror because we only need it to retrieve the reference | |
577 | * to the mirror set in write_callback(). | |
578 | */ | |
579 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a | 580 | |
1f965b19 | 581 | BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL)); |
1da177e4 LT |
582 | } |
583 | ||
584 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
585 | { | |
586 | int state; | |
587 | struct bio *bio; | |
588 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
589 | ||
590 | if (!writes->head) | |
591 | return; | |
592 | ||
593 | /* | |
594 | * Classify each write. | |
595 | */ | |
596 | bio_list_init(&sync); | |
597 | bio_list_init(&nosync); | |
598 | bio_list_init(&recover); | |
599 | ||
600 | while ((bio = bio_list_pop(writes))) { | |
1f965b19 HM |
601 | state = dm_rh_get_state(ms->rh, |
602 | dm_rh_bio_to_region(ms->rh, bio), 1); | |
1da177e4 | 603 | switch (state) { |
1f965b19 HM |
604 | case DM_RH_CLEAN: |
605 | case DM_RH_DIRTY: | |
1da177e4 LT |
606 | this_list = &sync; |
607 | break; | |
608 | ||
1f965b19 | 609 | case DM_RH_NOSYNC: |
1da177e4 LT |
610 | this_list = &nosync; |
611 | break; | |
612 | ||
1f965b19 | 613 | case DM_RH_RECOVERING: |
1da177e4 LT |
614 | this_list = &recover; |
615 | break; | |
616 | } | |
617 | ||
618 | bio_list_add(this_list, bio); | |
619 | } | |
620 | ||
621 | /* | |
622 | * Increment the pending counts for any regions that will | |
623 | * be written to (writes to recover regions are going to | |
624 | * be delayed). | |
625 | */ | |
1f965b19 HM |
626 | dm_rh_inc_pending(ms->rh, &sync); |
627 | dm_rh_inc_pending(ms->rh, &nosync); | |
628 | ms->log_failure = dm_rh_flush(ms->rh) ? 1 : 0; | |
1da177e4 LT |
629 | |
630 | /* | |
631 | * Dispatch io. | |
632 | */ | |
b80aa7a0 JB |
633 | if (unlikely(ms->log_failure)) { |
634 | spin_lock_irq(&ms->lock); | |
635 | bio_list_merge(&ms->failures, &sync); | |
636 | spin_unlock_irq(&ms->lock); | |
1f965b19 | 637 | wakeup_mirrord(ms); |
b80aa7a0 | 638 | } else |
fc1ff958 | 639 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 640 | do_write(ms, bio); |
1da177e4 LT |
641 | |
642 | while ((bio = bio_list_pop(&recover))) | |
1f965b19 | 643 | dm_rh_delay(ms->rh, bio); |
1da177e4 LT |
644 | |
645 | while ((bio = bio_list_pop(&nosync))) { | |
06386bbf | 646 | map_bio(get_default_mirror(ms), bio); |
1da177e4 LT |
647 | generic_make_request(bio); |
648 | } | |
649 | } | |
650 | ||
72f4b314 JB |
651 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
652 | { | |
653 | struct bio *bio; | |
654 | ||
655 | if (!failures->head) | |
656 | return; | |
657 | ||
b80aa7a0 | 658 | if (!ms->log_failure) { |
b34578a4 | 659 | while ((bio = bio_list_pop(failures))) { |
1f965b19 HM |
660 | ms->in_sync = 0; |
661 | dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0); | |
b34578a4 | 662 | } |
b80aa7a0 JB |
663 | return; |
664 | } | |
665 | ||
666 | /* | |
667 | * If the log has failed, unattempted writes are being | |
668 | * put on the failures list. We can't issue those writes | |
669 | * until a log has been marked, so we must store them. | |
670 | * | |
671 | * If a 'noflush' suspend is in progress, we can requeue | |
672 | * the I/O's to the core. This give userspace a chance | |
673 | * to reconfigure the mirror, at which point the core | |
674 | * will reissue the writes. If the 'noflush' flag is | |
675 | * not set, we have no choice but to return errors. | |
676 | * | |
677 | * Some writes on the failures list may have been | |
678 | * submitted before the log failure and represent a | |
679 | * failure to write to one of the devices. It is ok | |
680 | * for us to treat them the same and requeue them | |
681 | * as well. | |
682 | */ | |
683 | if (dm_noflush_suspending(ms->ti)) { | |
684 | while ((bio = bio_list_pop(failures))) | |
685 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
686 | return; | |
687 | } | |
688 | ||
689 | if (atomic_read(&ms->suspend)) { | |
690 | while ((bio = bio_list_pop(failures))) | |
691 | bio_endio(bio, -EIO); | |
692 | return; | |
693 | } | |
694 | ||
695 | spin_lock_irq(&ms->lock); | |
696 | bio_list_merge(&ms->failures, failures); | |
697 | spin_unlock_irq(&ms->lock); | |
698 | ||
a2aebe03 | 699 | delayed_wake(ms); |
72f4b314 JB |
700 | } |
