Noticed a misuse of 'to' while testing my event regression under
polkit ACLs, and decided to review the entire conf files for
other legibility bugs.

* daemon/libvirtd.conf: Use correct grammar.
* src/qemu/qemu.conf: Likewise.
Signed-off-by: NEric Blake <eblake@redhat.com>

https://bugzilla.redhat.com/show_bug.cgi?id=1058839

Commit f9f56340 for CVE-2014-0028 almost had the right idea - we
need to check the ACL rules to filter which events to send.  But
it overlooked one thing: the event dispatch queue is running in
the main loop thread, and therefore does not normally have a
current virIdentityPtr.  But filter checks can be based on current
identity, so when libvirtd.conf contains access_drivers=["polkit"],
we ended up rejecting access for EVERY event due to failure to
look up the current identity, even if it should have been allowed.

Furthermore, even for events that are triggered by API calls, it
is important to remember that the point of events is that they can
be copied across multiple connections, which may have separate
identities and permissions.  So even if events were dispatched
from a context where we have an identity, we must change to the
correct identity of the connection that will be receiving the
event, rather than basing a decision on the context that triggered
the event, when deciding whether to filter an event to a
particular connection.

If there were an easy way to get from virConnectPtr to the
appropriate virIdentityPtr, then object_event.c could adjust the
identity prior to checking whether to dispatch an event.  But
setting up that back-reference is a bit invasive.  Instead, it
is easier to delay the filtering check until lower down the
stack, at the point where we have direct access to the RPC
client object that owns an identity.  As such, this patch ends
up reverting a large portion of the framework of commit f9f56340.
We also have to teach 'make check' to special-case the fact that
the event registration filtering is done at the point of dispatch,
rather than the point of registration.  Note that even though we
don't actually use virConnectDomainEventRegisterCheckACL (because
the RegisterAny variant is sufficient), we still generate the
function for the purposes of documenting that the filtering
takes place.

Also note that I did not entirely delete the notion of a filter
from object_event.c; I still plan on using that for my upcoming
patch series for qemu monitor events in libvirt-qemu.so.  In
other words, while this patch changes ACL filtering to live in
remote.c and therefore we have no current client of the filtering
in object_event.c, the notion of filtering in object_event.c is
still useful down the road.

* src/check-aclrules.pl: Exempt event registration from having to
pass checkACL filter down call stack.
* daemon/remote.c (remoteRelayDomainEventCheckACL)
(remoteRelayNetworkEventCheckACL): New functions.
(remoteRelay*Event*): Use new functions.
* src/conf/domain_event.h (virDomainEventStateRegister)
(virDomainEventStateRegisterID): Drop unused parameter.
* src/conf/network_event.h (virNetworkEventStateRegisterID):
Likewise.
* src/conf/domain_event.c (virDomainEventFilter): Delete unused
function.
* src/conf/network_event.c (virNetworkEventFilter): Likewise.
* src/libxl/libxl_driver.c: Adjust caller.
* src/lxc/lxc_driver.c: Likewise.
* src/network/bridge_driver.c: Likewise.
* src/qemu/qemu_driver.c: Likewise.
* src/remote/remote_driver.c: Likewise.
* src/test/test_driver.c: Likewise.
* src/uml/uml_driver.c: Likewise.
* src/vbox/vbox_tmpl.c: Likewise.
* src/xen/xen_driver.c: Likewise.
Signed-off-by: NEric Blake <eblake@redhat.com>

I noticed this problem when adding systemd support to netcf, because I
setup the configure.ac to automatically prefer using systemd over
initscripts when possible - although I had copied the
install-data-local target from the example of libvirt's
"libvirt-guests" service more or less verbatim, "make distcheck" would
fail because it was trying to install the service file directly into
/lib/systemd/system rather than into
/home/user/some/unimportant/name/lib/systemd/system.

This is caused by the install/uninstall rules for the systemd unit
files relying on $(DESTDIR) pointing the installed files to the right
place, but in reality $(DESTDIR) is empty during this part of make
distcheck - it instead sets $(prefix) with the toplevel directory used
for its test build/install/uninstall cycle.

