Merge branch 'rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip...
[sfrench/cifs-2.6.git] / drivers / md / dm-exception-store.c
index 01590f3e00093ffd25e9fe0aa6e70d1472e607f4..a2e26c24214150cac23e92acf1391a212d60271b 100644 (file)
 /*
- * dm-exception-store.c
- *
  * Copyright (C) 2001-2002 Sistina Software (UK) Limited.
- * Copyright (C) 2006 Red Hat GmbH
+ * Copyright (C) 2006-2008 Red Hat GmbH
  *
  * This file is released under the GPL.
  */
 
-#include "dm-snap.h"
+#include "dm-exception-store.h"
 
+#include <linux/ctype.h>
 #include <linux/mm.h>
 #include <linux/pagemap.h>
 #include <linux/vmalloc.h>
 #include <linux/slab.h>
-#include <linux/dm-io.h>
-#include <linux/dm-kcopyd.h>
-
-#define DM_MSG_PREFIX "snapshots"
-#define DM_CHUNK_SIZE_DEFAULT_SECTORS 32       /* 16KB */
-
-/*-----------------------------------------------------------------
- * Persistent snapshots, by persistent we mean that the snapshot
- * will survive a reboot.
- *---------------------------------------------------------------*/
-
-/*
- * We need to store a record of which parts of the origin have
- * been copied to the snapshot device.  The snapshot code
- * requires that we copy exception chunks to chunk aligned areas
- * of the COW store.  It makes sense therefore, to store the
- * metadata in chunk size blocks.
- *
- * There is no backward or forward compatibility implemented,
- * snapshots with different disk versions than the kernel will
- * not be usable.  It is expected that "lvcreate" will blank out
- * the start of a fresh COW device before calling the snapshot
- * constructor.
- *
- * The first chunk of the COW device just contains the header.
- * After this there is a chunk filled with exception metadata,
- * followed by as many exception chunks as can fit in the
- * metadata areas.
- *
- * All on disk structures are in little-endian format.  The end
- * of the exceptions info is indicated by an exception with a
- * new_chunk of 0, which is invalid since it would point to the
- * header chunk.
- */
-
-/*
- * Magic for persistent snapshots: "SnAp" - Feeble isn't it.
- */
-#define SNAP_MAGIC 0x70416e53
-
-/*
- * The on-disk version of the metadata.
- */
-#define SNAPSHOT_DISK_VERSION 1
 
-struct disk_header {
-       uint32_t magic;
+#define DM_MSG_PREFIX "snapshot exception stores"
 
-       /*
-        * Is this snapshot valid.  There is no way of recovering
-        * an invalid snapshot.
-        */
-       uint32_t valid;
-
-       /*
-        * Simple, incrementing version. no backward
-        * compatibility.
-        */
-       uint32_t version;
-
-       /* In sectors */
-       uint32_t chunk_size;
-};
-
-struct disk_exception {
-       uint64_t old_chunk;
-       uint64_t new_chunk;
-};
-
-struct commit_callback {
-       void (*callback)(void *, int success);
-       void *context;
-};
-
-/*
- * The top level structure for a persistent exception store.
- */
-struct pstore {
-       struct dm_snapshot *snap;       /* up pointer to my snapshot */
-       int version;
-       int valid;
-       uint32_t exceptions_per_area;
+static LIST_HEAD(_exception_store_types);
+static DEFINE_SPINLOCK(_lock);
 
-       /*
-        * Now that we have an asynchronous kcopyd there is no
-        * need for large chunk sizes, so it wont hurt to have a
-        * whole chunks worth of metadata in memory at once.
-        */
-       void *area;
-
-       /*
-        * An area of zeros used to clear the next area.
-        */
-       void *zero_area;
-
-       /*
-        * Used to keep track of which metadata area the data in
-        * 'chunk' refers to.
-        */
-       chunk_t current_area;
-
-       /*
-        * The next free chunk for an exception.
-        */
-       chunk_t next_free;
-
-       /*
-        * The index of next free exception in the current
-        * metadata area.
-        */
-       uint32_t current_committed;
-
-       atomic_t pending_count;
-       uint32_t callback_count;
-       struct commit_callback *callbacks;
-       struct dm_io_client *io_client;
-
-       struct workqueue_struct *metadata_wq;
-};
-
-static unsigned sectors_to_pages(unsigned sectors)
-{
-       return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9);
-}
-
-static int alloc_area(struct pstore *ps)
+static struct dm_exception_store_type *__find_exception_store_type(const char *name)
 {
-       int r = -ENOMEM;
-       size_t len;
-
-       len = ps->snap->chunk_size << SECTOR_SHIFT;
-
-       /*
-        * Allocate the chunk_size block of memory that will hold
-        * a single metadata area.
-        */
-       ps->area = vmalloc(len);
-       if (!ps->area)
-               return r;
-
-       ps->zero_area = vmalloc(len);
-       if (!ps->zero_area) {
-               vfree(ps->area);
-               return r;
-       }
-       memset(ps->zero_area, 0, len);
+       struct dm_exception_store_type *type;
 
