blob: 98a633f3d6b002673e17e797b34c30903a2bfa39 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * dm-snapshot.c
3 *
4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
5 *
6 * This file is released under the GPL.
7 */
8
9#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/ctype.h>
11#include <linux/device-mapper.h>
12#include <linux/fs.h>
13#include <linux/init.h>
14#include <linux/kdev_t.h>
15#include <linux/list.h>
16#include <linux/mempool.h>
17#include <linux/module.h>
18#include <linux/slab.h>
19#include <linux/vmalloc.h>
20
21#include "dm-snap.h"
22#include "dm-bio-list.h"
23#include "kcopyd.h"
24
Alasdair G Kergon72d94862006-06-26 00:27:35 -070025#define DM_MSG_PREFIX "snapshots"
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027/*
28 * The percentage increment we will wake up users at
29 */
30#define WAKE_UP_PERCENT 5
31
32/*
33 * kcopyd priority of snapshot operations
34 */
35#define SNAPSHOT_COPY_PRIORITY 2
36
37/*
38 * Each snapshot reserves this many pages for io
39 */
40#define SNAPSHOT_PAGES 256
41
Adrian Bunkc642f9e2006-12-08 02:41:13 -080042static struct workqueue_struct *ksnapd;
David Howellsc4028952006-11-22 14:57:56 +000043static void flush_queued_bios(struct work_struct *work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -070044
Alasdair G Kergon028867a2007-07-12 17:26:32 +010045struct dm_snap_pending_exception {
46 struct dm_snap_exception e;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48 /*
49 * Origin buffers waiting for this to complete are held
50 * in a bio list
51 */
52 struct bio_list origin_bios;
53 struct bio_list snapshot_bios;
54
55 /*
Alasdair G Kergoneccf0812006-03-27 01:17:42 -080056 * Short-term queue of pending exceptions prior to submission.
57 */
58 struct list_head list;
59
60 /*
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080061 * The primary pending_exception is the one that holds
Alasdair G Kergon4b832e82006-10-03 01:15:30 -070062 * the ref_count and the list of origin_bios for a
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080063 * group of pending_exceptions. It is always last to get freed.
64 * These fields get set up when writing to the origin.
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +010066 struct dm_snap_pending_exception *primary_pe;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -080067
68 /*
69 * Number of pending_exceptions processing this chunk.
70 * When this drops to zero we must complete the origin bios.
71 * If incrementing or decrementing this, hold pe->snap->lock for
72 * the sibling concerned and not pe->primary_pe->snap->lock unless
73 * they are the same.
74 */
Alasdair G Kergon4b832e82006-10-03 01:15:30 -070075 atomic_t ref_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77 /* Pointer back to snapshot context */
78 struct dm_snapshot *snap;
79
80 /*
81 * 1 indicates the exception has already been sent to
82 * kcopyd.
83 */
84 int started;
85};
86
87/*
88 * Hash table mapping origin volumes to lists of snapshots and
89 * a lock to protect it
90 */
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *exception_cache;
92static struct kmem_cache *pending_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093static mempool_t *pending_pool;
94
95/*
96 * One of these per registered origin, held in the snapshot_origins hash
97 */
98struct origin {
99 /* The origin device */
100 struct block_device *bdev;
101
102 struct list_head hash_list;
103
104 /* List of snapshots for this origin */
105 struct list_head snapshots;
106};
107
108/*
109 * Size of the hash table for origin volumes. If we make this
110 * the size of the minors list then it should be nearly perfect
111 */
112#define ORIGIN_HASH_SIZE 256
113#define ORIGIN_MASK 0xFF
114static struct list_head *_origins;
115static struct rw_semaphore _origins_lock;
116
117static int init_origin_hash(void)
118{
119 int i;
120
121 _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
122 GFP_KERNEL);
123 if (!_origins) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700124 DMERR("unable to allocate memory");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 return -ENOMEM;
126 }
127
128 for (i = 0; i < ORIGIN_HASH_SIZE; i++)
129 INIT_LIST_HEAD(_origins + i);
130 init_rwsem(&_origins_lock);
131
132 return 0;
133}
134
135static void exit_origin_hash(void)
136{
137 kfree(_origins);
138}
139
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100140static unsigned origin_hash(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 return bdev->bd_dev & ORIGIN_MASK;
143}
144
145static struct origin *__lookup_origin(struct block_device *origin)
146{
147 struct list_head *ol;
148 struct origin *o;
149
150 ol = &_origins[origin_hash(origin)];
151 list_for_each_entry (o, ol, hash_list)
152 if (bdev_equal(o->bdev, origin))
153 return o;
154
155 return NULL;
156}
157
158static void __insert_origin(struct origin *o)
159{
160 struct list_head *sl = &_origins[origin_hash(o->bdev)];
161 list_add_tail(&o->hash_list, sl);
162}
163
164/*
165 * Make a note of the snapshot and its origin so we can look it
166 * up when the origin has a write on it.
