Re: [PATCH v5 27/28] ext4: xattr inode deduplication
From: Tahsin Erdogan
Date: Tue Jun 20 2017 - 05:50:06 EST
Thanks Andreas for the feedback. Please see my responses below:
> It would be preferable to allow a mount option like "no_mbcache" to disable
> the use of shared xattrs. In the Lustre case at least, there will never be
> shared large xattrs, and we've had a bunch of performance issues with mbcache
> due to lock contention among many server threads doing concurrent lookups and
> inserting many thousands of unique entries into the cache.
I have put nombcache mount option in a separate patch ("[PATCH 32/32]
ext4: add nombcache mount option"). I have named it nombcache instead
of no_mbcache to be consistent with other no* options. Let me know if
you prefer no_mbcache as the option name.
> This should follow the existing convention of always using s_csum_seed to seed
> the checksum, and change ext4_fill_super() to initialize s_csum_seed to ~0 if
> ext4_has_metadata_csum() is false, or always use the same value regardless of
> whether ext4_has_metadata_csum() is set or not.
Done.
> If it really necessary to have more than 2^32 references on a single shared
> inode then it would be better to avoid the re-use of i_mtime, which breaks
> the backref for unshared xattrs, and using i_size isn't enough of a guarantee
> that this is the correct parent inode in case of on-disk corruption.
I have now moved the lower 32bits of ref count from i_mtime to l_i_version.
> Should this be contingent on ext4_has_metadata_csum() feature being enabled, or
> alternately check if EXT4_XATTR_INODE_GET_PARENT() and i_generation match before
> returning an error. This will allow a smooth transition from existing filesystems
> that do not store the hash, but have only a single-use xattr inode with a parent
> backref.
I updated hash validation to fallback to parent backref check for
backward compatibility.
>> + /* Indirection block. */
>> + blocks += 1;
>
> Strictly speaking, this is only needed "if (blocks > EXT4_NDIR_BLOCKS)".
Ack. I didn't think it was worth going through exact calculation in
this case, let me know if you see value in doing that
I also updated the comment to mention extents.
>> + /* 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.
>> + */
>
> Just to clarify here, in the case of cloning an existing xattr block, are the
> refcounts being _incremented_ or _decremented_ on the existing ea_inodes? I'm
> trying to figure out if we really need to have credits for both old and new
> xattr inodes, as well as these additional credits. Since this is reserving
> about 110 blocks for every setxattr, this can add significant pressure on the
> journal if there are lots of threads creating files and/or setting xattrs.
Cloning causes incrementing xattr inode references.
>> +ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
>> + size_t value_len, u32 hash)
>> {
>> + struct inode *ea_inode;
>> + struct mb_cache_entry *ce;
>> + struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
>> + void *ea_data = NULL;
>> int err;
>
> This function should just return NULL if ea_inode_cache is NULL (e.g. in
> the case of "no_mbcache" mount option).
Done in later patch ("[PATCH 32/32] ext4: add nombcache mount option")
> Should skip mb_cache if EA_INODE_CACHE(inode) is NULL, or have a wrapper
> like ext4_xattr_inode_cache_insert() to match ext4_xattr_inode_cache_find()
> that does the same.
Added skip in patch ("[PATCH 32/32] ext4: add nombcache mount option")
On Tue, Jun 20, 2017 at 2:07 AM, Tahsin Erdogan <tahsin@xxxxxxxxxx> wrote:
> 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.
>
> Signed-off-by: Tahsin Erdogan <tahsin@xxxxxxxxxx>
> ---
> v5:
> - made ext4_meta_trans_blocks() static again since there are no
> remaining users outside of inode.c
> - initialize sbi->s_csum_seed when ea_inode feature is enabled
> - use l_i_version to hold lower 32 bits of the xattr ref count.
> This avoids clashes with old implementations which use i_mtime.
> Since l_i_version is not available in HURD_COMPAT mode, fail mount
> request when both ea_inode feature and HURD_COMPAT are set.
> - when hash validation fails, fall back to old implementation
> which has a backref to parent.
> - fixed checkpatch.pl warning about using unsigned alone
>
> v4:
> - eliminated xattr entry in the xattr inode to avoid complexity and
> recursion in xattr update path. Now the ref count and hash are stored
> in i_[c/m/a]time.tv_sec fields.
