~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~ [ freetext search ] ~ [ file search ] ~

Linux Cross Reference
Linux-2.6.17/drivers/md/dm-snap.c

Version: ~ [ 2.6.16 ] ~ [ 2.6.17 ] ~
Architecture: ~ [ ia64 ] ~ [ i386 ] ~ [ arm ] ~ [ ppc ] ~ [ sparc64 ] ~

  1 /*
  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>
 10 #include <linux/config.h>
 11 #include <linux/ctype.h>
 12 #include <linux/device-mapper.h>
 13 #include <linux/fs.h>
 14 #include <linux/init.h>
 15 #include <linux/kdev_t.h>
 16 #include <linux/list.h>
 17 #include <linux/mempool.h>
 18 #include <linux/module.h>
 19 #include <linux/slab.h>
 20 #include <linux/vmalloc.h>
 21 
 22 #include "dm-snap.h"
 23 #include "dm-bio-list.h"
 24 #include "kcopyd.h"
 25 
 26 /*
 27  * The percentage increment we will wake up users at
 28  */
 29 #define WAKE_UP_PERCENT 5
 30 
 31 /*
 32  * kcopyd priority of snapshot operations
 33  */
 34 #define SNAPSHOT_COPY_PRIORITY 2
 35 
 36 /*
 37  * Each snapshot reserves this many pages for io
 38  */
 39 #define SNAPSHOT_PAGES 256
 40 
 41 struct pending_exception {
 42         struct exception e;
 43 
 44         /*
 45          * Origin buffers waiting for this to complete are held
 46          * in a bio list
 47          */
 48         struct bio_list origin_bios;
 49         struct bio_list snapshot_bios;
 50 
 51         /*
 52          * Short-term queue of pending exceptions prior to submission.
 53          */
 54         struct list_head list;
 55 
 56         /*
 57          * The primary pending_exception is the one that holds
 58          * the sibling_count and the list of origin_bios for a
 59          * group of pending_exceptions.  It is always last to get freed.
 60          * These fields get set up when writing to the origin.
 61          */
 62         struct pending_exception *primary_pe;
 63 
 64         /*
 65          * Number of pending_exceptions processing this chunk.
 66          * When this drops to zero we must complete the origin bios.
 67          * If incrementing or decrementing this, hold pe->snap->lock for
 68          * the sibling concerned and not pe->primary_pe->snap->lock unless
 69          * they are the same.
 70          */
 71         atomic_t sibling_count;
 72 
 73         /* Pointer back to snapshot context */
 74         struct dm_snapshot *snap;
 75 
 76         /*
 77          * 1 indicates the exception has already been sent to
 78          * kcopyd.
 79          */
 80         int started;
 81 };
 82 
 83 /*
 84  * Hash table mapping origin volumes to lists of snapshots and
 85  * a lock to protect it
 86  */
 87 static kmem_cache_t *exception_cache;
 88 static kmem_cache_t *pending_cache;
 89 static mempool_t *pending_pool;
 90 
 91 /*
 92  * One of these per registered origin, held in the snapshot_origins hash
 93  */
 94 struct origin {
 95         /* The origin device */
 96         struct block_device *bdev;
 97 
 98         struct list_head hash_list;
 99 
100         /* List of snapshots for this origin */
101         struct list_head snapshots;
102 };
103 
104 /*
105  * Size of the hash table for origin volumes. If we make this
106  * the size of the minors list then it should be nearly perfect
107  */
108 #define ORIGIN_HASH_SIZE 256
109 #define ORIGIN_MASK      0xFF
110 static struct list_head *_origins;
111 static struct rw_semaphore _origins_lock;
112 
113 static int init_origin_hash(void)
114 {
115         int i;
116 
117         _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head),
118                            GFP_KERNEL);
119         if (!_origins) {
120                 DMERR("Device mapper: Snapshot: unable to allocate memory");
121                 return -ENOMEM;
122         }
123 
124         for (i = 0; i < ORIGIN_HASH_SIZE; i++)
125                 INIT_LIST_HEAD(_origins + i);
126         init_rwsem(&_origins_lock);
127 
128         return 0;
129 }
130 
131 static void exit_origin_hash(void)
132 {
133         kfree(_origins);
134 }
135 
136 static inline unsigned int origin_hash(struct block_device *bdev)
137 {
138         return bdev->bd_dev & ORIGIN_MASK;
139 }
140 
141 static struct origin *__lookup_origin(struct block_device *origin)
142 {
143         struct list_head *ol;
144         struct origin *o;
145 
146         ol = &_origins[origin_hash(origin)];
147         list_for_each_entry (o, ol, hash_list)
148                 if (bdev_equal(o->bdev, origin))
149                         return o;
150 
151         return NULL;
152 }
153 
154 static void __insert_origin(struct origin *o)
155 {
156         struct list_head *sl = &_origins[origin_hash(o->bdev)];
157         list_add_tail(&o->hash_list, sl);
158 }
159 
160 /*
161  * Make a note of the snapshot and its origin so we can look it
162  * up when the origin has a write on it.