167 */
168static int register_snapshot(struct dm_snapshot *snap)
169{
170 struct origin *o;
171 struct block_device *bdev = snap->origin->bdev;
172
173 down_write(&_origins_lock);
174 o = __lookup_origin(bdev);
175
176 if (!o) {
177 /* New origin */
178 o = kmalloc(sizeof(*o), GFP_KERNEL);
179 if (!o) {
180 up_write(&_origins_lock);
181 return -ENOMEM;
182 }
183
184 /* Initialise the struct */
185 INIT_LIST_HEAD(&o->snapshots);
186 o->bdev = bdev;
187
188 __insert_origin(o);
189 }
190
191 list_add_tail(&snap->list, &o->snapshots);
192
193 up_write(&_origins_lock);
194 return 0;
195}
196
197static void unregister_snapshot(struct dm_snapshot *s)
198{
199 struct origin *o;
200
201 down_write(&_origins_lock);
202 o = __lookup_origin(s->origin->bdev);
203
204 list_del(&s->list);
205 if (list_empty(&o->snapshots)) {
206 list_del(&o->hash_list);
207 kfree(o);
208 }
209
210 up_write(&_origins_lock);
211}
212
213/*
214 * Implementation of the exception hash tables.
215 */
216static int init_exception_table(struct exception_table *et, uint32_t size)
217{
218 unsigned int i;
219
220 et->hash_mask = size - 1;
221 et->table = dm_vcalloc(size, sizeof(struct list_head));
222 if (!et->table)
223 return -ENOMEM;
224
225 for (i = 0; i < size; i++)
226 INIT_LIST_HEAD(et->table + i);
227
228 return 0;
229}
230
Christoph Lametere18b8902006-12-06 20:33:20 -0800231static void exit_exception_table(struct exception_table *et, struct kmem_cache *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 struct list_head *slot;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100234 struct dm_snap_exception *ex, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 int i, size;
236
237 size = et->hash_mask + 1;
238 for (i = 0; i < size; i++) {
239 slot = et->table + i;
240
241 list_for_each_entry_safe (ex, next, slot, hash_list)
242 kmem_cache_free(mem, ex);
243 }
244
245 vfree(et->table);
246}
247
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100248static uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
250 return chunk & et->hash_mask;
251}
252
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100253static void insert_exception(struct exception_table *eh,
254 struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
257 list_add(&e->hash_list, l);
258}
259
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100260static void remove_exception(struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 list_del(&e->hash_list);
263}
264
265/*
266 * Return the exception data for a sector, or NULL if not
267 * remapped.
268 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100269static struct dm_snap_exception *lookup_exception(struct exception_table *et,
270 chunk_t chunk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 struct list_head *slot;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100273 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 slot = &et->table[exception_hash(et, chunk)];
276 list_for_each_entry (e, slot, hash_list)
277 if (e->old_chunk == chunk)
278 return e;
279
280 return NULL;
281}
282
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100283static struct dm_snap_exception *alloc_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100285 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 e = kmem_cache_alloc(exception_cache, GFP_NOIO);
288 if (!e)
289 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
290
291 return e;
292}
293
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100294static void free_exception(struct dm_snap_exception *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
296 kmem_cache_free(exception_cache, e);
297}
298
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100299static struct dm_snap_pending_exception *alloc_pending_exception(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300{
301 return mempool_alloc(pending_pool, GFP_NOIO);
302}
303
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100304static void free_pending_exception(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 mempool_free(pe, pending_pool);
307}
308
309int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
310{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100311 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313 e = alloc_exception();
314 if (!e)
315 return -ENOMEM;
316
317 e->old_chunk = old;
318 e->new_chunk = new;
319 insert_exception(&s->complete, e);
320 return 0;
321}
322
323/*
324 * Hard coded magic.
325 */
326static int calc_max_buckets(void)
327{
328 /* use a fixed size of 2MB */
329 unsigned long mem = 2 * 1024 * 1024;
330 mem /= sizeof(struct list_head);
331
332 return mem;
333}
334
335/*
336 * Rounds a number down to a power of 2.
337 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100338static uint32_t round_down(uint32_t n)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
340 while (n & (n - 1))
341 n &= (n - 1);
342 return n;
343}
344
345/*
346 * Allocate room for a suitable hash table.
347 */
348static int init_hash_tables(struct dm_snapshot *s)
349{
350 sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
351
352 /*
353 * Calculate based on the size of the original volume or
354 * the COW volume...
