If supported, use socket activation to pass a pre-opened listening
socket to the NBD server. This means we can guarantee to choose an
unused port even if there are multiple instances of virt-p2v running
(ie. when testing) or if there are unexpected services running on the
same machine.
This change applies to both NBD servers, since both now (or will
shortly) support this feature.
In preparation for using socket activation, move the nbd_local_port
global inside nbd.c and make it private.
This is largely code motion. However I rearranged the order in which
the ssh data connection and the NBD server are started up. Previously
the ssh connection was made first, and the NBD server was then
started. Now the NBD server starts first. This happens so that nbd.c
can choose the local port, and when we implement socket activation it
will allow nbd.c to open the socket and choose a free port too.
This is almost pure code motion, but I changed the name and prototype
of the function 'wait_nbd' to make its purpose clearer, and to remove
the unnecessary timeout setting (which is hard-coded).
Parse the output disk as URI, and use all its attributes just like
it is done for the input disk. The only change is that the fsync of the
output disk is limited now for local URIs only, since it will not work
with remote protocols.
Implement the "tgz" output format, i.e. "tar" compressed as gzip.
This was implemented in diskimage-builder upstream as
commit da41ee6012b064aa6901c871a1104a3a3933117e.
Implement a system similar to the operations in virt-sysprep, so each
output format has its own file and attributes (such as command line
arguments).
The result is that there is no more need to spread everywhere the job
required by each output format, such as checking for tools, or hooking
at the right point.
Make sure they are handled as options for the actual mkfs driver,
instead of mkfs itself. Also, improve the documentation of
--mkfs-options accordingly.
This was fixed in diskimage-builder upstream as
commit 5e2330d89c6c0e55b270464abea7e99a4d3246ad.
Add a new option to create files with the MD5 and SHA512 checksums of
all the disk image types produced.
This was implemented in diskimage-builder upstream as
commit 2ea5feca5c4b64867ac327736edfb20408f8840e and
commit 22952b7ea0543bb4f446752976d1d8ba232b021a.
Improve the error message that is shown when trying to selecting an
element, and another element provides it.
This was implemented in diskimage-builder upstream as
commit 452f7b8d5aaa6e85e3b01b970cd9cbc842b34b86.
We don't have to always extract all files from the OVA archive. The OVA,
as defined in the standard, is plain tar. We can work directly over the
tar archive if we use correct 'offset' and 'size' options when defining
the backing file for QEMU. This puts much lower requirement on available
disk space.
Since the virt-v2v behaviour for OVA input now depends on QEMU version
available this affects some of the tests. Expected result of the
affected also has to depend on the QEMU used thus such tests will have
two *.expected files.
Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
The function looks up file in tar archive and returns a tuple containing
which at byte it starts and how long the file is.
Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
The function returns version of qemu-img as a tuple (major, minor), or
the value (0,9) in case there was an error detecting the version.
Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
In commit 08f82f2e3d we increased the
memory size to 800MB (on x86) so that the semodule command would work.
However it has been discovered that another SELinux command takes
large amounts of RAM (setfiles during the SELinux relabel step).
Therefore increase the memory size again, this time to 2000MB.
The information whether the disk is gzip compressed or not is stored
in the OVF. There is no reason to do the detection.
Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
- Make module name explicit, so it's more obvious which module a
function is defined in.
- Group lines of code by feature.
- Capitalize some module names properly.
Just code motion, no functional change.
Commit 004de6cf45 was not sufficient to
solve this problem. We are also using gperf in guestfish. Rewrite
the code in the same way to avoid having to prototype the hash
function.
Add code in virt-p2v so that it can use nbdkit (with the file plugin)
as an alternative to qemu-nbd.
This is controlled through the virt-p2v --nbd command line option,
allowing you to select which server (out of qemu-nbd or nbdkit) you
prefer (with a fallback). The default is: --nbd=qemu-nbd,nbdkit so
qemu-nbd will be used preferentially.
When testing virt-p2v (eg. on the host machine) this prevents us from
testing two instances of virt-p2v at the same time because both will
try to use the same port.
Generate the random filename using our utility function
guestfs_int_random_string. This also means that we will not need to
call srandom() in guestfish or virt-edit.
The prototype of the hash function changed in gperf 3.1:
* The 'len' parameter of the hash function and of the lookup function is now
of type 'size_t' instead of 'unsigned int'. This makes it safe to call these
functions with strings of length > 4 GB, on 64-bit machines.
This commit moves the guestfs_int_string_to_errno function into the
third (functions) section of the gperf file so it is able to use the
lookup function without needing a prototype.
Thanks: Marius Cirsta
The prototype of the hash function changed in gperf 3.1:
* The 'len' parameter of the hash function and of the lookup function is now
of type 'size_t' instead of 'unsigned int'. This makes it safe to call these
functions with strings of length > 4 GB, on 64-bit machines.
This change tries to detect the required type in ./configure
Thanks: Marius Cirsta
Three more pieces of common code are moved under the common/
subdirectory. This is just code motion.
Note that windows.[ch] wasn't even being used by guestfish. That code
was only used in other virt tools.
This is mostly code motion but:
(1) I had to remove the compile-time COMPILING_GUESTFISH and
COMPILING_VIRT_RESCUE macros and replace them with runtime constants
and checks.
(2) I moved the fish/config.c file into this library.
Just code motion.
This commit makes it clearer what is a utility and what is part of the
library. It also makes it clear that we should rename:
guestfs-internal-frontend.h -> utils.h
guestfs-internal-frontend-cleanups.h -> cleanups.h (?)
but this commit does not make that change.
This commit, which is just code motion, moves the common XDR protocol
code (libprotocol) and the common errno handling (liberrnostring) into
libraries which are each built once and shared between the library and
daemon.
The error is:
gobject/docs/gtk-doc.make:52: warning: CLEANFILES multiply defined in condition TRUE ...
gobject/docs/Makefile.am:95: 'gobject/docs/gtk-doc.make' included from here
common-rules.mk:28: ... 'CLEANFILES' previously defined here
gobject/docs/Makefile.am:18: 'subdir-rules.mk' included from here
subdir-rules.mk:20: 'common-rules.mk' included from here