701 | ||
702 | static void trigger_event(struct work_struct *work) | |
703 | { | |
704 | struct mirror_set *ms = | |
705 | container_of(work, struct mirror_set, trigger_event); | |
706 | ||
707 | dm_table_event(ms->ti->table); | |
708 | } | |
709 | ||
1da177e4 LT |
710 | /*----------------------------------------------------------------- |
711 | * kmirrord | |
712 | *---------------------------------------------------------------*/ | |
a2aebe03 | 713 | static void do_mirror(struct work_struct *work) |
1da177e4 | 714 | { |
1f965b19 HM |
715 | struct mirror_set *ms = container_of(work, struct mirror_set, |
716 | kmirrord_work); | |
72f4b314 JB |
717 | struct bio_list reads, writes, failures; |
718 | unsigned long flags; | |
1da177e4 | 719 | |
72f4b314 | 720 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
721 | reads = ms->reads; |
722 | writes = ms->writes; | |
72f4b314 | 723 | failures = ms->failures; |
1da177e4 LT |
724 | bio_list_init(&ms->reads); |
725 | bio_list_init(&ms->writes); | |
72f4b314 JB |
726 | bio_list_init(&ms->failures); |
727 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 | 728 | |
1f965b19 | 729 | dm_rh_update_states(ms->rh, errors_handled(ms)); |
1da177e4 LT |
730 | do_recovery(ms); |
731 | do_reads(ms, &reads); | |
732 | do_writes(ms, &writes); | |
72f4b314 | 733 | do_failures(ms, &failures); |
7ff14a36 MP |
734 | |
735 | dm_table_unplug_all(ms->ti->table); | |
1da177e4 LT |
736 | } |
737 | ||
1da177e4 LT |
738 | /*----------------------------------------------------------------- |
739 | * Target functions | |
740 | *---------------------------------------------------------------*/ | |
741 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
742 | uint32_t region_size, | |
743 | struct dm_target *ti, | |
416cd17b | 744 | struct dm_dirty_log *dl) |
1da177e4 LT |
745 | { |
746 | size_t len; | |
747 | struct mirror_set *ms = NULL; | |
748 | ||
1da177e4 LT |
749 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
750 | ||
dd00cc48 | 751 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 752 | if (!ms) { |
72d94861 | 753 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
754 | return NULL; |
755 | } | |
756 | ||
1da177e4 LT |
757 | spin_lock_init(&ms->lock); |
758 | ||
759 | ms->ti = ti; | |
760 | ms->nr_mirrors = nr_mirrors; | |
761 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
762 | ms->in_sync = 0; | |
b80aa7a0 JB |
763 | ms->log_failure = 0; |
764 | atomic_set(&ms->suspend, 0); | |
72f4b314 | 765 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 766 | |
06386bbf JB |
767 | len = sizeof(struct dm_raid1_read_record); |
768 | ms->read_record_pool = mempool_create_kmalloc_pool(MIN_READ_RECORDS, | |
769 | len); | |
770 | if (!ms->read_record_pool) { | |
771 | ti->error = "Error creating mirror read_record_pool"; | |
772 | kfree(ms); | |
773 | return NULL; | |
774 | } | |
775 | ||
88be163a MB |
776 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
777 | if (IS_ERR(ms->io_client)) { | |
778 | ti->error = "Error creating dm_io client"; | |
06386bbf | 779 | mempool_destroy(ms->read_record_pool); |
88be163a MB |
780 | kfree(ms); |
781 | return NULL; | |
782 | } | |
783 | ||
1f965b19 HM |
784 | ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord, |
785 | wakeup_all_recovery_waiters, | |
786 | ms->ti->begin, MAX_RECOVERY, | |
787 | dl, region_size, ms->nr_regions); | |
788 | if (IS_ERR(ms->rh)) { | |
72d94861 | 789 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 790 | dm_io_client_destroy(ms->io_client); |
06386bbf | 791 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
792 | kfree(ms); |
793 | return NULL; | |
794 | } | |
795 | ||
796 | return ms; | |
797 | } | |
798 | ||
799 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
800 | unsigned int m) | |
801 | { | |
802 | while (m--) | |
803 | dm_put_device(ti, ms->mirror[m].dev); | |
804 | ||
88be163a | 805 | dm_io_client_destroy(ms->io_client); |
1f965b19 | 806 | dm_region_hash_destroy(ms->rh); |
06386bbf | 807 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
808 | kfree(ms); |
809 | } | |
810 | ||
1da177e4 LT |
811 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
812 | unsigned int mirror, char **argv) | |
813 | { | |
4ee218cd | 814 | unsigned long long offset; |