> - some clean up in ext4_xattr_set_entry() to reduce code duplication and
> complexity
>
> v3:
> - use s_csum_seed for hash calculations when available
> - return error on stored vs calculated hash mismatch
>
> v2:
> - make dependency on crc32c dynamic
> - update ext4_has_metadata_csum() and ext4_has_group_desc_csum() so that
> they do not misinterpret existence of EXT4_SB(sb)->s_chksum_driver
>
> fs/ext4/acl.c | 5 +-
> fs/ext4/ext4.h | 23 +-
> fs/ext4/inode.c | 13 +-
> fs/ext4/super.c | 37 +-
> fs/ext4/xattr.c | 1038 +++++++++++++++++++++++++++++++++++++++++--------------
> fs/ext4/xattr.h | 17 +-
> fs/mbcache.c | 9 +-
> 7 files changed, 848 insertions(+), 294 deletions(-)
>
> diff --git a/fs/ext4/acl.c b/fs/ext4/acl.c
> index 74f7ac539e00..8db03e5c78bc 100644
> --- a/fs/ext4/acl.c
> +++ b/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);
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index d79d8d7bee88..59e9488c4876 100644
> --- a/fs/ext4/ext4.h
> +++ b/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_mb_cache;
> + struct mb_cache *s_ea_inode_cache;
> spinlock_t s_es_lock ____cacheline_aligned_in_smp;
>
> /* Ratelimit ext4 messages. */
> @@ -2099,7 +2100,11 @@ 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
> @@ -2482,7 +2487,6 @@ 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,
> @@ -2709,19 +2713,20 @@ 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) |
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index cd007f9757d1..ea95bd9eab81 100644
> --- a/fs/ext4/inode.c
> +++ b/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);
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index b02a23ec92ca..2bfacd737bb6 100644
> --- a/fs/ext4/super.c
> +++ b/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_mb_cache) {
> ext4_xattr_destroy_cache(sbi->s_mb_cache);
> sbi->s_mb_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 an s_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_mb_cache) {
> ext4_xattr_destroy_cache(sbi->s_mb_cache);
> sbi->s_mb_cache = NULL;
> diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
> index 0484df8dadd1..d7e60358ec91 100644
> --- a/fs/ext4/xattr.c
> +++ b/fs/ext4/xattr.c
> @@ -108,6 +108,9 @@ const struct xattr_handler *ext4_xattr_handlers[] = {
> #define EXT4_GET_MB_CACHE(inode) (((struct ext4_sb_info *) \
> inode->i_sb->s_fs_info)->s_mb_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
> @@ -657,6 +734,101 @@ 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 super_block *sb,
> + struct buffer_head *block_bh,
> + size_t value_len)
> +{
> + 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
> + */
> + credits = 5;
> +
> + /* 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;
> +
> + /* Quota updates. */
> + credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
> +
> + /* 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)
> @@ -706,12 +878,140 @@ 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,
> - struct buffer_head *bh,
> - struct ext4_xattr_entry *first, bool block_csum,
> - struct ext4_xattr_inode_array **ea_inode_array,
> - int extra_credits)
> +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, bool skip_quota)
> {
> struct inode *ea_inode;
> struct ext4_xattr_entry *entry;
> @@ -748,10 +1048,16 @@ 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
> @@ -785,7 +1091,9 @@ 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 *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
> u32 hash, ref;
> @@ -808,6 +1116,14 @@ ext4_xattr_release_block(handle_t *handle, struct inode *inode,
> mb_cache_entry_delete(ext4_mb_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);
> @@ -879,8 +1195,8 @@ 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;
> @@ -948,7 +1264,7 @@ 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) };
> @@ -966,67 +1282,115 @@ 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);
> - err = ext4_inode_attach_jinode(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;
> + }
>
> - iput(ea_inode);
> + mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
> + ea_inode->i_ino, true /* reusable */);
>
> - return err;
> + *ret_inode = ea_inode;
> + return 0;
> }
>
> static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
> @@ -1034,9 +1398,37 @@ 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;
> @@ -1047,122 +1439,148 @@ static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
> min_offs = offs;
> }
> }
> - 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);
> - }
> - free += EXT4_XATTR_LEN(name_len);
> - }
> +
> + /* Check whether we have enough space. */
> if (i->value) {
> - size_t value_len = EXT4_XATTR_SIZE(i->value_len);
> + size_t free;
>
> - if (in_inode)
> - value_len = 0;
> + free = min_offs - ((void *)last - s->base) - sizeof(__u32);
> + if (!s->not_found)
> + free += EXT4_XATTR_LEN(name_len) + old_size;
>
> - if (free < EXT4_XATTR_LEN(name_len) + value_len)
> - return -ENOSPC;
> + if (free < EXT4_XATTR_LEN(name_len) + new_size) {
> + ret = -ENOSPC;
> + 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);
> - }
> - return 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);
>
> - /* 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;
> -
