xref: /openbmc/linux/drivers/md/dm-snap-persistent.c (revision 55494bf2947dccdf2d98b62374fea7365dfead84)
14db6bfe0SAlasdair G Kergon /*
24db6bfe0SAlasdair G Kergon  * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
34db6bfe0SAlasdair G Kergon  * Copyright (C) 2006-2008 Red Hat GmbH
44db6bfe0SAlasdair G Kergon  *
54db6bfe0SAlasdair G Kergon  * This file is released under the GPL.
64db6bfe0SAlasdair G Kergon  */
74db6bfe0SAlasdair G Kergon 
84db6bfe0SAlasdair G Kergon #include "dm-exception-store.h"
94db6bfe0SAlasdair G Kergon 
104db6bfe0SAlasdair G Kergon #include <linux/mm.h>
114db6bfe0SAlasdair G Kergon #include <linux/pagemap.h>
124db6bfe0SAlasdair G Kergon #include <linux/vmalloc.h>
13daaa5f7cSPaul Gortmaker #include <linux/export.h>
144db6bfe0SAlasdair G Kergon #include <linux/slab.h>
154db6bfe0SAlasdair G Kergon #include <linux/dm-io.h>
16*55494bf2SMikulas Patocka #include "dm-bufio.h"
174db6bfe0SAlasdair G Kergon 
184db6bfe0SAlasdair G Kergon #define DM_MSG_PREFIX "persistent snapshot"
194db6bfe0SAlasdair G Kergon #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32	/* 16KB */
204db6bfe0SAlasdair G Kergon 
214db6bfe0SAlasdair G Kergon /*-----------------------------------------------------------------
224db6bfe0SAlasdair G Kergon  * Persistent snapshots, by persistent we mean that the snapshot
234db6bfe0SAlasdair G Kergon  * will survive a reboot.
244db6bfe0SAlasdair G Kergon  *---------------------------------------------------------------*/
254db6bfe0SAlasdair G Kergon 
264db6bfe0SAlasdair G Kergon /*
274db6bfe0SAlasdair G Kergon  * We need to store a record of which parts of the origin have
284db6bfe0SAlasdair G Kergon  * been copied to the snapshot device.  The snapshot code
294db6bfe0SAlasdair G Kergon  * requires that we copy exception chunks to chunk aligned areas
304db6bfe0SAlasdair G Kergon  * of the COW store.  It makes sense therefore, to store the
314db6bfe0SAlasdair G Kergon  * metadata in chunk size blocks.
324db6bfe0SAlasdair G Kergon  *
334db6bfe0SAlasdair G Kergon  * There is no backward or forward compatibility implemented,
344db6bfe0SAlasdair G Kergon  * snapshots with different disk versions than the kernel will
354db6bfe0SAlasdair G Kergon  * not be usable.  It is expected that "lvcreate" will blank out
364db6bfe0SAlasdair G Kergon  * the start of a fresh COW device before calling the snapshot
374db6bfe0SAlasdair G Kergon  * constructor.
384db6bfe0SAlasdair G Kergon  *
394db6bfe0SAlasdair G Kergon  * The first chunk of the COW device just contains the header.
404db6bfe0SAlasdair G Kergon  * After this there is a chunk filled with exception metadata,
414db6bfe0SAlasdair G Kergon  * followed by as many exception chunks as can fit in the
424db6bfe0SAlasdair G Kergon  * metadata areas.
434db6bfe0SAlasdair G Kergon  *
444db6bfe0SAlasdair G Kergon  * All on disk structures are in little-endian format.  The end
454db6bfe0SAlasdair G Kergon  * of the exceptions info is indicated by an exception with a
464db6bfe0SAlasdair G Kergon  * new_chunk of 0, which is invalid since it would point to the
474db6bfe0SAlasdair G Kergon  * header chunk.
484db6bfe0SAlasdair G Kergon  */
494db6bfe0SAlasdair G Kergon 
504db6bfe0SAlasdair G Kergon /*
514db6bfe0SAlasdair G Kergon  * Magic for persistent snapshots: "SnAp" - Feeble isn't it.
524db6bfe0SAlasdair G Kergon  */
534db6bfe0SAlasdair G Kergon #define SNAP_MAGIC 0x70416e53
544db6bfe0SAlasdair G Kergon 
554db6bfe0SAlasdair G Kergon /*
564db6bfe0SAlasdair G Kergon  * The on-disk version of the metadata.
574db6bfe0SAlasdair G Kergon  */
584db6bfe0SAlasdair G Kergon #define SNAPSHOT_DISK_VERSION 1
594db6bfe0SAlasdair G Kergon 
604454a621SMikulas Patocka #define NUM_SNAPSHOT_HDR_CHUNKS 1
614454a621SMikulas Patocka 
624db6bfe0SAlasdair G Kergon struct disk_header {
63283a8328SAlasdair G Kergon 	__le32 magic;
644db6bfe0SAlasdair G Kergon 
654db6bfe0SAlasdair G Kergon 	/*
664db6bfe0SAlasdair G Kergon 	 * Is this snapshot valid.  There is no way of recovering
674db6bfe0SAlasdair G Kergon 	 * an invalid snapshot.
684db6bfe0SAlasdair G Kergon 	 */
69283a8328SAlasdair G Kergon 	__le32 valid;
704db6bfe0SAlasdair G Kergon 
714db6bfe0SAlasdair G Kergon 	/*
724db6bfe0SAlasdair G Kergon 	 * Simple, incrementing version. no backward
734db6bfe0SAlasdair G Kergon 	 * compatibility.
744db6bfe0SAlasdair G Kergon 	 */
75283a8328SAlasdair G Kergon 	__le32 version;
764db6bfe0SAlasdair G Kergon 
774db6bfe0SAlasdair G Kergon 	/* In sectors */
78283a8328SAlasdair G Kergon 	__le32 chunk_size;
79283a8328SAlasdair G Kergon } __packed;
804db6bfe0SAlasdair G Kergon 
814db6bfe0SAlasdair G Kergon struct disk_exception {
82283a8328SAlasdair G Kergon 	__le64 old_chunk;
83283a8328SAlasdair G Kergon 	__le64 new_chunk;
84283a8328SAlasdair G Kergon } __packed;
85283a8328SAlasdair G Kergon 
86283a8328SAlasdair G Kergon struct core_exception {
874db6bfe0SAlasdair G Kergon 	uint64_t old_chunk;
884db6bfe0SAlasdair G Kergon 	uint64_t new_chunk;
894db6bfe0SAlasdair G Kergon };
904db6bfe0SAlasdair G Kergon 
914db6bfe0SAlasdair G Kergon struct commit_callback {
924db6bfe0SAlasdair G Kergon 	void (*callback)(void *, int success);
934db6bfe0SAlasdair G Kergon 	void *context;
944db6bfe0SAlasdair G Kergon };
954db6bfe0SAlasdair G Kergon 
964db6bfe0SAlasdair G Kergon /*
974db6bfe0SAlasdair G Kergon  * The top level structure for a persistent exception store.
984db6bfe0SAlasdair G Kergon  */
994db6bfe0SAlasdair G Kergon struct pstore {
10071fab00aSJonathan Brassow 	struct dm_exception_store *store;
1014db6bfe0SAlasdair G Kergon 	int version;
1024db6bfe0SAlasdair G Kergon 	int valid;
1034db6bfe0SAlasdair G Kergon 	uint32_t exceptions_per_area;
1044db6bfe0SAlasdair G Kergon 
1054db6bfe0SAlasdair G Kergon 	/*
1064db6bfe0SAlasdair G Kergon 	 * Now that we have an asynchronous kcopyd there is no
1074db6bfe0SAlasdair G Kergon 	 * need for large chunk sizes, so it wont hurt to have a
1084db6bfe0SAlasdair G Kergon 	 * whole chunks worth of metadata in memory at once.
1094db6bfe0SAlasdair G Kergon 	 */
1104db6bfe0SAlasdair G Kergon 	void *area;
1114db6bfe0SAlasdair G Kergon 
1124db6bfe0SAlasdair G Kergon 	/*
1134db6bfe0SAlasdair G Kergon 	 * An area of zeros used to clear the next area.
