The aim of virSocketAddrPrefixToNetmask() is to initialize passed
virSocketAddr structure based on prefix length and family.
However, it doesn't set all members in the struct which may lead
to reads of uninitialized values:
==15421== Use of uninitialised value of size 8
==15421== at 0x50F297A: _itoa_word (in /lib64/libc-2.31.so)
==15421== by 0x510C8FE: __vfprintf_internal (in /lib64/libc-2.31.so)
==15421== by 0x5120295: __vsnprintf_internal (in /lib64/libc-2.31.so)
==15421== by 0x50F8969: snprintf (in /lib64/libc-2.31.so)
==15421== by 0x51BB602: getnameinfo (in /lib64/libc-2.31.so)
==15421== by 0x496DEE0: virSocketAddrFormatFull (virsocketaddr.c:486)
==15421== by 0x496DD9F: virSocketAddrFormat (virsocketaddr.c:444)
==15421== by 0x11871F: networkDnsmasqConfContents (bridge_driver.c:1404)
==15421== by 0x1118F5: testCompareXMLToConfFiles (networkxml2conftest.c:48)
==15421== by 0x111BAF: testCompareXMLToConfHelper (networkxml2conftest.c:112)
==15421== by 0x112679: virTestRun (testutils.c:142)
==15421== by 0x111D09: mymain (networkxml2conftest.c:144)
==15421== Uninitialised value was created by a stack allocation
==15421== at 0x1175D2: networkDnsmasqConfContents (bridge_driver.c:1056)
All callers expect the function to initialize the structure
fully.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Laine Stump <laine@redhat.com>
In 88957116c9 I've switched to -machine memory-backend=ID and
-object memory-backend-* because QEMU is obsoleting -mem-path
and -mem-prealloc. However, what I did not foresee was that using
-machine memory-backend in combination with -numa is not allowed
in QEMU. This was reported upstream and fortunately not released
yet.
The problem is that if domain has NUMA nodes then we will
generate memory-backend-* objects for NUMA nodes (because if QEMU
is new enough to expose default RAM ID it also supports -numa
memdev=) and adding non-NUMA memory backend is wrong.
Reported-by: Masayoshi Mizuma <msys.mizuma@gmail.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The @checkMACAddress string is allocated in
virVMXGetConfigString() but never freed.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
In our wrapper of g_dbus_connection_call_sync() in
virfirewalltest a string is duplicated and added onto a
virStringList. This leads to a memory leak because
virStringListAdd() duplicates the string itself.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
With us switching to glib more and more it is easy to get things
wrong (as can be seen in the previous commit). Set G_DEBUG
variable to "fatal-warnings" which causes GLib to abort the
program at the first call to g_warning() or g_critical().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
I've found two files under qemuxml2xmloutdata/ that are the same
as in qemuxml2argvdata/. Replace them with symlinks.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
So far, Libvirt configures memory-backend-* for memory hotplug,
possibly NUMA nodes and in a few other cases. This patch
switches to constructing the memory-backend-* command line for
all cases. To keep ability to migrate guests a little hack is
used: the ID of the object is set to the one that QEMU uses
internally anyways. These IDs are stable (first started to appear
somewhere around v0.13.0-rc0~96) and can't change.
In fact, this patch does exactly what QEMU does internally. The
reason for moving the logic into Libvirt is that QEMU wants to
deprecate the old style of specifying memory.