(This problem hasn't been seen when running "make distcheck" in
libvirt because libvirt will never build/install systemd support
unless explicitly told to do so on the configure commandline, and
"make distcheck" doesn't put the "--with-initscript=..." option on the
configure commandline.)

I verified that the same problem does exist in libvirt by modifying
libvirt's configure.ac to set:

  init_systemd=yes
  with_init_script=systemd+redhat

This forces a build/install of the systemd unit files during
distcheck, which yields an error like this:

/usr/bin/install -c -m 644 virtlockd.service \
  /lib/systemd/system/
libtool: install: warning: relinking `libvirt-qemu.la'
/usr/bin/install: cannot remove '/lib/systemd/system/virtlockd.service': Permission denied
make[4]: *** [install-systemd] Error 1

After adding $(prefix) to all the definitions of SYSTEMD_UNIT_DIR,
make distcheck now completes successfully with the modified
configure.ac, and the above lines change to something like this:

/usr/bin/install -c -m 644 virtlockd.service \
  /home/laine/devel/libvirt/libvirt-1.2.1/_inst/lib/systemd/system/

We haven't had a release with network events yet, so we are free
to fix the RPC so that it actually does what we want.  Doing
client-side filtering of per-network events is inefficient if a
connection is only interested in events on a single network out
of hundreds available on the server.  But to do server-side
per-network filtering, the server needs to know which network
to filter on - so we need to pass an optional network over on
registration.  Furthermore, it is possible to have a client with
both a global and per-network filter; in the existing code, the
server sends only one event and the client replicates to both
callbacks.  But with server-side filtering, the server will send
the event twice, so we need a way for the client to know which
callbackID is sending an event, to ensure that the client can
filter out events from a registration that does not match the
callbackID from the server.  Likewise, the existing style of
deregistering by eventID alone is fine; but in the new style,
we have to remember which callbackID to delete.

This patch fixes the RPC wire definition to contain all the
needed pieces of information, and hooks into the server and
client side improvements of the previous patches, in order to
switch over to full server-side filtering of network events.
Also, since we fixed this in time, all released versions of
libvirtd that support network events also support per-network
filtering, so we can hard-code that assumption into
network_event.c.

Converting domain events to server-side filtering will require
the introduction of new RPC numbers, as well as a server
feature bit that the client can use to tell whether to use
old-style (server only supports global events) or new-style
(server supports filtered events), so that is deferred to a
later set of patches.

* src/conf/network_event.c (virNetworkEventStateRegisterClient):
Assume server-side filtering.
* src/remote/remote_protocol.x
(remote_connect_network_event_register_any_args): Add network
argument.
(remote_connect_network_event_register_any_ret): Return callbackID
instead of count.
(remote_connect_network_event_deregister_any_args): Pass
callbackID instead of eventID.
(remote_connect_network_event_deregister_any_ret): Drop unused
type.
(remote_network_event_lifecycle_msg): Add callbackID.
* daemon/remote.c
(remoteDispatchConnectNetworkEventDeregisterAny): Drop unused arg,
and deal with callbackID from client.
(remoteRelayNetworkEventLifecycle): Pass callbackID.
(remoteDispatchConnectNetworkEventRegisterAny): Likewise, and
recognize non-NULL network.
* src/remote/remote_driver.c
(remoteConnectNetworkEventRegisterAny): Pass network, and track
server side id.
(remoteConnectNetworkEventDeregisterAny): Deregister by callback id.
(remoteNetworkBuildEventLifecycle): Pass remote id to event queue.
* src/remote_protocol-structs: Regenerate.
Signed-off-by: NEric Blake <eblake@redhat.com>

Right now, the daemon side of RPC events is hard-coded to at most
one callback per eventID.  But when there are hundreds of domains
or networks coupled and multiple conections, then sending every
event to every connection that wants an event, even for the
connections that only care about events for a particular object,
is inefficient.  In order to track more than one callback in the
server, we need to store callbacks by more than just their
eventID.  This patch rearranges the daemon side to store network
callbacks in a dynamic array, which can eventually be used for
multiple callbacks of the same eventID, although actual behavior
is unchanged without further patches to the RPC protocol.  For
ease of review, domain events are saved for a later patch, as
they touch more code.

