On Sat, Aug 30, 2025 at 03:30:11PM +0200, "Gustavo A. R. Silva" 
<[email protected]> wrote:
> Based on the comments above, it seems that the original code was expecting
> cgrp->ancestors[0] and cgrp_ancestor_storage to share the same addres in
> memory.

Fortunately, it doesn't matter what the address of cgrp_ancestor_storage
is. The important effect is that cgroup_root::cgrp is followed by
sufficient space to store a pointer (accessed via cgroup::ancestors[0]).

> However when I take a look at the pahole output, I see that these two members
> are actually misaligned by 56 bytes. See below:

So the root cgroup's ancestry may be saved inside the padding instead of
the dedicated storage. I don't think it causes immediate issues but it'd
be better not to write to these bytes. (Note that the layout depends on
kernel config.) Thanks for the report Gustavo!


> So, one solution for this is to use the TRAILING_OVERLAP() helper and
> move these members at the end of `struct cgroup_root`. With this the
> misalignment disappears (together with the 14722 warnings :) ), and now
> both cgrp->ancestors[0] and cgrp_ancestor_storage share the same address
> in memory. See below:

I didn't know TRAILING_OVERLAP() but it sounds like the tool for such
situations.
Why do you move struct cgroup at the end of struct cgroup_root?

(Actually, as I look at the macro's implementation, it should be
--- a/include/linux/stddef.h
+++ b/include/linux/stddef.h
@@ -110,7 +110,7 @@ enum {
                struct {                                                        
\
                        unsigned char __offset_to_##FAM[offsetof(TYPE, FAM)];   
\
                        MEMBERS                                                 
\
-               };                                                              
\
+               } __packed;                                                     
\
        }

 #endif
in order to avoid similar issues, no?)

Thanks,
Michal

Attachment: signature.asc
Description: PGP signature

Reply via email to