On 21/10/24 20:56, Chuck Lever III wrote:

Hello Cedric -

quoted

On Oct 21, 2024, at 1:19 AM, Cedric Blancher [off-list ref] wrote:

Good morning!

Where does the exports(5) man page in Linux come from?

http://git.linux-nfs.org/?p=steved/nfs-utils.git;a=summary

quoted

I'd like to get some items added for the refer= option:
- IPv4 syntax (w and w/o ports)
- IPv6 syntax (w and w/o ports)

IIRC "refer= IPv6 address" does not work with an alternate
port. Have you found that it does?

quoted

- hostname syntax (w and w/o ports)
- a note that hostnames used by a refer= must exist both on the client
and server
- Recommendation to use nfsref(5), if available

+1 All of the above seems to me sensible to document.

quoted

- a note that the local NFSv4 server most have an empty dir for each referral

My recollection is that the directory contents are obscured
by the referral, but the directory doesn't have to be empty.
This behavior makes it possible to simply "switch on" or
"switch off" the referral without having to move the
directory's file data.

Can you explain this a little more?

quoted

- autofs incompatibility, if you use a referral, and autofs encounters
it, you get a deadlock

This one seems like a bug that should be fixed instead of
documented. If you have a reproducer, please post it to this
list, cc: steved@redhat.com <mailto:steved@redhat.com>, raven@themaw.net <mailto:raven@themaw.net> .

A method of reproducing the problem would be ideal but a description of the

problem would also be useful. For example, the obvious question this prompts

is, is it actually autofs or is it the VFS automount support coming into 
play.

These two things are very different.


Ian

(I don't want to argue about where to file a bug again ;-)


--
Chuck Lever