ext4: xattr inode deduplication

Ext4 now supports xattr values that are up to 64k in size (vfs limit).
Large xattr values are stored in external inodes each one holding a
single value. Once written the data blocks of these inodes are immutable.

The real world use cases are expected to have a lot of value duplication
such as inherited acls etc. To reduce data duplication on disk, this patch
implements a deduplicator that allows sharing of xattr inodes.

The deduplication is based on an in-memory hash lookup that is a best
effort sharing scheme. When a xattr inode is read from disk (i.e.
getxattr() call), its crc32c hash is added to a hash table. Before
creating a new xattr inode for a value being set, the hash table is
checked to see if an existing inode holds an identical value. If such an
inode is found, the ref count on that inode is incremented. On value
removal the ref count is decremented and if it reaches zero the inode is
deleted.

The quota charging for such inodes is manually managed. Every reference
holder is charged the full size as if there was no sharing happening.
This is consistent with how xattr blocks are also charged.

[ Fixed up journal credits calculation to handle inline data and the
  rare case where an shared xattr block can get freed when two thread
  race on breaking the xattr block sharing. --tytso ]
Signed-off-by: Tahsin Erdogan <tahsin@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

fs/ext4/acl.c

@@ -238,7 +238,10 @@ ext4_set_acl(struct inode *inode, struct posix_acl *acl, int type)
 	if (error)
 		return error;
 retry:
-	credits = ext4_xattr_set_credits(inode, acl_size);
+	error = ext4_xattr_set_credits(inode, acl_size, &credits);
+	if (error)
+		return error;
+
 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
 	if (IS_ERR(handle))
 		return PTR_ERR(handle);

fs/ext4/ext4.h

@@ -1517,6 +1517,7 @@ struct ext4_sb_info {
 	long s_es_nr_inode;
 	struct ext4_es_stats s_es_stats;
 	struct mb_cache *s_ea_block_cache;
+	struct mb_cache *s_ea_inode_cache;
 	spinlock_t s_es_lock ____cacheline_aligned_in_smp;
 
 	/* Ratelimit ext4 messages. */
@@ -2100,7 +2101,11 @@ static inline struct ext4_inode *ext4_raw_inode(struct ext4_iloc *iloc)
 	return (struct ext4_inode *) (iloc->bh->b_data + iloc->offset);
 }
 
-#define ext4_is_quota_file(inode) IS_NOQUOTA(inode)
+static inline bool ext4_is_quota_file(struct inode *inode)
+{
+	return IS_NOQUOTA(inode) &&
+	       !(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL);
+}
 
 /*
  * This structure is stuffed into the struct file's private_data field
@@ -2493,7 +2498,6 @@ extern int ext4_truncate_restart_trans(handle_t *, struct inode *, int nblocks);
 extern void ext4_set_inode_flags(struct inode *);
 extern int ext4_alloc_da_blocks(struct inode *inode);
 extern void ext4_set_aops(struct inode *inode);
-extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
 extern int ext4_writepage_trans_blocks(struct inode *);
 extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
 extern int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
@@ -2720,19 +2724,20 @@ extern void ext4_group_desc_csum_set(struct super_block *sb, __u32 group,
 extern int ext4_register_li_request(struct super_block *sb,
 				    ext4_group_t first_not_zeroed);
 
-static inline int ext4_has_group_desc_csum(struct super_block *sb)
-{
-	return ext4_has_feature_gdt_csum(sb) ||
-	       EXT4_SB(sb)->s_chksum_driver != NULL;
-}
-
 static inline int ext4_has_metadata_csum(struct super_block *sb)
 {
 	WARN_ON_ONCE(ext4_has_feature_metadata_csum(sb) &&
 		     !EXT4_SB(sb)->s_chksum_driver);
 
-	return (EXT4_SB(sb)->s_chksum_driver != NULL);
+	return ext4_has_feature_metadata_csum(sb) &&
+	       (EXT4_SB(sb)->s_chksum_driver != NULL);
 }
+
+static inline int ext4_has_group_desc_csum(struct super_block *sb)
+{
+	return ext4_has_feature_gdt_csum(sb) || ext4_has_metadata_csum(sb);
+}
+
 static inline ext4_fsblk_t ext4_blocks_count(struct ext4_super_block *es)
 {
 	return ((ext4_fsblk_t)le32_to_cpu(es->s_blocks_count_hi) << 32) |

fs/ext4/inode.c

@@ -139,6 +139,8 @@ static void ext4_invalidatepage(struct page *page, unsigned int offset,
 				unsigned int length);
 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
+static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
+				  int pextents);
 
 /*
  * Test whether an inode is a fast symlink.
@@ -4843,8 +4845,15 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
 	}
 	brelse(iloc.bh);
 	ext4_set_inode_flags(inode);
-	if (ei->i_flags & EXT4_EA_INODE_FL)
+
+	if (ei->i_flags & EXT4_EA_INODE_FL) {
 		ext4_xattr_inode_set_class(inode);
+
+		inode_lock(inode);
+		inode->i_flags |= S_NOQUOTA;
+		inode_unlock(inode);
+	}
+
 	unlock_new_inode(inode);
 	return inode;
 
@@ -5503,7 +5512,7 @@ static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
  *
  * Also account for superblock, inode, quota and xattr blocks
  */
-int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
+static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 				  int pextents)
 {
 	ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);

fs/ext4/super.c

@@ -927,6 +927,10 @@ static void ext4_put_super(struct super_block *sb)
 		invalidate_bdev(sbi->journal_bdev);
 		ext4_blkdev_remove(sbi);
 	}
+	if (sbi->s_ea_inode_cache) {
+		ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
+		sbi->s_ea_inode_cache = NULL;
+	}
 	if (sbi->s_ea_block_cache) {
 		ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 		sbi->s_ea_block_cache = NULL;
@@ -1178,7 +1182,10 @@ static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
 	if (res)
 		return res;
 retry:
-	credits = ext4_xattr_set_credits(inode, len);
+	res = ext4_xattr_set_credits(inode, len, &credits);
+	if (res)
+		return res;
+
 	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
 	if (IS_ERR(handle))
 		return PTR_ERR(handle);
@@ -3445,7 +3452,8 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 	}
 