-       return 0;
-}
+       list_for_each_entry(type, &_exception_store_types, list)
+               if (!strcmp(name, type->name))
+                       return type;
 
-static void free_area(struct pstore *ps)
-{
-       vfree(ps->area);
-       ps->area = NULL;
-       vfree(ps->zero_area);
-       ps->zero_area = NULL;
+       return NULL;
 }
 
-struct mdata_req {
-       struct dm_io_region *where;
-       struct dm_io_request *io_req;
-       struct work_struct work;
-       int result;
-};
-
-static void do_metadata(struct work_struct *work)
+static struct dm_exception_store_type *_get_exception_store_type(const char *name)
 {
-       struct mdata_req *req = container_of(work, struct mdata_req, work);
+       struct dm_exception_store_type *type;
 
-       req->result = dm_io(req->io_req, 1, req->where, NULL);
-}
+       spin_lock(&_lock);
 
-/*
- * Read or write a chunk aligned and sized block of data from a device.
- */
-static int chunk_io(struct pstore *ps, chunk_t chunk, int rw, int metadata)
-{
-       struct dm_io_region where = {
-               .bdev = ps->snap->cow->bdev,
-               .sector = ps->snap->chunk_size * chunk,
-               .count = ps->snap->chunk_size,
-       };
-       struct dm_io_request io_req = {
-               .bi_rw = rw,
-               .mem.type = DM_IO_VMA,
-               .mem.ptr.vma = ps->area,
-               .client = ps->io_client,
-               .notify.fn = NULL,
-       };
-       struct mdata_req req;
-
-       if (!metadata)
-               return dm_io(&io_req, 1, &where, NULL);
-
-       req.where = &where;
-       req.io_req = &io_req;
+       type = __find_exception_store_type(name);
 
-       /*
-        * Issue the synchronous I/O from a different thread
-        * to avoid generic_make_request recursion.
-        */
-       INIT_WORK(&req.work, do_metadata);
-       queue_work(ps->metadata_wq, &req.work);
-       flush_workqueue(ps->metadata_wq);
+       if (type && !try_module_get(type->module))
+               type = NULL;
 
-       return req.result;
-}
+       spin_unlock(&_lock);
 
-/*
- * Convert a metadata area index to a chunk index.
- */
-static chunk_t area_location(struct pstore *ps, chunk_t area)
-{
-       return 1 + ((ps->exceptions_per_area + 1) * area);
+       return type;
 }
 
 /*
- * Read or write a metadata area.  Remembering to skip the first
- * chunk which holds the header.
+ * get_type
+ * @type_name
+ *
+ * Attempt to retrieve the dm_exception_store_type by name.  If not already
+ * available, attempt to load the appropriate module.
+ *
+ * Exstore modules are named "dm-exstore-" followed by the 'type_name'.
+ * Modules may contain multiple types.
+ * This function will first try the module "dm-exstore-<type_name>",
+ * then truncate 'type_name' on the last '-' and try again.
+ *
+ * For example, if type_name was "clustered-shared", it would search
+ * 'dm-exstore-clustered-shared' then 'dm-exstore-clustered'.
+ *
+ * 'dm-exception-store-<type_name>' is too long of a name in my
+ * opinion, which is why I've chosen to have the files
+ * containing exception store implementations be 'dm-exstore-<type_name>'.
+ * If you want your module to be autoloaded, you will follow this
+ * naming convention.
+ *
+ * Returns: dm_exception_store_type* on success, NULL on failure
  */
-static int area_io(struct pstore *ps, int rw)
+static struct dm_exception_store_type *get_type(const char *type_name)
 {
-       int r;
-       chunk_t chunk;
+       char *p, *type_name_dup;
+       struct dm_exception_store_type *type;
 