So far, only x84_64 test cases are changed, because tests for
other architectures use older capabilities, which still lack the
QEMU_CAPS_MACHINE_MEMORY_BACKEND capability and they don't report
the RAM ID.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1836043
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The machine structure has another (optional) attribute:
default-ram-id, which specifies the alias of the default RAM
object. While the alias is private, it can never change in order
to not break migration. QEMU uses the alias when allocating
regular, not NUMA memory. In order to switch to new command line
and maintain migration, save this ID.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The objects at @def and @mem pointers are only read and not
written. Make the arguments const to make that explicit.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
If a domain was using hugepages through memory-backend-file or
via -mem-path, we would turn prealloc on. But we are not doing
that for memory-backend-memfd. Fix this, because we need QEMU to
fully allocate hugepages.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
If user specifies immediate memory allocation in the domain XML,
they want QEMU to fully allocate its memory. And if the memory
was allocated using plain '-m' then we would honour it. But, if a
memory backend is used, then we don't set the prealloc attribute
of the backend.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
All three memory backends (-file, -ram and -memfd) have .prealloc
attribute. Since we are setting it only for -file, the
corresponding code lives only under if() that handles that
specific backend. But in near future we will want to set the
attribute for other backends too. Therefore, move the
corresponding code outside of the if().
This causes some .argv files to be changed, but the only change
happening there is move of the attribute (best viewed with:
'git show --color-words=.').
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
This originally started as bug 1595525 in which namespaces and
libusb used in QEMU were not playing nicely with each other. The
problem was that libusb built a cache of USB devices it saw
(which was a very limited set because of the namespace) and then
expected to receive udev events to keep the cache in sync. But
those udev events didn't come because on hotplug when we mknod()
devices in the namespace no udev event is generated. And what is
worse - libusb failed to open a device that wasn't in the cache.
Without going further into what the problem was, libusb added a
new API for opening USB devices that avoids using cache which
QEMU incorporated and exposes under "hostdevice" attribute.
What is even nicer is that QEMU uses qemu_open() for path
provided in the attribute and thus FD passing could be used.
Except qemu_open() expects so called FD sets instead of `getfd'
and these are not implemented in libvirt, yet.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1877218
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This capability tracks whether "usb-host" device has "hostdevice"
attribute. This attribute allows us to specify full path to the
USB device ("/dev/bus/usb/$bus/$dev") but more importantly, since
QEMU uses qemu_open() for this attribute it allows us to pass
pre-opened FD and have QEMU not bother with opening the file at
all.
The attribute was added in v5.1.0-rc0~71^2~1 QEMU commit.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
After v6.5.0-rc1~148 we started to rectify vCPU to guest NUMA
assignment - if there is a vCPU not assigned to any guest NUMA
node it is automatically assigned to node #0.
A month later I've made it possible to define guest NUMA nodes
without vCPUs (v6.6.0-rc1~250) - this is needed because of HMAT.
As a part of that I fixed all callers of
virDomainNumaGetNodeCpumask() (which returns a bitmap of vCPUs for
given node) to handle case when NULL is returned (i.e. no vCPUs
assigned to given node). But of course my patch was written
before aforementioned vCPU rectify patch but merged afterwards
and hence I missed the virDomainNumaFillCPUsInNode() caller.
And because we are dealing with a NULL pointer, of course this
leads to a crash. Just try to define a domain with at least two
NUMA nodes and no vCPU assignment to any of the nodes.
Fixes: a26f61ee0c
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1880289
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
In v6.7.0-rc1~86 I've tried to fix a problem where we were not
detecting NUMA nodes properly because we misused behaviour of a
libnuma API and as it turned out the behaviour was correct for
hosts with 64 CPUs in one NUMA node. So I changed the code to use
nodemask_isset(&numa_all_nodes, ..) instead and it fixed the
problem on such hosts. However, what I did not realize is that
numa_all_nodes does not reflect all NUMA nodes visible to
userspace, it contains only those nodes that the process
(libvirtd) an allocate memory from, which can be only a subset of
all NUMA nodes. The bitmask that contains all NUMA nodes visible
to userspace and which one I should have used is: numa_nodes_ptr.
For curious ones:
https://github.com/numactl/numactl/commit/4a22f2238234155e11e3e2717c011864722b767b
And as I was fixing virNumaGetNodeCPUs() I came to realize that
we already have a function that wraps the correct bitmask:
virNumaNodeIsAvailable().
