It turns out that Java assertions are disabled by default. You have
to add the 'java -ea' flag to the JVM. Who knew ..?
Because of this oversight, the tests weren't actually performing the
assertions that we wanted (although in fact none of the assertions
were failing).
This change enables assertions when running the tests.
The other programs have the variable, but the flag is not enabled
either because it doesn't make sense or because the implications are
not well understood.
This optional flag controls whether this API call will try to connect
to a running virtual machine 'guestfsd' process.
If the flag is given and the virtual machine is running, then the
libvirt XML is parsed looking for a suitable <channel> element, and
'guestfs_set_attach_method' is called with the corresponding
virtio-serial socket path.
Allow connections to a Unix domain socket which is connected
(via virtio-serial) to a guestfsd running free in an existing
guest.
In order to use this you have to add the following element
to the libvirt XML:
<channel type='unix'>
<source mode='bind' path='/tmp/socket'/>
<target type='virtio' name='org.libguestfs.channel.0'/>
</channel>
(or perform the equivalent on the qemu command line).
Then in guestfish, you can do:
guestfish \
attach-method unix:/tmp/socket : \
run : \
ll /
(or any other commands as desired).
This changes several aspects of the daemon. Currently:
* sysroot will be "" (ie. operate directly on /)
* CHROOT_IN/CHROOT_OUT are disabled
* autosync doesn't try to unmount everything
These allow you to get and set the attach method. The format
is one of:
* appliance
* unix:<path>
It's stored broken out into an enum and a string in the handle.
Since Fedora util-linux 2.19, the %post script does:
rm -f /etc/mtab
ln -s /proc/mounts /etc/mtab
We are no longer running %post scripts, so this means that /etc/mtab
is a plain file in the appliance. Usual 'mount' still updates it, but
for some reason mount.ntfs does *not* update it in Fedora 15, meaning
that you couldn't mount and then operate on NTFS partitions.
It seems better to always parse /proc/mounts (ie. what the kernel
thinks is mounted) unconditionally, rather than relying on the
capriciousness of the external mount command.
Therefore, parse /proc/mounts instead of /etc/mtab, but add a note
saying that in future we should really be parsing
/proc/self/mountinfo, but that needs a custom parser, and the format
is rather tricky:
http://lxr.linux.no/#linux+v2.6.37/Documentation/filesystems/proc.txt#L1462
Instead of explicitly calling umount-all; sync, we add a daemon
function called internal-autosync which does the same.
Apart from slightly simplifying the process of closing the handle, the
main advantage is we can modify the daemon for the standalone case so
that internal-autosync does not do the umount-all operation.
We used to maintain a global flag 'root_mounted' which tells us if the
user has mounted something on root (ie. on the sysroot directory).
This flag caused a lot of trouble (eg. RHBZ#599503) because it's hard
to keep the flag updated correctly when the user can do arbitrary
mounts and also use mkmountpoint.
Remove this flag and replace it with a test to see if something is
mounted on *or under* the sysroot. (It has to be *or under* because
of mkmountpoint and friends).
This also replaces a rather convoluted "have we mounted root yet"
check in the mount* APIs with a simpler check to see if the mountpoint
exists and is an ordinary directory.
On Debian we get this warning which I'm pretty sure is bogus:
fish.c:690: error: 'pcmd.cmd' may be used uninitialized in this
function [-Wuninitialized]
The new guestfish construct "<! cmd" executes the shell command
"cmd", and then anything printed to stdout by "cmd" is parsed
and executed as a guestfish command.
This allows some very hairy shell scripting with guestfish.
Note that 'time' and 'glob' (which both run subcommands) do not
correctly pass the exit_on_error flag in the remote case. This is not
a regression: the current code doesn't work either.
This also bumps the file size limit for "small text files"
up to 2 MB, since we want to parse Windows CD txtsetup.sif
files that are usually around 500K in size.
Previously it was including the final '/' character when calculating
the basename for the -a option eg:
Filesystem Size Used Available Use%
/Ubuntu1010x64:/dev/sda1 9.4G 2.3G 6.6G 25%
With this patch the '/' is not printed.
Fix guestfish (and other C tools) so that they ignore errors
when /etc/fstab contains bogus entries.
Update the documentation for inspect-get-mountpoints to emphasize
that callers must be aware of this when mounting the returned
values.
Add a regression test.
Update the example code ("inspect_vm") to reflect the way this
API ought to be called.
For more detail see:
https://bugzilla.redhat.com/show_bug.cgi?id=668574
