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 | ||
06386bbf | 8 | #include "dm-bio-record.h" |
1da177e4 | 9 | |
1da177e4 LT |
10 | #include <linux/init.h> |
11 | #include <linux/mempool.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/slab.h> | |
1da177e4 | 15 | #include <linux/workqueue.h> |
1f965b19 | 16 | #include <linux/device-mapper.h> |
a765e20e AK |
17 | #include <linux/dm-io.h> |
18 | #include <linux/dm-dirty-log.h> | |
19 | #include <linux/dm-kcopyd.h> | |
1f965b19 | 20 | #include <linux/dm-region-hash.h> |
1da177e4 | 21 | |
72d94861 | 22 | #define DM_MSG_PREFIX "raid1" |
1f965b19 HM |
23 | |
24 | #define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */ | |
88be163a | 25 | #define DM_IO_PAGES 64 |
1f965b19 | 26 | #define DM_KCOPYD_PAGES 64 |
72d94861 | 27 | |
a8e6afa2 | 28 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
f44db678 | 29 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
a8e6afa2 | 30 | |
33184048 | 31 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
1da177e4 | 32 | |
e4c8b3ba NB |
33 | /*----------------------------------------------------------------- |
34 | * Mirror set structures. | |
35 | *---------------------------------------------------------------*/ | |
72f4b314 JB |
36 | enum dm_raid1_error { |
37 | DM_RAID1_WRITE_ERROR, | |
64b30c46 | 38 | DM_RAID1_FLUSH_ERROR, |
72f4b314 JB |
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; |
04788507 | 61 | struct bio_list holds; /* bios are waiting until suspend */ |
e4c8b3ba | 62 | |
1f965b19 HM |
63 | struct dm_region_hash *rh; |
64 | struct dm_kcopyd_client *kcopyd_client; | |
88be163a | 65 | struct dm_io_client *io_client; |
06386bbf | 66 | mempool_t *read_record_pool; |
88be163a | 67 | |
e4c8b3ba NB |
68 | /* recovery */ |
69 | region_t nr_regions; | |
70 | int in_sync; | |
fc1ff958 | 71 | int log_failure; |
929be8fc | 72 | int leg_failure; |
b80aa7a0 | 73 | atomic_t suspend; |
e4c8b3ba | 74 | |
72f4b314 | 75 | atomic_t default_mirror; /* Default mirror */ |
e4c8b3ba | 76 | |
6ad36fe2 HS |
77 | struct workqueue_struct *kmirrord_wq; |
78 | struct work_struct kmirrord_work; | |
a2aebe03 MP |
79 | struct timer_list timer; |
80 | unsigned long timer_pending; | |
81 | ||
72f4b314 | 82 | struct work_struct trigger_event; |
6ad36fe2 | 83 | |
1f965b19 | 84 | unsigned nr_mirrors; |
e4c8b3ba NB |
85 | struct mirror mirror[0]; |
86 | }; | |
87 | ||
1f965b19 | 88 | static void wakeup_mirrord(void *context) |
1da177e4 | 89 | { |
1f965b19 | 90 | struct mirror_set *ms = context; |
1da177e4 | 91 | |
6ad36fe2 HS |
92 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); |
93 | } | |
94 | ||
a2aebe03 MP |
95 | static void delayed_wake_fn(unsigned long data) |
96 | { | |
97 | struct mirror_set *ms = (struct mirror_set *) data; | |
98 | ||
99 | clear_bit(0, &ms->timer_pending); | |
1f965b19 | 100 | wakeup_mirrord(ms); |
a2aebe03 MP |
101 | } |
102 | ||
103 | static void delayed_wake(struct mirror_set *ms) | |
104 | { | |
105 | if (test_and_set_bit(0, &ms->timer_pending)) | |
106 | return; | |
107 | ||
108 | ms->timer.expires = jiffies + HZ / 5; | |
109 | ms->timer.data = (unsigned long) ms; | |
110 | ms->timer.function = delayed_wake_fn; | |
111 | add_timer(&ms->timer); | |
112 | } | |
113 | ||
1f965b19 | 114 | static void wakeup_all_recovery_waiters(void *context) |
1da177e4 | 115 | { |
1f965b19 | 116 | wake_up_all(&_kmirrord_recovery_stopped); |
1da177e4 LT |
117 | } |
118 | ||
1f965b19 | 119 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) |
1da177e4 LT |
120 | { |
121 | unsigned long flags; | |
1da177e4 | 122 | int should_wake = 0; |
1f965b19 | 123 | struct bio_list *bl; |
1da177e4 | 124 | |
1f965b19 HM |
125 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; |
126 | spin_lock_irqsave(&ms->lock, flags); | |
127 | should_wake = !(bl->head); | |
128 | bio_list_add(bl, bio); | |
129 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 LT |
130 | |
131 | if (should_wake) | |
1f965b19 | 132 | wakeup_mirrord(ms); |
1da177e4 LT |
133 | } |
134 | ||
1f965b19 | 135 | static void dispatch_bios(void *context, struct bio_list *bio_list) |
1da177e4 | 136 | { |
1f965b19 HM |
137 | struct mirror_set *ms = context; |
138 | struct bio *bio; | |
1da177e4 | 139 | |
1f965b19 HM |
140 | while ((bio = bio_list_pop(bio_list))) |
141 | queue_bio(ms, bio, WRITE); | |
1da177e4 LT |
142 | } |
143 | ||
06386bbf JB |
144 | #define MIN_READ_RECORDS 20 |
145 | struct dm_raid1_read_record { | |
146 | struct mirror *m; | |
147 | struct dm_bio_details details; | |
148 | }; | |
149 | ||
95f8fac8 MP |
150 | static struct kmem_cache *_dm_raid1_read_record_cache; |
151 | ||
1da177e4 LT |
152 | /* |
153 | * Every mirror should look like this one. | |
154 | */ | |
155 | #define DEFAULT_MIRROR 0 | |
156 | ||
157 | /* | |
06386bbf JB |
158 | * This is yucky. We squirrel the mirror struct away inside |
159 | * bi_next for read/write buffers. This is safe since the bh | |
1da177e4 LT |
160 | * doesn't get submitted to the lower levels of block layer. |
161 | */ | |
06386bbf | 162 | static struct mirror *bio_get_m(struct bio *bio) |
1da177e4 | 163 | { |
06386bbf | 164 | return (struct mirror *) bio->bi_next; |
1da177e4 LT |
165 | } |
166 | ||
06386bbf | 167 | static void bio_set_m(struct bio *bio, struct mirror *m) |
1da177e4 | 168 | { |
06386bbf | 169 | bio->bi_next = (struct bio *) m; |
1da177e4 LT |
170 | } |
171 | ||
72f4b314 JB |
172 | static struct mirror *get_default_mirror(struct mirror_set *ms) |
173 | { | |
174 | return &ms->mirror[atomic_read(&ms->default_mirror)]; | |
175 | } | |
176 | ||
177 | static void set_default_mirror(struct mirror *m) | |
178 | { | |
179 | struct mirror_set *ms = m->ms; | |
180 | struct mirror *m0 = &(ms->mirror[0]); | |
181 | ||
182 | atomic_set(&ms->default_mirror, m - m0); | |
183 | } | |
184 | ||
87968ddd MP |
185 | static struct mirror *get_valid_mirror(struct mirror_set *ms) |
186 | { | |
187 | struct mirror *m; | |
188 | ||
189 | for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++) | |
190 | if (!atomic_read(&m->error_count)) | |
191 | return m; | |
192 | ||
193 | return NULL; | |
194 | } | |
195 | ||
72f4b314 JB |
196 | /* fail_mirror |
197 | * @m: mirror device to fail | |
198 | * @error_type: one of the enum's, DM_RAID1_*_ERROR | |
199 | * | |
200 | * If errors are being handled, record the type of | |
201 | * error encountered for this device. If this type | |
202 | * of error has already been recorded, we can return; | |
203 | * otherwise, we must signal userspace by triggering | |
204 | * an event. Additionally, if the device is the | |
205 | * primary device, we must choose a new primary, but | |
206 | * only if the mirror is in-sync. | |
207 | * | |
208 | * This function must not block. | |
209 | */ | |
210 | static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) | |
211 | { | |
212 | struct mirror_set *ms = m->ms; | |
213 | struct mirror *new; | |
214 | ||
929be8fc MP |
215 | ms->leg_failure = 1; |
216 | ||
72f4b314 JB |
217 | /* |
218 | * error_count is used for nothing more than a | |
219 | * simple way to tell if a device has encountered | |
220 | * errors. | |
221 | */ | |
222 | atomic_inc(&m->error_count); | |
223 | ||
224 | if (test_and_set_bit(error_type, &m->error_type)) | |
225 | return; | |
226 | ||
d460c65a JB |
227 | if (!errors_handled(ms)) |
228 | return; | |
229 | ||
72f4b314 JB |
230 | if (m != get_default_mirror(ms)) |
231 | goto out; | |
232 | ||
233 | if (!ms->in_sync) { | |
234 | /* | |
235 | * Better to issue requests to same failing device | |
236 | * than to risk returning corrupt data. | |
237 | */ | |
238 | DMERR("Primary mirror (%s) failed while out-of-sync: " | |
239 | "Reads may fail.", m->dev->name); | |
240 | goto out; | |
241 | } | |
242 | ||
87968ddd MP |
243 | new = get_valid_mirror(ms); |
244 | if (new) | |
245 | set_default_mirror(new); | |
246 | else | |
72f4b314 JB |
247 | DMWARN("All sides of mirror have failed."); |
248 | ||
249 | out: | |
250 | schedule_work(&ms->trigger_event); | |
251 | } | |
252 | ||
c0da3748 MP |
253 | static int mirror_flush(struct dm_target *ti) |
254 | { | |
255 | struct mirror_set *ms = ti->private; | |
256 | unsigned long error_bits; | |
257 | ||
258 | unsigned int i; | |
259 | struct dm_io_region io[ms->nr_mirrors]; | |
260 | struct mirror *m; | |
261 | struct dm_io_request io_req = { | |
d87f4c14 | 262 | .bi_rw = WRITE_FLUSH, |
c0da3748 MP |
263 | .mem.type = DM_IO_KMEM, |
264 | .mem.ptr.bvec = NULL, | |
265 | .client = ms->io_client, | |
266 | }; | |
267 | ||
268 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) { | |
269 | io[i].bdev = m->dev->bdev; | |
270 | io[i].sector = 0; | |
271 | io[i].count = 0; | |
272 | } | |
273 | ||
274 | error_bits = -1; | |
275 | dm_io(&io_req, ms->nr_mirrors, io, &error_bits); | |
276 | if (unlikely(error_bits != 0)) { | |
277 | for (i = 0; i < ms->nr_mirrors; i++) | |
278 | if (test_bit(i, &error_bits)) | |
279 | fail_mirror(ms->mirror + i, | |
64b30c46 | 280 | DM_RAID1_FLUSH_ERROR); |
c0da3748 MP |
281 | return -EIO; |
282 | } | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
1da177e4 LT |
287 | /*----------------------------------------------------------------- |
288 | * Recovery. | |
289 | * | |
290 | * When a mirror is first activated we may find that some regions | |
291 | * are in the no-sync state. We have to recover these by | |
292 | * recopying from the default mirror to all the others. | |
293 | *---------------------------------------------------------------*/ | |
4cdc1d1f | 294 | static void recovery_complete(int read_err, unsigned long write_err, |
1da177e4 LT |
295 | void *context) |
296 | { | |
1f965b19 HM |
297 | struct dm_region *reg = context; |
298 | struct mirror_set *ms = dm_rh_region_context(reg); | |
8f0205b7 | 299 | int m, bit = 0; |
1da177e4 | 300 | |
8f0205b7 | 301 | if (read_err) { |
f44db678 JB |
302 | /* Read error means the failure of default mirror. */ |
303 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
8f0205b7 JB |
304 | fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); |
305 | } | |
f44db678 | 306 | |
8f0205b7 | 307 | if (write_err) { |
4cdc1d1f | 308 | DMERR_LIMIT("Write error during recovery (error = 0x%lx)", |
f44db678 | 309 | write_err); |
8f0205b7 JB |
310 | /* |
311 | * Bits correspond to devices (excluding default mirror). | |
312 | * The default mirror cannot change during recovery. | |
313 | */ | |
314 | for (m = 0; m < ms->nr_mirrors; m++) { | |
315 | if (&ms->mirror[m] == get_default_mirror(ms)) | |
316 | continue; | |
317 | if (test_bit(bit, &write_err)) | |
318 | fail_mirror(ms->mirror + m, | |
319 | DM_RAID1_SYNC_ERROR); | |
320 | bit++; | |
321 | } | |
322 | } | |
f44db678 | 323 | |
1f965b19 | 324 | dm_rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
325 | } |
326 | ||
1f965b19 | 327 | static int recover(struct mirror_set *ms, struct dm_region *reg) |
1da177e4 LT |
328 | { |
329 | int r; | |
1f965b19 | 330 | unsigned i; |
eb69aca5 | 331 | struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; |
1da177e4 LT |
332 | struct mirror *m; |
333 | unsigned long flags = 0; | |
1f965b19 HM |
334 | region_t key = dm_rh_get_region_key(reg); |
335 | sector_t region_size = dm_rh_get_region_size(ms->rh); | |
1da177e4 LT |
336 | |
337 | /* fill in the source */ | |
72f4b314 | 338 | m = get_default_mirror(ms); |
1da177e4 | 339 | from.bdev = m->dev->bdev; |
1f965b19 HM |
340 | from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
341 | if (key == (ms->nr_regions - 1)) { | |
1da177e4 LT |
342 | /* |
343 | * The final region may be smaller than | |
344 | * region_size. | |
345 | */ | |
1f965b19 | 346 | from.count = ms->ti->len & (region_size - 1); |
1da177e4 | 347 | if (!from.count) |
1f965b19 | 348 | from.count = region_size; |
1da177e4 | 349 | } else |
1f965b19 | 350 | from.count = region_size; |
1da177e4 LT |
351 | |
352 | /* fill in the destinations */ | |
353 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
72f4b314 | 354 | if (&ms->mirror[i] == get_default_mirror(ms)) |
1da177e4 LT |
355 | continue; |
356 | ||
357 | m = ms->mirror + i; | |
358 | dest->bdev = m->dev->bdev; | |
1f965b19 | 359 | dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
1da177e4 LT |
360 | dest->count = from.count; |
361 | dest++; | |
362 | } | |
363 | ||
364 | /* hand to kcopyd */ | |
f7c83e2e JB |
365 | if (!errors_handled(ms)) |
366 | set_bit(DM_KCOPYD_IGNORE_ERROR, &flags); | |
367 | ||
eb69aca5 HM |
368 | r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, |
369 | flags, recovery_complete, reg); | |
1da177e4 LT |
370 | |
371 | return r; | |
372 | } | |
373 | ||
374 | static void do_recovery(struct mirror_set *ms) | |
375 | { | |
1f965b19 HM |
376 | struct dm_region *reg; |
377 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
1da177e4 | 378 | int r; |
1da177e4 LT |
379 | |
380 | /* | |
381 | * Start quiescing some regions. | |
382 | */ | |
1f965b19 | 383 | dm_rh_recovery_prepare(ms->rh); |
1da177e4 LT |
384 | |
385 | /* | |
386 | * Copy any already quiesced regions. | |
387 | */ | |
1f965b19 | 388 | while ((reg = dm_rh_recovery_start(ms->rh))) { |
1da177e4 LT |
389 | r = recover(ms, reg); |
390 | if (r) | |
1f965b19 | 391 | dm_rh_recovery_end(reg, 0); |
1da177e4 LT |
392 | } |
393 | ||
394 | /* | |
395 | * Update the in sync flag. | |
396 | */ | |
397 | if (!ms->in_sync && | |
398 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
399 | /* the sync is complete */ | |
400 | dm_table_event(ms->ti->table); | |
401 | ms->in_sync = 1; | |
402 | } | |
403 | } | |
404 | ||
405 | /*----------------------------------------------------------------- | |
406 | * Reads | |
407 | *---------------------------------------------------------------*/ | |
408 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
409 | { | |
06386bbf JB |
410 | struct mirror *m = get_default_mirror(ms); |
411 | ||
412 | do { | |
413 | if (likely(!atomic_read(&m->error_count))) | |
414 | return m; | |
415 | ||
416 | if (m-- == ms->mirror) | |
417 | m += ms->nr_mirrors; | |
418 | } while (m != get_default_mirror(ms)); | |
419 | ||
420 | return NULL; | |
421 | } | |
422 | ||
423 | static int default_ok(struct mirror *m) | |
424 | { | |
425 | struct mirror *default_mirror = get_default_mirror(m->ms); | |
426 | ||
427 | return !atomic_read(&default_mirror->error_count); | |
428 | } | |
429 | ||
430 | static int mirror_available(struct mirror_set *ms, struct bio *bio) | |
431 | { | |
1f965b19 HM |
432 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
433 | region_t region = dm_rh_bio_to_region(ms->rh, bio); | |
06386bbf | 434 | |
1f965b19 | 435 | if (log->type->in_sync(log, region, 0)) |
06386bbf JB |
436 | return choose_mirror(ms, bio->bi_sector) ? 1 : 0; |
437 | ||
438 | return 0; | |
1da177e4 LT |
439 | } |
440 | ||
441 | /* | |
442 | * remap a buffer to a particular mirror. | |
443 | */ | |
06386bbf JB |
444 | static sector_t map_sector(struct mirror *m, struct bio *bio) |
445 | { | |
4184153f MP |
446 | if (unlikely(!bio->bi_size)) |
447 | return 0; | |
b441a262 | 448 | return m->offset + dm_target_offset(m->ms->ti, bio->bi_sector); |
06386bbf JB |
449 | } |
450 | ||
451 | static void map_bio(struct mirror *m, struct bio *bio) | |
1da177e4 LT |
452 | { |
453 | bio->bi_bdev = m->dev->bdev; | |
06386bbf JB |
454 | bio->bi_sector = map_sector(m, bio); |
455 | } | |
456 | ||
22a1ceb1 | 457 | static void map_region(struct dm_io_region *io, struct mirror *m, |
06386bbf JB |
458 | struct bio *bio) |
459 | { | |
460 | io->bdev = m->dev->bdev; | |
461 | io->sector = map_sector(m, bio); | |
462 | io->count = bio->bi_size >> 9; | |
463 | } | |
464 | ||
04788507 MP |
465 | static void hold_bio(struct mirror_set *ms, struct bio *bio) |
466 | { | |
467 | /* | |
f0703040 TY |
468 | * Lock is required to avoid race condition during suspend |
469 | * process. | |
04788507 | 470 | */ |
f0703040 TY |
471 | spin_lock_irq(&ms->lock); |
472 | ||
04788507 | 473 | if (atomic_read(&ms->suspend)) { |
f0703040 TY |
474 | spin_unlock_irq(&ms->lock); |
475 | ||
476 | /* | |
477 | * If device is suspended, complete the bio. | |
478 | */ | |
04788507 MP |
479 | if (dm_noflush_suspending(ms->ti)) |
480 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
481 | else | |
482 | bio_endio(bio, -EIO); | |
483 | return; | |
484 | } | |
485 | ||
486 | /* | |
487 | * Hold bio until the suspend is complete. | |
488 | */ | |
04788507 MP |
489 | bio_list_add(&ms->holds, bio); |
490 | spin_unlock_irq(&ms->lock); | |
491 | } | |
492 | ||
06386bbf JB |
493 | /*----------------------------------------------------------------- |
494 | * Reads | |
495 | *---------------------------------------------------------------*/ | |
496 | static void read_callback(unsigned long error, void *context) | |
497 | { | |
498 | struct bio *bio = context; | |
499 | struct mirror *m; | |
500 | ||
501 | m = bio_get_m(bio); | |
502 | bio_set_m(bio, NULL); | |
503 | ||
504 | if (likely(!error)) { | |
505 | bio_endio(bio, 0); | |
506 | return; | |
507 | } | |
508 | ||
509 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
510 | ||
511 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
512 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
513 | "Trying alternative device.", | |
514 | m->dev->name); | |
515 | queue_bio(m->ms, bio, bio_rw(bio)); | |
516 | return; | |
517 | } | |
518 | ||
519 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
520 | m->dev->name); | |
521 | bio_endio(bio, -EIO); | |
522 | } | |
523 | ||
524 | /* Asynchronous read. */ | |
525 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
526 | { | |
22a1ceb1 | 527 | struct dm_io_region io; |
06386bbf JB |
528 | struct dm_io_request io_req = { |
529 | .bi_rw = READ, | |
530 | .mem.type = DM_IO_BVEC, | |
531 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
532 | .notify.fn = read_callback, | |
533 | .notify.context = bio, | |
534 | .client = m->ms->io_client, | |
535 | }; | |
536 | ||
537 | map_region(&io, m, bio); | |
538 | bio_set_m(bio, m); | |
1f965b19 HM |
539 | BUG_ON(dm_io(&io_req, 1, &io, NULL)); |
540 | } | |
541 | ||
542 | static inline int region_in_sync(struct mirror_set *ms, region_t region, | |
543 | int may_block) | |
544 | { | |
545 | int state = dm_rh_get_state(ms->rh, region, may_block); | |
546 | return state == DM_RH_CLEAN || state == DM_RH_DIRTY; | |
1da177e4 LT |
547 | } |
548 | ||
549 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
550 | { | |
551 | region_t region; | |
552 | struct bio *bio; | |
553 | struct mirror *m; | |
554 | ||
555 | while ((bio = bio_list_pop(reads))) { | |
1f965b19 | 556 | region = dm_rh_bio_to_region(ms->rh, bio); |
06386bbf | 557 | m = get_default_mirror(ms); |
1da177e4 LT |
558 | |
559 | /* | |
560 | * We can only read balance if the region is in sync. | |
561 | */ | |
1f965b19 | 562 | if (likely(region_in_sync(ms, region, 1))) |
1da177e4 | 563 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
564 | else if (m && atomic_read(&m->error_count)) |
565 | m = NULL; | |
1da177e4 | 566 | |
06386bbf JB |
567 | if (likely(m)) |
568 | read_async_bio(m, bio); | |
569 | else | |
570 | bio_endio(bio, -EIO); | |
1da177e4 LT |
571 | } |
572 | } | |
573 | ||
