Hold on to your hat, this is going to be a wild ride. As nearly
nothing in glibc, nor open() is a real function. Just look into
bits/fcntl2.h and you'll see that open() is actually a thin
wrapper that calls either __open_alias() or __open_2(). Now,
before 801ebb5edb the open() done in
virPCIDeviceConfigOpenInternal() had a constant oflags (we were
opening the pci config with O_RDWR). And since we were not
passing any mode nor O_CREAT the wrapper decided to call
__open_alias() which was open() provided by our mock. So far so
good. But after the referenced commit, the oflags is no longer
compile time constant and therefore the wrapper calls __open_2()
which we don't mock and thus the real __open_2() from glibc was
called and thus we did try to open real path from host's /sys.
This of course fails with variety of errors.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Quite a few of the tests have a need to mock the stat() / lstat()
functions and they are taking somewhat different & inconsistent
approaches none of which are actually fully correct. This is shown
by fact that 'make check' fails on 32-bit hosts. Investigation
revealed that the code was calling into the native C library impl,
not getting intercepted by our mocks.
The POSIX stat() function might resolve to any number of different
symbols in the C library.
The may be an additional stat64() function exposed by the headers
too.
On 64-bit hosts the stat & stat64 functions are identical, always
refering to the 64-bit ABI.
On 32-bit hosts they refer to the 32-bit & 64-bit ABIs respectively.
Libvirt uses _FILE_OFFSET_BITS=64 on 32-bit hosts, which causes the
C library to transparently rewrite stat() calls to be stat64() calls.
Libvirt will never see the 32-bit ABI from the traditional stat()
call. We cannot assume this rewriting is done using a macro. It might
be, but on GLibC it is done with a magic __asm__ statement to apply
the rewrite at link time instead of at preprocessing.
In GLibC there may be two additional functions exposed by the headers,
__xstat() and __xstat64(). When these exist, stat() and stat64() are
transparently rewritten to call __xstat() and __xstat64() respectively.
The former symbols will not actally exist in the library at all, only
the header. The leading "__" indicates the symbols are a private impl
detail of the C library that applications should not care about.
Unfortunately, because we are trying to mock replace the C library,
we need to know about this internal impl detail.
With all this in mind the list of functions we have to mock will depend
on several factors
- If _FILE_OFFSET_BITS is set, then we are on a 32-bit host, and we
only need to mock stat64 and __xstat64. The other stat / __xstat
functions exist, but we'll never call them so they can be ignored
for mocking.
- If _FILE_OFFSET_BITS is not set, then we are on a 64-bit host and
we should mock stat, stat64, __xstat & __xstat64. Either may be
called by app code.
- If __xstat & __xstat64 exist, then stat & stat64 will not exist
as symbols in the library, so the latter should not be mocked.
The same all applies to lstat()
These rules are complex enough that we don't want to duplicate them
across every mock file, so this centralizes all the logic in a helper
file virmockstathelper.c that should be #included when needed. The
code merely need to provide a filename rewriting callback called
virMockStatRedirect(). Optionally VIR_MOCK_STAT_HOOK can be defined
as a macro if further processing is needed inline.
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Vim treats *.h files as cpp ones with respect to syntax highlighting.
Thus "class" in _virNodeDevCapPCIDev highlighted mistakenly.
This can be fixed by filetype detection code tunables but it
is more convinient to skip this tuning by every project member.
Let's just use "klass" as field name instead of _class or class
and add syntax rule.
Signed-off-by: Nikolay Shirokovskiy <nshirokovskiy@virtuozzo.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In many files there are header comments that contain an Author:
statement, supposedly reflecting who originally wrote the code.
In a large collaborative project like libvirt, any non-trivial
file will have been modified by a large number of different
contributors. IOW, the Author: comments are quickly out of date,
omitting people who have made significant contribitions.
In some places Author: lines have been added despite the person
merely being responsible for creating the file by moving existing
code out of another file. IOW, the Author: lines give an incorrect
record of authorship.