163  */
164 static int register_snapshot(struct dm_snapshot *snap)
165 {
166         struct origin *o;
167         struct block_device *bdev = snap->origin->bdev;
168 
169         down_write(&_origins_lock);
170         o = __lookup_origin(bdev);
171 
172         if (!o) {
173                 /* New origin */
174                 o = kmalloc(sizeof(*o), GFP_KERNEL);
175                 if (!o) {
176                         up_write(&_origins_lock);
177                         return -ENOMEM;
178                 }
179 
180                 /* Initialise the struct */
181                 INIT_LIST_HEAD(&o->snapshots);
182                 o->bdev = bdev;
183 
184                 __insert_origin(o);
185         }
186 
187         list_add_tail(&snap->list, &o->snapshots);
188 
189         up_write(&_origins_lock);
190         return 0;
191 }
192 
193 static void unregister_snapshot(struct dm_snapshot *s)
194 {
195         struct origin *o;
196 
197         down_write(&_origins_lock);
198         o = __lookup_origin(s->origin->bdev);
199 
200         list_del(&s->list);
201         if (list_empty(&o->snapshots)) {
202                 list_del(&o->hash_list);
203                 kfree(o);
204         }
205 
206         up_write(&_origins_lock);
207 }
208 
209 /*
210  * Implementation of the exception hash tables.
211  */
212 static int init_exception_table(struct exception_table *et, uint32_t size)
213 {
214         unsigned int i;
215 
216         et->hash_mask = size - 1;
217         et->table = dm_vcalloc(size, sizeof(struct list_head));
218         if (!et->table)
219                 return -ENOMEM;
220 
221         for (i = 0; i < size; i++)
222                 INIT_LIST_HEAD(et->table + i);
223 
224         return 0;
225 }
226 
227 static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem)
228 {
229         struct list_head *slot;
230         struct exception *ex, *next;
231         int i, size;
232 
233         size = et->hash_mask + 1;
234         for (i = 0; i < size; i++) {
235                 slot = et->table + i;
236 
237                 list_for_each_entry_safe (ex, next, slot, hash_list)
238                         kmem_cache_free(mem, ex);
239         }
240 
241         vfree(et->table);
242 }
243 
244 static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk)
245 {
246         return chunk & et->hash_mask;
247 }
248 
249 static void insert_exception(struct exception_table *eh, struct exception *e)
250 {
251         struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)];
252         list_add(&e->hash_list, l);
253 }
254 
255 static inline void remove_exception(struct exception *e)
256 {
257         list_del(&e->hash_list);
258 }
259 
260 /*
261  * Return the exception data for a sector, or NULL if not
262  * remapped.
