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 MB |
63 | struct dm_io_client *io_client; |
64 | ||
e4c8b3ba NB |
65 | /* recovery */ |
66 | region_t nr_regions; | |
67 | int in_sync; | |
fc1ff958 | 68 | int log_failure; |
929be8fc | 69 | int leg_failure; |
b80aa7a0 | 70 | atomic_t suspend; |
e4c8b3ba | 71 | |
72f4b314 | 72 | atomic_t default_mirror; /* Default mirror */ |
e4c8b3ba | 73 | |
6ad36fe2 HS |
74 | struct workqueue_struct *kmirrord_wq; |
75 | struct work_struct kmirrord_work; | |
a2aebe03 MP |
76 | struct timer_list timer; |
77 | unsigned long timer_pending; | |
78 | ||
72f4b314 | 79 | struct work_struct trigger_event; |
6ad36fe2 | 80 | |
1f965b19 | 81 | unsigned nr_mirrors; |
e4c8b3ba NB |
82 | struct mirror mirror[0]; |
83 | }; | |
84 | ||
1f965b19 | 85 | static void wakeup_mirrord(void *context) |
1da177e4 | 86 | { |
1f965b19 | 87 | struct mirror_set *ms = context; |
1da177e4 | 88 | |
6ad36fe2 HS |
89 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); |
90 | } | |
91 | ||
a2aebe03 MP |
92 | static void delayed_wake_fn(unsigned long data) |
93 | { | |
94 | struct mirror_set *ms = (struct mirror_set *) data; | |
95 | ||
96 | clear_bit(0, &ms->timer_pending); | |
1f965b19 | 97 | wakeup_mirrord(ms); |
a2aebe03 MP |
98 | } |
99 | ||
100 | static void delayed_wake(struct mirror_set *ms) | |
101 | { | |
102 | if (test_and_set_bit(0, &ms->timer_pending)) | |
103 | return; | |
104 | ||
105 | ms->timer.expires = jiffies + HZ / 5; | |
106 | ms->timer.data = (unsigned long) ms; | |
107 | ms->timer.function = delayed_wake_fn; | |
108 | add_timer(&ms->timer); | |
109 | } | |
110 | ||
1f965b19 | 111 | static void wakeup_all_recovery_waiters(void *context) |
1da177e4 | 112 | { |
1f965b19 | 113 | wake_up_all(&_kmirrord_recovery_stopped); |
1da177e4 LT |
114 | } |
115 | ||
1f965b19 | 116 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) |
1da177e4 LT |
117 | { |
118 | unsigned long flags; | |
1da177e4 | 119 | int should_wake = 0; |
1f965b19 | 120 | struct bio_list *bl; |
1da177e4 | 121 | |
1f965b19 HM |
122 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; |
123 | spin_lock_irqsave(&ms->lock, flags); | |
124 | should_wake = !(bl->head); | |
125 | bio_list_add(bl, bio); | |
126 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 LT |
127 | |
128 | if (should_wake) | |
1f965b19 | 129 | wakeup_mirrord(ms); |
1da177e4 LT |
130 | } |
131 | ||
1f965b19 | 132 | static void dispatch_bios(void *context, struct bio_list *bio_list) |
1da177e4 | 133 | { |
1f965b19 HM |
134 | struct mirror_set *ms = context; |
135 | struct bio *bio; | |
1da177e4 | 136 | |
1f965b19 HM |
137 | while ((bio = bio_list_pop(bio_list))) |
138 | queue_bio(ms, bio, WRITE); | |
1da177e4 LT |
139 | } |
140 | ||
06386bbf JB |
141 | struct dm_raid1_read_record { |
142 | struct mirror *m; | |
143 | struct dm_bio_details details; | |
144 | }; | |
145 | ||
1da177e4 LT |
146 | /* |
147 | * Every mirror should look like this one. | |
148 | */ | |
149 | #define DEFAULT_MIRROR 0 | |
150 | ||
151 | /* | |
06386bbf JB |
152 | * This is yucky. We squirrel the mirror struct away inside |
153 | * bi_next for read/write buffers. This is safe since the bh | |
1da177e4 LT |
154 | * doesn't get submitted to the lower levels of block layer. |
155 | */ | |
06386bbf | 156 | static struct mirror *bio_get_m(struct bio *bio) |
1da177e4 | 157 | { |
06386bbf | 158 | return (struct mirror *) bio->bi_next; |
1da177e4 LT |
159 | } |
160 | ||
06386bbf | 161 | static void bio_set_m(struct bio *bio, struct mirror *m) |
1da177e4 | 162 | { |
06386bbf | 163 | bio->bi_next = (struct bio *) m; |
1da177e4 LT |
164 | } |
165 | ||
72f4b314 JB |
166 | static struct mirror *get_default_mirror(struct mirror_set *ms) |
167 | { | |
168 | return &ms->mirror[atomic_read(&ms->default_mirror)]; | |
169 | } | |
170 | ||
171 | static void set_default_mirror(struct mirror *m) | |
172 | { | |
173 | struct mirror_set *ms = m->ms; | |
174 | struct mirror *m0 = &(ms->mirror[0]); | |
175 | ||
176 | atomic_set(&ms->default_mirror, m - m0); | |
177 | } | |
178 | ||
87968ddd MP |
179 | static struct mirror *get_valid_mirror(struct mirror_set *ms) |
180 | { | |
181 | struct mirror *m; | |
182 | ||
183 | for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++) | |
184 | if (!atomic_read(&m->error_count)) | |
185 | return m; | |
186 | ||
187 | return NULL; | |
188 | } | |
189 | ||
72f4b314 JB |
190 | /* fail_mirror |
191 | * @m: mirror device to fail | |
192 | * @error_type: one of the enum's, DM_RAID1_*_ERROR | |
193 | * | |
194 | * If errors are being handled, record the type of | |
195 | * error encountered for this device. If this type | |
196 | * of error has already been recorded, we can return; | |
197 | * otherwise, we must signal userspace by triggering | |
198 | * an event. Additionally, if the device is the | |
199 | * primary device, we must choose a new primary, but | |
200 | * only if the mirror is in-sync. | |
201 | * | |
202 | * This function must not block. | |
203 | */ | |
204 | static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) | |
205 | { | |
206 | struct mirror_set *ms = m->ms; | |
207 | struct mirror *new; | |
208 | ||
929be8fc MP |
209 | ms->leg_failure = 1; |
210 | ||
72f4b314 JB |
211 | /* |
212 | * error_count is used for nothing more than a | |
213 | * simple way to tell if a device has encountered | |
214 | * errors. | |
215 | */ | |
216 | atomic_inc(&m->error_count); | |
217 | ||
218 | if (test_and_set_bit(error_type, &m->error_type)) | |
219 | return; | |
220 | ||
d460c65a JB |
221 | if (!errors_handled(ms)) |
222 | return; | |
223 | ||
72f4b314 JB |
224 | if (m != get_default_mirror(ms)) |
225 | goto out; | |
226 | ||
227 | if (!ms->in_sync) { | |
228 | /* | |
229 | * Better to issue requests to same failing device | |
230 | * than to risk returning corrupt data. | |
231 | */ | |
232 | DMERR("Primary mirror (%s) failed while out-of-sync: " | |
233 | "Reads may fail.", m->dev->name); | |
234 | goto out; | |
235 | } | |
236 | ||
87968ddd MP |