 	/* Load the checksum driver */
-	if (ext4_has_feature_metadata_csum(sb)) {
+	if (ext4_has_feature_metadata_csum(sb) ||
+	    ext4_has_feature_ea_inode(sb)) {
 		sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
 		if (IS_ERR(sbi->s_chksum_driver)) {
 			ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
@@ -3467,7 +3475,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 	/* Precompute checksum seed for all metadata */
 	if (ext4_has_feature_csum_seed(sb))
 		sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
-	else if (ext4_has_metadata_csum(sb))
+	else if (ext4_has_metadata_csum(sb) || ext4_has_feature_ea_inode(sb))
 		sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
 					       sizeof(es->s_uuid));
 
@@ -3597,6 +3605,16 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 				 "The Hurd can't support 64-bit file systems");
 			goto failed_mount;
 		}
+
+		/*
+		 * ea_inode feature uses l_i_version field which is not
+		 * available in HURD_COMPAT mode.
+		 */
+		if (ext4_has_feature_ea_inode(sb)) {
+			ext4_msg(sb, KERN_ERR,
+				 "ea_inode feature is not supported for Hurd");
+			goto failed_mount;
+		}
 	}
 
 	if (IS_EXT2_SB(sb)) {
@@ -4067,6 +4085,15 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 		goto failed_mount_wq;
 	}
 
+	if (ext4_has_feature_ea_inode(sb)) {
+		sbi->s_ea_inode_cache = ext4_xattr_create_cache();
+		if (!sbi->s_ea_inode_cache) {
+			ext4_msg(sb, KERN_ERR,
+				 "Failed to create ea_inode_cache");
+			goto failed_mount_wq;
+		}
+	}
+
 	if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
 	    (blocksize != PAGE_SIZE)) {
 		ext4_msg(sb, KERN_ERR,
@@ -4296,6 +4323,10 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 	if (EXT4_SB(sb)->rsv_conversion_wq)
 		destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
 failed_mount_wq:
+	if (sbi->s_ea_inode_cache) {
+		ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
+		sbi->s_ea_inode_cache = NULL;
+	}
 	if (sbi->s_ea_block_cache) {
 		ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
 		sbi->s_ea_block_cache = NULL;

fs/ext4/xattr.c

@@ -108,6 +108,9 @@ const struct xattr_handler *ext4_xattr_handlers[] = {
 #define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
 				inode->i_sb->s_fs_info)->s_ea_block_cache)
 
+#define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
+				inode->i_sb->s_fs_info)->s_ea_inode_cache)
+
 static int
 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 			struct inode *inode);
@@ -280,15 +283,44 @@ ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 	return cmp ? -ENODATA : 0;
 }
 
+static u32
+ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
+{
+	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
+}
+
+static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
+{
+	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
+	       ((u32)ea_inode->i_version);
+}
+
+static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
+{
+	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
+	ea_inode->i_version = (u32)ref_count;
+}
+
+static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
+{
+	return (u32)ea_inode->i_atime.tv_sec;
+}
+
+static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
+{
+	ea_inode->i_atime.tv_sec = hash;
+}
+
 /*
  * Read the EA value from an inode.
  */
 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 {
 	unsigned long block = 0;
-	struct buffer_head *bh = NULL;
+	struct buffer_head *bh;
 	int blocksize = ea_inode->i_sb->s_blocksize;
 	size_t csize, copied = 0;
+	void *copy_pos = buf;
 
 	while (copied < size) {
 		csize = (size - copied) > blocksize ? blocksize : size - copied;
@@ -298,10 +330,10 @@ static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 		if (!bh)
 			return -EFSCORRUPTED;
 
-		memcpy(buf, bh->b_data, csize);
+		memcpy(copy_pos, bh->b_data, csize);
 		brelse(bh);
 
-		buf += csize;
+		copy_pos += csize;
 		block += 1;
 		copied += csize;
 	}
@@ -317,29 +349,24 @@ static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 	inode = ext4_iget(parent->i_sb, ea_ino);
 	if (IS_ERR(inode)) {
 		err = PTR_ERR(inode);
-		ext4_error(parent->i_sb, "error while reading EA inode %lu "
-			   "err=%d", ea_ino, err);
+		ext4_error(parent->i_sb,
+			   "error while reading EA inode %lu err=%d", ea_ino,
+			   err);
 		return err;
 	}
 
 	if (is_bad_inode(inode)) {
-		ext4_error(parent->i_sb, "error while reading EA inode %lu "
-			   "is_bad_inode", ea_ino);
+		ext4_error(parent->i_sb,
+			   "error while reading EA inode %lu is_bad_inode",
+			   ea_ino);
 		err = -EIO;
 		goto error;
 	}
 
-	if (EXT4_XATTR_INODE_GET_PARENT(inode) != parent->i_ino ||
-	    inode->i_generation != parent->i_generation) {
-		ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
-			   "to parent is invalid.", ea_ino);
-		err = -EINVAL;
-		goto error;
-	}
-
 	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
-		ext4_error(parent->i_sb, "EA inode %lu does not have "
-			   "EXT4_EA_INODE_FL flag set.\n", ea_ino);
+		ext4_error(parent->i_sb,
+			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
+			    ea_ino);
 		err = -EINVAL;
 		goto error;
 	}
@@ -351,6 +378,20 @@ static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 	return err;
 }
 
+static int
+ext4_xattr_inode_verify_hash(struct inode *ea_inode, void *buffer, size_t size)
+{
+	u32 hash;
+
+	/* Verify stored hash matches calculated hash. */
+	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
+	if (hash != ext4_xattr_inode_get_hash(ea_inode))
+		return -EFSCORRUPTED;
+	return 0;
+}
+
+#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
+
 /*
  * Read the value from the EA inode.
  */
@@ -358,17 +399,53 @@ static int
 ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
 		     size_t size)
 {
+	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 	struct inode *ea_inode;
-	int ret;
+	int err;
 
-	ret = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
-	if (ret)
-		return ret;
+	err = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
+	if (err) {
+		ea_inode = NULL;
+		goto out;
+	}
 
-	ret = ext4_xattr_inode_read(ea_inode, buffer, size);
-	iput(ea_inode);
+	if (i_size_read(ea_inode) != size) {
+		ext4_warning_inode(ea_inode,
+				   "ea_inode file size=%llu entry size=%zu",
+				   i_size_read(ea_inode), size);
+		err = -EFSCORRUPTED;
+		goto out;
+	}
 