With this all in mind, the comments are useless as a means to identify
who to talk to about code in a particular file. Contributors will always
be better off using 'git log' and 'git blame' if they need to find the
author of a particular bit of code.
This commit thus deletes all Author: comments from the source and adds
a rule to prevent them reappearing.
The Copyright headers are similarly misleading and inaccurate, however,
we cannot delete these as they have legal meaning, despite being largely
inaccurate. In addition only the copyright holder is permitted to change
their respective copyright statement.
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
All of the ones being removed are pulled in by internal.h. The only
exception is sanlock which expects the application to include <stdint.h>
before sanlock's headers, because sanlock prototypes use fixed width
int, but they don't include stdint.h themselves, so we have to leave
that one in place.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Acked-by: Michal Privoznik <mprivozn@redhat.com>
It doesn't really make sense for us to have stdlib.h and string.h but
not stdio.h in the internal.h header.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Acked-by: Michal Privoznik <mprivozn@redhat.com>
There are only a couple remaining issues preventing it from
working on FreeBSD. Let's fix them.
With the mocking in place, qemumemlocktest and qemuxml2xmltest
can finally succeed on FreeBSD.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Clang complains about it:
error: second argument to 'va_arg' is of promotable type
'mode_t' (aka 'unsigned short'); this va_arg has undefined
behavior because arguments will be promoted to 'int'
[-Werror,-Wvarargs]
mode = va_arg(ap, mode_t);
^~~~~~
Work around the issue by passing int to va_arg() and casting
its return value to mode_t afterwards.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
We're using virFileCanonicalizePath() everywhere now, so
mocking this function has become entirely pointless.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Right-aligning backslashes when defining macros or using complex
commands in Makefiles looks cute, but as soon as any changes is
required to the code you end up with either distractingly broken
alignment or unnecessarily big diffs where most of the changes
are just pushing all backslashes a few characters to one side.
Generated using
$ git grep -El '[[:blank:]][[:blank:]]\\$' | \
grep -E '*\.([chx]|am|mk)$$' | \
while read f; do \
sed -Ei 's/[[:blank:]]*[[:blank:]]\\$/ \\/g' "$f"; \
done
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Later on we're going to need access to information about IOMMU
groups for host devices. Implement the support in virpcimock,
and start using that mock library in a few QEMU test cases.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Laine Stump <laine@laine.org>
It wasn't as great idea as I thought. Thing around stat() are
more complicated than that. Therefore we need to revert
86d1705a8a plus drop use of the macro as introduced in
later patches.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
There is some magic going on when it comes to stat() or lstat().
Basically, stat() can either be a regular function, an inline
function that calls __xstat(_STAT_VER, ...) or a macro that does
the same as the inline func. Don't ask why is that, just read the
documentation in sys/stat.h and make sure you have a bucket next
to you. Anyway, currently there will not be both stat and __xstat
symbols at the same time, as one of them gets overwritten to the
other one during compilation. But this is not true anymore once
we start chaining our mocking libraries. Therefore we need a
wrapper that calls desired function from glibc.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Instead of fakesysfsdir, which is very generic, use fakesysfspcidir and
fakesysfscgroupdir. This makes it explicit what part of the fake sysfs
filesystem they're referring to, and also leaves open the possibility of
handling files in two unrelated parts of the fake sysfs filesystem.
No functional changes.
We might need to mock files living outside PCI_SYSFS_PREFIX later on,
so it's better to treat the temporary directory we are passed via
the environment as the root of the fake filesystem and create
PCI_SYSFS_PREFIX inside it.
The environment variable name will be changed to reflect the new use
we're making of it in a later commit.
Since it is an abbreviation, PCI should always be fully
capitalized or full lower case, never Pci.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Windows doesn't allow : in filenames.
Commit 21685c955e added files with a : in
their names. This broke git operations on Windows as git is not able to
create those files on clone or pull.
Replace : with - in the offending filenames and adapt the test case.
This file is used by PCI detach and reattach APIs to probe for a driver
that handles a specific device.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
When determining if a device is behind a PCI bridge, the PCI device
class is checked by reading the config space. However, there are some
devices which have the wrong class on the config space, but the class is
initialized by Linux correctly as a PCI BRIDGE. This class can be read
by the sysfs file '/sys/bus/pci/devices/xxxx:xx:xx.x/class'.
One example of such bridge is IBM PCI Bridge 1014:03b9, which is
identified as a Host Bridge when reading the config space.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
'make distcheck' has been broken since commit 21685c9; basically,
it emulates the case of a read-only $(srcdir) (such as building
from a tarball exploded onto a CD-ROM), but we were creating our
fake pci device as a symlink into $(srcdir) and failing when that
requires opening the config file for writing:
3) testVirPCIDeviceReset ... libvirt: error : Failed to open config space file '/sys/bus/pci/devices/0000:00:01.0/config': Permission denied
Fix it by copying rather than symlinking.
* tests/virpcimock.c (make_file): Add parameter to allow binary
creation; adjust all callers.
(pci_device_new_from_stub): Copy rather than symlink.
Signed-off-by: Eric Blake <eblake@redhat.com>
While trying to debug a failure of virpcitest during 'make distcheck',
I noticed that with a VPATH build, 'cd tests; ./virpcitest' fails for
an entirely different reason. To reproduce the distcheck failure, I
had to run 'cd tests; abs_srcdir=/path/to/src ./virpcitest'. But we
document in HACKING that all of our tests are supposed to be runnable
without requiring extra environment variables.
The solution: hardcode the location of srcdir into the just-built
binaries, rather than requiring make to prepopulate environment
variables. With this, './virpcitest' passes even in a VPATH build
(provided that $(srcdir) is writable; a followup patch will fix the
conditions required by 'make distcheck'). [Note: the makefile must
still pass on directory variables to the test environment of shell
scripts, since those aren't compiled. So while this solves the case
of a compiled test, it still requires environment variables to pass
a VPATH build of any shell script test case that relies on srcdir.]
* tests/Makefile.am (AM_CFLAGS): Define abs_srcdir in all compiled
tests.
* tests/testutils.h (abs_srcdir): Quit declaring.
* tests/testutils.c (virtTestMain): Rely on define rather than
environment variable.
* tests/virpcimock.c (pci_device_new_from_stub): Rely on define.
* tests/cputest.c (mymain): Adjust abs_top_srcdir default.
* tests/qemuxml2argvtest.c (mymain): Likewise.
* tests/qemuxmlnstest.c (mymain): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
This addition, however, requires some refactoring to be done. First of
all, to match the best practice we should detach the device prior
resetting it. That's why testVirPCIDeviceDetach is detaching all devices
within 0000:00:01.0 and 0000:00:03.0 range. Then, the brand new test
will reset the 0000:00:02.0 device, so the last testVirPCIDeviceReattach
can reattach all the devices back.
In order to perform a PCI device reset, the dummy config file is not
sufficient anymore and must be replaced with real PCI config (binary
mess). Such config files are to be stored under tests/virpcitestdata/
and ought to have '.config' suffix.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
In the pci_driver_new function it is possible to set a list of
<vendor:device> IDs that the driver knows. These IDs are passed as
variable arguments and are processed the usual way using va_start() and
va_arg(). However, after all arguments has been processed, we should
call va_end() what we aren't currently doing.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This commit introduces yet another test under virpcitest:
virPCIDeviceDetach. However, in order to be able to do this, the
virpcimock needs to be extended to model the kernel behavior on PCI
device binding and unbinding (create 'driver' symlinks under the device
tree, check for device ID in driver's ID table, etc.)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Among with this test introduce virpcimock as we need to mock some
syscalls, e.g. redirect open() of a file under /sys/bus/pci to a
stub sysfs tree.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
