On 2016-09-29 9:32 am, Dean E. Weimer wrote:
I discovered, unfortunately by deleting a jail by accident, that my
backup process isn't working. At least it was only the operating
system part of the jail, I still have all the data so I just need to
reinstall the operating system. While the ports are in the process of
building I started to investigate the cause, because the backup logs
report everything was fine.

I have a custom pre-backup script I wrote that takes snapshots of my
ZFS datasets, and then mounts those under /mnt/backup with nullfs
mount points to the .zfs/snapshot/.. directories then I back them up
rather than the live file system, allowing me to stop some services
that don't restore from a running state correctly and then restart
after the snapshot so downtime is only a couple of minutes instead of
full length of backups.

It appeared to be running perfectly, without errors, but apparently
the script isn't reporting some nullfs mount failures, so why are the
failing, turns out it thinks the file name is too long? but looking at
the mount(2) man page it states this:

[ENAMETOOLONG] A component of a pathname exceeded 255 characters, or
                        the entire length of a path name exceeded 1023
                        characters.

I can see that at some point under this, I may reach that 1023 limit,
but what of the total 71 characters in this path has a problem?

/jails/unifi/ROOT/.zfs/snapshot/11.0-RELEASE-r306379-2016.09.28--bsnap

root@freebsd:/jails/unifi/ROOT/.zfs/snapshot # ls
ls: 11.0-RELEASE-r306379-2016.09.28--bsnap: File name too long

I thought, maybe its a ZFS specific error, and ran across this:
http://lists.freebsd.org/pipermail/freebsd-fs/2010-March/007964.html

[..snip..]
From looking at the code, I think you hitting this limit:

        /*
         * Be ultra-paranoid about making sure the type and fspath
         * variables will fit in our mp buffers, including the
         * terminating NUL.
         */
        if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
                return (ENAMETOOLONG);

in vfs_domount() or vfs_donmount().

This is FreeBSD limit caused by statfs structure:

/*
 * filesystem statistics
 */
[...]
#define MNAMELEN        88              /* size of on/from name bufs */
[...]
struct statfs {
[...]
        char    f_mntfromname[MNAMELEN];/* mounted filesystem */
        char    f_mntonname[MNAMELEN];  /* directory on which mounted */
};

When you list .zfs/snapshot/ directory (especially with -l option) ZFS
mounts snapshots on lookup and this is this mount that fails.
[..snip..]

I can seemingly due anything else with the snapshot, clone, send,
receive its just that I am unable to access the files on it through
.zfs/snapshot/..

I am trying to find what the limit is here from this, because this one
here works.

/jails/webmail/usr-local-subversion/.zfs/snapshot/usr-local-subversion--bsnap

its longer in total length than most of the ones that are failing.

/jails/unifi/ROOT/.zfs/snapshot/11.0-RELEASE-r306379-2016.09.28--bsnap

So it appears that its in the name, and not the mount point.

this one works as well, which is my ZFS boot environment on the main system
zraid/ROOT/11.0-RELEASE-r306379-2016.09.28
snapshot is /.zfs/snapshot/11.0-RELEASE-r306379-2016.09.28--bsnap

So its not just the last component of the zfs dataset name, which is
in this case the same.

I am trying to wrap my head around this and find where the limit is so
I can adjust my naming conventions used and actually get backups of
all of my data. Turns out all of my jail operating system paths aren't
being backed up, fortunately at least all of the data file systems for
the jails are.

I found a solution, I was naming the snapshots with the dataset name, which I think was causing the issue.

The following didn't seem to long to be an issue
/jails/unifi/ROOT/.zfs/snapshot/11.0-RELEASE-r306379-2016.09.28--bsnap

But apparently the snapshot name was
zraid/jails/unifi/11.0-RELEASE-r306379-2016.09.28@11.0-RELEASE-r306379-2016.09.28--bsnap

Still not sure how it adds up to too long, both full paths together aren't over 255, at 160, but apparently something else is added in there. I was able to easily modify my backup script to not include the last part of the dataset in the snapshot name and simply use -bsnap-, as the name. it appears to avoid all the issues, and my backups from last night include all the files.

/jails/unifi/ROOT/.zfs/snapshot/-bsnap-
zraid/jails/unifi/11.0-RELEASE-r306379-2016.09.28@-bsnap-

The total path now only adds up to 98, I haven't done any testing yet to find out where the limit is hit, The longest combination of these I had last night would have added up to 135, and that worked

--
Thanks,
   Dean E. Weimer
   http://www.dweimer.net/
_______________________________________________
freebsd-stable@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-stable
To unsubscribe, send any mail to "freebsd-stable-unsubscr...@freebsd.org"

Reply via email to