-	return ret;
+	err = ext4_xattr_inode_read(ea_inode, buffer, size);
+	if (err)
+		goto out;
+
+	err = ext4_xattr_inode_verify_hash(ea_inode, buffer, size);
+	/*
+	 * Compatibility check for old Lustre ea_inode implementation. Old
+	 * version does not have hash validation, but it has a backpointer
+	 * from ea_inode to the parent inode.
+	 */
+	if (err == -EFSCORRUPTED) {
+		if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != inode->i_ino ||
+		    ea_inode->i_generation != inode->i_generation) {
+			ext4_warning_inode(ea_inode,
+					   "EA inode hash validation failed");
+			goto out;
+		}
+		/* Do not add ea_inode to the cache. */
+		ea_inode_cache = NULL;
+	} else if (err)
+		goto out;
+
+	if (ea_inode_cache)
+		mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
+				      ext4_xattr_inode_get_hash(ea_inode),
+				      ea_inode->i_ino, true /* reusable */);
+out:
+	iput(ea_inode);
+	return err;
 }
 
 static int
@@ -656,6 +733,115 @@ static void ext4_xattr_update_super_block(handle_t *handle,
 	}
 }
 
+static inline size_t round_up_cluster(struct inode *inode, size_t length)
+{
+	struct super_block *sb = inode->i_sb;
+	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
+				    inode->i_blkbits);
+	size_t mask = ~(cluster_size - 1);
+
+	return (length + cluster_size - 1) & mask;
+}
+
+static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
+{
+	int err;
+
+	err = dquot_alloc_inode(inode);
+	if (err)
+		return err;
+	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
+	if (err)
+		dquot_free_inode(inode);
+	return err;
+}
+
+static void ext4_xattr_inode_free_quota(struct inode *inode, size_t len)
+{
+	dquot_free_space_nodirty(inode, round_up_cluster(inode, len));
+	dquot_free_inode(inode);
+}
+
+static int __ext4_xattr_set_credits(struct inode *inode,
+				    struct buffer_head *block_bh,
+				    size_t value_len)
+{
+	struct super_block *sb = inode->i_sb;
+	int credits;
+	int blocks;
+
+	/*
+	 * 1) Owner inode update
+	 * 2) Ref count update on old xattr block
+	 * 3) new xattr block
+	 * 4) block bitmap update for new xattr block
+	 * 5) group descriptor for new xattr block
+	 * 6) block bitmap update for old xattr block
+	 * 7) group descriptor for old block
+	 *
+	 * 6 & 7 can happen if we have two racing threads T_a and T_b
+	 * which are each trying to set an xattr on inodes I_a and I_b
+	 * which were both initially sharing an xattr block.
+	 */
+	credits = 7;
+
+	/* Quota updates. */
+	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
+
+	/*
+	 * In case of inline data, we may push out the data to a block,
+	 * so we need to reserve credits for this eventuality
+	 */
+	if (ext4_has_inline_data(inode))
+		credits += ext4_writepage_trans_blocks(inode) + 1;
+
+	/* We are done if ea_inode feature is not enabled. */
+	if (!ext4_has_feature_ea_inode(sb))
+		return credits;
+
+	/* New ea_inode, inode map, block bitmap, group descriptor. */
+	credits += 4;
+
+	/* Data blocks. */
+	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+
+	/* Indirection block or one level of extent tree. */
+	blocks += 1;
+
+	/* Block bitmap and group descriptor updates for each block. */
+	credits += blocks * 2;
+
+	/* Blocks themselves. */
+	credits += blocks;
+
+	/* Dereference ea_inode holding old xattr value.
+	 * Old ea_inode, inode map, block bitmap, group descriptor.
+	 */
+	credits += 4;
+
+	/* Data blocks for old ea_inode. */
+	blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
+
+	/* Indirection block or one level of extent tree for old ea_inode. */
+	blocks += 1;
+
+	/* Block bitmap and group descriptor updates for each block. */
+	credits += blocks * 2;
+
+	/* We may need to clone the existing xattr block in which case we need
+	 * to increment ref counts for existing ea_inodes referenced by it.
+	 */
+	if (block_bh) {
+		struct ext4_xattr_entry *entry = BFIRST(block_bh);
+
+		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
+			if (entry->e_value_inum)
+				/* Ref count update on ea_inode. */
+				credits += 1;
+	}
+	return credits;
+}
+
 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 				     int credits, struct buffer_head *bh,
 				     bool dirty, bool block_csum)
@@ -705,12 +891,140 @@ static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 	return 0;
 }
 
+static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
+				       int ref_change)
+{
+	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
+	struct ext4_iloc iloc;
+	s64 ref_count;
+	u32 hash;
+	int ret;
+
+	inode_lock(ea_inode);
+
+	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
+	if (ret) {
+		iloc.bh = NULL;
+		goto out;
+	}
+
+	ref_count = ext4_xattr_inode_get_ref(ea_inode);
+	ref_count += ref_change;
+	ext4_xattr_inode_set_ref(ea_inode, ref_count);
+
+	if (ref_change > 0) {
+		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
+			  ea_inode->i_ino, ref_count);
+
+		if (ref_count == 1) {
+			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
+				  ea_inode->i_ino, ea_inode->i_nlink);
+
+			set_nlink(ea_inode, 1);
+			ext4_orphan_del(handle, ea_inode);
+
+			hash = ext4_xattr_inode_get_hash(ea_inode);
+			mb_cache_entry_create(ea_inode_cache, GFP_NOFS, hash,
+					      ea_inode->i_ino,
+					      true /* reusable */);
+		}
+	} else {
+		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
+			  ea_inode->i_ino, ref_count);
+
+		if (ref_count == 0) {
+			WARN_ONCE(ea_inode->i_nlink != 1,
+				  "EA inode %lu i_nlink=%u",
+				  ea_inode->i_ino, ea_inode->i_nlink);
+
+			clear_nlink(ea_inode);
+			ext4_orphan_add(handle, ea_inode);
+
+			hash = ext4_xattr_inode_get_hash(ea_inode);
+			mb_cache_entry_delete(ea_inode_cache, hash,
+					      ea_inode->i_ino);
+		}
+	}
+
+	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
+	iloc.bh = NULL;
+	if (ret)
+		ext4_warning_inode(ea_inode,
+				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
+out:
+	brelse(iloc.bh);
+	inode_unlock(ea_inode);
+	return ret;
+}
+
+static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
+{
+	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
+}
+
+static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
+{
+	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
+}
+
+static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
+					struct ext4_xattr_entry *first)
+{
+	struct inode *ea_inode;
+	struct ext4_xattr_entry *entry;
+	struct ext4_xattr_entry *failed_entry;
+	unsigned int ea_ino;
+	int err, saved_err;
+
+	for (entry = first; !IS_LAST_ENTRY(entry);
+	     entry = EXT4_XATTR_NEXT(entry)) {
+		if (!entry->e_value_inum)
+			continue;
+		ea_ino = le32_to_cpu(entry->e_value_inum);
+		err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
+		if (err)
+			goto cleanup;
+		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
+		if (err) {
+			ext4_warning_inode(ea_inode, "inc ref error %d", err);
+			iput(ea_inode);
+			goto cleanup;
+		}
+		iput(ea_inode);
+	}
+	return 0;
+
+cleanup:
+	saved_err = err;
+	failed_entry = entry;
+
+	for (entry = first; entry != failed_entry;
+	     entry = EXT4_XATTR_NEXT(entry)) {
+		if (!entry->e_value_inum)
+			continue;
+		ea_ino = le32_to_cpu(entry->e_value_inum);
+		err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
+		if (err) {
+			ext4_warning(parent->i_sb,
+				     "cleanup ea_ino %u iget error %d", ea_ino,
+				     err);
+			continue;
+		}
+		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
+		if (err)
+			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
+					   err);
+		iput(ea_inode);
+	}
+	return saved_err;
+}
+
 static void
-ext4_xattr_inode_remove_all(handle_t *handle, struct inode *parent,
+ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
 			     struct buffer_head *bh,
 			     struct ext4_xattr_entry *first, bool block_csum,
 			     struct ext4_xattr_inode_array **ea_inode_array,
-			    int extra_credits)
+			     int extra_credits, bool skip_quota)
 {
 	struct inode *ea_inode;
 	struct ext4_xattr_entry *entry;
@@ -747,10 +1061,16 @@ ext4_xattr_inode_remove_all(handle_t *handle, struct inode *parent,
 			continue;
 		}
 