1da177e4 | 815 | |
4ee218cd | 816 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 817 | ti->error = "Invalid offset"; |
1da177e4 LT |
818 | return -EINVAL; |
819 | } | |
820 | ||
821 | if (dm_get_device(ti, argv[0], offset, ti->len, | |
822 | dm_table_get_mode(ti->table), | |
823 | &ms->mirror[mirror].dev)) { | |
72d94861 | 824 | ti->error = "Device lookup failure"; |
1da177e4 LT |
825 | return -ENXIO; |
826 | } | |
827 | ||
aa5617c5 | 828 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
829 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
830 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
831 | ms->mirror[mirror].offset = offset; |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
1da177e4 LT |
836 | /* |
837 | * Create dirty log: log_type #log_params <log_params> | |
838 | */ | |
416cd17b | 839 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1f965b19 HM |
840 | unsigned argc, char **argv, |
841 | unsigned *args_used) | |
1da177e4 | 842 | { |
1f965b19 | 843 | unsigned param_count; |
416cd17b | 844 | struct dm_dirty_log *dl; |
1da177e4 LT |
845 | |
846 | if (argc < 2) { | |
72d94861 | 847 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
848 | return NULL; |
849 | } | |
850 | ||
851 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 852 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
853 | return NULL; |
854 | } | |
855 | ||
856 | *args_used = 2 + param_count; | |
857 | ||
858 | if (argc < *args_used) { | |
72d94861 | 859 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
860 | return NULL; |
861 | } | |
862 | ||
416cd17b | 863 | dl = dm_dirty_log_create(argv[0], ti, param_count, argv + 2); |
1da177e4 | 864 | if (!dl) { |
72d94861 | 865 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
866 | return NULL; |
867 | } | |
868 | ||
1da177e4 LT |
869 | return dl; |
870 | } | |
871 | ||
a8e6afa2 JB |
872 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
873 | unsigned *args_used) | |
874 | { | |
875 | unsigned num_features; | |
876 | struct dm_target *ti = ms->ti; | |
877 | ||
878 | *args_used = 0; | |
879 | ||
880 | if (!argc) | |
881 | return 0; | |
882 | ||
883 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
884 | ti->error = "Invalid number of features"; | |
885 | return -EINVAL; | |
886 | } | |
887 | ||
888 | argc--; | |
889 | argv++; | |
890 | (*args_used)++; | |
891 | ||
892 | if (num_features > argc) { | |
893 | ti->error = "Not enough arguments to support feature count"; | |
894 | return -EINVAL; | |
895 | } | |
896 | ||
897 | if (!strcmp("handle_errors", argv[0])) | |
898 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
899 | else { | |
900 | ti->error = "Unrecognised feature requested"; | |
901 | return -EINVAL; | |
902 | } | |
903 | ||
904 | (*args_used)++; | |
905 | ||
906 | return 0; | |
907 | } | |
908 | ||
1da177e4 LT |
909 | /* |
910 | * Construct a mirror mapping: | |
911 | * | |
912 | * log_type #log_params <log_params> | |
913 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 914 | * [#features <features>] |
1da177e4 LT |
915 | * |
916 | * log_type is "core" or "disk" | |
917 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
918 | * |
919 | * If present, features must be "handle_errors". | |
1da177e4 | 920 | */ |
1da177e4 LT |
921 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
922 | { | |
923 | int r; | |
924 | unsigned int nr_mirrors, m, args_used; | |
925 | struct mirror_set *ms; | |
416cd17b | 926 | struct dm_dirty_log *dl; |
1da177e4 LT |
927 | |
928 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
929 | if (!dl) | |
930 | return -EINVAL; | |
931 | ||
932 | argv += args_used; | |
933 | argc -= args_used; | |
934 | ||
935 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
eb69aca5 | 936 | nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) { |
72d94861 | 937 | ti->error = "Invalid number of mirrors"; |
416cd17b | 938 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
939 | return -EINVAL; |
940 | } | |
941 | ||
942 | argv++, argc--; | |
943 | ||
a8e6afa2 JB |
944 | if (argc < nr_mirrors * 2) { |
945 | ti->error = "Too few mirror arguments"; | |
416cd17b | 946 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
947 | return -EINVAL; |
