- 30 Apr, 2008 40 commits
-
-
Oleg Nesterov authored
Currently the buggy /sbin/init hangs if SIGSEGV/etc happens. The kernel sends the signal, init dequeues it and ignores, returns from the exception, repeats the faulting instruction, and so on forever. Imho, such a behaviour is not good. I think that the explicit loud death of the buggy /sbin/init is better than the silent hang. Change force_sig_info() to clear SIGNAL_UNKILLABLE when the task should be really killed. Signed-off-by:
Oleg Nesterov <oleg@tv-sign.ru> Cc: Roland McGrath <roland@redhat.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Oleg Nesterov authored
Now that we rely on SIGNAL_UNKILLABLE flag, de_thread() doesn't need the nasty hack to kill the old ->child_reaper during the mt-exec. This also means we can avoid taking tasklist_lock around zap_other_threads(). Signed-off-by:
-
Oleg Nesterov authored
The global init has a lot of long standing problems with the unhandled fatal signals. - The "is_global_init(current)" check in get_signal_to_deliver() protects only the main thread. Sub-thread can dequee the fatal signal and shutdown the whole thread group except the main thread. If it dequeues SIGSTOP /sbin/init will be stopped, this is not right too. Note that we can't use is_global_init(->group_leader), this breaks exec and this can't solve other problems we have. - Even if afterwards ignored, the fatal signals sets SIGNAL_GROUP_EXIT on delivery. This breaks exec, has other bad implications, and this is just wrong. Introduce the new SIGNAL_UNKILLABLE flag to fix these problems. It also helps to solve some other problems addressed by the subsequent patches. Currently we use this flag for the global init only, but it could also be used by kthreads and (perhaps) by the sub-namespace inits. Signed-off-by:
"Eric W. Biederman" <ebiederm@xmission.com> Cc: Roland McGrath <roland@redhat.com> Signed-off-by:
-
Oleg Nesterov authored
Now that task_session() can't return a false NULL, check_kill_permission() doesn't need tasklist_lock. Signed-off-by:
-
Oleg Nesterov authored
This wasn't documented, but as Atsushi Tsuji pointed out check_kill_permission() needs tasklist_lock for task_session_nr(). I missed this fact when removed tasklist from the callers. Change check_kill_permission() to take tasklist_lock for the SIGCONT case. Re-order security checks so that we take tasklist_lock only if/when it is actually needed. This is a minimal fix for now, tasklist will be removed later. Also change the code to use task_session() instead of task_session_nr(). Also, remove the SIGCONT check from cap_task_kill(), it is bogus (and the whole function is bogus. Serge, Eric, why it is still alive?). Signed-off-by:
Atsushi Tsuji <a-tsuji@bk.jp.nec.com> Cc: Roland McGrath <roland@redhat.com> Cc: Serge Hallyn <serue@us.ibm.com> Signed-off-by:
-
Oleg Nesterov authored
send_signal() shouldn't call signalfd_notify() if it then fails with -EAGAIN. Harmless, just a paranoid cleanup. Also remove the comment. It is obsolete, signalfd_notify() was simplified and does a simple wakeup. Signed-off-by:
Davide Libenzi <davidel@xmailserver.org> Cc: Roland McGrath <roland@redhat.com> Signed-off-by:
-
Oleg Nesterov authored
A couple of small comments about how CLD_CONTINUED notification works. Signed-off-by:
-
Oleg Nesterov authored
Rename handle_stop_signal() to prepare_signal(), make it return a boolean, and move the callsites of sig_ignored() into it. No functional changes for now. But it would be nice to factor out the "should we drop this signal" checks as much as possible, before we try to fix the bugs with the sub-namespace init's signals (actually the global /sbin/init has some problems with signals too). Signed-off-by:
-
Oleg Nesterov authored
Move the callsite of print_fatal_signal() down, under "if (sig_kernel_coredump(signr))", so we don't need to check signr != SIGKILL. We are only interested in the sig_kernel_coredump() signals anyway, and due to the previous changes we almost never can see other fatal signals here except SIGKILL. Signed-off-by:
-
Oleg Nesterov authored