-		inode_lock(ea_inode);
-		clear_nlink(ea_inode);
-		ext4_orphan_add(handle, ea_inode);
-		inode_unlock(ea_inode);
+		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
+		if (err) {
+			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
+					   err);
+			continue;
+		}
+
+		if (!skip_quota)
+			ext4_xattr_inode_free_quota(parent,
+					      le32_to_cpu(entry->e_value_size));
 
 		/*
 		 * Forget about ea_inode within the same transaction that
@@ -784,7 +1104,9 @@ ext4_xattr_inode_remove_all(handle_t *handle, struct inode *parent,
  */
 static void
 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
-			 struct buffer_head *bh)
+			 struct buffer_head *bh,
+			 struct ext4_xattr_inode_array **ea_inode_array,
+			 int extra_credits)
 {
 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 	u32 hash, ref;
@@ -807,6 +1129,14 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
 		mb_cache_entry_delete(ea_block_cache, hash, bh->b_blocknr);
 		get_bh(bh);
 		unlock_buffer(bh);
+
+		if (ext4_has_feature_ea_inode(inode->i_sb))
+			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
+						     BFIRST(bh),
+						     true /* block_csum */,
+						     ea_inode_array,
+						     extra_credits,
+						     true /* skip_quota */);
 		ext4_free_blocks(handle, inode, bh, 0, 1,
 				 EXT4_FREE_BLOCKS_METADATA |
 				 EXT4_FREE_BLOCKS_FORGET);
@@ -878,8 +1208,8 @@ static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
 {
 	struct buffer_head *bh = NULL;
 	unsigned long block = 0;
-	unsigned blocksize = ea_inode->i_sb->s_blocksize;
-	unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
+	int blocksize = ea_inode->i_sb->s_blocksize;
+	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
 	int csize, wsize = 0;
 	int ret = 0;
 	int retries = 0;
@@ -945,7 +1275,7 @@ static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
  * Create an inode to store the value of a large EA.
  */
 static struct inode *ext4_xattr_inode_create(handle_t *handle,
-					     struct inode *inode)
+					     struct inode *inode, u32 hash)
 {
 	struct inode *ea_inode = NULL;
 	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
@@ -963,67 +1293,115 @@ static struct inode *ext4_xattr_inode_create(handle_t *handle,
 		ea_inode->i_fop = &ext4_file_operations;
 		ext4_set_aops(ea_inode);
 		ext4_xattr_inode_set_class(ea_inode);
-		ea_inode->i_generation = inode->i_generation;
-		EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
-
-		/*
-		 * A back-pointer from EA inode to parent inode will be useful
-		 * for e2fsck.
-		 */
-		EXT4_XATTR_INODE_SET_PARENT(ea_inode, inode->i_ino);
 		unlock_new_inode(ea_inode);
+		ext4_xattr_inode_set_ref(ea_inode, 1);
+		ext4_xattr_inode_set_hash(ea_inode, hash);
+		err = ext4_mark_inode_dirty(handle, ea_inode);
+		if (!err)
 			err = ext4_inode_attach_jinode(ea_inode);
 		if (err) {
 			iput(ea_inode);
 			return ERR_PTR(err);
 		}
+
+		/*
+		 * Xattr inodes are shared therefore quota charging is performed
+		 * at a higher level.
+		 */
+		dquot_free_inode(ea_inode);
+		dquot_drop(ea_inode);
+		inode_lock(ea_inode);
+		ea_inode->i_flags |= S_NOQUOTA;
+		inode_unlock(ea_inode);
 	}
 
 	return ea_inode;
 }
 
-/*
- * Unlink the inode storing the value of the EA.
- */
-int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
+static struct inode *
+ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
+			    size_t value_len, u32 hash)
 {
-	struct inode *ea_inode = NULL;
-	int err;
+	struct inode *ea_inode;
+	struct mb_cache_entry *ce;
+	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
+	void *ea_data;
 
-	err = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
-	if (err)
-		return err;
+	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
+	if (!ce)
+		return NULL;
 
-	clear_nlink(ea_inode);
-	iput(ea_inode);
+	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
+	if (!ea_data) {
+		mb_cache_entry_put(ea_inode_cache, ce);
+		return NULL;
+	}
 
-	return 0;
+	while (ce) {
+		ea_inode = ext4_iget(inode->i_sb, ce->e_value);
+		if (!IS_ERR(ea_inode) &&
+		    !is_bad_inode(ea_inode) &&
+		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
+		    i_size_read(ea_inode) == value_len &&
+		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
+		    !ext4_xattr_inode_verify_hash(ea_inode, ea_data,
+						  value_len) &&
+		    !memcmp(value, ea_data, value_len)) {
+			mb_cache_entry_touch(ea_inode_cache, ce);
+			mb_cache_entry_put(ea_inode_cache, ce);
+			kvfree(ea_data);
+			return ea_inode;
+		}
+
+		if (!IS_ERR(ea_inode))
+			iput(ea_inode);
+		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
+	}
+	kvfree(ea_data);
+	return NULL;
 }
 