355 */
356 cow_dev_size = get_dev_size(s->cow->bdev);
357 origin_dev_size = get_dev_size(s->origin->bdev);
358 max_buckets = calc_max_buckets();
359
360 hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
361 hash_size = min(hash_size, max_buckets);
362
363 /* Round it down to a power of 2 */
364 hash_size = round_down(hash_size);
365 if (init_exception_table(&s->complete, hash_size))
366 return -ENOMEM;
367
368 /*
369 * Allocate hash table for in-flight exceptions
370 * Make this smaller than the real hash table
371 */
372 hash_size >>= 3;
373 if (hash_size < 64)
374 hash_size = 64;
375
376 if (init_exception_table(&s->pending, hash_size)) {
377 exit_exception_table(&s->complete, exception_cache);
378 return -ENOMEM;
379 }
380
381 return 0;
382}
383
384/*
385 * Round a number up to the nearest 'size' boundary. size must
386 * be a power of 2.
387 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100388static ulong round_up(ulong n, ulong size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
390 size--;
391 return (n + size) & ~size;
392}
393
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700394static int set_chunk_size(struct dm_snapshot *s, const char *chunk_size_arg,
395 char **error)
396{
397 unsigned long chunk_size;
398 char *value;
399
400 chunk_size = simple_strtoul(chunk_size_arg, &value, 10);
401 if (*chunk_size_arg == '\0' || *value != '\0') {
402 *error = "Invalid chunk size";
403 return -EINVAL;
404 }
405
406 if (!chunk_size) {
407 s->chunk_size = s->chunk_mask = s->chunk_shift = 0;
408 return 0;
409 }
410
411 /*
412 * Chunk size must be multiple of page size. Silently
413 * round up if it's not.
414 */
415 chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
416
417 /* Check chunk_size is a power of 2 */
418 if (chunk_size & (chunk_size - 1)) {
419 *error = "Chunk size is not a power of 2";
420 return -EINVAL;
421 }
422
423 /* Validate the chunk size against the device block size */
424 if (chunk_size % (bdev_hardsect_size(s->cow->bdev) >> 9)) {
425 *error = "Chunk size is not a multiple of device blocksize";
426 return -EINVAL;
427 }
428
429 s->chunk_size = chunk_size;
430 s->chunk_mask = chunk_size - 1;
431 s->chunk_shift = ffs(chunk_size) - 1;
432
433 return 0;
434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
437 * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
438 */
439static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
440{
441 struct dm_snapshot *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 int r = -EINVAL;
443 char persistent;
444 char *origin_path;
445 char *cow_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700447 if (argc != 4) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700448 ti->error = "requires exactly 4 arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 r = -EINVAL;
450 goto bad1;
451 }
452
453 origin_path = argv[0];
454 cow_path = argv[1];
455 persistent = toupper(*argv[2]);
456
457 if (persistent != 'P' && persistent != 'N') {
458 ti->error = "Persistent flag is not P or N";
459 r = -EINVAL;
460 goto bad1;
461 }
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 s = kmalloc(sizeof(*s), GFP_KERNEL);
464 if (s == NULL) {
465 ti->error = "Cannot allocate snapshot context private "
466 "structure";
467 r = -ENOMEM;
468 goto bad1;
469 }
470
471 r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
472 if (r) {
473 ti->error = "Cannot get origin device";
474 goto bad2;
475 }
476
477 r = dm_get_device(ti, cow_path, 0, 0,
478 FMODE_READ | FMODE_WRITE, &s->cow);
479 if (r) {
480 dm_put_device(ti, s->origin);
481 ti->error = "Cannot get COW device";
482 goto bad2;
483 }
484
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700485 r = set_chunk_size(s, argv[3], &ti->error);
486 if (r)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 goto bad3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 s->type = persistent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 s->valid = 1;
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800492 s->active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 s->last_percent = 0;
494 init_rwsem(&s->lock);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700495 spin_lock_init(&s->pe_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 s->table = ti->table;
497
498 /* Allocate hash table for COW data */
499 if (init_hash_tables(s)) {
500 ti->error = "Unable to allocate hash table space";
501 r = -ENOMEM;
502 goto bad3;
503 }
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 s->store.snap = s;
506
507 if (persistent == 'P')
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700508 r = dm_create_persistent(&s->store);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 else
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -0700510 r = dm_create_transient(&s->store);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512 if (r) {
513 ti->error = "Couldn't create exception store";
514 r = -EINVAL;
515 goto bad4;
516 }
517
518 r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
519 if (r) {
520 ti->error = "Could not create kcopyd client";
521 goto bad5;
522 }
523
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800524 /* Metadata must only be loaded into one table at once */
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700525 r = s->store.read_metadata(&s->store);
Milan Broz07641472007-07-12 17:28:13 +0100526 if (r < 0) {
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700527 ti->error = "Failed to read snapshot metadata";
528 goto bad6;
