Regular VM shutdown triggers the error for existing session of virsh
console and it returns with non-zero exit code:
  error: internal error: console stream EOF

The message and status code are misleading because there's no real
error. virStreamRecv returns 0 correctly when EOF is reached.

Existing implementations of esx, fd, and remote streams behave the same
for virStreamFinish and virStreamAbort: they close the stream. So, we
can continue to use virStreamAbort to handle EOF and errors from
virStreamRecv but additonally we can report error if virStreamAbort
fails.

Fixes: 29f2b524 ("tools: console: pass stream/fd errors to user")
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: NMichal Privoznik <mprivozn@redhat.com>

A library has to be built with -flat_namespace to get all references to
global symbols indirected. That can also be achieved with two-level
namespace interposition but we're not using explicit symbol
interposition since it's more verbose and requires massive changes to
the mocks.

This provides a way to interpose a mock for virQEMUCapsProbeHostCPU from
qemucpumock and fixes domaincapstest on macOS.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

gnulib headers change stat, lstat and open to replacement functions,
even for function definitions. This effectively disables standard
library overrides in virfilewrapper and virmockstathelpers since they
are never reached.

Rename the functions and provide a declartion that uses correct
assembler name for the mocks.

This fixes firmware lookup in domaincapstest on macOS.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

Test executables and mocks have assumption that any symbol can be
replaced with LD_PRELOAD. That's not a case for macOS unless flat
namespace is used, because every external symbol reference records the
library to be looked up. And the symbols cannot be replaced unless dyld
interposing is used.

Setting DYLD_FORCE_FLAT_NAMESPACE changes symbol lookup behaviour to be
similar to Linux dynamic linker. It's more lightweight solution than
explicitly decorating all mock symbols as interpositions and building
libvirt as interposable dynamic library.

This fixes vircryptotest and allows to proceed other tests that rely on
mocks a little bit further.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

macOS syscall interface (/usr/lib/system/libsystem_kernel.dylib) has
three kinds of stat but only one of them can be used to fill
"struct stat": stat$INODE64.

virmockstathelpers looks up regular stat instead of stat$INODE64.  That
causes a failure in qemufirmwaretest because "struct stat" is laid out
differently from the values returned by stat.

Introduce VIR_MOCK_REAL_INIT_ALIASED that can be used to lookup
stat$INODE64 and lstat$INODE64 and use it to setup real functions on
macOS.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

>From ld(1):

  By default all references resolved to a dynamic library record the
  library to which they were resolved. At runtime, dyld uses that
  information to directly resolve symbols. The alternative is to use the
  -flat_namespace option.  With flat namespace, the library is not
  recorded.  At runtime, dyld will search each dynamic library in load
  order when resolving symbols. This is slower, but more like how other
  operating systems resolve symbols.

That fixes the set of tests that preload a mock library to replace
library symbols:
  qemublocktest
  qemumonitorjsontest
  viriscsitest
  virmacmaptest
  virnetserverclienttest
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

/tmp is a symbolic link to /private/tmp on macOS. That causes failures
in commandtest, because getcwd returns /private/tmp and the expected
output doesn't match to "CWD: /tmp".

Rathern than making a copy of commanddata solely for macOS, the /private
prefix is stripped.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

macOS has two kinds of loadable libraries: MH_BUNDLE, and MH_DYLIB.
bundle is used for plugins that are loaded with dlopen/dlsym/dlclose.
And there's no way to preload a bundle into an application. dynamic
linker (dyld) will reject it when finds it in DYLD_INSERT_LIBRARIES.

Unfortunately, a bundle is built if -module flag is provided to libtool.
The flag has been removed to build dylibs with ".dylib" suffix.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

In preparation libtool "-module" flag removal, add lib prefix to all
mock shared objects.

