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. */ | |
72d94861 | 25 | |
a8e6afa2 | 26 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
f44db678 | 27 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
a8e6afa2 | 28 | |
33184048 | 29 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
1da177e4 | 30 | |
e4c8b3ba NB |
31 | /*----------------------------------------------------------------- |
32 | * Mirror set structures. | |
33 | *---------------------------------------------------------------*/ | |
72f4b314 JB |
34 | enum dm_raid1_error { |
35 | DM_RAID1_WRITE_ERROR, | |
64b30c46 | 36 | DM_RAID1_FLUSH_ERROR, |
72f4b314 JB |
37 | DM_RAID1_SYNC_ERROR, |
38 | DM_RAID1_READ_ERROR | |
39 | }; | |
40 | ||
e4c8b3ba | 41 | struct mirror { |
aa5617c5 | 42 | struct mirror_set *ms; |
e4c8b3ba | 43 | atomic_t error_count; |
39ed7adb | 44 | unsigned long error_type; |
e4c8b3ba NB |
45 | struct dm_dev *dev; |
46 | sector_t offset; | |
47 | }; | |
48 | ||
49 | struct mirror_set { | |
50 | struct dm_target *ti; | |
51 | struct list_head list; | |
1f965b19 | 52 | |
a8e6afa2 | 53 | uint64_t features; |
e4c8b3ba | 54 | |
72f4b314 | 55 | spinlock_t lock; /* protects the lists */ |
e4c8b3ba NB |
56 | struct bio_list reads; |
57 | struct bio_list writes; | |
72f4b314 | 58 | struct bio_list failures; |
04788507 | 59 | struct bio_list holds; /* bios are waiting until suspend */ |
e4c8b3ba | 60 | |
1f965b19 HM |
61 | struct dm_region_hash *rh; |
62 | struct dm_kcopyd_client *kcopyd_client; | |
88be163a | 63 | struct dm_io_client *io_client; |
06386bbf | 64 | mempool_t *read_record_pool; |
88be163a | 65 | |
e4c8b3ba NB |
66 | /* recovery */ |
67 | region_t nr_regions; | |
68 | int in_sync; | |
fc1ff958 | 69 | int log_failure; |
929be8fc | 70 | int leg_failure; |
b80aa7a0 | 71 | atomic_t suspend; |
e4c8b3ba | 72 | |
72f4b314 | 73 | atomic_t default_mirror; /* Default mirror */ |
e4c8b3ba | 74 | |
6ad36fe2 HS |
75 | struct workqueue_struct *kmirrord_wq; |
76 | struct work_struct kmirrord_work; | |
a2aebe03 MP |
77 | struct timer_list timer; |
78 | unsigned long timer_pending; | |
79 | ||
72f4b314 | 80 | struct work_struct trigger_event; |
6ad36fe2 | 81 | |
1f965b19 | 82 | unsigned nr_mirrors; |
e4c8b3ba NB |
83 | struct mirror mirror[0]; |
84 | }; | |
85 | ||
1f965b19 | 86 | static void wakeup_mirrord(void *context) |
1da177e4 | 87 | { |
1f965b19 | 88 | struct mirror_set *ms = context; |
1da177e4 | 89 | |
6ad36fe2 HS |
90 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); |
91 | } | |
92 | ||
a2aebe03 MP |
93 | static void delayed_wake_fn(unsigned long data) |
94 | { | |
95 | struct mirror_set *ms = (struct mirror_set *) data; | |
96 | ||
97 | clear_bit(0, &ms->timer_pending); | |
1f965b19 | 98 | wakeup_mirrord(ms); |
a2aebe03 MP |
99 | } |
100 | ||
101 | static void delayed_wake(struct mirror_set *ms) | |
102 | { | |
103 | if (test_and_set_bit(0, &ms->timer_pending)) | |
104 | return; | |
105 | ||
106 | ms->timer.expires = jiffies + HZ / 5; | |
107 | ms->timer.data = (unsigned long) ms; | |
108 | ms->timer.function = delayed_wake_fn; | |
109 | add_timer(&ms->timer); | |
110 | } | |
111 | ||
1f965b19 | 112 | static void wakeup_all_recovery_waiters(void *context) |
1da177e4 | 113 | { |
1f965b19 | 114 | wake_up_all(&_kmirrord_recovery_stopped); |
1da177e4 LT |
115 | } |
116 | ||
1f965b19 | 117 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) |
1da177e4 LT |
118 | { |
119 | unsigned long flags; | |
1da177e4 | 120 | int should_wake = 0; |
1f965b19 | 121 | struct bio_list *bl; |
1da177e4 | 122 | |
1f965b19 HM |
123 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; |
124 | spin_lock_irqsave(&ms->lock, flags); | |
125 | should_wake = !(bl->head); | |
126 | bio_list_add(bl, bio); | |
127 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 LT |
128 | |
129 | if (should_wake) | |
1f965b19 | 130 | wakeup_mirrord(ms); |
1da177e4 LT |
131 | } |
132 | ||
1f965b19 | 133 | static void dispatch_bios(void *context, struct bio_list *bio_list) |
1da177e4 | 134 | { |
1f965b19 HM |
135 | struct mirror_set *ms = context; |
136 | struct bio *bio; | |
1da177e4 | 137 | |
1f965b19 HM |
138 | while ((bio = bio_list_pop(bio_list))) |
139 | queue_bio(ms, bio, WRITE); | |
1da177e4 LT |
140 | } |
141 | ||
06386bbf JB |
142 | #define MIN_READ_RECORDS 20 |
143 | struct dm_raid1_read_record { | |
144 | struct mirror *m; | |
145 | struct dm_bio_details details; | |
146 | }; | |
147 | ||
95f8fac8 MP |
148 | static struct kmem_cache *_dm_raid1_read_record_cache; |
149 | ||
1da177e4 LT |
150 | /* |
151 | * Every mirror should look like this one. | |
152 | */ | |
153 | #define DEFAULT_MIRROR 0 | |
154 | ||
155 | /* | |
06386bbf JB |
156 | * This is yucky. We squirrel the mirror struct away inside |
157 | * bi_next for read/write buffers. This is safe since the bh | |
1da177e4 LT |
158 | * doesn't get submitted to the lower levels of block layer. |
159 | */ | |
06386bbf | 160 | static struct mirror *bio_get_m(struct bio *bio) |
1da177e4 | 161 | { |
06386bbf | 162 | return (struct mirror *) bio->bi_next; |
1da177e4 LT |
163 | } |
164 | ||
06386bbf | 165 | static void bio_set_m(struct bio *bio, struct mirror *m) |
1da177e4 | 166 | { |
06386bbf | 167 | bio->bi_next = (struct bio *) m; |
1da177e4 LT |
168 | } |
169 | ||
72f4b314 JB |
170 | static struct mirror *get_default_mirror(struct mirror_set *ms) |
171 | { | |
172 | return &ms->mirror[atomic_read(&ms->default_mirror)]; | |
173 | } | |
174 | ||
175 | static void set_default_mirror(struct mirror *m) | |
176 | { | |
177 | struct mirror_set *ms = m->ms; | |
178 | struct mirror *m0 = &(ms->mirror[0]); | |
179 | ||
180 | atomic_set(&ms->default_mirror, m - m0); | |
181 | } | |
182 | ||
87968ddd MP |
183 | static struct mirror *get_valid_mirror(struct mirror_set *ms) |
184 | { | |
185 | struct mirror *m; | |
186 | ||
187 | for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++) | |
188 | if (!atomic_read(&m->error_count)) | |
189 | return m; | |
190 | ||
191 | return NULL; | |
192 | } | |
193 | ||
72f4b314 JB |
194 | /* fail_mirror |
195 | * @m: mirror device to fail | |
196 | * @error_type: one of the enum's, DM_RAID1_*_ERROR | |
197 | * | |
198 | * If errors are being handled, record the type of | |
199 | * error encountered for this device. If this type | |
200 | * of error has already been recorded, we can return; | |
201 | * otherwise, we must signal userspace by triggering | |
