On 11/12/26 14:17, Konstantin Khorenko wrote:
I won't apply this version, please send v6 with the changes below.
On Tue, 10 Nov 2026 18:26:49 +0100, Vladimir Riabchun
<[email protected]> wrote:
This field is not supposed to be inherited via fork, so report
a check for this was added as well.
- Item 6 of the cover letter reply: rewrite this sentence, and say in
the message that ve_capable() now uses overrided_ve and which VE
switches are allowed, refused, or skipped.
The sentence does not parse. The message also never says that
ve_capable() and ve_capable_noaudit() change, or what override_ve()
does for a CT caller. That is the part a reader of this patch needs.
+ /* CT tasks are not supposed to switch VE context, this is a breach */
+ if (new_ve != get_exec_env() && !ve_is_super(get_exec_env())) {
+ if (ve_is_super(new_ve)) {
+ /* Don't cry about it, this event we may allow. */
+ return ERR_PTR(-EPERM);
+ }
- Item 1 of the cover letter reply: return the current override
(old_ve) here, and ERR_PTR(-EPERM) for the cross-CT case below.
ERR_PTR(-EPERM) here means "go on, just do not switch the VE", so the
caller has to tell this error from a real one, and patch 5 does that
twice with PTR_ERR() != -EPERM. Returning old_ve gives the same
ve_capable() result: overrided_ve stays as it was, and revert_ve()
writes back the same value. Then any error pointer is a real refusal,
and the "this is a breach" comment no longer covers an allowed case.
I disagree. CT -> Host VE switch is not a normal case. We know it is
fine for overlayfs, but for other cases we should review it separately,
if this is a correct usage or CT is doing something nasty.
+ pr_err_ratelimited("Invalid VE override: %s -> %s\n",
+ ve_name(get_exec_env()),
+ ve_name(new_ve));
- Item 2 of the cover letter reply: do not print ve_name(new_ve).
new_ve is the creator VE of the overlay. It is pinned only by
get_ve(), and the CT may already be stopped and its cgroup removed.
ve_offline() frees ve->ve_name with plain kfree() and no RCU, so this
can read freed memory, or print "(null)".
+ WARN_ON(orig->overrided_ve);
- Item 3 of the cover letter reply: WARN_ON_ONCE.
If this ever fires, it fires on every fork in the window. One trace
is enough.
--
Best regards, Riabchun Vladimir
Linux Kernel Developer, Virtuozzo
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel
