From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cn.fujitsu.com ([59.151.112.132]:49498 "EHLO heian.cn.fujitsu.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1752422AbcAGBIr (ORCPT ); Wed, 6 Jan 2016 20:08:47 -0500 Received: from G08CNEXCHPEKD02.g08.fujitsu.local (unknown [10.167.33.83]) by cn.fujitsu.com (Postfix) with ESMTP id 9725141887DE for ; Thu, 7 Jan 2016 09:08:26 +0800 (CST) From: Qu Wenruo To: CC: Wang Xiaoguang Subject: [PATCH v3 13/16] btrfs: dedup: Add support for on-disk hash search Date: Thu, 7 Jan 2016 09:08:14 +0800 Message-ID: <1452128897-5433-14-git-send-email-quwenruo@cn.fujitsu.com> In-Reply-To: <1452128897-5433-1-git-send-email-quwenruo@cn.fujitsu.com> References: <1452128897-5433-1-git-send-email-quwenruo@cn.fujitsu.com> MIME-Version: 1.0 Content-Type: text/plain Sender: linux-btrfs-owner@vger.kernel.org List-ID: Now on-disk backend should be able to search hash now. Signed-off-by: Wang Xiaoguang Signed-off-by: Qu Wenruo --- v2: Newly introduced v3: None --- fs/btrfs/dedup.c | 169 ++++++++++++++++++++++++++++++++++++++++++++++++++++++- fs/btrfs/dedup.h | 3 + 2 files changed, 171 insertions(+), 1 deletion(-) diff --git a/fs/btrfs/dedup.c b/fs/btrfs/dedup.c index 5edb923..4bcdf5d 100644 --- a/fs/btrfs/dedup.c +++ b/fs/btrfs/dedup.c @@ -493,6 +493,172 @@ int btrfs_dedup_disable(struct btrfs_fs_info *fs_info) } /* + * Return 0 for not found + * Return >0 for found and set bytenr_ret + * Return <0 for error + */ +static int ondisk_search_hash(struct btrfs_dedup_info *dedup_info, u8 *hash, + u64 *bytenr_ret, u64 *num_bytes_ret) +{ + struct btrfs_path *path; + struct btrfs_key key; + struct btrfs_root *dedup_root = dedup_info->dedup_root; + u8 *buf = NULL; + u64 hash_key; + int hash_len = btrfs_dedup_sizes[dedup_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_DEDUP_HASH_ITEM_KEY; + key.offset = (u64)-1; + + ret = btrfs_search_slot(NULL, dedup_root, &key, path, 0, 0); + if (ret < 0) + goto out; + WARN_ON(ret == 0); + while (1) { + struct extent_buffer *node; + struct btrfs_dedup_hash_item *hash_item; + int slot; + + ret = btrfs_previous_item(dedup_root, path, hash_key, + BTRFS_DEDUP_HASH_ITEM_KEY); + if (ret < 0) + goto out; + if (ret > 0) { + ret = 0; + goto out; + } + + node = path->nodes[0]; + slot = path->slots[0]; + btrfs_item_key_to_cpu(node, &key, slot); + + if (key.type != BTRFS_DEDUP_HASH_ITEM_KEY || + memcmp(&key.objectid, hash + hash_len - 8, 8)) + break; + hash_item = btrfs_item_ptr(node, slot, + struct btrfs_dedup_hash_item); + read_extent_buffer(node, buf, (unsigned long)(hash_item + 1), + hash_len); + if (!memcmp(buf, hash, hash_len)) { + ret = 1; + *bytenr_ret = key.offset; + *num_bytes_ret = btrfs_dedup_hash_len(node, hash_item); + break; + } + } +out: + kfree(buf); + btrfs_free_path(path); + return ret; +} + +static int ondisk_search(struct inode *inode, u64 file_pos, + struct btrfs_dedup_hash *hash) +{ + int ret; + struct btrfs_root *root = BTRFS_I(inode)->root; + struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_trans_handle *trans = NULL; + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_head *head; + struct btrfs_dedup_info *dedup_info = fs_info->dedup_info; + u64 old_bytenr; + u64 bytenr; + u64 num_bytes; + + /* + * TODO: Opitmized the superhot mutex. + */ + mutex_lock(&dedup_info->ondisk_lock); + ret = ondisk_search_hash(dedup_info, hash->hash, &bytenr, &num_bytes); + mutex_unlock(&dedup_info->ondisk_lock); + if (ret <= 0) + goto out; + + trans = btrfs_join_transaction(root); + if (IS_ERR(trans)) + return PTR_ERR(trans); + +again: + delayed_refs = &trans->transaction->delayed_refs; + + spin_lock(&delayed_refs->lock); + head = btrfs_find_delayed_ref_head(trans, bytenr); + if (!head || head->processing == 1) { + /* + * Somebody else may be trying to run the refs, the found + * duplicated extent may be freed, so here we just + * choose to abort this dedup handle. + * XXX: we need to find a better method to improve it. + */ + spin_unlock(&delayed_refs->lock); + ret = 0; + goto out; + } + + ret = btrfs_delayed_ref_lock(trans, head); + spin_unlock(&delayed_refs->lock); + if (ret == -EAGAIN) { + mutex_lock(&dedup_info->ondisk_lock); + ret = ondisk_search_hash(dedup_info, hash->hash, &bytenr, + &num_bytes); + mutex_unlock(&dedup_info->ondisk_lock); + if (ret <= 0) + goto out; + goto again; + } + /* + * Still need to search the hash again to ensure the hash is not + * deleted in run_delayed_refs + */ + old_bytenr = bytenr; + mutex_lock(&dedup_info->ondisk_lock); + ret = ondisk_search_hash(dedup_info, hash->hash, &bytenr, &num_bytes); + if (ret <= 0) { + mutex_unlock(&dedup_info->ondisk_lock); + mutex_unlock(&head->mutex); + goto out; + } + + /* bytenr changed, we need to relock the delayed_ref head */ + if (old_bytenr != bytenr) { + mutex_unlock(&dedup_info->ondisk_lock); + mutex_unlock(&head->mutex); + goto again; + } + + /* + * finally, we found the matching hash, increase extent ref right now + * to avoid delayed ref run it + */ + btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, + root->root_key.objectid, + btrfs_ino(inode), file_pos); + hash->bytenr = bytenr; + hash->num_bytes = num_bytes; + mutex_unlock(&dedup_info->ondisk_lock); + mutex_unlock(&head->mutex); + ret = 1; +out: + if (trans) + btrfs_end_transaction(trans, root); + return ret; +} + +/* * Caller must ensure the corresponding ref head is not being run. */ static struct inmem_hash * @@ -646,7 +812,8 @@ int btrfs_dedup_search(struct inode *inode, u64 file_pos, if (dedup_info->backend == BTRFS_DEDUP_BACKEND_INMEMORY) ret = inmem_search(inode, file_pos, hash); - + if (dedup_info->backend == BTRFS_DEDUP_BACKEND_ONDISK) + ret = ondisk_search(inode, file_pos, hash); /* It's possible hash->bytenr/num_bytenr already changed */ if (ret == 0) { hash->num_bytes = 0; diff --git a/fs/btrfs/dedup.h b/fs/btrfs/dedup.h index 0503b98..03cc818 100644 --- a/fs/btrfs/dedup.h +++ b/fs/btrfs/dedup.h @@ -138,6 +138,9 @@ int btrfs_dedup_calc_hash(struct btrfs_root *root, struct inode *inode, * Return > 0 for a hash match, and the extent ref will be * INCREASED. * Return 0 for a hash miss. Nothing is done + * Return <0 for error. + * + * Only on-disk backedn may return error though. */ int btrfs_dedup_search(struct inode *inode, u64 file_pos, struct btrfs_dedup_hash *hash); -- 2.6.4