-       chunk = area_location(ps, ps->current_area);
+       type = _get_exception_store_type(type_name);
+       if (type)
+               return type;
 
-       r = chunk_io(ps, chunk, rw, 0);
-       if (r)
-               return r;
-
-       return 0;
-}
-
-static void zero_memory_area(struct pstore *ps)
-{
-       memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT);
-}
-
-static int zero_disk_area(struct pstore *ps, chunk_t area)
-{
-       struct dm_io_region where = {
-               .bdev = ps->snap->cow->bdev,
-               .sector = ps->snap->chunk_size * area_location(ps, area),
-               .count = ps->snap->chunk_size,
-       };
-       struct dm_io_request io_req = {
-               .bi_rw = WRITE,
-               .mem.type = DM_IO_VMA,
-               .mem.ptr.vma = ps->zero_area,
-               .client = ps->io_client,
-               .notify.fn = NULL,
-       };
-
-       return dm_io(&io_req, 1, &where, NULL);
-}
-
-static int read_header(struct pstore *ps, int *new_snapshot)
-{
-       int r;
-       struct disk_header *dh;
-       chunk_t chunk_size;
-       int chunk_size_supplied = 1;
-
-       /*
-        * Use default chunk size (or hardsect_size, if larger) if none supplied
-        */
-       if (!ps->snap->chunk_size) {
-               ps->snap->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
-                   bdev_hardsect_size(ps->snap->cow->bdev) >> 9);
-               ps->snap->chunk_mask = ps->snap->chunk_size - 1;
-               ps->snap->chunk_shift = ffs(ps->snap->chunk_size) - 1;
-               chunk_size_supplied = 0;
+       type_name_dup = kstrdup(type_name, GFP_KERNEL);
+       if (!type_name_dup) {
+               DMERR("No memory left to attempt load for \"%s\"", type_name);
+               return NULL;
        }
 
-       ps->io_client = dm_io_client_create(sectors_to_pages(ps->snap->
-                                                            chunk_size));
-       if (IS_ERR(ps->io_client))
-               return PTR_ERR(ps->io_client);
-
-       r = alloc_area(ps);
-       if (r)
-               return r;
-
-       r = chunk_io(ps, 0, READ, 1);
-       if (r)
-               goto bad;
-
-       dh = (struct disk_header *) ps->area;
-
-       if (le32_to_cpu(dh->magic) == 0) {
-               *new_snapshot = 1;
-               return 0;
-       }
-
-       if (le32_to_cpu(dh->magic) != SNAP_MAGIC) {
-               DMWARN("Invalid or corrupt snapshot");
-               r = -ENXIO;
-               goto bad;
+       while (request_module("dm-exstore-%s", type_name_dup) ||
+              !(type = _get_exception_store_type(type_name))) {
+               p = strrchr(type_name_dup, '-');
+               if (!p)
+                       break;
+               p[0] = '\0';
        }
 
-       *new_snapshot = 0;
-       ps->valid = le32_to_cpu(dh->valid);
-       ps->version = le32_to_cpu(dh->version);
-       chunk_size = le32_to_cpu(dh->chunk_size);
-
-       if (!chunk_size_supplied || ps->snap->chunk_size == chunk_size)
-               return 0;
-
-       DMWARN("chunk size %llu in device metadata overrides "
-              "table chunk size of %llu.",
-              (unsigned long long)chunk_size,
-              (unsigned long long)ps->snap->chunk_size);
-
-       /* We had a bogus chunk_size. Fix stuff up. */
-       free_area(ps);
-
-       ps->snap->chunk_size = chunk_size;
-       ps->snap->chunk_mask = chunk_size - 1;
-       ps->snap->chunk_shift = ffs(chunk_size) - 1;
+       if (!type)
+               DMWARN("Module for exstore type \"%s\" not found.", type_name);
 
-       r = dm_io_client_resize(sectors_to_pages(ps->snap->chunk_size),
-                               ps->io_client);
-       if (r)
-               return r;
-
-       r = alloc_area(ps);
-       return r;
+       kfree(type_name_dup);
 