263  */
264 static struct exception *lookup_exception(struct exception_table *et,
265                                           chunk_t chunk)
266 {
267         struct list_head *slot;
268         struct exception *e;
269 
270         slot = &et->table[exception_hash(et, chunk)];
271         list_for_each_entry (e, slot, hash_list)
272                 if (e->old_chunk == chunk)
273                         return e;
274 
275         return NULL;
276 }
277 
278 static inline struct exception *alloc_exception(void)
279 {
280         struct exception *e;
281 
282         e = kmem_cache_alloc(exception_cache, GFP_NOIO);
283         if (!e)
284                 e = kmem_cache_alloc(exception_cache, GFP_ATOMIC);
285 
286         return e;
287 }
288 
289 static inline void free_exception(struct exception *e)
290 {
291         kmem_cache_free(exception_cache, e);
292 }
293 
294 static inline struct pending_exception *alloc_pending_exception(void)
295 {
296         return mempool_alloc(pending_pool, GFP_NOIO);
297 }
298 
299 static inline void free_pending_exception(struct pending_exception *pe)
300 {
301         mempool_free(pe, pending_pool);
302 }
303 
304 int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new)
305 {
306         struct exception *e;
307 
308         e = alloc_exception();
309         if (!e)
310                 return -ENOMEM;
311 
312         e->old_chunk = old;
313         e->new_chunk = new;
314         insert_exception(&s->complete, e);
315         return 0;
316 }
317 
318 /*
319  * Hard coded magic.
320  */
321 static int calc_max_buckets(void)
322 {
323         /* use a fixed size of 2MB */
324         unsigned long mem = 2 * 1024 * 1024;
325         mem /= sizeof(struct list_head);
326 
327         return mem;
328 }
329 
330 /*
331  * Rounds a number down to a power of 2.
332  */
333 static inline uint32_t round_down(uint32_t n)
334 {
335         while (n & (n - 1))
336                 n &= (n - 1);
337         return n;
338 }
339 
340 /*
341  * Allocate room for a suitable hash table.
342  */
343 static int init_hash_tables(struct dm_snapshot *s)
344 {
345         sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets;
346 
347         /*
348          * Calculate based on the size of the original volume or
349          * the COW volume...
350          */
351         cow_dev_size = get_dev_size(s->cow->bdev);
352         origin_dev_size = get_dev_size(s->origin->bdev);
353         max_buckets = calc_max_buckets();
354 
355         hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift;
356         hash_size = min(hash_size, max_buckets);
357 
358         /* Round it down to a power of 2 */
359         hash_size = round_down(hash_size);
360         if (init_exception_table(&s->complete, hash_size))
361                 return -ENOMEM;
362 
363         /*
364          * Allocate hash table for in-flight exceptions
365          * Make this smaller than the real hash table
366          */
367         hash_size >>= 3;
368         if (hash_size < 64)
369                 hash_size = 64;
370 
371         if (init_exception_table(&s->pending, hash_size)) {
372                 exit_exception_table(&s->complete, exception_cache);
373                 return -ENOMEM;
374         }
375 
376         return 0;
377 }
378 
379 /*
380  * Round a number up to the nearest 'size' boundary.  size must
381  * be a power of 2.