574 | /*----------------------------------------------------------------- | |
575 | * Writes. | |
576 | * | |
577 | * We do different things with the write io depending on the | |
578 | * state of the region that it's in: | |
579 | * | |
580 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
581 | * RECOVERING: delay the io until recovery completes | |
582 | * NOSYNC: increment pending, just write to the default mirror | |
583 | *---------------------------------------------------------------*/ | |
72f4b314 | 584 | |
72f4b314 | 585 | |
1da177e4 LT |
586 | static void write_callback(unsigned long error, void *context) |
587 | { | |
72f4b314 | 588 | unsigned i, ret = 0; |
1da177e4 LT |
589 | struct bio *bio = (struct bio *) context; |
590 | struct mirror_set *ms; | |
72f4b314 JB |
591 | int should_wake = 0; |
592 | unsigned long flags; | |
1da177e4 | 593 | |
06386bbf JB |
594 | ms = bio_get_m(bio)->ms; |
595 | bio_set_m(bio, NULL); | |
1da177e4 LT |
596 | |
597 | /* | |
598 | * NOTE: We don't decrement the pending count here, | |
599 | * instead it is done by the targets endio function. | |
600 | * This way we handle both writes to SYNC and NOSYNC | |
601 | * regions with the same code. | |
602 | */ | |
60f355ea MP |
603 | if (likely(!error)) { |
604 | bio_endio(bio, ret); | |
605 | return; | |
606 | } | |
1da177e4 | 607 | |
72f4b314 JB |
608 | for (i = 0; i < ms->nr_mirrors; i++) |
609 | if (test_bit(i, &error)) | |
610 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
72f4b314 | 611 | |
60f355ea MP |
612 | /* |
613 | * Need to raise event. Since raising | |
614 | * events can block, we need to do it in | |
615 | * the main thread. | |
616 | */ | |
617 | spin_lock_irqsave(&ms->lock, flags); | |
618 | if (!ms->failures.head) | |
619 | should_wake = 1; | |
620 | bio_list_add(&ms->failures, bio); | |
621 | spin_unlock_irqrestore(&ms->lock, flags); | |
622 | if (should_wake) | |
623 | wakeup_mirrord(ms); | |
1da177e4 LT |
624 | } |
625 | ||
626 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
627 | { | |
628 | unsigned int i; | |
22a1ceb1 | 629 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 630 | struct mirror *m; |
88be163a | 631 | struct dm_io_request io_req = { |
d87f4c14 | 632 | .bi_rw = WRITE | (bio->bi_rw & WRITE_FLUSH_FUA), |
88be163a MB |
633 | .mem.type = DM_IO_BVEC, |
634 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
635 | .notify.fn = write_callback, | |
636 | .notify.context = bio, | |
637 | .client = ms->io_client, | |
638 | }; | |
1da177e4 | 639 | |
06386bbf JB |
640 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
641 | map_region(dest++, m, bio); | |
1da177e4 | 642 | |
06386bbf JB |
643 | /* |
644 | * Use default mirror because we only need it to retrieve the reference | |
645 | * to the mirror set in write_callback(). | |
646 | */ | |
647 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a | 648 | |
1f965b19 | 649 | BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL)); |
1da177e4 LT |
650 | } |
651 | ||
652 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
653 | { | |
654 | int state; | |
655 | struct bio *bio; | |
656 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
7513c2a7 JB |
657 | struct bio_list requeue; |
658 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
659 | region_t region; | |
1da177e4 LT |
660 | |
661 | if (!writes->head) | |
662 | return; | |
663 | ||
664 | /* | |
665 | * Classify each write. | |
666 | */ | |
667 | bio_list_init(&sync); | |
668 | bio_list_init(&nosync); | |
669 | bio_list_init(&recover); | |
7513c2a7 | 670 | bio_list_init(&requeue); |
1da177e4 LT |
671 | |
672 | while ((bio = bio_list_pop(writes))) { | |
d87f4c14 | 673 | if (bio->bi_rw & REQ_FLUSH) { |
4184153f MP |
674 | bio_list_add(&sync, bio); |
675 | continue; | |
676 | } | |
677 | ||
7513c2a7 JB |
678 | region = dm_rh_bio_to_region(ms->rh, bio); |
679 | ||
680 | if (log->type->is_remote_recovering && | |
681 | log->type->is_remote_recovering(log, region)) { | |
682 | bio_list_add(&requeue, bio); | |
683 | continue; | |
684 | } | |
685 | ||
686 | state = dm_rh_get_state(ms->rh, region, 1); | |
1da177e4 | 687 | switch (state) { |
1f965b19 HM |
688 | case DM_RH_CLEAN: |
689 | case DM_RH_DIRTY: | |
1da177e4 LT |
690 | this_list = &sync; |
691 | break; | |
692 | ||
1f965b19 | 693 | case DM_RH_NOSYNC: |
1da177e4 LT |
694 | this_list = &nosync; |
695 | break; | |
696 | ||
1f965b19 | 697 | case DM_RH_RECOVERING: |
1da177e4 LT |
698 | this_list = &recover; |
699 | break; | |
700 | } | |
701 | ||
702 | bio_list_add(this_list, bio); | |
703 | } | |
704 | ||
7513c2a7 JB |
705 | /* |
706 | * Add bios that are delayed due to remote recovery | |
707 | * back on to the write queue | |
708 | */ | |
709 | if (unlikely(requeue.head)) { | |
710 | spin_lock_irq(&ms->lock); | |
711 | bio_list_merge(&ms->writes, &requeue); | |
712 | spin_unlock_irq(&ms->lock); | |
69885683 | 713 | delayed_wake(ms); |
7513c2a7 JB |
714 | } |
715 | ||
1da177e4 LT |
716 | /* |
717 | * Increment the pending counts for any regions that will | |
718 | * be written to (writes to recover regions are going to | |
719 | * be delayed). | |
720 | */ | |
1f965b19 HM |
721 | dm_rh_inc_pending(ms->rh, &sync); |
722 | dm_rh_inc_pending(ms->rh, &nosync); | |
d2b69864 JB |
723 | |
724 | /* | |
725 | * If the flush fails on a previous call and succeeds here, | |
726 | * we must not reset the log_failure variable. We need | |
727 | * userspace interaction to do that. | |
728 | */ | |
729 | ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure; | |
1da177e4 LT |
730 | |
731 | /* | |
732 | * Dispatch io. | |
733 | */ | |
5528d17d | 734 | if (unlikely(ms->log_failure) && errors_handled(ms)) { |
b80aa7a0 JB |
735 | spin_lock_irq(&ms->lock); |
736 | bio_list_merge(&ms->failures, &sync); | |
737 | spin_unlock_irq(&ms->lock); | |
1f965b19 | 738 | wakeup_mirrord(ms); |
b80aa7a0 | 739 | } else |
fc1ff958 | 740 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 741 | do_write(ms, bio); |
1da177e4 LT |
742 | |
743 | while ((bio = bio_list_pop(&recover))) | |
1f965b19 | 744 | dm_rh_delay(ms->rh, bio); |
1da177e4 LT |
745 | |
746 | while ((bio = bio_list_pop(&nosync))) { | |
ede5ea0b MP |
747 | if (unlikely(ms->leg_failure) && errors_handled(ms)) { |
748 | spin_lock_irq(&ms->lock); | |
749 | bio_list_add(&ms->failures, bio); | |
750 | spin_unlock_irq(&ms->lock); | |