202 | * an event. Additionally, if the device is the | |
203 | * primary device, we must choose a new primary, but | |
204 | * only if the mirror is in-sync. | |
205 | * | |
206 | * This function must not block. | |
207 | */ | |
208 | static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) | |
209 | { | |
210 | struct mirror_set *ms = m->ms; | |
211 | struct mirror *new; | |
212 | ||
929be8fc MP |
213 | ms->leg_failure = 1; |
214 | ||
72f4b314 JB |
215 | /* |
216 | * error_count is used for nothing more than a | |
217 | * simple way to tell if a device has encountered | |
218 | * errors. | |
219 | */ | |
220 | atomic_inc(&m->error_count); | |
221 | ||
222 | if (test_and_set_bit(error_type, &m->error_type)) | |
223 | return; | |
224 | ||
d460c65a JB |
225 | if (!errors_handled(ms)) |
226 | return; | |
227 | ||
72f4b314 JB |
228 | if (m != get_default_mirror(ms)) |
229 | goto out; | |
230 | ||
231 | if (!ms->in_sync) { | |
232 | /* | |
233 | * Better to issue requests to same failing device | |
234 | * than to risk returning corrupt data. | |
235 | */ | |
236 | DMERR("Primary mirror (%s) failed while out-of-sync: " | |
237 | "Reads may fail.", m->dev->name); | |
238 | goto out; | |
239 | } | |
240 | ||
87968ddd MP |
241 | new = get_valid_mirror(ms); |
242 | if (new) | |
243 | set_default_mirror(new); | |
244 | else | |
72f4b314 JB |
245 | DMWARN("All sides of mirror have failed."); |
246 | ||
247 | out: | |
248 | schedule_work(&ms->trigger_event); | |
249 | } | |
250 | ||
c0da3748 MP |
251 | static int mirror_flush(struct dm_target *ti) |
252 | { | |
253 | struct mirror_set *ms = ti->private; | |
254 | unsigned long error_bits; | |
255 | ||
256 | unsigned int i; | |
257 | struct dm_io_region io[ms->nr_mirrors]; | |
258 | struct mirror *m; | |
259 | struct dm_io_request io_req = { | |
d87f4c14 | 260 | .bi_rw = WRITE_FLUSH, |
c0da3748 | 261 | .mem.type = DM_IO_KMEM, |
5fc2ffea | 262 | .mem.ptr.addr = NULL, |
c0da3748 MP |
263 | .client = ms->io_client, |
264 | }; | |
265 | ||
266 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) { | |
267 | io[i].bdev = m->dev->bdev; | |
268 | io[i].sector = 0; | |
269 | io[i].count = 0; | |
270 | } | |
271 | ||
272 | error_bits = -1; | |
273 | dm_io(&io_req, ms->nr_mirrors, io, &error_bits); | |
274 | if (unlikely(error_bits != 0)) { | |
275 | for (i = 0; i < ms->nr_mirrors; i++) | |
276 | if (test_bit(i, &error_bits)) | |
277 | fail_mirror(ms->mirror + i, | |
64b30c46 | 278 | DM_RAID1_FLUSH_ERROR); |
c0da3748 MP |
279 | return -EIO; |
280 | } | |
281 | ||
282 | return 0; | |
283 | } | |
284 | ||
1da177e4 LT |
285 | /*----------------------------------------------------------------- |
286 | * Recovery. | |
287 | * | |
288 | * When a mirror is first activated we may find that some regions | |
289 | * are in the no-sync state. We have to recover these by | |
290 | * recopying from the default mirror to all the others. | |
291 | *---------------------------------------------------------------*/ | |
4cdc1d1f | 292 | static void recovery_complete(int read_err, unsigned long write_err, |
1da177e4 LT |
293 | void *context) |
294 | { | |
1f965b19 HM |
295 | struct dm_region *reg = context; |
296 | struct mirror_set *ms = dm_rh_region_context(reg); | |
8f0205b7 | 297 | int m, bit = 0; |
1da177e4 | 298 | |
8f0205b7 | 299 | if (read_err) { |
f44db678 JB |
300 | /* Read error means the failure of default mirror. */ |
301 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
8f0205b7 JB |
302 | fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); |
303 | } | |
f44db678 | 304 | |
8f0205b7 | 305 | if (write_err) { |
4cdc1d1f | 306 | DMERR_LIMIT("Write error during recovery (error = 0x%lx)", |
f44db678 | 307 | write_err); |
8f0205b7 JB |
308 | /* |
309 | * Bits correspond to devices (excluding default mirror). | |
310 | * The default mirror cannot change during recovery. | |
311 | */ | |
312 | for (m = 0; m < ms->nr_mirrors; m++) { | |
313 | if (&ms->mirror[m] == get_default_mirror(ms)) | |
314 | continue; | |
315 | if (test_bit(bit, &write_err)) | |
316 | fail_mirror(ms->mirror + m, | |
317 | DM_RAID1_SYNC_ERROR); | |
318 | bit++; | |
319 | } | |
320 | } | |
f44db678 | 321 | |
1f965b19 | 322 | dm_rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
323 | } |
324 | ||
1f965b19 | 325 | static int recover(struct mirror_set *ms, struct dm_region *reg) |
1da177e4 LT |
326 | { |
327 | int r; | |
1f965b19 | 328 | unsigned i; |
eb69aca5 | 329 | struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; |
1da177e4 LT |
330 | struct mirror *m; |
331 | unsigned long flags = 0; | |
1f965b19 HM |
332 | region_t key = dm_rh_get_region_key(reg); |
333 | sector_t region_size = dm_rh_get_region_size(ms->rh); | |
1da177e4 LT |
334 | |
335 | /* fill in the source */ | |
72f4b314 | 336 | m = get_default_mirror(ms); |
1da177e4 | 337 | from.bdev = m->dev->bdev; |
1f965b19 HM |
338 | from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
339 | if (key == (ms->nr_regions - 1)) { | |
1da177e4 LT |
340 | /* |
341 | * The final region may be smaller than | |
342 | * region_size. | |
343 | */ | |
1f965b19 | 344 | from.count = ms->ti->len & (region_size - 1); |
1da177e4 | 345 | if (!from.count) |
1f965b19 | 346 | from.count = region_size; |
1da177e4 | 347 | } else |
1f965b19 | 348 | from.count = region_size; |
1da177e4 LT |
349 | |
350 | /* fill in the destinations */ | |
351 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
72f4b314 | 352 | if (&ms->mirror[i] == get_default_mirror(ms)) |
1da177e4 LT |
353 | continue; |
354 | ||
355 | m = ms->mirror + i; | |
356 | dest->bdev = m->dev->bdev; | |
1f965b19 | 357 | dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
1da177e4 LT |
358 | dest->count = from.count; |
359 | dest++; | |
360 | } | |
361 | ||
362 | /* hand to kcopyd */ | |
f7c83e2e JB |
363 | if (!errors_handled(ms)) |
364 | set_bit(DM_KCOPYD_IGNORE_ERROR, &flags); | |
365 | ||
eb69aca5 HM |
366 | r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, |
367 | flags, recovery_complete, reg); | |
1da177e4 LT |
368 | |
369 | return r; | |
370 | } | |
371 | ||
372 | static void do_recovery(struct mirror_set *ms) | |
373 | { | |
1f965b19 HM |
374 | struct dm_region *reg; |
375 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
1da177e4 | 376 | int r; |
1da177e4 LT |
377 | |
378 | /* | |
379 | * Start quiescing some regions. | |
380 | */ | |
1f965b19 | 381 | dm_rh_recovery_prepare(ms->rh); |
