illumos gate

Originally filed by Matt Ahrens:

When recovering a corrupt pool we are
sometimes able to open the pool read-only and then copy the data off the
corrupt pool into a new pool.  We would like to use zfs send to do copy the
data, but we can only send snapshots, not filesystems, and we can't create a
snapshot because the pool is readonly.

The workaround would be to use zfs send to copy the snapshots, and then rsync
to get the data from the filesystem.  This would roughly double the amount of
time that the recover takes, as rsync will need to read all the already-copied
data from both pools to determine which blocks are new.