1144db6bfe0SAlasdair G Kergon 	 */
1154db6bfe0SAlasdair G Kergon 	void *zero_area;
1164db6bfe0SAlasdair G Kergon 
1174db6bfe0SAlasdair G Kergon 	/*
11861578dcdSMikulas Patocka 	 * An area used for header. The header can be written
11961578dcdSMikulas Patocka 	 * concurrently with metadata (when invalidating the snapshot),
12061578dcdSMikulas Patocka 	 * so it needs a separate buffer.
12161578dcdSMikulas Patocka 	 */
12261578dcdSMikulas Patocka 	void *header_area;
12361578dcdSMikulas Patocka 
12461578dcdSMikulas Patocka 	/*
1254db6bfe0SAlasdair G Kergon 	 * Used to keep track of which metadata area the data in
1264db6bfe0SAlasdair G Kergon 	 * 'chunk' refers to.
1274db6bfe0SAlasdair G Kergon 	 */
1284db6bfe0SAlasdair G Kergon 	chunk_t current_area;
1294db6bfe0SAlasdair G Kergon 
1304db6bfe0SAlasdair G Kergon 	/*
1314db6bfe0SAlasdair G Kergon 	 * The next free chunk for an exception.
1324454a621SMikulas Patocka 	 *
1334454a621SMikulas Patocka 	 * When creating exceptions, all the chunks here and above are
1344454a621SMikulas Patocka 	 * free.  It holds the next chunk to be allocated.  On rare
1354454a621SMikulas Patocka 	 * occasions (e.g. after a system crash) holes can be left in
1364454a621SMikulas Patocka 	 * the exception store because chunks can be committed out of
1374454a621SMikulas Patocka 	 * order.
1384454a621SMikulas Patocka 	 *
1394454a621SMikulas Patocka 	 * When merging exceptions, it does not necessarily mean all the
1404454a621SMikulas Patocka 	 * chunks here and above are free.  It holds the value it would
1414454a621SMikulas Patocka 	 * have held if all chunks had been committed in order of
1424454a621SMikulas Patocka 	 * allocation.  Consequently the value may occasionally be
1434454a621SMikulas Patocka 	 * slightly too low, but since it's only used for 'status' and
1444454a621SMikulas Patocka 	 * it can never reach its minimum value too early this doesn't
1454454a621SMikulas Patocka 	 * matter.
1464db6bfe0SAlasdair G Kergon 	 */
1474454a621SMikulas Patocka 
1484db6bfe0SAlasdair G Kergon 	chunk_t next_free;
1494db6bfe0SAlasdair G Kergon 
1504db6bfe0SAlasdair G Kergon 	/*
1514db6bfe0SAlasdair G Kergon 	 * The index of next free exception in the current
1524db6bfe0SAlasdair G Kergon 	 * metadata area.
1534db6bfe0SAlasdair G Kergon 	 */
1544db6bfe0SAlasdair G Kergon 	uint32_t current_committed;
1554db6bfe0SAlasdair G Kergon 
1564db6bfe0SAlasdair G Kergon 	atomic_t pending_count;
1574db6bfe0SAlasdair G Kergon 	uint32_t callback_count;
1584db6bfe0SAlasdair G Kergon 	struct commit_callback *callbacks;
1594db6bfe0SAlasdair G Kergon 	struct dm_io_client *io_client;
1604db6bfe0SAlasdair G Kergon 
1614db6bfe0SAlasdair G Kergon 	struct workqueue_struct *metadata_wq;
1624db6bfe0SAlasdair G Kergon };
1634db6bfe0SAlasdair G Kergon 
1644db6bfe0SAlasdair G Kergon static int alloc_area(struct pstore *ps)
1654db6bfe0SAlasdair G Kergon {
1664db6bfe0SAlasdair G Kergon 	int r = -ENOMEM;
1674db6bfe0SAlasdair G Kergon 	size_t len;
1684db6bfe0SAlasdair G Kergon 
16971fab00aSJonathan Brassow 	len = ps->store->chunk_size << SECTOR_SHIFT;
1704db6bfe0SAlasdair G Kergon 
1714db6bfe0SAlasdair G Kergon 	/*
1724db6bfe0SAlasdair G Kergon 	 * Allocate the chunk_size block of memory that will hold
1734db6bfe0SAlasdair G Kergon 	 * a single metadata area.
1744db6bfe0SAlasdair G Kergon 	 */
1754db6bfe0SAlasdair G Kergon 	ps->area = vmalloc(len);
1764db6bfe0SAlasdair G Kergon 	if (!ps->area)
17761578dcdSMikulas Patocka 		goto err_area;
1784db6bfe0SAlasdair G Kergon 
179e29e65aaSJoe Perches 	ps->zero_area = vzalloc(len);
18061578dcdSMikulas Patocka 	if (!ps->zero_area)
18161578dcdSMikulas Patocka 		goto err_zero_area;
1824db6bfe0SAlasdair G Kergon 
18361578dcdSMikulas Patocka 	ps->header_area = vmalloc(len);
18461578dcdSMikulas Patocka 	if (!ps->header_area)
18561578dcdSMikulas Patocka 		goto err_header_area;
18661578dcdSMikulas Patocka 
1874db6bfe0SAlasdair G Kergon 	return 0;
18861578dcdSMikulas Patocka 
18961578dcdSMikulas Patocka err_header_area:
19061578dcdSMikulas Patocka 	vfree(ps->zero_area);
19161578dcdSMikulas Patocka 
19261578dcdSMikulas Patocka err_zero_area:
19361578dcdSMikulas Patocka 	vfree(ps->area);
19461578dcdSMikulas Patocka 
19561578dcdSMikulas Patocka err_area:
19661578dcdSMikulas Patocka 	return r;
1974db6bfe0SAlasdair G Kergon }
1984db6bfe0SAlasdair G Kergon 
1994db6bfe0SAlasdair G Kergon static void free_area(struct pstore *ps)
2004db6bfe0SAlasdair G Kergon {
201a32079ceSJonathan Brassow 	if (ps->area)
2024db6bfe0SAlasdair G Kergon 		vfree(ps->area);
2034db6bfe0SAlasdair G Kergon 	ps->area = NULL;
204a32079ceSJonathan Brassow 
205a32079ceSJonathan Brassow 	if (ps->zero_area)
2064db6bfe0SAlasdair G Kergon 		vfree(ps->zero_area);
2074db6bfe0SAlasdair G Kergon 	ps->zero_area = NULL;
20861578dcdSMikulas Patocka 
20961578dcdSMikulas Patocka 	if (ps->header_area)
21061578dcdSMikulas Patocka 		vfree(ps->header_area);
21161578dcdSMikulas Patocka 	ps->header_area = NULL;
2124db6bfe0SAlasdair G Kergon }
2134db6bfe0SAlasdair G Kergon 
2144db6bfe0SAlasdair G Kergon struct mdata_req {
2154db6bfe0SAlasdair G Kergon 	struct dm_io_region *where;
2164db6bfe0SAlasdair G Kergon 	struct dm_io_request *io_req;
2174db6bfe0SAlasdair G Kergon 	struct work_struct work;
2184db6bfe0SAlasdair G Kergon 	int result;
2194db6bfe0SAlasdair G Kergon };
2204db6bfe0SAlasdair G Kergon 
2214db6bfe0SAlasdair G Kergon static void do_metadata(struct work_struct *work)
2224db6bfe0SAlasdair G Kergon {
2234db6bfe0SAlasdair G Kergon 	struct mdata_req *req = container_of(work, struct mdata_req, work);
2244db6bfe0SAlasdair G Kergon 
2254db6bfe0SAlasdair G Kergon 	req->result = dm_io(req->io_req, 1, req->where, NULL);
2264db6bfe0SAlasdair G Kergon }
2274db6bfe0SAlasdair G Kergon 
2284db6bfe0SAlasdair G Kergon /*
2294db6bfe0SAlasdair G Kergon  * Read or write a chunk aligned and sized block of data from a device.