1da177e4 LT |
382 | |
383 | /* | |
384 | * Copy any already quiesced regions. | |
385 | */ | |
1f965b19 | 386 | while ((reg = dm_rh_recovery_start(ms->rh))) { |
1da177e4 LT |
387 | r = recover(ms, reg); |
388 | if (r) | |
1f965b19 | 389 | dm_rh_recovery_end(reg, 0); |
1da177e4 LT |
390 | } |
391 | ||
392 | /* | |
393 | * Update the in sync flag. | |
394 | */ | |
395 | if (!ms->in_sync && | |
396 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
397 | /* the sync is complete */ | |
398 | dm_table_event(ms->ti->table); | |
399 | ms->in_sync = 1; | |
400 | } | |
401 | } | |
402 | ||
403 | /*----------------------------------------------------------------- | |
404 | * Reads | |
405 | *---------------------------------------------------------------*/ | |
406 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
407 | { | |
06386bbf JB |
408 | struct mirror *m = get_default_mirror(ms); |
409 | ||
410 | do { | |
411 | if (likely(!atomic_read(&m->error_count))) | |
412 | return m; | |
413 | ||
414 | if (m-- == ms->mirror) | |
415 | m += ms->nr_mirrors; | |
416 | } while (m != get_default_mirror(ms)); | |
417 | ||
418 | return NULL; | |
419 | } | |
420 | ||
421 | static int default_ok(struct mirror *m) | |
422 | { | |
423 | struct mirror *default_mirror = get_default_mirror(m->ms); | |
424 | ||
425 | return !atomic_read(&default_mirror->error_count); | |
426 | } | |
427 | ||
428 | static int mirror_available(struct mirror_set *ms, struct bio *bio) | |
429 | { | |
1f965b19 HM |
430 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
431 | region_t region = dm_rh_bio_to_region(ms->rh, bio); | |
06386bbf | 432 | |
1f965b19 | 433 | if (log->type->in_sync(log, region, 0)) |
06386bbf JB |
434 | return choose_mirror(ms, bio->bi_sector) ? 1 : 0; |
435 | ||
436 | return 0; | |
1da177e4 LT |
437 | } |
438 | ||
439 | /* | |
440 | * remap a buffer to a particular mirror. | |
441 | */ | |
06386bbf JB |
442 | static sector_t map_sector(struct mirror *m, struct bio *bio) |
443 | { | |
4184153f MP |
444 | if (unlikely(!bio->bi_size)) |
445 | return 0; | |
b441a262 | 446 | return m->offset + dm_target_offset(m->ms->ti, bio->bi_sector); |
06386bbf JB |
447 | } |
448 | ||
449 | static void map_bio(struct mirror *m, struct bio *bio) | |
1da177e4 LT |
450 | { |
451 | bio->bi_bdev = m->dev->bdev; | |
06386bbf JB |
452 | bio->bi_sector = map_sector(m, bio); |
453 | } | |
454 | ||
22a1ceb1 | 455 | static void map_region(struct dm_io_region *io, struct mirror *m, |
06386bbf JB |
456 | struct bio *bio) |
457 | { | |
458 | io->bdev = m->dev->bdev; | |
459 | io->sector = map_sector(m, bio); | |
460 | io->count = bio->bi_size >> 9; | |
461 | } | |
462 | ||
04788507 MP |
463 | static void hold_bio(struct mirror_set *ms, struct bio *bio) |
464 | { | |
465 | /* | |
f0703040 TY |
466 | * Lock is required to avoid race condition during suspend |
467 | * process. | |
04788507 | 468 | */ |
f0703040 TY |
469 | spin_lock_irq(&ms->lock); |
470 | ||
04788507 | 471 | if (atomic_read(&ms->suspend)) { |
f0703040 TY |
472 | spin_unlock_irq(&ms->lock); |
473 | ||
474 | /* | |
475 | * If device is suspended, complete the bio. | |
476 | */ | |
04788507 MP |
477 | if (dm_noflush_suspending(ms->ti)) |
478 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
479 | else | |
480 | bio_endio(bio, -EIO); | |
481 | return; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Hold bio until the suspend is complete. | |
486 | */ | |
04788507 MP |
487 | bio_list_add(&ms->holds, bio); |
488 | spin_unlock_irq(&ms->lock); | |
489 | } | |
490 | ||
06386bbf JB |
491 | /*----------------------------------------------------------------- |
492 | * Reads | |
493 | *---------------------------------------------------------------*/ | |
494 | static void read_callback(unsigned long error, void *context) | |
495 | { | |
496 | struct bio *bio = context; | |
497 | struct mirror *m; | |
498 | ||
499 | m = bio_get_m(bio); | |
500 | bio_set_m(bio, NULL); | |
501 | ||
502 | if (likely(!error)) { | |
503 | bio_endio(bio, 0); | |
504 | return; | |
505 | } | |
506 | ||
507 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
508 | ||
509 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
510 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
511 | "Trying alternative device.", | |
512 | m->dev->name); | |
513 | queue_bio(m->ms, bio, bio_rw(bio)); | |
514 | return; | |
515 | } | |
516 | ||
517 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
518 | m->dev->name); | |
519 | bio_endio(bio, -EIO); | |
520 | } | |
521 | ||
522 | /* Asynchronous read. */ | |
523 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
524 | { | |
22a1ceb1 | 525 | struct dm_io_region io; |
06386bbf JB |
526 | struct dm_io_request io_req = { |
527 | .bi_rw = READ, | |
528 | .mem.type = DM_IO_BVEC, | |
529 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
530 | .notify.fn = read_callback, | |
531 | .notify.context = bio, | |
532 | .client = m->ms->io_client, | |
533 | }; | |
534 | ||
535 | map_region(&io, m, bio); | |
536 | bio_set_m(bio, m); | |
1f965b19 HM |
537 | BUG_ON(dm_io(&io_req, 1, &io, NULL)); |
538 | } | |
539 | ||
540 | static inline int region_in_sync(struct mirror_set *ms, region_t region, | |
541 | int may_block) | |
542 | { | |
543 | int state = dm_rh_get_state(ms->rh, region, may_block); | |
544 | return state == DM_RH_CLEAN || state == DM_RH_DIRTY; | |
1da177e4 LT |
545 | } |
546 | ||
547 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
548 | { | |
549 | region_t region; | |
550 | struct bio *bio; | |
551 | struct mirror *m; | |
552 | ||
553 | while ((bio = bio_list_pop(reads))) { | |
1f965b19 | 554 | region = dm_rh_bio_to_region(ms->rh, bio); |
06386bbf | 555 | m = get_default_mirror(ms); |
1da177e4 LT |
556 | |
557 | /* | |
558 | * We can only read balance if the region is in sync. | |
559 | */ | |
1f965b19 | 560 | if (likely(region_in_sync(ms, region, 1))) |
1da177e4 | 561 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
562 | else if (m && atomic_read(&m->error_count)) |
563 | m = NULL; | |
1da177e4 | 564 | |
06386bbf JB |
565 | if (likely(m)) |
566 | read_async_bio(m, bio); | |
567 | else | |
568 | bio_endio(bio, -EIO); | |
1da177e4 LT |
569 | } |
570 | } | |
571 | ||
572 | /*----------------------------------------------------------------- | |
573 | * Writes. | |
574 | * | |
575 | * We do different things with the write io depending on the | |
