aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/md/dm-exception-store.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/md/dm-exception-store.c')
-rw-r--r--drivers/md/dm-exception-store.c749
1 files changed, 19 insertions, 730 deletions
diff --git a/drivers/md/dm-exception-store.c b/drivers/md/dm-exception-store.c
index c5c9a2652315..74777e0f80df 100644
--- a/drivers/md/dm-exception-store.c
+++ b/drivers/md/dm-exception-store.c
@@ -1,757 +1,46 @@
1/* 1/*
2 * dm-exception-store.c
3 *
4 * Copyright (C) 2001-2002 Sistina Software (UK) Limited. 2 * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
5 * Copyright (C) 2006 Red Hat GmbH 3 * Copyright (C) 2006-2008 Red Hat GmbH
6 * 4 *
7 * This file is released under the GPL. 5 * This file is released under the GPL.
8 */ 6 */
9 7
10#include "dm-exception-store.h" 8#include "dm-exception-store.h"
11#include "dm-snap.h"
12 9
13#include <linux/mm.h> 10#include <linux/mm.h>
14#include <linux/pagemap.h> 11#include <linux/pagemap.h>
15#include <linux/vmalloc.h> 12#include <linux/vmalloc.h>
16#include <linux/slab.h> 13#include <linux/slab.h>
17#include <linux/dm-io.h> 14#include <linux/device-mapper.h>
18#include <linux/dm-kcopyd.h>
19
20#define DM_MSG_PREFIX "snapshots"
21#define DM_CHUNK_SIZE_DEFAULT_SECTORS 32 /* 16KB */
22
23/*-----------------------------------------------------------------
24 * Persistent snapshots, by persistent we mean that the snapshot
25 * will survive a reboot.
26 *---------------------------------------------------------------*/
27
28/*
29 * We need to store a record of which parts of the origin have
30 * been copied to the snapshot device. The snapshot code
31 * requires that we copy exception chunks to chunk aligned areas
32 * of the COW store. It makes sense therefore, to store the
33 * metadata in chunk size blocks.
34 *
35 * There is no backward or forward compatibility implemented,
36 * snapshots with different disk versions than the kernel will
37 * not be usable. It is expected that "lvcreate" will blank out
38 * the start of a fresh COW device before calling the snapshot
39 * constructor.
40 *
41 * The first chunk of the COW device just contains the header.
42 * After this there is a chunk filled with exception metadata,
43 * followed by as many exception chunks as can fit in the
44 * metadata areas.
45 *
46 * All on disk structures are in little-endian format. The end
47 * of the exceptions info is indicated by an exception with a
48 * new_chunk of 0, which is invalid since it would point to the
49 * header chunk.
50 */
51
52/*
53 * Magic for persistent snapshots: "SnAp" - Feeble isn't it.
54 */
55#define SNAP_MAGIC 0x70416e53
56
57/*
58 * The on-disk version of the metadata.
59 */
60#define SNAPSHOT_DISK_VERSION 1
61
62struct disk_header {
63 uint32_t magic;
64
65 /*
66 * Is this snapshot valid. There is no way of recovering
67 * an invalid snapshot.
68 */
69 uint32_t valid;
70
71 /*
72 * Simple, incrementing version. no backward
73 * compatibility.
74 */
75 uint32_t version;
76
77 /* In sectors */
78 uint32_t chunk_size;
79};
80
81struct disk_exception {
82 uint64_t old_chunk;
83 uint64_t new_chunk;
84};
85
86struct commit_callback {
87 void (*callback)(void *, int success);
88 void *context;
89};
90
91/*
92 * The top level structure for a persistent exception store.
93 */
94struct pstore {
95 struct dm_snapshot *snap; /* up pointer to my snapshot */
96 int version;
97 int valid;
98 uint32_t exceptions_per_area;
99
100 /*
101 * Now that we have an asynchronous kcopyd there is no
102 * need for large chunk sizes, so it wont hurt to have a
103 * whole chunks worth of metadata in memory at once.
