subdirs trying to be mounted when using ghosting
Hi guys
We're having a problem when trying to access a subdir of an automount directory that isn't mounted. Instead of looking up the mount entry for the directory, a subdir is looked up instead. If the exact dir of the entry is accessed, then it mounts fine.
i.e.
ls /net/alpha/job/xmen3 -> no such file or directory ls /net/alpha -> lists OK
We use a yp map for auto.net, it's not -host based mounting, heres the line from auto.master
/net yp:auto.net --ghost
Here's the entry from auto.net for /net/alpha which seems to be causing the problem (but I understand we experienced this for other mount points, including nfs servers. alpha is a lustre cluster)
alpha -fstype=lustre mds1a://mdt1a/alpha
We're running autofs 4.1.4-8 from RHEL4 RHN repo rebuilt for Fedora Core 3. Also kernel-2.6.12-1.1381 from FC3 updates.
We've got debugging turned on the automounter so here you can see the logs when running the commands listed above
Jan 24 15:38:49 rj081 automount[3186]: handle_packet: type = 0
Jan 24 15:38:49 rj081 automount[3186]: handle_packet_missing: token 269, name alpha/job
Jan 24 15:38:49 rj081 automount[3186]: attempting to mount entry /net/alpha/job
Jan 24 15:38:49 rj081 automount[25673]: lookup(yp): looking up alpha/job
Jan 24 15:38:49 rj081 automount[25673]: ret = 8
Jan 24 15:38:49 rj081 automount[25673]: failed to mount /net/alpha/job
Jan 24 15:38:49 rj081 automount[25673]: umount_multi: path=/net/alpha/job incl=1
Jan 24 15:38:49 rj081 automount[25673]: umount_multi: no mounts found under /net/alpha/job
Jan 24 15:38:51 rj081 automount[3137]: sig 14 switching from 1 to 2
Jan 24 15:38:51 rj081 automount[3137]: get_pkt: state 1, next 2
Jan 24 15:38:51 rj081 automount[3137]: st_expire(): state = 1
Jan 24 15:38:51 rj081 automount[3137]: expire_proc: exp_proc=25674
Jan 24 15:38:51 rj081 automount[3292]: sig 14 switching from 1 to 2
Jan 24 15:38:51 rj081 automount[3292]: get_pkt: state 1, next 2
Jan 24 15:38:51 rj081 automount[3292]: st_expire(): state = 1
Jan 24 15:38:51 rj081 automount[3292]: expire_proc: exp_proc=25675
Jan 24 15:38:51 rj081 automount[3137]: handle_child: got pid 25674, sig 0 (0), stat 0
Jan 24 15:38:51 rj081 automount[3137]: sigchld: exp 25674 finished, switching from 2 to 1
Jan 24 15:38:51 rj081 automount[3137]: get_pkt: state 2, next 1
Jan 24 15:38:51 rj081 automount[3137]: st_ready(): state = 2
Jan 24 15:38:51 rj081 automount[3292]: handle_child: got pid 25675, sig 0 (0), stat 0
Jan 24 15:38:51 rj081 automount[3292]: sigchld: exp 25675 finished, switching from 2 to 1
Jan 24 15:38:51 rj081 automount[3292]: get_pkt: state 2, next 1
Jan 24 15:38:51 rj081 automount[3292]: st_ready(): state = 2
Jan 24 15:38:51 rj081 automount[3186]: handle_child: got pid 25673, sig 0 (0), stat 1
Jan 24 15:38:51 rj081 automount[3186]: sig_child: found pending iop pid 25673: signalled 0 (sig 0), exit status 1
Jan 24 15:38:51 rj081 automount[3186]: send_fail: token=269
Jan 24 15:38:51 rj081 automount[3186]: handle_packet: type = 0
Jan 24 15:38:51 rj081 automount[3186]: handle_packet_missing: token 270, name alpha
Jan 24 15:38:51 rj081 automount[3186]: attempting to mount entry /net/alpha
Jan 24 15:38:51 rj081 automount[25677]: lookup(yp): looking up alpha
Jan 24 15:38:51 rj081 automount[25677]: ret = 1
Jan 24 15:38:51 rj081 automount[25677]: lookup(yp): alpha -> -fstype=lustre mds1a://mdt1a/alpha
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): expanded entry: -fstype=lustre mds1a://mdt1a/alpha
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): dequote("fstype=lustre") -> fstype=lustre
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): gathered options: fstype=lustre
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): dequote("mds1a://mdt1a/alpha") -> mds1a://mdt1a/alpha
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): core of entry: options=fstype=lustre, loc=mds1a://mdt1a/alpha
Jan 24 15:38:51 rj081 automount[25677]: parse(sun): mounting root /net, mountpoint alpha, what mds1a://mdt1a/alpha, fstype lustre, options
Jan 24 15:38:51 rj081 automount[25677]: do_mount mds1a://mdt1a/alpha /net/alpha type lustre options using module generic
Jan 24 15:38:51 rj081 automount[25677]: mount(generic): calling mkdir_path /net/alpha
Jan 24 15:38:51 rj081 automount[25677]: mount(generic): calling mount -t lustre mds1a://mdt1a/alpha /net/alpha
Jan 24 15:38:52 rj081 automount[25677]: mount(generic): mounted mds1a://mdt1a/alpha type lustre on /net/alpha
Jan 24 15:39:31 rj081 automount[3186]: handle_child: got pid 25677, sig 0 (0), stat 0
Jan 24 15:39:31 rj081 automount[3186]: sig_child: found pending iop pid 25677: signalled 0 (sig 0), exit status 0Signature · 3 lines
Johan van den Dorpe Systems Administrator, Framestore CFC 020 7106 2441
==> Regarding [autofs] subdirs trying to be mounted when using ghosting; Johan van den Dorpe <[email protected]> adds:
johan.vandendorpe> Hi guys We're having a problem when trying to access a johan.vandendorpe> subdir of an automount directory that isn't johan.vandendorpe> mounted. Instead of looking up the mount entry for the johan.vandendorpe> directory, a subdir is looked up instead. If the exact johan.vandendorpe> dir of the entry is accessed, then it mounts fine.
