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 <[email protected]>
Closes #764
Closes #1140

5 files changed, 96 insertions, 23 deletions

diff --git a/config/kernel-dirty-inode.m4 b/config/kernel-dirty-inode.m4
new file mode 100644
index 000000000..2ededf154
--- /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 <linux/fs.h>
+	],[
+		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 2312730c5..aab3a167b 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 dd25fbcbc..5da5eaf1d 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 415ba7191..8ec4db26f 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -218,32 +218,11 @@ 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
 	 * only called once when the last reference is dropped.
@@ -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 fd4f691e1..d4d4e1b67 100644
--- a/module/zfs/zpl_super.c
+++ b/module/zfs/zpl_super.c
@@ -49,6 +49,25 @@ zpl_inode_destroy(struct inode *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
  * unhashed and has no links the default policy is to evict it
@@ -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