2304db6bfe0SAlasdair G Kergon  */
23102d2fd31SMikulas Patocka static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw,
23202d2fd31SMikulas Patocka 		    int metadata)
2334db6bfe0SAlasdair G Kergon {
2344db6bfe0SAlasdair G Kergon 	struct dm_io_region where = {
235fc56f6fbSMike Snitzer 		.bdev = dm_snap_cow(ps->store->snap)->bdev,
23671fab00aSJonathan Brassow 		.sector = ps->store->chunk_size * chunk,
23771fab00aSJonathan Brassow 		.count = ps->store->chunk_size,
2384db6bfe0SAlasdair G Kergon 	};
2394db6bfe0SAlasdair G Kergon 	struct dm_io_request io_req = {
2404db6bfe0SAlasdair G Kergon 		.bi_rw = rw,
2414db6bfe0SAlasdair G Kergon 		.mem.type = DM_IO_VMA,
24202d2fd31SMikulas Patocka 		.mem.ptr.vma = area,
2434db6bfe0SAlasdair G Kergon 		.client = ps->io_client,
2444db6bfe0SAlasdair G Kergon 		.notify.fn = NULL,
2454db6bfe0SAlasdair G Kergon 	};
2464db6bfe0SAlasdair G Kergon 	struct mdata_req req;
2474db6bfe0SAlasdair G Kergon 
2484db6bfe0SAlasdair G Kergon 	if (!metadata)
2494db6bfe0SAlasdair G Kergon 		return dm_io(&io_req, 1, &where, NULL);
2504db6bfe0SAlasdair G Kergon 
2514db6bfe0SAlasdair G Kergon 	req.where = &where;
2524db6bfe0SAlasdair G Kergon 	req.io_req = &io_req;
2534db6bfe0SAlasdair G Kergon 
2544db6bfe0SAlasdair G Kergon 	/*
2554db6bfe0SAlasdair G Kergon 	 * Issue the synchronous I/O from a different thread
2564db6bfe0SAlasdair G Kergon 	 * to avoid generic_make_request recursion.
2574db6bfe0SAlasdair G Kergon 	 */
258ca1cab37SAndrew Morton 	INIT_WORK_ONSTACK(&req.work, do_metadata);
2594db6bfe0SAlasdair G Kergon 	queue_work(ps->metadata_wq, &req.work);
2605ea330a7SMikulas Patocka 	flush_workqueue(ps->metadata_wq);
261c1a64160SChuansheng Liu 	destroy_work_on_stack(&req.work);
2624db6bfe0SAlasdair G Kergon 
2634db6bfe0SAlasdair G Kergon 	return req.result;
2644db6bfe0SAlasdair G Kergon }
2654db6bfe0SAlasdair G Kergon 
2664db6bfe0SAlasdair G Kergon /*
2674db6bfe0SAlasdair G Kergon  * Convert a metadata area index to a chunk index.
2684db6bfe0SAlasdair G Kergon  */
2694db6bfe0SAlasdair G Kergon static chunk_t area_location(struct pstore *ps, chunk_t area)
2704db6bfe0SAlasdair G Kergon {
27187c961cbSTomohiro Kusumi 	return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area);
2724db6bfe0SAlasdair G Kergon }
2734db6bfe0SAlasdair G Kergon 
274e9c6a182SMikulas Patocka static void skip_metadata(struct pstore *ps)
275e9c6a182SMikulas Patocka {
276e9c6a182SMikulas Patocka 	uint32_t stride = ps->exceptions_per_area + 1;
277e9c6a182SMikulas Patocka 	chunk_t next_free = ps->next_free;
278e9c6a182SMikulas Patocka 	if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS)
279e9c6a182SMikulas Patocka 		ps->next_free++;
280e9c6a182SMikulas Patocka }
281e9c6a182SMikulas Patocka 
2824db6bfe0SAlasdair G Kergon /*
2834db6bfe0SAlasdair G Kergon  * Read or write a metadata area.  Remembering to skip the first
2844db6bfe0SAlasdair G Kergon  * chunk which holds the header.
2854db6bfe0SAlasdair G Kergon  */
2864db6bfe0SAlasdair G Kergon static int area_io(struct pstore *ps, int rw)
2874db6bfe0SAlasdair G Kergon {
2884db6bfe0SAlasdair G Kergon 	int r;
2894db6bfe0SAlasdair G Kergon 	chunk_t chunk;
2904db6bfe0SAlasdair G Kergon 
2914db6bfe0SAlasdair G Kergon 	chunk = area_location(ps, ps->current_area);
2924db6bfe0SAlasdair G Kergon 
29302d2fd31SMikulas Patocka 	r = chunk_io(ps, ps->area, chunk, rw, 0);
2944db6bfe0SAlasdair G Kergon 	if (r)
2954db6bfe0SAlasdair G Kergon 		return r;
2964db6bfe0SAlasdair G Kergon 
2974db6bfe0SAlasdair G Kergon 	return 0;
2984db6bfe0SAlasdair G Kergon }
2994db6bfe0SAlasdair G Kergon 
3004db6bfe0SAlasdair G Kergon static void zero_memory_area(struct pstore *ps)
3014db6bfe0SAlasdair G Kergon {
30271fab00aSJonathan Brassow 	memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT);
3034db6bfe0SAlasdair G Kergon }
3044db6bfe0SAlasdair G Kergon 
3054db6bfe0SAlasdair G Kergon static int zero_disk_area(struct pstore *ps, chunk_t area)
3064db6bfe0SAlasdair G Kergon {
30702d2fd31SMikulas Patocka 	return chunk_io(ps, ps->zero_area, area_location(ps, area), WRITE, 0);
3084db6bfe0SAlasdair G Kergon }
3094db6bfe0SAlasdair G Kergon 
3104db6bfe0SAlasdair G Kergon static int read_header(struct pstore *ps, int *new_snapshot)
3114db6bfe0SAlasdair G Kergon {
3124db6bfe0SAlasdair G Kergon 	int r;
3134db6bfe0SAlasdair G Kergon 	struct disk_header *dh;
314df96eee6SMikulas Patocka 	unsigned chunk_size;
3154db6bfe0SAlasdair G Kergon 	int chunk_size_supplied = 1;
316ae0b7448SMikulas Patocka 	char *chunk_err;
3174db6bfe0SAlasdair G Kergon 
3184db6bfe0SAlasdair G Kergon 	/*
319df96eee6SMikulas Patocka 	 * Use default chunk size (or logical_block_size, if larger)
320df96eee6SMikulas Patocka 	 * if none supplied
3214db6bfe0SAlasdair G Kergon 	 */
32271fab00aSJonathan Brassow 	if (!ps->store->chunk_size) {
32371fab00aSJonathan Brassow 		ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
324fc56f6fbSMike Snitzer 		    bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
325fc56f6fbSMike Snitzer 					    bdev) >> 9);
32671fab00aSJonathan Brassow 		ps->store->chunk_mask = ps->store->chunk_size - 1;
32771fab00aSJonathan Brassow 		ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1;
3284db6bfe0SAlasdair G Kergon 		chunk_size_supplied = 0;
3294db6bfe0SAlasdair G Kergon 	}
3304db6bfe0SAlasdair G Kergon 
331bda8efecSMikulas Patocka 	ps->io_client = dm_io_client_create();
3324db6bfe0SAlasdair G Kergon 	if (IS_ERR(ps->io_client))
3334db6bfe0SAlasdair G Kergon 		return PTR_ERR(ps->io_client);
3344db6bfe0SAlasdair G Kergon 
3354db6bfe0SAlasdair G Kergon 	r = alloc_area(ps);
3364db6bfe0SAlasdair G Kergon 	if (r)
3374db6bfe0SAlasdair G Kergon 		return r;
3384db6bfe0SAlasdair G Kergon 
33961578dcdSMikulas Patocka 	r = chunk_io(ps, ps->header_area, 0, READ, 1);
3404db6bfe0SAlasdair G Kergon 	if (r)
3414db6bfe0SAlasdair G Kergon 		goto bad;
3424db6bfe0SAlasdair G Kergon 
34361578dcdSMikulas Patocka 	dh = ps->header_area;
3444db6bfe0SAlasdair G Kergon 
3454db6bfe0SAlasdair G Kergon 	if (le32_to_cpu(dh->magic) == 0) {
3464db6bfe0SAlasdair G Kergon 		*new_snapshot = 1;
3474db6bfe0SAlasdair G Kergon 		return 0;
3484db6bfe0SAlasdair G Kergon 	}
3494db6bfe0SAlasdair G Kergon 
3504db6bfe0SAlasdair G Kergon 	if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
3514db6bfe0SAlasdair G Kergon 		DMWARN("Invalid or corrupt snapshot");
