Re: mountd: Possible bug in next_mnt()

[Richard Weinberger <richard@nod.at>]

----- Ursprüngliche Mail -----
>> Well, the goal of my mail was not sending a ready-to-apply patch.
>> It was a question. To me next_mnt() looks wrong but I'm not sure whether
>> the current handling of "/" is desired for some special case I'm not aware of.
>> 
>> I'll happily send a patch after we agree that next_mnt() is wrong.
> I'm still trying to reproduce problem... I have
> 
> /etc/nfs.conf: rootdir=/export
> 
> /etc/exports:
> /home *(rw,sec=sys:krb5:krb5i:krb5p)
> /tmp *(rw,fsid=666,all_squash)
> / *(rw,fsid=root,all_squash)
> 
> I'm not seeing the problem... Where does the crossmount come in?

Chris reported the problem to me while he was testing my re-export/crossmount patches.
I can try reproducing without re-exporting later.

In theory you should see the problem as follows:

1. Have rootdir=/export in your nfs.conf
2. /export is some filesystem that contains more mounts
3. /export/fs1 is a different filesytem
4. /export/fs2 is a different filesytem too
5. /etc/exports contains: / *(rw,fsid=root,all_squash,crossmount)

Client mounts / to /nfs and then tries to access /nfs/fs1.
Then nfsd_fh() iterates over all exports, finds one with NFSEXP_CROSSMOUNT set.
Using next_mnt() it finds possible sub-mounts. But for the "/" case next_mnt()
returns none -> nfsd_fh() fails -> client cannot enter /nfs/fs1.

Thanks,
//richard