Skip to contentSkip to conversation

review: propogate return codes from flush routines

PATCH Lachlan McIlroy 2 participants 4 messages Sep 4, 2006 – Sep 6, 2006

lore
Share & tools
lore
#0 lore raw

Here's a patch to handle error return values in fs_flush_pages and
fs_flushinval_pages. It changes the prototype of fs_flushinval_pages
so we can propogate the errors and handle them at higher layers. I also modified xfs_itruncate_start so that it could propogate the error further.

I've changed the necessary prototypes on 2.4 to keep the build happy but haven't bothered to fix the error handling in fs_flush_pages or fs_flushinval_pages for 2.4.

The motivation behind this change was the recent BUG reported due to a direct I/O read trying to write to delayed alloc extents. While the exact cause of this problem is not known it is possible that fs_flushinval_pages ignored an error while flushing, truncated the pages on the file anyway, and failed to convert all delayed alloc extents.

Lachlan

xfs/linux-2.4/xfs_fs_subr.c renamed from xfs/linux-2.4/xfs_fs_subr.c_1.48 +2 -1
@@ -35,7 +35,7 @@
 		truncate_inode_pages(ip->i_mapping, first);
 }
 
-void
+int
 fs_flushinval_pages(
 	bhv_desc_t	*bdp,
 	xfs_off_t	first,
@@ -53,6 +53,7 @@
 		filemap_fdatawait(ip->i_mapping);
 		truncate_inode_pages(ip->i_mapping, first);
 	}
+	return 0;
 }
 
 int
xfs/linux-2.4/xfs_fs_subr.h renamed from xfs/linux-2.4/xfs_fs_subr.h_1.17 +1 -1
@@ -23,7 +23,7 @@
 extern int  fs_nosys(void);
 extern void fs_noval(void);
 extern void fs_tosspages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
-extern void fs_flushinval_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
+extern int  fs_flushinval_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
 extern int  fs_flush_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, uint64_t, int);
 
 #endif	/* __XFS_FS_SUBR_H__ */
xfs/linux-2.4/xfs_vnode.h renamed from xfs/linux-2.4/xfs_vnode.h_1.113 +1 -1
@@ -183,7 +183,7 @@
 typedef void	(*vop_link_removed_t)(bhv_desc_t *, bhv_vnode_t *, int);
 typedef void	(*vop_vnode_change_t)(bhv_desc_t *, bhv_vchange_t, __psint_t);
 typedef void	(*vop_ptossvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
-typedef void	(*vop_pflushinvalvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
+typedef int	(*vop_pflushinvalvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
 typedef int	(*vop_pflushvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t,
 				uint64_t, int);
 typedef int	(*vop_iflush_t)(bhv_desc_t *, int);
xfs/linux-2.6/xfs_fs_subr.c renamed from xfs/linux-2.6/xfs_fs_subr.c_1.47 +11 -8
@@ -35,7 +35,7 @@
 		truncate_inode_pages(ip->i_mapping, first);
 }
 
-void
+int
 fs_flushinval_pages(
 	bhv_desc_t	*bdp,
 	xfs_off_t	first,
@@ -44,13 +44,16 @@
 {
 	bhv_vnode_t	*vp = BHV_TO_VNODE(bdp);
 	struct inode	*ip = vn_to_inode(vp);
+	int		ret = 0;
 
 	if (VN_CACHED(vp)) {
 		if (VN_TRUNC(vp))
 			VUNTRUNCATE(vp);
-		filemap_write_and_wait(ip->i_mapping);
-		truncate_inode_pages(ip->i_mapping, first);
+		ret = filemap_write_and_wait(ip->i_mapping);
+		if (!ret)
+			truncate_inode_pages(ip->i_mapping, first);
 	}
+	return ret;
 }
 
 int
@@ -63,14 +66,14 @@
 {
 	bhv_vnode_t	*vp = BHV_TO_VNODE(bdp);
 	struct inode	*ip = vn_to_inode(vp);
+	int		ret = 0;
 
 	if (VN_DIRTY(vp)) {
 		if (VN_TRUNC(vp))
 			VUNTRUNCATE(vp);
-		filemap_fdatawrite(ip->i_mapping);
-		if (flags & XFS_B_ASYNC)
-			return 0;
-		filemap_fdatawait(ip->i_mapping);
+		ret = filemap_fdatawrite(ip->i_mapping);
+		if (!ret && !(flags & XFS_B_ASYNC))
+			ret = filemap_fdatawait(ip->i_mapping);
 	}
-	return 0;
+	return ret;
 }
xfs/linux-2.6/xfs_fs_subr.h renamed from xfs/linux-2.6/xfs_fs_subr.h_1.13 +1 -1
@@ -23,7 +23,7 @@
 extern int  fs_nosys(void);
 extern void fs_noval(void);
 extern void fs_tosspages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
-extern void fs_flushinval_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
+extern int  fs_flushinval_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
 extern int  fs_flush_pages(bhv_desc_t *, xfs_off_t, xfs_off_t, uint64_t, int);
 