 /*
  * Add value of the EA in an inode.
  */
-static int ext4_xattr_inode_set(handle_t *handle, struct inode *inode,
-				unsigned long *ea_ino, const void *value,
-				size_t value_len)
+static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
+					  const void *value, size_t value_len,
+					  struct inode **ret_inode)
 {
 	struct inode *ea_inode;
+	u32 hash;
 	int err;
 
+	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
+	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
+	if (ea_inode) {
+		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
+		if (err) {
+			iput(ea_inode);
+			return err;
+		}
+
+		*ret_inode = ea_inode;
+		return 0;
+	}
+
 	/* Create an inode for the EA value */
-	ea_inode = ext4_xattr_inode_create(handle, inode);
+	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
 	if (IS_ERR(ea_inode))
 		return PTR_ERR(ea_inode);
 
 	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
-	if (err)
-		clear_nlink(ea_inode);
-	else
-		*ea_ino = ea_inode->i_ino;
-
+	if (err) {
+		ext4_xattr_inode_dec_ref(handle, ea_inode);
 		iput(ea_inode);
-
 		return err;
+	}
+
+	mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
+			      ea_inode->i_ino, true /* reusable */);
+
+	*ret_inode = ea_inode;
+	return 0;
 }
 
 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
@@ -1031,9 +1409,37 @@ static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
 				handle_t *handle, struct inode *inode)
 {
 	struct ext4_xattr_entry *last;
-	size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
+	struct ext4_xattr_entry *here = s->here;
+	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
 	int in_inode = i->in_inode;
-	int rc;
+	struct inode *old_ea_inode = NULL;
+	struct inode *new_ea_inode = NULL;
+	size_t old_size, new_size;
+	int ret;
+
+	/* Space used by old and new values. */
+	old_size = (!s->not_found && !here->e_value_inum) ?
+			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
+	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
+
+	/*
+	 * Optimization for the simple case when old and new values have the
+	 * same padded sizes. Not applicable if external inodes are involved.
+	 */
+	if (new_size && new_size == old_size) {
+		size_t offs = le16_to_cpu(here->e_value_offs);
+		void *val = s->base + offs;
+
+		here->e_value_size = cpu_to_le32(i->value_len);
+		if (i->value == EXT4_ZERO_XATTR_VALUE) {
+			memset(val, 0, new_size);
+		} else {
+			memcpy(val, i->value, i->value_len);
+			/* Clear padding bytes. */
+			memset(val + i->value_len, 0, new_size - i->value_len);
+		}
+		return 0;
+	}
 
 	/* Compute min_offs and last. */
 	last = s->first;
@@ -1044,122 +1450,148 @@ static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
 				min_offs = offs;
 		}
 	}
+
+	/* Check whether we have enough space. */
+	if (i->value) {
+		size_t free;
+
 		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
-	if (!s->not_found) {
-		if (!in_inode &&
-		    !s->here->e_value_inum && s->here->e_value_size) {
-			size_t size = le32_to_cpu(s->here->e_value_size);
-			free += EXT4_XATTR_SIZE(size);
+		if (!s->not_found)
+			free += EXT4_XATTR_LEN(name_len) + old_size;
+
+		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
+			ret = -ENOSPC;
+			goto out;
 		}
-		free += EXT4_XATTR_LEN(name_len);
 	}
-	if (i->value) {
-		size_t value_len = EXT4_XATTR_SIZE(i->value_len);
 
-		if (in_inode)
-			value_len = 0;
+	/*
+	 * Getting access to old and new ea inodes is subject to failures.
+	 * Finish that work before doing any modifications to the xattr data.
+	 */
+	if (!s->not_found && here->e_value_inum) {
+		ret = ext4_xattr_inode_iget(inode,
+					    le32_to_cpu(here->e_value_inum),
+					    &old_ea_inode);
+		if (ret) {
+			old_ea_inode = NULL;
+			goto out;
+		}
+	}
+	if (i->value && in_inode) {
+		WARN_ON_ONCE(!i->value_len);
 
-		if (free < EXT4_XATTR_LEN(name_len) + value_len)
-			return -ENOSPC;
+		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
+		if (ret)
+			goto out;
+
+		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
+						     i->value_len,
+						     &new_ea_inode);
+		if (ret) {
+			new_ea_inode = NULL;
+			ext4_xattr_inode_free_quota(inode, i->value_len);
+			goto out;
+		}
 	}
 
-	if (i->value && s->not_found) {
-		/* Insert the new name. */
-		size_t size = EXT4_XATTR_LEN(name_len);
-		size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
-		memmove((void *)s->here + size, s->here, rest);
-		memset(s->here, 0, size);
-		s->here->e_name_index = i->name_index;
-		s->here->e_name_len = name_len;
-		memcpy(s->here->e_name, i->name, name_len);
-	} else {
-		if (!s->here->e_value_inum && s->here->e_value_size &&
-		    s->here->e_value_offs > 0) {
-			void *first_val = s->base + min_offs;
-			size_t offs = le16_to_cpu(s->here->e_value_offs);
-			void *val = s->base + offs;
-			size_t size = EXT4_XATTR_SIZE(
-				le32_to_cpu(s->here->e_value_size));
-
-			if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
-				/* The old and the new value have the same
-				   size. Just replace. */
-				s->here->e_value_size =
-					cpu_to_le32(i->value_len);
-				if (i->value == EXT4_ZERO_XATTR_VALUE) {
-					memset(val, 0, size);
-				} else {
-					/* Clear pad bytes first. */
-					memset(val + size - EXT4_XATTR_PAD, 0,
-					       EXT4_XATTR_PAD);
-					memcpy(val, i->value, i->value_len);
+	if (old_ea_inode) {
+		/* We are ready to release ref count on the old_ea_inode. */
+		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
+		if (ret) {
+			/* Release newly required ref count on new_ea_inode. */
+			if (new_ea_inode) {
+				int err;
+
+				err = ext4_xattr_inode_dec_ref(handle,
+							       new_ea_inode);
+				if (err)
+					ext4_warning_inode(new_ea_inode,
+						  "dec ref new_ea_inode err=%d",
+						  err);
+				ext4_xattr_inode_free_quota(inode,
+							    i->value_len);
 			}
-				return 0;
+			goto out;
 		}
 
