On Monday 17 September 2007, Bob Proulx wrote: > Markus Schoder wrote: > > Package: coreutils > > Version: 5.97-5.4 > > Severity: normal > > I have recreated the problem and have reduced it to something related > to the use of openat() and fallback code in the 5.97-5.4 NMU whereas > the previous 5.97-5.3 used getcwd(). Methinks the NMU changed more > than intended. Certainly more than was documented in the changelog. > > > First some pre-conditions: > > 1. Current directory is /tmp > > 2. /tmp is bind mounted to /var/chroot/sid-ia32/tmp > > The key info here is that openat() is now attempted and fails (for me > on my older amd64 kernel) and the fallback code requires /proc to be > available. This was not required in the getcwd() version. > > > 3. ia32 is a program that does a chroot to /var/chroot/sid-ia32 and > > then execs a shell. > > Using 'sudo chroot /path/chroot /bin/bash' works fine too.
For me this causes the directory to change to / where the problem does not happen. > > Given those here is a command sequence that shows the problem: > > > > /tmp$ ia32 > > /tmp$ /bin/pwd > > /bin/pwd: couldn't find directory entry in `../..' with matching > > i-node > > Confirmed. pwd tries to open /proc/self/fd and various and fails > because it is not there. > > Workaround: Also bind mount /proc into the chroot. I always had /proc bind mounted. That cannot be the whole story but thanks for looking into this. -- Markus -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