576 | * state of the region that it's in: | |
577 | * | |
578 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
579 | * RECOVERING: delay the io until recovery completes | |
580 | * NOSYNC: increment pending, just write to the default mirror | |
581 | *---------------------------------------------------------------*/ | |
72f4b314 | 582 | |
72f4b314 | 583 | |
1da177e4 LT |
584 | static void write_callback(unsigned long error, void *context) |
585 | { | |
72f4b314 | 586 | unsigned i, ret = 0; |
1da177e4 LT |
587 | struct bio *bio = (struct bio *) context; |
588 | struct mirror_set *ms; | |
72f4b314 JB |
589 | int should_wake = 0; |
590 | unsigned long flags; | |
1da177e4 | 591 | |
06386bbf JB |
592 | ms = bio_get_m(bio)->ms; |
593 | bio_set_m(bio, NULL); | |
1da177e4 LT |
594 | |
595 | /* | |
596 | * NOTE: We don't decrement the pending count here, | |
597 | * instead it is done by the targets endio function. | |
598 | * This way we handle both writes to SYNC and NOSYNC | |
599 | * regions with the same code. | |
600 | */ | |
60f355ea MP |
601 | if (likely(!error)) { |
602 | bio_endio(bio, ret); | |
603 | return; | |
604 | } | |
1da177e4 | 605 | |
72f4b314 JB |
606 | for (i = 0; i < ms->nr_mirrors; i++) |
607 | if (test_bit(i, &error)) | |
608 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
72f4b314 | 609 | |
60f355ea MP |
610 | /* |
611 | * Need to raise event. Since raising | |
612 | * events can block, we need to do it in | |
613 | * the main thread. | |
614 | */ | |
615 | spin_lock_irqsave(&ms->lock, flags); | |
616 | if (!ms->failures.head) | |
617 | should_wake = 1; | |
618 | bio_list_add(&ms->failures, bio); | |
619 | spin_unlock_irqrestore(&ms->lock, flags); | |
620 | if (should_wake) | |
621 | wakeup_mirrord(ms); | |
1da177e4 LT |
622 | } |
623 | ||
624 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
625 | { | |
626 | unsigned int i; | |
22a1ceb1 | 627 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 628 | struct mirror *m; |
88be163a | 629 | struct dm_io_request io_req = { |
d87f4c14 | 630 | .bi_rw = WRITE | (bio->bi_rw & WRITE_FLUSH_FUA), |
88be163a MB |
631 | .mem.type = DM_IO_BVEC, |
632 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
633 | .notify.fn = write_callback, | |
634 | .notify.context = bio, | |
635 | .client = ms->io_client, | |
636 | }; | |
1da177e4 | 637 | |
5fc2ffea MS |
638 | if (bio->bi_rw & REQ_DISCARD) { |
639 | io_req.bi_rw |= REQ_DISCARD; | |
640 | io_req.mem.type = DM_IO_KMEM; | |
641 | io_req.mem.ptr.addr = NULL; | |
642 | } | |
643 | ||
06386bbf JB |
644 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
645 | map_region(dest++, m, bio); | |
1da177e4 | 646 | |
06386bbf JB |
647 | /* |
648 | * Use default mirror because we only need it to retrieve the reference | |
649 | * to the mirror set in write_callback(). | |
650 | */ | |
651 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a | 652 | |
1f965b19 | 653 | BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL)); |
1da177e4 LT |
654 | } |
655 | ||
656 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
657 | { | |
658 | int state; | |
659 | struct bio *bio; | |
660 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
7513c2a7 JB |
661 | struct bio_list requeue; |
662 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
663 | region_t region; | |
1da177e4 LT |
664 | |
665 | if (!writes->head) | |
666 | return; | |
667 | ||
668 | /* | |
669 | * Classify each write. | |
670 | */ | |
671 | bio_list_init(&sync); | |
672 | bio_list_init(&nosync); | |
673 | bio_list_init(&recover); | |
7513c2a7 | 674 | bio_list_init(&requeue); |
1da177e4 LT |
675 | |
676 | while ((bio = bio_list_pop(writes))) { | |
5fc2ffea MS |
677 | if ((bio->bi_rw & REQ_FLUSH) || |
678 | (bio->bi_rw & REQ_DISCARD)) { | |
4184153f MP |
679 | bio_list_add(&sync, bio); |
680 | continue; | |
681 | } | |
682 | ||
7513c2a7 JB |
683 | region = dm_rh_bio_to_region(ms->rh, bio); |
684 | ||
685 | if (log->type->is_remote_recovering && | |
686 | log->type->is_remote_recovering(log, region)) { | |
687 | bio_list_add(&requeue, bio); | |
688 | continue; | |
689 | } | |
690 | ||
691 | state = dm_rh_get_state(ms->rh, region, 1); | |
1da177e4 | 692 | switch (state) { |
1f965b19 HM |
693 | case DM_RH_CLEAN: |
694 | case DM_RH_DIRTY: | |
1da177e4 LT |
695 | this_list = &sync; |
696 | break; | |
697 | ||
1f965b19 | 698 | case DM_RH_NOSYNC: |
1da177e4 LT |
699 | this_list = &nosync; |
700 | break; | |
701 | ||
1f965b19 | 702 | case DM_RH_RECOVERING: |
1da177e4 LT |
703 | this_list = &recover; |
704 | break; | |
705 | } | |
706 | ||
707 | bio_list_add(this_list, bio); | |
708 | } | |
709 | ||
7513c2a7 JB |
710 | /* |
711 | * Add bios that are delayed due to remote recovery | |
712 | * back on to the write queue | |
713 | */ | |
714 | if (unlikely(requeue.head)) { | |
715 | spin_lock_irq(&ms->lock); | |
716 | bio_list_merge(&ms->writes, &requeue); | |
717 | spin_unlock_irq(&ms->lock); | |
69885683 | 718 | delayed_wake(ms); |
7513c2a7 JB |
719 | } |
720 | ||
1da177e4 LT |
721 | /* |
722 | * Increment the pending counts for any regions that will | |
723 | * be written to (writes to recover regions are going to | |
724 | * be delayed). | |
725 | */ | |
1f965b19 HM |
726 | dm_rh_inc_pending(ms->rh, &sync); |
727 | dm_rh_inc_pending(ms->rh, &nosync); | |
d2b69864 JB |
728 | |
729 | /* | |
730 | * If the flush fails on a previous call and succeeds here, | |
731 | * we must not reset the log_failure variable. We need | |
732 | * userspace interaction to do that. | |
733 | */ | |
734 | ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure; | |
1da177e4 LT |
735 | |
736 | /* | |
737 | * Dispatch io. | |
738 | */ | |
5528d17d | 739 | if (unlikely(ms->log_failure) && errors_handled(ms)) { |
b80aa7a0 JB |
740 | spin_lock_irq(&ms->lock); |
741 | bio_list_merge(&ms->failures, &sync); | |
742 | spin_unlock_irq(&ms->lock); | |
1f965b19 | 743 | wakeup_mirrord(ms); |
b80aa7a0 | 744 | } else |
fc1ff958 | 745 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 746 | do_write(ms, bio); |
1da177e4 LT |
747 | |
748 | while ((bio = bio_list_pop(&recover))) | |
1f965b19 | 749 | dm_rh_delay(ms->rh, bio); |
1da177e4 LT |
750 | |
751 | while ((bio = bio_list_pop(&nosync))) { | |
ede5ea0b MP |
752 | if (unlikely(ms->leg_failure) && errors_handled(ms)) { |
753 | spin_lock_irq(&ms->lock); | |
754 | bio_list_add(&ms->failures, bio); | |
755 | spin_unlock_irq(&ms->lock); | |
