From 8780c53961e668211682d40ad36946294c3145d8 Mon Sep 17 00:00:00 2001 From: Brian Behlendorf Date: Tue, 11 Dec 2012 16:58:44 -0800 Subject: [PATCH] Update SAs when an inode is dirtied Revert the portion of commit d3aa3ea which always resulted in the SAs being update when an mmap()'ed file was closed. That change accidentally resulted in unexpected ctime updates which upset tools like git. That was always a horrible hack and I'm happy it will never make it in to a tagged release. The right fix is something I initially resisted doing because I was worried about the additional overhead. However, in hindsight the overhead isn't as bad as I feared. This patch implemented the sops->dirty_inode() callback which is unsurprisingly called when an inode is dirtied. We leverage this callback to keep the znode SAs strictly in sync with the inode. However, for now we're going to go slowly to avoid introducing any new unexpected issues by only updating the atime, mtime, and ctime. This will cover the callpath of most concern to us. ->filemap_page_mkwrite->file_update_time->update_time-> mark_inode_dirty_sync->__mark_inode_dirty->dirty_inode Signed-off-by: Brian Behlendorf Closes #764 Closes #1140 --- config/kernel-dirty-inode.m4 | 23 ++++++++++++ config/kernel.m4 | 1 + include/sys/zfs_vnops.h | 1 + module/zfs/zfs_vnops.c | 73 +++++++++++++++++++++++++----------- module/zfs/zpl_super.c | 21 ++++++++++- 5 files changed, 96 insertions(+), 23 deletions(-) create mode 100644 config/kernel-dirty-inode.m4 diff --git a/config/kernel-dirty-inode.m4 b/config/kernel-dirty-inode.m4 new file mode 100644 index 0000000000..2ededf1544 --- /dev/null +++ b/config/kernel-dirty-inode.m4 @@ -0,0 +1,23 @@ +dnl # +dnl # 3.0 API change +dnl # The sops->dirty_inode() callbacks were updated to take a flags +dnl # argument. This allows the greater control over whether the +dnl # filesystem needs to push out a transaction or not. +dnl # +AC_DEFUN([ZFS_AC_KERNEL_DIRTY_INODE_WITH_FLAGS], [ + AC_MSG_CHECKING([whether sops->dirty_inode() wants flags]) + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + void (*dirty_inode) (struct inode *, int) = NULL; + struct super_operations sops __attribute__ ((unused)); + + sops.dirty_inode = dirty_inode; + ],[ + AC_MSG_RESULT([yes]) + AC_DEFINE(HAVE_DIRTY_INODE_WITH_FLAGS, 1, + [sops->dirty_inode() wants flags]) + ],[ + AC_MSG_RESULT([no]) + ]) +]) diff --git a/config/kernel.m4 b/config/kernel.m4 index 2312730c55..aab3a167b8 100644 --- a/config/kernel.m4 +++ b/config/kernel.m4 @@ -47,6 +47,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_KERNEL_SHOW_OPTIONS ZFS_AC_KERNEL_FSYNC ZFS_AC_KERNEL_EVICT_INODE + ZFS_AC_KERNEL_DIRTY_INODE_WITH_FLAGS ZFS_AC_KERNEL_NR_CACHED_OBJECTS ZFS_AC_KERNEL_FREE_CACHED_OBJECTS ZFS_AC_KERNEL_FALLOCATE diff --git a/include/sys/zfs_vnops.h b/include/sys/zfs_vnops.h index dd25fbcbc6..5da5eaf1d9 100644 --- a/include/sys/zfs_vnops.h +++ b/include/sys/zfs_vnops.h @@ -75,6 +75,7 @@ extern int zfs_setsecattr(struct inode *ip, vsecattr_t *vsecp, int flag, extern int zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages); extern int zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc); +extern int zfs_dirty_inode(struct inode *ip, int flags); extern int zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len, unsigned long vm_flags); diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c index 415ba71911..8ec4db26f3 100644 --- a/module/zfs/zfs_vnops.c +++ b/module/zfs/zfs_vnops.c @@ -218,31 +218,10 @@ zfs_close(struct inode *ip, int flag, cred_t *cr) { znode_t *zp = ITOZ(ip); zfs_sb_t *zsb = ITOZSB(ip); - int error = 0; ZFS_ENTER(zsb); ZFS_VERIFY_ZP(zp); - /* - * When closing an mmap()'ed file ensure the inode atime, mtime, and - * ctime are written to disk. These values may have been updated in - * memory by filemap_page_mkwrite() bit are not yet reflected in the - * znode since writepage() may occur after the close. - */ - if (zp->z_is_mapped) { - vattr_t *vap; - - vap = kmem_zalloc(sizeof(vattr_t), KM_SLEEP); - vap->va_mask = ATTR_ATIME | ATTR_MTIME | ATTR_CTIME; - vap->va_atime = ip->i_atime; - vap->va_mtime = ip->i_mtime; - vap->va_ctime = ip->i_ctime; - - error = zfs_setattr(ip, vap, 0, cr); - - kmem_free(vap, sizeof(vattr_t)); - } - /* * Zero the synchronous opens in the znode. Under Linux the * zfs_close() hook is not symmetric with zfs_open(), it is @@ -256,7 +235,7 @@ zfs_close(struct inode *ip, int flag, cred_t *cr) VERIFY(zfs_vscan(ip, cr, 1) == 0); ZFS_EXIT(zsb); - return (error); + return (0); } EXPORT_SYMBOL(zfs_close); @@ -3920,6 +3899,56 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc) return (err); } +/* + * Update the system attributes when the inode has been dirtied. For the + * moment we're conservative and only update the atime, mtime, and ctime. + */ +int +zfs_dirty_inode(struct inode *ip, int flags) +{ + znode_t *zp = ITOZ(ip); + zfs_sb_t *zsb = ITOZSB(ip); + dmu_tx_t *tx; + uint64_t atime[2], mtime[2], ctime[2]; + sa_bulk_attr_t bulk[3]; + int error; + int cnt = 0; + + ZFS_ENTER(zsb); + ZFS_VERIFY_ZP(zp); + + tx = dmu_tx_create(zsb->z_os); + + dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); + zfs_sa_upgrade_txholds(tx, zp); + + error = dmu_tx_assign(tx, TXG_WAIT); + if (error) { + dmu_tx_abort(tx); + goto out; + } + + mutex_enter(&zp->z_lock); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zsb), NULL, &atime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zsb), NULL, &mtime, 16); + SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zsb), NULL, &ctime, 16); + + /* Preserve the mtime and ctime provided by the inode */ + ZFS_TIME_ENCODE(&ip->i_atime, atime); + ZFS_TIME_ENCODE(&ip->i_mtime, mtime); + ZFS_TIME_ENCODE(&ip->i_ctime, ctime); + zp->z_atime_dirty = 0; + + error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); + mutex_exit(&zp->z_lock); + + dmu_tx_commit(tx); +out: + ZFS_EXIT(zsb); + return (error); +} +EXPORT_SYMBOL(zfs_dirty_inode); + /*ARGSUSED*/ void zfs_inactive(struct inode *ip) diff --git a/module/zfs/zpl_super.c b/module/zfs/zpl_super.c index fd4f691e19..d4d4e1b670 100644 --- a/module/zfs/zpl_super.c +++ b/module/zfs/zpl_super.c @@ -48,6 +48,25 @@ zpl_inode_destroy(struct inode *ip) zfs_inode_destroy(ip); } +/* + * Called from __mark_inode_dirty() to reflect that something in the + * inode has changed. We use it to ensure the znode system attributes + * are always strictly update to date with respect to the inode. + */ +#ifdef HAVE_DIRTY_INODE_WITH_FLAGS +static void +zpl_dirty_inode(struct inode *ip, int flags) +{ + zfs_dirty_inode(ip, flags); +} +#else +static void +zpl_dirty_inode(struct inode *ip) +{ + zfs_dirty_inode(ip, 0); +} +#endif /* HAVE_DIRTY_INODE_WITH_FLAGS */ + /* * When ->drop_inode() is called its return value indicates if the * inode should be evicted from the inode cache. If the inode is @@ -306,7 +325,7 @@ zpl_free_cached_objects(struct super_block *sb, int nr_to_scan) const struct super_operations zpl_super_operations = { .alloc_inode = zpl_inode_alloc, .destroy_inode = zpl_inode_destroy, - .dirty_inode = NULL, + .dirty_inode = zpl_dirty_inode, .write_inode = NULL, .drop_inode = NULL, #ifdef HAVE_EVICT_INODE