On 04/01/2016 02:35 AM, Qu Wenruo wrote:
Now on-disk backend should be able to search hash now.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@xxxxxxxxxxxxxx>
Signed-off-by: Qu Wenruo <quwenruo@xxxxxxxxxxxxxx>
---
fs/btrfs/dedupe.c | 167 ++++++++++++++++++++++++++++++++++++++++++++++++------
fs/btrfs/dedupe.h | 1 +
2 files changed, 151 insertions(+), 17 deletions(-)
diff --git a/fs/btrfs/dedupe.c b/fs/btrfs/dedupe.c
index a274c1c..00f2a01 100644
--- a/fs/btrfs/dedupe.c
+++ b/fs/btrfs/dedupe.c
@@ -652,6 +652,112 @@ int btrfs_dedupe_disable(struct btrfs_fs_info *fs_info)
}
/*
+ * Compare ondisk hash with src.
+ * Return 0 if hash matches.
+ * Return non-zero for hash mismatch
+ *
+ * Caller should ensure the slot contains a valid hash item.
+ */
+static int memcmp_ondisk_hash(const struct btrfs_key *key,
+ struct extent_buffer *node, int slot,
+ int hash_len, const u8 *src)
+{
+ u64 offset;
+ int ret;
+
+ /* Return value doesn't make sense in this case though */
+ if (WARN_ON(hash_len <= 8 || key->type != BTRFS_DEDUPE_HASH_ITEM_KEY))
No magic numbers please.
+ return -EINVAL;
+
+ /* compare the hash exlcuding the last 64 bits */
+ offset = btrfs_item_ptr_offset(node, slot);
+ ret = memcmp_extent_buffer(node, src, offset, hash_len - 8);
+ if (ret)
+ return ret;
+ return memcmp(&key->objectid, src + hash_len - 8, 8);
+}
+
+ /*
+ * Return 0 for not found
+ * Return >0 for found and set bytenr_ret
+ * Return <0 for error
+ */
+static int ondisk_search_hash(struct btrfs_dedupe_info *dedupe_info, u8 *hash,
+ u64 *bytenr_ret, u32 *num_bytes_ret)
+{
+ struct btrfs_path *path;
+ struct btrfs_key key;
+ struct btrfs_root *dedupe_root = dedupe_info->dedupe_root;
+ u8 *buf = NULL;
+ u64 hash_key;
+ int hash_len = btrfs_dedupe_sizes[dedupe_info->hash_type];
+ int ret;
+
+ path = btrfs_alloc_path();
+ if (!path)
+ return -ENOMEM;
+
+ buf = kmalloc(hash_len, GFP_NOFS);
+ if (!buf) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ memcpy(&hash_key, hash + hash_len - 8, 8);
+ key.objectid = hash_key;
+ key.type = BTRFS_DEDUPE_HASH_ITEM_KEY;
+ key.offset = (u64)-1;
+
+ ret = btrfs_search_slot(NULL, dedupe_root, &key, path, 0, 0);
+ if (ret < 0)
+ goto out;
+ WARN_ON(ret == 0);
+ while (1) {
+ struct extent_buffer *node;
+ struct btrfs_dedupe_hash_item *hash_item;
+ int slot;
+
+ ret = btrfs_previous_item(dedupe_root, path, hash_key,
+ BTRFS_DEDUPE_HASH_ITEM_KEY);
+ if (ret < 0)
+ break;
+ if (ret > 0) {
+ ret = 0;
+ break;
+ }
+
+ node = path->nodes[0];
+ slot = path->slots[0];
+ btrfs_item_key_to_cpu(node, &key, slot);
+
+ /*
+ * Type of objectid mismatch means no previous item may
+ * hit, exit searching
+ */
+ if (key.type != BTRFS_DEDUPE_HASH_ITEM_KEY ||
+ memcmp(&key.objectid, &hash_key, 8))
+ break;
+ hash_item = btrfs_item_ptr(node, slot,
+ struct btrfs_dedupe_hash_item);
+ /*
+ * If the hash mismatch, it's still possible that previous item
+ * has the desired hash.
+ */
+ if (memcmp_ondisk_hash(&key, node, slot, hash_len, hash))
+ continue;
+ /* Found */
+ ret = 1;
+ *bytenr_ret = key.offset;
+ *num_bytes_ret = dedupe_info->blocksize;
+ break;
+ }
+out:
+ kfree(buf);
+ btrfs_free_path(path);
+ return ret;
+}
+
+/*
* Caller must ensure the corresponding ref head is not being run.
*/
static struct inmem_hash *
@@ -681,9 +787,36 @@ inmem_search_hash(struct btrfs_dedupe_info *dedupe_info, u8 *hash)
return NULL;
}
-static int inmem_search(struct btrfs_dedupe_info *dedupe_info,
- struct inode *inode, u64 file_pos,
- struct btrfs_dedupe_hash *hash)
+/* Wapper for different backends, caller needs to hold dedupe_info->lock */
+static inline int generic_search_hash(struct btrfs_dedupe_info *dedupe_info,
+ u8 *hash, u64 *bytenr_ret,
+ u32 *num_bytes_ret)
+{
+ if (dedupe_info->backend == BTRFS_DEDUPE_BACKEND_INMEMORY) {
+ struct inmem_hash *found_hash;
+ int ret;
+
+ found_hash = inmem_search_hash(dedupe_info, hash);
+ if (found_hash) {
+ ret = 1;
+ *bytenr_ret = found_hash->bytenr;
+ *num_bytes_ret = found_hash->num_bytes;
+ } else {
+ ret = 0;
+ *bytenr_ret = 0;
+ *num_bytes_ret = 0;
Why set it to 0 only in the INMEMORY case? If they need to be zero'ed
perhaps do it at the start of the helper? Thanks,
Josef
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
