On Thu, Mar 27, 2008 at 02:04:10PM -0400, Michael Stone wrote: > If someone can show that it is a coreutils problem, and have > some kind of reproducible case I'd be happy to look at it more. Ok, but I'd still like to keep the bug open in coreutils, so people can find it. > point I don't see what I can do. (I will point out that this problem did > not exist when the package was built, because it would have failed the > build tests--so what could have changed apart from the rm binary...) The difference is that the buildd's run a oldabi kernel and oldabi userland. This is the "arm" debian port. Now we also have new abi, EABI, which is the debian port "armel". Both are little endian. Matthias and some others run a EABI kernel with a oldabi emulation, and use the emulation layer to run oldabi arm port in a chroot. In this chroot new versions of coreutils/findutils which use *at syscalls are failing. Now after a quick relook, I think the issue is fstatat64. both find and rm call it, but arch/arm/kernel/sys_oabi-compat.c doesn't have a shim to convert the stat64 struct oldabi <-> eabi for this function. Fix is probably trivial, if we want to continue to support the oldabi emulation layer. -- "rm -rf" only sounds scary if you don't have backups
Attachment:
signature.asc
Description: Digital signature