Milan Broz07641472007-07-12 17:28:13 +0100529 } else if (r > 0) {
530 s->valid = 0;
531 DMWARN("Snapshot is marked invalid.");
Mark McLoughlinf9cea4f2006-10-03 01:15:25 -0700532 }
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800533
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700534 bio_list_init(&s->queued_bios);
David Howellsc4028952006-11-22 14:57:56 +0000535 INIT_WORK(&s->queued_bios_work, flush_queued_bios);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 /* Add snapshot to the list of snapshots for this origin */
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800538 /* Exceptions aren't triggered till snapshot_resume() is called */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 if (register_snapshot(s)) {
540 r = -EINVAL;
541 ti->error = "Cannot register snapshot origin";
542 goto bad6;
543 }
544
545 ti->private = s;
Alasdair G Kergonc51c2752006-06-26 00:27:18 -0700546 ti->split_io = s->chunk_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 return 0;
549
550 bad6:
551 kcopyd_client_destroy(s->kcopyd_client);
552
553 bad5:
554 s->store.destroy(&s->store);
555
556 bad4:
557 exit_exception_table(&s->pending, pending_cache);
558 exit_exception_table(&s->complete, exception_cache);
559
560 bad3:
561 dm_put_device(ti, s->cow);
562 dm_put_device(ti, s->origin);
563
564 bad2:
565 kfree(s);
566
567 bad1:
568 return r;
569}
570
Milan Broz31c93a02006-12-08 02:41:11 -0800571static void __free_exceptions(struct dm_snapshot *s)
572{
573 kcopyd_client_destroy(s->kcopyd_client);
574 s->kcopyd_client = NULL;
575
576 exit_exception_table(&s->pending, pending_cache);
577 exit_exception_table(&s->complete, exception_cache);
578
579 s->store.destroy(&s->store);
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582static void snapshot_dtr(struct dm_target *ti)
583{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100584 struct dm_snapshot *s = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700586 flush_workqueue(ksnapd);
587
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800588 /* Prevent further origin writes from using this snapshot. */
589 /* After this returns there can be no new kcopyd jobs. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 unregister_snapshot(s);
591
Milan Broz31c93a02006-12-08 02:41:11 -0800592 __free_exceptions(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 dm_put_device(ti, s->origin);
595 dm_put_device(ti, s->cow);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 kfree(s);
598}
599
600/*
601 * Flush a list of buffers.
602 */
603static void flush_bios(struct bio *bio)
604{
605 struct bio *n;
606
607 while (bio) {
608 n = bio->bi_next;
609 bio->bi_next = NULL;
610 generic_make_request(bio);
611 bio = n;
612 }
613}
614
David Howellsc4028952006-11-22 14:57:56 +0000615static void flush_queued_bios(struct work_struct *work)
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700616{
David Howellsc4028952006-11-22 14:57:56 +0000617 struct dm_snapshot *s =
618 container_of(work, struct dm_snapshot, queued_bios_work);
Alasdair G Kergonca3a9312006-10-03 01:15:30 -0700619 struct bio *queued_bios;
620 unsigned long flags;
621
622 spin_lock_irqsave(&s->pe_lock, flags);
623 queued_bios = bio_list_get(&s->queued_bios);
624 spin_unlock_irqrestore(&s->pe_lock, flags);
625
626 flush_bios(queued_bios);
627}
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629/*
630 * Error a list of buffers.
631 */
632static void error_bios(struct bio *bio)
633{
634 struct bio *n;
635
636 while (bio) {
637 n = bio->bi_next;
638 bio->bi_next = NULL;
NeilBrown6712ecf2007-09-27 12:47:43 +0200639 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 bio = n;
641 }
642}
643
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700644static void __invalidate_snapshot(struct dm_snapshot *s, int err)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800645{
646 if (!s->valid)
647 return;
648
649 if (err == -EIO)
650 DMERR("Invalidating snapshot: Error reading/writing.");
651 else if (err == -ENOMEM)
652 DMERR("Invalidating snapshot: Unable to allocate exception.");
653
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800654 if (s->store.drop_snapshot)
655 s->store.drop_snapshot(&s->store);
656
657 s->valid = 0;
658
659 dm_table_event(s->table);
660}
661
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100662static void get_pending_exception(struct dm_snap_pending_exception *pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700663{
664 atomic_inc(&pe->ref_count);
665}
666
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100667static struct bio *put_pending_exception(struct dm_snap_pending_exception *pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700668{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100669 struct dm_snap_pending_exception *primary_pe;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700670 struct bio *origin_bios = NULL;
671
672 primary_pe = pe->primary_pe;
673
674 /*
675 * If this pe is involved in a write to the origin and
676 * it is the last sibling to complete then release
677 * the bios for the original write to the origin.
678 */
679 if (primary_pe &&
680 atomic_dec_and_test(&primary_pe->ref_count))
681 origin_bios = bio_list_get(&primary_pe->origin_bios);
682
683 /*
684 * Free the pe if it's not linked to an origin write or if
685 * it's not itself a primary pe.
686 */
687 if (!primary_pe || primary_pe != pe)
688 free_pending_exception(pe);
689
690 /*
691 * Free the primary pe if nothing references it.
692 */
693 if (primary_pe && !atomic_read(&primary_pe->ref_count))
694 free_pending_exception(primary_pe);
695
696 return origin_bios;
697}
698
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100699static void pending_complete(struct dm_snap_pending_exception *pe, int success)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100701 struct dm_snap_exception *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 struct dm_snapshot *s = pe->snap;
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700703 struct bio *origin_bios = NULL;
704 struct bio *snapshot_bios = NULL;
705 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800707 if (!success) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 /* Read/write error - snapshot is unusable */
709 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700710 __invalidate_snapshot(s, -EIO);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700711 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800712 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 }
714
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800715 e = alloc_exception();
716 if (!e) {
717 down_write(&s->lock);
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700718 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700719 error = 1;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800720 goto out;
721 }
722 *e = pe->e;
723
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700724 down_write(&s->lock);
725 if (!s->valid) {
726 free_exception(e);
727 error = 1;
728 goto out;
729 }
730
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800731 /*
732 * Add a proper exception, and remove the
733 * in-flight exception from the list.
734 */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800735 insert_exception(&s->complete, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800736
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 out:
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700738 remove_exception(&pe->e);
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700739 snapshot_bios = bio_list_get(&pe->snapshot_bios);
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700740 origin_bios = put_pending_exception(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Alasdair G Kergon9d493fa2006-10-03 01:15:29 -0700742 up_write(&s->lock);
743
744 /* Submit any pending write bios */
745 if (error)
746 error_bios(snapshot_bios);
747 else
748 flush_bios(snapshot_bios);
749
750 flush_bios(origin_bios);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
753static void commit_callback(void *context, int success)
754{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100755 struct dm_snap_pending_exception *pe = context;
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 pending_complete(pe, success);
758}
759
760/*
761 * Called when the copy I/O has finished. kcopyd actually runs
762 * this code so don't block.
763 */
764static void copy_callback(int read_err, unsigned int write_err, void *context)
765{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100766 struct dm_snap_pending_exception *pe = context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 struct dm_snapshot *s = pe->snap;
768
769 if (read_err || write_err)
770 pending_complete(pe, 0);
771
772 else
773 /* Update the metadata if we are persistent */
774 s->store.commit_exception(&s->store, &pe->e, commit_callback,
775 pe);
776}
777
778/*
779 * Dispatches the copy operation to kcopyd.
780 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100781static void start_copy(struct dm_snap_pending_exception *pe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782{
783 struct dm_snapshot *s = pe->snap;
784 struct io_region src, dest;
785 struct block_device *bdev = s->origin->bdev;
786 sector_t dev_size;
787
788 dev_size = get_dev_size(bdev);
789
790 src.bdev = bdev;
791 src.sector = chunk_to_sector(s, pe->e.old_chunk);
792 src.count = min(s->chunk_size, dev_size - src.sector);
793
794 dest.bdev = s->cow->bdev;
795 dest.sector = chunk_to_sector(s, pe->e.new_chunk);
796 dest.count = src.count;
797
798 /* Hand over to kcopyd */
799 kcopyd_copy(s->kcopyd_client,
800 &src, 1, &dest, 0, copy_callback, pe);
801}
802
803/*
804 * Looks to see if this snapshot already has a pending exception
805 * for this chunk, otherwise it allocates a new one and inserts
806 * it into the pending table.
807 *
808 * NOTE: a write lock must be held on snap->lock before calling
809 * this.
810 */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100811static struct dm_snap_pending_exception *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812__find_pending_exception(struct dm_snapshot *s, struct bio *bio)
813{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100814 struct dm_snap_exception *e;
815 struct dm_snap_pending_exception *pe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 chunk_t chunk = sector_to_chunk(s, bio->bi_sector);
817
818 /*
819 * Is there a pending exception for this already ?
820 */
821 e = lookup_exception(&s->pending, chunk);
822 if (e) {
823 /* cast the exception to a pending exception */
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100824 pe = container_of(e, struct dm_snap_pending_exception, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800825 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
827
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800828 /*
829 * Create a new pending exception, we don't want
830 * to hold the lock while we do this.
831 */
832 up_write(&s->lock);
833 pe = alloc_pending_exception();
834 down_write(&s->lock);
835
836 if (!s->valid) {
837 free_pending_exception(pe);
838 return NULL;
839 }
840
841 e = lookup_exception(&s->pending, chunk);
842 if (e) {
843 free_pending_exception(pe);
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100844 pe = container_of(e, struct dm_snap_pending_exception, e);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800845 goto out;
846 }
847
848 pe->e.old_chunk = chunk;
849 bio_list_init(&pe->origin_bios);
850 bio_list_init(&pe->snapshot_bios);
851 pe->primary_pe = NULL;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700852 atomic_set(&pe->ref_count, 0);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800853 pe->snap = s;
854 pe->started = 0;
855
856 if (s->store.prepare_exception(&s->store, &pe->e)) {
857 free_pending_exception(pe);
858 return NULL;
859 }
860
Alasdair G Kergon4b832e82006-10-03 01:15:30 -0700861 get_pending_exception(pe);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800862 insert_exception(&s->pending, &pe->e);
863
864 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 return pe;
866}
867
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100868static void remap_exception(struct dm_snapshot *s, struct dm_snap_exception *e,
869 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
871 bio->bi_bdev = s->cow->bdev;
872 bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
873 (bio->bi_sector & s->chunk_mask);
874}
875
876static int snapshot_map(struct dm_target *ti, struct bio *bio,
877 union map_info *map_context)
878{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100879 struct dm_snap_exception *e;
880 struct dm_snapshot *s = ti->private;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800881 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 chunk_t chunk;
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100883 struct dm_snap_pending_exception *pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 chunk = sector_to_chunk(s, bio->bi_sector);
886
887 /* Full snapshots are not usable */
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800888 /* To get here the table must be live so s->active is always set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if (!s->valid)
Alasdair G Kergonf6a80ea2005-07-12 15:53:01 -0700890 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700892 /* FIXME: should only take write lock if we need
893 * to copy an exception */
894 down_write(&s->lock);
895
896 if (!s->valid) {
897 r = -EIO;
898 goto out_unlock;
899 }
900
901 /* If the block is already remapped - use that, else remap it */
902 e = lookup_exception(&s->complete, chunk);
903 if (e) {
904 remap_exception(s, e, bio);
905 goto out_unlock;
906 }
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 /*
909 * Write to snapshot - higher level takes care of RW/RO
910 * flags so we should only get this if we are
911 * writeable.
912 */
913 if (bio_rw(bio) == WRITE) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800914 pe = __find_pending_exception(s, bio);
915 if (!pe) {
Alasdair G Kergon695368a2006-10-03 01:15:31 -0700916 __invalidate_snapshot(s, -ENOMEM);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800917 r = -EIO;
918 goto out_unlock;
919 }
920
921 remap_exception(s, &pe->e, bio);
922 bio_list_add(&pe->snapshot_bios, bio);
923
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800924 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700925
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800926 if (!pe->started) {
927 /* this is protected by snap->lock */
928 pe->started = 1;
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700929 up_write(&s->lock);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -0800930 start_copy(pe);
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700931 goto out;
932 }
933 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 /*
935 * FIXME: this read path scares me because we
936 * always use the origin when we have a pending
937 * exception. However I can't think of a
938 * situation where this is wrong - ejt.
939 */
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700940 bio->bi_bdev = s->origin->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Alasdair G Kergonba40a2a2006-10-03 01:15:28 -0700942 out_unlock:
943 up_write(&s->lock);
944 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return r;
946}
947
948static void snapshot_resume(struct dm_target *ti)
949{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100950 struct dm_snapshot *s = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -0800952 down_write(&s->lock);
953 s->active = 1;
954 up_write(&s->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955}
956
957static int snapshot_status(struct dm_target *ti, status_type_t type,
958 char *result, unsigned int maxlen)
959{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100960 struct dm_snapshot *snap = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
962 switch (type) {
963 case STATUSTYPE_INFO:
964 if (!snap->valid)
965 snprintf(result, maxlen, "Invalid");
966 else {
967 if (snap->store.fraction_full) {
968 sector_t numerator, denominator;
969 snap->store.fraction_full(&snap->store,
970 &numerator,
971 &denominator);
Andrew Morton4ee218c2006-03-27 01:17:48 -0800972 snprintf(result, maxlen, "%llu/%llu",
973 (unsigned long long)numerator,
974 (unsigned long long)denominator);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 }
976 else
977 snprintf(result, maxlen, "Unknown");
978 }
979 break;
980
981 case STATUSTYPE_TABLE:
982 /*
983 * kdevname returns a static pointer so we need
984 * to make private copies if the output is to
985 * make sense.
986 */
Andrew Morton4ee218c2006-03-27 01:17:48 -0800987 snprintf(result, maxlen, "%s %s %c %llu",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 snap->origin->name, snap->cow->name,
Andrew Morton4ee218c2006-03-27 01:17:48 -0800989 snap->type,
990 (unsigned long long)snap->chunk_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 break;
992 }
993
994 return 0;
995}
996
997/*-----------------------------------------------------------------
998 * Origin methods
999 *---------------------------------------------------------------*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000static int __origin_write(struct list_head *snapshots, struct bio *bio)
1001{
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001002 int r = DM_MAPIO_REMAPPED, first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 struct dm_snapshot *snap;
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001004 struct dm_snap_exception *e;
1005 struct dm_snap_pending_exception *pe, *next_pe, *primary_pe = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 chunk_t chunk;
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001007 LIST_HEAD(pe_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 /* Do all the snapshots on this origin */
1010 list_for_each_entry (snap, snapshots, list) {
1011
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001012 down_write(&snap->lock);
1013
Alasdair G Kergonaa14ede2006-02-01 03:04:50 -08001014 /* Only deal with valid and active snapshots */
1015 if (!snap->valid || !snap->active)
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001016 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001018 /* Nothing to do if writing beyond end of snapshot */
1019 if (bio->bi_sector >= dm_table_get_size(snap->table))
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001020 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 /*
1023 * Remember, different snapshots can have
1024 * different chunk sizes.
1025 */
1026 chunk = sector_to_chunk(snap, bio->bi_sector);
1027
1028 /*
1029 * Check exception table to see if block
1030 * is already remapped in this snapshot
1031 * and trigger an exception if not.
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001032 *
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001033 * ref_count is initialised to 1 so pending_complete()
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001034 * won't destroy the primary_pe while we're inside this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 */
1036 e = lookup_exception(&snap->complete, chunk);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001037 if (e)
1038 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001040 pe = __find_pending_exception(snap, bio);
1041 if (!pe) {
Alasdair G Kergon695368a2006-10-03 01:15:31 -07001042 __invalidate_snapshot(snap, -ENOMEM);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001043 goto next_snapshot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 }
1045
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001046 if (!primary_pe) {
1047 /*
1048 * Either every pe here has same
1049 * primary_pe or none has one yet.
1050 */
1051 if (pe->primary_pe)
1052 primary_pe = pe->primary_pe;
1053 else {
1054 primary_pe = pe;
1055 first = 1;
1056 }
1057
1058 bio_list_add(&primary_pe->origin_bios, bio);
1059
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001060 r = DM_MAPIO_SUBMITTED;
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001061 }
1062
1063 if (!pe->primary_pe) {
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001064 pe->primary_pe = primary_pe;
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001065 get_pending_exception(primary_pe);
Alasdair G Kergon76df1c62006-03-27 01:17:45 -08001066 }
1067
1068 if (!pe->started) {
1069 pe->started = 1;
1070 list_add_tail(&pe->list, &pe_queue);
1071 }
1072
1073 next_snapshot:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 up_write(&snap->lock);
1075 }
1076
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001077 if (!primary_pe)
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001078 return r;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001079
1080 /*
1081 * If this is the first time we're processing this chunk and
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001082 * ref_count is now 1 it means all the pending exceptions
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001083 * got completed while we were in the loop above, so it falls to
1084 * us here to remove the primary_pe and submit any origin_bios.
1085 */
1086
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001087 if (first && atomic_dec_and_test(&primary_pe->ref_count)) {
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001088 flush_bios(bio_list_get(&primary_pe->origin_bios));
1089 free_pending_exception(primary_pe);
1090 /* If we got here, pe_queue is necessarily empty. */
Alasdair G Kergon4b832e82006-10-03 01:15:30 -07001091 return r;
Alasdair G Kergonb4b610f2006-03-27 01:17:44 -08001092 }
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 /*
1095 * Now that we have a complete pe list we can start the copying.
1096 */
Alasdair G Kergoneccf0812006-03-27 01:17:42 -08001097 list_for_each_entry_safe(pe, next_pe, &pe_queue, list)
1098 start_copy(pe);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 return r;
1101}
1102
1103/*
1104 * Called on a write from the origin driver.
1105 */
1106static int do_origin(struct dm_dev *origin, struct bio *bio)
1107{
1108 struct origin *o;
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001109 int r = DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
1111 down_read(&_origins_lock);
1112 o = __lookup_origin(origin->bdev);
1113 if (o)
1114 r = __origin_write(&o->snapshots, bio);
1115 up_read(&_origins_lock);
1116
1117 return r;
1118}
1119
1120/*
1121 * Origin: maps a linear range of a device, with hooks for snapshotting.
1122 */
1123
1124/*
1125 * Construct an origin mapping: <dev_path>
1126 * The context for an origin is merely a 'struct dm_dev *'
1127 * pointing to the real device.
1128 */
1129static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1130{
1131 int r;
1132 struct dm_dev *dev;
1133
1134 if (argc != 1) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001135 ti->error = "origin: incorrect number of arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 return -EINVAL;
1137 }
1138
1139 r = dm_get_device(ti, argv[0], 0, ti->len,
1140 dm_table_get_mode(ti->table), &dev);
1141 if (r) {
1142 ti->error = "Cannot get target device";
1143 return r;
1144 }
1145
1146 ti->private = dev;
1147 return 0;
1148}
1149
1150static void origin_dtr(struct dm_target *ti)
1151{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001152 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 dm_put_device(ti, dev);
1154}
1155
1156static int origin_map(struct dm_target *ti, struct bio *bio,
1157 union map_info *map_context)
1158{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001159 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 bio->bi_bdev = dev->bdev;
1161
1162 /* Only tell snapshots if this is a write */
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -08001163 return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
1165
1166#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
1167
1168/*
1169 * Set the target "split_io" field to the minimum of all the snapshots'
1170 * chunk sizes.
