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 = { | |
262 | .bi_rw = WRITE_BARRIER, | |
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; | |
06386bbf JB |
448 | return m->offset + (bio->bi_sector - m->ms->ti->begin); |
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 | /* | |
468 | * If device is suspended, complete the bio. | |
469 | */ | |
470 | if (atomic_read(&ms->suspend)) { | |
471 | if (dm_noflush_suspending(ms->ti)) | |
472 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
473 | else | |
474 | bio_endio(bio, -EIO); | |
475 | return; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Hold bio until the suspend is complete. | |
480 | */ | |
481 | spin_lock_irq(&ms->lock); | |
482 | bio_list_add(&ms->holds, bio); | |
483 | spin_unlock_irq(&ms->lock); | |
484 | } | |
485 | ||
06386bbf JB |
486 | /*----------------------------------------------------------------- |
487 | * Reads | |
488 | *---------------------------------------------------------------*/ | |
489 | static void read_callback(unsigned long error, void *context) | |
490 | { | |
491 | struct bio *bio = context; | |
492 | struct mirror *m; | |
493 | ||
494 | m = bio_get_m(bio); | |
495 | bio_set_m(bio, NULL); | |
496 | ||
497 | if (likely(!error)) { | |
498 | bio_endio(bio, 0); | |
499 | return; | |
500 | } | |
501 | ||
502 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
503 | ||
504 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
505 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
506 | "Trying alternative device.", | |
507 | m->dev->name); | |
508 | queue_bio(m->ms, bio, bio_rw(bio)); | |
509 | return; | |
510 | } | |
511 | ||
512 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
513 | m->dev->name); | |
514 | bio_endio(bio, -EIO); | |
515 | } | |
516 | ||
517 | /* Asynchronous read. */ | |
518 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
519 | { | |
22a1ceb1 | 520 | struct dm_io_region io; |
06386bbf JB |
521 | struct dm_io_request io_req = { |
522 | .bi_rw = READ, | |
523 | .mem.type = DM_IO_BVEC, | |
524 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
525 | .notify.fn = read_callback, | |
526 | .notify.context = bio, | |
527 | .client = m->ms->io_client, | |
528 | }; | |
529 | ||
530 | map_region(&io, m, bio); | |
531 | bio_set_m(bio, m); | |
1f965b19 HM |
532 | BUG_ON(dm_io(&io_req, 1, &io, NULL)); |
533 | } | |
534 | ||
535 | static inline int region_in_sync(struct mirror_set *ms, region_t region, | |
536 | int may_block) | |
537 | { | |
538 | int state = dm_rh_get_state(ms->rh, region, may_block); | |
539 | return state == DM_RH_CLEAN || state == DM_RH_DIRTY; | |
1da177e4 LT |
540 | } |
541 | ||
542 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
543 | { | |
544 | region_t region; | |
545 | struct bio *bio; | |
546 | struct mirror *m; | |
547 | ||
548 | while ((bio = bio_list_pop(reads))) { | |
1f965b19 | 549 | region = dm_rh_bio_to_region(ms->rh, bio); |
06386bbf | 550 | m = get_default_mirror(ms); |
1da177e4 LT |
551 | |
552 | /* | |
553 | * We can only read balance if the region is in sync. | |
554 | */ | |
1f965b19 | 555 | if (likely(region_in_sync(ms, region, 1))) |
1da177e4 | 556 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
557 | else if (m && atomic_read(&m->error_count)) |
558 | m = NULL; | |
1da177e4 | 559 | |
06386bbf JB |
560 | if (likely(m)) |
561 | read_async_bio(m, bio); | |
562 | else | |
563 | bio_endio(bio, -EIO); | |
1da177e4 LT |
564 | } |
565 | } | |
566 | ||
567 | /*----------------------------------------------------------------- | |
568 | * Writes. | |
569 | * | |
570 | * We do different things with the write io depending on the | |
571 | * state of the region that it's in: | |
572 | * | |
573 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
574 | * RECOVERING: delay the io until recovery completes | |
575 | * NOSYNC: increment pending, just write to the default mirror | |
576 | *---------------------------------------------------------------*/ | |
72f4b314 | 577 | |
72f4b314 | 578 | |
1da177e4 LT |
579 | static void write_callback(unsigned long error, void *context) |
580 | { | |
72f4b314 | 581 | unsigned i, ret = 0; |
1da177e4 LT |
582 | struct bio *bio = (struct bio *) context; |
583 | struct mirror_set *ms; | |
72f4b314 JB |
584 | int should_wake = 0; |
585 | unsigned long flags; | |
1da177e4 | 586 | |
06386bbf JB |
587 | ms = bio_get_m(bio)->ms; |
588 | bio_set_m(bio, NULL); | |