 #endif	/* __XFS_FS_SUBR_H__ */
xfs/linux-2.6/xfs_lrw.c renamed from xfs/linux-2.6/xfs_lrw.c_1.250 +9 -3
@@ -200,7 +200,7 @@
 	struct file		*file = iocb->ki_filp;
 	struct inode		*inode = file->f_mapping->host;
 	size_t			size = 0;
-	ssize_t			ret;
+	ssize_t			ret = 0;
 	xfs_fsize_t		n;
 	xfs_inode_t		*ip;
 	xfs_mount_t		*mp;
@@ -272,9 +272,13 @@
 
 	if (unlikely(ioflags & IO_ISDIRECT)) {
 		if (VN_CACHED(vp))
-			bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
+			ret = bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
 						 -1, FI_REMAPF_LOCKED);
 		mutex_unlock(&inode->i_mutex);
+		if (ret) {
+			xfs_iunlock(ip, XFS_IOLOCK_SHARED);
+			return ret;
+		}
 	}
 
 	xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
@@ -802,8 +806,10 @@
 		if (need_flush) {
 			xfs_inval_cached_trace(io, pos, -1,
 					ctooff(offtoct(pos)), -1);
-			bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
+			error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
 					-1, FI_REMAPF_LOCKED);
+			if (error)
+				goto out_unlock_internal;
 		}
 
 		if (need_i_mutex) {
xfs/linux-2.6/xfs_vnode.h renamed from xfs/linux-2.6/xfs_vnode.h_1.125 +1 -1
@@ -196,7 +196,7 @@
 typedef void	(*vop_link_removed_t)(bhv_desc_t *, bhv_vnode_t *, int);
 typedef void	(*vop_vnode_change_t)(bhv_desc_t *, bhv_vchange_t, __psint_t);
 typedef void	(*vop_ptossvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
-typedef void	(*vop_pflushinvalvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
+typedef int	(*vop_pflushinvalvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t, int);
 typedef int	(*vop_pflushvp_t)(bhv_desc_t *, xfs_off_t, xfs_off_t,
 				uint64_t, int);
 typedef int	(*vop_iflush_t)(bhv_desc_t *, int);
xfs/xfs_dfrag.c renamed from xfs/xfs_dfrag.c_1.55 +3 -1
@@ -200,7 +200,9 @@
 
 	if (VN_CACHED(tvp) != 0) {
 		xfs_inval_cached_trace(&tip->i_iocore, 0, -1, 0, -1);
-		bhv_vop_flushinval_pages(tvp, 0, -1, FI_REMAPF_LOCKED);
+		error = bhv_vop_flushinval_pages(tvp, 0, -1, FI_REMAPF_LOCKED);
+		if (error)
+			goto error0;
 	}
 
 	/* Verify O_DIRECT for ftmp */
xfs/xfs_inode.c renamed from xfs/xfs_inode.c_1.451 +5 -3
@@ -1421,7 +1421,7 @@
  * must be called again with all the same restrictions as the initial
  * call.
  */
-void
+int
 xfs_itruncate_start(
 	xfs_inode_t	*ip,
 	uint		flags,
@@ -1431,6 +1431,7 @@
 	xfs_off_t	toss_start;
 	xfs_mount_t	*mp;
 	bhv_vnode_t	*vp;
+	int		error = 0;
 
 	ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
 	ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
@@ -1468,7 +1469,7 @@
 		 * file size, so there is no way that the data extended
 		 * out there.
 		 */
-		return;
+		return 0;
 	}
 	last_byte = xfs_file_last_byte(ip);
 	xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
@@ -1477,7 +1478,7 @@
 		if (flags & XFS_ITRUNC_DEFINITE) {
 			bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
 		} else {
-			bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
+			error = bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
 		}
 	}
 
@@ -1486,6 +1487,7 @@
 		ASSERT(VN_CACHED(vp) == 0);
 	}
 #endif
+	return error;
 }
 
