Re: zfs and device name changes
Stephen Borrill <[email protected]> Fri, 27 Mar 2026 12:17:06 +0000 (GMT)
| Newsgroups | gmane.os.netbsd.devel.kernel |
|---|---|
| Message-ID | <Pine.NEB.4.64.2603271211440.2896@ugly> |
On Fri, 27 Mar 2026, Takashi YAMAMOTO wrote: > hi, > > On Thu, Mar 26, 2026 at 12:41 PM Taylor R Campbell <[email protected]> wrote: >> >>> Date: Tue, 24 Mar 2026 09:14:37 +0900 >>> From: Takashi YAMAMOTO <[email protected]> >>> >>> the attached patch is my attempt to make zfs a bit more robust against >>> device name changes. >>> the identical patch is available at github too: >>> https://github.com/yamt/netbsd-src/commit/32283c2e362034301c3da218a05849c04ee20c2a >>> >>> while it seems working as far as i tested, i'd be happy if someone can review it >>> as my knowledge of zfs (well, and recent netbsd in general) is weak. >> >> I don't understand why all this new code is needed. Doesn't zfs >> already have logic to scan all disks/partitions/wedges and find the >> vdevs by guid? > > which code are you talking about? > it's entirely possible i'm missing something as i'm new to the code base. > >> >> I am under the impression that /etc/zfs/zpool.cache may bypass the >> scan so this doesn't work in some circumstances, but in my years of >> using zfs on various machines with frequent device renumbering of cgd >> volumes and dkN wedges, I have never encountered this type of trouble >> myself, and I'm not sure what I'm doing differently. > > do you mean zfs finds vdevs after renumbering without zpool import? > it doesn't match my experience. > without this patch, i had to use zpool export/import after: > - modify gpt in a way affecting dk numbering > - swapping qemu disk images Naively: # zpool create tank mirror xbd2 xbd3 mirror xbd4 xbd5 # zpool status pool: tank state: ONLINE scan: none requested config: NAME STATE READ WRITE CKSUM tank ONLINE 0 0 0 mirror-0 ONLINE 0 0 0 xbd2 ONLINE 0 0 0 xbd3 ONLINE 0 0 0 mirror-1 ONLINE 0 0 0 xbd4 ONLINE 0 0 0 xbd5 ONLINE 0 0 0 errors: No known data errors # halt -p ** Remove xbd1 to simulate failed/disconnected disk ** means xbd2 -> xbd1, xbd3 -> xbd2, etc. After boot: # zpool status pool: tank state: UNAVAIL status: One or more devices could not be opened. There are insufficient replicas for the pool to continue functioning. action: Attach the missing device and online it using 'zpool online'. see: http://illumos.org/msg/ZFS-8000-3C scan: none requested config: NAME STATE READ WRITE CKSUM tank UNAVAIL 0 0 0 mirror-0 UNAVAIL 0 0 0 6289893268167966748 FAULTED 0 0 0 was /dev/xbd2 4017376292647041077 FAULTED 0 0 0 was /dev/xbd3 mirror-1 UNAVAIL 0 0 0 4378765686596708079 FAULTED 0 0 0 was /dev/xbd4 6863498524284650610 UNAVAIL 0 0 0 was /dev/xbd5 After yamt's patch: dmesg shows: ZFS WARNING: vdev guid mismatch for /dev/xbd2, actual 37c09674044d7835 expected 574a309a2445c01c ZFS: trying to find a vdev (/dev/xbd2) by guid 574a309a2445c01c ZFS WARNING: vdev guid mismatch for /dev/xbd3, actual 3cc48031384426ef expected 37c09674044d7835 ZFS: trying to find a vdev (/dev/xbd3) by guid 37c09674044d7835 ZFS WARNING: vdev guid mismatch for /dev/xbd4, actual 5f400b8b20678472 expected 3cc48031384426ef ZFS: trying to find a vdev (/dev/xbd4) by guid 3cc48031384426ef ZFS: trying to find a vdev (/dev/xbd5) by guid 5f400b8b20678472 # zpool status pool: tank state: UNAVAIL status: One or more devices could not be opened. There are insufficient replicas for the pool to continue functioning. action: Attach the missing device and online it using 'zpool online'. see: http://illumos.org/msg/ZFS-8000-3C scan: none requested config: NAME STATE READ WRITE CKSUM tank UNAVAIL 0 0 0 mirror-0 UNAVAIL 0 0 0 6289893268167966748 UNAVAIL 0 0 0 was /dev/xbd2 4017376292647041077 UNAVAIL 0 0 0 was /dev/xbd3 mirror-1 UNAVAIL 0 0 0 4378765686596708079 UNAVAIL 0 0 0 was /dev/xbd4 6863498524284650610 UNAVAIL 0 0 0 was /dev/xbd5 -- Stephen