1da177e4 LT |
589 | |
590 | /* | |
591 | * NOTE: We don't decrement the pending count here, | |
592 | * instead it is done by the targets endio function. | |
593 | * This way we handle both writes to SYNC and NOSYNC | |
594 | * regions with the same code. | |
595 | */ | |
60f355ea MP |
596 | if (likely(!error)) { |
597 | bio_endio(bio, ret); | |
598 | return; | |
599 | } | |
1da177e4 | 600 | |
72f4b314 JB |
601 | for (i = 0; i < ms->nr_mirrors; i++) |
602 | if (test_bit(i, &error)) | |
603 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
72f4b314 | 604 | |
60f355ea MP |
605 | /* |
606 | * Need to raise event. Since raising | |
607 | * events can block, we need to do it in | |
608 | * the main thread. | |
609 | */ | |
610 | spin_lock_irqsave(&ms->lock, flags); | |
611 | if (!ms->failures.head) | |
612 | should_wake = 1; | |
613 | bio_list_add(&ms->failures, bio); | |
614 | spin_unlock_irqrestore(&ms->lock, flags); | |
615 | if (should_wake) | |
616 | wakeup_mirrord(ms); | |
1da177e4 LT |
617 | } |
618 | ||
619 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
620 | { | |
621 | unsigned int i; | |
22a1ceb1 | 622 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 623 | struct mirror *m; |
88be163a | 624 | struct dm_io_request io_req = { |
4184153f | 625 | .bi_rw = WRITE | (bio->bi_rw & WRITE_BARRIER), |
88be163a MB |
626 | .mem.type = DM_IO_BVEC, |
627 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
628 | .notify.fn = write_callback, | |
629 | .notify.context = bio, | |
630 | .client = ms->io_client, | |
631 | }; | |
1da177e4 | 632 | |
06386bbf JB |
633 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
634 | map_region(dest++, m, bio); | |
1da177e4 | 635 | |
06386bbf JB |
636 | /* |
637 | * Use default mirror because we only need it to retrieve the reference | |
638 | * to the mirror set in write_callback(). | |
639 | */ | |
640 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a | 641 | |
1f965b19 | 642 | BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL)); |
1da177e4 LT |
643 | } |
644 | ||
645 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
646 | { | |
647 | int state; | |
648 | struct bio *bio; | |
649 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
7513c2a7 JB |
650 | struct bio_list requeue; |
651 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
652 | region_t region; | |
1da177e4 LT |
653 | |
654 | if (!writes->head) | |
655 | return; | |
656 | ||
657 | /* | |
658 | * Classify each write. | |
659 | */ | |
660 | bio_list_init(&sync); | |
661 | bio_list_init(&nosync); | |
662 | bio_list_init(&recover); | |
7513c2a7 | 663 | bio_list_init(&requeue); |
1da177e4 LT |
664 | |
665 | while ((bio = bio_list_pop(writes))) { | |
4184153f MP |
666 | if (unlikely(bio_empty_barrier(bio))) { |
667 | bio_list_add(&sync, bio); | |
668 | continue; | |
669 | } | |
670 | ||
7513c2a7 JB |
671 | region = dm_rh_bio_to_region(ms->rh, bio); |
672 | ||
673 | if (log->type->is_remote_recovering && | |
674 | log->type->is_remote_recovering(log, region)) { | |
675 | bio_list_add(&requeue, bio); | |
676 | continue; | |
677 | } | |
678 | ||
679 | state = dm_rh_get_state(ms->rh, region, 1); | |
1da177e4 | 680 | switch (state) { |
1f965b19 HM |
681 | case DM_RH_CLEAN: |
682 | case DM_RH_DIRTY: | |
1da177e4 LT |
683 | this_list = &sync; |
684 | break; | |
685 | ||
1f965b19 | 686 | case DM_RH_NOSYNC: |
1da177e4 LT |
687 | this_list = &nosync; |
688 | break; | |
689 | ||
1f965b19 | 690 | case DM_RH_RECOVERING: |
1da177e4 LT |
691 | this_list = &recover; |
692 | break; | |
693 | } | |
694 | ||
695 | bio_list_add(this_list, bio); | |
696 | } | |
697 | ||
7513c2a7 JB |
698 | /* |
699 | * Add bios that are delayed due to remote recovery | |
700 | * back on to the write queue | |
701 | */ | |
702 | if (unlikely(requeue.head)) { | |
703 | spin_lock_irq(&ms->lock); | |
704 | bio_list_merge(&ms->writes, &requeue); | |
705 | spin_unlock_irq(&ms->lock); | |
69885683 | 706 | delayed_wake(ms); |
7513c2a7 JB |
707 | } |
708 | ||
1da177e4 LT |
709 | /* |
710 | * Increment the pending counts for any regions that will | |
711 | * be written to (writes to recover regions are going to | |
712 | * be delayed). | |
713 | */ | |
1f965b19 HM |
714 | dm_rh_inc_pending(ms->rh, &sync); |
715 | dm_rh_inc_pending(ms->rh, &nosync); | |
d2b69864 JB |
716 | |
717 | /* | |
718 | * If the flush fails on a previous call and succeeds here, | |
