From: oleg at redhat.com (Oleg Nesterov) Date: Sat, 20 Apr 2019 13:51:26 +0200 Subject: [PATCH RFC 1/2] Add polling support to pidfd In-Reply-To: References: <20190417130940.GC32622@redhat.com> <20190419190247.GB251571@google.com> <20190419191858.iwcvqm6fihbkaata@brauner.io> <20190419194902.GE251571@google.com> <20190419212002.GB44851@google.com> Message-ID: <20190420115125.GC17468@redhat.com> On 04/19, Christian Brauner wrote: > > > Just for the exit notification purposes, the states are: > > if process has exit_state == 0, block. > > if process is zombie/dead but not reaped, then return POLLIN > > if process is reaped, then return POLLIN | POLLHUP > > Oleg was explicitly against EXIT_ZOMBIE/DEAD thing, no? He said so in a > prior mail. Has this been addressed? Yes, please don't use EXIT_XXX codes, nobody should ever use them except the core kernel exit/wait paths. For example, EXIT_DEAD means that the task auto- reaps or its parent had to temporary drop tasklist. Just check ->exit_state != 0 && thread_group_empty(). Note that we need thread_group_empty() only for the case when the task is traced, in this case we have an extra notification for debugger which can confuse pidfd_poll(). And of course, everything will differ if/when we will need to monitor the sub-threads. And btw I don't think it needs tasklist_lock, but lets discuss this when we have a new version based on pidfd. Oleg.