-bad:
-       free_area(ps);
-       return r;
+       return type;
 }
 
-static int write_header(struct pstore *ps)
+static void put_type(struct dm_exception_store_type *type)
 {
-       struct disk_header *dh;
-
-       memset(ps->area, 0, ps->snap->chunk_size << SECTOR_SHIFT);
-
-       dh = (struct disk_header *) ps->area;
-       dh->magic = cpu_to_le32(SNAP_MAGIC);
-       dh->valid = cpu_to_le32(ps->valid);
-       dh->version = cpu_to_le32(ps->version);
-       dh->chunk_size = cpu_to_le32(ps->snap->chunk_size);
-
-       return chunk_io(ps, 0, WRITE, 1);
+       spin_lock(&_lock);
+       module_put(type->module);
+       spin_unlock(&_lock);
 }
 
-/*
- * Access functions for the disk exceptions, these do the endian conversions.
- */
-static struct disk_exception *get_exception(struct pstore *ps, uint32_t index)
+int dm_exception_store_type_register(struct dm_exception_store_type *type)
 {
-       BUG_ON(index >= ps->exceptions_per_area);
-
-       return ((struct disk_exception *) ps->area) + index;
-}
+       int r = 0;
 
-static void read_exception(struct pstore *ps,
-                          uint32_t index, struct disk_exception *result)
-{
-       struct disk_exception *e = get_exception(ps, index);
+       spin_lock(&_lock);
+       if (!__find_exception_store_type(type->name))
+               list_add(&type->list, &_exception_store_types);
+       else
+               r = -EEXIST;
+       spin_unlock(&_lock);
 
-       /* copy it */
-       result->old_chunk = le64_to_cpu(e->old_chunk);
-       result->new_chunk = le64_to_cpu(e->new_chunk);
+       return r;
 }
+EXPORT_SYMBOL(dm_exception_store_type_register);
 
-static void write_exception(struct pstore *ps,
-                           uint32_t index, struct disk_exception *de)
+int dm_exception_store_type_unregister(struct dm_exception_store_type *type)
 {
-       struct disk_exception *e = get_exception(ps, index);
+       spin_lock(&_lock);
 
-       /* copy it */
-       e->old_chunk = cpu_to_le64(de->old_chunk);
-       e->new_chunk = cpu_to_le64(de->new_chunk);
-}
-
-/*
- * Registers the exceptions that are present in the current area.
- * 'full' is filled in to indicate if the area has been
- * filled.
- */
-static int insert_exceptions(struct pstore *ps, int *full)
-{
-       int r;
-       unsigned int i;
-       struct disk_exception de;
-
-       /* presume the area is full */
-       *full = 1;
-
-       for (i = 0; i < ps->exceptions_per_area; i++) {
-               read_exception(ps, i, &de);
-
-               /*
-                * If the new_chunk is pointing at the start of
-                * the COW device, where the first metadata area
-                * is we know that we've hit the end of the
-                * exceptions.  Therefore the area is not full.
-                */
-               if (de.new_chunk == 0LL) {
-                       ps->current_committed = i;
-                       *full = 0;
-                       break;
-               }
-
-               /*
-                * Keep track of the start of the free chunks.
-                */
-               if (ps->next_free <= de.new_chunk)
-                       ps->next_free = de.new_chunk + 1;
-
-               /*
-                * Otherwise we add the exception to the snapshot.
-                */
-               r = dm_add_exception(ps->snap, de.old_chunk, de.new_chunk);
-               if (r)
-                       return r;
+       if (!__find_exception_store_type(type->name)) {
+               spin_unlock(&_lock);
+               return -EINVAL;
        }
 
-       return 0;
-}
+       list_del(&type->list);
 
-static int read_exceptions(struct pstore *ps)
-{
-       int r, full = 1;
-
-       /*
-        * Keeping reading chunks and inserting exceptions until
-        * we find a partially full area.
-        */
-       for (ps->current_area = 0; full; ps->current_area++) {
-               r = area_io(ps, READ);
-               if (r)
-                       return r;
-
-               r = insert_exceptions(ps, &full);
-               if (r)
-                       return r;
-       }
-
-       ps->current_area--;
+       spin_unlock(&_lock);
 
        return 0;
 }
+EXPORT_SYMBOL(dm_exception_store_type_unregister);
 