3524db6bfe0SAlasdair G Kergon 		r = -ENXIO;
3534db6bfe0SAlasdair G Kergon 		goto bad;
3544db6bfe0SAlasdair G Kergon 	}
3554db6bfe0SAlasdair G Kergon 
3564db6bfe0SAlasdair G Kergon 	*new_snapshot = 0;
3574db6bfe0SAlasdair G Kergon 	ps->valid = le32_to_cpu(dh->valid);
3584db6bfe0SAlasdair G Kergon 	ps->version = le32_to_cpu(dh->version);
3594db6bfe0SAlasdair G Kergon 	chunk_size = le32_to_cpu(dh->chunk_size);
3604db6bfe0SAlasdair G Kergon 
361ae0b7448SMikulas Patocka 	if (ps->store->chunk_size == chunk_size)
3624db6bfe0SAlasdair G Kergon 		return 0;
3634db6bfe0SAlasdair G Kergon 
364ae0b7448SMikulas Patocka 	if (chunk_size_supplied)
365df96eee6SMikulas Patocka 		DMWARN("chunk size %u in device metadata overrides "
366df96eee6SMikulas Patocka 		       "table chunk size of %u.",
367df96eee6SMikulas Patocka 		       chunk_size, ps->store->chunk_size);
3684db6bfe0SAlasdair G Kergon 
3694db6bfe0SAlasdair G Kergon 	/* We had a bogus chunk_size. Fix stuff up. */
3704db6bfe0SAlasdair G Kergon 	free_area(ps);
3714db6bfe0SAlasdair G Kergon 
372ae0b7448SMikulas Patocka 	r = dm_exception_store_set_chunk_size(ps->store, chunk_size,
373ae0b7448SMikulas Patocka 					      &chunk_err);
374ae0b7448SMikulas Patocka 	if (r) {
375df96eee6SMikulas Patocka 		DMERR("invalid on-disk chunk size %u: %s.",
376df96eee6SMikulas Patocka 		      chunk_size, chunk_err);
377ae0b7448SMikulas Patocka 		return r;
378ae0b7448SMikulas Patocka 	}
3794db6bfe0SAlasdair G Kergon 
3804db6bfe0SAlasdair G Kergon 	r = alloc_area(ps);
3814db6bfe0SAlasdair G Kergon 	return r;
3824db6bfe0SAlasdair G Kergon 
3834db6bfe0SAlasdair G Kergon bad:
3844db6bfe0SAlasdair G Kergon 	free_area(ps);
3854db6bfe0SAlasdair G Kergon 	return r;
3864db6bfe0SAlasdair G Kergon }
3874db6bfe0SAlasdair G Kergon 
3884db6bfe0SAlasdair G Kergon static int write_header(struct pstore *ps)
3894db6bfe0SAlasdair G Kergon {
3904db6bfe0SAlasdair G Kergon 	struct disk_header *dh;
3914db6bfe0SAlasdair G Kergon 
39261578dcdSMikulas Patocka 	memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT);
3934db6bfe0SAlasdair G Kergon 
39461578dcdSMikulas Patocka 	dh = ps->header_area;
3954db6bfe0SAlasdair G Kergon 	dh->magic = cpu_to_le32(SNAP_MAGIC);
3964db6bfe0SAlasdair G Kergon 	dh->valid = cpu_to_le32(ps->valid);
3974db6bfe0SAlasdair G Kergon 	dh->version = cpu_to_le32(ps->version);
39871fab00aSJonathan Brassow 	dh->chunk_size = cpu_to_le32(ps->store->chunk_size);
3994db6bfe0SAlasdair G Kergon 
40061578dcdSMikulas Patocka 	return chunk_io(ps, ps->header_area, 0, WRITE, 1);
4014db6bfe0SAlasdair G Kergon }
4024db6bfe0SAlasdair G Kergon 
4034db6bfe0SAlasdair G Kergon /*
4044db6bfe0SAlasdair G Kergon  * Access functions for the disk exceptions, these do the endian conversions.
4054db6bfe0SAlasdair G Kergon  */
4062cadabd5SMikulas Patocka static struct disk_exception *get_exception(struct pstore *ps, void *ps_area,
4072cadabd5SMikulas Patocka 					    uint32_t index)
4084db6bfe0SAlasdair G Kergon {
4094db6bfe0SAlasdair G Kergon 	BUG_ON(index >= ps->exceptions_per_area);
4104db6bfe0SAlasdair G Kergon 
4112cadabd5SMikulas Patocka 	return ((struct disk_exception *) ps_area) + index;
4124db6bfe0SAlasdair G Kergon }
4134db6bfe0SAlasdair G Kergon 
4142cadabd5SMikulas Patocka static void read_exception(struct pstore *ps, void *ps_area,
415283a8328SAlasdair G Kergon 			   uint32_t index, struct core_exception *result)
4164db6bfe0SAlasdair G Kergon {
4172cadabd5SMikulas Patocka 	struct disk_exception *de = get_exception(ps, ps_area, index);
4184db6bfe0SAlasdair G Kergon 
4194db6bfe0SAlasdair G Kergon 	/* copy it */
420283a8328SAlasdair G Kergon 	result->old_chunk = le64_to_cpu(de->old_chunk);
421283a8328SAlasdair G Kergon 	result->new_chunk = le64_to_cpu(de->new_chunk);
4224db6bfe0SAlasdair G Kergon }
4234db6bfe0SAlasdair G Kergon 
4244db6bfe0SAlasdair G Kergon static void write_exception(struct pstore *ps,
425283a8328SAlasdair G Kergon 			    uint32_t index, struct core_exception *e)
4264db6bfe0SAlasdair G Kergon {
4272cadabd5SMikulas Patocka 	struct disk_exception *de = get_exception(ps, ps->area, index);
4284db6bfe0SAlasdair G Kergon 
4294db6bfe0SAlasdair G Kergon 	/* copy it */
430283a8328SAlasdair G Kergon 	de->old_chunk = cpu_to_le64(e->old_chunk);
431283a8328SAlasdair G Kergon 	de->new_chunk = cpu_to_le64(e->new_chunk);
4324db6bfe0SAlasdair G Kergon }
4334db6bfe0SAlasdair G Kergon 
4344454a621SMikulas Patocka static void clear_exception(struct pstore *ps, uint32_t index)
4354454a621SMikulas Patocka {
4362cadabd5SMikulas Patocka 	struct disk_exception *de = get_exception(ps, ps->area, index);
4374454a621SMikulas Patocka 
4384454a621SMikulas Patocka 	/* clear it */
439283a8328SAlasdair G Kergon 	de->old_chunk = 0;
440283a8328SAlasdair G Kergon 	de->new_chunk = 0;
4414454a621SMikulas Patocka }
4424454a621SMikulas Patocka 
4434db6bfe0SAlasdair G Kergon /*
4444db6bfe0SAlasdair G Kergon  * Registers the exceptions that are present in the current area.
4454db6bfe0SAlasdair G Kergon  * 'full' is filled in to indicate if the area has been
4464db6bfe0SAlasdair G Kergon  * filled.
4474db6bfe0SAlasdair G Kergon  */
4482cadabd5SMikulas Patocka static int insert_exceptions(struct pstore *ps, void *ps_area,
449a159c1acSJonathan Brassow 			     int (*callback)(void *callback_context,
450a159c1acSJonathan Brassow 					     chunk_t old, chunk_t new),
451a159c1acSJonathan Brassow 			     void *callback_context,
452a159c1acSJonathan Brassow 			     int *full)
4534db6bfe0SAlasdair G Kergon {
4544db6bfe0SAlasdair G Kergon 	int r;
4554db6bfe0SAlasdair G Kergon 	unsigned int i;
456283a8328SAlasdair G Kergon 	struct core_exception e;
4574db6bfe0SAlasdair G Kergon 
4584db6bfe0SAlasdair G Kergon 	/* presume the area is full */
4594db6bfe0SAlasdair G Kergon 	*full = 1;
4604db6bfe0SAlasdair G Kergon 
4614db6bfe0SAlasdair G Kergon 	for (i = 0; i < ps->exceptions_per_area; i++) {
4622cadabd5SMikulas Patocka 		read_exception(ps, ps_area, i, &e);
4634db6bfe0SAlasdair G Kergon 
4644db6bfe0SAlasdair G Kergon 		/*
4654db6bfe0SAlasdair G Kergon 		 * If the new_chunk is pointing at the start of
4664db6bfe0SAlasdair G Kergon 		 * the COW device, where the first metadata area
4674db6bfe0SAlasdair G Kergon 		 * is we know that we've hit the end of the
4684db6bfe0SAlasdair G Kergon 		 * exceptions.  Therefore the area is not full.
