nfsd: Fix handling of VOP_PATHCONF() for attribute replies

[rmacklem]

Summary

des@ reported a panic in the NFSv4 server, where
nfsv4_fillattr() did a VOP_PATHCONF() without having
"vp" locked. kib@ suggested D51398 to fix it.

I think for the case of Readdirplus reply, relocking the
vnode can cause deadlocks. As such, this patch handles
VOP_PATHCONF() in the same way that the code checks
for ACL support, by doing the VOP_PATHCONF() before
the calls to nfsv4_filllattr() where the vnode is still locked.

des@, please test.

Test Plan

Tested when running a kernel with DEBUG_VFS_LOCKS
option enabled.

[kib]

sys/fs/nfsserver/nfs_nfsdport.c

2952

You can write these two lines as

has_hiddensystem = pathval > 0;

2957

Similarly there.

sys/fs/nfsserver/nfs_nfsdserv.c

318

Even better there.

[des]

I am unable to apply this patch. If possible, please regenerate it using git arc update.

[rmacklem]

In D51410#1174312, @des wrote:

I am unable to apply this patch. If possible, please regenerate it using git arc update.

It is now in main as 392a82b2.

[rmacklem]

Commit has been done. Didn't automagically close for
some reason?

[des]

In D51410#1174418, @rmacklem wrote:

Commit has been done. Didn't automagically close for some reason?

You misspelled the metadata in the commit. Please use git arc in the future.

[rmacklem]

In D51410#1174435, @des wrote:

In D51410#1174418, @rmacklem wrote:

Commit has been done. Didn't automagically close for some reason?

You misspelled the metadata in the commit. Please use git arc in the future.

The line in git log is:
Differential Revision: https://reviews.freebsd.org/D51410
looks ok to me?