382  */
383 static inline ulong round_up(ulong n, ulong size)
384 {
385         size--;
386         return (n + size) & ~size;
387 }
388 
389 static void read_snapshot_metadata(struct dm_snapshot *s)
390 {
391         if (s->store.read_metadata(&s->store)) {
392                 down_write(&s->lock);
393                 s->valid = 0;
394                 up_write(&s->lock);
395 
396                 dm_table_event(s->table);
397         }
398 }
399 
400 /*
401  * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size>
402  */
403 static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
404 {
405         struct dm_snapshot *s;
406         unsigned long chunk_size;
407         int r = -EINVAL;
408         char persistent;
409         char *origin_path;
410         char *cow_path;
411         char *value;
412         int blocksize;
413 
414         if (argc < 4) {
415                 ti->error = "dm-snapshot: requires exactly 4 arguments";
416                 r = -EINVAL;
417                 goto bad1;
418         }
419 
420         origin_path = argv[0];
421         cow_path = argv[1];
422         persistent = toupper(*argv[2]);
423 
424         if (persistent != 'P' && persistent != 'N') {
425                 ti->error = "Persistent flag is not P or N";
426                 r = -EINVAL;
427                 goto bad1;
428         }
429 
430         chunk_size = simple_strtoul(argv[3], &value, 10);
431         if (chunk_size == 0 || value == NULL) {
432                 ti->error = "Invalid chunk size";
433                 r = -EINVAL;
434                 goto bad1;
435         }
436 
437         s = kmalloc(sizeof(*s), GFP_KERNEL);
438         if (s == NULL) {
439                 ti->error = "Cannot allocate snapshot context private "
440                     "structure";
441                 r = -ENOMEM;
442                 goto bad1;
443         }
444 
445         r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin);
446         if (r) {
447                 ti->error = "Cannot get origin device";
448                 goto bad2;
449         }
450 
451         r = dm_get_device(ti, cow_path, 0, 0,
452                           FMODE_READ | FMODE_WRITE, &s->cow);
453         if (r) {
454                 dm_put_device(ti, s->origin);
455                 ti->error = "Cannot get COW device";
456                 goto bad2;
457         }
458 
459         /*
460          * Chunk size must be multiple of page size.  Silently
461          * round up if it's not.
462          */
463         chunk_size = round_up(chunk_size, PAGE_SIZE >> 9);
464 
465         /* Validate the chunk size against the device block size */
466         blocksize = s->cow->bdev->bd_disk->queue->hardsect_size;
467         if (chunk_size % (blocksize >> 9)) {
468                 ti->error = "Chunk size is not a multiple of device blocksize";
469                 r = -EINVAL;
470                 goto bad3;
471         }
472 
473         /* Check chunk_size is a power of 2 */
474         if (chunk_size & (chunk_size - 1)) {
475                 ti->error = "Chunk size is not a power of 2";
476                 r = -EINVAL;
477                 goto bad3;
478         }
479 
480         s->chunk_size = chunk_size;
481         s->chunk_mask = chunk_size - 1;
482         s->type = persistent;
483         s->chunk_shift = ffs(chunk_size) - 1;
484 
485         s->valid = 1;
486         s->active = 0;
487         s->last_percent = 0;
488         init_rwsem(&s->lock);
489         s->table = ti->table;
490 
491         /* Allocate hash table for COW data */
492         if (init_hash_tables(s)) {
493                 ti->error = "Unable to allocate hash table space";
494                 r = -ENOMEM;
495                 goto bad3;
496         }
497 
498         /*
499          * Check the persistent flag - done here because we need the iobuf
500          * to check the LV header
501          */
502         s->store.snap = s;
503 
504         if (persistent == 'P')
505                 r = dm_create_persistent(&s->store, chunk_size);
506         else
507                 r = dm_create_transient(&s->store, s, blocksize);
508 
509         if (r) {
510                 ti->error = "Couldn't create exception store";
511                 r = -EINVAL;
512                 goto bad4;
513         }
514 
515         r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
516         if (r) {
517                 ti->error = "Could not create kcopyd client";
518                 goto bad5;
519         }
520 
521         /* Metadata must only be loaded into one table at once */
522         read_snapshot_metadata(s);
523 
524         /* Add snapshot to the list of snapshots for this origin */
525         /* Exceptions aren't triggered till snapshot_resume() is called */
526         if (register_snapshot(s)) {
527                 r = -EINVAL;
528                 ti->error = "Cannot register snapshot origin";
529                 goto bad6;
530         }
531 
532         ti->private = s;
533         ti->split_io = chunk_size;
534 
535         return 0;
536 
537  bad6:
538         kcopyd_client_destroy(s->kcopyd_client);
539 
540  bad5:
541         s->store.destroy(&s->store);
542 
543  bad4:
544         exit_exception_table(&s->pending, pending_cache);
545         exit_exception_table(&s->complete, exception_cache);
546 
547  bad3:
548         dm_put_device(ti, s->cow);
549         dm_put_device(ti, s->origin);
550 
551  bad2:
552         kfree(s);
553 
554  bad1:
555         return r;
556 }
557 
558 static void snapshot_dtr(struct dm_target *ti)
559 {
560         struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
561 
562         /* Prevent further origin writes from using this snapshot. */
563         /* After this returns there can be no new kcopyd jobs. */
564         unregister_snapshot(s);
565 
566         kcopyd_client_destroy(s->kcopyd_client);
567 
568         exit_exception_table(&s->pending, pending_cache);
569         exit_exception_table(&s->complete, exception_cache);
570 
571         /* Deallocate memory used */
572         s->store.destroy(&s->store);
573 
574         dm_put_device(ti, s->origin);
575         dm_put_device(ti, s->cow);
576 
577         kfree(s);
578 }
579 
580 /*
581  * Flush a list of buffers.
