Re: kern/58901: nfs server should allow specifying fsid in exports(5)

["Taylor R Campbell via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR kern/58901; it has been noted by GNATS.

From: Taylor R Campbell <riastradh%NetBSD.org@localhost>
To: gnats-bugs%NetBSD.org@localhost, netbsd-bugs%NetBSD.org@localhost
Cc: yamt%NetBSD.org@localhost
Subject: Re: kern/58901: nfs server should allow specifying fsid in exports(5)
Date: Wed, 26 Aug 2026 03:51:26 +0000

 The attached patch aims to implement a new exports(5) option
 
 -fsid=0x12345678.0x0abcdef9
 
 to set the exported fsid to {0x12345678,0x0abcdef9}, or
 
 -fsid=0x01020304.0x05060708:0x12345678.0x0abcdef9
 
 to verify that the local fsid is {0x01020304,0x05060708}, and if it
 is, export it as {0x12345678,0x0abcdef9}.  (If not, mountd(8) will
 fail to export the mount point at all, so you can't accidentally
 export a file system that has the wrong local fsid internally.)
 
 I initially used `,' instead of `.' as the separator but that fell
 afoul of the mountd(8) parser for exports(5) options somehow.
 
 I have only verified that /usr/tests/fs/nfs and /usr/tests/fs/vfs
 pass, and (by inserting panics into it) that it's using the new code
 paths, and (by examining debug messages) that it parses the option
 sensibly; I haven't verified that it actually exports the fsid as
 requested.  (Perhaps showmount(8) should sprout an option to dump the
 raw fhandles of the mount points to make this easier.)
 
 This should relieve some of the pressure we have to keep fsids
 consistent, so we can perhaps, e.g., start using the struct fs::fs_id
 member of the ffs superblock we've been initializing but not using for
 decades (PR kern/58902: ffs ignores fs_id), and/or get rid of that
 awful hash function in makefstype, and/or enforce fsid uniqueness
 locally (PR kern/60135: fsid may not be unique).