Stop segfaulting on unmount error case

[rincebrain]

Motivation and Context

After a ZTS run errored out, trying to zpool export testpool2 segfaulted on trying to deref a NULL zhp.

I'm not a fan of my tools segfaulting, personally, unless the world is on fire.

(If there's a better way to extract the error code once this happens, I'm all ears, but with NULL zhp, we've got no hdl...could special-case getting back something negative more than 1, I guess, but that's a big change just to smuggle an error out...)

Description

Add a NULL check.

How Has This Been Tested?

It passes testing, but unfortunately I unwedged the system with some zpool online commands and haven't had it reproduce since, so I'm basing this working on having a core dump saying it was NULL dereferencing there.

#0  0x00007f06cc5eb6a0 in unmount_one (zhp=0x0, mountpoint=0x55e4c18bf350 "/var/tmp/zpool_scrub_offline_device", flags=<optimized out>) at /usr/src/debug/zfs-2.1.99-538_g2408c4a3c.fc34.x86_64/lib/libzfs/libzfs_mount.c:598
        libzfs_err = 2007
        error = <optimized out>
(gdb) list
598                     return (zfs_error_fmt(zhp->zfs_hdl, libzfs_err,
599                         dgettext(TEXT_DOMAIN, "cannot unmount '%s'"),
600                         mountpoint));

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Library ABI change (libzfs, libzfs_core, libnvpair, libuutil and libzfsbootenv)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[lundman]

I am probably at fault here. Looks good!

[szubersk]

If zhp == NULL invoking unmount_one doesn't make sense at all. Most likely null value is injected from

zfs/lib/libzfs/libzfs_mount.c

Line 1636 in d41856a

if (unmount_one(sets[i].dataset, sets[i].mountpoint,

and the if-guard should go there, much like it is done few lines below:

zfs/lib/libzfs/libzfs_mount.c

Line 1642 in d41856a

if (sets[i].dataset)

and

zfs/lib/libzfs/libzfs_mount.c

Line 1652 in d41856a

if (sets[i].dataset)

[rincebrain]

I mostly put the guard in there because of the comment at

zfs/lib/libzfs/libzfs_mount.c

Line 1599 in d41856a

/*

suggesting that sometimes calling with a NULL was deliberately permitted - and if you look, do_unmount on Linux doesn't actually use the zhp arg anywhere, so it could have something reasonable happen when called there.

I don't particularly care either way, I just went for the option that changed less behavior.

[szubersk]

You're right, Rich. Indeed zhp is used in neither of OS-specific implementations, what really matters is the mountpoint. Quite a surprise to be honest :)