582  */
583 static void flush_bios(struct bio *bio)
584 {
585         struct bio *n;
586 
587         while (bio) {
588                 n = bio->bi_next;
589                 bio->bi_next = NULL;
590                 generic_make_request(bio);
591                 bio = n;
592         }
593 }
594 
595 /*
596  * Error a list of buffers.
597  */
598 static void error_bios(struct bio *bio)
599 {
600         struct bio *n;
601 
602         while (bio) {
603                 n = bio->bi_next;
604                 bio->bi_next = NULL;
605                 bio_io_error(bio, bio->bi_size);
606                 bio = n;
607         }
608 }
609 
610 static inline void error_snapshot_bios(struct pending_exception *pe)
611 {
612         error_bios(bio_list_get(&pe->snapshot_bios));
613 }
614 
615 static struct bio *__flush_bios(struct pending_exception *pe)
616 {
617         /*
618          * If this pe is involved in a write to the origin and
619          * it is the last sibling to complete then release
620          * the bios for the original write to the origin.
621          */
622 
623         if (pe->primary_pe &&
624             atomic_dec_and_test(&pe->primary_pe->sibling_count))
625                 return bio_list_get(&pe->primary_pe->origin_bios);
626 
627         return NULL;
628 }
629 
630 static void __invalidate_snapshot(struct dm_snapshot *s,
631                                 struct pending_exception *pe, int err)
632 {
633         if (!s->valid)
634                 return;
635 
636         if (err == -EIO)
637                 DMERR("Invalidating snapshot: Error reading/writing.");
638         else if (err == -ENOMEM)
639                 DMERR("Invalidating snapshot: Unable to allocate exception.");
640 
641         if (pe)
642                 remove_exception(&pe->e);
643 
644         if (s->store.drop_snapshot)
645                 s->store.drop_snapshot(&s->store);
646 
647         s->valid = 0;
648 
649         dm_table_event(s->table);
650 }
651 
652 static void pending_complete(struct pending_exception *pe, int success)
653 {
654         struct exception *e;
655         struct pending_exception *primary_pe;
656         struct dm_snapshot *s = pe->snap;
657         struct bio *flush = NULL;
658 
659         if (!success) {
660                 /* Read/write error - snapshot is unusable */
661                 down_write(&s->lock);
662                 __invalidate_snapshot(s, pe, -EIO);
663                 flush = __flush_bios(pe);
664                 up_write(&s->lock);
665 
666                 error_snapshot_bios(pe);
667                 goto out;
668         }
669 
670         e = alloc_exception();
671         if (!e) {
672                 down_write(&s->lock);
673                 __invalidate_snapshot(s, pe, -ENOMEM);
674                 flush = __flush_bios(pe);
675                 up_write(&s->lock);
676 
677                 error_snapshot_bios(pe);
678                 goto out;
679         }
680         *e = pe->e;
681 
682         /*
683          * Add a proper exception, and remove the
684          * in-flight exception from the list.
