I have a couple of boxes here which make daily snapshots of their filesystems. One just makes a snapshot at 7am, called '7am' which it does by deleting the previous days and making a new one called '7am'. The other has snapshots called 'today', 'yesterday', '2daysago' etc, up to a week. It does this by deleting the '5daysago' snapshot, renaming all the others to shuffle them down by one, and then creating a new snapshot for 'today'. On the box with the snapshots being created every day with the same name I quickly end up with unavailable snapshots, and the error message: 'Bad file descriptor'. On the machine which is creating dailys which do not have the same name this does not happen. Interesting - and unexpected. The machines are running identical kernels, being 7-STABLE form a few days ago. -pete.
On Fri, 21 Nov 2008 13:39:20 +0000 Pete French <petefrench@ticketswitch.com> wrote about Curious failure of ZFS snapshots: PF> On the box with the snapshots being created every day with the same PF> name I quickly end up with unavailable snapshots, and the error PF> message: 'Bad file descriptor'. On the machine which is creating PF> dailys which do not have the same name this does not happen. PF> PF> Interesting - and unexpected. The machines are running identical PF> kernels, being 7-STABLE form a few days ago. I have a similar setup (creating daily.0 every night and rotating the rest) for some home directories here... Right now 3 of them are fine, and one is showing the same problem you described: mclane# ll /tank/home/pt/.zfs/ ls: snapshot: Bad file descriptor total 0 Note that zfs still thinks the snapshots are there: mclane# zfs list -r tank/home/pt NAME USED AVAIL REFER MOUNTPOINT tank/home/pt 454M 1.78T 262M /tank/home/pt tank/home/pt@weekly.1 32.5M - 260M - tank/home/pt@daily.6 29.6M - 262M - tank/home/pt@weekly.0 45.3K - 262M - tank/home/pt@daily.5 45.3K - 262M - tank/home/pt@daily.4 28.2M - 262M - tank/home/pt@daily.3 28.7M - 262M - tank/home/pt@daily.2 27.8M - 262M - tank/home/pt@daily.1 374K - 262M - tank/home/pt@hourly.7 45.3K - 262M - tank/home/pt@hourly.6 45.3K - 262M - tank/home/pt@hourly.5 45.3K - 262M - tank/home/pt@hourly.4 45.3K - 262M - tank/home/pt@daily.0 45.3K - 262M - tank/home/pt@hourly.3 45.3K - 262M - tank/home/pt@hourly.2 45.3K - 262M - tank/home/pt@hourly.1 45.3K - 262M - tank/home/pt@hourly.0 45.3K - 262M - I even do autoamtic incremental backups via send/receive. There are no errors (up to today) and the snapshots are accessible on the backup machine. I am a bit troubled by unaccessible snapshots. Does anyone else here have the same problem (or can even offer a solution)? cu Gerrit
On Fri, 21 Nov 2008 15:15:18 +0100 Gerrit K?hn <gerrit@pmp.uni-hannover.de> wrote about Re: Curious failure of ZFS snapshots: GK> Right now 3 of them are fine, and one is showing the same problem you GK> described: GK> GK> mclane# ll /tank/home/pt/.zfs/ GK> ls: snapshot: Bad file descriptor GK> total 0 [...] One addition: I just tried to unmount tank/home/pt (in the hope the after a remount everything would be fine again). This paniced the kernel, even the coredump did not finish. After rebooting everything came back fine, though. Now the snapshots are accessible again. cu Gerrit