751 | wakeup_mirrord(ms); | |
752 | } else { | |
929be8fc MP |
753 | map_bio(get_default_mirror(ms), bio); |
754 | generic_make_request(bio); | |
755 | } | |
1da177e4 LT |
756 | } |
757 | } | |
758 | ||
72f4b314 JB |
759 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
760 | { | |
761 | struct bio *bio; | |
762 | ||
0f398a84 | 763 | if (likely(!failures->head)) |
72f4b314 JB |
764 | return; |
765 | ||
b80aa7a0 JB |
766 | /* |
767 | * If the log has failed, unattempted writes are being | |
0f398a84 | 768 | * put on the holds list. We can't issue those writes |
b80aa7a0 JB |
769 | * until a log has been marked, so we must store them. |
770 | * | |
771 | * If a 'noflush' suspend is in progress, we can requeue | |
772 | * the I/O's to the core. This give userspace a chance | |
773 | * to reconfigure the mirror, at which point the core | |
774 | * will reissue the writes. If the 'noflush' flag is | |
775 | * not set, we have no choice but to return errors. | |
776 | * | |
777 | * Some writes on the failures list may have been | |
778 | * submitted before the log failure and represent a | |
779 | * failure to write to one of the devices. It is ok | |
780 | * for us to treat them the same and requeue them | |
781 | * as well. | |
782 | */ | |
0f398a84 | 783 | while ((bio = bio_list_pop(failures))) { |
60f355ea | 784 | if (!ms->log_failure) { |
0f398a84 | 785 | ms->in_sync = 0; |
c58098be | 786 | dm_rh_mark_nosync(ms->rh, bio); |
0f398a84 | 787 | } |
60f355ea MP |
788 | |
789 | /* | |
790 | * If all the legs are dead, fail the I/O. | |
791 | * If we have been told to handle errors, hold the bio | |
792 | * and wait for userspace to deal with the problem. | |
793 | * Otherwise pretend that the I/O succeeded. (This would | |
794 | * be wrong if the failed leg returned after reboot and | |
795 | * got replicated back to the good legs.) | |
796 | */ | |
797 | if (!get_valid_mirror(ms)) | |
798 | bio_endio(bio, -EIO); | |
799 | else if (errors_handled(ms)) | |
800 | hold_bio(ms, bio); | |
801 | else | |
802 | bio_endio(bio, 0); | |
b80aa7a0 | 803 | } |
72f4b314 JB |
804 | } |
805 | ||
806 | static void trigger_event(struct work_struct *work) | |
807 | { | |
808 | struct mirror_set *ms = | |
809 | container_of(work, struct mirror_set, trigger_event); | |
810 | ||
811 | dm_table_event(ms->ti->table); | |
812 | } | |
813 | ||
1da177e4 LT |
814 | /*----------------------------------------------------------------- |
815 | * kmirrord | |
816 | *---------------------------------------------------------------*/ | |
a2aebe03 | 817 | static void do_mirror(struct work_struct *work) |
1da177e4 | 818 | { |
1f965b19 HM |
819 | struct mirror_set *ms = container_of(work, struct mirror_set, |
820 | kmirrord_work); | |
72f4b314 JB |
821 | struct bio_list reads, writes, failures; |
822 | unsigned long flags; | |
1da177e4 | 823 | |
72f4b314 | 824 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
825 | reads = ms->reads; |
826 | writes = ms->writes; | |
72f4b314 | 827 | failures = ms->failures; |
1da177e4 LT |
828 | bio_list_init(&ms->reads); |
829 | bio_list_init(&ms->writes); | |
72f4b314 JB |
830 | bio_list_init(&ms->failures); |
831 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 | 832 | |
1f965b19 | 833 | dm_rh_update_states(ms->rh, errors_handled(ms)); |
1da177e4 LT |
834 | do_recovery(ms); |
835 | do_reads(ms, &reads); | |
836 | do_writes(ms, &writes); | |
72f4b314 | 837 | do_failures(ms, &failures); |
7ff14a36 MP |
838 | |
839 | dm_table_unplug_all(ms->ti->table); | |
1da177e4 LT |
840 | } |
841 | ||
1da177e4 LT |
842 | /*----------------------------------------------------------------- |
843 | * Target functions | |
844 | *---------------------------------------------------------------*/ | |
845 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
846 | uint32_t region_size, | |
847 | struct dm_target *ti, | |
416cd17b | 848 | struct dm_dirty_log *dl) |
1da177e4 LT |
849 | { |
850 | size_t len; | |
851 | struct mirror_set *ms = NULL; | |
852 | ||
1da177e4 LT |
853 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
854 | ||
dd00cc48 | 855 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 856 | if (!ms) { |
72d94861 | 857 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
858 | return NULL; |
859 | } | |
860 | ||
1da177e4 | 861 | spin_lock_init(&ms->lock); |
5339fc2d MP |
862 | bio_list_init(&ms->reads); |
863 | bio_list_init(&ms->writes); | |
864 | bio_list_init(&ms->failures); | |
865 | bio_list_init(&ms->holds); | |
1da177e4 LT |
866 | |
867 | ms->ti = ti; | |
868 | ms->nr_mirrors = nr_mirrors; | |
869 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
870 | ms->in_sync = 0; | |
b80aa7a0 | 871 | ms->log_failure = 0; |
929be8fc | 872 | ms->leg_failure = 0; |
b80aa7a0 | 873 | atomic_set(&ms->suspend, 0); |
72f4b314 | 874 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 875 | |
95f8fac8 MP |
876 | ms->read_record_pool = mempool_create_slab_pool(MIN_READ_RECORDS, |
877 | _dm_raid1_read_record_cache); | |
878 | ||
06386bbf JB |
879 | if (!ms->read_record_pool) { |
880 | ti->error = "Error creating mirror read_record_pool"; | |
881 | kfree(ms); | |
882 | return NULL; | |
883 | } | |
884 | ||
88be163a MB |
885 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
886 | if (IS_ERR(ms->io_client)) { | |
887 | ti->error = "Error creating dm_io client"; | |
06386bbf | 888 | mempool_destroy(ms->read_record_pool); |
88be163a MB |
889 | kfree(ms); |
890 | return NULL; | |
891 | } | |
892 | ||
1f965b19 HM |
893 | ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord, |
894 | wakeup_all_recovery_waiters, | |
895 | ms->ti->begin, MAX_RECOVERY, | |
896 | dl, region_size, ms->nr_regions); | |
897 | if (IS_ERR(ms->rh)) { | |
72d94861 | 898 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 899 | dm_io_client_destroy(ms->io_client); |
06386bbf | 900 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
901 | kfree(ms); |
902 | return NULL; | |
903 | } | |
904 | ||
905 | return ms; | |
906 | } | |
907 | ||
908 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
909 | unsigned int m) | |
910 | { | |
911 | while (m--) | |
912 | dm_put_device(ti, ms->mirror[m].dev); | |
913 | ||
88be163a | 914 | dm_io_client_destroy(ms->io_client); |
1f965b19 | 915 | dm_region_hash_destroy(ms->rh); |
06386bbf | 916 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
917 | kfree(ms); |