Fixed by Ian in more recent kernels. Patch was something like the attached, though you'd do better to drop this part, I think:
- return (dentry->d_time - jiffies <= AUTOFS_NEGATIVE_TIMEOUT);
+ return (dentry->d_time - jiffies <= 0);
-Jeff
fs/autofs4/expire.c +2 -2
@@ -177,7 +177,7 @@ resume:DPRINTK ("dentry %p %.*s" , dentry, (int )dentry->d_name.len, dentry->d_name.name); -if (!list_empty (&dentry->d_subdirs)) { +if (!simple_empty_nolock (dentry)) { this_parent = dentry; goto repeat; }
@@ -269,7 +269,7 @@ static struct dentry *autofs4_expire(str goto next; } -if (simple_empty (dentry) ) +if (simple_empty (dentry)) goto next;/* Case 2: tree mount, expire iff entire tree is not busy */
fs/autofs4/root.c +2 -2
@@ -386,13 +386,13 @@ static int autofs4_revalidate(struct den/* Negative dentry.. invalidate if "old" */ if (dentry->d_inode ==NULL ) -return (dentry->d_time - jiffies <=AUTOFS_NEGATIVE_TIMEOUT ); +return (dentry->d_time - jiffies <=0 );/* Check for a non-mountpoint directory with no contents */ spin_lock (&dcache_lock);if (S_ISDIR (dentry->d_inode->i_mode) && !d_mountpoint (dentry) && -list_empty (&dentry->d_subdirs)) { +simple_empty_nolock (dentry)) {DPRINTK ("dentry=%p %.*s, emptydir" , dentry, dentry->d_name.len, dentry->d_name.name);spin_unlock (&dcache_lock);
Quoted 10 lines · Jeff Moyer
==> Regarding [autofs] subdirs trying to be mounted when using ghosting; Johan van den Dorpe <[email protected]> adds:
johan.vandendorpe> Hi guys We're having a problem when trying to access a johan.vandendorpe> subdir of an automount directory that isn't johan.vandendorpe> mounted. Instead of looking up the mount entry for the johan.vandendorpe> directory, a subdir is looked up instead. If the exact johan.vandendorpe> dir of the entry is accessed, then it mounts fine.
Fixed by Ian in more recent kernels. Patch was something like the attached...
Thanks very much Jeff, Ian. No errors this morning, but after the last kernel change it was a couple of days before they cropped up... so we'll see how we go.
Thanks again.
Signature · 3 lines
Johan van den Dorpe Systems Administrator, Framestore CFC 020 7106 2441
I'm seeing a very similar problem on clients using a 2.4.30 kernel (with the 20041227 autofs4 patch) - do you (or anyone else) know if this patch will work on 2.4.x clients? The code this patch touches appears to be identical in the 2.4.x autofs4 source.
Thanks
James Pearson
Quoted 63 lines · Jeff Moyer
==> Regarding [autofs] subdirs trying to be mounted when using ghosting; Johan van den Dorpe <[email protected]> adds:
johan.vandendorpe> Hi guys We're having a problem when trying to access a johan.vandendorpe> subdir of an automount directory that isn't johan.vandendorpe> mounted. Instead of looking up the mount entry for the johan.vandendorpe> directory, a subdir is looked up instead. If the exact johan.vandendorpe> dir of the entry is accessed, then it mounts fine.
Fixed by Ian in more recent kernels. Patch was something like the attached, though you'd do better to drop this part, I think:
fs/autofs4/root.c +1
-2-return (dentry->d_time - jiffies <=AUTOFS_NEGATIVE_TIMEOUT ); +return (dentry->d_time - jiffies <=0 ); -Jeffdiff -Nurp
linux-2.6.12.orig/fs/autofs4/expire.clinux-2.6.12/fs/autofs4/expire.cfs/autofs4/expire.c +2
-2@@ -177,7 +177,7 @@ resume:DPRINTK ("dentry %p %.*s" , dentry, (int )dentry->d_name.len, dentry->d_name.name); -if (!list_empty (&dentry->d_subdirs)) { +if (!simple_empty_nolock (dentry)) { this_parent = dentry; goto repeat; }@@ -269,7 +269,7 @@ static struct dentry *autofs4_expire(str goto next; } -if (simple_empty (dentry) ) +if (simple_empty (dentry)) goto next;/* Case 2: tree mount, expire iff entire tree is not busy */ diff -Nurp
linux-2.6.12.orig/fs/autofs4/root.clinux-2.6.12/fs/autofs4/root.cfs/autofs4/root.c +2
-2@@ -386,13 +386,13 @@ static int autofs4_revalidate(struct den/* Negative dentry.. invalidate if "old" */ if (dentry->d_inode ==NULL ) -return (dentry->d_time - jiffies <=AUTOFS_NEGATIVE_TIMEOUT ); +return (dentry->d_time - jiffies <=0 );/* Check for a non-mountpoint directory with no contents */ spin_lock (&dcache_lock);if (S_ISDIR (dentry->d_inode->i_mode) && !d_mountpoint (dentry) && -list_empty (&dentry->d_subdirs)) { +simple_empty_nolock (dentry)) {DPRINTK ("dentry=%p %.*s, emptydir" , dentry, dentry->d_name.len, dentry->d_name.name);spin_unlock (&dcache_lock);_______________________________________________
autofs mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/autofs