Eric Sandeen wrote:
> Christoph Hellwig wrote:
>
>>> return -ENOMEM;
>>> + /* set to high value to try and avoid collisions with loop below */
>>> + inode->i_ino = 0xffffffff;
>>> + insert_inode_hash(inode);
>> This is odd. Can't we just add some constant base to the loop below?
>>
> I thought the same thing, but Jeff pointed out that
> nfsctl_transaction_write does ops based on inode numbers, and they maybe
> can't move around?
>
> rv = write_op[ino](file, data, size);
>
> However, nfsd's call to simple_fill_super already sends in a set of
> files starting at inode 2:
>
> enum {
> NFSD_Root = 1,
> NFSD_Svc,
> ...
>
> static struct tree_descr nfsd_files[] = {
> [NFSD_Svc] = {".svc", &transaction_ops, S_IWUSR},
> ...
> return simple_fill_super(sb, 0x6e667364, nfsd_files);
>
> which does...
>
> for (i = 0; !files->name || files->name[0]; i++, files++) {
> if (!files->name)
> continue;
> ...
> inode->i_ino = i;
>
> So I think it's already expecting the root inode at one, and the other
> files starting at 2?
>
Very interesting -- I saw that nfsctl depended on certain values for i_ino,
but I didn't notice that it had reserved the first slot for NFSD_Root. The
only other caller of simple_fill_super that seems to depend on the inode
numbers is selinuxfs, and that one has this:
enum sel_inos {
SEL_ROOT_INO = 2,
...so it looks like we can probably use 1 for the root inode (though we might
want to fix selinuxfs to set SEL_ROOT_INO = 1). I'll have a closer look to be
sure that this is all correct and respin this patch since that seems like it
would be less magic-number-y.
Thanks,
Jeff
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