685          */
686         down_write(&s->lock);
687         if (!s->valid) {
688                 flush = __flush_bios(pe);
689                 up_write(&s->lock);
690 
691                 free_exception(e);
692 
693                 error_snapshot_bios(pe);
694                 goto out;
695         }
696 
697         insert_exception(&s->complete, e);
698         remove_exception(&pe->e);
699         flush = __flush_bios(pe);
700 
701         up_write(&s->lock);
702 
703         /* Submit any pending write bios */
704         flush_bios(bio_list_get(&pe->snapshot_bios));
705 
706  out:
707         primary_pe = pe->primary_pe;
708 
709         /*
710          * Free the pe if it's not linked to an origin write or if
711          * it's not itself a primary pe.
712          */
713         if (!primary_pe || primary_pe != pe)
714                 free_pending_exception(pe);
715 
716         /*
717          * Free the primary pe if nothing references it.
718          */
719         if (primary_pe && !atomic_read(&primary_pe->sibling_count))
720                 free_pending_exception(primary_pe);
721 
722         if (flush)
723                 flush_bios(flush);
724 }
725 
726 static void commit_callback(void *context, int success)
727 {
728         struct pending_exception *pe = (struct pending_exception *) context;
729         pending_complete(pe, success);
730 }
731 
732 /*
733  * Called when the copy I/O has finished.  kcopyd actually runs
734  * this code so don't block.
735  */
736 static void copy_callback(int read_err, unsigned int write_err, void *context)
737 {
738         struct pending_exception *pe = (struct pending_exception *) context;
739         struct dm_snapshot *s = pe->snap;
740 
741         if (read_err || write_err)
742                 pending_complete(pe, 0);
743 
744         else
745                 /* Update the metadata if we are persistent */
746                 s->store.commit_exception(&s->store, &pe->e, commit_callback,
747                                           pe);
748 }
749 
750 /*
751  * Dispatches the copy operation to kcopyd.
752  */
753 static void start_copy(struct pending_exception *pe)
754 {
755         struct dm_snapshot *s = pe->snap;
756         struct io_region src, dest;
757         struct block_device *bdev = s->origin->bdev;
758         sector_t dev_size;
759 
760         dev_size = get_dev_size(bdev);
761 
762         src.bdev = bdev;
763         src.sector = chunk_to_sector(s, pe->e.old_chunk);
764         src.count = min(s->chunk_size, dev_size - src.sector);
765 
766         dest.bdev = s->cow->bdev;
767         dest.sector = chunk_to_sector(s, pe->e.new_chunk);
768         dest.count = src.count;
769 
770         /* Hand over to kcopyd */
771         kcopyd_copy(s->kcopyd_client,
772                     &src, 1, &dest, 0, copy_callback, pe);
773 }
774 
775 /*
776  * Looks to see if this snapshot already has a pending exception
777  * for this chunk, otherwise it allocates a new one and inserts
778  * it into the pending table.
779  *
780  * NOTE: a write lock must be held on snap->lock before calling
781  * this.
782  */
783 static struct pending_exception *
784 __find_pending_exception(struct dm_snapshot *s, struct bio *bio)
785 {
786         struct exception *e;
787         struct pending_exception *pe;
788         chunk_t chunk = sector_to_chunk(s, bio->bi_sector);
789 
790         /*
791          * Is there a pending exception for this already ?
792          */
793         e = lookup_exception(&s->pending, chunk);
794         if (e) {
795                 /* cast the exception to a pending exception */
796                 pe = container_of(e, struct pending_exception, e);
797                 goto out;
798         }
799 
800         /*
801          * Create a new pending exception, we don't want
802          * to hold the lock while we do this.