719 | * we must not reset the log_failure variable. We need | |
720 | * userspace interaction to do that. | |
721 | */ | |
722 | ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure; | |
1da177e4 LT |
723 | |
724 | /* | |
725 | * Dispatch io. | |
726 | */ | |
5528d17d | 727 | if (unlikely(ms->log_failure) && errors_handled(ms)) { |
b80aa7a0 JB |
728 | spin_lock_irq(&ms->lock); |
729 | bio_list_merge(&ms->failures, &sync); | |
730 | spin_unlock_irq(&ms->lock); | |
1f965b19 | 731 | wakeup_mirrord(ms); |
b80aa7a0 | 732 | } else |
fc1ff958 | 733 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 734 | do_write(ms, bio); |
1da177e4 LT |
735 | |
736 | while ((bio = bio_list_pop(&recover))) | |
1f965b19 | 737 | dm_rh_delay(ms->rh, bio); |
1da177e4 LT |
738 | |
739 | while ((bio = bio_list_pop(&nosync))) { | |
929be8fc MP |
740 | if (unlikely(ms->leg_failure) && errors_handled(ms)) |
741 | hold_bio(ms, bio); | |
742 | else { | |
743 | map_bio(get_default_mirror(ms), bio); | |
744 | generic_make_request(bio); | |
745 | } | |
1da177e4 LT |
746 | } |
747 | } | |
748 | ||
72f4b314 JB |
749 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
750 | { | |
751 | struct bio *bio; | |
752 | ||
0f398a84 | 753 | if (likely(!failures->head)) |
72f4b314 JB |
754 | return; |
755 | ||
b80aa7a0 JB |
756 | /* |
757 | * If the log has failed, unattempted writes are being | |
0f398a84 | 758 | * put on the holds list. We can't issue those writes |
b80aa7a0 JB |
759 | * until a log has been marked, so we must store them. |
760 | * | |
761 | * If a 'noflush' suspend is in progress, we can requeue | |
762 | * the I/O's to the core. This give userspace a chance | |
763 | * to reconfigure the mirror, at which point the core | |
764 | * will reissue the writes. If the 'noflush' flag is | |
765 | * not set, we have no choice but to return errors. | |
766 | * | |
767 | * Some writes on the failures list may have been | |
768 | * submitted before the log failure and represent a | |
769 | * failure to write to one of the devices. It is ok | |
770 | * for us to treat them the same and requeue them | |
771 | * as well. | |
772 | */ | |
0f398a84 | 773 | while ((bio = bio_list_pop(failures))) { |
60f355ea | 774 | if (!ms->log_failure) { |
0f398a84 | 775 | ms->in_sync = 0; |
c58098be | 776 | dm_rh_mark_nosync(ms->rh, bio); |
0f398a84 | 777 | } |
60f355ea MP |
778 | |
779 | /* | |
780 | * If all the legs are dead, fail the I/O. | |
781 | * If we have been told to handle errors, hold the bio | |
782 | * and wait for userspace to deal with the problem. | |
783 | * Otherwise pretend that the I/O succeeded. (This would | |
784 | * be wrong if the failed leg returned after reboot and | |
785 | * got replicated back to the good legs.) | |
786 | */ | |
787 | if (!get_valid_mirror(ms)) | |
788 | bio_endio(bio, -EIO); | |
789 | else if (errors_handled(ms)) | |
790 | hold_bio(ms, bio); | |
791 | else | |
792 | bio_endio(bio, 0); | |
b80aa7a0 | 793 | } |
72f4b314 JB |
794 | } |
795 | ||
796 | static void trigger_event(struct work_struct *work) | |
797 | { | |
798 | struct mirror_set *ms = | |
799 | container_of(work, struct mirror_set, trigger_event); | |
800 | ||
801 | dm_table_event(ms->ti->table); | |
802 | } | |
803 | ||
1da177e4 LT |
804 | /*----------------------------------------------------------------- |
805 | * kmirrord | |
806 | *---------------------------------------------------------------*/ | |
a2aebe03 | 807 | static void do_mirror(struct work_struct *work) |
1da177e4 | 808 | { |
1f965b19 HM |
809 | struct mirror_set *ms = container_of(work, struct mirror_set, |
810 | kmirrord_work); | |
72f4b314 JB |
811 | struct bio_list reads, writes, failures; |
812 | unsigned long flags; | |
1da177e4 | 813 | |
72f4b314 | 814 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
815 | reads = ms->reads; |
816 | writes = ms->writes; | |
72f4b314 | 817 | failures = ms->failures; |
1da177e4 LT |
818 | bio_list_init(&ms->reads); |
819 | bio_list_init(&ms->writes); | |
72f4b314 JB |
820 | bio_list_init(&ms->failures); |
821 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 | 822 | |
1f965b19 | 823 | dm_rh_update_states(ms->rh, errors_handled(ms)); |
1da177e4 LT |
824 | do_recovery(ms); |
825 | do_reads(ms, &reads); | |
826 | do_writes(ms, &writes); | |
72f4b314 | 827 | do_failures(ms, &failures); |
7ff14a36 MP |
828 | |
829 | dm_table_unplug_all(ms->ti->table); | |
1da177e4 LT |
830 | } |
831 | ||
1da177e4 LT |