Fixes: 24d7d85208
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1876956
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
This function was introduced in the 2.0.6 release which happened
in December 2010. I think it is safe to assume that all libnuma
we deal with have the function.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In 8e1804f9f6 I've tried to fix the following use case: domain
is started with path to UEFI only and relies on libvirt to figure
out corresponding NVRAM template to create a per-domain copy
from. The fix consisted of having a check tailored exactly for
this use case and if it's hit then using FW autoselection to
figure it out. Unfortunately, the NVRAM template is not saved in
the inactive XML (well, the domain might be transient anyway).
Then, as a part of that check we see whether the per-domain copy
doesn't exist already and if it does then no template is looked
up hence no template will appear in the live XML.
This works, until the domain is migrated. At the destination, the
per-domain copy will not exist so we need to know the template to
create the per-domain copy from. But we don't even get to the
check because we are not starting a fresh new domain and thus the
qemuFirmwareFillDomain() function quits early.
The solution is to switch order of these two checks. That is
evaluate the check for the old style before checking flags.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1852910
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Even though this was brought up in upstream discussion [1] it
missed my patches: users should prefer <oemStrings/> over fwcfg.
The reason is that fwcfg is considered somewhat internal to QEMU
and it has limited number of slots and neither of these applies
to <oemStrings/>.
While I'm at it, I'm fixing the example too (because it contains
incorrect element name) and clarifying sysfs/ exposure.
1: https://www.redhat.com/archives/libvir-list/2020-May/msg00957.html
Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
When starting a domain with <numatune/> set libvirt translates
given NUMA nodes into a set of host CPUs which is then used to
QEMU process affinity. But, if the numatune contains a
non-existent NUMA node then the translation fails with no error
reported. This is because virNumaNodesetToCPUset() calls
virNumaGetNodeCPUs() and expects it to report an error on
failure. Well, it does except for non-existent NUMA nodes. While
this behaviour might look strange it is actually desired because
of how we construct host capabilities. The virNumaGetNodeCPUs()
is called from virCapabilitiesHostNUMAInitReal() where we do not
want any error reported for non-existent NUMA nodes.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1724866
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
A wireshark plugin must declare what major and minor version it
was built with as these are checked when wireshark loads plugins.
On the top of that, we use major + minor + micro to adapt to
changed API between releases. So far, we were getting these
version numbers from wireshark/config.h.
And while most distributions install wireshark/config.h file some
don't. On distros shipping it it's hack^Wsaved during built by
packaging system and installed later. But some distros are not
doing that. At least not for new enough wireshark because as of
wireshark's commit v2.9.0~1273 the ws_version.h is installed
which contains the version macros we need and is installed by
wireshark itself.
But of course, some distros which have new enough wireshark
packaged do not ship ws_version.h and stick to the hack. That is
why we can't simply bump the minimal version and switch to the
new header file. We need a configure check and adopt our code to
deal with both ways. At least for the time being.
Based on Andrea's original patch:
https://www.redhat.com/archives/libvir-list/2020-September/msg00156.html
Closes: https://gitlab.com/libvirt/libvirt/-/issues/74
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
When starting a domain, qemuProcessLaunch() iterates over all
VIR_PERF_EVENT_* values and (possibly) enables them. While there
is nothing wrong with the code, the for loop where it's done makes
it harder to jump onto next block of code.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When building and populating domain NS a couple of functions are
called that append paths to a string list. This string list is
then inspected, one item at the time by
qemuNamespacePrepareOneItem() which gathers all the info for
given path (stat buffer, possible link target, ACLs, SELinux
label) using qemuNamespaceMknodItemInit(). If the path needs to
be created in the domain's private /dev then it's added onto this
qemuNamespaceMknodData list which is freed later in the process.
But, if the path does not need to be created in the domain's
private /dev, then the memory allocated by
qemuNamespaceMknodItemInit() is not freed anywhere leading to a
leak.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In the past we had to declare @cfg and then explicitly unref it.
But now, with glib we can use g_autoptr() which will do the unref
automatically and thus is more bulletproof.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Laine Stump <laine@redhat.com>
In the qemuInterfacePrepareSlirp() function, the qemu driver
config is obtained (via virQEMUDriverGetConfig()), but it is
never unrefed leading to mangled refcounter.