948 | } | |
949 | ||
950 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
951 | if (!ms) { | |
416cd17b | 952 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
953 | return -ENOMEM; |
954 | } | |
955 | ||
956 | /* Get the mirror parameter sets */ | |
957 | for (m = 0; m < nr_mirrors; m++) { | |
958 | r = get_mirror(ms, ti, m, argv); | |
959 | if (r) { | |
960 | free_context(ms, ti, m); | |
961 | return r; | |
962 | } | |
963 | argv += 2; | |
964 | argc -= 2; | |
965 | } | |
966 | ||
967 | ti->private = ms; | |
1f965b19 | 968 | ti->split_io = dm_rh_get_region_size(ms->rh); |
1da177e4 | 969 | |
6ad36fe2 HS |
970 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
971 | if (!ms->kmirrord_wq) { | |
972 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
973 | r = -ENOMEM; |
974 | goto err_free_context; | |
6ad36fe2 HS |
975 | } |
976 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
977 | init_timer(&ms->timer); |
978 | ms->timer_pending = 0; | |
72f4b314 | 979 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 980 | |
a8e6afa2 | 981 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
982 | if (r) |
983 | goto err_destroy_wq; | |
a8e6afa2 JB |
984 | |
985 | argv += args_used; | |
986 | argc -= args_used; | |
987 | ||
f44db678 JB |
988 | /* |
989 | * Any read-balancing addition depends on the | |
990 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
991 | * This is because the decision to balance depends | |
992 | * on the sync state of a region. If the above | |
993 | * flag is not present, we ignore errors; and | |
994 | * the sync state may be inaccurate. | |
995 | */ | |
996 | ||
a8e6afa2 JB |
997 | if (argc) { |
998 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
999 | r = -EINVAL; |
1000 | goto err_destroy_wq; | |
a8e6afa2 JB |
1001 | } |
1002 | ||
1f965b19 | 1003 | r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client); |
a72cf737 DM |
1004 | if (r) |
1005 | goto err_destroy_wq; | |
1da177e4 | 1006 | |
1f965b19 | 1007 | wakeup_mirrord(ms); |
1da177e4 | 1008 | return 0; |
a72cf737 DM |
1009 | |
1010 | err_destroy_wq: | |
1011 | destroy_workqueue(ms->kmirrord_wq); | |
1012 | err_free_context: | |
1013 | free_context(ms, ti, ms->nr_mirrors); | |
1014 | return r; | |
1da177e4 LT |
1015 | } |
1016 | ||
1017 | static void mirror_dtr(struct dm_target *ti) | |
1018 | { | |
1019 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1020 | ||
a2aebe03 | 1021 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1022 | flush_workqueue(ms->kmirrord_wq); |
18776c73 | 1023 | flush_scheduled_work(); |
eb69aca5 | 1024 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1025 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1026 | free_context(ms, ti, ms->nr_mirrors); |
1027 | } | |
1028 | ||
1da177e4 LT |
1029 | /* |
1030 | * Mirror mapping function | |
1031 | */ | |
1032 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1033 | union map_info *map_context) | |
1034 | { | |
1035 | int r, rw = bio_rw(bio); | |
1036 | struct mirror *m; | |
1037 | struct mirror_set *ms = ti->private; | |
06386bbf | 1038 | struct dm_raid1_read_record *read_record = NULL; |
1f965b19 | 1039 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 LT |
1040 | |
1041 | if (rw == WRITE) { | |
06386bbf | 1042 | /* Save region for mirror_end_io() handler */ |
1f965b19 | 1043 | map_context->ll = dm_rh_bio_to_region(ms->rh, bio); |
1da177e4 | 1044 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1045 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1046 | } |
1047 | ||
1f965b19 | 1048 | r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0); |
1da177e4 LT |
1049 | if (r < 0 && r != -EWOULDBLOCK) |
1050 | return r; | |
1051 | ||
1da177e4 | 1052 | /* |
06386bbf | 1053 | * If region is not in-sync queue the bio. |
1da177e4 | 1054 | */ |
06386bbf JB |
1055 | if (!r || (r == -EWOULDBLOCK)) { |
1056 | if (rw == READA) | |
1057 | return -EWOULDBLOCK; | |
1da177e4 | 1058 | |
1da177e4 | 1059 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1060 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1061 | } |
1062 | ||
06386bbf JB |
1063 | /* |
1064 | * The region is in-sync and we can perform reads directly. | |
1065 | * Store enough information so we can retry if it fails. | |