-static struct pstore *get_info(struct exception_store *store)
-{
-       return (struct pstore *) store->context;
-}
-
-static void persistent_fraction_full(struct exception_store *store,
-                                    sector_t *numerator, sector_t *denominator)
+/*
+ * Round a number up to the nearest 'size' boundary.  size must
+ * be a power of 2.
+ */
+static ulong round_up(ulong n, ulong size)
 {
-       *numerator = get_info(store)->next_free * store->snap->chunk_size;
-       *denominator = get_dev_size(store->snap->cow->bdev);
+       size--;
+       return (n + size) & ~size;
 }
 
-static void persistent_destroy(struct exception_store *store)
+static int set_chunk_size(struct dm_exception_store *store,
+                         const char *chunk_size_arg, char **error)
 {
-       struct pstore *ps = get_info(store);
+       unsigned long chunk_size_ulong;
+       char *value;
 
-       destroy_workqueue(ps->metadata_wq);
-       dm_io_client_destroy(ps->io_client);
-       vfree(ps->callbacks);
-       free_area(ps);
-       kfree(ps);
-}
-
-static int persistent_read_metadata(struct exception_store *store)
-{
-       int r, uninitialized_var(new_snapshot);
-       struct pstore *ps = get_info(store);
+       chunk_size_ulong = simple_strtoul(chunk_size_arg, &value, 10);
+       if (*chunk_size_arg == '\0' || *value != '\0') {
+               *error = "Invalid chunk size";
+               return -EINVAL;
+       }
 
-       /*
-        * Read the snapshot header.
-        */
-       r = read_header(ps, &new_snapshot);
-       if (r)
-               return r;
+       if (!chunk_size_ulong) {
+               store->chunk_size = store->chunk_mask = store->chunk_shift = 0;
+               return 0;
+       }
 
        /*
-        * Now we know correct chunk_size, complete the initialisation.
+        * Chunk size must be multiple of page size.  Silently
+        * round up if it's not.
         */
-       ps->exceptions_per_area = (ps->snap->chunk_size << SECTOR_SHIFT) /
-                                 sizeof(struct disk_exception);
-       ps->callbacks = dm_vcalloc(ps->exceptions_per_area,
-                       sizeof(*ps->callbacks));
-       if (!ps->callbacks)
-               return -ENOMEM;
+       chunk_size_ulong = round_up(chunk_size_ulong, PAGE_SIZE >> 9);
 
-       /*
-        * Do we need to setup a new snapshot ?
-        */
-       if (new_snapshot) {
-               r = write_header(ps);
-               if (r) {
-                       DMWARN("write_header failed");
-                       return r;
-               }
-
-               ps->current_area = 0;
-               zero_memory_area(ps);
-               r = zero_disk_area(ps, 0);
-               if (r) {
-                       DMWARN("zero_disk_area(0) failed");
-                       return r;
-               }
-       } else {
-               /*
-                * Sanity checks.
-                */
-               if (ps->version != SNAPSHOT_DISK_VERSION) {
-                       DMWARN("unable to handle snapshot disk version %d",
-                              ps->version);
-                       return -EINVAL;
-               }
-
-               /*
-                * Metadata are valid, but snapshot is invalidated
-                */
-               if (!ps->valid)
-                       return 1;
-
-               /*
-                * Read the metadata.
-                */
-               r = read_exceptions(ps);
-               if (r)
-                       return r;
+       /* Check chunk_size is a power of 2 */
+       if (!is_power_of_2(chunk_size_ulong)) {
+               *error = "Chunk size is not a power of 2";
+               return -EINVAL;
        }
 
-       return 0;
-}
-
-static int persistent_prepare(struct exception_store *store,
-                             struct dm_snap_exception *e)
-{
-       struct pstore *ps = get_info(store);
-       uint32_t stride;
-       chunk_t next_free;
-       sector_t size = get_dev_size(store->snap->cow->bdev);
-
-       /* Is there enough room ? */
-       if (size < ((ps->next_free + 1) * store->snap->chunk_size))
-               return -ENOSPC;
-
-       e->new_chunk = ps->next_free;
+       /* Validate the chunk size against the device block size */
+       if (chunk_size_ulong % (bdev_hardsect_size(store->cow->bdev) >> 9)) {
+               *error = "Chunk size is not a multiple of device blocksize";
+               return -EINVAL;
+       }
 