Fixes: 9145b3f1cc
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Laine Stump <laine@redhat.com>
In 6.7.0 release I've changed how domain namespace is built and
populated. Previously it used to be done from a pre-exec hook
(ran in the forked off child, just before dropping all privileges
and exec()-ing QEMU), which not only meant we had to have two
different code paths for creating a node in domain's namespace
(one for this pre-exec hook, the other for hotplug ran from the
daemon), it also proved problematic because it was leaking FDs
into QEMU process.
To mitigate this problem, we've not only ditched libdevmapper
from the NS population process, I've also dropped the pre-exec
code and let the NS be populated from the daemon (using the
hotplug code). But, I was not careful when doing so, because the
pre-exec code was tolerant to files that doesn't exist, while
this new code isn't. For instance, the very first thing that is
done when the new NS is created is it's populated with
@defaultDeviceACL which contain files like /dev/null, /dev/zero,
/dev/random and /dev/kvm (and others). While the rest will
probably exist every time, /dev/kvm might not and thus the new
code I wrote has to be tolerant to that.
Of course, users can override the @defaultDeviceACL (by setting
cgroup_device_acl in qemu.conf) and remove /dev/kvm (which is
acceptable workaround), but we definitely want libvirt to work
out of the box even on hosts without KVM.
Fixes: 9048dc4e62
Reported-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As it turned out my previous commits which switched from HAVE_ to
WITH_ and dropped stdarg.h detection were a bit too aggressive.
Because of reasons described in 9ea3424a17 we need to define
HAVE_STDARG_H before including readline otherwise macos build
fails. Honestly, I still don't fully understand the problem so I
am not going to bother you with "explanation".
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This allows us to drop include of readline header files from
virsh.c and virt-admin.c because they needed it only because of
the add_history() function.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Currently, we are mixing: #if HAVE_BLAH with #if WITH_BLAH.
Things got way better with Pavel's work on meson, but apparently,
mixing these two lead to confusing and easy to miss bugs (see
31fb929eca for instance). While we were forced to use HAVE_
prefix with autotools, we are free to chose our own prefix with
meson and since WITH_ prefix appears to be more popular let's use
it everywhere.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In packet-libvirt.c in wireshark dissector we include rpc/types.h
but guard the include with a condition (that is supposed to be
true if we detected during configure phase that the host system
has the header file). Thing is, it looks like we were never doing
the configure check and thus the file was never included and yet,
the NSS plugin works. Drop the include then.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In meson.build, we check for presence of unshare() function
explicitly (even though there is the function usability check a
few hundred lines below), but never have any code depending on
HAVE_UNSHARE. The same applies to stdarg.h and sys/sysctl.h
header files - either we simply include them or guard their
include using different conditionals.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
There are couple of conditional #includes at the beginning of
virfile.c and they try to be nice and document #endifs. But they
are mostly wrong because either they have the condition in the
comment inverted or the comment refers to a different condition
than they belong to. Just remove the comments as these #includes
are single line mostly.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
There are two places where we try to check whether the host
system has net/if.h before including it. But the check is missing
'_H' suffix.
Fixes: 7f3eb533f4
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When switching to meson, some of HAVE_* macros were renamed to
WITH_ because they did not reflect whether the build platform has
or doesn't have something, but whether we are building with some
functionality turned on or off. This is the case with
HAVE_BSD_NSS macro too. As a result, the NSS plugin built on BSD
did not expose nss_module_register() function which made the
plugin unusable:
https://www.redhat.com/archives/libvir-list/2020-September/msg00000.html
Fixes: c742687055
Reported-by: Roman Bogorodskiy <bogorodskiy@gmail.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Roman Bogorodskiy <bogorodskiy@gmail.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
When editing a domain with hotplug enabled, I removed the only
NUMA node it had and got no error. I got the error later though,
when starting the domain. This is not as user friendly as it can
be. Move the validation call out from command line generator and
into domain validator (which is called prior to starting cmd line
generation anyway).
