On May 17, 2023 1:25:08 AM PDT, Christian Brauner <brauner@xxxxxxxxxx> wrote: >On Wed, May 17, 2023 at 12:40:03AM -0700, syzbot wrote: >> Hello, >> >> syzbot found the following issue on: >> >> HEAD commit: 065efa589871 Add linux-next specific files for 20230517 >> git tree: linux-next >> console output: https://syzkaller.appspot.com/x/log.txt?x=17f27bb2280000 >> kernel config: https://syzkaller.appspot.com/x/.config?x=821eeb02ef201bcc >> dashboard link: https://syzkaller.appspot.com/bug?extid=ac3b41786a2d0565b6d5 >> compiler: gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.2 >> >> Downloadable assets: >> disk image: https://storage.googleapis.com/syzbot-assets/dbbd691e9e5a/disk-065efa58.raw.xz >> vmlinux: https://storage.googleapis.com/syzbot-assets/e5b9541c3979/vmlinux-065efa58.xz >> kernel image: https://storage.googleapis.com/syzbot-assets/44cf3f3aaabb/bzImage-065efa58.xz >> >> IMPORTANT: if you fix the issue, please add the following tag to the commit: >> Reported-by: syzbot+ac3b41786a2d0565b6d5@xxxxxxxxxxxxxxxxxxxxxxxxx >> >> ================================================================================ >> UBSAN: array-index-out-of-bounds in kernel/pid.c:244:15 > >Only way I see this happening is if the logic in >kernel/pid_namespace.c:create_pid_cachep() which sets the object size >for the struct pid allocation of this pid namespace based on >parent_pid_namespace->level + 1 is broken. The way this works is: > > struct pid > { > [snip] > struct upid numbers[1]; I was *just* looking at this fake flex array during LSS last week. It was one of two core structs still using the ancient 1-element style. > }; > > create_pid_namespace() > { > unsigned int level = parent_pid_ns->level + 1; > ns->pid_cachep = create_pid_cachep(level); > } > >and then during fork: > > alloc_pid() > { > pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); > } > >So effectively, the wrong level must've been set in >create_pid_namespace() so that the flexible array allocation is too >small. > >I don't have time to debug this tbh. Ccing Kees maybe there's some >flexible array stuff going on I'm unaware of. Yes, I think it's due to: https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=for-next/hardening&id=2d47c6956ab3c8b580a59d7704aab3e2a4882b6c This makes the sanitizer treat only [0]-arrays as flex arrays. Though I wonder why Clang hasn't warned about this yet. Regardless, we'll need to fix struct pid. Since it uses a static initializer for "numbers[0]", this will need a bit of a tweak, but I've got patches for this. I hadn't sent them yet because I was still studying the use of the "levels" member which is off by one for the count of "numbers" elements, which some code already has to work around (using "<=" when iterating and "+ 1" for some outputs)... -- Kees Cook