-       /*
-        * Move onto the next free pending, making sure to take
-        * into account the location of the metadata chunks.
-        */
-       stride = (ps->exceptions_per_area + 1);
-       next_free = ++ps->next_free;
-       if (sector_div(next_free, stride) == 1)
-               ps->next_free++;
+       store->chunk_size = chunk_size_ulong;
+       store->chunk_mask = chunk_size_ulong - 1;
+       store->chunk_shift = ffs(chunk_size_ulong) - 1;
 
-       atomic_inc(&ps->pending_count);
        return 0;
 }
 
-static void persistent_commit(struct exception_store *store,
-                             struct dm_snap_exception *e,
-                             void (*callback) (void *, int success),
-                             void *callback_context)
+int dm_exception_store_create(struct dm_target *ti, int argc, char **argv,
+                             unsigned *args_used,
+                             struct dm_exception_store **store)
 {
-       unsigned int i;
-       struct pstore *ps = get_info(store);
-       struct disk_exception de;
-       struct commit_callback *cb;
-
-       de.old_chunk = e->old_chunk;
-       de.new_chunk = e->new_chunk;
-       write_exception(ps, ps->current_committed++, &de);
-
-       /*
-        * Add the callback to the back of the array.  This code
-        * is the only place where the callback array is
-        * manipulated, and we know that it will never be called
-        * multiple times concurrently.
-        */
-       cb = ps->callbacks + ps->callback_count++;
-       cb->callback = callback;
-       cb->context = callback_context;
-
-       /*
-        * If there are exceptions in flight and we have not yet
-        * filled this metadata area there's nothing more to do.
-        */
-       if (!atomic_dec_and_test(&ps->pending_count) &&
-           (ps->current_committed != ps->exceptions_per_area))
-               return;
-
-       /*
-        * If we completely filled the current area, then wipe the next one.
-        */
-       if ((ps->current_committed == ps->exceptions_per_area) &&
-            zero_disk_area(ps, ps->current_area + 1))
-               ps->valid = 0;
-
-       /*
-        * Commit exceptions to disk.
-        */
-       if (ps->valid && area_io(ps, WRITE))
-               ps->valid = 0;
-
-       /*
-        * Advance to the next area if this one is full.
-        */
-       if (ps->current_committed == ps->exceptions_per_area) {
-               ps->current_committed = 0;
-               ps->current_area++;
-               zero_memory_area(ps);
+       int r = 0;
+       struct dm_exception_store_type *type;
+       struct dm_exception_store *tmp_store;
+       char persistent;
+
+       if (argc < 3) {
+               ti->error = "Insufficient exception store arguments";
+               return -EINVAL;
        }
 
-       for (i = 0; i < ps->callback_count; i++) {
-               cb = ps->callbacks + i;
-               cb->callback(cb->context, ps->valid);
+       tmp_store = kmalloc(sizeof(*tmp_store), GFP_KERNEL);
+       if (!tmp_store) {
+               ti->error = "Exception store allocation failed";
+               return -ENOMEM;
        }
 
-       ps->callback_count = 0;
-}
+       persistent = toupper(*argv[1]);
+       if (persistent != 'P' && persistent != 'N') {
+               ti->error = "Persistent flag is not P or N";
+               return -EINVAL;
+       }
 
-static void persistent_drop(struct exception_store *store)
-{
-       struct pstore *ps = get_info(store);
+       type = get_type(argv[1]);
+       if (!type) {
+               ti->error = "Exception store type not recognised";
+               r = -EINVAL;
+               goto bad_type;
+       }
 
-       ps->valid = 0;
-       if (write_header(ps))
-               DMWARN("write header failed");
-}
+       tmp_store->type = type;
+       tmp_store->ti = ti;
 
-int dm_create_persistent(struct exception_store *store)
-{
-       struct pstore *ps;
+       r = dm_get_device(ti, argv[0], 0, 0,
+                         FMODE_READ | FMODE_WRITE, &tmp_store->cow);
+       if (r) {
+               ti->error = "Cannot get COW device";
+               goto bad_cow;
+       }
 
-       /* allocate the pstore */
-       ps = kmalloc(sizeof(*ps), GFP_KERNEL);
-       if (!ps)
-               return -ENOMEM;
+       r = set_chunk_size(tmp_store, argv[2], &ti->error);
+       if (r)
+               goto bad_cow;
 