 /*
xfs/xfs_inode.h renamed from xfs/xfs_inode.h_1.215 +1 -1
@@ -439,7 +439,7 @@
 uint		xfs_dic2xflags(struct xfs_dinode_core *);
 int		xfs_ifree(struct xfs_trans *, xfs_inode_t *,
 			   struct xfs_bmap_free *);
-void		xfs_itruncate_start(xfs_inode_t *, uint, xfs_fsize_t);
+int		xfs_itruncate_start(xfs_inode_t *, uint, xfs_fsize_t);
 int		xfs_itruncate_finish(struct xfs_trans **, xfs_inode_t *,
 				     xfs_fsize_t, int, int);
 int		xfs_iunlink(struct xfs_trans *, xfs_inode_t *);
xfs/xfs_utils.c renamed from xfs/xfs_utils.c_1.72 +5 -1
@@ -420,7 +420,11 @@
 	 * in a transaction.
 	 */
 	xfs_ilock(ip, XFS_IOLOCK_EXCL);
-	xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
+	error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0);
+	if (error) {
+		xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+		return error;
+	}
 
 	tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE);
 	if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
xfs/xfs_vfsops.c renamed from xfs/xfs_vfsops.c_1.511 +1 -1
@@ -1150,7 +1150,7 @@
 			if (XFS_FORCED_SHUTDOWN(mp)) {
 				bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
 			} else {
-				bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
+				error = bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
 			}
 
 			xfs_ilock(ip, XFS_ILOCK_SHARED);
xfs/xfs_vnodeops.c renamed from xfs/xfs_vnodeops.c_1.682 +13 -3
@@ -1258,8 +1258,12 @@
 		 * do that within a transaction.
 		 */
 		xfs_ilock(ip, XFS_IOLOCK_EXCL);
-		xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
+		error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
 				    ip->i_d.di_size);
+		if (error) {
+			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+			return error;
+		}
 
 		error = xfs_trans_reserve(tp, 0,
 					  XFS_ITRUNCATE_LOG_RES(mp),
@@ -1676,7 +1680,11 @@
 		 */
 		xfs_ilock(ip, XFS_IOLOCK_EXCL);
 
-		xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
+		error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
+		if (error) {
+			xfs_iunlock(ip, XFS_IOLOCK_EXCL);
+			return VN_INACTIVE_CACHE;
+		}
 
 		error = xfs_trans_reserve(tp, 0,
 					  XFS_ITRUNCATE_LOG_RES(mp),
@@ -4332,8 +4340,10 @@
 	if (VN_CACHED(vp) != 0) {
 		xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
 				ctooff(offtoct(ioffset)), -1);
-		bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
+		error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
 				-1, FI_REMAPF_LOCKED);
+		if (error)
+			goto out_unlock_iolock;
 	}
 
 	/*
#1 lore raw
Quoted 4 lines · Lachlan McIlroy

Here's a patch to handle error return values in fs_flush_pages and fs_flushinval_pages. It changes the prototype of fs_flushinval_pages so we can propogate the errors and handle them at higher layers. I also modified xfs_itruncate_start so that it could propogate the error further.

IMHO this is always a good idea. Although I guess the only concern can be getting the right error back (and a useful one).

Quoted 5 lines

The motivation behind this change was the recent BUG reported due to a direct I/O read trying to write to delayed alloc extents. While the exact cause of this problem is not known it is possible that fs_flushinval_pages ignored an error while flushing, truncated the pages on the file anyway, and failed to convert all delayed alloc extents.

from a quick look the patch seems to do as advertised.

i probably just haven't looked hard enough - but I'm assuming the layers higher up deal with the error and: report to user, write log message or something if there's a really catastrophic error?

Signature · 2 lines
Stewart Smith ([email protected])
http://www.flamingspork.com/
#2 lore raw
Quoted 9 lines · Stewart Smith
Quoted 3 lines · Lachlan McIlroy

Here's a patch to handle error return values in fs_flush_pages and
fs_flushinval_pages. It changes the prototype of fs_flushinval_pages
so we can propogate the errors and handle them at higher layers. I also modified xfs_itruncate_start so that it could propogate the error further.

IMHO this is always a good idea. Although I guess the only concern can be getting the right error back (and a useful one).

In all but one case the patch passes the error back up to the next layer unmodified. In xfs_inactive() I've converted the error into VN_INACTIVE_CACHE since that follows the existing convention.

Quoted 9 lines
Quoted 7 lines

The motivation behind this change was the recent BUG reported due to a direct I/O read trying to write to delayed alloc extents. While the exact cause of this problem is not known it is possible that fs_flushinval_pages ignored an error while flushing, truncated the pages on the file anyway, and failed to convert all delayed alloc extents.

Quoted 6 lines

from a quick look the patch seems to do as advertised.

i probably just haven't looked hard enough - but I'm assuming the layers higher up deal with the error and: report to user, write log message or something if there's a really catastrophic error?

What's a really catastrophic error?

I'll let the higher layers handle the error in their own way - most likely any errors will result in a failed system call.

#3 lore raw
Quoted 1 line · Lachlan McIlroy

What's a really catastrophic error?

I think I've just spent too much time with HA database clusters... good temporary errors "out of resource", okay temp errors "node failed" or "lost connection", bad errors "cluster failure", catastrophic "and your data is gone too and backups haven't been working". :)

I sometimes forget that the rest of the world can just have "errors".

Signature · 2 lines
Stewart Smith ([email protected])
http://www.flamingspork.com/