In selinux driver there's virSecuritySELinuxSetFileconImpl()
which is responsible for actual setting of SELinux label on given
file and handling possible failures. In fhe failure handling code
we decide whether failure is fatal or not. But there is a bug:
depending on SELinux mode (Permissive vs. Enforcing) the ENOENT
is either ignored or considered fatal. This not correct - ENOENT
must always be fatal - QEMU will fail opening it anyways.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2004850
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
---
 src/security/security_selinux.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/src/security/security_selinux.c b/src/security/security_selinux.c
index 050acee2b0..7e8c4fb4f2 100644
--- a/src/security/security_selinux.c
+++ b/src/security/security_selinux.c
@@ -1283,9 +1283,11 @@ virSecuritySELinuxSetFileconImpl(const char *path,
         } else {
             /* However, don't claim error if SELinux is in Enforcing mode and
              * we are running as unprivileged user and we really did see EPERM.
-             * Otherwise we want to return error if SELinux is Enforcing. */
-            if (security_getenforce() == 1 &&
-                (setfilecon_errno != EPERM || privileged)) {
+             * Otherwise we want to return error if SELinux is Enforcing, or we
+             * saw EPERM regardless of SELinux mode. */
+            if (setfilecon_errno == ENOENT ||
+                (security_getenforce() == 1 &&
+                 (setfilecon_errno != EPERM || privileged))) {
                 virReportSystemError(setfilecon_errno,
                                      _("unable to set security context '%s' on '%s'"),
                                      tcon, path);
-- 
2.32.0

On a Monday in 2021, Michal Privoznik wrote:
>In selinux driver there's virSecuritySELinuxSetFileconImpl()
>which is responsible for actual setting of SELinux label on given
>file and handling possible failures. In fhe failure handling code
>we decide whether failure is fatal or not. But there is a bug:
>depending on SELinux mode (Permissive vs. Enforcing) the ENOENT
>is either ignored or considered fatal.

>This not correct - ENOENT
>must always be fatal - QEMU will fail opening it anyways.
>
>Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2004850

It won't get as far as trying to start QEMU. The error message in the linked bug:
   error: unable to stat: /var/lib/libvirt/images/slic.dat: No such file or directory
comes from the DAC driver.

IIUC in virSecurityStackTransactionCommit we happily commit the SELinux
changes, fail to commit the DAC changes, but the rollback calling
virSecurityManagerTransactionAbort does nothing.

And since qemuSecuritySetAllLabel does not complete successfully, qemuProcessLaunch
does not ask its callers to restore the labels.

Jano

>Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
>---
> src/security/security_selinux.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>

On 9/20/21 5:57 PM, Ján Tomko wrote:
> On a Monday in 2021, Michal Privoznik wrote:
>> In selinux driver there's virSecuritySELinuxSetFileconImpl()
>> which is responsible for actual setting of SELinux label on given
>> file and handling possible failures. In fhe failure handling code
>> we decide whether failure is fatal or not. But there is a bug:
>> depending on SELinux mode (Permissive vs. Enforcing) the ENOENT
>> is either ignored or considered fatal.
> 
>> This not correct - ENOENT
>> must always be fatal - QEMU will fail opening it anyways.
>>
>> Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2004850
> 
> It won't get as far as trying to start QEMU. The error message in the
> linked bug:
>   error: unable to stat: /var/lib/libvirt/images/slic.dat: No such file
> or directory
> comes from the DAC driver.

Correct. I should have rephrased that.

> 
> IIUC in virSecurityStackTransactionCommit we happily commit the SELinux
> changes, fail to commit the DAC changes, but the rollback calling
> virSecurityManagerTransactionAbort does nothing.

Indeed.

> 
> And since qemuSecuritySetAllLabel does not complete successfully,
> qemuProcessLaunch
> does not ask its callers to restore the labels.

Yes.

Michal