[Devel] [PATCH VZ10 2/8] ve/mount: thread owning ve through alloc_vfsmnt/clone_mnt/copy_tree

[Pavel Tikhomirov]

On 5/14/26 16:50, Konstantin Khorenko wrote:
> On 4/29/26 15:41, Pavel Tikhomirov wrote:
>> Add owner_ve parameter to alloc_vfsmnt(), clone_mnt() and copy_tree(),
>> to identify correct ownership of mount for accounting. NULL preserves
>> the existing behaviour of taking current ve via get_exec_env().
>>
>> This will be used to derive correct ownership of newly created mounts in
>> case of simultaneous creation of new ve namespace and mount namespace,
>> were we would like new mounts to have the new ve as an owner.
> 
> s/were/where
> 
>>
>> There is no change in behaviour yet: copy_mnt_ns() still resolves the
>> owner from current ve via get_exec_env().
>>
> ...
> 
>> @@ -1325,13 +1325,13 @@ vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
>>   EXPORT_SYMBOL_GPL(vfs_submount);
>>     static struct mount *clone_mnt(struct mount *old, struct dentry *root,
>> -                    int flag)
>> +                int flag, struct ve_struct *owner_ve)
> 
> So here we add an extra argument ve_struct without CONFIG_VE guard.

> 
>>   {
>>       struct super_block *sb = old->mnt.mnt_sb;
>>       struct mount *mnt;
>>       int err;
>>   -    mnt = alloc_vfsmnt(old->mnt_devname);
>> +    mnt = alloc_vfsmnt(old->mnt_devname, owner_ve);
>>       if (!mnt)
>>           return ERR_PTR(-ENOMEM);
>>   
> ...
> 
>>   @@ -4235,7 +4238,11 @@ struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
>>       copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
>>       if (user_ns != ns->user_ns)
>>           copy_flags |= CL_SHARED_TO_SLAVE;
>> -    new = copy_tree(old, old->mnt.mnt_root, copy_flags);
>> +#ifdef CONFIG_VE
>> +    new = copy_tree(old, old->mnt.mnt_root, copy_flags, new_ns->ve_owner);
>> +#else
>> +    new = copy_tree(old, old->mnt.mnt_root, copy_flags, NULL);
>> +#endif
> 
> but here we try to play more fair and use ve_owner only in case CONFIG_VE is used.
> i understand that it's because we have added ve_owner in struct mnt_namespace under CONFIG_VE only.
> 
> But may be to add ve_owner unconditionally and set to non-NULL only in case of CONFIG_VE?
> Then here we'll have just
> +    new = copy_tree(old, old->mnt.mnt_root, copy_flags, new_ns->ve_owner);
> 

Yes, probably since we already have mount->ve_owner not under CONFIG_VE, we can also have mnt_namespace->ve_owner the same way, then this place could be simplified. Done in:

[PATCH VZ10] ve/mntns: make mnt_namespace->ve_owner present in not CONFIG_VE case

> 
>>       if (IS_ERR(new)) {
>>           namespace_unlock();
>>           ns_free_inum(&new_ns->ns);

-- 
Best regards, Pavel Tikhomirov
Senior Software Developer, Virtuozzo.