On 4/4/16 11:12 PM, Ian Kent wrote: > On Mon, 2016-04-04 at 20:55 -0400, Jeff Mahoney wrote: >> Hi all - >> >> I'm investigating an issue where our user updated from SLE11 to SLE12 >> and the hostname for autofs mounts stopped appearing in the 'mount' >> output. I know why this happens, what the reason for it being this >> way >> is, and that it can be overridden by setting the >> use_hostname_for_mounts=yes option (and the side effects). Before I >> go >> and write up the patch implementing it, is there a reason why autofs >> doesn't pass the hostname in the mount source and specify the ip >> address >> using the -oaddr=<ipaddr> option? Was this tried and found to be >> wanting somehow? > > TBH I didn't even consider doing that. > > But the daemon needs to be conservative in the options it uses because > maps may be used with other platforms and unknown or non-standard > options can cause mount failures. > > While Linux mount.nfs(8) has the sloppy option (which maintainers seem > to try and remove from time to time) other platforms probably don't. > > The addr=<address> option isn't listed as a valid option in nfs(5) and > neither is it listed in on-line Solaris documentation that I could find > so I'm not sure making the daemon use it is a good idea. > > Maybe there's another way... > > Note that I do need to be able to mount to a specific IP address while > specifying the host name in the mount command but don't know of a way todo it. After looking at it again, the kernel supports doing this but nfs-utils explicitly doesn't. A closer look at nfs-utils also shows that, given a hostname, mount.nfs will iterate over all available addresses matching the specification (e.g. constrained by proto=) and try to mount using each one. It will properly pass an addr= field to the kernel but it ignores the one passed on the mount command line. It will use getaddrinfo() to get the list of addresses for the remote host and then iterate over them, retrying until it finds one that works. So that part of it works, but the prioritization implemented by the replicated modules isn't. I wonder if it would make sense to move that to nfs-utils instead. It seems like it would be generally useful. -Jeff -- Jeff Mahoney SUSE Labs
Attachment:
signature.asc
Description: OpenPGP digital signature