237 | new = get_valid_mirror(ms); |
238 | if (new) | |
239 | set_default_mirror(new); | |
240 | else | |
72f4b314 JB |
241 | DMWARN("All sides of mirror have failed."); |
242 | ||
243 | out: | |
244 | schedule_work(&ms->trigger_event); | |
245 | } | |
246 | ||
c0da3748 MP |
247 | static int mirror_flush(struct dm_target *ti) |
248 | { | |
249 | struct mirror_set *ms = ti->private; | |
250 | unsigned long error_bits; | |
251 | ||
252 | unsigned int i; | |
253 | struct dm_io_region io[ms->nr_mirrors]; | |
254 | struct mirror *m; | |
255 | struct dm_io_request io_req = { | |
d87f4c14 | 256 | .bi_rw = WRITE_FLUSH, |
c0da3748 | 257 | .mem.type = DM_IO_KMEM, |
5fc2ffea | 258 | .mem.ptr.addr = NULL, |
c0da3748 MP |
259 | .client = ms->io_client, |
260 | }; | |
261 | ||
262 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) { | |
263 | io[i].bdev = m->dev->bdev; | |
264 | io[i].sector = 0; | |
265 | io[i].count = 0; | |
266 | } | |
267 | ||
268 | error_bits = -1; | |
269 | dm_io(&io_req, ms->nr_mirrors, io, &error_bits); | |
270 | if (unlikely(error_bits != 0)) { | |
271 | for (i = 0; i < ms->nr_mirrors; i++) | |
272 | if (test_bit(i, &error_bits)) | |
273 | fail_mirror(ms->mirror + i, | |
64b30c46 | 274 | DM_RAID1_FLUSH_ERROR); |
c0da3748 MP |
275 | return -EIO; |
276 | } | |
277 | ||
278 | return 0; | |
279 | } | |
280 | ||
1da177e4 LT |
281 | /*----------------------------------------------------------------- |
282 | * Recovery. | |
283 | * | |
284 | * When a mirror is first activated we may find that some regions | |
285 | * are in the no-sync state. We have to recover these by | |
286 | * recopying from the default mirror to all the others. | |
287 | *---------------------------------------------------------------*/ | |
4cdc1d1f | 288 | static void recovery_complete(int read_err, unsigned long write_err, |
1da177e4 LT |
289 | void *context) |
290 | { | |
1f965b19 HM |
291 | struct dm_region *reg = context; |
292 | struct mirror_set *ms = dm_rh_region_context(reg); | |
8f0205b7 | 293 | int m, bit = 0; |
1da177e4 | 294 | |
8f0205b7 | 295 | if (read_err) { |
f44db678 JB |
296 | /* Read error means the failure of default mirror. */ |
297 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
8f0205b7 JB |
298 | fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); |
299 | } | |
f44db678 | 300 | |
8f0205b7 | 301 | if (write_err) { |
4cdc1d1f | 302 | DMERR_LIMIT("Write error during recovery (error = 0x%lx)", |
f44db678 | 303 | write_err); |
8f0205b7 JB |
304 | /* |
305 | * Bits correspond to devices (excluding default mirror). | |
306 | * The default mirror cannot change during recovery. | |
307 | */ | |
308 | for (m = 0; m < ms->nr_mirrors; m++) { | |
309 | if (&ms->mirror[m] == get_default_mirror(ms)) | |
310 | continue; | |
311 | if (test_bit(bit, &write_err)) | |
312 | fail_mirror(ms->mirror + m, | |
313 | DM_RAID1_SYNC_ERROR); | |
314 | bit++; | |
315 | } | |
316 | } | |
f44db678 | 317 | |
1f965b19 | 318 | dm_rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
319 | } |
320 | ||
1f965b19 | 321 | static int recover(struct mirror_set *ms, struct dm_region *reg) |
1da177e4 LT |
322 | { |
323 | int r; | |
1f965b19 | 324 | unsigned i; |
eb69aca5 | 325 | struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; |
1da177e4 LT |
326 | struct mirror *m; |
327 | unsigned long flags = 0; | |
1f965b19 HM |
328 | region_t key = dm_rh_get_region_key(reg); |
329 | sector_t region_size = dm_rh_get_region_size(ms->rh); | |
1da177e4 LT |
330 | |
331 | /* fill in the source */ | |
72f4b314 | 332 | m = get_default_mirror(ms); |
1da177e4 | 333 | from.bdev = m->dev->bdev; |
1f965b19 HM |
334 | from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
335 | if (key == (ms->nr_regions - 1)) { | |
1da177e4 LT |
336 | /* |
337 | * The final region may be smaller than | |
338 | * region_size. | |
339 | */ | |
1f965b19 | 340 | from.count = ms->ti->len & (region_size - 1); |
1da177e4 | 341 | if (!from.count) |
1f965b19 | 342 | from.count = region_size; |
1da177e4 | 343 | } else |
1f965b19 | 344 | from.count = region_size; |
1da177e4 LT |
345 | |
346 | /* fill in the destinations */ | |
347 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
72f4b314 | 348 | if (&ms->mirror[i] == get_default_mirror(ms)) |
1da177e4 LT |
349 | continue; |
350 | ||
351 | m = ms->mirror + i; | |
352 | dest->bdev = m->dev->bdev; | |
1f965b19 | 353 | dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key); |
1da177e4 LT |
354 | dest->count = from.count; |
355 | dest++; | |
356 | } | |
357 | ||
358 | /* hand to kcopyd */ | |
f7c83e2e JB |
359 | if (!errors_handled(ms)) |
360 | set_bit(DM_KCOPYD_IGNORE_ERROR, &flags); | |
361 | ||
eb69aca5 HM |
362 | r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, |
363 | flags, recovery_complete, reg); | |
1da177e4 LT |
364 | |
365 | return r; | |
366 | } | |
367 | ||
368 | static void do_recovery(struct mirror_set *ms) | |
369 | { | |
1f965b19 HM |
370 | struct dm_region *reg; |
371 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
1da177e4 | 372 | int r; |
1da177e4 LT |
373 | |
374 | /* | |
375 | * Start quiescing some regions. | |
376 | */ | |
1f965b19 | 377 | dm_rh_recovery_prepare(ms->rh); |
1da177e4 LT |
378 | |
379 | /* | |
380 | * Copy any already quiesced regions. | |
381 | */ | |
1f965b19 | 382 | while ((reg = dm_rh_recovery_start(ms->rh))) { |
1da177e4 LT |
383 | r = recover(ms, reg); |
384 | if (r) | |
1f965b19 | 385 | dm_rh_recovery_end(reg, 0); |
1da177e4 LT |
386 | } |
387 | ||
388 | /* | |
389 | * Update the in sync flag. | |
390 | */ | |
391 | if (!ms->in_sync && | |
392 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
393 | /* the sync is complete */ | |
394 | dm_table_event(ms->ti->table); | |
395 | ms->in_sync = 1; | |
396 | } | |
397 | } | |
398 | ||
399 | /*----------------------------------------------------------------- | |
400 | * Reads | |
401 | *---------------------------------------------------------------*/ | |
402 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
403 | { | |