When doing this, I had to remove memory-hotplug-nonuma xml2xml
test case because there is no way the test case can succeed,
obviously.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
So far, the <cell/> element can have two types of children
elements: <distances/> and <cache/> (which can be repeated more
times). However, there is no reason to require specific order in
input XML. Allow elements to be interleaved.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Even if namespaces are disabled, then due to a missing check at the
beginning of qemuDomainBuildNamespace(), the domain startup code
still tries to populate (nonexistent) domain's namespace.
Fixes: 8da362fe62
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
This function is not changing the domain definition, it's only
reading from it. The function is going to be used from another
function which already takes const virDomainDef. Make the @def
const to avoid typecasting it.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This is similar to one of previous patches.
When receiving stream (on virStorageVolUpload() and subsequent
virStreamSparseSendAll()) we may receive a hole. If the volume we
are saving the incoming data into is a regular file we just
lseek() and ftruncate() to create the hole. But this won't work
if the file is a block device. If that is the case we must write
zeroes so that any subsequent reader reads nothing just zeroes
(just like they would from a hole in a regular file).
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1852528
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
This is very similar to previous commit.
The virshStreamInData() callback is used by virStreamSparseSendAll()
to detect whether the file the data is read from is in data or hole
section. The SendAll() will then send corresponding type of virStream
message to make server create a hole or write actual data. But the
callback uses virFileInData() even for block devices, which results in
an error. Just like in previous commit, emulate a DATA section
for block devices.
Partially resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1852528
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
When handling sparse stream, a thread is executed. This thread
runs a read() or write() loop (depending what API is called; in
this case it's virStorageVolDownload() and this the thread run
read() loop). The read() is handled in virFDStreamThreadDoRead()
which is then data/hole section aware, meaning it uses
virFileInData() to detect data and hole sections and sends
TYPE_DATA or TYPE_HOLE virStream messages accordingly.
However, virFileInData() does not work with block devices. Simply
because block devices don't have data and hole sections. What we
can do though, is to mimic being always in a DATA section.
Partially resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1852528
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
This callback is called when the server sends us STREAM_HOLE
meaning there is no real data, only zeroes. For regular files
we would just seek() beyond EOF and ftruncate() to create the
hole. But for block devices this won't work. Not only we can't
seek() beyond EOF, and ftruncate() will fail, this approach won't
fill the device with zeroes. We have to do it manually.
Partially resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1852528
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
We can't use virFileInData() with block devices, but we can
emulate being in data section all the time (vol-upload case).
Alternatively, we can't just lseek() beyond EOF with block
devices to create a hole, we will have to write zeroes
(vol-download case). But to decide we need to know if the FD we
are reading data from / writing data to is a block device. Store
this information in _virshStreamCallbackData.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
These callback will need to know more that the FD they are
working on. Pass the structure that is passed to other stream
callbacks (e.g. virshStreamSource() or virshStreamSourceSkip())
instead of inventing a new one.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
For libvirt, the volume is just a binary blob and it doesn't
interpret data on volume upload/download. But as it turns out,
this unspoken assumption is not clear to our users. Document it
explicitly.
Suggested in: https://bugzilla.redhat.com/show_bug.cgi?id=1851023#c17
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
In a very distant past, we came around machines that has not
continuous node IDs. This made us error out when constructing
capabilities XML. We resolved it by utilizing strange behaviour
of numa_node_to_cpus() in which it returned a mask with all bits
set for a non-existent node. However, this is not the only case
when it returns all ones mask - if the node exists and has enough
CPUs to fill the mask up (e.g. 128 CPUs).
The fix consists of using nodemask_isset(&numa_all_nodes, ..)
prior to calling numa_node_to_cpus() to determine if the node
exists.