+		ext4_xattr_inode_free_quota(inode,
+					    le32_to_cpu(here->e_value_size));
+	}
+
+	/* No failures allowed past this point. */
+
+	if (!s->not_found && here->e_value_offs) {
 		/* Remove the old value. */
-			memmove(first_val + size, first_val, val - first_val);
-			memset(first_val, 0, size);
-			s->here->e_value_size = 0;
-			s->here->e_value_offs = 0;
-			min_offs += size;
+		void *first_val = s->base + min_offs;
+		size_t offs = le16_to_cpu(here->e_value_offs);
+		void *val = s->base + offs;
+
+		memmove(first_val + old_size, first_val, val - first_val);
+		memset(first_val, 0, old_size);
+		min_offs += old_size;
 
 		/* Adjust all value offsets. */
 		last = s->first;
 		while (!IS_LAST_ENTRY(last)) {
 			size_t o = le16_to_cpu(last->e_value_offs);
+
 			if (!last->e_value_inum &&
 			    last->e_value_size && o < offs)
-					last->e_value_offs =
-						cpu_to_le16(o + size);
+				last->e_value_offs = cpu_to_le16(o + old_size);
 			last = EXT4_XATTR_NEXT(last);
 		}
 	}
-		if (s->here->e_value_inum) {
-			ext4_xattr_inode_unlink(inode,
-					    le32_to_cpu(s->here->e_value_inum));
-			s->here->e_value_inum = 0;
-		}
+
 	if (!i->value) {
-			/* Remove the old name. */
+		/* Remove old name. */
 		size_t size = EXT4_XATTR_LEN(name_len);
+
 		last = ENTRY((void *)last - size);
-			memmove(s->here, (void *)s->here + size,
-				(void *)last - (void *)s->here + sizeof(__u32));
+		memmove(here, (void *)here + size,
+			(void *)last - (void *)here + sizeof(__u32));
 		memset(last, 0, size);
-		}
+	} else if (s->not_found) {
+		/* Insert new name. */
+		size_t size = EXT4_XATTR_LEN(name_len);
+		size_t rest = (void *)last - (void *)here + sizeof(__u32);
+
+		memmove((void *)here + size, here, rest);
+		memset(here, 0, size);
+		here->e_name_index = i->name_index;
+		here->e_name_len = name_len;
+		memcpy(here->e_name, i->name, name_len);
+	} else {
+		/* This is an update, reset value info. */
+		here->e_value_inum = 0;
+		here->e_value_offs = 0;
+		here->e_value_size = 0;
 	}
 
 	if (i->value) {
-		/* Insert the new value. */
+		/* Insert new value. */
 		if (in_inode) {
-			unsigned long ea_ino =
-				le32_to_cpu(s->here->e_value_inum);
-			rc = ext4_xattr_inode_set(handle, inode, &ea_ino,
-						  i->value, i->value_len);
-			if (rc)
-				goto out;
-			s->here->e_value_inum = cpu_to_le32(ea_ino);
-			s->here->e_value_offs = 0;
+			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
 		} else if (i->value_len) {
-			size_t size = EXT4_XATTR_SIZE(i->value_len);
-			void *val = s->base + min_offs - size;
-			s->here->e_value_offs = cpu_to_le16(min_offs - size);
-			s->here->e_value_inum = 0;
+			void *val = s->base + min_offs - new_size;
+
+			here->e_value_offs = cpu_to_le16(min_offs - new_size);
 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
-				memset(val, 0, size);
+				memset(val, 0, new_size);
 			} else {
-				/* Clear the pad bytes first. */
-				memset(val + size - EXT4_XATTR_PAD, 0,
-				       EXT4_XATTR_PAD);
 				memcpy(val, i->value, i->value_len);
+				/* Clear padding bytes. */
+				memset(val + i->value_len, 0,
+				       new_size - i->value_len);
 			}
 		}
-		s->here->e_value_size = cpu_to_le32(i->value_len);
+		here->e_value_size = cpu_to_le32(i->value_len);
 	}
-
+	ret = 0;
 out:
-	return rc;
+	iput(old_ea_inode);
+	iput(new_ea_inode);
+	return ret;
 }
 
 struct ext4_xattr_block_find {
@@ -1221,6 +1653,8 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 	struct mb_cache_entry *ce = NULL;
 	int error = 0;
 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
+	struct inode *ea_inode = NULL;
+	size_t old_ea_inode_size = 0;
 
 #define header(x) ((struct ext4_xattr_header *)(x))
 
@@ -1275,6 +1709,24 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 			header(s->base)->h_refcount = cpu_to_le32(1);
 			s->here = ENTRY(s->base + offset);
 			s->end = s->base + bs->bh->b_size;
+
+			/*
+			 * If existing entry points to an xattr inode, we need
+			 * to prevent ext4_xattr_set_entry() from decrementing
+			 * ref count on it because the reference belongs to the
+			 * original block. In this case, make the entry look
+			 * like it has an empty value.
+			 */
+			if (!s->not_found && s->here->e_value_inum) {
+				/*
+				 * Defer quota free call for previous inode
+				 * until success is guaranteed.
+				 */
+				old_ea_inode_size = le32_to_cpu(
+							s->here->e_value_size);
+				s->here->e_value_inum = 0;
+				s->here->e_value_size = 0;
+			}
 		}
 	} else {
 		/* Allocate a buffer where we construct the new block. */
@@ -1296,6 +1748,24 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 		goto bad_block;
 	if (error)
 		goto cleanup;
+
+	if (i->value && s->here->e_value_inum) {
+		unsigned int ea_ino;
+
+		/*
+		 * A ref count on ea_inode has been taken as part of the call to
+		 * ext4_xattr_set_entry() above. We would like to drop this
+		 * extra ref but we have to wait until the xattr block is
+		 * initialized and has its own ref count on the ea_inode.
+		 */
+		ea_ino = le32_to_cpu(s->here->e_value_inum);
+		error = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
+		if (error) {
+			ea_inode = NULL;
+			goto cleanup;
+		}
+	}
+
 	if (!IS_LAST_ENTRY(s->first))
 		ext4_xattr_rehash(header(s->base), s->here);
 
