David Howells <dhowells@xxxxxxxxxx> writes:
> Paulo Alcantara <pc@xxxxxxxxxxxxx> wrote:
>
>> I don't know why it was designed that way, but the reason we have two
>> different superblocks with ${opts} being the same is because cifs.ko
>> relies on the value of cifs_sb_info::prepath to build paths out of
>> dentries. See build_path_from_dentry(). So, when you access
>> /mnt/2/foo, cifs.ko will build a path like '[optional tree name prefix]
>> + cifs_sb_info::prepath + \foo' and then reuse connections
>> (server+session+tcon) from first superblock to perform I/O on that file.
>
> Yep. You don't *need* prepath. You could always build from the sb->s_root
> without a prepath and have mnt->mnt_root offset the place the VFS thinks you
> are:
>
> [rootdir]/ <--- s_root points here
> |
> v
> foo/
> |
> v
> bar/ <--- mnt_root points here
> |
> v
> a
>
> Without prepath, you build back up the tree { a, bar/, foo/, [rootdir] } with
> prepath you insert the prepath at the end.
>
> Bind mounts just make the VFS think it's starting midway down, but you build
> up back to s_root.
>
> Think of a mount as just referring to a subtree of the tree inside the
> superblock. The prepath is just an optimisation - but possibly one that makes
> sense for cifs if you're having to do pathname fabrication a lot.
Thanks alot Dave! Great explanation. We also need to remember that
those prefix paths can also be changed over reconnect. That is, if
you're currently mounted to a DFS link target '\srv1\share' and client
failovers to next target '\srv2\share\foo\bar', cifs_sb_info::prepath
will be set to '\foo\bar'. And if you mounted the DFS link as
`mount.cifs //dfs/link/some/dir`, cifs_sb_info::prepath would be set to
'\some\dir\foo\bar'. Yeah, a lot of corner cases to handle...
Anyways, don't worry much about all of this as we can handle in
follow-up patches.
FWIW, patch looks good:
Acked-by: Paulo Alcantara (Red Hat) <pc@xxxxxxxxxxxxx>
