Skip to contentSkip to conversation

replicated mounts patch (local subnet mounts now get passed addr= option)

PATCH Jeff Moyer 2 participants 2 messages Aug 2, 2005 2 messages not in the archive Kernel automounter

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

Jeff Moyer

author
#2 lore raw

==> Regarding Re: [autofs] [PATCH] replicated mounts patch (local subnet mounts now get passed addr= option); Ian Kent <[email protected]> adds:

Quoted 13 lines · Jeff Layton
Quoted 13 lines

Hi All, Below is the latest iteration of my mount_nfs.c overhaul patch. This one now passes the addr= option to the mount command when the NFS server is deemed to be "local" (which in my patch is different than a bind mount).

This patch, in conjunction with the patch to mount that I sent to the util-linux maintainer should give us the proper behavior when the NFS server is a multihomed host and has an interface on the same subnet as the client.

I've unfortunately still not heard back from Adrian Bunk on my mount patch, however, so I have no idea if that's going to go in or not.

Quoted 6 lines
Quoted 6 lines

Still though, passing the addr= option shouldn't be harmful since the current version of mount just ignores it. We just may not get the desired behavior without that.

Ian, are you still planning to consider this for inclusion in 4.1.5 or 5.x?

raven> My plan is fairly simple.

raven> Get a basic implementation that provides direct mounts and release raven> as first alpha. I'm likely to label it 5.0.0 as the best way to raven> support the needed change in the protocol between the kernel and the raven> daemon is to bump the kernel module protocol version (to 5.00 as raven> well).

raven> Release 4.1.5 soon after, another bug fix release to have as stable raven> a base as possible before starting work on 4.1.6 which will include raven> analysis and possible merge of the LDAP rework , your mount rework raven> and the run in forground patches.

The run in the foreground patches don't actually add any functionality. Can we push those to v5? It seems like a pretty silly change to make in a stable release, especially since it isn't difficult to debug autofs the way it stands today, via logging.

When you're fixing to merge in the LDAP stuff, can we get that reposted to the list?

raven> Following that (and probably while working on the above) I will raven> return to working on 5.0.0. Forward porting the patches from raven> 4.1.6. New stuff that will actually be in 5.0.0 is not final raven> yet. There are two critical areas that still require a lot of raven> work. First is lazy mount/umount of multi-mount mount entries. This raven> is really difficult. Second is implementing a utility to startup and raven> shutdown autofs to get the complexity out of the init script.

Joy! Can I add to the list a re-write of the parser? Yes, I'll sign up to do the leg work. This seems to be the most error-prone and fragile bit of code.

raven> Another issue is the patch for mount that's heading for util-linux raven> some time (and has been in the FC mount) which causes no less than 7 raven> ports per mount to be opened. autofs opens 2 as well giving a total raven> of 9 per mount. Only one is needed for the actual mount. As I said raven> on the NFS list, I'm wondering if it's possible to use the SO_LINGER raven> option to help.

Steve Dickson fixed this. In fact, he trimmed down the number of reserved ports used by quite a bit. Perhaps you should try out the latest rawhide packages?

-Jeff

maintainer
#3 lore raw
Quoted 36 lines · Jeff Moyer

==> Regarding Re: [autofs] [PATCH] replicated mounts patch (local subnet mounts now get passed addr= option); Ian Kent <[email protected]> adds:

Quoted 13 lines · Jeff Layton
Quoted 13 lines

Hi All, Below is the latest iteration of my mount_nfs.c overhaul patch. This one now passes the addr= option to the mount command when the NFS server is deemed to be "local" (which in my patch is different than a bind mount).

This patch, in conjunction with the patch to mount that I sent to the util-linux maintainer should give us the proper behavior when the NFS server is a multihomed host and has an interface on the same subnet as the client.

I've unfortunately still not heard back from Adrian Bunk on my mount patch, however, so I have no idea if that's going to go in or not.

Quoted 6 lines
Quoted 6 lines

Still though, passing the addr= option shouldn't be harmful since the current version of mount just ignores it. We just may not get the desired behavior without that.

Ian, are you still planning to consider this for inclusion in 4.1.5 or 5.x?

raven> My plan is fairly simple.

raven> Get a basic implementation that provides direct mounts and release raven> as first alpha. I'm likely to label it 5.0.0 as the best way to raven> support the needed change in the protocol between the kernel and the raven> daemon is to bump the kernel module protocol version (to 5.00 as raven> well).

raven> Release 4.1.5 soon after, another bug fix release to have as stable raven> a base as possible before starting work on 4.1.6 which will include raven> analysis and possible merge of the LDAP rework , your mount rework raven> and the run in forground patches.

Quoted 4 lines

The run in the foreground patches don't actually add any functionality. Can we push those to v5? It seems like a pretty silly change to make in a stable release, especially since it isn't difficult to debug autofs the way it stands today, via logging.

I'm not to concerned myself as I also find the logging enough. So the priority on this is low.

Quoted 3 lines

When you're fixing to merge in the LDAP stuff, can we get that reposted to the list?

Yep. No problem.

Quoted 9 lines

raven> Following that (and probably while working on the above) I will raven> return to working on 5.0.0. Forward porting the patches from raven> 4.1.6. New stuff that will actually be in 5.0.0 is not final raven> yet. There are two critical areas that still require a lot of raven> work. First is lazy mount/umount of multi-mount mount entries. This raven> is really difficult. Second is implementing a utility to startup and raven> shutdown autofs to get the complexity out of the init script.

Quoted 3 lines

Joy! Can I add to the list a re-write of the parser? Yes, I'll sign up to do the leg work. This seems to be the most error-prone and fragile bit of code.

You mean the master map parser I guess.
That would be the bulk of the utility I think.

Fire away.

Quoted 11 lines

raven> Another issue is the patch for mount that's heading for util-linux raven> some time (and has been in the FC mount) which causes no less than 7 raven> ports per mount to be opened. autofs opens 2 as well giving a total raven> of 9 per mount. Only one is needed for the actual mount. As I said raven> on the NFS list, I'm wondering if it's possible to use the SO_LINGER raven> option to help.

Steve Dickson fixed this. In fact, he trimmed down the number of reserved ports used by quite a bit. Perhaps you should try out the latest rawhide packages?

Will do.
But my curiosity has stired so I'm seeking an answer to how using the SO_LINGER option on a socket that has no outstanding io behaves.

Ian