832 | /*----------------------------------------------------------------- |
833 | * Target functions | |
834 | *---------------------------------------------------------------*/ | |
835 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
836 | uint32_t region_size, | |
837 | struct dm_target *ti, | |
416cd17b | 838 | struct dm_dirty_log *dl) |
1da177e4 LT |
839 | { |
840 | size_t len; | |
841 | struct mirror_set *ms = NULL; | |
842 | ||
1da177e4 LT |
843 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
844 | ||
dd00cc48 | 845 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 846 | if (!ms) { |
72d94861 | 847 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
848 | return NULL; |
849 | } | |
850 | ||
1da177e4 | 851 | spin_lock_init(&ms->lock); |
5339fc2d MP |
852 | bio_list_init(&ms->reads); |
853 | bio_list_init(&ms->writes); | |
854 | bio_list_init(&ms->failures); | |
855 | bio_list_init(&ms->holds); | |
1da177e4 LT |
856 | |
857 | ms->ti = ti; | |
858 | ms->nr_mirrors = nr_mirrors; | |
859 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
860 | ms->in_sync = 0; | |
b80aa7a0 | 861 | ms->log_failure = 0; |
929be8fc | 862 | ms->leg_failure = 0; |
b80aa7a0 | 863 | atomic_set(&ms->suspend, 0); |
72f4b314 | 864 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 865 | |
95f8fac8 MP |
866 | ms->read_record_pool = mempool_create_slab_pool(MIN_READ_RECORDS, |
867 | _dm_raid1_read_record_cache); | |
868 | ||
06386bbf JB |
869 | if (!ms->read_record_pool) { |
870 | ti->error = "Error creating mirror read_record_pool"; | |
871 | kfree(ms); | |
872 | return NULL; | |
873 | } | |
874 | ||
88be163a MB |
875 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
876 | if (IS_ERR(ms->io_client)) { | |
877 | ti->error = "Error creating dm_io client"; | |
06386bbf | 878 | mempool_destroy(ms->read_record_pool); |
88be163a MB |
879 | kfree(ms); |
880 | return NULL; | |
881 | } | |
882 | ||
1f965b19 HM |
883 | ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord, |
884 | wakeup_all_recovery_waiters, | |
885 | ms->ti->begin, MAX_RECOVERY, | |
886 | dl, region_size, ms->nr_regions); | |
887 | if (IS_ERR(ms->rh)) { | |
72d94861 | 888 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 889 | dm_io_client_destroy(ms->io_client); |
06386bbf | 890 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
891 | kfree(ms); |
892 | return NULL; | |
893 | } | |
894 | ||
895 | return ms; | |
896 | } | |
897 | ||
898 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
899 | unsigned int m) | |
900 | { | |
901 | while (m--) | |
902 | dm_put_device(ti, ms->mirror[m].dev); | |
903 | ||
88be163a | 904 | dm_io_client_destroy(ms->io_client); |
1f965b19 | 905 | dm_region_hash_destroy(ms->rh); |
06386bbf | 906 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
907 | kfree(ms); |
908 | } | |
909 | ||
1da177e4 LT |
910 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
911 | unsigned int mirror, char **argv) | |
912 | { | |
4ee218cd | 913 | unsigned long long offset; |
1da177e4 | 914 | |
4ee218cd | 915 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 916 | ti->error = "Invalid offset"; |
1da177e4 LT |
917 | return -EINVAL; |
918 | } | |
919 | ||
920 | if (dm_get_device(ti, argv[0], offset, ti->len, | |
921 | dm_table_get_mode(ti->table), | |
922 | &ms->mirror[mirror].dev)) { | |
72d94861 | 923 | ti->error = "Device lookup failure"; |
1da177e4 LT |
924 | return -ENXIO; |
925 | } | |
926 | ||
aa5617c5 | 927 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
928 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
929 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
930 | ms->mirror[mirror].offset = offset; |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
1da177e4 LT |
935 | /* |
936 | * Create dirty log: log_type #log_params <log_params> | |
937 | */ | |
416cd17b | 938 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1f965b19 HM |
939 | unsigned argc, char **argv, |
940 | unsigned *args_used) | |
1da177e4 | 941 | { |
1f965b19 | 942 | unsigned param_count; |
416cd17b | 943 | struct dm_dirty_log *dl; |
1da177e4 LT |
944 | |
945 | if (argc < 2) { | |
72d94861 | 946 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
947 | return NULL; |
948 | } | |
949 | ||
950 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 951 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
952 | return NULL; |
953 | } | |
954 | ||
955 | *args_used = 2 + param_count; | |
956 | ||
957 | if (argc < *args_used) { | |
72d94861 | 958 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
959 | return NULL; |
960 | } | |
961 | ||
c0da3748 MP |
962 | dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count, |
963 | argv + 2); | |
1da177e4 | 964 | if (!dl) { |
72d94861 | 965 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
966 | return NULL; |
967 | } | |
968 | ||
1da177e4 LT |
969 | return dl; |
970 | } | |
971 | ||
a8e6afa2 JB |
972 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
973 | unsigned *args_used) | |
974 | { | |
975 | unsigned num_features; | |
976 | struct dm_target *ti = ms->ti; | |
977 | ||
978 | *args_used = 0; | |
979 | ||
980 | if (!argc) | |
981 | return 0; | |
982 | ||
983 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
984 | ti->error = "Invalid number of features"; | |
985 | return -EINVAL; | |
986 | } | |
987 | ||
988 | argc--; | |
989 | argv++; | |
990 | (*args_used)++; | |
991 | ||
992 | if (num_features > argc) { | |
993 | ti->error = "Not enough arguments to support feature count"; | |
994 | return -EINVAL; | |
995 | } | |
996 | ||
997 | if (!strcmp("handle_errors", argv[0])) | |
998 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
999 | else { | |
1000 | ti->error = "Unrecognised feature requested"; | |
1001 | return -EINVAL; | |
1002 | } | |
1003 | ||
1004 | (*args_used)++; | |
1005 | ||
1006 | return 0; | |
1007 | } | |
1008 | ||
1da177e4 LT |
1009 | /* |
1010 | * Construct a mirror mapping: | |
1011 | * | |
1012 | * log_type #log_params <log_params> | |
1013 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1014 | * [#features <features>] |
1da177e4 LT |
1015 | * |
1016 | * log_type is "core" or "disk" | |
1017 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1018 | * |
1019 | * If present, features must be "handle_errors". | |
1da177e4 | 1020 | */ |
1da177e4 LT |
1021 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1022 | { | |
1023 | int r; | |
1024 | unsigned int nr_mirrors, m, args_used; | |
1025 | struct mirror_set *ms; | |
416cd17b | 1026 | struct dm_dirty_log *dl; |
1da177e4 LT |
1027 | |
1028 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
1029 | if (!dl) | |
1030 | return -EINVAL; | |
1031 | ||
1032 | argv += args_used; | |
1033 | argc -= args_used; | |
1034 | ||
1035 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
eb69aca5 | 1036 | nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) { |
72d94861 | 1037 | ti->error = "Invalid number of mirrors"; |
416cd17b | 1038 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1039 | return -EINVAL; |
1040 | } | |
1041 | ||
1042 | argv++, argc--; | |
1043 | ||
a8e6afa2 JB |
1044 | if (argc < nr_mirrors * 2) { |
1045 | ti->error = "Too few mirror arguments"; | |
416cd17b | 1046 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1047 | return -EINVAL; |
1048 | } | |
1049 | ||
1050 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
1051 | if (!ms) { | |
416cd17b | 1052 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1053 | return -ENOMEM; |
1054 | } | |
1055 | ||
1056 | /* Get the mirror parameter sets */ | |
1057 | for (m = 0; m < nr_mirrors; m++) { | |
1058 | r = get_mirror(ms, ti, m, argv); | |
1059 | if (r) { | |
1060 | free_context(ms, ti, m); | |
1061 | return r; | |
1062 | } | |
1063 | argv += 2; | |
1064 | argc -= 2; | |
1065 | } | |
1066 | ||
1067 | ti->private = ms; | |
1f965b19 | 1068 | ti->split_io = dm_rh_get_region_size(ms->rh); |
4184153f | 1069 | ti->num_flush_requests = 1; |
1da177e4 | 1070 | |
6ad36fe2 HS |
1071 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
1072 | if (!ms->kmirrord_wq) { | |
1073 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1074 | r = -ENOMEM; |
1075 | goto err_free_context; | |
6ad36fe2 HS |
1076 | } |
1077 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
1078 | init_timer(&ms->timer); |
1079 | ms->timer_pending = 0; | |
72f4b314 | 1080 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 1081 | |
a8e6afa2 | 1082 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1083 | if (r) |
1084 | goto err_destroy_wq; | |
a8e6afa2 JB |
1085 | |
1086 | argv += args_used; | |
1087 | argc -= args_used; | |
1088 | ||
f44db678 JB |
1089 | /* |
1090 | * Any read-balancing addition depends on the | |
1091 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1092 | * This is because the decision to balance depends | |