756 | wakeup_mirrord(ms); | |
757 | } else { | |
929be8fc MP |
758 | map_bio(get_default_mirror(ms), bio); |
759 | generic_make_request(bio); | |
760 | } | |
1da177e4 LT |
761 | } |
762 | } | |
763 | ||
72f4b314 JB |
764 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
765 | { | |
766 | struct bio *bio; | |
767 | ||
0f398a84 | 768 | if (likely(!failures->head)) |
72f4b314 JB |
769 | return; |
770 | ||
b80aa7a0 JB |
771 | /* |
772 | * If the log has failed, unattempted writes are being | |
0f398a84 | 773 | * put on the holds list. We can't issue those writes |
b80aa7a0 JB |
774 | * until a log has been marked, so we must store them. |
775 | * | |
776 | * If a 'noflush' suspend is in progress, we can requeue | |
777 | * the I/O's to the core. This give userspace a chance | |
778 | * to reconfigure the mirror, at which point the core | |
779 | * will reissue the writes. If the 'noflush' flag is | |
780 | * not set, we have no choice but to return errors. | |
781 | * | |
782 | * Some writes on the failures list may have been | |
783 | * submitted before the log failure and represent a | |
784 | * failure to write to one of the devices. It is ok | |
785 | * for us to treat them the same and requeue them | |
786 | * as well. | |
787 | */ | |
0f398a84 | 788 | while ((bio = bio_list_pop(failures))) { |
60f355ea | 789 | if (!ms->log_failure) { |
0f398a84 | 790 | ms->in_sync = 0; |
c58098be | 791 | dm_rh_mark_nosync(ms->rh, bio); |
0f398a84 | 792 | } |
60f355ea MP |
793 | |
794 | /* | |
795 | * If all the legs are dead, fail the I/O. | |
796 | * If we have been told to handle errors, hold the bio | |
797 | * and wait for userspace to deal with the problem. | |
798 | * Otherwise pretend that the I/O succeeded. (This would | |
799 | * be wrong if the failed leg returned after reboot and | |
800 | * got replicated back to the good legs.) | |
801 | */ | |
802 | if (!get_valid_mirror(ms)) | |
803 | bio_endio(bio, -EIO); | |
804 | else if (errors_handled(ms)) | |
805 | hold_bio(ms, bio); | |
806 | else | |
807 | bio_endio(bio, 0); | |
b80aa7a0 | 808 | } |
72f4b314 JB |
809 | } |
810 | ||
811 | static void trigger_event(struct work_struct *work) | |
812 | { | |
813 | struct mirror_set *ms = | |
814 | container_of(work, struct mirror_set, trigger_event); | |
815 | ||
816 | dm_table_event(ms->ti->table); | |
817 | } | |
818 | ||
1da177e4 LT |
819 | /*----------------------------------------------------------------- |
820 | * kmirrord | |
821 | *---------------------------------------------------------------*/ | |
a2aebe03 | 822 | static void do_mirror(struct work_struct *work) |
1da177e4 | 823 | { |
1f965b19 HM |
824 | struct mirror_set *ms = container_of(work, struct mirror_set, |
825 | kmirrord_work); | |
72f4b314 JB |
826 | struct bio_list reads, writes, failures; |
827 | unsigned long flags; | |
1da177e4 | 828 | |
72f4b314 | 829 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
830 | reads = ms->reads; |
831 | writes = ms->writes; | |
72f4b314 | 832 | failures = ms->failures; |
1da177e4 LT |
833 | bio_list_init(&ms->reads); |
834 | bio_list_init(&ms->writes); | |
72f4b314 JB |
835 | bio_list_init(&ms->failures); |
836 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 | 837 | |
1f965b19 | 838 | dm_rh_update_states(ms->rh, errors_handled(ms)); |
1da177e4 LT |
839 | do_recovery(ms); |
840 | do_reads(ms, &reads); | |
841 | do_writes(ms, &writes); | |
72f4b314 | 842 | do_failures(ms, &failures); |
1da177e4 LT |
843 | } |
844 | ||
1da177e4 LT |
845 | /*----------------------------------------------------------------- |
846 | * Target functions | |
847 | *---------------------------------------------------------------*/ | |
848 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
849 | uint32_t region_size, | |
850 | struct dm_target *ti, | |
416cd17b | 851 | struct dm_dirty_log *dl) |
1da177e4 LT |
852 | { |
853 | size_t len; | |
854 | struct mirror_set *ms = NULL; | |
855 | ||
1da177e4 LT |
856 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
857 | ||
dd00cc48 | 858 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 859 | if (!ms) { |
72d94861 | 860 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
861 | return NULL; |
862 | } | |
863 | ||
1da177e4 | 864 | spin_lock_init(&ms->lock); |
5339fc2d MP |
865 | bio_list_init(&ms->reads); |
866 | bio_list_init(&ms->writes); | |
867 | bio_list_init(&ms->failures); | |
868 | bio_list_init(&ms->holds); | |
1da177e4 LT |
869 | |
870 | ms->ti = ti; | |
871 | ms->nr_mirrors = nr_mirrors; | |
872 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
873 | ms->in_sync = 0; | |
b80aa7a0 | 874 | ms->log_failure = 0; |
929be8fc | 875 | ms->leg_failure = 0; |
b80aa7a0 | 876 | atomic_set(&ms->suspend, 0); |
72f4b314 | 877 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 878 | |
95f8fac8 MP |
879 | ms->read_record_pool = mempool_create_slab_pool(MIN_READ_RECORDS, |
880 | _dm_raid1_read_record_cache); | |
881 | ||
06386bbf JB |
882 | if (!ms->read_record_pool) { |
883 | ti->error = "Error creating mirror read_record_pool"; | |
884 | kfree(ms); | |
885 | return NULL; | |
886 | } | |
887 | ||
bda8efec | 888 | ms->io_client = dm_io_client_create(); |
88be163a MB |
889 | if (IS_ERR(ms->io_client)) { |
890 | ti->error = "Error creating dm_io client"; | |
06386bbf | 891 | mempool_destroy(ms->read_record_pool); |
88be163a MB |
892 | kfree(ms); |
893 | return NULL; | |
894 | } | |
895 | ||
1f965b19 HM |
896 | ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord, |
897 | wakeup_all_recovery_waiters, | |
898 | ms->ti->begin, MAX_RECOVERY, | |
899 | dl, region_size, ms->nr_regions); | |
900 | if (IS_ERR(ms->rh)) { | |
72d94861 | 901 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 902 | dm_io_client_destroy(ms->io_client); |
06386bbf | 903 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
904 | kfree(ms); |
905 | return NULL; | |
906 | } | |
907 | ||
908 | return ms; | |
909 | } | |
910 | ||
911 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
912 | unsigned int m) | |
913 | { | |
914 | while (m--) | |
915 | dm_put_device(ti, ms->mirror[m].dev); | |
916 | ||
88be163a | 917 | dm_io_client_destroy(ms->io_client); |
1f965b19 | 918 | dm_region_hash_destroy(ms->rh); |
06386bbf | 919 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
920 | kfree(ms); |
921 | } | |
922 | ||
1da177e4 LT |
923 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
