It doesn't set the right at all, that is the problem - which worked here iirc.

Do you have a custom umask set at the point where you launch pppd? That would cause the problem ... introducing a chmod 644 into the dns script should help anyway.

Yes I have custom umask (umask 066). But I think that it should change the rights since I run pon script in a shell so should I change my umask every time that I run the script? I think not.

I have the same issue since the update of ppp 2.4.4-8 in conjunction with rp-pppoe 3.10-1 and filesystem 2009.01-2. Downgrading to ppp 2.4.4-7 helped before, but now it doesn't help anymore.

In the past resolv.conf was saved in /etc/ppp with the permissions root:root 644 as needed and a symlink to this file was set in /etc.

Now resolv.conf is saved directly in /etc, but with the permissions root:root 600. If I chmod this file to 644 manually, the permissions are always reset to 600, if I disconnect with pppoe-stop and reconnect with pppoe-start.

As a workaround I had to modify the script /usr/sbin/pppoe-start and add a `chmod 644 /etc/resolv.conf` after the "Connected!" message.

Btw., umask for my user root is set to 022. So this is not a umask issue.

This is bad anyway, as rp-pppoe shouldn't even touch that, because initializing the ppp connection (including ip-up handling and settings) is a problem of pppd, not of rp-pppoe. The fact that rp-pppoe does DNS handling at all (apparently by setting that symlink) is a bug on its own.

I won't deal with rp-pppoe's stupidness at all (the package is superfluous anyway, so I'll completely ignore it) and only fix this in ppp by adding a proper chmod call to the 00-dns.sh script. I don't know when I'll get to that, in the meantime add a chmod 644 /etc/resolv.conf at the end of /etc/ppp/ip-up.d/00-dns.sh.

I, too, don't think, that this is rp-pppoe related. But if you want to add a chmod to the 00-dns.sh script, I'd suggest to remove this file completely, because it's generally superfluous.

I don't think, that creating a backup of resolv.conf at every ppp(oe) connection is necessary, especially because this backup is still saved in /etc/ppp. Though /etc/ppp/resolv.conf is always copied to /etc. Why not keep /etc/ppp/resolv.conf as a backup (also not really necessary) and keep /etc/resolv.conf untouched. I'd say this would be more KISS like and save disk space. If someone needs more copies of resolv.conf for different DNS servers, these copies can be made manually.

I just see, that the backup of resolv.conf is written back to resolv.conf at disconnection by /etc/ppp/ip-down.d/00-dns.sh. This file could or should be removed, too, if /etc/ppp/ip-up.d/00-dns.sh will be removed.

No, this is necessary. If you establish a ppp connection with USEPEERDNS, the upstream DNS server will not be set unless you move /etc/ppp/resolv.conf to /etc/. Apps like wvdial, kppp or plain ppp with a config file rely on that. The fact that rp-pppoe does a job that ppp should do is simply wrong.

I uninstalled rp-pppoe and configured ppp with rp-pppoe.so. If I now connect with pon and disconnect with poff, the permissions of /etc/resolv.conf aren't changed anymore. Maybe the first time they were set to 600 - I didn't pay attention - but the permissions aren't changed anymore at dial-up. I also tried to connect with networkmanager. The same here. The script /etc/ppp/ip-up.d/00-dns.sh seems to be executed, but the permissions of /etc/resolv.conf haven't been changed. So I guess my problem was related to rp-pppoe maybe in conjunction with some possible changes in ppp. So I'd suggest not to add the chmod to the script /etc/ppp/ip-up.d/00-dns.sh but to /usr/sbin/pppoe-start.

The chmod is necessary anyway in case of non-standard umasks (betek's original problem), adding it won't hurt then.

Should be fixed in ppp 2.4.4-9 in testing.