1093 | * on the sync state of a region. If the above | |
1094 | * flag is not present, we ignore errors; and | |
1095 | * the sync state may be inaccurate. | |
1096 | */ | |
1097 | ||
a8e6afa2 JB |
1098 | if (argc) { |
1099 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1100 | r = -EINVAL; |
1101 | goto err_destroy_wq; | |
a8e6afa2 JB |
1102 | } |
1103 | ||
1f965b19 | 1104 | r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client); |
a72cf737 DM |
1105 | if (r) |
1106 | goto err_destroy_wq; | |
1da177e4 | 1107 | |
1f965b19 | 1108 | wakeup_mirrord(ms); |
1da177e4 | 1109 | return 0; |
a72cf737 DM |
1110 | |
1111 | err_destroy_wq: | |
1112 | destroy_workqueue(ms->kmirrord_wq); | |
1113 | err_free_context: | |
1114 | free_context(ms, ti, ms->nr_mirrors); | |
1115 | return r; | |
1da177e4 LT |
1116 | } |
1117 | ||
1118 | static void mirror_dtr(struct dm_target *ti) | |
1119 | { | |
1120 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1121 | ||
a2aebe03 | 1122 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1123 | flush_workqueue(ms->kmirrord_wq); |
18776c73 | 1124 | flush_scheduled_work(); |
eb69aca5 | 1125 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1126 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1127 | free_context(ms, ti, ms->nr_mirrors); |
1128 | } | |
1129 | ||
1da177e4 LT |
1130 | /* |
1131 | * Mirror mapping function | |
1132 | */ | |
1133 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1134 | union map_info *map_context) | |
1135 | { | |
1136 | int r, rw = bio_rw(bio); | |
1137 | struct mirror *m; | |
1138 | struct mirror_set *ms = ti->private; | |
06386bbf | 1139 | struct dm_raid1_read_record *read_record = NULL; |
1f965b19 | 1140 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 LT |
1141 | |
1142 | if (rw == WRITE) { | |
06386bbf | 1143 | /* Save region for mirror_end_io() handler */ |
1f965b19 | 1144 | map_context->ll = dm_rh_bio_to_region(ms->rh, bio); |
1da177e4 | 1145 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1146 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1147 | } |
1148 | ||
1f965b19 | 1149 | r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0); |
1da177e4 LT |
1150 | if (r < 0 && r != -EWOULDBLOCK) |
1151 | return r; | |
1152 | ||
1da177e4 | 1153 | /* |
06386bbf | 1154 | * If region is not in-sync queue the bio. |
1da177e4 | 1155 | */ |
06386bbf JB |
1156 | if (!r || (r == -EWOULDBLOCK)) { |
1157 | if (rw == READA) | |
1158 | return -EWOULDBLOCK; | |
1da177e4 | 1159 | |
1da177e4 | 1160 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1161 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1162 | } |
1163 | ||
06386bbf JB |
1164 | /* |
1165 | * The region is in-sync and we can perform reads directly. | |
1166 | * Store enough information so we can retry if it fails. | |
1167 | */ | |
1da177e4 | 1168 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1169 | if (unlikely(!m)) |
1da177e4 LT |
1170 | return -EIO; |
1171 | ||
06386bbf JB |
1172 | read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO); |
1173 | if (likely(read_record)) { | |
1174 | dm_bio_record(&read_record->details, bio); | |
1175 | map_context->ptr = read_record; | |
1176 | read_record->m = m; | |
1177 | } | |
1178 | ||
1179 | map_bio(m, bio); | |
1180 | ||
d2a7ad29 | 1181 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1182 | } |
1183 | ||
1184 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1185 | int error, union map_info *map_context) | |
1186 | { | |
1187 | int rw = bio_rw(bio); | |
1188 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1189 | struct mirror *m = NULL; |
1190 | struct dm_bio_details *bd = NULL; | |
1191 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1192 | |
1193 | /* | |
1194 | * We need to dec pending if this was a write. | |
1195 | */ | |
06386bbf | 1196 | if (rw == WRITE) { |
4184153f MP |
1197 | if (likely(!bio_empty_barrier(bio))) |
1198 | dm_rh_dec(ms->rh, map_context->ll); | |
06386bbf JB |
1199 | return error; |
1200 | } | |
1da177e4 | 1201 | |
06386bbf JB |
1202 | if (error == -EOPNOTSUPP) |
1203 | goto out; | |
1204 | ||
1f98a13f | 1205 | if ((error == -EWOULDBLOCK) && bio_rw_flagged(bio, BIO_RW_AHEAD)) |
06386bbf JB |
1206 | goto out; |
1207 | ||
1208 | if (unlikely(error)) { | |
1209 | if (!read_record) { | |
1210 | /* | |
1211 | * There wasn't enough memory to record necessary | |
1212 | * information for a retry or there was no other | |
1213 | * mirror in-sync. | |
