On Jun 20, 2017, at 3:14 AM, Tahsin Erdogan <tahsin@xxxxxxxxxx> wrote:
>
> The main purpose of mb cache is to achieve deduplication in
> extended attributes. In use cases where opportunity for deduplication
> is unlikely, it only adds overhead.
>
> Add a mount option to explicitly turn off mb cache.
>
> Suggested-by: Andreas Dilger <adilger@xxxxxxxxx>
> Signed-off-by: Tahsin Erdogan <tahsin@xxxxxxxxxx>
> ---
> fs/ext4/ext4.h | 1 +
> fs/ext4/super.c | 33 ++++++++++++++++++++++-----------
> fs/ext4/xattr.c | 55 ++++++++++++++++++++++++++++++++++++-------------------
> 3 files changed, 59 insertions(+), 30 deletions(-)
>
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 59e9488c4876..21aca1f87187 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -1149,6 +1149,7 @@ struct ext4_inode_info {
> #define EXT4_MOUNT_DIOREAD_NOLOCK 0x400000 /* Enable support for dio read nolocking */
> #define EXT4_MOUNT_JOURNAL_CHECKSUM 0x800000 /* Journal checksums */
> #define EXT4_MOUNT_JOURNAL_ASYNC_COMMIT 0x1000000 /* Journal Async Commit */
> +#define EXT4_MOUNT_NO_MBCACHE 0x2000000 /* No mbcache usage */
We've been using 0x00001 for EXT4_MOUNT_NO_MBCACHE.
> #define EXT4_MOUNT_DELALLOC 0x8000000 /* Delalloc support */
> #define EXT4_MOUNT_DATA_ERR_ABORT 0x10000000 /* Abort on file data write */
> #define EXT4_MOUNT_BLOCK_VALIDITY 0x20000000 /* Block validity checking */
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index 4b15bf674d45..4262a1b9b5b2 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -1336,7 +1336,7 @@ enum {
> Opt_inode_readahead_blks, Opt_journal_ioprio,
> Opt_dioread_nolock, Opt_dioread_lock,
> Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
> - Opt_max_dir_size_kb, Opt_nojournal_checksum,
> + Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
> };
>
> static const match_table_t tokens = {
> @@ -1419,6 +1419,7 @@ static const match_table_t tokens = {
> {Opt_noinit_itable, "noinit_itable"},
> {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
> {Opt_test_dummy_encryption, "test_dummy_encryption"},
> + {Opt_nombcache, "nombcache"},
For compatibility, please also add:
{Opt_nombcache, "no_mbcache"},
> {Opt_removed, "check=none"}, /* mount option from ext2/3 */
> {Opt_removed, "nocheck"}, /* mount option from ext2/3 */
> {Opt_removed, "reservation"}, /* mount option from ext2/3 */
> @@ -1626,6 +1627,7 @@ static const struct mount_opts {
> {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
> {Opt_max_dir_size_kb, 0, MOPT_GTE0},
> {Opt_test_dummy_encryption, 0, MOPT_GTE0},
> + {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
> {Opt_err, 0, 0}
> };
>
> @@ -4080,19 +4082,22 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
> sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
>
> no_journal:
> - sbi->s_mb_cache = ext4_xattr_create_cache();
> - if (!sbi->s_mb_cache) {
> - ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
> - 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) {
> + if (!test_opt(sb, NO_MBCACHE)) {
> + sbi->s_mb_cache = ext4_xattr_create_cache();
Should be named "s_mb_ea_block_cache" or similar?
> + if (!sbi->s_mb_cache) {
> ext4_msg(sb, KERN_ERR,
> - "Failed to create an s_ea_inode_cache");
> + "Failed to create xattr block mb_cache");
> goto failed_mount_wq;
> }
> +
> + if (ext4_has_feature_ea_inode(sb)) {
> + sbi->s_ea_inode_cache = ext4_xattr_create_cache();
Should be named "s_mb_ea_inode_cache", or alternately use "s_ea_block_cache"
above?
> + if (!sbi->s_ea_inode_cache) {
> + ext4_msg(sb, KERN_ERR,
> + "Failed to create ea_inode mb_cache");
> + goto failed_mount_wq;
> + }
> + }
> }
>
> if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
> @@ -4989,6 +4994,12 @@ static int ext4_remount(struct super_block *sb, int *flags,
> }
> }
>
> + if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_NO_MBCACHE) {
> + ext4_msg(sb, KERN_ERR, "can't enable nombcache during remount");
> + err = -EINVAL;
> + goto restore_opts;
> + }
It appears that this restriction also applies to enabling mbcache during
remount as well, so the message should be updated? In particular, if
NO_MBCACHE is currently set but isn't passed during the remount then this
will always cause the remount to fail. That makes it harder for users to
just run something like "mount -o remount,ro".
It seems we are a bit inconsistent with how we handle changing mount options.
In some cases, if an option is set to be enabled but cannot be, it is actually
ignored (e.g. DAX, JOURNAL_CHECKSUM). In other cases this returns an error.
In this case it is probably easiest to just ignore the change to this option,
as is done with JOURNAL_CHECKSUM and DAX.
> if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
> ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
> "dax flag with busy inodes while remounting");
> diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
> index 9d1b600dc827..65ead540c684 100644
> --- a/fs/ext4/xattr.c
> +++ b/fs/ext4/xattr.c
> @@ -1200,11 +1206,13 @@ ext4_xattr_release_block(handle_t *handle,
> if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
> struct mb_cache_entry *ce;
This could go inside the "if (ext4_mb_cache)" block.
> - ce = mb_cache_entry_get(ext4_mb_cache, hash,
> - bh->b_blocknr);
> - if (ce) {
> - ce->e_reusable = 1;
> - mb_cache_entry_put(ext4_mb_cache, ce);
> + if (ext4_mb_cache) {
> + ce = mb_cache_entry_get(ext4_mb_cache, hash,
> + bh->b_blocknr);
> + if (ce) {
> + ce->e_reusable = 1;
> + mb_cache_entry_put(ext4_mb_cache, ce);
> + }
> }
> }
Cheers, Andreas
Attachment:
signature.asc
Description: Message signed with OpenPGP
