FreeNAS Code

The ""No" .AppleDouble?" setting should be "Enable .AppleDouble?"

--keep-newer-files errors out if the files in the update are newer than
what's on disk; let's remove the option for now to unbreak upgrades.

Mute noise from ix-multipath and <80 cols cleanup.

notifier.notifier.main() always prints out the result of the callee
(in this case notifier.notifier.multipath_sync(..)). This change below
mutes the print output so it doesn't always print out 'None' at boot.

Reviewed-by: zippybr

Add xdd to the image.

make updatep* more robust

1. Use mount -o noatime to match other logic done by delphij in r9942
to reduce potential filesystem metadata corruption race by reducing
necessary filesystem updates.
2. Use temporary directories instead of hardcoded mountpoints. The
problem was twofold with the previous code:
a. Simultaneous upgrades from multiple sources could cause
corruption; this still can still happen if the update is run after
fsck+tunefs is run -- but the race is reduced.
b. Mounting the new root slice in /mnt in updatep1 has the side
effect that data could be lost in the window where */etc/fstab was
being edited; the potential window for this happening was
miniscule, but it could result in up to a couple seconds worth of
data loss if the root device media was under heavy I/O at the time
and a process or processes wrote to the root media while the pseudo
temporary mountpoint was mounted.
3. Use tar cf - | tar xpvf - instead of cp -Rf to preserve additional
file properties (cp -Rf mashes hardlinks for example).
4. Drop overly complicated pseudo mktemp -d code as real mktemp -d
is being employed.
5. Move the trap set operation further up in the code to apply to both
the first and second mount case in updatep1.
6. Fix the trap reset operation at the end of updatep2 (the previous
code executed "1" if a SIGINT or SIGTERM was received or atexit(3)
was called at the end of /bin/sh's lifetime).

More work will need to be done to "lock the system" so multiple
concurrent upgrades cannot be attempted or the user is at least
prompted to break an upgrade lock, but this is a step forward.

Verified via 2 GUI upgrades (s1 -> s2 -> s1 again) that the logic
functioned as expected.

Signed-off-by: Garrett Cooper <yanegomi@...>

Unmount /data before reboot on failure in ix-update

/data isn't unmounted before reboot on failure which can cause filesystem
corruption if blocks aren't synched to the backing media (especially
because shutdown -r now isn't being called in ix-update). The attached
change unmounts /data which also handles sync'ing of the filesystem
according to mount(2) semantics.

Signed-off-by: Garrett Cooper <yanegomi@...>

Add system .shrc and add mountrw alias to .cshrc and .shrc

1. Adds .shrc for the root system account (it's the interactive user
rc file used by /bin/sh, but also /usr/local/bin/bash).
2. Adds aliases for mount -wu -o noatime that we can provide to users,
support techs, devs, etc, to decrease potential for filesystem
metadata corruption. This wasn't done via a generalized script for the
simple reason that it's a generic idea that for the most part just
impacts interactive use.

Signed-off-by: Garrett Cooper <yanegomi@...>

Use /var/home indirection instead of manipulating /home for AD/LDAP.

This change adds a single layer of indirection from /home -> /var/home
in the install image such that we no longer have to touch /home in the
when home directories are specified in the CIFS configuration when
ix-activedirectory and ix-ldap are run.

Verified sane via the AD code path with both trusted domains being set
and not being set. Please note that the ix-ldap code path is a simpler
pattern than the one found in ix-activedirectory, and thus covering
ix-activedirectory is sufficient.

Signed-off-by: Garrett Cooper <yanegomi@...>

Get iX repo and sf repo back in synch.

Do-Not-Merge: already in iX repo.

Add autotuning hooks to ix-sysctl.

Do-Not-Merge: already in iX repo.