@@ -1406,6 +1876,22 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 						 EXT4_FREE_BLOCKS_METADATA);
 				goto cleanup;
 			}
+			error = ext4_xattr_inode_inc_ref_all(handle, inode,
+						      ENTRY(header(s->base)+1));
+			if (error)
+				goto getblk_failed;
+			if (ea_inode) {
+				/* Drop the extra ref on ea_inode. */
+				error = ext4_xattr_inode_dec_ref(handle,
+								 ea_inode);
+				if (error)
+					ext4_warning_inode(ea_inode,
+							   "dec ref error=%d",
+							   error);
+				iput(ea_inode);
+				ea_inode = NULL;
+			}
+
 			lock_buffer(new_bh);
 			error = ext4_journal_get_create_access(handle, new_bh);
 			if (error) {
@@ -1425,15 +1911,38 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
 		}
 	}
 
+	if (old_ea_inode_size)
+		ext4_xattr_inode_free_quota(inode, old_ea_inode_size);
+
 	/* Update the inode. */
 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
 
 	/* Drop the previous xattr block. */
-	if (bs->bh && bs->bh != new_bh)
-		ext4_xattr_release_block(handle, inode, bs->bh);
+	if (bs->bh && bs->bh != new_bh) {
+		struct ext4_xattr_inode_array *ea_inode_array = NULL;
+
+		ext4_xattr_release_block(handle, inode, bs->bh,
+					 &ea_inode_array,
+					 0 /* extra_credits */);
+		ext4_xattr_inode_array_free(ea_inode_array);
+	}
 	error = 0;
 
 cleanup:
+	if (ea_inode) {
+		int error2;
+
+		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
+		if (error2)
+			ext4_warning_inode(ea_inode, "dec ref error=%d",
+					   error2);
+
+		/* If there was an error, revert the quota charge. */
+		if (error)
+			ext4_xattr_inode_free_quota(inode,
+						    i_size_read(ea_inode));
+		iput(ea_inode);
+	}
 	if (ce)
 		mb_cache_entry_put(ea_block_cache, ce);
 	brelse(new_bh);
@@ -1558,6 +2067,22 @@ static int ext4_xattr_value_same(struct ext4_xattr_search *s,
 	return !memcmp(value, i->value, i->value_len);
 }
 
+static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
+{
+	struct buffer_head *bh;
+	int error;
+
+	if (!EXT4_I(inode)->i_file_acl)
+		return NULL;
+	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
+	if (!bh)
+		return ERR_PTR(-EIO);
+	error = ext4_xattr_check_block(inode, bh);
+	if (error)
+		return ERR_PTR(error);
+	return bh;
+}
+
 /*
  * ext4_xattr_set_handle()
  *
@@ -1600,9 +2125,18 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 
 	/* Check journal credits under write lock. */
 	if (ext4_handle_valid(handle)) {
+		struct buffer_head *bh;
 		int credits;
 
-		credits = ext4_xattr_set_credits(inode, value_len);
+		bh = ext4_xattr_get_block(inode);
+		if (IS_ERR(bh)) {
+			error = PTR_ERR(bh);
+			goto cleanup;
+		}
+
+		credits = __ext4_xattr_set_credits(inode, bh, value_len);
+		brelse(bh);
+
 		if (!ext4_handle_has_enough_credits(handle, credits)) {
 			error = -ENOSPC;
 			goto cleanup;
@@ -1638,6 +2172,7 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 		if (flags & XATTR_CREATE)
 			goto cleanup;
 	}
+
 	if (!value) {
 		if (!is.s.not_found)
 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
@@ -1706,34 +2241,29 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	return error;
 }
 
-int ext4_xattr_set_credits(struct inode *inode, size_t value_len)
+int ext4_xattr_set_credits(struct inode *inode, size_t value_len, int *credits)
 {
-	struct super_block *sb = inode->i_sb;
-	int credits;
-
-	if (!EXT4_SB(sb)->s_journal)
-		return 0;
-
-	credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb);
+	struct buffer_head *bh;
+	int err;
 
-	/*
-	 * In case of inline data, we may push out the data to a block,
-	 * so we need to reserve credits for this eventuality
-	 */
-	if (ext4_has_inline_data(inode))
-		credits += ext4_writepage_trans_blocks(inode) + 1;
+	*credits = 0;
 
-	if (ext4_has_feature_ea_inode(sb)) {
-		int nrblocks = (value_len + sb->s_blocksize - 1) >>
-					sb->s_blocksize_bits;
+	if (!EXT4_SB(inode->i_sb)->s_journal)
+		return 0;
 
-		/* For new inode */
-		credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
+	down_read(&EXT4_I(inode)->xattr_sem);
 
-		/* For data blocks of EA inode */
-		credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
+	bh = ext4_xattr_get_block(inode);
+	if (IS_ERR(bh)) {
+		err = PTR_ERR(bh);
+	} else {
+		*credits = __ext4_xattr_set_credits(inode, bh, value_len);
+		brelse(bh);
+		err = 0;
 	}
-	return credits;
+
+	up_read(&EXT4_I(inode)->xattr_sem);
+	return err;
 }
 
 /*
@@ -1758,7 +2288,10 @@ ext4_xattr_set(struct inode *inode, int name_index, const char *name,
 		return error;
 
 retry:
-	credits = ext4_xattr_set_credits(inode, value_len);
+	error = ext4_xattr_set_credits(inode, value_len, &credits);
+	if (error)
+		return error;
+
 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
 	if (IS_ERR(handle)) {
 		error = PTR_ERR(handle);
@@ -2064,10 +2597,10 @@ int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
 	return error;
 }
 
-
 #define EIA_INCR 16 /* must be 2^n */
 #define EIA_MASK (EIA_INCR - 1)
-/* Add the large xattr @inode into @ea_inode_array for later deletion.
+
+/* Add the large xattr @inode into @ea_inode_array for deferred iput().
  * If @ea_inode_array is new or full it will be grown and the old
  * contents copied over.
  */
@@ -2112,21 +2645,19 @@ ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
  * ext4_xattr_delete_inode()
  *
  * Free extended attribute resources associated with this inode. Traverse
- * all entries and unlink any xattr inodes associated with this inode. This
- * is called immediately before an inode is freed. We have exclusive
- * access to the inode. If an orphan inode is deleted it will also delete any
- * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
- * to ensure they belong to the parent inode and were not deleted already.
+ * all entries and decrement reference on any xattr inodes associated with this
+ * inode. This is called immediately before an inode is freed. We have exclusive
+ * access to the inode. If an orphan inode is deleted it will also release its
+ * references on xattr block and xattr inodes.
  */
-int
-ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
+int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 			    struct ext4_xattr_inode_array **ea_inode_array,
 			    int extra_credits)
 {
 	struct buffer_head *bh = NULL;
 	struct ext4_xattr_ibody_header *header;
-	struct ext4_inode *raw_inode;
 	struct ext4_iloc iloc = { .bh = NULL };
+	struct ext4_xattr_entry *entry;
 	int error;
 
 	error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
@@ -2138,28 +2669,33 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 		goto cleanup;
 	}
 
