Skip to contentSkip to conversation

Automount 5 LDAP filtering question

DISCUSSION Victor Danilchenko 2 participants 3 messages Mar 23, 2007 Kernel automounter

lore
Share & tools
lore
#0 lore raw

Hi all,

We have a widely deployed autofs net with LDAP as the map source. So
far, it has worked marvelously well. We have recently installed an FC6 host, and ran into some problems. All but one of them are solved, but that one remains a thorn in my side.

The problem is this. We have individual systems indirectly automounting
each other. As such, it is useful for them to be able to keep both the automounted and the local mountpoints in the same tree, for performance reasons (rather than mount self over NFS from the /exports/myhost directory into the /nfs/myhost directory, for example). With automount 4, it was easy. Since it was getting the LDAP master map via a dedicated utility which printed the formatted map to STDOUT, we just modified the way this utility is called:

/usr/lib/autofs/autofs-ldap-auto-master | grep -v auto.hostname

This way, each automounting host in a given group would get the entire
auto.master map from LDAP, but without that host's own automount entry -- it would have the local volumes already mounted in the same location. Every host's volumes are just transparently accessible in /nfs/<hostname>/*

In comes automount 5. There is no longer the wrapper script, automount
knows how to use the LDAP master map directly. We got it all to work with out current LDAP data tree, with one 'but': I can find no facility to exclude self's automount entry.

Is there perhaps some undocumented way to specify an additional LDAP
filter clause that could be conjoined with the standard '(objectClass=<DEFAULT_ENTRY_OBJECT_CLASS>)' filter? I can't imagine nobody else has run into the problem of excluding self's mounts from the LDAP master map in a cross-mount situation...

Something conceptually equivalent to this, to go into
/etc/sysconfig/autofs:

EXTRA_FILTER_CLAUSE="!(cn=/nfs/myhost)"

So that if the EXTRA_FILTER_CLAUSE is defined, the "(<DEFAULT_FILTER>)"
is replaced with:

"(&(<DEFAULT_FILTER>)(EXTRA_FILTER_CLAUSE))"

Is there anything in Automount 5 that would provide this sort of
functionality?

Signature · 3 lines
|  Victor  Danilchenko  | Students nowadays, complaining they only get |
| [email protected] | 10MBs of disk space! In my day we were lucky |
|   CSCF   |   5-4231   | if we had one file, and that was /dev/null.  |
#1 lore raw
Quoted 5 lines · Victor Danilchenko

The problem is this. We have individual systems indirectly automounting
each other. As such, it is useful for them to be able to keep both the automounted and the local mountpoints in the same tree, for performance reasons (rather than mount self over NFS from the /exports/myhost directory into the /nfs/myhost directory, for example). With automount

Which operating system is this? We're running Linux (with autofs-4), and autofs is smart enough to recognize and do a bind mount of local filesystems, so there is no overhead. We don't do anything special; however, the local filesystems are in one place (/h[1-9], /m[1-9]) while the automounted references are elsewhere (/net/$HOSTNAME). Here's an excerpt from /etc/mtab on a typical host, Sunset:

/dev/sdb2 /m1 ext3 rw,acl,user_xattr,quota 0 0 automount(pid4220) /net autofs rw,fd=4,pgrp=4220,minproto=2,maxproto=4 0 0 automount(pid4094) /net/sunset autofs rw,fd=4,pgrp=4220...
/m1 /net/sunset/m1 none rw,bind 0 0 <== Bind mount here
automount(pid21051) /net/julia autofs rw,fd=4,pgrp=4220...
julia:/h1 /net/julia/h1 nfs rw,,addr=128.97.4.5... <== NFS mount here

So at least on Linux you don't need to exclude local filesystems from the automount map. I believe Solaris also has this situation covered but I can't remember just what it does.

James F. Carter Voice 310 825 2897 FAX 310 206 6673
UCLA-Mathnet; 6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA 90095-1555
Email: [email protected] http://www.math.ucla.edu/~jimc (q.v. for PGP key)

#2 lore raw
Quoted 11 lines · Jim Carter
Quoted 5 lines · Victor Danilchenko

The problem is this. We have individual systems indirectly automounting
each other. As such, it is useful for them to be able to keep both the automounted and the local mountpoints in the same tree, for performance reasons (rather than mount self over NFS from the /exports/myhost directory into the /nfs/myhost directory, for example). With automount

Which operating system is this? We're running Linux (with autofs-4), and autofs is smart enough to recognize and do a bind mount of local filesystems, so there is no overhead.

Interesting. I didn't know that. I am sure it's the same way on Fedora
and RHEL. I hope someone will offer a way to customize the LDAP filter, but if it's not possible, we will go with your idea, thanks.

Thanks a bunch.

Quoted 19 lines

We don't do anything special;
however, the local filesystems are in one place (/h[1-9], /m[1-9]) while
the automounted references are elsewhere (/net/$HOSTNAME). Here's an
excerpt from /etc/mtab on a typical host, Sunset:

/dev/sdb2 /m1 ext3 rw,acl,user_xattr,quota 0 0 automount(pid4220) /net autofs rw,fd=4,pgrp=4220,minproto=2,maxproto=4 0 0 automount(pid4094) /net/sunset autofs rw,fd=4,pgrp=4220...
/m1 /net/sunset/m1 none rw,bind 0 0 <== Bind mount here
automount(pid21051) /net/julia autofs rw,fd=4,pgrp=4220...
julia:/h1 /net/julia/h1 nfs rw,,addr=128.97.4.5... <== NFS mount here

So at least on Linux you don't need to exclude local filesystems from the automount map. I believe Solaris also has this situation covered but I can't remember just what it does.

James F. Carter Voice 310 825 2897 FAX 310 206 6673
UCLA-Mathnet; 6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA 90095-1555
Email: [email protected] http://www.math.ucla.edu/~jimc (q.v. for PGP key)

Signature · 4 lines
|  Victor  Danilchenko  + Unix: Your gun, Your bullet,        |
| [email protected] |       Your foot, Your choice.       |
|   CSCF   |   5-4231   | MS:   Same as Unix, BUT: No choice, |
+-----------------------+       and We Aim Higher.            |