1171 */
1172static void origin_resume(struct dm_target *ti)
1173{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001174 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct dm_snapshot *snap;
1176 struct origin *o;
1177 chunk_t chunk_size = 0;
1178
1179 down_read(&_origins_lock);
1180 o = __lookup_origin(dev->bdev);
1181 if (o)
1182 list_for_each_entry (snap, &o->snapshots, list)
1183 chunk_size = min_not_zero(chunk_size, snap->chunk_size);
1184 up_read(&_origins_lock);
1185
1186 ti->split_io = chunk_size;
1187}
1188
1189static int origin_status(struct dm_target *ti, status_type_t type, char *result,
1190 unsigned int maxlen)
1191{
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001192 struct dm_dev *dev = ti->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194 switch (type) {
1195 case STATUSTYPE_INFO:
1196 result[0] = '\0';
1197 break;
1198
1199 case STATUSTYPE_TABLE:
1200 snprintf(result, maxlen, "%s", dev->name);
1201 break;
1202 }
1203
1204 return 0;
1205}
1206
1207static struct target_type origin_target = {
1208 .name = "snapshot-origin",
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001209 .version = {1, 5, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 .module = THIS_MODULE,
1211 .ctr = origin_ctr,
1212 .dtr = origin_dtr,
1213 .map = origin_map,
1214 .resume = origin_resume,
1215 .status = origin_status,
1216};
1217
1218static struct target_type snapshot_target = {
1219 .name = "snapshot",
Mark McLoughlin4c7e3bf2006-10-03 01:15:25 -07001220 .version = {1, 5, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 .module = THIS_MODULE,
1222 .ctr = snapshot_ctr,
1223 .dtr = snapshot_dtr,
1224 .map = snapshot_map,
1225 .resume = snapshot_resume,
1226 .status = snapshot_status,
1227};
1228
1229static int __init dm_snapshot_init(void)
1230{
1231 int r;
1232
1233 r = dm_register_target(&snapshot_target);
1234 if (r) {
1235 DMERR("snapshot target register failed %d", r);
1236 return r;
1237 }
1238
1239 r = dm_register_target(&origin_target);
1240 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -07001241 DMERR("Origin target register failed %d", r);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 goto bad1;
1243 }
1244
1245 r = init_origin_hash();
1246 if (r) {
1247 DMERR("init_origin_hash failed.");
1248 goto bad2;
1249 }
1250
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001251 exception_cache = KMEM_CACHE(dm_snap_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (!exception_cache) {
1253 DMERR("Couldn't create exception cache.");
1254 r = -ENOMEM;
1255 goto bad3;
1256 }
1257
Alasdair G Kergon028867a2007-07-12 17:26:32 +01001258 pending_cache = KMEM_CACHE(dm_snap_pending_exception, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (!pending_cache) {
1260 DMERR("Couldn't create pending cache.");
1261 r = -ENOMEM;
1262 goto bad4;
1263 }
1264
Matthew Dobson93d23412006-03-26 01:37:50 -08001265 pending_pool = mempool_create_slab_pool(128, pending_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (!pending_pool) {
1267 DMERR("Couldn't create pending pool.");
1268 r = -ENOMEM;
1269 goto bad5;
1270 }
1271
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001272 ksnapd = create_singlethread_workqueue("ksnapd");
1273 if (!ksnapd) {
1274 DMERR("Failed to create ksnapd workqueue.");
1275 r = -ENOMEM;
1276 goto bad6;
1277 }
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 return 0;
1280
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001281 bad6:
1282 mempool_destroy(pending_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 bad5:
1284 kmem_cache_destroy(pending_cache);
1285 bad4:
1286 kmem_cache_destroy(exception_cache);
1287 bad3:
1288 exit_origin_hash();
1289 bad2:
1290 dm_unregister_target(&origin_target);
1291 bad1:
1292 dm_unregister_target(&snapshot_target);
1293 return r;
1294}
1295
1296static void __exit dm_snapshot_exit(void)
1297{
1298 int r;
1299
Alasdair G Kergonca3a9312006-10-03 01:15:30 -07001300 destroy_workqueue(ksnapd);
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 r = dm_unregister_target(&snapshot_target);
1303 if (r)
1304 DMERR("snapshot unregister failed %d", r);
1305
1306 r = dm_unregister_target(&origin_target);
1307 if (r)
1308 DMERR("origin unregister failed %d", r);
1309
1310 exit_origin_hash();
1311 mempool_destroy(pending_pool);
1312 kmem_cache_destroy(pending_cache);
1313 kmem_cache_destroy(exception_cache);
1314}
1315
1316/* Module hooks */
1317module_init(dm_snapshot_init);
1318module_exit(dm_snapshot_exit);
1319
1320MODULE_DESCRIPTION(DM_NAME " snapshot target");
1321MODULE_AUTHOR("Joe Thornber");
1322MODULE_LICENSE("GPL");