handle_stop_signal() clears SIGNAL_STOP_DEQUEUED when sig == SIGKILL. Remove this nasty special case. It was needed to prevent the race with group stop and exit caused by thread-specific SIGKILL. Now that we use complete_signal() for private signals too this is not needed, complete_signal() will notice SIGKILL and abort the soon-to-begin group stop. Except: the target thread is dead (has PF_EXITING). But in that case we should not just clear SIGNAL_STOP_DEQUEUED and nothing more. We should either kill the whole thread group, or silently ignore the signal. I suspect we are not right wrt zombie leaders, but this is another issue which and should be fixed separately. Note that this check can't abort the group stop if it was already started/finished, this check only adds a subtle side effect if we race with the thread which has already dequeued sig_kernel_stop() signal and temporary released ->siglock. Signed-off-by:
-
Oleg Nesterov authored
We export send_sigqueue() and send_group_sigqueue() for the only user, posix_timer_event(). This is a bit silly, because both are just trivial helpers on top of do_send_sigqueue() and because the we pass the unused .si_signo parameter. Kill them both, rename do_send_sigqueue() to send_sigqueue(), and export it. Signed-off-by:
-
Oleg Nesterov authored
Suggested by Pavel Emelyanov. send_sigqueue/send_group_sigqueue are only differ in how they lock ->siglock. Unify them. send_group_sigqueue() uses spin_lock() because it knows the task can't exit, but in that case lock_task_sighand() can't fail and doesn't hurt. Note that the "sig" argument is ignored, it is always equal to ->si_signo. Signed-off-by:
Pavel Emelyanov <xemul@openvz.org> Signed-off-by:
-
Pavel Emelyanov authored
Factor out complete_signal() callsites. This change completely unifies the helpers sending the specific/group signals. Signed-off-by:
-
Oleg Nesterov authored
Based on Pavel Emelyanov's suggestion. Rename __group_complete_signal() to complete_signal() and use it to process the specific signals too. To do this we simply add the "int group" argument. This allows us to greatly simply the signal-sending code and adds a useful behaviour change. We can avoid the unneeded wakeups for the private signals because wants_signal() is more clever than sigismember(blocked), but more importantly we now take into account the fatal specific signals too. The latter allows us to kill some subtle checks in handle_stop_signal() and makes the specific/group signal's behaviour more consistent. For example, currently sigtimedwait(FATAL_SIGNAL) behaves differently depending on was the signal sent by kill() or tkill() if the signal was not blocked. And. This allows us to tweak/fix the behaviour when the specific signal is sent to the dying/dead ->group_leader. Signed-off-by:
-
Oleg Nesterov authored
send_signal() is used either with ->pending or with ->signal->shared_pending. Change it to take "int group" instead, this argument will be re-used later. Signed-off-by:
-
Oleg Nesterov authored
Move the unchanged definition of __group_complete_signal() so that send_signal can see it. To simplify the reading of the next patches. Signed-off-by:
-
Oleg Nesterov authored
Suggested by Roland McGrath. Initialize signal->curr_target in copy_signal(). This way ->curr_target is never == NULL, we can kill the check in __group_complete_signal's hot path. Signed-off-by:
-
Oleg Nesterov authored
The comment in send_sig_info() is wrong, tasklist_lock can't help. The caller must ensure the task can't go away, otherwise ->sighand can be NULL even before we take the lock. p->sighand could be changed by exec(), but I can't imagine how it is possible to prevent exit(), but not exec(). Since the things seem to work, I assume all callers are correct. However, drm_vbl_send_signals() looks broken. block_all_signals() which is solely used by drm is definitely broken. Signed-off-by:
-
Oleg Nesterov authored
Convert do_tkill() to use rcu_read_lock() + lock_task_sighand() to avoid taking tasklist lock. Note that we don't return an error if lock_task_sighand() fails, we pretend the task dies after receiving the signal. Otherwise, we should fight with the nasty races with mt-exec without having any advantage. Signed-off-by:
-
Oleg Nesterov authored
Move handle_stop_signal() into send_signal(). This factors out a couple of callsites and allows us to do further unifications. Also, with this change specific_send_sig_info() does handle_stop_signal(). Not that this is really important, we never send STOP/CONT via send_sig() and friends, but still this looks more consistent. The only (afaics) special case is get_signal_to_deliver(). If the traced task dequeues SIGCONT, it can re-send it to itself after ptrace_stop() if the signal was blocked by debugger. In that case handle_stop_signal() is unnecessary, but hopefully not a problem. Signed-off-by:
-
Oleg Nesterov authored
handle_stop_signal() was changed, now send_group_sigqueue() doesn't need tasklist_lock. Signed-off-by:
-
Oleg Nesterov authored
Cosmetic, cache p->signal to make the code a bit more readable. Signed-off-by:
-
Oleg Nesterov authored
send_group_sigqueue() calls handle_stop_signal(), send_sigqueue() doesn't. This is not consistent and in fact I'd say this is (minor) bug. Move handle_stop_signal() from send_group_sigqueue() to do_send_sigqueue(), the latter is called by send_sigqueue() too. Signed-off-by:
-
Oleg Nesterov authored
lock_task_sighand() was changed, send_sigqueue() doesn't need rcu_read_lock() any longer. Signed-off-by:
-
Oleg Nesterov authored
Cache the values of current->signal/sighand. Shrinks .text a bit and makes the code more readable. Also, remove "sigset_t *mask", it is pointless because in fact we save the constant offset. Signed-off-by:
-
Oleg Nesterov authored
Cache the value of p->signal, and change the code to use while_each_thread() helper. Signed-off-by:
-
Oleg Nesterov authored
Now that handle_stop_signal() doesn't drop ->siglock, we can't see both ->group_stop_count && SIGNAL_STOP_STOPPED. Merge two "if" branches. As Roland pointed out, we never actually needed 2 do_notify_parent_cldstop() calls. Signed-off-by:
-
Oleg Nesterov authored
Previously handle_stop_signal(SIGCONT) could drop ->siglock. That is why kill_pid_info(SIGCONT) takes tasklist_lock to make sure the target task can't go away after unlock. Not needed now. Signed-off-by:
-
Oleg Nesterov authored
Based on discussion with Jiri and Roland. In short: currently handle_stop_signal(SIGCONT, p) sends the notification to p->parent, with this patch p itself notifies its parent when it becomes running. handle_stop_signal(SIGCONT) has to drop ->siglock temporary in order to notify the parent with do_notify_parent_cldstop(). This leads to multiple problems: - as Jiri Kosina pointed out, the stopped task can resume without actually seeing SIGCONT which may have a handler. - we race with another sig_kernel_stop() signal which may come in that window. - we race with sig_fatal() signals which may set SIGNAL_GROUP_EXIT in that window. - we can't avoid taking tasklist_lock() while sending SIGCONT. With this patch handle_stop_signal() just sets the new SIGNAL_CLD_CONTINUED flag in p->signal->flags and returns. The notification is sent by the first task which returns from finish_stop() (there should be at least one) or any other signalled thread from get_signal_to_deliver(). This is a user-visible change. Say, currently kill(SIGCONT, stopped_child) can't return without seeing SIGCHLD, with this patch SIGCHLD can be delayed unpredictably. Another difference is that if the child is ptraced by another process, CLD_CONTINUED may be delivered to ->real_parent after ptrace_detach() while currently it always goes to the tracer which doesn't actually need this notification. Hopefully not a problem. The patch asks for the futher obvious cleanups, I'll send them separately. Signed-off-by:
-
Oleg Nesterov authored
Every implementation of ->task_kill() does nothing when the signal comes from the kernel. This is correct, but means that check_kill_permission() should call security_task_kill() only for SI_FROMUSER() case, and we can remove the same check from ->task_kill() implementations. (sadly, check_kill_permission() is the last user of signal->session/__session but we can't s/task_session_nr/task_session/ here). NOTE: Eric W. Biederman pointed out cap_task_kill() should die, and I think he is very right. Signed-off-by:
-
Pavel Emelyanov authored
Both functions do the same thing after proper locking, but with different sigpending structs, so move the common code into a helper. After this we have 4 places that look very similar: send_sigqueue: calls do_send_sigqueue and signal_wakeup send_group_sigqueue: calls do_send_sigqueue and __group_complete_signal __group_send_sig_info: calls send_signal and __group_complete_signal specific_send_sig_info: calls send_signal and signal_wakeup Besides, send_signal performs actions similar to do_send_sigqueue's and __group_complete_signal - to signal_wakeup. It looks like they can be consolidated gracefully. Oleg said: Personally, I think this change is very good. But send_sigqueue() and send_group_sigqueue() have a very subtle difference which I was never able to understand. Let's suppose that sigqueue is already queued, and the signal is ignored (the latter means we should re-schedule cpu timer or handle overrruns). In that case send_sigqueue() returns 0, but send_group_sigqueue() returns 1. I think this is not the problem (in fact, I think this patch makes the behaviour more correct), but I hope Thomas can take a look and confirm. Signed-off-by:
-
Pavel Emelyanov authored
The signr variable may be declared without initialization - it is set ro the return value from __dequeue_signal() right at the function beginning. Besides, after recalc_sigpending() two checks for signr to be not 0 may be merged into one. Both if-s become easier to read. Thanks to Oleg for pointing out mistakes in the first version of this patch. Signed-off-by:
-
Pavel Emelyanov authored
Both sig_ignored() and do_sigaction() check for signr to be explicitly or implicitly ignored. Introduce a helper for them. This patch is aimed to help handling signals by pid namespace's init, and was derived from one of Oleg's patches https://lists.linux-foundation.org/pipermail/containers/2007-December/009308.html so, if he doesn't mind, he should be considered as an author. Signed-off-by:
-
Oleg Nesterov authored
lock_task_sighand() was changed, and do_task_stat() doesn't need rcu_read_lock any longer. sighand->siglock protects all "interesting" fields. Except: it doesn't protect ->tty->pgrp, but neither does rcu_read_lock(), this should be fixed. Signed-off-by:
-
Oleg Nesterov authored
Just a trivial example, more to come. k_getrusage() holds rcu_read_lock() because it was previously required by lock_task_sighand(). Unneeded now. Signed-off-by:
-
Oleg Nesterov authored
Most of the callers of lock_task_sighand() doesn't actually need rcu_lock(). lock_task_sighand() needs it only to safely play with tsk->sighand, it can take the lock itself. Signed-off-by:
-
Oleg Nesterov authored
do_group_exit() checks SIGNAL_GROUP_EXIT to avoid taking sighand->siglock. Since ed5d2cac exec() doesn't set this flag, we should use signal_group_exit(). This is not needed for correctness, but can speedup the multithreaded exec and makes the code more consistent. Signed-off-by:
-
Oleg Nesterov authored
do_signal_stop() needs signal_group_exit() but checks sig->group_exit_task. This (optimization) is correct, SIGNAL_STOP_DEQUEUED and SIGNAL_GROUP_EXIT are mutually exclusive, but looks confusing. Use signal_group_exit(), this is not fastpath, the code clarity is more important. Signed-off-by:
-
Pavel Emelyanov authored
Two callers for send_signal() - the specific_send_sig_info and the __group_send_sig_info - both check for sig to be ignored or already queued. Move these checks into send_signal() and make it return 1 to indicate that the signal is dropped, but there's no error in this. Besides, merge comments and spell-check them. [oleg@tv-sign.ru: simplifications] Signed-off-by:
-
Pavel Emelyanov authored
This makes the code more readable, due to less brackets and small letters in name. I also move it above the send_signal() as a preparation for the 3rd patch. Signed-off-by:
-