803          */
804         up_write(&s->lock);
805         pe = alloc_pending_exception();
806         down_write(&s->lock);
807 
808         if (!s->valid) {
809                 free_pending_exception(pe);
810                 return NULL;
811         }
812 
813         e = lookup_exception(&s->pending, chunk);
814         if (e) {
815                 free_pending_exception(pe);
816                 pe = container_of(e, struct pending_exception, e);
817                 goto out;
818         }
819 
820         pe->e.old_chunk = chunk;
821         bio_list_init(&pe->origin_bios);
822         bio_list_init(&pe->snapshot_bios);
823         pe->primary_pe = NULL;
824         atomic_set(&pe->sibling_count, 1);
825         pe->snap = s;
826         pe->started = 0;
827 
828         if (s->store.prepare_exception(&s->store, &pe->e)) {
829                 free_pending_exception(pe);
830                 return NULL;
831         }
832 
833         insert_exception(&s->pending, &pe->e);
834 
835  out:
836         return pe;
837 }
838 
839 static inline void remap_exception(struct dm_snapshot *s, struct exception *e,
840                                    struct bio *bio)
841 {
842         bio->bi_bdev = s->cow->bdev;
843         bio->bi_sector = chunk_to_sector(s, e->new_chunk) +
844                 (bio->bi_sector & s->chunk_mask);
845 }
846 
847 static int snapshot_map(struct dm_target *ti, struct bio *bio,
848                         union map_info *map_context)
849 {
850         struct exception *e;
851         struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
852         int copy_needed = 0;
853         int r = 1;
854         chunk_t chunk;
855         struct pending_exception *pe = NULL;
856 
857         chunk = sector_to_chunk(s, bio->bi_sector);
858 
859         /* Full snapshots are not usable */
860         /* To get here the table must be live so s->active is always set. */
861         if (!s->valid)
862                 return -EIO;
863 
864         if (unlikely(bio_barrier(bio)))
865                 return -EOPNOTSUPP;
866 
867         /*
868          * Write to snapshot - higher level takes care of RW/RO
869          * flags so we should only get this if we are
870          * writeable.
871          */
872         if (bio_rw(bio) == WRITE) {
873 
874                 /* FIXME: should only take write lock if we need
875                  * to copy an exception */
876                 down_write(&s->lock);
877 
878                 if (!s->valid) {
879                         r = -EIO;
880                         goto out_unlock;
881                 }
882 
883                 /* If the block is already remapped - use that, else remap it */
884                 e = lookup_exception(&s->complete, chunk);
885                 if (e) {
886                         remap_exception(s, e, bio);
887                         goto out_unlock;
888                 }
889 
890                 pe = __find_pending_exception(s, bio);
891                 if (!pe) {
892                         __invalidate_snapshot(s, pe, -ENOMEM);
893                         r = -EIO;
894                         goto out_unlock;
895                 }
896 
897                 remap_exception(s, &pe->e, bio);
898                 bio_list_add(&pe->snapshot_bios, bio);
899 
900                 if (!pe->started) {
901                         /* this is protected by snap->lock */
902                         pe->started = 1;
903                         copy_needed = 1;
904                 }
905 
906                 r = 0;
907 
908  out_unlock:
909                 up_write(&s->lock);
910 
911                 if (copy_needed)
912                         start_copy(pe);
913         } else {
914                 /*
915                  * FIXME: this read path scares me because we
916                  * always use the origin when we have a pending
917                  * exception.  However I can't think of a
918                  * situation where this is wrong - ejt.
