On 2021/11/29 7:18, Martin Wilck wrote:
>> /*
>> @@ -329,6 +331,20 @@ void set_path_removed(struct path *pp)
>> condlog(0, "%s: internal error: mpp == NULL", pp-
>>> dev);
>> return;
>> }
>> +
>> + /*
>> + * The path is removed, clear it from mp->paths and mpp->pgs.
>> + */
>> + i = find_slot(mpp->paths, pp);
>> + if (i != -1)
>> + vector_del_slot(mpp->paths, i);
>> +
>> + vector_foreach_slot(mpp->pg, pgp, j) {
>> + i = find_slot(pgp->paths, (void *)pp);
>> + if (i != -1)
>> + vector_del_slot(pgp->paths, i);
>> + }
>> +
>
> No, we can't do this. It would invalidate all the work we did for
> INIT_REMOVED. It's the very idea of INIT_REMOVED that we do NOT
> immediately clear the path from our data structures. The main problem
> is that INIT_REMOVED paths may still be part of the map in the kernel.
> Thus when we re-read the kernel state, the path will be re-added to
> mpp->paths.
>
> We need to rework the free_multipath() logic. It might be sufficient
> to simply clear the references to the devices in mpp before calling
> orphan_paths(). Can you try the attached patch, please?
>
Thanks for replay. The attached patch resolves this error.
I hopes that "Reported-by: Lixiaokeng <lixiaokeng@huawei>" is added to
the patch. Thanks again.
Regards,
Lixiaokeng
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/dm-devel