Fixes: 628c935747
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1860231
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
After previous cleanups, some labels in some functions have
nothing but 'return' statement in them. Drop the labels and
replace 'goto'-s with respective return statements.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Again, instead of closing FDs explicitly, we can automatically
close them when they go out of their respective scopes.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
A cleanup function can be declared for virFDStreamMsg type so
that the structure doesn't have to be freed explicitly.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
All callers of virFDStreamMsgQueuePush() have the same pattern:
they explicitly set @msg passed to NULL to avoid freeing it later
on. Well, the function can take address of the pointer and clear
it for them.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
The buffer that allocated in the virFDStreamThreadDoRead() can be
automatically freed, or if saved into the message structure it
can be stolen.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
So far, only ENOENT is ignored (to deal with kernels without
devmapper). However, as reported on the list, under certain
scenarios a different error can occur. For instance, when libvirt
is running inside a container which doesn't have permissions to
talk to the devmapper. If this is the case, then open() returns
-1 and sets errno=EPERM.
Assuming that multipath devices are fairly narrow use case and
using them in a restricted container is even more narrow the best
fix seems to be to ignore all open errors BUT produce a warning
on failure. To avoid flooding logs with warnings on kernels
without devmapper the level is reduced to a plain debug message.
Reported-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Reviewed-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
When adding support for HMAT, in f0611fe883 I've introduced a
check which aims to validate /domain/cpu/numa/interconnects. As a
part of that, there is a loop which checks whether all <latency/>
with @cache attribute refer to an existing cache level. For
instance:
<cpu mode='host-model' check='partial'>
<numa>
<cell id='0' cpus='0-5' memory='512000' unit='KiB' discard='yes'>
<cache level='1' associativity='direct' policy='writeback'>
<size value='8' unit='KiB'/>
<line value='5' unit='B'/>
</cache>
</cell>
<interconnects>
<latency initiator='0' target='0' cache='1' type='access' value='5'/>
<bandwidth initiator='0' target='0' type='access' value='204800' unit='KiB'/>
</interconnects>
</numa>
</cpu>
This XML defines that accessing L1 cache of node #0 from node #0
has latency of 5ns.
However, the loop was not written properly. Well, the check in
it, as it was always checking for the first cache in the target
node and not the rest. Therefore, the following example errors
out:
<cpu mode='host-model' check='partial'>
<numa>
<cell id='0' cpus='0-5' memory='512000' unit='KiB' discard='yes'>
<cache level='3' associativity='direct' policy='writeback'>
<size value='10' unit='KiB'/>
<line value='8' unit='B'/>
</cache>
<cache level='1' associativity='direct' policy='writeback'>
<size value='8' unit='KiB'/>
<line value='5' unit='B'/>
</cache>
</cell>
<interconnects>
<latency initiator='0' target='0' cache='1' type='access' value='5'/>
<bandwidth initiator='0' target='0' type='access' value='204800' unit='KiB'/>
</interconnects>
</numa>
</cpu>
This errors out even though it is a valid configuration. The L1
cache under node #0 is still present.
Fixes: f0611fe883
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Laine Stump <laine@redhat.com>
In one of my latest patch (v6.6.0~30) I was trying to remove
libdevmapper use in favor of our own implementation. However, the
code did not take into account that device mapper can be not
compiled into the kernel (e.g. be a separate module that's not
loaded) in which case /proc/devices won't have the device-mapper
major number and thus virDevMapperGetTargets() and/or
virIsDevMapperDevice() fails.
However, such failure is safe to ignore, because if device mapper
is missing then there can't be any multipath devices and thus we
don't need to allow the deps in CGroups, nor create them in the
domain private namespace, etc.
Fixes: 2249455654
Reported-by: Andrea Bolognani <abologna@redhat.com>
Reported-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Tested-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
The device mapper major is needed in virIsDevMapperDevice() which
determines whether given device is managed by device-mapper. This
number is obtained by parsing /proc/devices and then stored in a
global variable so that the file doesn't have to be parsed again.
However, as it turns out this logic is flawed - the major number
is not static and can change as it can be specified as a
parameter when loading the dm-mod module.