While at it, introduce VIR_TEST_MOCK macros that makes path out of mock
name to be used with VIR_TEST_PRELOAD or VIR_TEST_MAIN_PRELOAD.  That,
hopefully, improves readability, reduces line length and allows to
tailor VIR_TEST_MOCK for specific platform if it has shared library
suffix different from ".so".
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

LD_PRELOAD has no effect on macOS. Instead, dyld(1) provides a way for
symbol hooking via DYLD_INSERT_LIBRARIES. The variable should contain
colon-separated paths to the dylibs to be inserted.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

getnameinfo on macOS formats certain IPv6 addresses as IPv4-translated
addresses. The following pattern has been observed:
  ::ffff is formated as ::0.0.255.255
  ::fffe is formated as ::0.0.255.254
  ::ffff:0 is formated as ::255.255.0.0
  ::fffe:0 is formated as ::255.254.0.0
  ::ffff:0:0 is formated as ::ffff:0.0.0.0
  ::fffe:0:0 is formated as ::fffe:0:0
  ::ffff:0:0:0 is formated as ::ffff:0:0:0

The getnameinfo behavior causes a failure for:
  DO_TEST_PARSE_AND_FORMAT("::ffff", AF_UNSPEC, true);

Use non-ambigious IPv6 for parse/format testing.
Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

getaddrinfo on macOS doesn't interpret octal IPv4 addresses. Only
inet_aton can be used for that. Therefore, from macOS standpoint
"0177.0.0.01" is not the same as "127.0.0.1".

The issue was also discovered by python and dotnet core:
  https://bugs.python.org/issue27612
  https://github.com/dotnet/corefx/issues/8362Signed-off-by: NRoman Bolshakov <r.bolshakov@yadro.com>

After my previous patches we have virPCIDeviceBindToStub() and
virPCIDeviceUnbindFromStub() which really do nothing but call
virPCIDeviceBindToStubWithOverride() and
virPCIDeviceUnbindFromStubWithOverride() respectively.
Drop "WithOverride" from the names and drop the thin wrappers.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>

Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

This enum was introduced to model how RHEL-7 kernel behaves - for
some reason going with the old way (via new_id + bind) fails but
using driver_override succeeds. Well, we don't need to care about
that anymore since we don't create new_id file.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

Now that PCI attach/detach happens solely via driver_override
these two files are no longer needed.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

Now that nothing supports "pci-stub" driver (aka KVM style of PCI
device assignment) there is no need for virpcimock to create it.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

As stated in 84f9358b all kernels that we are interested in
have 'drivers_override'. Drop the other, older style of
overriding PCI device driver - newid.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

This function is no longer used after previous commit.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

Now that no one uses KVM style of PCI assignment we can safely
remove 'pci-stub' backend.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

The KVM assignment is going to be removed shortly. Don't let the
hostdev module configure it.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

After previous commits, the function is not used anymore.
Remove it.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

There are two places where we need to create virPCIDevice from
given virDomainHostdevDef. In both places the code is duplicated.
Move them into a single function and call it from those two
places.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

The KVM assignment was removed in qemu driver in previous commit.
Remove it from domaincapstest too which is hard coding it.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

KVM style of PCI devices assignment was dropped in kernel in
favor of vfio pci (see kernel commit v4.12-rc1~68^2~65). Since
vfio is around for quite some time now and is far superior
discourage people in using KVM style.

Ideally, I'd make QEMU_CAPS_VFIO_PCI implicitly assumed but turns
out qemu-3.0.0 doesn't support vfio-pci device for RISC-V.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>
Tested-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

https://bugzilla.redhat.com/show_bug.cgi?id=1711789

Starting up or building some types of pools may take a very long
time (e.g. a misconfigured NFS). Holding the pool object locked
throughout the whole time hurts concurrency, e.g. if there's
another thread that is listing all the pools.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

In near future the storage pool object lock will be released
during startPool and buildPool callback (in some backends). But
this means that another thread may acquire the pool object lock
and change its definition rendering the former thread access not
only stale definition but also access freed memory
(virStoragePoolObjAssignDef() will free old def when setting a
new one).