06386bbf JB |
404 | struct mirror *m = get_default_mirror(ms); |
405 | ||
406 | do { | |
407 | if (likely(!atomic_read(&m->error_count))) | |
408 | return m; | |
409 | ||
410 | if (m-- == ms->mirror) | |
411 | m += ms->nr_mirrors; | |
412 | } while (m != get_default_mirror(ms)); | |
413 | ||
414 | return NULL; | |
415 | } | |
416 | ||
417 | static int default_ok(struct mirror *m) | |
418 | { | |
419 | struct mirror *default_mirror = get_default_mirror(m->ms); | |
420 | ||
421 | return !atomic_read(&default_mirror->error_count); | |
422 | } | |
423 | ||
424 | static int mirror_available(struct mirror_set *ms, struct bio *bio) | |
425 | { | |
1f965b19 HM |
426 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
427 | region_t region = dm_rh_bio_to_region(ms->rh, bio); | |
06386bbf | 428 | |
1f965b19 | 429 | if (log->type->in_sync(log, region, 0)) |
06386bbf JB |
430 | return choose_mirror(ms, bio->bi_sector) ? 1 : 0; |
431 | ||
432 | return 0; | |
1da177e4 LT |
433 | } |
434 | ||
435 | /* | |
436 | * remap a buffer to a particular mirror. | |
437 | */ | |
06386bbf JB |
438 | static sector_t map_sector(struct mirror *m, struct bio *bio) |
439 | { | |
4184153f MP |
440 | if (unlikely(!bio->bi_size)) |
441 | return 0; | |
b441a262 | 442 | return m->offset + dm_target_offset(m->ms->ti, bio->bi_sector); |
06386bbf JB |
443 | } |
444 | ||
445 | static void map_bio(struct mirror *m, struct bio *bio) | |
1da177e4 LT |
446 | { |
447 | bio->bi_bdev = m->dev->bdev; | |
06386bbf JB |
448 | bio->bi_sector = map_sector(m, bio); |
449 | } | |
450 | ||
22a1ceb1 | 451 | static void map_region(struct dm_io_region *io, struct mirror *m, |
06386bbf JB |
452 | struct bio *bio) |
453 | { | |
454 | io->bdev = m->dev->bdev; | |
455 | io->sector = map_sector(m, bio); | |
456 | io->count = bio->bi_size >> 9; | |
457 | } | |
458 | ||
04788507 MP |
459 | static void hold_bio(struct mirror_set *ms, struct bio *bio) |
460 | { | |
461 | /* | |
f0703040 TY |
462 | * Lock is required to avoid race condition during suspend |
463 | * process. | |
04788507 | 464 | */ |
f0703040 TY |
465 | spin_lock_irq(&ms->lock); |
466 | ||
04788507 | 467 | if (atomic_read(&ms->suspend)) { |
f0703040 TY |
468 | spin_unlock_irq(&ms->lock); |
469 | ||
470 | /* | |
471 | * If device is suspended, complete the bio. | |
472 | */ | |
04788507 MP |
473 | if (dm_noflush_suspending(ms->ti)) |
474 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
475 | else | |
476 | bio_endio(bio, -EIO); | |
477 | return; | |
478 | } | |
479 | ||
480 | /* | |
481 | * Hold bio until the suspend is complete. | |
482 | */ | |
04788507 MP |
483 | bio_list_add(&ms->holds, bio); |
484 | spin_unlock_irq(&ms->lock); | |
485 | } | |
486 | ||
06386bbf JB |
487 | /*----------------------------------------------------------------- |
488 | * Reads | |
489 | *---------------------------------------------------------------*/ | |
490 | static void read_callback(unsigned long error, void *context) | |
491 | { | |
492 | struct bio *bio = context; | |
493 | struct mirror *m; | |
494 | ||
495 | m = bio_get_m(bio); | |
496 | bio_set_m(bio, NULL); | |
497 | ||
498 | if (likely(!error)) { | |
499 | bio_endio(bio, 0); | |
500 | return; | |
501 | } | |
502 | ||
503 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
504 | ||
505 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
506 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
507 | "Trying alternative device.", | |
508 | m->dev->name); | |
509 | queue_bio(m->ms, bio, bio_rw(bio)); | |
510 | return; | |
511 | } | |
512 | ||
513 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
514 | m->dev->name); | |
515 | bio_endio(bio, -EIO); | |
516 | } | |
517 | ||
518 | /* Asynchronous read. */ | |
519 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
520 | { | |
22a1ceb1 | 521 | struct dm_io_region io; |
06386bbf JB |
522 | struct dm_io_request io_req = { |
523 | .bi_rw = READ, | |
524 | .mem.type = DM_IO_BVEC, | |
525 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
526 | .notify.fn = read_callback, | |
527 | .notify.context = bio, | |
528 | .client = m->ms->io_client, | |
529 | }; | |
530 | ||
531 | map_region(&io, m, bio); | |
532 | bio_set_m(bio, m); | |
1f965b19 HM |
533 | BUG_ON(dm_io(&io_req, 1, &io, NULL)); |
534 | } | |
535 | ||
536 | static inline int region_in_sync(struct mirror_set *ms, region_t region, | |
537 | int may_block) | |
538 | { | |
539 | int state = dm_rh_get_state(ms->rh, region, may_block); | |
540 | return state == DM_RH_CLEAN || state == DM_RH_DIRTY; | |
1da177e4 LT |
541 | } |
542 | ||
543 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
544 | { | |
545 | region_t region; | |
546 | struct bio *bio; | |
547 | struct mirror *m; | |
548 | ||
549 | while ((bio = bio_list_pop(reads))) { | |
1f965b19 | 550 | region = dm_rh_bio_to_region(ms->rh, bio); |
06386bbf | 551 | m = get_default_mirror(ms); |
1da177e4 LT |
552 | |
553 | /* | |
554 | * We can only read balance if the region is in sync. | |
555 | */ | |
1f965b19 | 556 | if (likely(region_in_sync(ms, region, 1))) |
1da177e4 | 557 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
558 | else if (m && atomic_read(&m->error_count)) |
559 | m = NULL; | |
1da177e4 | 560 | |
06386bbf JB |
561 | if (likely(m)) |
562 | read_async_bio(m, bio); | |
563 | else | |
564 | bio_endio(bio, -EIO); | |
1da177e4 LT |
565 | } |
566 | } | |
567 | ||
568 | /*----------------------------------------------------------------- | |
569 | * Writes. | |
570 | * | |
571 | * We do different things with the write io depending on the | |
572 | * state of the region that it's in: | |
573 | * | |
574 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
575 | * RECOVERING: delay the io until recovery completes | |
576 | * NOSYNC: increment pending, just write to the default mirror | |
577 | *---------------------------------------------------------------*/ | |
72f4b314 | 578 | |
72f4b314 | 579 | |
1da177e4 LT |
580 | static void write_callback(unsigned long error, void *context) |
581 | { | |
