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 18:26:52 +0100, Vladimir Riabchun 
<[email protected]> wrote:
Modify ovl_override_creds so that it returns both old cred and old VE.
This will also protect us from future rebase issues - all new users of
old_cred and ovl_override_creds will fail with compilation error.

- Item 8 of the cover letter reply: add a Fixes tag for ec01a90d0ef9
   ("ovl: replace capable by ve_capable for indexing feature"), limit
   this claim to ovl_override_creds() callers, say why cross-CT use
   must fail, and fix "swich" and "Stroing".

ec01a90d0ef9 turned capable() into ve_capable() in
ovl_can_decode_fh() and ovl_decode_real_fh(). That made the result
depend on the VE, which is the bug.

The compile error catches new ovl_override_creds() callers only.
ovl_creds() users, such as backing_file_ctx.cred in read/write/splice/
mmap, still switch creds without the VE.

Yes, but it is a separate issue with other ve_capable users.
Created a task for it: https://virtuozzo.atlassian.net/browse/VSTOR-155347


We don't allow cross-CT overlayfs usages, it is broken and reported
as -EINVAL. But we want to allow usage of host-owned overlayfs by CT.

"It is broken" does not say why the error is needed. If a CT-B task
kept its own VE on a CT-A overlay, CT-A creds would be checked
against CT-B's user namespace. That fails at run time, while the
mount-time check passed, and it is the same oops again. Please say
that in the message, since Pavel's "just skip the override" looks
like the simpler choice otherwise.

+       /*
+        * We want to allow passing host overlayfs to CT.
+        * It is fine to keep CT ve - host creds should work.
+        * Cross-CT mounts are forbidden.
+        */
+       if (IS_ERR(desc->ve) && PTR_ERR(desc->ve) != -EPERM)
+               return PTR_ERR(desc->ve);

- Item 1 of the cover letter reply: test only IS_ERR() here and in
   ovl_revert_creds(), once override_ve() returns old_ve for the VE0
   case. Fix the comment: mounting works, using the overlay from
   another CT is refused.

The user-visible errno also changes then: stat, open and lookup on a
foreign CT's overlay return EPERM rather than EINVAL. EINVAL there
reads as a bad argument from the caller.

Two other comments will fix.



--
Best regards, Riabchun Vladimir
Linux Kernel Developer, Virtuozzo

_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to