One way out of this would be to have the pool appear as active
because our code deals with obj->def and obj->newdef just fine.
But we can't declare a pool as active if it's not started or
still building up. Therefore, have a boolean flag that is very
similar and forces virStoragePoolObjAssignDef() to store new
definition in obj->newdef even for an inactive pool. In turn, we
have to move the definition to correct place when unsetting the
flag. But that's as easy as calling
virStoragePoolUpdateInactive().

Technically speaking, change made to
storageDriverAutostartCallback() is not needed because until
storage driver is initialized no storage API can run therefore
there can't be anyone wanting to change the pool's definition.
But I'm doing the change there for consistency anyways.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

If there's a persistent storage and user tries to start a new one
with the same name and UUID (e.g. to test new configuration) it
may happen that upon failure we lose the persistent defintion.
Fortunately, we don't remove it from the disk only from the
internal list of the pools.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

This flag can be used to denote that the definition we're trying
to assign to a pool object is live definition and thus the
inactive definition should be saved into ->newDef.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

Separate storage pool definition assignment into a function.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

There will be more boolean information that we want to pass to
this function. Instead of having them in separate arguments per
each one, use @flags.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

This function is doing much more than plain assigning pool
definition to a pool object. Rename it.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

There is no need for this function to call
virStoragePoolObjEndAPI(). The object is perfectly usable after
return from this function. In fact, all callers will call
virStoragePoolObjEndAPI() eventually.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

The function comment mistakenly refers to 'poolptr' when in fact
the variable is named 'objptr'.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

Turns out there's one callback that might remove a storage pool
during its run: storagePoolUpdateAllState() call
storagePoolUpdateStateCallback() which may call
virStoragePoolUpdateInactive() which in turn may call
virStoragePoolObjRemove(). Problem is that the
UpdateStateCallback() sees a storage pool object with just two
references: one for each hash table holding the object. If the
function ends up calling ObjRemove() then upon removing the
object from hash tables those references are gone and thus any
subsequent call touching the object is invalid.

The solution to this problem is to grab reference for the object
we are running iterator with.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

The fact that we're removing a pool object from the list of pools
doesn't mean we want to unlock it. It violates locking policy
too as object locking and unlocking is not done on the same
level.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJán Tomko <jtomko@redhat.com>

It may happen that we leave some XATTRs behind. For instance, on
a sudden power loss, the host just shuts down without calling
restore on domain paths. This creates a problem, because when the
host starts up again, the XATTRs are there but they don't reflect
the true state and this may result in libvirt denying start of a
domain.

To solve this, save a unique timestamp (host boot time) among
with our XATTRs.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1741140Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJiri Denemark <jdenemar@redhat.com>

This module contains function to get host boot time.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NJiri Denemark <jdenemar@redhat.com>

If user has two domains, each have the same disk (configured for
RW) but each runs with different seclabel then we deny start of
the second domain because in order to do that we would need to
relabel the disk but that would cut the first domain off. Even if
we did not do that, qemu would fail to start because it would be
unable to lock the disk image for the second time. So far, this
behaviour is expected. But what is not expected is that we
increase the refcounter in XATTRs and leave it like that.

What happens is that when the second domain starts,
virSecuritySetRememberedLabel() is called, and since there are
XATTRs from the first domain it increments the refcounter and
returns it (refcounter == 2 at this point). Then callers
(virSecurityDACSetOwnership() and
virSecuritySELinuxSetFileconHelper()) realize that refcounter is
greater than 1 and desired seclabel doesn't match the one the
disk image already has and an error is produced. But the
refcounter is never decremented.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1740024Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NMartin Kletzander <mkletzan@redhat.com>

Also get rid of the temporary 'type' variable.
Signed-off-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NMichal Privoznik <mprivozn@redhat.com>