924 | unsigned int mirror, char **argv) | |
925 | { | |
4ee218cd | 926 | unsigned long long offset; |
1da177e4 | 927 | |
4ee218cd | 928 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 929 | ti->error = "Invalid offset"; |
1da177e4 LT |
930 | return -EINVAL; |
931 | } | |
932 | ||
8215d6ec | 933 | if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), |
1da177e4 | 934 | &ms->mirror[mirror].dev)) { |
72d94861 | 935 | ti->error = "Device lookup failure"; |
1da177e4 LT |
936 | return -ENXIO; |
937 | } | |
938 | ||
aa5617c5 | 939 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
940 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
941 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
942 | ms->mirror[mirror].offset = offset; |
943 | ||
944 | return 0; | |
945 | } | |
946 | ||
1da177e4 LT |
947 | /* |
948 | * Create dirty log: log_type #log_params <log_params> | |
949 | */ | |
416cd17b | 950 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1f965b19 HM |
951 | unsigned argc, char **argv, |
952 | unsigned *args_used) | |
1da177e4 | 953 | { |
1f965b19 | 954 | unsigned param_count; |
416cd17b | 955 | struct dm_dirty_log *dl; |
1da177e4 LT |
956 | |
957 | if (argc < 2) { | |
72d94861 | 958 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
959 | return NULL; |
960 | } | |
961 | ||
962 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 963 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
964 | return NULL; |
965 | } | |
966 | ||
967 | *args_used = 2 + param_count; | |
968 | ||
969 | if (argc < *args_used) { | |
72d94861 | 970 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
971 | return NULL; |
972 | } | |
973 | ||
c0da3748 MP |
974 | dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count, |
975 | argv + 2); | |
1da177e4 | 976 | if (!dl) { |
72d94861 | 977 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
978 | return NULL; |
979 | } | |
980 | ||
1da177e4 LT |
981 | return dl; |
982 | } | |
983 | ||
a8e6afa2 JB |
984 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
985 | unsigned *args_used) | |
986 | { | |
987 | unsigned num_features; | |
988 | struct dm_target *ti = ms->ti; | |
989 | ||
990 | *args_used = 0; | |
991 | ||
992 | if (!argc) | |
993 | return 0; | |
994 | ||
995 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
996 | ti->error = "Invalid number of features"; | |
997 | return -EINVAL; | |
998 | } | |
999 | ||
1000 | argc--; | |
1001 | argv++; | |
1002 | (*args_used)++; | |
1003 | ||
1004 | if (num_features > argc) { | |
1005 | ti->error = "Not enough arguments to support feature count"; | |
1006 | return -EINVAL; | |
1007 | } | |
1008 | ||
1009 | if (!strcmp("handle_errors", argv[0])) | |
1010 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
1011 | else { | |
1012 | ti->error = "Unrecognised feature requested"; | |
1013 | return -EINVAL; | |
1014 | } | |
1015 | ||
1016 | (*args_used)++; | |
1017 | ||
1018 | return 0; | |
1019 | } | |
1020 | ||
1da177e4 LT |
1021 | /* |
1022 | * Construct a mirror mapping: | |
1023 | * | |
1024 | * log_type #log_params <log_params> | |
1025 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1026 | * [#features <features>] |
1da177e4 LT |
1027 | * |
1028 | * log_type is "core" or "disk" | |
1029 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1030 | * |
1031 | * If present, features must be "handle_errors". | |
1da177e4 | 1032 | */ |
1da177e4 LT |
1033 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1034 | { | |
1035 | int r; | |
1036 | unsigned int nr_mirrors, m, args_used; | |
1037 | struct mirror_set *ms; | |
416cd17b | 1038 | struct dm_dirty_log *dl; |
1da177e4 LT |
1039 | |
1040 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
1041 | if (!dl) | |
1042 | return -EINVAL; | |
1043 | ||
1044 | argv += args_used; | |
1045 | argc -= args_used; | |
1046 | ||
1047 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
eb69aca5 | 1048 | nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) { |
72d94861 | 1049 | ti->error = "Invalid number of mirrors"; |
416cd17b | 1050 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1051 | return -EINVAL; |
1052 | } | |
1053 | ||
1054 | argv++, argc--; | |
1055 | ||
a8e6afa2 JB |
1056 | if (argc < nr_mirrors * 2) { |
1057 | ti->error = "Too few mirror arguments"; | |
416cd17b | 1058 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1059 | return -EINVAL; |
1060 | } | |
1061 | ||
1062 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
1063 | if (!ms) { | |
416cd17b | 1064 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1065 | return -ENOMEM; |
1066 | } | |
1067 | ||
1068 | /* Get the mirror parameter sets */ | |
1069 | for (m = 0; m < nr_mirrors; m++) { | |
1070 | r = get_mirror(ms, ti, m, argv); | |
1071 | if (r) { | |
1072 | free_context(ms, ti, m); | |
1073 | return r; | |
1074 | } | |
1075 | argv += 2; | |
1076 | argc -= 2; | |
1077 | } | |
1078 | ||
1079 | ti->private = ms; | |
1f965b19 | 1080 | ti->split_io = dm_rh_get_region_size(ms->rh); |
4184153f | 1081 | ti->num_flush_requests = 1; |
5fc2ffea | 1082 | ti->num_discard_requests = 1; |
1da177e4 | 1083 | |
9c4376de TH |
1084 | ms->kmirrord_wq = alloc_workqueue("kmirrord", |
1085 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
6ad36fe2 HS |
1086 | if (!ms->kmirrord_wq) { |
1087 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1088 | r = -ENOMEM; |
1089 | goto err_free_context; | |
6ad36fe2 HS |
1090 | } |
1091 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
1092 | init_timer(&ms->timer); |
1093 | ms->timer_pending = 0; | |
72f4b314 | 1094 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 1095 | |
a8e6afa2 | 1096 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1097 | if (r) |
1098 | goto err_destroy_wq; | |
a8e6afa2 JB |
1099 | |
1100 | argv += args_used; | |
1101 | argc -= args_used; | |
1102 | ||
f44db678 JB |
1103 | /* |
1104 | * Any read-balancing addition depends on the | |
1105 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1106 | * This is because the decision to balance depends | |
1107 | * on the sync state of a region. If the above | |
1108 | * flag is not present, we ignore errors; and | |
1109 | * the sync state may be inaccurate. | |
1110 | */ | |
1111 | ||
a8e6afa2 JB |
1112 | if (argc) { |
1113 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1114 | r = -EINVAL; |
1115 | goto err_destroy_wq; | |
a8e6afa2 JB |
1116 | } |
1117 | ||
fa34ce73 MP |
1118 | ms->kcopyd_client = dm_kcopyd_client_create(); |
1119 | if (IS_ERR(ms->kcopyd_client)) { | |
1120 | r = PTR_ERR(ms->kcopyd_client); | |
a72cf737 | 1121 | goto err_destroy_wq; |
fa34ce73 | 1122 | } |
1da177e4 | 1123 | |
1f965b19 | 1124 | wakeup_mirrord(ms); |
1da177e4 | 1125 | return 0; |
a72cf737 DM |
1126 | |
1127 | err_destroy_wq: | |
1128 | destroy_workqueue(ms->kmirrord_wq); | |
1129 | err_free_context: | |
1130 | free_context(ms, ti, ms->nr_mirrors); | |
1131 | return r; | |
1da177e4 LT |
1132 | } |
1133 | ||
1134 | static void mirror_dtr(struct dm_target *ti) | |
1135 | { | |
1136 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1137 | ||
a2aebe03 | 1138 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1139 | flush_workqueue(ms->kmirrord_wq); |
d5ffa387 | 1140 | flush_work_sync(&ms->trigger_event); |
eb69aca5 | 1141 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1142 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1143 | free_context(ms, ti, ms->nr_mirrors); |
1144 | } | |
1145 | ||
1da177e4 LT |
1146 | /* |
1147 | * Mirror mapping function | |
1148 | */ | |
1149 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1150 | union map_info *map_context) | |
1151 | { | |
1152 | int r, rw = bio_rw(bio); | |
1153 | struct mirror *m; | |
1154 | struct mirror_set *ms = ti->private; | |
06386bbf | 1155 | struct dm_raid1_read_record *read_record = NULL; |
1f965b19 | 1156 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 LT |
1157 | |
1158 | if (rw == WRITE) { | |
06386bbf | 1159 | /* Save region for mirror_end_io() handler */ |
1f965b19 | 1160 | map_context->ll = dm_rh_bio_to_region(ms->rh, bio); |
1da177e4 | 1161 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1162 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1163 | } |
1164 | ||
1f965b19 | 1165 | r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0); |
1da177e4 LT |
1166 | if (r < 0 && r != -EWOULDBLOCK) |
1167 | return r; | |
1168 | ||
1da177e4 | 1169 | /* |
06386bbf | 1170 | * If region is not in-sync queue the bio. |
1da177e4 | 1171 | */ |
06386bbf JB |
1172 | if (!r || (r == -EWOULDBLOCK)) { |
1173 | if (rw == READA) | |
1174 | return -EWOULDBLOCK; | |
1da177e4 | 1175 | |
1da177e4 | 1176 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1177 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1178 | } |
1179 | ||
06386bbf JB |
1180 | /* |
1181 | * The region is in-sync and we can perform reads directly. | |
1182 | * Store enough information so we can retry if it fails. | |
1183 | */ | |
1da177e4 | 1184 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1185 | if (unlikely(!m)) |
1da177e4 LT |
1186 | return -EIO; |
1187 | ||
06386bbf JB |
1188 | read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO); |
1189 | if (likely(read_record)) { | |
1190 | dm_bio_record(&read_record->details, bio); | |
1191 | map_context->ptr = read_record; | |
1192 | read_record->m = m; | |
1193 | } | |
1194 | ||
1195 | map_bio(m, bio); | |
1196 | ||
d2a7ad29 | 1197 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1201 | int error, union map_info *map_context) | |
1202 | { | |
1203 | int rw = bio_rw(bio); | |
1204 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1205 | struct mirror *m = NULL; |
1206 | struct dm_bio_details *bd = NULL; | |
1207 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1208 | |
1209 | /* | |
1210 | * We need to dec pending if this was a write. | |
1211 | */ | |
06386bbf | 1212 | if (rw == WRITE) { |
d87f4c14 | 1213 | if (!(bio->bi_rw & REQ_FLUSH)) |
4184153f | 1214 | dm_rh_dec(ms->rh, map_context->ll); |
06386bbf JB |
1215 | return error; |
1216 | } | |
1da177e4 | 1217 | |
06386bbf JB |
1218 | if (error == -EOPNOTSUPP) |
1219 | goto out; | |
1220 | ||
7b6d91da | 1221 | if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) |
06386bbf JB |
1222 | goto out; |
1223 | ||
1224 | if (unlikely(error)) { | |
1225 | if (!read_record) { | |
1226 | /* | |
1227 | * There wasn't enough memory to record necessary | |
1228 | * information for a retry or there was no other | |
1229 | * mirror in-sync. | |
1230 | */ | |
e03f1a84 | 1231 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1232 | return -EIO; |
1233 | } | |
e03f1a84 AB |
1234 | |
1235 | m = read_record->m; | |
1236 | ||
06386bbf JB |
1237 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1238 | m->dev->name); | |
1239 | ||
06386bbf JB |
1240 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1241 | ||
1242 | /* | |
1243 | * A failed read is requeued for another attempt using an intact | |
1244 | * mirror. | |
1245 | */ | |
1246 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1247 | bd = &read_record->details; | |
1248 | ||
1249 | dm_bio_restore(bd, bio); | |
1250 | mempool_free(read_record, ms->read_record_pool); | |
1251 | map_context->ptr = NULL; | |
1252 | queue_bio(ms, bio, rw); | |
1253 | return 1; | |
1254 | } | |
1255 | DMERR("All replicated volumes dead, failing I/O"); | |
1256 | } | |
1257 | ||
1258 | out: | |
1259 | if (read_record) { | |
1260 | mempool_free(read_record, ms->read_record_pool); | |
1261 | map_context->ptr = NULL; | |
1262 | } | |
1263 | ||
1264 | return error; | |
1da177e4 LT |
1265 | } |
1266 | ||
b80aa7a0 | 1267 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1268 | { |
1269 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1f965b19 | 1270 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 | 1271 | |
04788507 MP |
1272 | struct bio_list holds; |
1273 | struct bio *bio; | |
1274 | ||
b80aa7a0 JB |
1275 | atomic_set(&ms->suspend, 1); |
1276 | ||
f0703040 TY |
1277 | /* |
1278 | * Process bios in the hold list to start recovery waiting | |
1279 | * for bios in the hold list. After the process, no bio has | |
1280 | * a chance to be added in the hold list because ms->suspend | |
1281 | * is set. | |
1282 | */ | |
1283 | spin_lock_irq(&ms->lock); | |
1284 | holds = ms->holds; | |
1285 | bio_list_init(&ms->holds); | |
1286 | spin_unlock_irq(&ms->lock); | |
1287 | ||
1288 | while ((bio = bio_list_pop(&holds))) | |
1289 | hold_bio(ms, bio); | |
1290 | ||
b80aa7a0 JB |
1291 | /* |
1292 | * We must finish up all the work that we've | |
1293 | * generated (i.e. recovery work). | |
1294 | */ | |
1f965b19 | 1295 | dm_rh_stop_recovery(ms->rh); |
33184048 | 1296 | |
33184048 | 1297 | wait_event(_kmirrord_recovery_stopped, |
1f965b19 | 1298 | !dm_rh_recovery_in_flight(ms->rh)); |
33184048 | 1299 | |