> - /* 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 = EXT4_XATTR_NEXT(last);
> - }
> + 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 (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 (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);
> + }
> + goto out;
> }
> - if (!i->value) {
> - /* Remove the 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));
> - memset(last, 0, size);
> +
> + 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. */
> + 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 + old_size);
> + last = EXT4_XATTR_NEXT(last);
> }
> }
>
> + if (!i->value) {
> + /* Remove old name. */
> + size_t size = EXT4_XATTR_LEN(name_len);
> +
> + last = ENTRY((void *)last - size);
> + 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 {
> @@ -1224,6 +1642,8 @@ ext4_xattr_block_set(handle_t *handle, struct inode *inode,
> struct mb_cache_entry *ce = NULL;
> int error = 0;
> struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
> + struct inode *ea_inode = NULL;
> + size_t old_ea_inode_size = 0;
>
> #define header(x) ((struct ext4_xattr_header *)(x))
>
> @@ -1278,6 +1698,24 @@ 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. */
> @@ -1299,6 +1737,24 @@ 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);
>
> @@ -1409,6 +1865,22 @@ 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) {
> @@ -1428,15 +1900,38 @@ 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(ext4_mb_cache, ce);
> brelse(new_bh);
> @@ -1561,6 +2056,22 @@ 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()
> *
> @@ -1603,9 +2114,18 @@ 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->i_sb, bh, value_len);
> + brelse(bh);
> +
> if (!ext4_handle_has_enough_credits(handle, credits)) {
> error = -ENOSPC;
> goto cleanup;
> @@ -1641,6 +2161,7 @@ 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);
> @@ -1709,34 +2230,29 @@ 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;
> + struct buffer_head *bh;
> + int err;
>
> - credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb);
> + *credits = 0;
>
> - /*
> - * 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;
> -
> - 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->i_sb, bh, value_len);
> + brelse(bh);
> + err = 0;
> }
> - return credits;
> +
> + up_read(&EXT4_I(inode)->xattr_sem);
> + return err;
> }
>
> /*
> @@ -1761,7 +2277,10 @@ 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);
> @@ -2067,10 +2586,10 @@ 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.
> */
> @@ -2115,21 +2634,19 @@ 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,
> - struct ext4_xattr_inode_array **ea_inode_array,
> - int extra_credits)
> +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,
> @@ -2141,66 +2658,71 @@ 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)
> - goto cleanup;
> -
> - error = ext4_journal_get_write_access(handle, iloc.bh);
> - if (error)
> - goto cleanup;
> + error = ext4_get_inode_loc(inode, &iloc);
> + if (error) {
> + EXT4_ERROR_INODE(inode, "inode loc (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);
> + error = ext4_journal_get_write_access(handle, iloc.bh);
> + if (error) {
> + EXT4_ERROR_INODE(inode, "write access (error %d)",
> + error);
> + goto cleanup;
> + }
>
> -delete_external_ea:
> - if (!EXT4_I(inode)->i_file_acl) {
> - error = 0;
> - goto cleanup;
> - }
> - bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
> - if (!bh) {
> - EXT4_ERROR_INODE(inode, "block %llu read error",
> - EXT4_I(inode)->i_file_acl);
> - 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;
> - 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_has_feature_ea_inode(inode->i_sb)) {
> - error = ext4_journal_get_write_access(handle, bh);
> - if (error) {
> - EXT4_ERROR_INODE(inode, "write access %llu",
> + 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",
> EXT4_I(inode)->i_file_acl);
> + error = -EIO;
> + goto cleanup;
> + }
> + 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;
> }
> - ext4_xattr_inode_remove_all(handle, inode, bh,
> - BFIRST(bh),
> - true /* block_csum */,
> - ea_inode_array,
> - extra_credits);
> - }
>
> - ext4_xattr_release_block(handle, inode, bh);
> - /* Update i_file_acl within the same transaction that releases block. */
> - EXT4_I(inode)->i_file_acl = 0;
> - error = ext4_mark_inode_dirty(handle, inode);
> - if (error) {
> - EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
> - error);
> - goto cleanup;
> + if (ext4_has_feature_ea_inode(inode->i_sb)) {
> + 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, 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) {
> + EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
> + error);
> + goto cleanup;
> + }
> }
> + error = 0;
> cleanup:
> brelse(iloc.bh);
> brelse(bh);
> @@ -2209,17 +2731,13 @@ 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);
> }
>
> diff --git a/fs/ext4/xattr.h b/fs/ext4/xattr.h
> index b2005a2716d9..67616cb9a059 100644
> --- a/fs/ext4/xattr.h
> +++ b/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);
> diff --git a/fs/mbcache.c b/fs/mbcache.c
> index 45a8d52dc991..d818fd236787 100644
> --- a/fs/mbcache.c
> +++ b/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
> --
> 2.13.1.518.g3df882009-goog
>