918 | } | |
919 | ||
1da177e4 LT |
920 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
921 | unsigned int mirror, char **argv) | |
922 | { | |
4ee218cd | 923 | unsigned long long offset; |
1da177e4 | 924 | |
4ee218cd | 925 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 926 | ti->error = "Invalid offset"; |
1da177e4 LT |
927 | return -EINVAL; |
928 | } | |
929 | ||
8215d6ec | 930 | if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), |
1da177e4 | 931 | &ms->mirror[mirror].dev)) { |
72d94861 | 932 | ti->error = "Device lookup failure"; |
1da177e4 LT |
933 | return -ENXIO; |
934 | } | |
935 | ||
aa5617c5 | 936 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
937 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
938 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
939 | ms->mirror[mirror].offset = offset; |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
1da177e4 LT |
944 | /* |
945 | * Create dirty log: log_type #log_params <log_params> | |
946 | */ | |
416cd17b | 947 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1f965b19 HM |
948 | unsigned argc, char **argv, |
949 | unsigned *args_used) | |
1da177e4 | 950 | { |
1f965b19 | 951 | unsigned param_count; |
416cd17b | 952 | struct dm_dirty_log *dl; |
1da177e4 LT |
953 | |
954 | if (argc < 2) { | |
72d94861 | 955 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
956 | return NULL; |
957 | } | |
958 | ||
959 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 960 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
961 | return NULL; |
962 | } | |
963 | ||
964 | *args_used = 2 + param_count; | |
965 | ||
966 | if (argc < *args_used) { | |
72d94861 | 967 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
968 | return NULL; |
969 | } | |
970 | ||
c0da3748 MP |
971 | dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count, |
972 | argv + 2); | |
1da177e4 | 973 | if (!dl) { |
72d94861 | 974 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
975 | return NULL; |
976 | } | |
977 | ||
1da177e4 LT |
978 | return dl; |
979 | } | |
980 | ||
a8e6afa2 JB |
981 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
982 | unsigned *args_used) | |
983 | { | |
984 | unsigned num_features; | |
985 | struct dm_target *ti = ms->ti; | |
986 | ||
987 | *args_used = 0; | |
988 | ||
989 | if (!argc) | |
990 | return 0; | |
991 | ||
992 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
993 | ti->error = "Invalid number of features"; | |
994 | return -EINVAL; | |
995 | } | |
996 | ||
997 | argc--; | |
998 | argv++; | |
999 | (*args_used)++; | |
1000 | ||
1001 | if (num_features > argc) { | |
1002 | ti->error = "Not enough arguments to support feature count"; | |
1003 | return -EINVAL; | |
1004 | } | |
1005 | ||
1006 | if (!strcmp("handle_errors", argv[0])) | |
1007 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
1008 | else { | |
1009 | ti->error = "Unrecognised feature requested"; | |
1010 | return -EINVAL; | |
1011 | } | |
1012 | ||
1013 | (*args_used)++; | |
1014 | ||
1015 | return 0; | |
1016 | } | |
1017 | ||
1da177e4 LT |
1018 | /* |
1019 | * Construct a mirror mapping: | |
1020 | * | |
1021 | * log_type #log_params <log_params> | |
1022 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1023 | * [#features <features>] |
1da177e4 LT |
1024 | * |
1025 | * log_type is "core" or "disk" | |
1026 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1027 | * |
1028 | * If present, features must be "handle_errors". | |
1da177e4 | 1029 | */ |
1da177e4 LT |
1030 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1031 | { | |
1032 | int r; | |
1033 | unsigned int nr_mirrors, m, args_used; | |
1034 | struct mirror_set *ms; | |
416cd17b | 1035 | struct dm_dirty_log *dl; |
1da177e4 LT |
1036 | |
1037 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
1038 | if (!dl) | |
1039 | return -EINVAL; | |
1040 | ||
1041 | argv += args_used; | |
1042 | argc -= args_used; | |
1043 | ||
1044 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
eb69aca5 | 1045 | nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) { |
72d94861 | 1046 | ti->error = "Invalid number of mirrors"; |
416cd17b | 1047 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1048 | return -EINVAL; |
1049 | } | |
1050 | ||
1051 | argv++, argc--; | |
1052 | ||
a8e6afa2 JB |
1053 | if (argc < nr_mirrors * 2) { |
1054 | ti->error = "Too few mirror arguments"; | |
416cd17b | 1055 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1056 | return -EINVAL; |
1057 | } | |
1058 | ||
1059 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
1060 | if (!ms) { | |
416cd17b | 1061 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1062 | return -ENOMEM; |
1063 | } | |
1064 | ||
1065 | /* Get the mirror parameter sets */ | |
1066 | for (m = 0; m < nr_mirrors; m++) { | |
1067 | r = get_mirror(ms, ti, m, argv); | |
1068 | if (r) { | |
1069 | free_context(ms, ti, m); | |
1070 | return r; | |
1071 | } | |
1072 | argv += 2; | |
1073 | argc -= 2; | |
1074 | } | |
1075 | ||
1076 | ti->private = ms; | |
1f965b19 | 1077 | ti->split_io = dm_rh_get_region_size(ms->rh); |
4184153f | 1078 | ti->num_flush_requests = 1; |
1da177e4 | 1079 | |
6ad36fe2 HS |
1080 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
1081 | if (!ms->kmirrord_wq) { | |
1082 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1083 | r = -ENOMEM; |
1084 | goto err_free_context; | |
6ad36fe2 HS |
1085 | } |
1086 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
1087 | init_timer(&ms->timer); |
1088 | ms->timer_pending = 0; | |
72f4b314 | 1089 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 1090 | |
a8e6afa2 | 1091 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1092 | if (r) |
1093 | goto err_destroy_wq; | |
a8e6afa2 JB |
1094 | |
1095 | argv += args_used; | |
1096 | argc -= args_used; | |
1097 | ||
f44db678 JB |
1098 | /* |
1099 | * Any read-balancing addition depends on the | |
1100 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1101 | * This is because the decision to balance depends | |
1102 | * on the sync state of a region. If the above | |
1103 | * flag is not present, we ignore errors; and | |
1104 | * the sync state may be inaccurate. | |
1105 | */ | |
1106 | ||
a8e6afa2 JB |
1107 | if (argc) { |
1108 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1109 | r = -EINVAL; |
1110 | goto err_destroy_wq; | |
a8e6afa2 JB |
1111 | } |
1112 | ||