1066 | */ | |
1da177e4 | 1067 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1068 | if (unlikely(!m)) |
1da177e4 LT |
1069 | return -EIO; |
1070 | ||
06386bbf JB |
1071 | read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO); |
1072 | if (likely(read_record)) { | |
1073 | dm_bio_record(&read_record->details, bio); | |
1074 | map_context->ptr = read_record; | |
1075 | read_record->m = m; | |
1076 | } | |
1077 | ||
1078 | map_bio(m, bio); | |
1079 | ||
d2a7ad29 | 1080 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1081 | } |
1082 | ||
1083 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1084 | int error, union map_info *map_context) | |
1085 | { | |
1086 | int rw = bio_rw(bio); | |
1087 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1088 | struct mirror *m = NULL; |
1089 | struct dm_bio_details *bd = NULL; | |
1090 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * We need to dec pending if this was a write. | |
1094 | */ | |
06386bbf | 1095 | if (rw == WRITE) { |
1f965b19 | 1096 | dm_rh_dec(ms->rh, map_context->ll); |
06386bbf JB |
1097 | return error; |
1098 | } | |
1da177e4 | 1099 | |
06386bbf JB |
1100 | if (error == -EOPNOTSUPP) |
1101 | goto out; | |
1102 | ||
1103 | if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio)) | |
1104 | goto out; | |
1105 | ||
1106 | if (unlikely(error)) { | |
1107 | if (!read_record) { | |
1108 | /* | |
1109 | * There wasn't enough memory to record necessary | |
1110 | * information for a retry or there was no other | |
1111 | * mirror in-sync. | |
1112 | */ | |
e03f1a84 | 1113 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1114 | return -EIO; |
1115 | } | |
e03f1a84 AB |
1116 | |
1117 | m = read_record->m; | |
1118 | ||
06386bbf JB |
1119 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1120 | m->dev->name); | |
1121 | ||
06386bbf JB |
1122 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1123 | ||
1124 | /* | |
1125 | * A failed read is requeued for another attempt using an intact | |
1126 | * mirror. | |
1127 | */ | |
1128 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1129 | bd = &read_record->details; | |
1130 | ||
1131 | dm_bio_restore(bd, bio); | |
1132 | mempool_free(read_record, ms->read_record_pool); | |
1133 | map_context->ptr = NULL; | |
1134 | queue_bio(ms, bio, rw); | |
1135 | return 1; | |
1136 | } | |
1137 | DMERR("All replicated volumes dead, failing I/O"); | |
1138 | } | |
1139 | ||
1140 | out: | |
1141 | if (read_record) { | |
1142 | mempool_free(read_record, ms->read_record_pool); | |
1143 | map_context->ptr = NULL; | |
1144 | } | |
1145 | ||
1146 | return error; | |
1da177e4 LT |
1147 | } |
1148 | ||
b80aa7a0 | 1149 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1150 | { |
1151 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1f965b19 | 1152 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 | 1153 | |
b80aa7a0 JB |
1154 | atomic_set(&ms->suspend, 1); |
1155 | ||
1156 | /* | |
1157 | * We must finish up all the work that we've | |
1158 | * generated (i.e. recovery work). | |
1159 | */ | |
1f965b19 | 1160 | dm_rh_stop_recovery(ms->rh); |
33184048 | 1161 | |
33184048 | 1162 | wait_event(_kmirrord_recovery_stopped, |
1f965b19 | 1163 | !dm_rh_recovery_in_flight(ms->rh)); |
33184048 | 1164 | |
b80aa7a0 JB |
1165 | if (log->type->presuspend && log->type->presuspend(log)) |
1166 | /* FIXME: need better error handling */ | |
1167 | DMWARN("log presuspend failed"); | |
1168 | ||
1169 | /* | |
1170 | * Now that recovery is complete/stopped and the | |
1171 | * delayed bios are queued, we need to wait for | |
1172 | * the worker thread to complete. This way, | |
1173 | * we know that all of our I/O has been pushed. | |
1174 | */ | |
1175 | flush_workqueue(ms->kmirrord_wq); | |
1176 | } | |
1177 | ||
1178 | static void mirror_postsuspend(struct dm_target *ti) | |
1179 | { | |
1180 | struct mirror_set *ms = ti->private; | |
1f965b19 | 1181 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 | 1182 | |
6b3df0d7 | 1183 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1184 | /* FIXME: need better error handling */ |
b80aa7a0 | 1185 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | static void mirror_resume(struct dm_target *ti) | |