Unfortunately, I was not able to come up with a good solution and
thus the /proc/devices file is being parsed every time we need
the device mapper major.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Tested-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
We can use qemuDomainSetupInput() to obtain the path that we
need to unlink() from within domain's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
We can use qemuDomainSetupRNG() to obtain the path that we
need to unlink() from within domain's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
We can use qemuDomainSetupChardev() to obtain the path that we
need to unlink() from within domain's namespace. Note, while
previously we unlinked only VIR_DOMAIN_CHR_TYPE_DEV chardevs,
with this change we unlink some other types too - exactly those
types we created when plugging the device in.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
We can use qemuDomainSetupMemory() to obtain the path that we
need to unlink() from within domain's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In my attempt to deduplicate the code, we can use
qemuDomainSetupHostdev() to obtain the list of paths to unlink
and then pass it to qemuDomainNamespaceUnlinkPaths() to unlink
them in a single fork.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
So far, the only caller qemuDomainNamespaceUnlinkPath() will
always pass a single path to unlink, but similarly to
qemuDomainNamespaceMknodPaths() - there are a few callers that
would like to pass two or more files to unlink at once (held in a
string list). Make the @paths argument a string list then.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Simirarly to qemuDomainAttachDeviceMknodHelper() which was
modified just a couple of commits ago, modify the unlink helper
which is called on device detach so that it can unlink multiple
files in one go instead of forking off for every single one of
them.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
After previous cleanup, creating /dev nodes from pre-exec hook is
no longer needed and thus can be removed.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain SEV into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain loader into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain RNGs into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain inputs into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain graphics (render node)
into daemon's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain TPM into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain chardevs into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain memory (nvdimms) into
daemon's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain hostdevs into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in one of previous commits, populating domain's
namespace from pre-exec() hook is dangerous. This commit moves
population of the namespace with domain disks into daemon's
namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in previous commit, populating domain's namespace
from pre-exec() hook is dangerous. This commit moves population
of the namespace with basic /dev nodes (e.g. /dev/null, /dev/kvm,
etc.) into daemon's namespace.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Okay, here is the deal. Currently, the way we build namespace is
very fragile. It is done from pre-exec hook when starting a
domain, after we mass closed all FDs and before we drop
privileges and exec() QEMU. This fact poses some limitations onto
the namespace build code, e.g. it has to make sure not to keep
any FD opened (not even through a library call), because it would
be leaked to QEMU. Also, it has to call only async signal safe
functions. These requirements are hard to meet - in fact as of my
commit v6.2.0-rc1~235 we are leaking a FD into QEMU by calling
libdevmapper functions.
To solve this issue and avoid similar problems in the future, we
should change our paradigm. We already have functions which can
populate domain's namespace with nodes from the daemon context.
If we use them to populate the namespace and keep only the bare
minimum in the pre-exec hook, we've mitigated the risk.
Therefore, the old qemuDomainBuildNamespace() is renamed to
qemuDomainUnshareNamespace() and new qemuDomainBuildNamespace()
function is introduced. So far, the new function is basically a
NOP and domain's namespace is still populated from the pre-exec
hook - next patches will fix it.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The aim to make it look as close to
qemuDomainNamespaceSetupDisk() as possible. The latter will call
the former and this change makes that diff easier to read.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Every caller does the same - counts the number of items in a
string list they have, only to pass the number to
qemuDomainNamespaceMknodPaths(). This is needless - the function
can accept the string list and count the items itself.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
While the previous commit prepared the helper function run in a
forked off helper (with corresponding struct), this commit
modifies the caller, which now create all files requested in a
single process and does not fork off for every single path.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
So far, when attaching a device needs two or more /dev nodes
created into a domain, we fork off and run the helper for every
node separately. For majority of devices this is okay, because
they need no or one node created anyway. But the idea is to use
this attach code to build the namespace when starting a domain,
in which case there will be way more nodes than one.
