xfs: don't bump the i_version on an atime update in xfs_vn_update_time
Subsystems (2)
RFC PATCH 1/1 xfs: don't bump the i_version on an atime update in xfs_vn_update_time
xfs will update the i_version when updating only the atime value, which is not desirable for any of the current consumers of i_version. Doing so leads to unnecessary cache invalidations on NFS and extra measurement activity in IMA.
Add a new XFS_ILOG_NOIVER flag, and use that to indicate that the transaction should not update the i_version. Set that value in xfs_vn_update_time if we're only updating the atime.
--- fs/xfs/libxfs/xfs_log_format.h | 2 +- fs/xfs/libxfs/xfs_trans_inode.c | 2 +- fs/xfs/xfs_iops.c | 10 +++++++--- 3 files changed, 9 insertions(+), 5 deletions(-)
Dave,
How about this for an alternate approach? This just explicitly ensures that we don't bump the i_version on an atime-only update, and seems to fix the testcase I have.
fs/xfs/libxfs/xfs_log_format.h +1 -1
@@ -323,7 +323,7 @@ struct xfs_inode_log_format_32 {#define XFS_ILOG_ABROOT 0x100/* log i_af.i_broot */ #define XFS_ILOG_DOWNER 0x200/* change the data fork owner on replay */ #define XFS_ILOG_AOWNER 0x400/* change the attr fork owner on replay */ - +#define XFS_ILOG_NOIVER 0x800/* don't bump i_version */ /* * The timestamps are dirty, but not necessarily anything else in the inode
fs/xfs/libxfs/xfs_trans_inode.c +1 -1
@@ -126,7 +126,7 @@ xfs_trans_log_inode(* unconditionally. */ if (!test_and_set_bit (XFS_LI_DIRTY , &iip->ili_item.li_flags)) { -if (IS_I_VERSION (inode) && +if (!(flags &XFS_ILOG_NOIVER ) &&IS_I_VERSION (inode) &&inode_maybe_inc_iversion (inode, flags &XFS_ILOG_CORE )) iversion_flags =XFS_ILOG_CORE ; }
fs/xfs/xfs_iops.c +7 -3
@@ -1021,7 +1021,7 @@ xfs_vn_update_time( {struct xfs_inode *ip =XFS_I (inode);struct xfs_mount *mp = ip->i_mount; -int log_flags =XFS_ILOG_TIMESTAMP ; +int log_flags =XFS_ILOG_TIMESTAMP |XFS_ILOG_NOIVER ;struct xfs_trans *tp;int error;
@@ -1041,10 +1041,14 @@ xfs_vn_update_time(return error;xfs_ilock (ip,XFS_ILOCK_EXCL ); -if (flags &S_CTIME ) +if (flags &S_CTIME ) { inode->i_ctime = *now; -if (flags &S_MTIME ) + log_flags &= ~XFS_ILOG_NOIVER ; + } +if (flags &S_MTIME ) { inode->i_mtime = *now; + log_flags &= ~XFS_ILOG_NOIVER ; + }if (flags &S_ATIME ) inode->i_atime = *now;
Signature · 1 line
2.37.2
Quoted 8 lines · Jeff Layton
xfs will update the
i_versionwhen updating only the atime value, which is not desirable for any of the current consumers ofi_version. Doing so leads to unnecessary cache invalidations on NFS and extra measurement activity in IMA.Add a new
XFS_ILOG_NOIVERflag, and use that to indicate that the transaction should not update thei_version. Set that value inxfs_vn_update_timeif we're only updating the atime.
I suppose that would work, since explicit atime updates from futimens() go through notify_change/setattr and not ->update_time.
--D
Quoted 72 lines
Signed-off-by Jeff Layton <[email protected]>--- fs/xfs/libxfs/xfs_log_format.h | 2 +- fs/xfs/libxfs/xfs_trans_inode.c | 2 +- fs/xfs/xfs_iops.c | 10 +++++++--- 3 files changed, 9 insertions(+), 5 deletions(-)Dave,
How about this for an alternate approach? This just explicitly ensures that we don't bump the
i_versionon an atime-only update, and seems to fix the testcase I have.fs/xfs/libxfs/xfs_log_format.h +1
-1@@ -323,7 +323,7 @@ struct xfs_inode_log_format_32 {#define XFS_ILOG_ABROOT 0x100/* log i_af.i_broot */ #define XFS_ILOG_DOWNER 0x200/* change the data fork owner on replay */ #define XFS_ILOG_AOWNER 0x400/* change the attr fork owner on replay */ - +#define XFS_ILOG_NOIVER 0x800/* don't bump i_version */ /* * The timestamps are dirty, but not necessarily anything else in the inode fs/xfs/libxfs/xfs_trans_inode.c +1
-1@@ -126,7 +126,7 @@ xfs_trans_log_inode(* unconditionally. */ if (!test_and_set_bit (XFS_LI_DIRTY , &iip->ili_item.li_flags)) { -if (IS_I_VERSION (inode) && +if (!(flags &XFS_ILOG_NOIVER ) &&IS_I_VERSION (inode) &&inode_maybe_inc_iversion (inode, flags &XFS_ILOG_CORE )) iversion_flags =XFS_ILOG_CORE ; }fs/xfs/xfs_iops.c +7
-3@@ -1021,7 +1021,7 @@ xfs_vn_update_time( {struct xfs_inode *ip =XFS_I (inode);struct xfs_mount *mp = ip->i_mount; -int log_flags =XFS_ILOG_TIMESTAMP ; +int log_flags =XFS_ILOG_TIMESTAMP |XFS_ILOG_NOIVER ;struct xfs_trans *tp;int error;@@ -1041,10 +1041,14 @@ xfs_vn_update_time(return error;xfs_ilock (ip,XFS_ILOCK_EXCL ); -if (flags &S_CTIME ) +if (flags &S_CTIME ) { inode->i_ctime = *now; -if (flags &S_MTIME ) + log_flags &= ~XFS_ILOG_NOIVER ; + } +if (flags &S_MTIME ) { inode->i_mtime = *now; + log_flags &= ~XFS_ILOG_NOIVER ; + }if (flags &S_ATIME ) inode->i_atime = *now;Signature · 1 line
2.37.2
Quoted 16 lines · Jeff Layton
xfs will update the
i_versionwhen updating only the atime value, which is not desirable for any of the current consumers ofi_version. Doing so leads to unnecessary cache invalidations on NFS and extra measurement activity in IMA.Add a new
XFS_ILOG_NOIVERflag, and use that to indicate that the transaction should not update thei_version. Set that value inxfs_vn_update_timeif we're only updating the atime.Signed-off-by Jeff Layton <[email protected]>--- fs/xfs/libxfs/xfs_log_format.h | 2 +- fs/xfs/libxfs/xfs_trans_inode.c | 2 +- fs/xfs/xfs_iops.c | 10 +++++++--- 3 files changed, 9 insertions(+), 5 deletions(-)
Quoted 5 lines
Dave,
How about this for an alternate approach? This just explicitly ensures that we don't bump the
i_versionon an atime-only update, and seems to fix the testcase I have.
This just duplicates lazytime functionality, only now users can't opt-in or out.
atime update filtering is a VFS function so behaviour is common across all filesystems. Deficiencies in VFS filtering behaviour should not be hacked around by individual filesystems, the VFS filtering should be fixed.
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]