On 1/29/26 3:24 AM, Jan Beulich wrote:
On 29.01.2026 03:42, Stefano Stabellini wrote:
On Wed, 28 Jan 2026, Jan Beulich wrote:
On 23.01.2026 02:06, Stefano Stabellini wrote:
@@ -742,17 +758,36 @@ static long 
guest_console_write(XEN_GUEST_HANDLE_PARAM(char) buffer,
          if ( copy_from_guest(kbuf, buffer, kcount) )
              return -EFAULT;
- if ( is_hardware_domain(cd) )
+        /*
+         * Take both cons->lock and console_lock:
+         * - cons->lock protects cons->buf and cons->idx
+         * - console_lock protects console_send and is_focus_domain
+         *   checks
+         *
+         * The order must be respected. guest_printk takes the
+         * console_lock so it is important that cons->lock is taken
+         * first.
+         */
+        spin_lock(&cons->lock);
+        nrspin_lock_irq(&console_lock);
+        if ( is_focus_domain(cd) )

Why would any of the domains possibly being permitted to be "focus" suddenly
gain direct access here? Full access in do_console_io() is still prevented by
the XSM check there, afaict. Cc-ing Daniel, as it's not quite clear to me
whether to introduce another XSM check here, whether to check ->is_console
directly, or yet something else.

The XSM check still happens first in do_console_io() via
xsm_console_io(XSM_OTHER, current->domain, cmd), which validates that
the domain has permission to use console_io hypercalls. The focus check
is an additional restriction that only allows reading when the domain
has focus: it doesn't grant new permissions. Dom0less domains with
input_allowed = true are already permitted by XSM policy to use
console_io;

Are they? I don't see any XSM or Flask code checking that flag. What the
dummy xsm_console_io() checks is ->is_console.

Unless I am misunderstanding what you are asking here, I don't see why XSM would be concerned with this check. The `is_focus_domain()` conditional is not an access decision but a decision whether write to the console or buffer the write.


However, what indeed I didn't pay attention to when writing the original
comment is that guest_console_write() has only a single caller,
do_console_io(). So there's no concern in this regard here as long as no
new caller appears.


Correct, the `xsm_console_io()` hook is the access check if the guest is allowed to read/write to the console. Any paths to this function should be guarded by a call to this hook.

v/r,
dps

Reply via email to