1214 | */ | |
e03f1a84 | 1215 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1216 | return -EIO; |
1217 | } | |
e03f1a84 AB |
1218 | |
1219 | m = read_record->m; | |
1220 | ||
06386bbf JB |
1221 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1222 | m->dev->name); | |
1223 | ||
06386bbf JB |
1224 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1225 | ||
1226 | /* | |
1227 | * A failed read is requeued for another attempt using an intact | |
1228 | * mirror. | |
1229 | */ | |
1230 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1231 | bd = &read_record->details; | |
1232 | ||
1233 | dm_bio_restore(bd, bio); | |
1234 | mempool_free(read_record, ms->read_record_pool); | |
1235 | map_context->ptr = NULL; | |
1236 | queue_bio(ms, bio, rw); | |
1237 | return 1; | |
1238 | } | |
1239 | DMERR("All replicated volumes dead, failing I/O"); | |
1240 | } | |
1241 | ||
1242 | out: | |
1243 | if (read_record) { | |
1244 | mempool_free(read_record, ms->read_record_pool); | |
1245 | map_context->ptr = NULL; | |
1246 | } | |
1247 | ||
1248 | return error; | |
1da177e4 LT |
1249 | } |
1250 | ||
b80aa7a0 | 1251 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1252 | { |
1253 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1f965b19 | 1254 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 | 1255 | |
04788507 MP |
1256 | struct bio_list holds; |
1257 | struct bio *bio; | |
1258 | ||
b80aa7a0 JB |
1259 | atomic_set(&ms->suspend, 1); |
1260 | ||
1261 | /* | |
1262 | * We must finish up all the work that we've | |
1263 | * generated (i.e. recovery work). | |
1264 | */ | |
1f965b19 | 1265 | dm_rh_stop_recovery(ms->rh); |
33184048 | 1266 | |
33184048 | 1267 | wait_event(_kmirrord_recovery_stopped, |
1f965b19 | 1268 | !dm_rh_recovery_in_flight(ms->rh)); |
33184048 | 1269 | |
b80aa7a0 JB |
1270 | if (log->type->presuspend && log->type->presuspend(log)) |
1271 | /* FIXME: need better error handling */ | |
1272 | DMWARN("log presuspend failed"); | |
1273 | ||
1274 | /* | |
1275 | * Now that recovery is complete/stopped and the | |
1276 | * delayed bios are queued, we need to wait for | |
1277 | * the worker thread to complete. This way, | |
1278 | * we know that all of our I/O has been pushed. | |
1279 | */ | |
1280 | flush_workqueue(ms->kmirrord_wq); | |
04788507 MP |
1281 | |
1282 | /* | |
1283 | * Now set ms->suspend is set and the workqueue flushed, no more | |
1284 | * entries can be added to ms->hold list, so process it. | |
1285 | * | |
1286 | * Bios can still arrive concurrently with or after this | |
1287 | * presuspend function, but they cannot join the hold list | |
1288 | * because ms->suspend is set. | |
1289 | */ | |
1290 | spin_lock_irq(&ms->lock); | |
1291 | holds = ms->holds; | |
1292 | bio_list_init(&ms->holds); | |
1293 | spin_unlock_irq(&ms->lock); | |
1294 | ||
1295 | while ((bio = bio_list_pop(&holds))) | |
1296 | hold_bio(ms, bio); | |
b80aa7a0 JB |
1297 | } |
1298 | ||
1299 | static void mirror_postsuspend(struct dm_target *ti) | |
1300 | { | |
1301 | struct mirror_set *ms = ti->private; | |
1f965b19 | 1302 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 | 1303 | |
6b3df0d7 | 1304 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1305 | /* FIXME: need better error handling */ |
b80aa7a0 | 1306 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1307 | } |
1308 | ||
1309 | static void mirror_resume(struct dm_target *ti) | |
1310 | { | |
b80aa7a0 | 1311 | struct mirror_set *ms = ti->private; |
1f965b19 | 1312 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 JB |
1313 | |
1314 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1315 | if (log->type->resume && log->type->resume(log)) |
1316 | /* FIXME: need better error handling */ | |
1317 | DMWARN("log resume failed"); | |
1f965b19 | 1318 | dm_rh_start_recovery(ms->rh); |
1da177e4 LT |
1319 | } |
1320 | ||
af195ac8 JB |
1321 | /* |
1322 | * device_status_char | |
1323 | * @m: mirror device/leg we want the status of | |
1324 | * | |
1325 | * We return one character representing the most severe error | |
1326 | * we have encountered. | |
1327 | * A => Alive - No failures | |
1328 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1329 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1330 | * R => Read - A read failure occurred, mirror data unaffected | |
1331 | * | |
1332 | * Returns: <char> | |
1333 | */ | |
1334 | static char device_status_char(struct mirror *m) | |
1335 | { | |
1336 | if (!atomic_read(&(m->error_count))) | |
1337 | return 'A'; | |
1338 | ||
64b30c46 MP |
1339 | return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' : |
1340 | (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
af195ac8 JB |
1341 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : |
1342 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1343 | } | |
1344 | ||
1345 | ||
1da177e4 LT |
1346 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1347 | char *result, unsigned int maxlen) | |
1348 | { | |
315dcc22 | 1349 | unsigned int m, sz = 0; |
1da177e4 | 1350 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1f965b19 | 1351 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
af195ac8 | 1352 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1353 | |
1da177e4 LT |
1354 | switch (type) { |
1355 | case STATUSTYPE_INFO: | |
1356 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1357 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1358 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1359 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1360 | } | |
1361 | buffer[m] = '\0'; | |
1da177e4 | 1362 | |
af195ac8 | 1363 | DMEMIT("%llu/%llu 1 %s ", |
1f965b19 | 1364 | (unsigned long long)log->type->get_sync_count(log), |
af195ac8 | 1365 | (unsigned long long)ms->nr_regions, buffer); |
315dcc22 | 1366 | |
1f965b19 | 1367 | sz += log->type->status(log, type, result+sz, maxlen-sz); |
315dcc22 | 1368 | |
1da177e4 LT |
1369 | break; |
1370 | ||
1371 | case STATUSTYPE_TABLE: | |
1f965b19 | 1372 | sz = log->type->status(log, type, result, maxlen); |
315dcc22 | 1373 | |
e52b8f6d | 1374 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1375 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1376 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1377 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1378 | |
1379 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1380 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1381 | } |
1382 | ||
1383 | return 0; | |
1384 | } | |
1385 | ||
af4874e0 MS |
1386 | static int mirror_iterate_devices(struct dm_target *ti, |
1387 | iterate_devices_callout_fn fn, void *data) | |
1388 | { | |
1389 | struct mirror_set *ms = ti->private; | |
1390 | int ret = 0; | |
1391 | unsigned i; | |
1392 | ||
1393 | for (i = 0; !ret && i < ms->nr_mirrors; i++) | |
1394 | ret = fn(ti, ms->mirror[i].dev, | |
5dea271b | 1395 | ms->mirror[i].offset, ti->len, data); |
af4874e0 MS |
1396 | |
1397 | return ret; | |
1398 | } | |
1399 | ||
1da177e4 LT |
1400 | static struct target_type mirror_target = { |
1401 | .name = "mirror", | |
af4874e0 | 1402 | .version = {1, 12, 0}, |
1da177e4 LT |
1403 | .module = THIS_MODULE, |
1404 | .ctr = mirror_ctr, | |
1405 | .dtr = mirror_dtr, | |
1406 | .map = mirror_map, | |
1407 | .end_io = mirror_end_io, | |
b80aa7a0 | 1408 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1409 | .postsuspend = mirror_postsuspend, |
1410 | .resume = mirror_resume, | |
1411 | .status = mirror_status, | |
af4874e0 | 1412 | .iterate_devices = mirror_iterate_devices, |
1da177e4 LT |
1413 | }; |
1414 | ||
1415 | static int __init dm_mirror_init(void) | |
1416 | { | |
1417 | int r; | |
1418 | ||
95f8fac8 MP |
1419 | _dm_raid1_read_record_cache = KMEM_CACHE(dm_raid1_read_record, 0); |
1420 | if (!_dm_raid1_read_record_cache) { | |
1421 | DMERR("Can't allocate dm_raid1_read_record cache"); | |
1422 | r = -ENOMEM; | |
1423 | goto bad_cache; | |
1424 | } | |
1425 | ||
1da177e4 | 1426 | r = dm_register_target(&mirror_target); |
95f8fac8 | 1427 | if (r < 0) { |
0cd33124 | 1428 | DMERR("Failed to register mirror target"); |
95f8fac8 MP |
1429 | goto bad_target; |
1430 | } | |
1431 | ||
1432 | return 0; | |
1da177e4 | 1433 | |
95f8fac8 MP |
1434 | bad_target: |
1435 | kmem_cache_destroy(_dm_raid1_read_record_cache); | |
1436 | bad_cache: | |
1da177e4 LT |
1437 | return r; |
1438 | } | |
1439 | ||
1440 | static void __exit dm_mirror_exit(void) | |
1441 | { | |
10d3bd09 | 1442 | dm_unregister_target(&mirror_target); |
95f8fac8 | 1443 | kmem_cache_destroy(_dm_raid1_read_record_cache); |
1da177e4 LT |
1444 | } |
1445 | ||
1446 | /* Module hooks */ | |
1447 | module_init(dm_mirror_init); | |
1448 | module_exit(dm_mirror_exit); | |
1449 | ||
1450 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1451 | MODULE_AUTHOR("Joe Thornber"); | |
1452 | MODULE_LICENSE("GPL"); |