On Thu, Jan 26, 2023 at 07:30:19PM +0100, Oleg Nesterov wrote: > On 01/26, Andrei Vagin wrote: > > > > On Thu, Jan 26, 2023 at 7:07 AM Oleg Nesterov <oleg@xxxxxxxxxx> wrote: > > > > > > IIUC, PTRACE_O_SUSPEND_SYSCALL_USER_DISPATCH is needed to run the injected > > > code, and this also needs to change the state of the traced process. If > > > the tracer (CRIU) dies while the tracee runs this code, I guess the tracee > > > will have other problems? > > > > Our injected code can reheal itself if something goes wrong. The hack > > here is that we inject > > the code with a signal frame and it calls rt_segreturn to resume the process. > > What will happen if CRIU dies and clears ->ptrace right before > syscall_user_dispatch() checks PT_SUSPEND_SYSCALL_USER_DISPATCH ? > > How the tracee will react to SIGSYS with unexpected .si_syscall ? > > > I don't expect that > > the syscall user dispatch > > is used by many applications, > > Agreed, so the case when CRIU will need to do the additional > PTRACE_SET_SYSCALL_USER_DISPATCH_CONFIG twice to disable and then re-enable > syscall_user_dispatch is unlikely. > > > so I don't strongly insist on > > PTRACE_O_SUSPEND_SYSCALL_USER_DISPATCH. > > I too won't argue too much. but so far I do not feel there is enough > justification for this feature ... > > Oleg. > I'm not married to the idea, just want to make sure I have the tools needed to make checkpoint work. The option seems like the easiest way given the exclusion area issue. One idea is to overwrite a portion of the exclusion area, but this obviously can increase the complexity of the checkpoint process. Another idea is to disable/enable SUD via get/set, but this produces potential detach issues. @Andrei i'm happy to take this to IRC or somewhere else out of band to discuss the checkpoint specifics, if that would be preferable and you are interested.