1f965b19 | 1113 | r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client); |
a72cf737 DM |
1114 | if (r) |
1115 | goto err_destroy_wq; | |
1da177e4 | 1116 | |
1f965b19 | 1117 | wakeup_mirrord(ms); |
1da177e4 | 1118 | return 0; |
a72cf737 DM |
1119 | |
1120 | err_destroy_wq: | |
1121 | destroy_workqueue(ms->kmirrord_wq); | |
1122 | err_free_context: | |
1123 | free_context(ms, ti, ms->nr_mirrors); | |
1124 | return r; | |
1da177e4 LT |
1125 | } |
1126 | ||
1127 | static void mirror_dtr(struct dm_target *ti) | |
1128 | { | |
1129 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1130 | ||
a2aebe03 | 1131 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1132 | flush_workqueue(ms->kmirrord_wq); |
18776c73 | 1133 | flush_scheduled_work(); |
eb69aca5 | 1134 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1135 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1136 | free_context(ms, ti, ms->nr_mirrors); |
1137 | } | |
1138 | ||
1da177e4 LT |
1139 | /* |
1140 | * Mirror mapping function | |
1141 | */ | |
1142 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1143 | union map_info *map_context) | |
1144 | { | |
1145 | int r, rw = bio_rw(bio); | |
1146 | struct mirror *m; | |
1147 | struct mirror_set *ms = ti->private; | |
06386bbf | 1148 | struct dm_raid1_read_record *read_record = NULL; |
1f965b19 | 1149 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 LT |
1150 | |
1151 | if (rw == WRITE) { | |
06386bbf | 1152 | /* Save region for mirror_end_io() handler */ |
1f965b19 | 1153 | map_context->ll = dm_rh_bio_to_region(ms->rh, bio); |
1da177e4 | 1154 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1155 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1156 | } |
1157 | ||
1f965b19 | 1158 | r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0); |
1da177e4 LT |
1159 | if (r < 0 && r != -EWOULDBLOCK) |
1160 | return r; | |
1161 | ||
1da177e4 | 1162 | /* |
06386bbf | 1163 | * If region is not in-sync queue the bio. |
1da177e4 | 1164 | */ |
06386bbf JB |
1165 | if (!r || (r == -EWOULDBLOCK)) { |
1166 | if (rw == READA) | |
1167 | return -EWOULDBLOCK; | |
1da177e4 | 1168 | |
1da177e4 | 1169 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1170 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1171 | } |
1172 | ||
06386bbf JB |
1173 | /* |
1174 | * The region is in-sync and we can perform reads directly. | |
1175 | * Store enough information so we can retry if it fails. | |
1176 | */ | |
1da177e4 | 1177 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1178 | if (unlikely(!m)) |
1da177e4 LT |
1179 | return -EIO; |
1180 | ||
06386bbf JB |
1181 | read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO); |
1182 | if (likely(read_record)) { | |
1183 | dm_bio_record(&read_record->details, bio); | |
1184 | map_context->ptr = read_record; | |
1185 | read_record->m = m; | |
1186 | } | |
1187 | ||
1188 | map_bio(m, bio); | |
1189 | ||
d2a7ad29 | 1190 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1194 | int error, union map_info *map_context) | |
1195 | { | |
1196 | int rw = bio_rw(bio); | |
1197 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1198 | struct mirror *m = NULL; |
1199 | struct dm_bio_details *bd = NULL; | |
1200 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1201 | |
1202 | /* | |
1203 | * We need to dec pending if this was a write. | |
1204 | */ | |
06386bbf | 1205 | if (rw == WRITE) { |
d87f4c14 | 1206 | if (!(bio->bi_rw & REQ_FLUSH)) |
4184153f | 1207 | dm_rh_dec(ms->rh, map_context->ll); |
06386bbf JB |
1208 | return error; |
1209 | } | |
1da177e4 | 1210 | |
06386bbf JB |
1211 | if (error == -EOPNOTSUPP) |
1212 | goto out; | |
1213 | ||
7b6d91da | 1214 | if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) |
06386bbf JB |
1215 | goto out; |
1216 | ||
1217 | if (unlikely(error)) { | |
1218 | if (!read_record) { | |
1219 | /* | |
1220 | * There wasn't enough memory to record necessary | |
1221 | * information for a retry or there was no other | |
1222 | * mirror in-sync. | |
1223 | */ | |
e03f1a84 | 1224 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1225 | return -EIO; |
1226 | } | |
e03f1a84 AB |
1227 | |
1228 | m = read_record->m; | |
1229 | ||
06386bbf JB |
1230 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1231 | m->dev->name); | |
1232 | ||
06386bbf JB |
1233 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1234 | ||
1235 | /* | |
1236 | * A failed read is requeued for another attempt using an intact | |
1237 | * mirror. | |
1238 | */ | |
1239 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1240 | bd = &read_record->details; | |
1241 | ||
1242 | dm_bio_restore(bd, bio); | |
1243 | mempool_free(read_record, ms->read_record_pool); | |
1244 | map_context->ptr = NULL; | |
1245 | queue_bio(ms, bio, rw); | |
1246 | return 1; | |
1247 | } | |
1248 | DMERR("All replicated volumes dead, failing I/O"); | |
1249 | } | |
1250 | ||
1251 | out: | |
1252 | if (read_record) { | |
1253 | mempool_free(read_record, ms->read_record_pool); | |
1254 | map_context->ptr = NULL; | |
1255 | } | |
1256 | ||
1257 | return error; | |
1da177e4 LT |
1258 | } |
1259 | ||
b80aa7a0 | 1260 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1261 | { |
1262 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1f965b19 | 1263 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 | 1264 | |
04788507 MP |
1265 | struct bio_list holds; |
1266 | struct bio *bio; | |
1267 | ||
b80aa7a0 JB |
1268 | atomic_set(&ms->suspend, 1); |
1269 | ||
f0703040 TY |
1270 | /* |
1271 | * Process bios in the hold list to start recovery waiting | |
1272 | * for bios in the hold list. After the process, no bio has | |
1273 | * a chance to be added in the hold list because ms->suspend | |
1274 | * is set. | |
1275 | */ | |
1276 | spin_lock_irq(&ms->lock); | |
1277 | holds = ms->holds; | |
1278 | bio_list_init(&ms->holds); | |
1279 | spin_unlock_irq(&ms->lock); | |
1280 | ||
1281 | while ((bio = bio_list_pop(&holds))) | |
1282 | hold_bio(ms, bio); | |
1283 | ||
b80aa7a0 JB |
1284 | /* |
1285 | * We must finish up all the work that we've | |
1286 | * generated (i.e. recovery work). | |
1287 | */ | |
1f965b19 | 1288 | dm_rh_stop_recovery(ms->rh); |
33184048 | 1289 | |
33184048 | 1290 | wait_event(_kmirrord_recovery_stopped, |
1f965b19 | 1291 | !dm_rh_recovery_in_flight(ms->rh)); |
33184048 | 1292 | |
b80aa7a0 JB |