4694db6bfe0SAlasdair G Kergon 		 */
470283a8328SAlasdair G Kergon 		if (e.new_chunk == 0LL) {
4714db6bfe0SAlasdair G Kergon 			ps->current_committed = i;
4724db6bfe0SAlasdair G Kergon 			*full = 0;
4734db6bfe0SAlasdair G Kergon 			break;
4744db6bfe0SAlasdair G Kergon 		}
4754db6bfe0SAlasdair G Kergon 
4764db6bfe0SAlasdair G Kergon 		/*
4774db6bfe0SAlasdair G Kergon 		 * Keep track of the start of the free chunks.
4784db6bfe0SAlasdair G Kergon 		 */
479283a8328SAlasdair G Kergon 		if (ps->next_free <= e.new_chunk)
480283a8328SAlasdair G Kergon 			ps->next_free = e.new_chunk + 1;
4814db6bfe0SAlasdair G Kergon 
4824db6bfe0SAlasdair G Kergon 		/*
4834db6bfe0SAlasdair G Kergon 		 * Otherwise we add the exception to the snapshot.
4844db6bfe0SAlasdair G Kergon 		 */
485283a8328SAlasdair G Kergon 		r = callback(callback_context, e.old_chunk, e.new_chunk);
4864db6bfe0SAlasdair G Kergon 		if (r)
4874db6bfe0SAlasdair G Kergon 			return r;
4884db6bfe0SAlasdair G Kergon 	}
4894db6bfe0SAlasdair G Kergon 
4904db6bfe0SAlasdair G Kergon 	return 0;
4914db6bfe0SAlasdair G Kergon }
4924db6bfe0SAlasdair G Kergon 
493a159c1acSJonathan Brassow static int read_exceptions(struct pstore *ps,
494a159c1acSJonathan Brassow 			   int (*callback)(void *callback_context, chunk_t old,
495a159c1acSJonathan Brassow 					   chunk_t new),
496a159c1acSJonathan Brassow 			   void *callback_context)
4974db6bfe0SAlasdair G Kergon {
4984db6bfe0SAlasdair G Kergon 	int r, full = 1;
499*55494bf2SMikulas Patocka 	struct dm_bufio_client *client;
500*55494bf2SMikulas Patocka 
501*55494bf2SMikulas Patocka 	client = dm_bufio_client_create(dm_snap_cow(ps->store->snap)->bdev,
502*55494bf2SMikulas Patocka 					ps->store->chunk_size << SECTOR_SHIFT,
503*55494bf2SMikulas Patocka 					1, 0, NULL, NULL);
504*55494bf2SMikulas Patocka 
505*55494bf2SMikulas Patocka 	if (IS_ERR(client))
506*55494bf2SMikulas Patocka 		return PTR_ERR(client);
5074db6bfe0SAlasdair G Kergon 
5084db6bfe0SAlasdair G Kergon 	/*
5094db6bfe0SAlasdair G Kergon 	 * Keeping reading chunks and inserting exceptions until
5104db6bfe0SAlasdair G Kergon 	 * we find a partially full area.
5114db6bfe0SAlasdair G Kergon 	 */
5124db6bfe0SAlasdair G Kergon 	for (ps->current_area = 0; full; ps->current_area++) {
513*55494bf2SMikulas Patocka 		struct dm_buffer *bp;
514*55494bf2SMikulas Patocka 		void *area;
515*55494bf2SMikulas Patocka 		chunk_t chunk = area_location(ps, ps->current_area);
5164db6bfe0SAlasdair G Kergon 
517*55494bf2SMikulas Patocka 		area = dm_bufio_read(client, chunk, &bp);
518*55494bf2SMikulas Patocka 		if (unlikely(IS_ERR(area))) {
519*55494bf2SMikulas Patocka 			r = PTR_ERR(area);
520*55494bf2SMikulas Patocka 			goto ret_destroy_bufio;
521*55494bf2SMikulas Patocka 		}
522*55494bf2SMikulas Patocka 
523*55494bf2SMikulas Patocka 		r = insert_exceptions(ps, area, callback, callback_context,
5242cadabd5SMikulas Patocka 				      &full);
525*55494bf2SMikulas Patocka 
526*55494bf2SMikulas Patocka 		dm_bufio_release(bp);
527*55494bf2SMikulas Patocka 
528*55494bf2SMikulas Patocka 		dm_bufio_forget(client, chunk);
529*55494bf2SMikulas Patocka 
530*55494bf2SMikulas Patocka 		if (unlikely(r))
531*55494bf2SMikulas Patocka 			goto ret_destroy_bufio;
5324db6bfe0SAlasdair G Kergon 	}
5334db6bfe0SAlasdair G Kergon 
5344db6bfe0SAlasdair G Kergon 	ps->current_area--;
5354db6bfe0SAlasdair G Kergon 
536e9c6a182SMikulas Patocka 	skip_metadata(ps);
537e9c6a182SMikulas Patocka 
538*55494bf2SMikulas Patocka 	r = 0;
539*55494bf2SMikulas Patocka 
540*55494bf2SMikulas Patocka ret_destroy_bufio:
541*55494bf2SMikulas Patocka 	dm_bufio_client_destroy(client);
542*55494bf2SMikulas Patocka 
543*55494bf2SMikulas Patocka 	return r;
5444db6bfe0SAlasdair G Kergon }
5454db6bfe0SAlasdair G Kergon 
5464db6bfe0SAlasdair G Kergon static struct pstore *get_info(struct dm_exception_store *store)
5474db6bfe0SAlasdair G Kergon {
5484db6bfe0SAlasdair G Kergon 	return (struct pstore *) store->context;
5494db6bfe0SAlasdair G Kergon }
5504db6bfe0SAlasdair G Kergon 
551985903bbSMike Snitzer static void persistent_usage(struct dm_exception_store *store,
552985903bbSMike Snitzer 			     sector_t *total_sectors,
553985903bbSMike Snitzer 			     sector_t *sectors_allocated,
554985903bbSMike Snitzer 			     sector_t *metadata_sectors)
5554db6bfe0SAlasdair G Kergon {
556985903bbSMike Snitzer 	struct pstore *ps = get_info(store);
557985903bbSMike Snitzer 
558985903bbSMike Snitzer 	*sectors_allocated = ps->next_free * store->chunk_size;
559fc56f6fbSMike Snitzer 	*total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
560985903bbSMike Snitzer 
561985903bbSMike Snitzer 	/*
562985903bbSMike Snitzer 	 * First chunk is the fixed header.
563985903bbSMike Snitzer 	 * Then there are (ps->current_area + 1) metadata chunks, each one
564985903bbSMike Snitzer 	 * separated from the next by ps->exceptions_per_area data chunks.
