Download raw body.
sysupgrade vs NFS
On Sat, Jun 14, 2025 at 06:28:24PM +0000, Klemens Nanni wrote:
> 14.06.2025 19:24, Miod Vallat пишет:
> > NFS mounts are not available during upgrade, therefore if you
> > don't use the -b option to select something different than
> > /home/_sysupgrade and /home is mounted on NFS, the upgrade fails.
>
> Aha, here the installer skips NFS completely wrt. DNS concerns.
>
> Two experiments for more insights:
>
> 100.64.73.1:/vm /home nfs rw,nodev,nosuid,soft,intr 0 0
>
> @@ -2712,7 +2715,6 @@ munge_fstab() {
> while read _dev _mp _fstype _opt _rest; do
> # Drop irrelevant lines and filesystems.
> [[ $_dev == @(/dev/vnd*|\#*) ||
> - $_fstype == nfs ||
> ! -f /sbin/mount_$_fstype ||
> $_opt == *noauto* ||
> $_opt == *xx* ]] && continue
>
> The function does "mount non-ffs filesystems read only":
>
> Installing base77.tgz 8% |** | 43520 KB 00:42 ETA
> tar: Unable to set file uid/gid of ./home: Read-only file system
> tar: Could not set permissions on ./home: Read-only file system
> tar: Could not set permissions on ./home: Read-only file system
> Installing base77.tgz 100% |**************************| 496 MB 00:39
> tar: Access/modification time set failed on: /mnt/home: Read-only file system
> Installation of base77.tgz failed. Continue anyway? [no] no
> ^C
>
> If it weren't for the chmod'ing the directory itself, it'd work
> (at least without DNS); adding "continue anyway" and "are you *sure*"
> to /auto_upgrade.conf and restarting (A)utoinstall makes it through.
>
>
>
> >From there, I tried these answers interjected between sysupgrade and reboot:
>
> Location of sets = nfs
> Server IP address or hostname = 100.64.73.1
> Filesystem on server to mount = /vm
> Pathname to the sets = /_sysupgrade
> Directory does not contain SHA256.sig. Continue without verification = yes
>
> That did the trick even with read-write mounting the sets dir:
>
> Location of sets? (disk http nfs or 'done') [http] nfs
> Server IP address or hostname? 100.64.73.1
> Filesystem on server to mount? /vm
> Use TCP transport? (requires TCP-capable NFS server) [no] no
> NFS Portmap: RPC: Port mapper failure - RPC: Unable to send
> Pathname to the sets? (or 'done') [7.7/amd64] /_sysupgrade
>
> Select sets by entering a set name, a file name pattern or 'all'. De-select
> sets by prepending a '-', e.g.: '-game*'. Selected sets are labelled '[X]'.
> [X] bsd [X] comp77.tgz [X] xbase77.tgz [X] xserv77.tgz
> [X] bsd.rd [X] man77.tgz [X] xshare77.tgz
> [X] base77.tgz [X] game77.tgz [X] xfont77.tgz
> Set name(s)? (or 'abort' or 'done') [done] done
> Directory does not contain SHA256.sig. Continue without verification? [no] yes
> Installing bsd 100% |**************************| 31312 KB 00:01
> Installing bsd.rd 100% |**************************| 4685 KB 00:00
> Installing base77.tgz 100% |**************************| 496 MB 00:37
> Installing comp77.tgz 100% |**************************| 87512 KB 00:15
> Installing man77.tgz 100% |**************************| 8272 KB 00:04
> Installing game77.tgz 100% |**************************| 2752 KB 00:00
> Installing xbase77.tgz 100% |**************************| 72981 KB 00:07
> Installing xshare77.tgz 100% |**************************| 4556 KB 00:06
> Installing xfont77.tgz 100% |**************************| 23021 KB 00:04
> Installing xserv77.tgz 100% |**************************| 11775 KB 00:01
> Installing BUILDINFO 100% |**************************| 54 00:00
> Location of sets? (disk http nfs or 'done') [done] done
> Making all device nodes... done.
> fw_update: add none; update none; keep intel
> Relinking to create unique kernel...uid 0 on /mnt/usr: file system full
>
>
> (3G vm.img was to small, hence relink failure, but that's unrelated.)
>
>
> Mark, perhaps you want to improve munge_fstab() to allow NFS in certain
> situations? Or have sysupgrade produce respective answers when the dir
> is on NFS? Or just roll with a custom file?
>
> Hope this helps.
>
>
1/ My first name is "Marc". Surely you can remember that.
2/ This is quickly turning into a "garage shed" issue. I've been looking
at making it run with NFS, which has its own challenges. We can work on
that together if you choose.
-> in the meantime, preventing sysupgrade from wasting our time seems like
a good idea. Several people seem to agree.
(BTW, detecting nfs efficiently and helping sysupgrade creating a proper
auto update response file will be needed anyway, so why not try to help
in the first step already ? unless you got a full patch to making it work
over nfs already ?...)
sysupgrade vs NFS