Dmitry Vyukov
Dmitry Vyukov
> The current implementation seems to be generating segments and writing them into offsets into a file and then mounting that. Correct. > We have different flags, options for each...
> If we want to continue with syzkaller, it is better to enable this env variable for all the targets and extend documentation to make this new flag searchable. +1
These types are effectively unused, we only reference then in the `sendmsg$autorun` syscall only to prevent unused warnings. It references dozens on policies, so there is no chance to generate...
Manager ignores are matched against kernel console output. So I suspect the "KASAN:.*Read" ignore we have is not working, since these things are on separate lines: ``` BUG: KASAN: slab-use-after-free...
I think we can remove it now, it's not doing what it's supposed to be doing, and we don't plan to fix it.
This may be a good use case for bug tags.
Kernel can already dump some things on some panics (locks, ftrace, task list, backtrace for all CPUs). So an alternative may be to extend kernel to dump more info, and...
> Should we replace the socket() call by syz_init_net_socket() or something similar? The socket should be in the test net namespace to reset the right state. > Creating a cgroup...
> You mean calling setup_cgroups() not in the runner executor process before creating the procs Probably not as simple as moving the call, but, yes, something like that. > or...
> If we recreate the proc on the SYZFAIL (which we afaik already do), we are going to hit the same problem as the failing program will just configure it...