1293 | if (log->type->presuspend && log->type->presuspend(log)) |
1294 | /* FIXME: need better error handling */ | |
1295 | DMWARN("log presuspend failed"); | |
1296 | ||
1297 | /* | |
1298 | * Now that recovery is complete/stopped and the | |
1299 | * delayed bios are queued, we need to wait for | |
1300 | * the worker thread to complete. This way, | |
1301 | * we know that all of our I/O has been pushed. | |
1302 | */ | |
1303 | flush_workqueue(ms->kmirrord_wq); | |
1304 | } | |
1305 | ||
1306 | static void mirror_postsuspend(struct dm_target *ti) | |
1307 | { | |
1308 | struct mirror_set *ms = ti->private; | |
1f965b19 | 1309 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 | 1310 | |
6b3df0d7 | 1311 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1312 | /* FIXME: need better error handling */ |
b80aa7a0 | 1313 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1314 | } |
1315 | ||
1316 | static void mirror_resume(struct dm_target *ti) | |
1317 | { | |
b80aa7a0 | 1318 | struct mirror_set *ms = ti->private; |
1f965b19 | 1319 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 JB |
1320 | |
1321 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1322 | if (log->type->resume && log->type->resume(log)) |
1323 | /* FIXME: need better error handling */ | |
1324 | DMWARN("log resume failed"); | |
1f965b19 | 1325 | dm_rh_start_recovery(ms->rh); |
1da177e4 LT |
1326 | } |
1327 | ||
af195ac8 JB |
1328 | /* |
1329 | * device_status_char | |
1330 | * @m: mirror device/leg we want the status of | |
1331 | * | |
1332 | * We return one character representing the most severe error | |
1333 | * we have encountered. | |
1334 | * A => Alive - No failures | |
1335 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1336 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1337 | * R => Read - A read failure occurred, mirror data unaffected | |
1338 | * | |
1339 | * Returns: <char> | |
1340 | */ | |
1341 | static char device_status_char(struct mirror *m) | |
1342 | { | |
1343 | if (!atomic_read(&(m->error_count))) | |
1344 | return 'A'; | |
1345 | ||
64b30c46 MP |
1346 | return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' : |
1347 | (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
af195ac8 JB |
1348 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : |
1349 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1350 | } | |
1351 | ||
1352 | ||
1da177e4 LT |
1353 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1354 | char *result, unsigned int maxlen) | |
1355 | { | |
315dcc22 | 1356 | unsigned int m, sz = 0; |
1da177e4 | 1357 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1f965b19 | 1358 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
af195ac8 | 1359 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1360 | |
1da177e4 LT |
1361 | switch (type) { |
1362 | case STATUSTYPE_INFO: | |
1363 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1364 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1365 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1366 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1367 | } | |
1368 | buffer[m] = '\0'; | |
1da177e4 | 1369 | |
af195ac8 | 1370 | DMEMIT("%llu/%llu 1 %s ", |
1f965b19 | 1371 | (unsigned long long)log->type->get_sync_count(log), |
af195ac8 | 1372 | (unsigned long long)ms->nr_regions, buffer); |
315dcc22 | 1373 | |
1f965b19 | 1374 | sz += log->type->status(log, type, result+sz, maxlen-sz); |
315dcc22 | 1375 | |
1da177e4 LT |
1376 | break; |
1377 | ||
1378 | case STATUSTYPE_TABLE: | |
1f965b19 | 1379 | sz = log->type->status(log, type, result, maxlen); |
315dcc22 | 1380 | |
e52b8f6d | 1381 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1382 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1383 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1384 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1385 | |
1386 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1387 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1388 | } |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | ||
af4874e0 MS |
1393 | static int mirror_iterate_devices(struct dm_target *ti, |
1394 | iterate_devices_callout_fn fn, void *data) | |
1395 | { | |
1396 | struct mirror_set *ms = ti->private; | |
1397 | int ret = 0; | |
1398 | unsigned i; | |
1399 | ||
1400 | for (i = 0; !ret && i < ms->nr_mirrors; i++) | |
1401 | ret = fn(ti, ms->mirror[i].dev, | |
5dea271b | 1402 | ms->mirror[i].offset, ti->len, data); |
af4874e0 MS |
1403 | |
1404 | return ret; | |
1405 | } | |
1406 | ||
1da177e4 LT |
1407 | static struct target_type mirror_target = { |
1408 | .name = "mirror", | |
af4874e0 | 1409 | .version = {1, 12, 0}, |
1da177e4 LT |
1410 | .module = THIS_MODULE, |
1411 | .ctr = mirror_ctr, | |
1412 | .dtr = mirror_dtr, | |
1413 | .map = mirror_map, | |
1414 | .end_io = mirror_end_io, | |
b80aa7a0 | 1415 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1416 | .postsuspend = mirror_postsuspend, |
1417 | .resume = mirror_resume, | |
1418 | .status = mirror_status, | |
af4874e0 | 1419 | .iterate_devices = mirror_iterate_devices, |
1da177e4 LT |
1420 | }; |
1421 | ||
1422 | static int __init dm_mirror_init(void) | |
1423 | { | |
1424 | int r; | |
1425 | ||
95f8fac8 MP |
1426 | _dm_raid1_read_record_cache = KMEM_CACHE(dm_raid1_read_record, 0); |
1427 | if (!_dm_raid1_read_record_cache) { | |
1428 | DMERR("Can't allocate dm_raid1_read_record cache"); | |
1429 | r = -ENOMEM; | |
1430 | goto bad_cache; | |
1431 | } | |
1432 | ||
1da177e4 | 1433 | r = dm_register_target(&mirror_target); |
95f8fac8 | 1434 | if (r < 0) { |
0cd33124 | 1435 | DMERR("Failed to register mirror target"); |
95f8fac8 MP |
1436 | goto bad_target; |
1437 | } | |
1438 | ||
1439 | return 0; | |
1da177e4 | 1440 | |
95f8fac8 MP |
1441 | bad_target: |
1442 | kmem_cache_destroy(_dm_raid1_read_record_cache); | |
1443 | bad_cache: | |
1da177e4 LT |
1444 | return r; |
1445 | } | |
1446 | ||
1447 | static void __exit dm_mirror_exit(void) | |
1448 | { | |
10d3bd09 | 1449 | dm_unregister_target(&mirror_target); |
95f8fac8 | 1450 | kmem_cache_destroy(_dm_raid1_read_record_cache); |
1da177e4 LT |
1451 | } |
1452 | ||
1453 | /* Module hooks */ | |
1454 | module_init(dm_mirror_init); | |
1455 | module_exit(dm_mirror_exit); | |
1456 | ||
1457 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1458 | MODULE_AUTHOR("Joe Thornber"); | |
1459 | MODULE_LICENSE("GPL"); |