104 */
105 void *area;
106
107 /*
108 * An area of zeros used to clear the next area.
109 */
110 void *zero_area;
111
112 /*
113 * Used to keep track of which metadata area the data in
114 * 'chunk' refers to.
115 */
116 chunk_t current_area;
117
118 /*
119 * The next free chunk for an exception.
120 */
121 chunk_t next_free;
122
123 /*
124 * The index of next free exception in the current
125 * metadata area.
126 */
127 uint32_t current_committed;
128
129 atomic_t pending_count;
130 uint32_t callback_count;
131 struct commit_callback *callbacks;
132 struct dm_io_client *io_client;
133
134 struct workqueue_struct *metadata_wq;
135};
136
137static unsigned sectors_to_pages(unsigned sectors)
138{
139 return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9);
140}
141
142static int alloc_area(struct pstore *ps)
143{
144 int r = -ENOMEM;
145 size_t len;
146
147 len = ps->snap->chunk_size << SECTOR_SHIFT;
148
149 /*
150 * Allocate the chunk_size block of memory that will hold
151 * a single metadata area.
152 */
153 ps->area = vmalloc(len);
154 if (!ps->area)
155 return r;
156
157 ps->zero_area = vmalloc(len);
158 if (!ps->zero_area) {
159 vfree(ps->area);
160 return r;
161 }
162 memset(ps->zero_area, 0, len);
163
164 return 0;
165}
166
167static void free_area(struct pstore *ps)
168{
169 vfree(ps->area);
170 ps->area = NULL;
171 vfree(ps->zero_area);
172 ps->zero_area = NULL;
173}
174
175struct mdata_req {
176 struct dm_io_region *where;
177 struct dm_io_request *io_req;
178 struct work_struct work;
179 int result;
180};
181
182static void do_metadata(struct work_struct *work)
183{
184 struct mdata_req *req = container_of(work, struct mdata_req, work);
185
186 req->result = dm_io(req->io_req, 1, req->where, NULL);
187}
188
189/*
190 * Read or write a chunk aligned and sized block of data from a device.
191 */
192static int chunk_io(struct pstore *ps, chunk_t chunk, int rw, int metadata)
193{
194 struct dm_io_region where = {
195 .bdev = ps->snap->cow->bdev,
196 .sector = ps->snap->chunk_size * chunk,
197 .count = ps->snap->chunk_size,
198 };
199 struct dm_io_request io_req = {
200 .bi_rw = rw,
201 .mem.type = DM_IO_VMA,
202 .mem.ptr.vma = ps->area,
203 .client = ps->io_client,
204 .notify.fn = NULL,
205 };
206 struct mdata_req req;
207
208 if (!metadata)
209 return dm_io(&io_req, 1, &where, NULL);
210
211 req.where = &where;
212 req.io_req = &io_req;
213
214 /*
215 * Issue the synchronous I/O from a different thread
216 * to avoid generic_make_request recursion.
217 */
218 INIT_WORK(&req.work, do_metadata);
219 queue_work(ps->metadata_wq, &req.work);
220 flush_workqueue(ps->metadata_wq);
221
222 return req.result;
223}
224
225/*
226 * Convert a metadata area index to a chunk index.
227 */
228static chunk_t area_location(struct pstore *ps, chunk_t area)
229{
230 return 1 + ((ps->exceptions_per_area + 1) * area);
231}
232
233/*
234 * Read or write a metadata area. Remembering to skip the first
235 * chunk which holds the header.
236 */
237static int area_io(struct pstore *ps, int rw)
238{
239 int r;
240 chunk_t chunk;
241
242 chunk = area_location(ps, ps->current_area);
243
244 r = chunk_io(ps, chunk, rw, 0);
245 if (r)
246 return r;
247
248 return 0;
249}
250
251static void zero_memory_area(struct pstore *ps)
252{
253 memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT);
254}
255
256static int zero_disk_area(struct pstore *ps, chunk_t area)
257{
258 struct dm_io_region where = {
259 .bdev = ps->snap->cow->bdev,
260 .sector = ps->snap->chunk_size * area_location(ps, area),
261 .count = ps->snap->chunk_size,
262 };
263 struct dm_io_request io_req = {
264 .bi_rw = WRITE,
265 .mem.type = DM_IO_VMA,
266 .mem.ptr.vma = ps->zero_area,
267 .client = ps->io_client,
268 .notify.fn = NULL,
269 };
270
271 return dm_io(&io_req, 1, &where, NULL);
272}
273
274static int read_header(struct pstore *ps, int *new_snapshot)
275{
276 int r;
277 struct disk_header *dh;
278 chunk_t chunk_size;
279 int chunk_size_supplied = 1;
280
281 /*
282 * Use default chunk size (or hardsect_size, if larger) if none supplied
283 */
284 if (!ps->snap->chunk_size) {
285 ps->snap->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
286 bdev_hardsect_size(ps->snap->cow->bdev) >> 9);
287 ps->snap->chunk_mask = ps->snap->chunk_size - 1;
288 ps->snap->chunk_shift = ffs(ps->snap->chunk_size) - 1;
289 chunk_size_supplied = 0;
290 }
291
292 ps->io_client = dm_io_client_create(sectors_to_pages(ps->snap->
293 chunk_size));
294 if (IS_ERR(ps->io_client))
295 return PTR_ERR(ps->io_client);
296
297 r = alloc_area(ps);
298 if (r)
299 return r;
300
301 r = chunk_io(ps, 0, READ, 1);
302 if (r)
303 goto bad;
304
305 dh = (struct disk_header *) ps->area;
306
307 if (le32_to_cpu(dh->magic) == 0) {
308 *new_snapshot = 1;
309 return 0;
310 }
311
312 if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
313 DMWARN("Invalid or corrupt snapshot");
314 r = -ENXIO;
315 goto bad;
316 }
317
318 *new_snapshot = 0;
319 ps->valid = le32_to_cpu(dh->valid);
320 ps->version = le32_to_cpu(dh->version);
321 chunk_size = le32_to_cpu(dh->chunk_size);
322
323 if (!chunk_size_supplied || ps->snap->chunk_size == chunk_size)
324 return 0;
325
326 DMWARN("chunk size %llu in device metadata overrides "
327 "table chunk size of %llu.",
328 (unsigned long long)chunk_size,
329 (unsigned long long)ps->snap->chunk_size);
330
331 /* We had a bogus chunk_size. Fix stuff up. */
332 free_area(ps);
333
334 ps->snap->chunk_size = chunk_size;
335 ps->snap->chunk_mask = chunk_size - 1;
336 ps->snap->chunk_shift = ffs(chunk_size) - 1;
337
338 r = dm_io_client_resize(sectors_to_pages(ps->snap->chunk_size),
339 ps->io_client);
340 if (r)
341 return r;
342
343 r = alloc_area(ps);
344 return r;
345
346bad:
347 free_area(ps);
348 return r;
349}
350
351static int write_header(struct pstore *ps)
352{
353 struct disk_header *dh;
354
355 memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT);
356
357 dh = (struct disk_header *) ps->area;
358 dh->magic = cpu_to_le32(SNAP_MAGIC);
359 dh->valid = cpu_to_le32(ps->valid);
360 dh->version = cpu_to_le32(ps->version);
361 dh->chunk_size = cpu_to_le32(ps->snap->chunk_size);
362
363 return chunk_io(ps, 0, WRITE, 1);
364}
365
366/*
367 * Access functions for the disk exceptions, these do the endian conversions.
368 */
369static struct disk_exception *get_exception(struct pstore *ps, uint32_t index)
370{
371 BUG_ON(index >= ps->exceptions_per_area);
372
373 return ((struct disk_exception *) ps->area) + index;
374}
375 15
376static void read_exception(struct pstore *ps, 16#define DM_MSG_PREFIX "snapshot exception stores"
377 uint32_t index, struct disk_exception *result)
378{
379 struct disk_exception *e = get_exception(ps, index);
380
381 /* copy it */
382 result->old_chunk = le64_to_cpu(e->old_chunk);
383 result->new_chunk = le64_to_cpu(e->new_chunk);
384}
385
386static void write_exception(struct pstore *ps,
387 uint32_t index, struct disk_exception *de)
388{
389 struct disk_exception *e = get_exception(ps, index);
390
391 /* copy it */
392 e->old_chunk = cpu_to_le64(de->old_chunk);
393 e->new_chunk = cpu_to_le64(de->new_chunk);
394}
395 17
396/* 18int dm_exception_store_init(void)
397 * Registers the exceptions that are present in the current area.
398 * 'full' is filled in to indicate if the area has been
399 * filled.
400 */
401static int insert_exceptions(struct pstore *ps, int *full)
402{ 19{
403 int r; 20 int r;
404 unsigned int i;
405 struct disk_exception de;
406
407 /* presume the area is full */
408 *full = 1;
409
410 for (i = 0; i < ps->exceptions_per_area; i++) {
411 read_exception(ps, i, &de);
412
413 /*
414 * If the new_chunk is pointing at the start of
415 * the COW device, where the first metadata area
416 * is we know that we've hit the end of the
417 * exceptions. Therefore the area is not full.
418 */
419 if (de.new_chunk == 0LL) {
420 ps->current_committed = i;
421 *full = 0;
422 break;
423 }
424
425 /*
426 * Keep track of the start of the free chunks.
427 */
428 if (ps->next_free <= de.new_chunk)
429 ps->next_free = de.new_chunk + 1;
430
431 /*
432 * Otherwise we add the exception to the snapshot.
433 */
434 r = dm_add_exception(ps->snap, de.old_chunk, de.new_chunk);
435 if (r)
436 return r;
437 }
438
439 return 0;
440}
441
442static int read_exceptions(struct pstore *ps)
443{
444 int r, full = 1;
445
446 /*
447 * Keeping reading chunks and inserting exceptions until
448 * we find a partially full area.
449 */
450 for (ps->current_area = 0; full; ps->current_area++) {
451 r = area_io(ps, READ);
452 if (r)
453 return r;
454 21
455 r = insert_exceptions(ps, &full); 22 r = dm_transient_snapshot_init();
456 if (r) 23 if (r) {
457 return r; 24 DMERR("Unable to register transient exception store type.");
25 goto transient_fail;
458 } 26 }
459 27
460 ps->current_area--; 28 r = dm_persistent_snapshot_init();
461 29 if (r) {
462 return 0; 30 DMERR("Unable to register persistent exception store type");
463} 31 goto persistent_fail;
464
465static struct pstore *get_info(struct dm_exception_store *store)
466{
467 return (struct pstore *) store->context;
468}
469
470static void persistent_fraction_full(struct dm_exception_store *store,
471 sector_t *numerator, sector_t *denominator)
472{
473 *numerator = get_info(store)->next_free * store->snap->chunk_size;
474 *denominator = get_dev_size(store->snap->cow->bdev);
475}
476
477static void persistent_destroy(struct dm_exception_store *store)
478{
479 struct pstore *ps = get_info(store);
480
481 destroy_workqueue(ps->metadata_wq);
482 dm_io_client_destroy(ps->io_client);
483 vfree(ps->callbacks);
484 free_area(ps);
485 kfree(ps);
486}
487
488static int persistent_read_metadata(struct dm_exception_store *store)
489{
490 int r, uninitialized_var(new_snapshot);
491 struct pstore *ps = get_info(store);
492
493 /*
494 * Read the snapshot header.
495 */
496 r = read_header(ps, &new_snapshot);
497 if (r)
498 return r;
499
500 /*
501 * Now we know correct chunk_size, complete the initialisation.
502 */
503 ps->exceptions_per_area = (ps->snap->chunk_size << SECTOR_SHIFT) /
504 sizeof(struct disk_exception);
505 ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
506 sizeof(*ps->callbacks));
507 if (!ps->callbacks)
508 return -ENOMEM;
509
510 /*
511 * Do we need to setup a new snapshot ?
512 */
513 if (new_snapshot) {
514 r = write_header(ps);
515 if (r) {
516 DMWARN("write_header failed");
517 return r;
518 }
519
520 ps->current_area = 0;
521 zero_memory_area(ps);
522 r = zero_disk_area(ps, 0);
523 if (r) {
524 DMWARN("zero_disk_area(0) failed");
525 return r;
526 }
527 } else {
528 /*
529 * Sanity checks.
530 */
531 if (ps->version != SNAPSHOT_DISK_VERSION) {
532 DMWARN("unable to handle snapshot disk version %d",
533 ps->version);
534 return -EINVAL;
535 }
536
537 /*
538 * Metadata are valid, but snapshot is invalidated
539 */
540 if (!ps->valid)
541 return 1;
542
543 /*
544 * Read the metadata.
545 */
546 r = read_exceptions(ps);
547 if (r)
548 return r;
549 } 32 }
550 33
551 return 0; 34 return 0;
552}
553
554static int persistent_prepare(struct dm_exception_store *store,
555 struct dm_snap_exception *e)
556{
557 struct pstore *ps = get_info(store);
558 uint32_t stride;
559 chunk_t next_free;
560 sector_t size = get_dev_size(store->snap->cow->bdev);
561
562 /* Is there enough room ? */
563 if (size < ((ps->next_free + 1) * store->snap->chunk_size))
564 return -ENOSPC;
565 35
566 e->new_chunk = ps->next_free; 36persistent_fail:
567 37 dm_persistent_snapshot_exit();
568 /* 38transient_fail:
569 * Move onto the next free pending, making sure to take 39 return r;
570 * into account the location of the metadata chunks.
571 */
572 stride = (ps->exceptions_per_area + 1);
573 next_free = ++ps->next_free;
574 if (sector_div(next_free, stride) == 1)
575 ps->next_free++;
576
577 atomic_inc(&ps->pending_count);
578 return 0;
579}
580
581static void persistent_commit(struct dm_exception_store *store,
582 struct dm_snap_exception *e,
583 void (*callback) (void *, int success),
584 void *callback_context)
585{
586 unsigned int i;
587 struct pstore *ps = get_info(store);
588 struct disk_exception de;
589 struct commit_callback *cb;
590
591 de.old_chunk = e->old_chunk;
592 de.new_chunk = e->new_chunk;
593 write_exception(ps, ps->current_committed++, &de);
594
595 /*
596 * Add the callback to the back of the array. This code
597 * is the only place where the callback array is
598 * manipulated, and we know that it will never be called
599 * multiple times concurrently.
600 */
601 cb = ps->callbacks + ps->callback_count++;
602 cb->callback = callback;
603 cb->context = callback_context;
604
605 /*
606 * If there are exceptions in flight and we have not yet
607 * filled this metadata area there's nothing more to do.
608 */
609 if (!atomic_dec_and_test(&ps->pending_count) &&
610 (ps->current_committed != ps->exceptions_per_area))
611 return;
612
613 /*
614 * If we completely filled the current area, then wipe the next one.
615 */
616 if ((ps->current_committed == ps->exceptions_per_area) &&
617 zero_disk_area(ps, ps->current_area + 1))
618 ps->valid = 0;
619
620 /*
621 * Commit exceptions to disk.
622 */
623 if (ps->valid && area_io(ps, WRITE))
624 ps->valid = 0;
625
626 /*
627 * Advance to the next area if this one is full.
628 */
629 if (ps->current_committed == ps->exceptions_per_area) {
630 ps->current_committed = 0;
631 ps->current_area++;
632 zero_memory_area(ps);
633 }
634
635 for (i = 0; i < ps->callback_count; i++) {
636 cb = ps->callbacks + i;
637 cb->callback(cb->context, ps->valid);
638 }
639
640 ps->callback_count = 0;
641}
642
643static void persistent_drop(struct dm_exception_store *store)
644{
645 struct pstore *ps = get_info(store);
646
647 ps->valid = 0;
648 if (write_header(ps))
649 DMWARN("write header failed");
650}
651
652int dm_create_persistent(struct dm_exception_store *store)
653{
654 struct pstore *ps;
655
656 /* allocate the pstore */
657 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
658 if (!ps)
659 return -ENOMEM;
660
661 ps->snap = store->snap;
662 ps->valid = 1;
663 ps->version = SNAPSHOT_DISK_VERSION;
664 ps->area = NULL;
665 ps->next_free = 2; /* skipping the header and first area */
666 ps->current_committed = 0;
667
668 ps->callback_count = 0;
669 atomic_set(&ps->pending_count, 0);
670 ps->callbacks = NULL;
671
672 ps->metadata_wq = create_singlethread_workqueue("ksnaphd");
673 if (!ps->metadata_wq) {
674 kfree(ps);
675 DMERR("couldn't start header metadata update thread");
676 return -ENOMEM;
677 }
678
679 store->destroy = persistent_destroy;
680 store->read_metadata = persistent_read_metadata;
681 store->prepare_exception = persistent_prepare;
682 store->commit_exception = persistent_commit;
683 store->drop_snapshot = persistent_drop;
684 store->fraction_full = persistent_fraction_full;
685 store->context = ps;
686
687 return 0;
688}
689
690/*-----------------------------------------------------------------
691 * Implementation of the store for non-persistent snapshots.
692 *---------------------------------------------------------------*/
693struct transient_c {
694 sector_t next_free;
695};
696
697static void transient_destroy(struct dm_exception_store *store)
698{
699 kfree(store->context);
700}
701
702static int transient_read_metadata(struct dm_exception_store *store)
703{
704 return 0;
705}
706
707static int transient_prepare(struct dm_exception_store *store,
708 struct dm_snap_exception *e)
709{
710 struct transient_c *tc = (struct transient_c *) store->context;
711 sector_t size = get_dev_size(store->snap->cow->bdev);
712
713 if (size < (tc->next_free + store->snap->chunk_size))
714 return -1;
715
716 e->new_chunk = sector_to_chunk(store->snap, tc->next_free);
717 tc->next_free += store->snap->chunk_size;
718
719 return 0;
720}
721
722static void transient_commit(struct dm_exception_store *store,
723 struct dm_snap_exception *e,
724 void (*callback) (void *, int success),
725 void *callback_context)
726{
727 /* Just succeed */
728 callback(callback_context, 1);
729}
730
731static void transient_fraction_full(struct dm_exception_store *store,
732 sector_t *numerator, sector_t *denominator)
733{
734 *numerator = ((struct transient_c *) store->context)->next_free;
735 *denominator = get_dev_size(store->snap->cow->bdev);
736} 40}
737 41
738int dm_create_transient(struct dm_exception_store *store) 42void dm_exception_store_exit(void)
739{ 43{
740 struct transient_c *tc; 44 dm_persistent_snapshot_exit();
741 45 dm_transient_snapshot_exit();
742 store->destroy = transient_destroy;
743 store->read_metadata = transient_read_metadata;
744 store->prepare_exception = transient_prepare;
745 store->commit_exception = transient_commit;
746 store->drop_snapshot = NULL;
747 store->fraction_full = transient_fraction_full;
748
749 tc = kmalloc(sizeof(struct transient_c), GFP_KERNEL);
750 if (!tc)
751 return -ENOMEM;
752
753 tc->next_free = 0;
754 store->context = tc;
755
756 return 0;
757} 46}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-23 05:53:12 -0500