-       ps->snap = store->snap;
-       ps->valid = 1;
-       ps->version = SNAPSHOT_DISK_VERSION;
-       ps->area = NULL;
-       ps->next_free = 2;      /* skipping the header and first area */
-       ps->current_committed = 0;
-
-       ps->callback_count = 0;
-       atomic_set(&ps->pending_count, 0);
-       ps->callbacks = NULL;
-
-       ps->metadata_wq = create_singlethread_workqueue("ksnaphd");
-       if (!ps->metadata_wq) {
-               kfree(ps);
-               DMERR("couldn't start header metadata update thread");
-               return -ENOMEM;
+       r = type->ctr(tmp_store, 0, NULL);
+       if (r) {
+               ti->error = "Exception store type constructor failed";
+               goto bad_ctr;
        }
 
-       store->destroy = persistent_destroy;
-       store->read_metadata = persistent_read_metadata;
-       store->prepare_exception = persistent_prepare;
-       store->commit_exception = persistent_commit;
-       store->drop_snapshot = persistent_drop;
-       store->fraction_full = persistent_fraction_full;
-       store->context = ps;
-
+       *args_used = 3;
+       *store = tmp_store;
        return 0;
-}
-
-/*-----------------------------------------------------------------
- * Implementation of the store for non-persistent snapshots.
- *---------------------------------------------------------------*/
-struct transient_c {
-       sector_t next_free;
-};
 
-static void transient_destroy(struct exception_store *store)
-{
-       kfree(store->context);
+bad_ctr:
+       dm_put_device(ti, tmp_store->cow);
+bad_cow:
+       put_type(type);
+bad_type:
+       kfree(tmp_store);
+       return r;
 }
+EXPORT_SYMBOL(dm_exception_store_create);
 
-static int transient_read_metadata(struct exception_store *store)
+void dm_exception_store_destroy(struct dm_exception_store *store)
 {
-       return 0;
+       store->type->dtr(store);
+       dm_put_device(store->ti, store->cow);
+       put_type(store->type);
+       kfree(store);
 }
+EXPORT_SYMBOL(dm_exception_store_destroy);
 
-static int transient_prepare(struct exception_store *store,
-                            struct dm_snap_exception *e)
+int dm_exception_store_init(void)
 {
-       struct transient_c *tc = (struct transient_c *) store->context;
-       sector_t size = get_dev_size(store->snap->cow->bdev);
+       int r;
 
-       if (size < (tc->next_free + store->snap->chunk_size))
-               return -1;
+       r = dm_transient_snapshot_init();
+       if (r) {
+               DMERR("Unable to register transient exception store type.");
+               goto transient_fail;
+       }
 
-       e->new_chunk = sector_to_chunk(store->snap, tc->next_free);
-       tc->next_free += store->snap->chunk_size;
+       r = dm_persistent_snapshot_init();
+       if (r) {
+               DMERR("Unable to register persistent exception store type");
+               goto persistent_fail;
+       }
 
        return 0;
-}
-
-static void transient_commit(struct exception_store *store,
-                            struct dm_snap_exception *e,
-                            void (*callback) (void *, int success),
-                            void *callback_context)
-{
-       /* Just succeed */
-       callback(callback_context, 1);
-}
 
-static void transient_fraction_full(struct exception_store *store,
-                                   sector_t *numerator, sector_t *denominator)
-{
-       *numerator = ((struct transient_c *) store->context)->next_free;
-       *denominator = get_dev_size(store->snap->cow->bdev);
+persistent_fail:
+       dm_persistent_snapshot_exit();
+transient_fail:
+       return r;
 }
 
-int dm_create_transient(struct exception_store *store)
+void dm_exception_store_exit(void)
 {
-       struct transient_c *tc;
-
-       store->destroy = transient_destroy;
-       store->read_metadata = transient_read_metadata;
-       store->prepare_exception = transient_prepare;
-       store->commit_exception = transient_commit;
-       store->drop_snapshot = NULL;
-       store->fraction_full = transient_fraction_full;
-
-       tc = kmalloc(sizeof(struct transient_c), GFP_KERNEL);
-       if (!tc)
-               return -ENOMEM;
-
-       tc->next_free = 0;
-       store->context = tc;
-
-       return 0;
+       dm_persistent_snapshot_exit();
+       dm_transient_snapshot_exit();
 }