-	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
-		goto delete_external_ea;
+	if (ext4_has_feature_ea_inode(inode->i_sb) &&
+	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 
 		error = ext4_get_inode_loc(inode, &iloc);
-	if (error)
+		if (error) {
+			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
 			goto cleanup;
+		}
 
 		error = ext4_journal_get_write_access(handle, iloc.bh);
-	if (error)
+		if (error) {
+			EXT4_ERROR_INODE(inode, "write access (error %d)",
+					 error);
 			goto cleanup;
+		}
 
-	raw_inode = ext4_raw_inode(&iloc);
-	header = IHDR(inode, raw_inode);
-	ext4_xattr_inode_remove_all(handle, inode, iloc.bh, IFIRST(header),
-				    false /* block_csum */, ea_inode_array,
-				    extra_credits);
-
-delete_external_ea:
-	if (!EXT4_I(inode)->i_file_acl) {
-		error = 0;
-		goto cleanup;
+		header = IHDR(inode, ext4_raw_inode(&iloc));
+		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
+			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
+						     IFIRST(header),
+						     false /* block_csum */,
+						     ea_inode_array,
+						     extra_credits,
+						     false /* skip_quota */);
 	}
+
+	if (EXT4_I(inode)->i_file_acl) {
 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 		if (!bh) {
 			EXT4_ERROR_INODE(inode, "block %llu read error",
@@ -2167,30 +2703,28 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 			error = -EIO;
 			goto cleanup;
 		}
-	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
-	    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
-		EXT4_ERROR_INODE(inode, "bad block %llu",
-				 EXT4_I(inode)->i_file_acl);
-		error = -EFSCORRUPTED;
+		error = ext4_xattr_check_block(inode, bh);
+		if (error) {
+			EXT4_ERROR_INODE(inode, "bad block %llu (error %d)",
+					 EXT4_I(inode)->i_file_acl, error);
 			goto cleanup;
 		}
 
 		if (ext4_has_feature_ea_inode(inode->i_sb)) {
-		error = ext4_journal_get_write_access(handle, bh);
-		if (error) {
-			EXT4_ERROR_INODE(inode, "write access %llu",
-					 EXT4_I(inode)->i_file_acl);
-			goto cleanup;
-		}
-		ext4_xattr_inode_remove_all(handle, inode, bh,
-					    BFIRST(bh),
-					    true /* block_csum */,
-					    ea_inode_array,
-					    extra_credits);
+			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
+			     entry = EXT4_XATTR_NEXT(entry))
+				if (entry->e_value_inum)
+					ext4_xattr_inode_free_quota(inode,
+					      le32_to_cpu(entry->e_value_size));
+
 		}
 
-	ext4_xattr_release_block(handle, inode, bh);
-	/* Update i_file_acl within the same transaction that releases block. */
+		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
+					 extra_credits);
+		/*
+		 * Update i_file_acl value in the same transaction that releases
+		 * block.
+		 */
 		EXT4_I(inode)->i_file_acl = 0;
 		error = ext4_mark_inode_dirty(handle, inode);
 		if (error) {
@@ -2198,6 +2732,8 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 					 error);
 			goto cleanup;
 		}
+	}
+	error = 0;
 cleanup:
 	brelse(iloc.bh);
 	brelse(bh);
@@ -2206,17 +2742,13 @@ ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 
 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
 {
-	struct inode	*ea_inode;
-	int		idx = 0;
+	int idx;
 
 	if (ea_inode_array == NULL)
 		return;
 
-	for (; idx < ea_inode_array->count; ++idx) {
-		ea_inode = ea_inode_array->inodes[idx];
-		clear_nlink(ea_inode);
-		iput(ea_inode);
-	}
+	for (idx = 0; idx < ea_inode_array->count; ++idx)
+		iput(ea_inode_array->inodes[idx]);
 	kfree(ea_inode_array);
 }
 

fs/ext4/xattr.h

@@ -69,19 +69,6 @@ struct ext4_xattr_entry {
 		EXT4_I(inode)->i_extra_isize))
 #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
 
-/*
- * Link EA inode back to parent one using i_mtime field.
- * Extra integer type conversion added to ignore higher
- * bits in i_mtime.tv_sec which might be set by ext4_get()
- */
-#define EXT4_XATTR_INODE_SET_PARENT(inode, inum)      \
-do {                                                  \
-      (inode)->i_mtime.tv_sec = inum;                 \
-} while(0)
-
-#define EXT4_XATTR_INODE_GET_PARENT(inode)            \
-((__u32)(inode)->i_mtime.tv_sec)
-
 /*
  * The minimum size of EA value when you start storing it in an external inode
  * size of block - size of header - size of 1 entry - 4 null bytes
@@ -165,9 +152,9 @@ extern ssize_t ext4_listxattr(struct dentry *, char *, size_t);
 extern int ext4_xattr_get(struct inode *, int, const char *, void *, size_t);
 extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
 extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
-extern int ext4_xattr_set_credits(struct inode *inode, size_t value_len);
+extern int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
+				  int *credits);
 
-extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
 extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 				   struct ext4_xattr_inode_array **array,
 				   int extra_credits);

fs/mbcache.c

@@ -13,10 +13,11 @@
  * mb_cache_entry_delete()).
  *
  * Ext2 and ext4 use this cache for deduplication of extended attribute blocks.
- * They use hash of a block contents as a key and block number as a value.
- * That's why keys need not be unique (different xattr blocks may end up having
- * the same hash). However block number always uniquely identifies a cache
- * entry.
+ * Ext4 also uses it for deduplication of xattr values stored in inodes.
+ * They use hash of data as a key and provide a value that may represent a
+ * block or inode number. That's why keys need not be unique (hash of different
+ * data may be the same). However user provided value always uniquely
+ * identifies a cache entry.
  *
  * We provide functions for creation and removal of entries, search by key,
  * and a special "delete entry with given key-value pair" operation. Fixed