b80aa7a0 JB |
1300 | if (log->type->presuspend && log->type->presuspend(log)) |
1301 | /* FIXME: need better error handling */ | |
1302 | DMWARN("log presuspend failed"); | |
1303 | ||
1304 | /* | |
1305 | * Now that recovery is complete/stopped and the | |
1306 | * delayed bios are queued, we need to wait for | |
1307 | * the worker thread to complete. This way, | |
1308 | * we know that all of our I/O has been pushed. | |
1309 | */ | |
1310 | flush_workqueue(ms->kmirrord_wq); | |
1311 | } | |
1312 | ||
1313 | static void mirror_postsuspend(struct dm_target *ti) | |
1314 | { | |
1315 | struct mirror_set *ms = ti->private; | |
1f965b19 | 1316 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 | 1317 | |
6b3df0d7 | 1318 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1319 | /* FIXME: need better error handling */ |
b80aa7a0 | 1320 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1321 | } |
1322 | ||
1323 | static void mirror_resume(struct dm_target *ti) | |
1324 | { | |
b80aa7a0 | 1325 | struct mirror_set *ms = ti->private; |
1f965b19 | 1326 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 JB |
1327 | |
1328 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1329 | if (log->type->resume && log->type->resume(log)) |
1330 | /* FIXME: need better error handling */ | |
1331 | DMWARN("log resume failed"); | |
1f965b19 | 1332 | dm_rh_start_recovery(ms->rh); |
1da177e4 LT |
1333 | } |
1334 | ||
af195ac8 JB |
1335 | /* |
1336 | * device_status_char | |
1337 | * @m: mirror device/leg we want the status of | |
1338 | * | |
1339 | * We return one character representing the most severe error | |
1340 | * we have encountered. | |
1341 | * A => Alive - No failures | |
1342 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1343 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1344 | * R => Read - A read failure occurred, mirror data unaffected | |
1345 | * | |
1346 | * Returns: <char> | |
1347 | */ | |
1348 | static char device_status_char(struct mirror *m) | |
1349 | { | |
1350 | if (!atomic_read(&(m->error_count))) | |
1351 | return 'A'; | |
1352 | ||
64b30c46 MP |
1353 | return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' : |
1354 | (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
af195ac8 JB |
1355 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : |
1356 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1357 | } | |
1358 | ||
1359 | ||
1da177e4 LT |
1360 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1361 | char *result, unsigned int maxlen) | |
1362 | { | |
315dcc22 | 1363 | unsigned int m, sz = 0; |
1da177e4 | 1364 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1f965b19 | 1365 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
af195ac8 | 1366 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1367 | |
1da177e4 LT |
1368 | switch (type) { |
1369 | case STATUSTYPE_INFO: | |
1370 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1371 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1372 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1373 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1374 | } | |
1375 | buffer[m] = '\0'; | |
1da177e4 | 1376 | |
af195ac8 | 1377 | DMEMIT("%llu/%llu 1 %s ", |
1f965b19 | 1378 | (unsigned long long)log->type->get_sync_count(log), |
af195ac8 | 1379 | (unsigned long long)ms->nr_regions, buffer); |
315dcc22 | 1380 | |
1f965b19 | 1381 | sz += log->type->status(log, type, result+sz, maxlen-sz); |
315dcc22 | 1382 | |
1da177e4 LT |
1383 | break; |
1384 | ||
1385 | case STATUSTYPE_TABLE: | |
1f965b19 | 1386 | sz = log->type->status(log, type, result, maxlen); |
315dcc22 | 1387 | |
e52b8f6d | 1388 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1389 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1390 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1391 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1392 | |
1393 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1394 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1395 | } |
1396 | ||
1397 | return 0; | |
1398 | } | |
1399 | ||
af4874e0 MS |
1400 | static int mirror_iterate_devices(struct dm_target *ti, |
1401 | iterate_devices_callout_fn fn, void *data) | |
1402 | { | |
1403 | struct mirror_set *ms = ti->private; | |
1404 | int ret = 0; | |
1405 | unsigned i; | |
1406 | ||
1407 | for (i = 0; !ret && i < ms->nr_mirrors; i++) | |
1408 | ret = fn(ti, ms->mirror[i].dev, | |
5dea271b | 1409 | ms->mirror[i].offset, ti->len, data); |
af4874e0 MS |
1410 | |
1411 | return ret; | |
1412 | } | |
1413 | ||
1da177e4 LT |
1414 | static struct target_type mirror_target = { |
1415 | .name = "mirror", | |
9c4376de | 1416 | .version = {1, 12, 1}, |
1da177e4 LT |
1417 | .module = THIS_MODULE, |
1418 | .ctr = mirror_ctr, | |
1419 | .dtr = mirror_dtr, | |
1420 | .map = mirror_map, | |
1421 | .end_io = mirror_end_io, | |
b80aa7a0 | 1422 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1423 | .postsuspend = mirror_postsuspend, |
1424 | .resume = mirror_resume, | |
1425 | .status = mirror_status, | |
af4874e0 | 1426 | .iterate_devices = mirror_iterate_devices, |
1da177e4 LT |
1427 | }; |
1428 | ||
1429 | static int __init dm_mirror_init(void) | |
1430 | { | |
1431 | int r; | |
1432 | ||
95f8fac8 MP |
1433 | _dm_raid1_read_record_cache = KMEM_CACHE(dm_raid1_read_record, 0); |
1434 | if (!_dm_raid1_read_record_cache) { | |
1435 | DMERR("Can't allocate dm_raid1_read_record cache"); | |
1436 | r = -ENOMEM; | |
1437 | goto bad_cache; | |
1438 | } | |
1439 | ||
1da177e4 | 1440 | r = dm_register_target(&mirror_target); |
95f8fac8 | 1441 | if (r < 0) { |
0cd33124 | 1442 | DMERR("Failed to register mirror target"); |
95f8fac8 MP |
1443 | goto bad_target; |
1444 | } | |
1445 | ||
1446 | return 0; | |
1da177e4 | 1447 | |
95f8fac8 MP |
1448 | bad_target: |
1449 | kmem_cache_destroy(_dm_raid1_read_record_cache); | |
1450 | bad_cache: | |
1da177e4 LT |
1451 | return r; |
1452 | } | |
1453 | ||
1454 | static void __exit dm_mirror_exit(void) | |
1455 | { | |
10d3bd09 | 1456 | dm_unregister_target(&mirror_target); |
95f8fac8 | 1457 | kmem_cache_destroy(_dm_raid1_read_record_cache); |
1da177e4 LT |
1458 | } |
1459 | ||
1460 | /* Module hooks */ | |
1461 | module_init(dm_mirror_init); | |
1462 | module_exit(dm_mirror_exit); | |
1463 | ||
1464 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1465 | MODULE_AUTHOR("Joe Thornber"); | |
1466 | MODULE_LICENSE("GPL"); |