919                  */
920 
921                 /* Do reads */
922                 down_read(&s->lock);
923 
924                 if (!s->valid) {
925                         up_read(&s->lock);
926                         return -EIO;
927                 }
928 
929                 /* See if it it has been remapped */
930                 e = lookup_exception(&s->complete, chunk);
931                 if (e)
932                         remap_exception(s, e, bio);
933                 else
934                         bio->bi_bdev = s->origin->bdev;
935 
936                 up_read(&s->lock);
937         }
938 
939         return r;
940 }
941 
942 static void snapshot_resume(struct dm_target *ti)
943 {
944         struct dm_snapshot *s = (struct dm_snapshot *) ti->private;
945 
946         down_write(&s->lock);
947         s->active = 1;
948         up_write(&s->lock);
949 }
950 
951 static int snapshot_status(struct dm_target *ti, status_type_t type,
952                            char *result, unsigned int maxlen)
953 {
954         struct dm_snapshot *snap = (struct dm_snapshot *) ti->private;
955 
956         switch (type) {
957         case STATUSTYPE_INFO:
958                 if (!snap->valid)
959                         snprintf(result, maxlen, "Invalid");
960                 else {
961                         if (snap->store.fraction_full) {
962                                 sector_t numerator, denominator;
963                                 snap->store.fraction_full(&snap->store,
964                                                           &numerator,
965                                                           &denominator);
966                                 snprintf(result, maxlen, "%llu/%llu",
967                                         (unsigned long long)numerator,
968                                         (unsigned long long)denominator);
969                         }
970                         else
971                                 snprintf(result, maxlen, "Unknown");
972                 }
973                 break;
974 
975         case STATUSTYPE_TABLE:
976                 /*
977                  * kdevname returns a static pointer so we need
978                  * to make private copies if the output is to
979                  * make sense.
980                  */
981                 snprintf(result, maxlen, "%s %s %c %llu",
982                          snap->origin->name, snap->cow->name,
983                          snap->type,
984                          (unsigned long long)snap->chunk_size);
985                 break;
986         }
987 
988         return 0;
989 }
990 
991 /*-----------------------------------------------------------------
992  * Origin methods
993  *---------------------------------------------------------------*/
994 static int __origin_write(struct list_head *snapshots, struct bio *bio)
995 {
996         int r = 1, first = 0;
997         struct dm_snapshot *snap;
998         struct exception *e;
999         struct pending_exception *pe, *next_pe, *primary_pe = NULL;
1000         chunk_t chunk;
1001         LIST_HEAD(pe_queue);
1002 
1003         /* Do all the snapshots on this origin */
1004         list_for_each_entry (snap, snapshots, list) {
1005 
1006                 down_write(&snap->lock);
1007 
1008                 /* Only deal with valid and active snapshots */
1009                 if (!snap->valid || !snap->active)
1010                         goto next_snapshot;
1011 
1012                 /* Nothing to do if writing beyond end of snapshot */
1013                 if (bio->bi_sector >= dm_table_get_size(snap->table))
1014                         goto next_snapshot;
1015 
1016                 /*
1017                  * Remember, different snapshots can have
1018                  * different chunk sizes.
1019                  */
1020                 chunk = sector_to_chunk(snap, bio->bi_sector);
1021 
1022                 /*
1023                  * Check exception table to see if block
1024                  * is already remapped in this snapshot
1025                  * and trigger an exception if not.
1026                  *
1027                  * sibling_count is initialised to 1 so pending_complete()
1028                  * won't destroy the primary_pe while we're inside this loop.
1029                  */
1030                 e = lookup_exception(&snap->complete, chunk);
1031                 if (e)
1032                         goto next_snapshot;
1033 
1034                 pe = __find_pending_exception(snap, bio);
1035                 if (!pe) {
1036                         __invalidate_snapshot(snap, pe, ENOMEM);
1037                         goto next_snapshot;
1038                 }
1039 
1040                 if (!primary_pe) {
1041                         /*
1042                          * Either every pe here has same
1043                          * primary_pe or none has one yet.
1044                          */
1045                         if (pe->primary_pe)
1046                                 primary_pe = pe->primary_pe;
1047                         else {
1048                                 primary_pe = pe;
1049                                 first = 1;
1050                         }
1051 
1052                         bio_list_add(&primary_pe->origin_bios, bio);
1053 
1054                         r = 0;
1055                 }
1056 
1057                 if (!pe->primary_pe) {
1058                         atomic_inc(&primary_pe->sibling_count);
1059                         pe->primary_pe = primary_pe;
1060                 }
1061 
1062                 if (!pe->started) {
1063                         pe->started = 1;
1064                         list_add_tail(&pe->list, &pe_queue);
1065                 }
1066 
1067  next_snapshot:
1068                 up_write(&snap->lock);
1069         }
1070 
1071         if (!primary_pe)