72f4b314 | 582 | unsigned i, ret = 0; |
1da177e4 LT |
583 | struct bio *bio = (struct bio *) context; |
584 | struct mirror_set *ms; | |
72f4b314 JB |
585 | int should_wake = 0; |
586 | unsigned long flags; | |
1da177e4 | 587 | |
06386bbf JB |
588 | ms = bio_get_m(bio)->ms; |
589 | bio_set_m(bio, NULL); | |
1da177e4 LT |
590 | |
591 | /* | |
592 | * NOTE: We don't decrement the pending count here, | |
593 | * instead it is done by the targets endio function. | |
594 | * This way we handle both writes to SYNC and NOSYNC | |
595 | * regions with the same code. | |
596 | */ | |
60f355ea MP |
597 | if (likely(!error)) { |
598 | bio_endio(bio, ret); | |
599 | return; | |
600 | } | |
1da177e4 | 601 | |
72f4b314 JB |
602 | for (i = 0; i < ms->nr_mirrors; i++) |
603 | if (test_bit(i, &error)) | |
604 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
72f4b314 | 605 | |
60f355ea MP |
606 | /* |
607 | * Need to raise event. Since raising | |
608 | * events can block, we need to do it in | |
609 | * the main thread. | |
610 | */ | |
611 | spin_lock_irqsave(&ms->lock, flags); | |
612 | if (!ms->failures.head) | |
613 | should_wake = 1; | |
614 | bio_list_add(&ms->failures, bio); | |
615 | spin_unlock_irqrestore(&ms->lock, flags); | |
616 | if (should_wake) | |
617 | wakeup_mirrord(ms); | |
1da177e4 LT |
618 | } |
619 | ||
620 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
621 | { | |
622 | unsigned int i; | |
22a1ceb1 | 623 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 624 | struct mirror *m; |
88be163a | 625 | struct dm_io_request io_req = { |
d87f4c14 | 626 | .bi_rw = WRITE | (bio->bi_rw & WRITE_FLUSH_FUA), |
88be163a MB |
627 | .mem.type = DM_IO_BVEC, |
628 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
629 | .notify.fn = write_callback, | |
630 | .notify.context = bio, | |
631 | .client = ms->io_client, | |
632 | }; | |
1da177e4 | 633 | |
5fc2ffea MS |
634 | if (bio->bi_rw & REQ_DISCARD) { |
635 | io_req.bi_rw |= REQ_DISCARD; | |
636 | io_req.mem.type = DM_IO_KMEM; | |
637 | io_req.mem.ptr.addr = NULL; | |
638 | } | |
639 | ||
06386bbf JB |
640 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
641 | map_region(dest++, m, bio); | |
1da177e4 | 642 | |
06386bbf JB |
643 | /* |
644 | * Use default mirror because we only need it to retrieve the reference | |
645 | * to the mirror set in write_callback(). | |
646 | */ | |
647 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a | 648 | |
1f965b19 | 649 | BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL)); |
1da177e4 LT |
650 | } |
651 | ||
652 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
653 | { | |
654 | int state; | |
655 | struct bio *bio; | |
656 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
7513c2a7 JB |
657 | struct bio_list requeue; |
658 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); | |
659 | region_t region; | |
1da177e4 LT |
660 | |
661 | if (!writes->head) | |
662 | return; | |
663 | ||
664 | /* | |
665 | * Classify each write. | |
666 | */ | |
667 | bio_list_init(&sync); | |
668 | bio_list_init(&nosync); | |
669 | bio_list_init(&recover); | |
7513c2a7 | 670 | bio_list_init(&requeue); |
1da177e4 LT |
671 | |
672 | while ((bio = bio_list_pop(writes))) { | |
5fc2ffea MS |
673 | if ((bio->bi_rw & REQ_FLUSH) || |
674 | (bio->bi_rw & REQ_DISCARD)) { | |
4184153f MP |
675 | bio_list_add(&sync, bio); |
676 | continue; | |
677 | } | |
678 | ||
7513c2a7 JB |
679 | region = dm_rh_bio_to_region(ms->rh, bio); |
680 | ||
681 | if (log->type->is_remote_recovering && | |
682 | log->type->is_remote_recovering(log, region)) { | |
683 | bio_list_add(&requeue, bio); | |
684 | continue; | |
685 | } | |
686 | ||
687 | state = dm_rh_get_state(ms->rh, region, 1); | |
1da177e4 | 688 | switch (state) { |
1f965b19 HM |
689 | case DM_RH_CLEAN: |
690 | case DM_RH_DIRTY: | |
1da177e4 LT |
691 | this_list = &sync; |
692 | break; | |
693 | ||
1f965b19 | 694 | case DM_RH_NOSYNC: |
1da177e4 LT |
695 | this_list = &nosync; |
696 | break; | |
697 | ||
1f965b19 | 698 | case DM_RH_RECOVERING: |
1da177e4 LT |
699 | this_list = &recover; |
700 | break; | |
701 | } | |
702 | ||
703 | bio_list_add(this_list, bio); | |
704 | } | |
705 | ||
7513c2a7 JB |
706 | /* |
707 | * Add bios that are delayed due to remote recovery | |
708 | * back on to the write queue | |
709 | */ | |
710 | if (unlikely(requeue.head)) { | |
711 | spin_lock_irq(&ms->lock); | |
712 | bio_list_merge(&ms->writes, &requeue); | |
713 | spin_unlock_irq(&ms->lock); | |
69885683 | 714 | delayed_wake(ms); |
7513c2a7 JB |
715 | } |
716 | ||
1da177e4 LT |
717 | /* |
718 | * Increment the pending counts for any regions that will | |
719 | * be written to (writes to recover regions are going to | |
720 | * be delayed). | |
721 | */ | |
1f965b19 HM |
722 | dm_rh_inc_pending(ms->rh, &sync); |
723 | dm_rh_inc_pending(ms->rh, &nosync); | |
d2b69864 JB |
724 | |
725 | /* | |
726 | * If the flush fails on a previous call and succeeds here, | |
727 | * we must not reset the log_failure variable. We need | |
728 | * userspace interaction to do that. | |
729 | */ | |
730 | ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure; | |
1da177e4 LT |
731 | |
732 | /* | |
733 | * Dispatch io. | |
734 | */ | |
5528d17d | 735 | if (unlikely(ms->log_failure) && errors_handled(ms)) { |
b80aa7a0 JB |
736 | spin_lock_irq(&ms->lock); |
737 | bio_list_merge(&ms->failures, &sync); | |
738 | spin_unlock_irq(&ms->lock); | |
1f965b19 | 739 | wakeup_mirrord(ms); |
b80aa7a0 | 740 | } else |
fc1ff958 | 741 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 742 | do_write(ms, bio); |
1da177e4 LT |
743 | |
744 | while ((bio = bio_list_pop(&recover))) | |
1f965b19 | 745 | dm_rh_delay(ms->rh, bio); |
1da177e4 LT |
746 | |
747 | while ((bio = bio_list_pop(&nosync))) { | |
ede5ea0b MP |
748 | if (unlikely(ms->leg_failure) && errors_handled(ms)) { |
749 | spin_lock_irq(&ms->lock); | |
750 | bio_list_add(&ms->failures, bio); | |
751 | spin_unlock_irq(&ms->lock); | |
752 | wakeup_mirrord(ms); | |
753 | } else { | |
929be8fc MP |
754 | map_bio(get_default_mirror(ms), bio); |
755 | generic_make_request(bio); | |
756 | } | |
1da177e4 LT |
757 | } |
758 | } | |
759 | ||
72f4b314 JB |
760 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
761 | { | |
762 | struct bio *bio; | |
763 | ||
0f398a84 | 764 | if (likely(!failures->head)) |
72f4b314 JB |
765 | return; |
766 | ||
b80aa7a0 JB |
767 | /* |
768 | * If the log has failed, unattempted writes are being | |
0f398a84 | 769 | * put on the holds list. We can't issue those writes |
b80aa7a0 JB |
770 | * until a log has been marked, so we must store them. |
771 | * | |
772 | * If a 'noflush' suspend is in progress, we can requeue | |
773 | * the I/O's to the core. This give userspace a chance | |
774 | * to reconfigure the mirror, at which point the core | |
775 | * will reissue the writes. If the 'noflush' flag is | |
776 | * not set, we have no choice but to return errors. | |
777 | * | |
778 | * Some writes on the failures list may have been | |
779 | * submitted before the log failure and represent a | |
780 | * failure to write to one of the devices. It is ok | |
781 | * for us to treat them the same and requeue them | |
782 | * as well. | |
783 | */ | |
0f398a84 | 784 | while ((bio = bio_list_pop(failures))) { |
60f355ea | 785 | if (!ms->log_failure) { |
0f398a84 | 786 | ms->in_sync = 0; |
c58098be | 787 | dm_rh_mark_nosync(ms->rh, bio); |
0f398a84 | 788 | } |
60f355ea MP |
789 | |
790 | /* | |
791 | * If all the legs are dead, fail the I/O. | |
792 | * If we have been told to handle errors, hold the bio | |
793 | * and wait for userspace to deal with the problem. | |
794 | * Otherwise pretend that the I/O succeeded. (This would | |
795 | * be wrong if the failed leg returned after reboot and | |
796 | * got replicated back to the good legs.) | |
797 | */ | |
798 | if (!get_valid_mirror(ms)) | |
799 | bio_endio(bio, -EIO); | |
800 | else if (errors_handled(ms)) | |
801 | hold_bio(ms, bio); | |
802 | else | |
803 | bio_endio(bio, 0); | |
b80aa7a0 | 804 | } |
72f4b314 JB |
805 | } |
806 | ||
807 | static void trigger_event(struct work_struct *work) | |
808 | { | |
809 | struct mirror_set *ms = | |
810 | container_of(work, struct mirror_set, trigger_event); | |
811 | ||
812 | dm_table_event(ms->ti->table); | |
813 | } | |
814 | ||
1da177e4 LT |
815 | /*----------------------------------------------------------------- |
816 | * kmirrord | |
817 | *---------------------------------------------------------------*/ | |
a2aebe03 | 818 | static void do_mirror(struct work_struct *work) |
1da177e4 | 819 | { |
1f965b19 HM |
820 | struct mirror_set *ms = container_of(work, struct mirror_set, |
821 | kmirrord_work); | |
72f4b314 JB |
822 | struct bio_list reads, writes, failures; |
823 | unsigned long flags; | |
1da177e4 | 824 | |
72f4b314 | 825 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
826 | reads = ms->reads; |
827 | writes = ms->writes; | |
72f4b314 | 828 | failures = ms->failures; |
1da177e4 LT |
829 | bio_list_init(&ms->reads); |
830 | bio_list_init(&ms->writes); | |
72f4b314 JB |
831 | bio_list_init(&ms->failures); |
832 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 | 833 | |
1f965b19 | 834 | dm_rh_update_states(ms->rh, errors_handled(ms)); |
1da177e4 LT |
835 | do_recovery(ms); |
836 | do_reads(ms, &reads); | |
837 | do_writes(ms, &writes); | |
72f4b314 | 838 | do_failures(ms, &failures); |
1da177e4 LT |
839 | } |
840 | ||
1da177e4 LT |
841 | /*----------------------------------------------------------------- |
842 | * Target functions | |
843 | *---------------------------------------------------------------*/ | |
844 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
845 | uint32_t region_size, | |
846 | struct dm_target *ti, | |
416cd17b | 847 | struct dm_dirty_log *dl) |
1da177e4 LT |
848 | { |
849 | size_t len; | |
850 | struct mirror_set *ms = NULL; | |
851 | ||
1da177e4 LT |
852 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
853 | ||
dd00cc48 | 854 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 855 | if (!ms) { |
72d94861 | 856 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
857 | return NULL; |
858 | } | |
859 | ||
1da177e4 | 860 | spin_lock_init(&ms->lock); |
5339fc2d MP |
861 | bio_list_init(&ms->reads); |
862 | bio_list_init(&ms->writes); | |
863 | bio_list_init(&ms->failures); | |
864 | bio_list_init(&ms->holds); | |
1da177e4 LT |
865 | |
866 | ms->ti = ti; | |
867 | ms->nr_mirrors = nr_mirrors; | |
868 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
869 | ms->in_sync = 0; | |
b80aa7a0 | 870 | ms->log_failure = 0; |
929be8fc | 871 | ms->leg_failure = 0; |
b80aa7a0 | 872 | atomic_set(&ms->suspend, 0); |
72f4b314 | 873 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 874 | |
bda8efec | 875 | ms->io_client = dm_io_client_create(); |
88be163a MB |
876 | if (IS_ERR(ms->io_client)) { |
877 | ti->error = "Error creating dm_io client"; | |
878 | kfree(ms); | |
879 | return NULL; | |
880 | } | |
881 | ||
1f965b19 HM |
882 | ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord, |
883 | wakeup_all_recovery_waiters, | |
884 | ms->ti->begin, MAX_RECOVERY, | |
885 | dl, region_size, ms->nr_regions); | |
886 | if (IS_ERR(ms->rh)) { | |
72d94861 | 887 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 888 | dm_io_client_destroy(ms->io_client); |
1da177e4 LT |
889 | kfree(ms); |
890 | return NULL; | |
891 | } | |
892 | ||
893 | return ms; | |
894 | } | |
895 | ||
896 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
897 | unsigned int m) | |
898 | { | |
899 | while (m--) | |
900 | dm_put_device(ti, ms->mirror[m].dev); | |
901 | ||
88be163a | 902 | dm_io_client_destroy(ms->io_client); |
1f965b19 | 903 | dm_region_hash_destroy(ms->rh); |
1da177e4 LT |
904 | kfree(ms); |
905 | } | |
906 | ||
1da177e4 LT |
907 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, |
908 | unsigned int mirror, char **argv) | |
909 | { | |
4ee218cd | 910 | unsigned long long offset; |
31998ef1 | 911 | char dummy; |
1da177e4 | 912 | |
31998ef1 | 913 | if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1) { |
72d94861 | 914 | ti->error = "Invalid offset"; |
1da177e4 LT |
915 | return -EINVAL; |
916 | } | |
917 | ||
8215d6ec | 918 | if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), |
1da177e4 | 919 | &ms->mirror[mirror].dev)) { |
72d94861 | 920 | ti->error = "Device lookup failure"; |
1da177e4 LT |
921 | return -ENXIO; |
922 | } | |
923 | ||
aa5617c5 | 924 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
925 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
926 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
927 | ms->mirror[mirror].offset = offset; |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
1da177e4 LT |
932 | /* |
933 | * Create dirty log: log_type #log_params <log_params> | |
934 | */ | |
416cd17b | 935 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1f965b19 HM |
936 | unsigned argc, char **argv, |
937 | unsigned *args_used) | |
1da177e4 | 938 | { |
1f965b19 | 939 | unsigned param_count; |
416cd17b | 940 | struct dm_dirty_log *dl; |
31998ef1 | 941 | char dummy; |
1da177e4 LT |
942 | |
943 | if (argc < 2) { | |
72d94861 | 944 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
945 | return NULL; |
946 | } | |
947 | ||
31998ef1 | 948 | if (sscanf(argv[1], "%u%c", ¶m_count, &dummy) != 1) { |
72d94861 | 949 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
950 | return NULL; |
951 | } | |
952 | ||
953 | *args_used = 2 + param_count; | |
954 | ||
955 | if (argc < *args_used) { | |
72d94861 | 956 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
957 | return NULL; |
958 | } | |
959 | ||
c0da3748 MP |
960 | dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count, |
961 | argv + 2); | |
1da177e4 | 962 | if (!dl) { |
72d94861 | 963 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
964 | return NULL; |
965 | } | |
966 | ||
1da177e4 LT |
967 | return dl; |
968 | } | |
969 | ||
a8e6afa2 JB |
970 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
971 | unsigned *args_used) | |
972 | { | |
973 | unsigned num_features; | |
974 | struct dm_target *ti = ms->ti; | |
31998ef1 | 975 | char dummy; |
a8e6afa2 JB |
976 | |
977 | *args_used = 0; | |
978 | ||
979 | if (!argc) | |
980 | return 0; | |
981 | ||
31998ef1 | 982 | if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) { |
a8e6afa2 JB |
983 | ti->error = "Invalid number of features"; |
984 | return -EINVAL; | |
985 | } | |
986 | ||
987 | argc--; | |
988 | argv++; | |
989 | (*args_used)++; | |
990 | ||
991 | if (num_features > argc) { | |
992 | ti->error = "Not enough arguments to support feature count"; | |
993 | return -EINVAL; | |
994 | } | |
995 | ||
996 | if (!strcmp("handle_errors", argv[0])) | |
997 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
998 | else { | |
999 | ti->error = "Unrecognised feature requested"; | |
1000 | return -EINVAL; | |
1001 | } | |
1002 | ||
1003 | (*args_used)++; | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | ||
1da177e4 LT |
1008 | /* |
1009 | * Construct a mirror mapping: | |
1010 | * | |
1011 | * log_type #log_params <log_params> | |
1012 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1013 | * [#features <features>] |
1da177e4 LT |
1014 | * |
1015 | * log_type is "core" or "disk" | |
1016 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1017 | * |
1018 | * If present, features must be "handle_errors". | |
1da177e4 | 1019 | */ |
1da177e4 LT |
1020 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1021 | { | |
1022 | int r; | |
1023 | unsigned int nr_mirrors, m, args_used; | |
1024 | struct mirror_set *ms; | |
416cd17b | 1025 | struct dm_dirty_log *dl; |
31998ef1 | 1026 | char dummy; |
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 | ||
31998ef1 | 1035 | if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 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; | |
542f9038 MS |
1068 | |
1069 | r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh)); | |
1070 | if (r) | |
1071 | goto err_free_context; | |
1072 | ||
4184153f | 1073 | ti->num_flush_requests = 1; |
5fc2ffea | 1074 | ti->num_discard_requests = 1; |
39cf0ed2 | 1075 | ti->per_bio_data_size = sizeof(struct dm_raid1_read_record); |
0ac55489 | 1076 | ti->discard_zeroes_data_unsupported = true; |
1da177e4 | 1077 | |
9c4376de TH |
1078 | ms->kmirrord_wq = alloc_workqueue("kmirrord", |
1079 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
6ad36fe2 HS |
1080 | if (!ms->kmirrord_wq) { |
1081 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1082 | r = -ENOMEM; |
1083 | goto err_free_context; | |
6ad36fe2 HS |
1084 | } |
1085 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
1086 | init_timer(&ms->timer); |
1087 | ms->timer_pending = 0; | |
72f4b314 | 1088 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 1089 | |
a8e6afa2 | 1090 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1091 | if (r) |
1092 | goto err_destroy_wq; | |
a8e6afa2 JB |
1093 | |
1094 | argv += args_used; | |
1095 | argc -= args_used; | |
1096 | ||
f44db678 JB |
1097 | /* |
1098 | * Any read-balancing addition depends on the | |
1099 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1100 | * This is because the decision to balance depends | |
1101 | * on the sync state of a region. If the above | |
1102 | * flag is not present, we ignore errors; and | |
1103 | * the sync state may be inaccurate. | |
1104 | */ | |
1105 | ||
a8e6afa2 JB |
1106 | if (argc) { |
1107 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1108 | r = -EINVAL; |
1109 | goto err_destroy_wq; | |
a8e6afa2 JB |
1110 | } |
1111 | ||
fa34ce73 MP |
1112 | ms->kcopyd_client = dm_kcopyd_client_create(); |
1113 | if (IS_ERR(ms->kcopyd_client)) { | |
1114 | r = PTR_ERR(ms->kcopyd_client); | |
a72cf737 | 1115 | goto err_destroy_wq; |
fa34ce73 | 1116 | } |
1da177e4 | 1117 | |
1f965b19 | 1118 | wakeup_mirrord(ms); |
1da177e4 | 1119 | return 0; |
a72cf737 DM |
1120 | |
1121 | err_destroy_wq: | |
1122 | destroy_workqueue(ms->kmirrord_wq); | |
1123 | err_free_context: | |
1124 | free_context(ms, ti, ms->nr_mirrors); | |
1125 | return r; | |
1da177e4 LT |
1126 | } |
1127 | ||
1128 | static void mirror_dtr(struct dm_target *ti) | |
1129 | { | |
1130 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1131 | ||
a2aebe03 | 1132 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1133 | flush_workqueue(ms->kmirrord_wq); |
43829731 | 1134 | flush_work(&ms->trigger_event); |
eb69aca5 | 1135 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1136 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1137 | free_context(ms, ti, ms->nr_mirrors); |
1138 | } | |
1139 | ||
1da177e4 LT |
1140 | /* |
1141 | * Mirror mapping function | |
1142 | */ | |
1143 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1144 | union map_info *map_context) | |
1145 | { | |
1146 | int r, rw = bio_rw(bio); | |
1147 | struct mirror *m; | |
1148 | struct mirror_set *ms = ti->private; | |
39cf0ed2 | 1149 | struct dm_raid1_read_record *read_record; |
1f965b19 | 1150 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 LT |
1151 | |
1152 | if (rw == WRITE) { | |
06386bbf | 1153 | /* Save region for mirror_end_io() handler */ |
1f965b19 | 1154 | map_context->ll = dm_rh_bio_to_region(ms->rh, bio); |
1da177e4 | 1155 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1156 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1157 | } |
1158 | ||
1f965b19 | 1159 | r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0); |
1da177e4 LT |
1160 | if (r < 0 && r != -EWOULDBLOCK) |
1161 | return r; | |
1162 | ||
1da177e4 | 1163 | /* |
06386bbf | 1164 | * If region is not in-sync queue the bio. |
1da177e4 | 1165 | */ |
06386bbf JB |
1166 | if (!r || (r == -EWOULDBLOCK)) { |
1167 | if (rw == READA) | |
1168 | return -EWOULDBLOCK; | |
1da177e4 | 1169 | |
1da177e4 | 1170 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1171 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1172 | } |
1173 | ||
06386bbf JB |
1174 | /* |
1175 | * The region is in-sync and we can perform reads directly. | |
1176 | * Store enough information so we can retry if it fails. | |
1177 | */ | |
1da177e4 | 1178 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1179 | if (unlikely(!m)) |
1da177e4 LT |
1180 | return -EIO; |
1181 | ||
39cf0ed2 | 1182 | read_record = dm_per_bio_data(bio, sizeof(struct dm_raid1_read_record)); |
7c27213b MP |
1183 | dm_bio_record(&read_record->details, bio); |
1184 | map_context->ptr = read_record; | |
1185 | read_record->m = m; | |
06386bbf JB |
1186 | |
1187 | map_bio(m, bio); | |
1188 | ||
d2a7ad29 | 1189 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1190 | } |
1191 | ||
1192 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1193 | int error, union map_info *map_context) | |
1194 | { | |
1195 | int rw = bio_rw(bio); | |
1196 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1197 | struct mirror *m = NULL; |
1198 | struct dm_bio_details *bd = NULL; | |
1199 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1200 | |
1201 | /* | |
1202 | * We need to dec pending if this was a write. | |
1203 | */ | |
06386bbf | 1204 | if (rw == WRITE) { |
751f188d | 1205 | if (!(bio->bi_rw & (REQ_FLUSH | REQ_DISCARD))) |
4184153f | 1206 | dm_rh_dec(ms->rh, map_context->ll); |
06386bbf JB |
1207 | return error; |
1208 | } | |
1da177e4 | 1209 | |
06386bbf JB |
1210 | if (error == -EOPNOTSUPP) |
1211 | goto out; | |
1212 | ||
7b6d91da | 1213 | if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) |
06386bbf JB |
1214 | goto out; |
1215 | ||
1216 | if (unlikely(error)) { | |
1217 | if (!read_record) { | |
1218 | /* | |
1219 | * There wasn't enough memory to record necessary | |
1220 | * information for a retry or there was no other | |
1221 | * mirror in-sync. | |
1222 | */ | |
e03f1a84 | 1223 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1224 | return -EIO; |
1225 | } | |
e03f1a84 AB |
1226 | |
1227 | m = read_record->m; | |
1228 | ||
06386bbf JB |
1229 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1230 | m->dev->name); | |
1231 | ||
06386bbf JB |
1232 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1233 | ||
1234 | /* | |
1235 | * A failed read is requeued for another attempt using an intact | |
1236 | * mirror. | |
1237 | */ | |
1238 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1239 | bd = &read_record->details; | |
1240 | ||
1241 | dm_bio_restore(bd, bio); | |
06386bbf JB |
1242 | map_context->ptr = NULL; |
1243 | queue_bio(ms, bio, rw); | |
19cbbc60 | 1244 | return DM_ENDIO_INCOMPLETE; |
06386bbf JB |
1245 | } |
1246 | DMERR("All replicated volumes dead, failing I/O"); | |
1247 | } | |
1248 | ||
1249 | out: | |
39cf0ed2 | 1250 | map_context->ptr = NULL; |
06386bbf JB |
1251 | |
1252 | return error; | |
1da177e4 LT |
1253 | } |
1254 | ||
b80aa7a0 | 1255 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1256 | { |
1257 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1f965b19 | 1258 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
1da177e4 | 1259 | |
04788507 MP |
1260 | struct bio_list holds; |
1261 | struct bio *bio; | |
1262 | ||
b80aa7a0 JB |
1263 | atomic_set(&ms->suspend, 1); |
1264 | ||
f0703040 TY |
1265 | /* |
1266 | * Process bios in the hold list to start recovery waiting | |
1267 | * for bios in the hold list. After the process, no bio has | |
1268 | * a chance to be added in the hold list because ms->suspend | |
1269 | * is set. | |
1270 | */ | |
1271 | spin_lock_irq(&ms->lock); | |
1272 | holds = ms->holds; | |
1273 | bio_list_init(&ms->holds); | |
1274 | spin_unlock_irq(&ms->lock); | |
1275 | ||
1276 | while ((bio = bio_list_pop(&holds))) | |
1277 | hold_bio(ms, bio); | |
1278 | ||
b80aa7a0 JB |
1279 | /* |
1280 | * We must finish up all the work that we've | |
1281 | * generated (i.e. recovery work). | |
1282 | */ | |
1f965b19 | 1283 | dm_rh_stop_recovery(ms->rh); |
33184048 | 1284 | |
33184048 | 1285 | wait_event(_kmirrord_recovery_stopped, |
1f965b19 | 1286 | !dm_rh_recovery_in_flight(ms->rh)); |
33184048 | 1287 | |
b80aa7a0 JB |
1288 | if (log->type->presuspend && log->type->presuspend(log)) |
1289 | /* FIXME: need better error handling */ | |
1290 | DMWARN("log presuspend failed"); | |
1291 | ||
1292 | /* | |
1293 | * Now that recovery is complete/stopped and the | |
1294 | * delayed bios are queued, we need to wait for | |
1295 | * the worker thread to complete. This way, | |
1296 | * we know that all of our I/O has been pushed. | |
1297 | */ | |
1298 | flush_workqueue(ms->kmirrord_wq); | |
1299 | } | |
1300 | ||
1301 | static void mirror_postsuspend(struct dm_target *ti) | |
1302 | { | |
1303 | struct mirror_set *ms = ti->private; | |
1f965b19 | 1304 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 | 1305 | |
6b3df0d7 | 1306 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1307 | /* FIXME: need better error handling */ |
b80aa7a0 | 1308 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1309 | } |
1310 | ||
1311 | static void mirror_resume(struct dm_target *ti) | |
1312 | { | |
b80aa7a0 | 1313 | struct mirror_set *ms = ti->private; |
1f965b19 | 1314 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
b80aa7a0 JB |
1315 | |
1316 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1317 | if (log->type->resume && log->type->resume(log)) |
1318 | /* FIXME: need better error handling */ | |
1319 | DMWARN("log resume failed"); | |
1f965b19 | 1320 | dm_rh_start_recovery(ms->rh); |
1da177e4 LT |
1321 | } |
1322 | ||
af195ac8 JB |
1323 | /* |
1324 | * device_status_char | |
1325 | * @m: mirror device/leg we want the status of | |
1326 | * | |
1327 | * We return one character representing the most severe error | |
1328 | * we have encountered. | |
1329 | * A => Alive - No failures | |
1330 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1331 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1332 | * R => Read - A read failure occurred, mirror data unaffected | |
1333 | * | |
1334 | * Returns: <char> | |
1335 | */ | |
1336 | static char device_status_char(struct mirror *m) | |
1337 | { | |
1338 | if (!atomic_read(&(m->error_count))) | |
1339 | return 'A'; | |
1340 | ||
64b30c46 MP |
1341 | return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' : |
1342 | (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
af195ac8 JB |
1343 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : |
1344 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1345 | } | |
1346 | ||
1347 | ||
1da177e4 | 1348 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1f4e0ff0 | 1349 | unsigned status_flags, char *result, unsigned maxlen) |
1da177e4 | 1350 | { |
315dcc22 | 1351 | unsigned int m, sz = 0; |
1da177e4 | 1352 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1f965b19 | 1353 | struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh); |
af195ac8 | 1354 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1355 | |
1da177e4 LT |
1356 | switch (type) { |
1357 | case STATUSTYPE_INFO: | |
1358 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1359 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1360 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1361 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1362 | } | |
1363 | buffer[m] = '\0'; | |
1da177e4 | 1364 | |
af195ac8 | 1365 | DMEMIT("%llu/%llu 1 %s ", |
1f965b19 | 1366 | (unsigned long long)log->type->get_sync_count(log), |
af195ac8 | 1367 | (unsigned long long)ms->nr_regions, buffer); |
315dcc22 | 1368 | |
1f965b19 | 1369 | sz += log->type->status(log, type, result+sz, maxlen-sz); |
315dcc22 | 1370 | |
1da177e4 LT |
1371 | break; |
1372 | ||
1373 | case STATUSTYPE_TABLE: | |
1f965b19 | 1374 | sz = log->type->status(log, type, result, maxlen); |
315dcc22 | 1375 | |
e52b8f6d | 1376 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1377 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1378 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1379 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1380 | |
1381 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1382 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1383 | } |
1384 | ||
1385 | return 0; | |
1386 | } | |
1387 | ||
af4874e0 MS |
1388 | static int mirror_iterate_devices(struct dm_target *ti, |
1389 | iterate_devices_callout_fn fn, void *data) | |
1390 | { | |
1391 | struct mirror_set *ms = ti->private; | |
1392 | int ret = 0; | |
1393 | unsigned i; | |
1394 | ||
1395 | for (i = 0; !ret && i < ms->nr_mirrors; i++) | |
1396 | ret = fn(ti, ms->mirror[i].dev, | |
5dea271b | 1397 | ms->mirror[i].offset, ti->len, data); |
af4874e0 MS |
1398 | |
1399 | return ret; | |
1400 | } | |
1401 | ||
1da177e4 LT |
1402 | static struct target_type mirror_target = { |
1403 | .name = "mirror", | |
39cf0ed2 | 1404 | .version = {1, 13, 1}, |
1da177e4 LT |
1405 | .module = THIS_MODULE, |
1406 | .ctr = mirror_ctr, | |
1407 | .dtr = mirror_dtr, | |
1408 | .map = mirror_map, | |
1409 | .end_io = mirror_end_io, | |
b80aa7a0 | 1410 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1411 | .postsuspend = mirror_postsuspend, |
1412 | .resume = mirror_resume, | |
1413 | .status = mirror_status, | |
af4874e0 | 1414 | .iterate_devices = mirror_iterate_devices, |
1da177e4 LT |
1415 | }; |
1416 | ||
1417 | static int __init dm_mirror_init(void) | |
1418 | { | |
1419 | int r; | |
1420 | ||
1da177e4 | 1421 | r = dm_register_target(&mirror_target); |
95f8fac8 | 1422 | if (r < 0) { |
0cd33124 | 1423 | DMERR("Failed to register mirror target"); |
95f8fac8 MP |
1424 | goto bad_target; |
1425 | } | |
1426 | ||
1427 | return 0; | |
1da177e4 | 1428 | |
95f8fac8 | 1429 | bad_target: |
1da177e4 LT |
1430 | return r; |
1431 | } | |
1432 | ||
1433 | static void __exit dm_mirror_exit(void) | |
1434 | { | |
10d3bd09 | 1435 | dm_unregister_target(&mirror_target); |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | /* Module hooks */ | |
1439 | module_init(dm_mirror_init); | |
1440 | module_exit(dm_mirror_exit); | |
1441 | ||
1442 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1443 | MODULE_AUTHOR("Joe Thornber"); | |
1444 | MODULE_LICENSE("GPL"); |