To achieve this, the recursive approach for handling symlinks has
to be turned into an iterative one.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When attaching a device into a domain, the corresponding /dev
node might need to be created in the domain's namespace. For some
types of files we call mknod(), for symlinks we call symlink(),
but for others - which exist in the host namespace - we need to
so called 'bind mount' them (which is a way of passing a
file/directory between mount namespaces). There is this condition
in qemuDomainAttachDeviceMknodRecursive() which decides whether a
bind mount will be used, move it into a separate function so that
it can be reused later.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This structure is going to be used from not only device attach
code, but also when building the namespace. Moreover, the code
lives in a separate file so the chances of clashing with another
name are minimal.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
It's not really a problem since this is a helper process that
dies as soon as the helper function returns, but the cleanup code
will be replaced with a function soon and this change prepares
the code for that.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
While qemuDomainNamespaceMknodPaths() doesn't actually create
files in the namespace in one go (it forks for each path), it a
few commits time it will.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Functions that create a device node after domain startup (used
from hotplug) will get a list of paths they want to create and
eventually call qemuDomainNamespaceMknodPaths() which then checks
whether domain mount namespace is enabled in the first place.
Alternatively, on device hotunplug, we might want to delete a
path inside domain namespace in which case
qemuDomainNamespaceUnlinkPaths() checks whether the namespace is
enabled. While this is not dangerous, it certainly burns a couple
of CPU cycles needlessly.
Check whether mount namespace is enabled upfront.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
There is a lot of functions called from
qemuDomainBuildNamespace() that accept @cfg
(virQEMUDriverConfigPtr) as an argument and don't use it.
Historically, it was done so that all qemuDomainSetupAll*()
functions look the same.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The name of this function is not very helpful, because it doesn't
create anything, it just flips a bit in a bitmask when domain is
starting up. Move the function internals into qemu_process.c and
forget the function ever existed.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The qemu_domain.c file is big as is and we should split it into
separate semantic blocks. Start with code that handles domain
namespaces.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In contrast with other releases, this time NEWS is pretty much
updated. I've identified only a couple of features/bug fixes
worth mentioning that were not mentioned yet.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Document the fix of leaking /dev/mapper/control to QEMU (fixed in
v6.6.0-rc1-3-g2249455654).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
CVE-2020-14339
When building domain's private /dev in a namespace, libdevmapper
is consulted for getting full dependency tree of domain's disks.
The reason is that for a multipath devices all dependent devices
must be created in the namespace and allowed in CGroups.
However, this approach is very fragile as building of namespace
happens in the forked off child process, after mass close of FDs
and just before dropping privileges and execing QEMU. And it so
happens that when calling libdevmapper APIs, one of them opens
/dev/mapper/control and saves the FD into a global variable. The
FD is kept open until the lib is unlinked or dm_lib_release() is
called explicitly. We are doing neither.
However, the virDevMapperGetTargets() function is called also
from libvirtd (when setting up CGroups) and thus has to be thread
safe. Unfortunately, libdevmapper APIs are not thread safe (nor
async signal safe) and thus we can't use them. Reimplement what
libdevmapper would do using plain C (ioctl()-s, /proc/devices
parsing, /dev/mapper dirwalking, and so on).
Fixes: a30078cb83
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1858260
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Since we have VIR_AUTOSTRINGLIST we can use it to free string
lists used in the function automatically.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
There are two distinct WITH_DEVMAPPER sections in the file, for
different functions each. Rearrange the code to make some of
future commits smaller.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
In 9536379da4 and 8b0cb0e666 I've tried to call
virNetSocketCheckProtocolByLookup() only if we are suspecting the
host is IPv4 or IPv6 capable (because we've found an interface
with such address). However, the code was missing dereference of
the boolean variables and thus was comparing pointers against
NULL.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Historically, if we found an interface with an IPv6 address we
did not blindly trust that host is IPv6 capable (as in we can
successfully translate IPv4 addresses) but used getaddrinfo() to
confirm it. Turns out, we have use the same argument for IPv4.
For instance, in an namespace created by the following steps,
getaddrinfo("127.0.0.1", ...) fails (demonstrating by "Socket
TCP/IPv4 Accept" test case failing in virnetsockettest):
unshare -n
ip link set lo up
ip link add dummy0 type dummy
ip link set dummy0 up
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik