Nathan Lynch <ntl@xxxxxxxxx> writes:
> Last commit is ed3b275 "allow error string during checkpoint while
> holding a spinlock".
>
> The failure seems to involve a vma corresponding to
> /usr/lib/gconv/gconv-modules.cache which is mapped read-only and shared,
> e.g.
>
> open("/usr/lib/gconv/gconv-modules.cache", O_RDONLY) = 3
> mmap2(NULL, 26048, PROT_READ, MAP_SHARED, 3, 0) = 0xb7f52000
>
> I believe the second check in filemap_restore() is where the restart
> goes awry; the object is of type CKPT_VMA_SHM_FILE but doesn't have
> VM_SHARED set in the flags (it does have VM_MAY_SHARE set, however).
As Serge suggested in other bug report, updated to latest user-cr and
got same result.
Also, I forgot to mention that if I modify the check in filemap_restore
to allow VM_MAY_SHARE, the restart operation appears to complete
successfully but the kernel gets an oops on an address in that vma when
the task runs.
_______________________________________________
Containers mailing list
Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/containers