While at it, fix a bug where a malicious client could send a
negative eventID to cause network event registration to access
outside of array bounds (thankfully not a CVE, since domain
events were already doing the bounds check, and since network
events have not been released).

* daemon/libvirtd.h (daemonClientPrivate): Alter the tracking of
network events.
* daemon/remote.c (daemonClientEventCallback): New struct.
(remoteEventCallbackFree): New function.
(remoteClientInitHook, remoteRelayNetworkEventLifecycle)
(remoteClientFreeFunc)
(remoteDispatchConnectNetworkEventRegisterAny): Track network
callbacks differently.
(remoteDispatchConnectNetworkEventDeregisterAny): Enforce bounds.
Signed-off-by: NEric Blake <eblake@redhat.com>

We point to the manpages where available and redirect to libvirt's
homepage as a last resort.

into remoteDispatchObjectEventSend as it will later be used for both
the domain and network events.

- Use SIGUSR1, not SIGHUP, on reload. At present, virtlockd only
  responds to the former.

- Fix PID file for virtlockd.

- Do not start virtlockd in any runlevels by default. It needs to be
  explicitly selected in libvirt's qemu.conf anyway, so there is no
  need to have it running on all systems regardless.

- Fix chkconfig priorities to ensure virtlockd is started before
  libvirtd is started, and stopped after libvirtd is stopped.

- Add "Should-Start: virtlockd" to the libvirtd initscript's LSB header,
  for the same reason.

- Add "Default-Stop" to both libvirtd and virtlockd initscripts. LSB
  does not guarantee that this defaults to the inverse of
  "Default-Start".
Signed-off-by: NMichael Chapman <mike@very.puzzling.org>
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

https://bugzilla.redhat.com/show_bug.cgi?id=1033061

Currently, initialization of drivers is done in a separate thread. This
is done for several reasons: a driver that is initialized may require
running event loop, it may take ages to initialize driver (e.g. due to
autostarting domains). While the thread is spawn and run, the main()
continues its execution. However, if something goes bad, or the event
loop is just exited (e.g. due to a --timeout or SIGINT) we try to
cleanup all the drivers. So we have two threads running Initialize() and
Cleanup() concurrently. This may result in accessing stale pointers -
e.g. netcf driver will free() itself in stateCleanup callback, while the
init thread may come, open a dummy connection in order to autostart some
domains and voilà: do_open() iterates over interface drivers and
accesses stale netcf driver.

The fix consists in not running stateCleanup if the init thread is still
running.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

Unconditional use of getenv is not secure in setuid env.
While not all libvirt code runs in a setuid env (since
much of it only exists inside libvirtd) this is not always
clear to developers. So make all the code paranoid, even
if it only ever runs inside libvirtd.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The libvirt.so library has far too many library deps to allow
linking against it from setuid programs. Those libraries can
do stuff in __attribute__((constructor) functions which is
not setuid safe.

The virt-login-shell needs to link directly against individual
files that it uses, with all library deps turned off except
for libxml2 and libselinux.

Create a libvirt-setuid-rpc-client.la library which is linked
to by virt-login-shell. A config-post.h file allows this library
to disable all external deps except libselinux and libxml2.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

If we fail to get domain, we had to judge whether
it's null or not when doing 'cleanup'.
Signed-off-by: NChen Hanxiao <chenhanxiao@cn.fujitsu.com>

The libvirtd server pushes data out to clients. It does not
know what protocol version the client might have, so must be
conservative and use the old payload limits. ie send no more
than 256kb of data per packet.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

When a client disconnects from libvirtd, all event callbacks
must be removed. This involves running the public API

  virConnectDomainEventDeregisterAny

This code does not run in normal API dispatch context, so no
identity was set. The result was that the access control drivers
denied the attempt to deregister callbacks. The callbacks thus
continued to trigger after the client was free'd causing fairly
predictable use of free memory & a crash.

This can be triggered by any client with readonly access when
the ACL drivers are active.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

Caused by commit 012c25e8 splitting out a convenience library.

