This avoids conflicts with the globally installed libguestfs
appliance, or lets us build in multiple local directories at the same
time without conflicts.
Cherry picked from commit f7d18c84dd
and backported to stable-1.8.
qemu-img used to allow you to specify a fractional image size in bytes
(or at least, it used to ignore the part after the decimal place). In
qemu-img 0.14 it no longer does this so we round down the size to a
whole number of bytes.
(cherry picked from commit eda9826d25)
I have diffed the output from the original virt-df with this
new version, and they agree very closely. Some differences:
- Old virt-df have a divide-by-zero error in cases where the
number of used inodes was 0. New virt-df fixes this.
- New virt-df uses gnulib human_readable library which displays
numbers to 3 significant figures for -h output (old version
used an ad hoc function).
This tool replaces virt-list-filesystems and virt-list-partitions with
a new tool written in C with a more uniform command line structure
and output.
This existing Perl tools are deprecated but remain indefinitely.
With changes in the core API since 1.5, virt-cat was little
more than a Perl wrapper which did some command line argument
processing. Thus it could easily be rewritten in C.
This version also shares core command line argument processing
with guestfish and guestmount, so the options have changed
slightly (old-style command line *is* supported).
virt-cat -a disk.img file [file ...]
virt-cat -d domname file [file ...]
Several other guestfish options are supported including encryption,
and with the new style multiple files can be downloaded. See the
man page for details.
We were generating this dummy 'Fedora' image already in the
tools directory. It contains just enough Fedora-like files
to fool our inspection code and thus to test the tools.
This is general enough that we can use it everywhere.
Update the following tools to use the C API for inspection:
- virt-cat
- virt-edit
- virt-ls
- virt-tar
- virt-win-reg
None of the tools in the tools/ directory now use the deprecated
Perl inspection APIs.
This was probably not a security issue, but this change
makes the code cleaner by not opening the tmp file twice.
Also be more careful about error checking in close syscall.
This adds two new options: --format specifies the format of the
input disk, and --output-format specified the format of the output
disk.
Requiring the format of the output disk seems a bit strange at first:
after all, this is the disk that the virt-resize user has to create.
However it is needed because we sometimes reopen this disk, after
copying data over the first sector, and in theory a raw-format guest
could write a qcow2 header here and have it copied to the output
disk, which we would subsequently reopen.
Sys::Guestfs::Lib is changed in two ways: firstly we take the format
string from libvirt and pass it to add_drive_opts. Secondly we allow
an extra format => parameter to open_guest which allows the
format to be specified for disk images.
All the tools are changed to add an extra --format parameter allowing
the format to be specified for direct disk images.
This tests a number of things which have caused problems for us:
- resizing PVs and LV content
- handling GPT format disks
- using qcow2 as a target disk format
- shrinking disk images
Note that the disk content is empty (not a real VM), but this is
adequate since all we want to test are the operations and calculations
done by virt-resize. We are not interested here in whether e2fsprogs
and LVM actually works.
Previously we copied the bootloader data directly from the
source disk image to the target disk image using host file
operations (before launching libguestfs). This has two problems:
firstly it has no chance of working with qcow2, and secondly
it didn't behave properly with GPT.
This changes the code so that everything is done through
libguestfs. Block device sizes are now calculated properly
for qcow2 (RHBZ#633096) because this is done using the libguestfs
blockdev_getsize64 call. The partition table is still created
by parted, but to workaround a bug in parted this is done before
copying the bootloader. Finally the bootloader copy is done
using the new APIs pread-device and pwrite-device.
Shrinking now works, at least for simple cases (RHBZ#633766).
With this commit you will see a plain progress bar during the
lengthy copy operations, similar to below:
Summary of changes:
/dev/sda1: partition will be left alone
/dev/sda2: partition will be resized from 7.5G to 9.5G
/dev/sda2: content will be expanded using the 'pvresize' method
Copying /dev/sda1 ...
[############################################################################]
Copying /dev/sda2 ...
[########################################------------------------------------]
This commit greatly improves the performance of the 'virt-df'
command by batching as many disks as possible onto a single appliance.
In many situations this means the appliance is launched only once,
versus one launch per domain as before.
However doing it this way is a lot more complex:
(1) Because of limits in Linux and virtio-blk, we can only attach
26 disks maximum at a time to the appliance.
(2) We have to use LVM filters (lvm-set-filter) to confine LVM to
the disks of a single guest.
