[Cluster-devel] [PATCH] mount.gfs2 - gfs2 mounts doubled up in mtab

[Cluster-devel] [PATCH] mount.gfs2 - gfs2 mounts doubled up in mtab

[Abhijith Das]

When the -o remount option is used with mount.gfs2 it fails to remove the original mtab entry because it can't find it. Instead it simply adds the mount entry corresponding to the remount, thereby doubling the entries in mtab. This patch corrects the logic to find the mtab entry so mount.gfs2 finds and removes the original entry correctly before add the the remount entry.

Signed-off-by: Abhi Das <adas@redhat.com>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bz706141-upstream.patch
Type: text/x-patch
Size: 1143 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/cluster-devel/attachments/20110720/c485d19c/attachment.bin>

[Cluster-devel] [PATCH] mount.gfs2 - gfs2 mounts doubled up in mtab

[Steven Whitehouse]

Hi,

On Wed, 2011-07-20 at 16:10 -0400, Abhijith Das wrote:
> When the -o remount option is used with mount.gfs2 it fails to remove the original mtab entry because it can't find it. Instead it simply adds the mount entry corresponding to the remount, thereby doubling the entries in mtab. This patch corrects the logic to find the mtab entry so mount.gfs2 finds and removes the original entry correctly before add the the remount entry.
> 
> Signed-off-by: Abhi Das <adas@redhat.com>

+               sscanf(line, "%s %s %s", device, path, type);

This seems to remove the error checking for argument assignment from
sscanf,

Steve.

[Cluster-devel] [PATCH] mount.gfs2 - gfs2 mounts doubled up in mtab

[Abhijith Das]

----- Original Message -----
> From: "Steven Whitehouse" <swhiteho@redhat.com>
> To: "Abhijith Das" <adas@redhat.com>
> Cc: "cluster-devel" <cluster-devel@redhat.com>
> Sent: Friday, July 22, 2011 3:35:53 AM
> Subject: Re: [Cluster-devel] [PATCH] mount.gfs2 - gfs2 mounts doubled up in mtab
> Hi,
> 
> On Wed, 2011-07-20 at 16:10 -0400, Abhijith Das wrote:
> > When the -o remount option is used with mount.gfs2 it fails to
> > remove the original mtab entry because it can't find it. Instead it
> > simply adds the mount entry corresponding to the remount, thereby
> > doubling the entries in mtab. This patch corrects the logic to find
> > the mtab entry so mount.gfs2 finds and removes the original entry
> > correctly before add the the remount entry.
> >
> > Signed-off-by: Abhi Das <adas@redhat.com>
> 
> + sscanf(line, "%s %s %s", device, path, type);
> 
> This seems to remove the error checking for argument assignment from
> sscanf,
> 
> Steve.

Hi,

This was fixed before checking into gfs2-utils.git and cluster.git. Here's the patch that actually went in:

Thanks!
--Abhi
Red Hat Filesystems


diff --git a/gfs2/mount/mtab.c b/gfs2/mount/mtab.c
index 148ff14..fbe5f63 100644
--- a/gfs2/mount/mtab.c
+++ b/gfs2/mount/mtab.c
@@ -120,6 +120,7 @@ void del_mtab_entry(struct mount_options *mo)
 	mode_t old_umask;
 	struct stat sbuf;
 	int found = 0;
+	char *abs_path, *abs_dev;
 
 	if (ignore_mtab())
 		return;
@@ -139,13 +140,19 @@ void del_mtab_entry(struct mount_options *mo)
 	while (fgets(line, PATH_MAX, mtab)) {
 		/* exclude the line matching the fs being unmounted
 		   from the next version of mtab */
-
-		if ((sscanf(line, "%s %s %s", device, path, type) == 3) &&
-		    (strncmp(type, "gfs", 3) == 0) &&
-		    (strcmp(path, mo->dir) == 0) &&
-		    (strcmp(device, mo->dev) == 0)) {
-			found = 1;
-			continue;
+		if (sscanf(line, "%s %s %s", device, path, type) == 3) {
+			abs_path = realpath(path, NULL);
+			abs_dev = realpath(device, NULL);
+			if ((strcmp(type, "gfs2") == 0) &&
+			    (strcmp(abs_path, mo->dir) == 0) &&
+			    (strcmp(abs_dev, mo->dev) == 0)) {
+				found = 1;
+				free(abs_path);
+				free(abs_dev);
+				continue;
+			}
+			free(abs_path);
+			free(abs_dev);
 		}
 
 		/* all other lines from mtab are included in