  CC       libvirtd_conf_la-libvirtd-config.lo
In file included from ../src/rpc/virnetmessage.h:24:0,
                 from ../src/rpc/virnetserverprogram.h:27,
                 from ../src/rpc/virnetserver.h:32,
                 from libvirtd-config.c:31:
../src/rpc/virnetprotocol.h:9:21: fatal error: rpc/rpc.h: No such file
or directory

* daemon/Makefile.am (libvirtd_conf_la_CFLAGS): Add XDR_CFLAGS.
Signed-off-by: NEric Blake <eblake@redhat.com>

Signed-off-by: NGiuseppe Scrivano <gscrivan@redhat.com>
Signed-off-by: NEric Blake <eblake@redhat.com>

A previous patch used existing #define for the various files in /etc/pki
instead of hardcoding them in the help output. However I missed that
remote_driver.h contains #define for more paths that are present
in the daemon help output.
This commit uses the existing constants for the path to the
configuration file and to the libvirt sockets.

The 'stats' variable was not initialized to NULL, so if some
early validation of the RPC call fails, it is possible to jump
to the 'cleanup' label and VIR_FREE an uninitialized pointer.
This is a security flaw, since the API can be called from a
readonly connection which can trigger the validation checks.

This was introduced in release v0.9.1 onwards by

  commit 158ba873
  Author: Daniel P. Berrange <berrange@redhat.com>
  Date:   Wed Apr 13 16:21:35 2011 +0100

    Merge all returns paths from dispatcher into single path
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

With the existing pkcheck (pid, start time) tuple for identifying
the process, there is a race condition, where a process can make
a libvirt RPC call and in another thread exec a setuid application,
causing it to change to effective UID 0. This in turn causes polkit
to do its permission check based on the wrong UID.

To address this, libvirt must get the UID the caller had at time
of connect() (from SO_PEERCRED) and pass a (pid, start time, uid)
triple to the pkcheck program.

This fix requires that libvirt is re-built against a version of
polkit that has the fix for its CVE-2013-4288, so that libvirt
can see 'pkg-config --variable pkcheck_supports_uid polkit-gobject-1'
Signed-off-by: NColin Walters <walters@redhat.com>
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

There are constants for these paths in remote_driver.h so we can
use these rather than duplicating them in the help output.
Signed-off-by: NChristophe Fergeau <cfergeau@redhat.com>
Signed-off-by: NEric Blake <eblake@redhat.com>

The help message indicates that the CA certificate is
$sysconfdir/pki/CA/caert.pem while the actual path is
$sysconfdir/pki/CA/cacert.pem
Signed-off-by: NChristophe Fergeau <cfergeau@redhat.com>
Signed-off-by: NEric Blake <eblake@redhat.com>

If we use subdir-objects with automake, any reference to a
cross-directory .c file will result in automake creating
rules that track dependency in the cross directory.  But this
presents a problem during 'make distclean' - if the cross
directory is cleaned up first, then the daemon directory will
be left with dangling references to .Po dependency files that
no longer exist.

Meanwhile, referring to the cross-directory .c file means
that we are compiling the file twice - once in src, and once
in daemon.  Better is to compile just once in src into a
convenience library, and then use that library from daemon.

The tests directory had a similar situation of a cross-directory
.c file; to solve that, we actually need a convenience library.

* daemon/Makefile.am (DAEMON_SOURCES): Drop .c files...
(libvirtd_LDADD): ...and instead use library.
(libvirtd_conf_la_SOURCES): Declare a new convenience library.
(libvirtd_LDFLAGS): Drop duplicate flag.
* tests/Makefile.am (libvirtdconftest_SOURCES): Drop .c file...
(libvirtdconftest_LDADD): ..and instead use library.
Signed-off-by: NEric Blake <eblake@redhat.com>

Trying to enable automake's subdir-objects option resulted in
the creation of literal directories such as src/$(srcdir)/remote/.
I traced this to the fact that we had used a literal $(srcdir)
in a location that later fed an automake *_SOURCES variable.
This has also been reported as an automake bug:
http://debbugs.gnu.org/cgi/bugreport.cgi?bug=13928
but it's better to fix our code than to wait for an automake fix.

Some things to remember that affect VPATH builds, and where an
in-tree build is blissfully unaware of the issues: if a VPATH
build fails to find a file that was used as a prereq of any
other target, then the rule for that file will expand $@ to
prefer the current build dir (bad because a VPATH build on a
fresh checkout will then stick $@ in the current directory
instead of the desired srcdir); conversely, if a VPATH build
finds the file in srcdir but decides it needs to be rebuilt,
then the rule for that file will expand $@ to include the
directory where it was found out-of-date (bad for an explicit
listing of $(srcdir)/$@ because an incremental VPATH build will
then expand srcdir twice).  As we want these files to go into
srcdir unconditionally, we have to massage or avoid $@ for any
recipe that involves one of these files.

Therefore, this patch removes all uses of $(srcdir) from any
generated file name that later feeds a *_SOURCES variable, and
then rewrites all the recipes to generate those files to
hard-code their creation into srcdir without the use of $@.

* src/Makefile.am (REMOTE_DRIVER_GENERATED): Drop $(srcdir); VPATH
builds know how to find the files, and automake subdir-objects
fails with it in place.
(LXC_MONITOR_PROTOCOL_GENERATED, (LXC_MONITOR_GENERATED)
(ACCESS_DRIVER_GENERATED, LOCK_PROTOCOL_GENERATED): Likewise.
(*_client_bodies.h): Hard-code rules to write into srcdir, as
VPATH tries to build $@ locally if missing.
(util/virkeymaps.h): Likewise.
(lxc/lxc_monitor_dispatch.h): Likewise.
(access/viraccessapi*): Likewise.
(locking/lock_daemon_dispatch_stubs.h): Likewise.
* daemon/Makeflie.am (DAEMON_GENERATED, remote_dispatch.h):
Likewise.
Signed-off-by: NEric Blake <eblake@redhat.com>

fixup DAEMON_GENERATED

Automake has builtin support to prevent botched conditional nesting,
but only if you use:
if FOO
else !FOO
endif !FOO

An example error message when using the wrong name:

daemon/Makefile.am:378: error: else reminder (LIBVIRT_INIT_SCRIPT_SYSTEMD_TRUE) incompatible with current conditional: LIBVIRT_INIT_SCRIPT_SYSTEMD_FALSE
daemon/Makefile.am:381: error: endif reminder (LIBVIRT_INIT_SCRIPT_SYSTEMD_TRUE) incompatible with current conditional: LIBVIRT_INIT_SCRIPT_SYSTEMD_FALSE

As our makefiles tend to have quite a bit of nested conditionals,
it's better to take advantage of the benefits of the build system
double-checking that our conditionals are well-nested, but that
requires a syntax check to enforce our usage style.

Alas, unlike C preprocessor and spec files, we can't use indentation
to make it easier to see how deeply nesting goes.

* cfg.mk (sc_makefile_conditionals): New rule.
* daemon/Makefile.am: Enforce the style.
* gnulib/tests/Makefile.am: Likewise.
* python/Makefile.am: Likewise.
* src/Makefile.am: Likewise.
* tests/Makefile.am: Likewise.
* tools/Makefile.am: Likewise.
Signed-off-by: NEric Blake <eblake@redhat.com>

Fixes for argument layouts of various functions in daemon/remote.c

The return values for the virConnectListAllSecrets call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllNWFilters call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllNodeDevices call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllInterfaces call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllNetworks call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virStoragePoolListAllVolumes call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllStoragePools call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virConnectListAllDomains call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virDomain{SnapshotListAllChildren,ListAllSnapshots}
calls were not bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The return values for the virDomainGetJobStats call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

The parameters for the virDomainMigrate*Params RPC calls were
not bounds checks, meaning a malicious client can cause libvirtd
to consume arbitrary memory

This issue was introduced in the 1.1.0 release of libvirt
Signed-off-by: NDaniel P. Berrange <berrange@redhat.com>

and use it when available

by adding it to AM_LDFLAGS instead of every linking rule and
by avoiding a forked grep.