565985903bbSMike Snitzer 	 */
5664454a621SMikulas Patocka 	*metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
5674454a621SMikulas Patocka 			    store->chunk_size;
5684db6bfe0SAlasdair G Kergon }
5694db6bfe0SAlasdair G Kergon 
570493df71cSJonathan Brassow static void persistent_dtr(struct dm_exception_store *store)
5714db6bfe0SAlasdair G Kergon {
5724db6bfe0SAlasdair G Kergon 	struct pstore *ps = get_info(store);
5734db6bfe0SAlasdair G Kergon 
5744db6bfe0SAlasdair G Kergon 	destroy_workqueue(ps->metadata_wq);
575a32079ceSJonathan Brassow 
576a32079ceSJonathan Brassow 	/* Created in read_header */
577a32079ceSJonathan Brassow 	if (ps->io_client)
5784db6bfe0SAlasdair G Kergon 		dm_io_client_destroy(ps->io_client);
5794db6bfe0SAlasdair G Kergon 	free_area(ps);
580a32079ceSJonathan Brassow 
581a32079ceSJonathan Brassow 	/* Allocated in persistent_read_metadata */
582a32079ceSJonathan Brassow 	if (ps->callbacks)
583a32079ceSJonathan Brassow 		vfree(ps->callbacks);
584a32079ceSJonathan Brassow 
5854db6bfe0SAlasdair G Kergon 	kfree(ps);
5864db6bfe0SAlasdair G Kergon }
5874db6bfe0SAlasdair G Kergon 
588a159c1acSJonathan Brassow static int persistent_read_metadata(struct dm_exception_store *store,
589a159c1acSJonathan Brassow 				    int (*callback)(void *callback_context,
590a159c1acSJonathan Brassow 						    chunk_t old, chunk_t new),
591a159c1acSJonathan Brassow 				    void *callback_context)
5924db6bfe0SAlasdair G Kergon {
5934db6bfe0SAlasdair G Kergon 	int r, uninitialized_var(new_snapshot);
5944db6bfe0SAlasdair G Kergon 	struct pstore *ps = get_info(store);
5954db6bfe0SAlasdair G Kergon 
5964db6bfe0SAlasdair G Kergon 	/*
5974db6bfe0SAlasdair G Kergon 	 * Read the snapshot header.
5984db6bfe0SAlasdair G Kergon 	 */
5994db6bfe0SAlasdair G Kergon 	r = read_header(ps, &new_snapshot);
6004db6bfe0SAlasdair G Kergon 	if (r)
6014db6bfe0SAlasdair G Kergon 		return r;
6024db6bfe0SAlasdair G Kergon 
6034db6bfe0SAlasdair G Kergon 	/*
6044db6bfe0SAlasdair G Kergon 	 * Now we know correct chunk_size, complete the initialisation.
6054db6bfe0SAlasdair G Kergon 	 */
60671fab00aSJonathan Brassow 	ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) /
60771fab00aSJonathan Brassow 				  sizeof(struct disk_exception);
6084db6bfe0SAlasdair G Kergon 	ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
6094db6bfe0SAlasdair G Kergon 				   sizeof(*ps->callbacks));
6104db6bfe0SAlasdair G Kergon 	if (!ps->callbacks)
6114db6bfe0SAlasdair G Kergon 		return -ENOMEM;
6124db6bfe0SAlasdair G Kergon 
6134db6bfe0SAlasdair G Kergon 	/*
6144db6bfe0SAlasdair G Kergon 	 * Do we need to setup a new snapshot ?
6154db6bfe0SAlasdair G Kergon 	 */
6164db6bfe0SAlasdair G Kergon 	if (new_snapshot) {
6174db6bfe0SAlasdair G Kergon 		r = write_header(ps);
6184db6bfe0SAlasdair G Kergon 		if (r) {
6194db6bfe0SAlasdair G Kergon 			DMWARN("write_header failed");
6204db6bfe0SAlasdair G Kergon 			return r;
6214db6bfe0SAlasdair G Kergon 		}
6224db6bfe0SAlasdair G Kergon 
6234db6bfe0SAlasdair G Kergon 		ps->current_area = 0;
6244db6bfe0SAlasdair G Kergon 		zero_memory_area(ps);
6254db6bfe0SAlasdair G Kergon 		r = zero_disk_area(ps, 0);
626f5acc834SJon Brassow 		if (r)
6274db6bfe0SAlasdair G Kergon 			DMWARN("zero_disk_area(0) failed");
6284db6bfe0SAlasdair G Kergon 		return r;
6294db6bfe0SAlasdair G Kergon 	}
6304db6bfe0SAlasdair G Kergon 	/*
6314db6bfe0SAlasdair G Kergon 	 * Sanity checks.
6324db6bfe0SAlasdair G Kergon 	 */
6334db6bfe0SAlasdair G Kergon 	if (ps->version != SNAPSHOT_DISK_VERSION) {
6344db6bfe0SAlasdair G Kergon 		DMWARN("unable to handle snapshot disk version %d",
6354db6bfe0SAlasdair G Kergon 		       ps->version);
6364db6bfe0SAlasdair G Kergon 		return -EINVAL;
6374db6bfe0SAlasdair G Kergon 	}
6384db6bfe0SAlasdair G Kergon 
6394db6bfe0SAlasdair G Kergon 	/*
6404db6bfe0SAlasdair G Kergon 	 * Metadata are valid, but snapshot is invalidated
6414db6bfe0SAlasdair G Kergon 	 */
6424db6bfe0SAlasdair G Kergon 	if (!ps->valid)
6434db6bfe0SAlasdair G Kergon 		return 1;
6444db6bfe0SAlasdair G Kergon 
6454db6bfe0SAlasdair G Kergon 	/*
6464db6bfe0SAlasdair G Kergon 	 * Read the metadata.
6474db6bfe0SAlasdair G Kergon 	 */
648a159c1acSJonathan Brassow 	r = read_exceptions(ps, callback, callback_context);
6494db6bfe0SAlasdair G Kergon 
650f5acc834SJon Brassow 	return r;
6514db6bfe0SAlasdair G Kergon }
6524db6bfe0SAlasdair G Kergon 
653a159c1acSJonathan Brassow static int persistent_prepare_exception(struct dm_exception_store *store,
6541d4989c8SJon Brassow 					struct dm_exception *e)
6554db6bfe0SAlasdair G Kergon {
6564db6bfe0SAlasdair G Kergon 	struct pstore *ps = get_info(store);
657fc56f6fbSMike Snitzer 	sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
6584db6bfe0SAlasdair G Kergon 
6594db6bfe0SAlasdair G Kergon 	/* Is there enough room ? */
660d0216849SJonathan Brassow 	if (size < ((ps->next_free + 1) * store->chunk_size))
6614db6bfe0SAlasdair G Kergon 		return -ENOSPC;
6624db6bfe0SAlasdair G Kergon 
6634db6bfe0SAlasdair G Kergon 	e->new_chunk = ps->next_free;
6644db6bfe0SAlasdair G Kergon 
6654db6bfe0SAlasdair G Kergon 	/*
6664db6bfe0SAlasdair G Kergon 	 * Move onto the next free pending, making sure to take
6674db6bfe0SAlasdair G Kergon 	 * into account the location of the metadata chunks.
6684db6bfe0SAlasdair G Kergon 	 */
6694db6bfe0SAlasdair G Kergon 	ps->next_free++;
670e9c6a182SMikulas Patocka 	skip_metadata(ps);
6714db6bfe0SAlasdair G Kergon 
6724db6bfe0SAlasdair G Kergon 	atomic_inc(&ps->pending_count);
6734db6bfe0SAlasdair G Kergon 	return 0;
6744db6bfe0SAlasdair G Kergon }
6754db6bfe0SAlasdair G Kergon 
676a159c1acSJonathan Brassow static void persistent_commit_exception(struct dm_exception_store *store,
6771d4989c8SJon Brassow 					struct dm_exception *e,
6784db6bfe0SAlasdair G Kergon 					void (*callback) (void *, int success),
6794db6bfe0SAlasdair G Kergon 					void *callback_context)
6804db6bfe0SAlasdair G Kergon {
6814db6bfe0SAlasdair G Kergon 	unsigned int i;
6824db6bfe0SAlasdair G Kergon 	struct pstore *ps = get_info(store);
683283a8328SAlasdair G Kergon 	struct core_exception ce;
6844db6bfe0SAlasdair G Kergon 	struct commit_callback *cb;
6854db6bfe0SAlasdair G Kergon 
686283a8328SAlasdair G Kergon 	ce.old_chunk = e->old_chunk;
687283a8328SAlasdair G Kergon 	ce.new_chunk = e->new_chunk;
688283a8328SAlasdair G Kergon 	write_exception(ps, ps->current_committed++, &ce);
6894db6bfe0SAlasdair G Kergon 
6904db6bfe0SAlasdair G Kergon 	/*
6914db6bfe0SAlasdair G Kergon 	 * Add the callback to the back of the array.  This code
6924db6bfe0SAlasdair G Kergon 	 * is the only place where the callback array is
6934db6bfe0SAlasdair G Kergon 	 * manipulated, and we know that it will never be called
6944db6bfe0SAlasdair G Kergon 	 * multiple times concurrently.
6954db6bfe0SAlasdair G Kergon 	 */
6964db6bfe0SAlasdair G Kergon 	cb = ps->callbacks + ps->callback_count++;
6974db6bfe0SAlasdair G Kergon 	cb->callback = callback;
6984db6bfe0SAlasdair G Kergon 	cb->context = callback_context;
6994db6bfe0SAlasdair G Kergon 
7004db6bfe0SAlasdair G Kergon 	/*
7014db6bfe0SAlasdair G Kergon 	 * If there are exceptions in flight and we have not yet
7024db6bfe0SAlasdair G Kergon 	 * filled this metadata area there's nothing more to do.
7034db6bfe0SAlasdair G Kergon 	 */
7044db6bfe0SAlasdair G Kergon 	if (!atomic_dec_and_test(&ps->pending_count) &&
7054db6bfe0SAlasdair G Kergon 	    (ps->current_committed != ps->exceptions_per_area))
7064db6bfe0SAlasdair G Kergon 		return;
7074db6bfe0SAlasdair G Kergon 
7084db6bfe0SAlasdair G Kergon 	/*
7094db6bfe0SAlasdair G Kergon 	 * If we completely filled the current area, then wipe the next one.
7104db6bfe0SAlasdair G Kergon 	 */
7114db6bfe0SAlasdair G Kergon 	if ((ps->current_committed == ps->exceptions_per_area) &&
7124db6bfe0SAlasdair G Kergon 	    zero_disk_area(ps, ps->current_area + 1))
7134db6bfe0SAlasdair G Kergon 		ps->valid = 0;
7144db6bfe0SAlasdair G Kergon 
7154db6bfe0SAlasdair G Kergon 	/*
7164db6bfe0SAlasdair G Kergon 	 * Commit exceptions to disk.
7174db6bfe0SAlasdair G Kergon 	 */
718d87f4c14STejun Heo 	if (ps->valid && area_io(ps, WRITE_FLUSH_FUA))
7194db6bfe0SAlasdair G Kergon 		ps->valid = 0;
7204db6bfe0SAlasdair G Kergon 
7214db6bfe0SAlasdair G Kergon 	/*
7224db6bfe0SAlasdair G Kergon 	 * Advance to the next area if this one is full.
7234db6bfe0SAlasdair G Kergon 	 */
7244db6bfe0SAlasdair G Kergon 	if (ps->current_committed == ps->exceptions_per_area) {
7254db6bfe0SAlasdair G Kergon 		ps->current_committed = 0;
7264db6bfe0SAlasdair G Kergon 		ps->current_area++;
7274db6bfe0SAlasdair G Kergon 		zero_memory_area(ps);
7284db6bfe0SAlasdair G Kergon 	}
7294db6bfe0SAlasdair G Kergon 
7304db6bfe0SAlasdair G Kergon 	for (i = 0; i < ps->callback_count; i++) {
7314db6bfe0SAlasdair G Kergon 		cb = ps->callbacks + i;
7324db6bfe0SAlasdair G Kergon 		cb->callback(cb->context, ps->valid);
7334db6bfe0SAlasdair G Kergon 	}
7344db6bfe0SAlasdair G Kergon 
7354db6bfe0SAlasdair G Kergon 	ps->callback_count = 0;
7364db6bfe0SAlasdair G Kergon }
7374db6bfe0SAlasdair G Kergon 
7384454a621SMikulas Patocka static int persistent_prepare_merge(struct dm_exception_store *store,
7394454a621SMikulas Patocka 				    chunk_t *last_old_chunk,
7404454a621SMikulas Patocka 				    chunk_t *last_new_chunk)
7414454a621SMikulas Patocka {
7424454a621SMikulas Patocka 	struct pstore *ps = get_info(store);
743283a8328SAlasdair G Kergon 	struct core_exception ce;
7444454a621SMikulas Patocka 	int nr_consecutive;
7454454a621SMikulas Patocka 	int r;
7464454a621SMikulas Patocka 
7474454a621SMikulas Patocka 	/*
7484454a621SMikulas Patocka 	 * When current area is empty, move back to preceding area.
7494454a621SMikulas Patocka 	 */
7504454a621SMikulas Patocka 	if (!ps->current_committed) {
7514454a621SMikulas Patocka 		/*
7524454a621SMikulas Patocka 		 * Have we finished?
7534454a621SMikulas Patocka 		 */
7544454a621SMikulas Patocka 		if (!ps->current_area)
7554454a621SMikulas Patocka 			return 0;
7564454a621SMikulas Patocka 
7574454a621SMikulas Patocka 		ps->current_area--;
7584454a621SMikulas Patocka 		r = area_io(ps, READ);
7594454a621SMikulas Patocka 		if (r < 0)
7604454a621SMikulas Patocka 			return r;
7614454a621SMikulas Patocka 		ps->current_committed = ps->exceptions_per_area;
7624454a621SMikulas Patocka 	}
7634454a621SMikulas Patocka 
7642cadabd5SMikulas Patocka 	read_exception(ps, ps->area, ps->current_committed - 1, &ce);
765283a8328SAlasdair G Kergon 	*last_old_chunk = ce.old_chunk;
766283a8328SAlasdair G Kergon 	*last_new_chunk = ce.new_chunk;
7674454a621SMikulas Patocka 
7684454a621SMikulas Patocka 	/*
7694454a621SMikulas Patocka 	 * Find number of consecutive chunks within the current area,
7704454a621SMikulas Patocka 	 * working backwards.
7714454a621SMikulas Patocka 	 */
7724454a621SMikulas Patocka 	for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
7734454a621SMikulas Patocka 	     nr_consecutive++) {
7742cadabd5SMikulas Patocka 		read_exception(ps, ps->area,
7752cadabd5SMikulas Patocka 			       ps->current_committed - 1 - nr_consecutive, &ce);
776283a8328SAlasdair G Kergon 		if (ce.old_chunk != *last_old_chunk - nr_consecutive ||
777283a8328SAlasdair G Kergon 		    ce.new_chunk != *last_new_chunk - nr_consecutive)
7784454a621SMikulas Patocka 			break;
7794454a621SMikulas Patocka 	}
7804454a621SMikulas Patocka 
7814454a621SMikulas Patocka 	return nr_consecutive;
7824454a621SMikulas Patocka }
7834454a621SMikulas Patocka 
7844454a621SMikulas Patocka static int persistent_commit_merge(struct dm_exception_store *store,
7854454a621SMikulas Patocka 				   int nr_merged)
7864454a621SMikulas Patocka {
7874454a621SMikulas Patocka 	int r, i;
7884454a621SMikulas Patocka 	struct pstore *ps = get_info(store);
7894454a621SMikulas Patocka 
7904454a621SMikulas Patocka 	BUG_ON(nr_merged > ps->current_committed);
7914454a621SMikulas Patocka 
7924454a621SMikulas Patocka 	for (i = 0; i < nr_merged; i++)
7934454a621SMikulas Patocka 		clear_exception(ps, ps->current_committed - 1 - i);
7944454a621SMikulas Patocka 
795762a80d9SMikulas Patocka 	r = area_io(ps, WRITE_FLUSH_FUA);
7964454a621SMikulas Patocka 	if (r < 0)
7974454a621SMikulas Patocka 		return r;
7984454a621SMikulas Patocka 
7994454a621SMikulas Patocka 	ps->current_committed -= nr_merged;
8004454a621SMikulas Patocka 
8014454a621SMikulas Patocka 	/*
8024454a621SMikulas Patocka 	 * At this stage, only persistent_usage() uses ps->next_free, so
8034454a621SMikulas Patocka 	 * we make no attempt to keep ps->next_free strictly accurate
8044454a621SMikulas Patocka 	 * as exceptions may have been committed out-of-order originally.
8054454a621SMikulas Patocka 	 * Once a snapshot has become merging, we set it to the value it
8064454a621SMikulas Patocka 	 * would have held had all the exceptions been committed in order.
8074454a621SMikulas Patocka 	 *
8084454a621SMikulas Patocka 	 * ps->current_area does not get reduced by prepare_merge() until
8094454a621SMikulas Patocka 	 * after commit_merge() has removed the nr_merged previous exceptions.
8104454a621SMikulas Patocka 	 */
81187c961cbSTomohiro Kusumi 	ps->next_free = area_location(ps, ps->current_area) +
81287c961cbSTomohiro Kusumi 			ps->current_committed + 1;
8134454a621SMikulas Patocka 
8144454a621SMikulas Patocka 	return 0;
8154454a621SMikulas Patocka }
8164454a621SMikulas Patocka 
817a159c1acSJonathan Brassow static void persistent_drop_snapshot(struct dm_exception_store *store)
8184db6bfe0SAlasdair G Kergon {
8194db6bfe0SAlasdair G Kergon 	struct pstore *ps = get_info(store);
8204db6bfe0SAlasdair G Kergon 
8214db6bfe0SAlasdair G Kergon 	ps->valid = 0;
8224db6bfe0SAlasdair G Kergon 	if (write_header(ps))
8234db6bfe0SAlasdair G Kergon 		DMWARN("write header failed");
8244db6bfe0SAlasdair G Kergon }
8254db6bfe0SAlasdair G Kergon 
826493df71cSJonathan Brassow static int persistent_ctr(struct dm_exception_store *store,
827493df71cSJonathan Brassow 			  unsigned argc, char **argv)
8284db6bfe0SAlasdair G Kergon {
8294db6bfe0SAlasdair G Kergon 	struct pstore *ps;
8304db6bfe0SAlasdair G Kergon 
8314db6bfe0SAlasdair G Kergon 	/* allocate the pstore */
832a32079ceSJonathan Brassow 	ps = kzalloc(sizeof(*ps), GFP_KERNEL);
8334db6bfe0SAlasdair G Kergon 	if (!ps)
8344db6bfe0SAlasdair G Kergon 		return -ENOMEM;
8354db6bfe0SAlasdair G Kergon 
83671fab00aSJonathan Brassow 	ps->store = store;
8374db6bfe0SAlasdair G Kergon 	ps->valid = 1;
8384db6bfe0SAlasdair G Kergon 	ps->version = SNAPSHOT_DISK_VERSION;
8394db6bfe0SAlasdair G Kergon 	ps->area = NULL;
84061578dcdSMikulas Patocka 	ps->zero_area = NULL;
84161578dcdSMikulas Patocka 	ps->header_area = NULL;
8424454a621SMikulas Patocka 	ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
8434db6bfe0SAlasdair G Kergon 	ps->current_committed = 0;
8444db6bfe0SAlasdair G Kergon 
8454db6bfe0SAlasdair G Kergon 	ps->callback_count = 0;
8464db6bfe0SAlasdair G Kergon 	atomic_set(&ps->pending_count, 0);
8474db6bfe0SAlasdair G Kergon 	ps->callbacks = NULL;
8484db6bfe0SAlasdair G Kergon 
849239c8dd5STejun Heo 	ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0);
8504db6bfe0SAlasdair G Kergon 	if (!ps->metadata_wq) {
8514db6bfe0SAlasdair G Kergon 		kfree(ps);
8524db6bfe0SAlasdair G Kergon 		DMERR("couldn't start header metadata update thread");
8534db6bfe0SAlasdair G Kergon 		return -ENOMEM;
8544db6bfe0SAlasdair G Kergon 	}
8554db6bfe0SAlasdair G Kergon 
8564db6bfe0SAlasdair G Kergon 	store->context = ps;
8574db6bfe0SAlasdair G Kergon 
8584db6bfe0SAlasdair G Kergon 	return 0;
8594db6bfe0SAlasdair G Kergon }
8604db6bfe0SAlasdair G Kergon 
8611e302a92SJonathan Brassow static unsigned persistent_status(struct dm_exception_store *store,
862493df71cSJonathan Brassow 				  status_type_t status, char *result,
8631e302a92SJonathan Brassow 				  unsigned maxlen)
864493df71cSJonathan Brassow {
8651e302a92SJonathan Brassow 	unsigned sz = 0;
8661e302a92SJonathan Brassow 
8671e302a92SJonathan Brassow 	switch (status) {
8681e302a92SJonathan Brassow 	case STATUSTYPE_INFO:
8691e302a92SJonathan Brassow 		break;
8701e302a92SJonathan Brassow 	case STATUSTYPE_TABLE:
871fc56f6fbSMike Snitzer 		DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
8721e302a92SJonathan Brassow 	}
873493df71cSJonathan Brassow 
874493df71cSJonathan Brassow 	return sz;
875493df71cSJonathan Brassow }
876493df71cSJonathan Brassow 
877493df71cSJonathan Brassow static struct dm_exception_store_type _persistent_type = {
878493df71cSJonathan Brassow 	.name = "persistent",
879493df71cSJonathan Brassow 	.module = THIS_MODULE,
880493df71cSJonathan Brassow 	.ctr = persistent_ctr,
881493df71cSJonathan Brassow 	.dtr = persistent_dtr,
882493df71cSJonathan Brassow 	.read_metadata = persistent_read_metadata,
883493df71cSJonathan Brassow 	.prepare_exception = persistent_prepare_exception,
884493df71cSJonathan Brassow 	.commit_exception = persistent_commit_exception,
8854454a621SMikulas Patocka 	.prepare_merge = persistent_prepare_merge,
8864454a621SMikulas Patocka 	.commit_merge = persistent_commit_merge,
887493df71cSJonathan Brassow 	.drop_snapshot = persistent_drop_snapshot,
888985903bbSMike Snitzer 	.usage = persistent_usage,
889493df71cSJonathan Brassow 	.status = persistent_status,
890493df71cSJonathan Brassow };
891493df71cSJonathan Brassow 
892493df71cSJonathan Brassow static struct dm_exception_store_type _persistent_compat_type = {
893493df71cSJonathan Brassow 	.name = "P",
894493df71cSJonathan Brassow 	.module = THIS_MODULE,
895493df71cSJonathan Brassow 	.ctr = persistent_ctr,
896493df71cSJonathan Brassow 	.dtr = persistent_dtr,
897493df71cSJonathan Brassow 	.read_metadata = persistent_read_metadata,
898493df71cSJonathan Brassow 	.prepare_exception = persistent_prepare_exception,
899493df71cSJonathan Brassow 	.commit_exception = persistent_commit_exception,
9004454a621SMikulas Patocka 	.prepare_merge = persistent_prepare_merge,
9014454a621SMikulas Patocka 	.commit_merge = persistent_commit_merge,
902493df71cSJonathan Brassow 	.drop_snapshot = persistent_drop_snapshot,
903985903bbSMike Snitzer 	.usage = persistent_usage,
904493df71cSJonathan Brassow 	.status = persistent_status,
905493df71cSJonathan Brassow };
906493df71cSJonathan Brassow 
9074db6bfe0SAlasdair G Kergon int dm_persistent_snapshot_init(void)
9084db6bfe0SAlasdair G Kergon {
909493df71cSJonathan Brassow 	int r;
910493df71cSJonathan Brassow 
911493df71cSJonathan Brassow 	r = dm_exception_store_type_register(&_persistent_type);
912493df71cSJonathan Brassow 	if (r) {
913493df71cSJonathan Brassow 		DMERR("Unable to register persistent exception store type");
914493df71cSJonathan Brassow 		return r;
915493df71cSJonathan Brassow 	}
916493df71cSJonathan Brassow 
917493df71cSJonathan Brassow 	r = dm_exception_store_type_register(&_persistent_compat_type);
918493df71cSJonathan Brassow 	if (r) {
919493df71cSJonathan Brassow 		DMERR("Unable to register old-style persistent exception "
920493df71cSJonathan Brassow 		      "store type");
921493df71cSJonathan Brassow 		dm_exception_store_type_unregister(&_persistent_type);
922493df71cSJonathan Brassow 		return r;
923493df71cSJonathan Brassow 	}
924493df71cSJonathan Brassow 
925493df71cSJonathan Brassow 	return r;
9264db6bfe0SAlasdair G Kergon }
9274db6bfe0SAlasdair G Kergon 
9284db6bfe0SAlasdair G Kergon void dm_persistent_snapshot_exit(void)
9294db6bfe0SAlasdair G Kergon {
930493df71cSJonathan Brassow 	dm_exception_store_type_unregister(&_persistent_type);
931493df71cSJonathan Brassow 	dm_exception_store_type_unregister(&_persistent_compat_type);
9324db6bfe0SAlasdair G Kergon }
933