NAME

zfsd — ZFS fault management daemon

SYNOPSIS

zfsd [-d]

DESCRIPTION

zfsd attempts to resolve ZFS faults that the kernel can't resolve by itself. It listens to devctl(4) events, which are how the kernel notifies userland of events such as I/O errors and disk removals. zfsd attempts to resolve these faults by activating or deactivating hot spares and onlining offline vdevs.

The following options are available:

-d: Run in the foreground instead of daemonizing.

System administrators never interact with zfsd directly. Instead, they control its behavior indirectly through zpool configuration. There are two ways to influence zfsd: assigning hot spares and setting pool properties. Currently, only the autoreplace property has any effect. See zpool(8) for details.

zfsd will attempt to resolve the following types of fault:

device removal: When a leaf vdev disappears, zfsd will activate any available hot spare.
device arrival: When a new GEOM device appears, zfsd will attempt to read its ZFS label, if any. If it matches a previously removed vdev on an active pool, zfsd will online it. Once resilvering completes, any active hot spare will detach automatically.
If the new device has no ZFS label but its physical path matches the physical path of a previously removed vdev on an active pool, and that pool has the autoreplace property set, then zfsd will replace the missing vdev with the newly arrived device. Once resilvering completes, any active hot spare will detach automatically.
vdev degrade or fault events: If a vdev becomes degraded or faulted, zfsd will activate any available hot spare.
I/O errors: By default, if a leaf vdev generates more than 50 I/O errors in a 60 second period, then zfsd will mark that vdev as FAULTED. ZFS will no longer issue any I/Os to it. zfsd will activate a hot spare if one is available. The defaults can be changed by setting the io_n and/or io_t vdev properties. See vdevprops(7) for details.
I/O delays: By default, if a leaf vdev generates more than delayed 8 I/O events in a 60 second period, then zfsd will mark that vdev as FAULTED. ZFS will no longer issue any I/Os to it. zfsd will activate a hot spare if one is available. The defaults can be changed by setting the slow_io_n and/or slow_io_t vdev properties. See vdevprops(7) for details.
Checksum errors: By default, if a leaf vdev generates more than 50 checksum errors in a 60 second period, then zfsd will mark that vdev as DEGRADED. ZFS will still use it, but zfsd will also activate a hot spare if one is available. The defaults can be changed by setting the checksum_n and/or checksum_t vdev properties. See vdevprops(7) for details.
Spare addition: If the system administrator adds a hot spare to a pool that is already degraded, zfsd will activate the spare.
Resilver complete: zfsd will detach any hot spare once a permanent replacement finishes resilvering.
Physical path change: If the physical path of an existing disk changes, zfsd will attempt to replace any missing disk with the same physical path, if its pool's autoreplace property is set.

zfsd will log interesting events and its actions to syslog with facility daemon and identity [zfsd].

FILES

/var/db/zfsd/cases: When zfsd exits, it serializes any unresolved casefiles here, then reads them back in when next it starts up.

HISTORY

zfsd first appeared in FreeBSD 11.0.

AUTHORS

zfsd was originally written by Justin Gibbs <gibbs@FreeBSD.org> and
Alan Somers <asomers@FreeBSD.org>

TODO

In the future, zfsd should be able to resume a pool that became suspended due to device removals, if enough missing devices have returned.