ZFS rollback not working/corruption?

Up to Table of Contents

This FAQ applies to: Any version.

#65 — Cache leads to data corruption after rollback

Important notice: risk of data loss

The details of this issue are still under investigation, but we take care to immediately notify any users of 'destructive inplace snapshot operations' (including, but possibly not limited to 'zfs rollback') that this can (at least in some configurations) lead to inconsistent page caching.

The workaround for now will be to follow any structural file system change by a full cache flush by means of

echo 3 > /proc/sys/vm/drop_caches

forcing the kernel/fuse layers to forget any cached data that may have become stale.

Of course this is a temporary workaround, and this behaviour is interpreted as a bug against zfs-fuse.

Mitigating factors

Passing any of the options --disable-block-cache or --disable-page-cache seems to mitigate this behaviour. You are advised to keep using the explicit flush mentioned above until it has been confirmed that these options make behaviour safe.

Reference

Issue #65

Add comment

You can add a comment by filling out the form below. Plain text formatting.

Info: You are not logged in. You may optionally enter your username and password below. If you don't enter anything, this comment will be posted as 'Anonymous User'.

Name

Password

Subject (Required)

Comment (Required)

Enter the word below (Required)