As motivation for this; we have had two breakouts of dib in recent
memory. One was a failure to unmount through symlinks in the core
code (I335316019ef948758392b03e91f9869102a472b9) and the other was
removing host keys on the build-system
(Ib01d71ff9415a0ae04d963f6e380aab9ac2260ce).
For the most part, dib runs unprivileged. Bits of the core code are
hopefully well tested (modulo bugs like the first one!). We give free
reign inside the chroot (although there is still some potential there
for adverse external affects via bind mounts). Where we could be a
bit safer (and could have prevented at least the second of these
breakouts) is with some better checking that the "sudo" calls
*outside* the chroot at least looked sane.
This adds a basic check that we're using chroot or image paths when
calling sudo in those parts of elements that run *outside* the chroot.
Various files are updated to accomodate this check; mostly by just
ignoring it for existing code (I have not audited these calls).
Nobody is pretending this type of checking makes dib magically safe,
or removes the issues with it needing to do things as root during the
build. But this can help find egregious errors like the key removal.
Change-Id: I161a5aea1d29dcdc7236f70d372c53246ec73749
A TODO was placed on the partitioning section of the vm element to
replace sfdisk with a saner (and less arcane) way of partitioning. It
suggested parted for replacement. This changeset should reproduce the
same disk label and partition layout as sfdisk, but with less ioctl
errors and version dependency. It will also ensure partition alignment.
Change-Id: I5d8d75131458b73bfb05f80f1bfa7e2970e004b3
As described in the comments, sfdisk was rewritten for util-linux 2.26
(as shipped in F22) and now interprets arguments a sectors, rather
than cylinders.
The current partitioning line is "1 - - *" (start/size/type/bootable)
which means you start getting:
---
/usr/sbin/grub2-install: warning: this msdos-style partition label has
no post-MBR gap; embedding won't be possible.
/usr/sbin/grub2-install: warning: Embedding is not possible. GRUB can
only be installed in this setup by using blocklists. However,
blocklists are UNRELIABLE and their use is discoura ged..
/usr/sbin/grub2-install: error: will not proceed with blocklists.
---
when building images, because the start is interpreted by the new
sfdisk as sector 1 and it crams the partition right next to the MBR.
Specifying "-" for the size is undefined in the man page; even reading
the source it's not totally clear what "-" for the size does [2]. In
any case, the alignment is wrong in sectors or cylinders; we want to
be a multiple of 4KiB for best performance.
The intent here is to create one single, Linux, bootable, partition
taking up the whole disk starting at 1MiB, so "2048 + L *" makes this
clear.
We use the -uS argument to ensure both versions treat this start-value
as a sector offset (newer sfdisk essentially ignores the argument).
As described in the comments, bugs in the older sfdisk necessitate
usage of "--force".
Although we could choose more or less, it seems most common to align
to a 1MiB boundary (i.e. starting at sector 2048). libguestfs has
some disucssion around --alignment and where it sets it's default to
this [3]. The 2.26-era sfdisk also defaults to putting partitions
here. 1MiB should be enough for GPT schemes in the future as well.
[1] https://github.com/karelzak/util-linux/blob/master/libfdisk/src/script.c#L1050
[2] https://bugzilla.redhat.com/show_bug.cgi?id=1249893
[3] http://libguestfs.org/virt-resize.1.html
Change-Id: I2c2966f98d1d5ad4ebb433ea148b3b26c65dc1b5
There is a wide variety of tracing options through the various shell
scripts. Some use "set -eux", others explicity set xtrace and others
do nothing. There is a "-x" option to bin/disk-image-create but it
doesn't flow down to the many scripts it calls.
This adds a global integer variable set by disk-image-create
DIB_DEBUG_TRACE. All scripts have a stanza added to detect this and
turn on tracing. Any other tracing methods are rolled into this. So
the standard header is
---
if [ "${DIB_DEBUG_TRACE:-0}" -gt 0 ]; then
set -x
fi
set -eu
set -o pipefail
---
Multiple -x options can be specified to dib-create-image, which
increases the value of DIB_DEBUG_TRACE. If script authors feel their
script should only trace at higher levels, they should modify the
"-gt" value. If they feel it should trace by default, they can modify
the default value also.
Changes to pachset 16 : scripts which currently trace themselves by
default have retained this behaviour with DIB_DEBUG_TRACE defaulting
to "1". This was done by running [1] on patch set 15. See the thread
beginning at [2]
dib-lint is also updated to look for the variable being matched.
[1] https://gist.github.com/ianw/71bbda9e6acc74ccd0fd
[2] http://lists.openstack.org/pipermail/openstack-dev/2014-November/051575.html
Change-Id: I6c5a962260741dcf6f89da9a33b96372a719b7b0
Patch adds support to create PowerPC image with vm element.
It creates 2 partitions, one for PReP boot and other for root and
installs grub-ieee1275.
Change-Id: I4675ef2b82aa69b63e63a1cc7db01b0c0e6f9fff
Closes-Bug: 1418739
When running inside a Docker container, we cannot rely on devices in
/dev/mapper to be automagically created by udev, because we probably
don't have a udev at all. To work around this, run dmsetup mknodes
after every kpartx run.
Change-Id: If7e30579224ce54c5ed26d08974d8293c144719a
As with the previous similar changes, this is intended to catch
problems as they happen instead of ignoring them and continuing on
to potentially fail later. Setting this on all existing scripts
will allow us to enforce use via Jenkins.
Change-Id: Iad2d490c86dceab148ea9ab08f457c49a5d5352e
Fixes problems found by set -eu and pipefail, including:
-Many unset variables
-Commands that can fail under normal circumstances, which breaks
with set -e. This change swallows those expected errors to allow
our existing error code to handle them.
-The dkms element was not finding Fedora kernel versions correctly.
This may be an issue for other distros too, but since Fedora was
working fine without this functionality I only changed it to print
a warning message rather than failing the build when it happens.
-The ramdisk init script will not be set -eu because if it fails
the result is a kernel panic, which can be tricky to debug.
However, in testing with set -e a few failing commands were found
and have been fixed in this patch.
Change-Id: I44cf98dfc80cfcaec54b88cc83be80a3dbf2cec3
If the loop partition isn't created as a device by the kernel, go ahead and use
kpartx to create the device.
Change-Id: I53290b7724d0cb45a1fc9225ec096025db8978dd
Qemu-nbd does not perform well with older versions of qemu due to
the lack of writeback caching mode. It also only builds qcow2 images
and there is a desire for raw image support. Finally, qemu-nbd makes
it very difficult to build images concurrently due to the somewhat
opaque nature of how it selects a /dev/nbd# device. losetup, on
the other hand, makes this process very straight forward.
Change-Id: I309fad8af4fd1e8d1720c17b65e1897a76d5e897
Co-Author: Clint Byrum <clint@fewbar.com>
Flavour is overloaded in openstack due to it being used by nova. Element
seems to have the same feeling of combinability without using a term already
in active use in the openstack community.
Change-Id: Ia4c028d4062a8f69c66665821c94dd4bcdf06031