1189 | { | |
b80aa7a0 | 1190 | struct mirror_set *ms = ti->private; |
1f965b19 | 1191 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 JB |
1192 | |
1193 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1194 | if (log->type->resume && log->type->resume(log)) |
1195 | /* FIXME: need better error handling */ | |
1196 | DMWARN("log resume failed"); | |
1f965b19 | 1197 | dm_rh_start_recovery(ms->rh); |
1da177e4 LT |
1198 | } |
1199 | ||
af195ac8 JB |
1200 | /* |
1201 | * device_status_char | |
1202 | * @m: mirror device/leg we want the status of | |
1203 | * | |
1204 | * We return one character representing the most severe error | |
1205 | * we have encountered. | |
1206 | * A => Alive - No failures | |
1207 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1208 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1209 | * R => Read - A read failure occurred, mirror data unaffected | |
1210 | * | |
1211 | * Returns: <char> | |
1212 | */ | |
1213 | static char device_status_char(struct mirror *m) | |
1214 | { | |
1215 | if (!atomic_read(&(m->error_count))) | |
1216 | return 'A'; | |
1217 | ||
1218 | return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
1219 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : | |
1220 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1221 | } | |
1222 | ||
1223 | ||
1da177e4 LT |
1224 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1225 | char *result, unsigned int maxlen) | |
1226 | { | |
315dcc22 | 1227 | unsigned int m, sz = 0; |
1da177e4 | 1228 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1f965b19 | 1229 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
af195ac8 | 1230 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1231 | |
1da177e4 LT |
1232 | switch (type) { |
1233 | case STATUSTYPE_INFO: | |
1234 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1235 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1236 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1237 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1238 | } | |
1239 | buffer[m] = '\0'; | |
1da177e4 | 1240 | |
af195ac8 | 1241 | DMEMIT("%llu/%llu 1 %s ", |
1f965b19 | 1242 | (unsigned long long)log->type->get_sync_count(log), |
af195ac8 | 1243 | (unsigned long long)ms->nr_regions, buffer); |
315dcc22 | 1244 | |
1f965b19 | 1245 | sz += log->type->status(log, type, result+sz, maxlen-sz); |
315dcc22 | 1246 | |
1da177e4 LT |
1247 | break; |
1248 | ||
1249 | case STATUSTYPE_TABLE: | |
1f965b19 | 1250 | sz = log->type->status(log, type, result, maxlen); |
315dcc22 | 1251 | |
e52b8f6d | 1252 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1253 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1254 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1255 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1256 | |
1257 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1258 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1259 | } |
1260 | ||
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | static struct target_type mirror_target = { | |
1265 | .name = "mirror", | |
af195ac8 | 1266 | .version = {1, 0, 20}, |
1da177e4 LT |
1267 | .module = THIS_MODULE, |
1268 | .ctr = mirror_ctr, | |
1269 | .dtr = mirror_dtr, | |
1270 | .map = mirror_map, | |
1271 | .end_io = mirror_end_io, | |
b80aa7a0 | 1272 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1273 | .postsuspend = mirror_postsuspend, |
1274 | .resume = mirror_resume, | |
1275 | .status = mirror_status, | |
1276 | }; | |
1277 | ||
1278 | static int __init dm_mirror_init(void) | |
1279 | { | |
1280 | int r; | |
1281 | ||
1da177e4 | 1282 | r = dm_register_target(&mirror_target); |
769aef30 | 1283 | if (r < 0) |
0cd33124 | 1284 | DMERR("Failed to register mirror target"); |
1da177e4 LT |
1285 | |
1286 | return r; | |
1287 | } | |
1288 | ||
1289 | static void __exit dm_mirror_exit(void) | |
1290 | { | |
10d3bd09 | 1291 | dm_unregister_target(&mirror_target); |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | /* Module hooks */ | |
1295 | module_init(dm_mirror_init); | |
1296 | module_exit(dm_mirror_exit); | |
1297 | ||
1298 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1299 | MODULE_AUTHOR("Joe Thornber"); | |
1300 | MODULE_LICENSE("GPL"); |