4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/freezer.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
62 #if defined(CONFIG_SYSTEM_LOAD_ANALYZER)
63 #include <linux/load_analyzer.h>
66 static void exit_mm(struct task_struct * tsk);
68 static void __unhash_process(struct task_struct *p, bool group_dead)
71 detach_pid(p, PIDTYPE_PID);
73 detach_pid(p, PIDTYPE_PGID);
74 detach_pid(p, PIDTYPE_SID);
76 list_del_rcu(&p->tasks);
77 list_del_init(&p->sibling);
78 __this_cpu_dec(process_counts);
80 list_del_rcu(&p->thread_group);
81 list_del_rcu(&p->thread_node);
85 * This function expects the tasklist_lock write-locked.
87 static void __exit_signal(struct task_struct *tsk)
89 struct signal_struct *sig = tsk->signal;
90 bool group_dead = thread_group_leader(tsk);
91 struct sighand_struct *sighand;
92 struct tty_struct *uninitialized_var(tty);
93 cputime_t utime, stime;
95 sighand = rcu_dereference_check(tsk->sighand,
96 lockdep_tasklist_lock_is_held());
97 spin_lock(&sighand->siglock);
99 posix_cpu_timers_exit(tsk);
101 posix_cpu_timers_exit_group(tsk);
106 * This can only happen if the caller is de_thread().
107 * FIXME: this is the temporary hack, we should teach
108 * posix-cpu-timers to handle this case correctly.
110 if (unlikely(has_group_leader_pid(tsk)))
111 posix_cpu_timers_exit_group(tsk);
114 * If there is any task waiting for the group exit
117 if (sig->notify_count > 0 && !--sig->notify_count)
118 wake_up_process(sig->group_exit_task);
120 if (tsk == sig->curr_target)
121 sig->curr_target = next_thread(tsk);
123 * Accumulate here the counters for all threads but the
124 * group leader as they die, so they can be added into
125 * the process-wide totals when those are taken.
126 * The group leader stays around as a zombie as long
127 * as there are other threads. When it gets reaped,
128 * the exit.c code will add its counts into these totals.
129 * We won't ever get here for the group leader, since it
130 * will have been the last reference on the signal_struct.
132 task_cputime(tsk, &utime, &stime);
135 sig->gtime += task_gtime(tsk);
136 sig->min_flt += tsk->min_flt;
137 sig->maj_flt += tsk->maj_flt;
138 sig->nvcsw += tsk->nvcsw;
139 sig->nivcsw += tsk->nivcsw;
140 sig->inblock += task_io_get_inblock(tsk);
141 sig->oublock += task_io_get_oublock(tsk);
142 task_io_accounting_add(&sig->ioac, &tsk->ioac);
143 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
147 __unhash_process(tsk, group_dead);
150 * Do this under ->siglock, we can race with another thread
151 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
153 flush_sigqueue(&tsk->pending);
155 spin_unlock(&sighand->siglock);
157 __cleanup_sighand(sighand);
158 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
160 flush_sigqueue(&sig->shared_pending);
165 static void delayed_put_task_struct(struct rcu_head *rhp)
167 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
169 perf_event_delayed_put(tsk);
170 trace_sched_process_free(tsk);
171 put_task_struct(tsk);
175 void release_task(struct task_struct * p)
177 struct task_struct *leader;
180 /* don't need to get the RCU readlock here - the process is dead and
181 * can't be modifying its own credentials. But shut RCU-lockdep up */
183 atomic_dec(&__task_cred(p)->user->processes);
188 write_lock_irq(&tasklist_lock);
189 ptrace_release_task(p);
193 * If we are the last non-leader member of the thread
194 * group, and the leader is zombie, then notify the
195 * group leader's parent process. (if it wants notification.)
198 leader = p->group_leader;
199 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
201 * If we were the last child thread and the leader has
202 * exited already, and the leader's parent ignores SIGCHLD,
203 * then we are the one who should release the leader.
205 zap_leader = do_notify_parent(leader, leader->exit_signal);
207 leader->exit_state = EXIT_DEAD;
210 write_unlock_irq(&tasklist_lock);
212 call_rcu(&p->rcu, delayed_put_task_struct);
215 if (unlikely(zap_leader))
220 * This checks not only the pgrp, but falls back on the pid if no
221 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
224 * The caller must hold rcu lock or the tasklist lock.
226 struct pid *session_of_pgrp(struct pid *pgrp)
228 struct task_struct *p;
229 struct pid *sid = NULL;
231 p = pid_task(pgrp, PIDTYPE_PGID);
233 p = pid_task(pgrp, PIDTYPE_PID);
235 sid = task_session(p);
241 * Determine if a process group is "orphaned", according to the POSIX
242 * definition in 2.2.2.52. Orphaned process groups are not to be affected
243 * by terminal-generated stop signals. Newly orphaned process groups are
244 * to receive a SIGHUP and a SIGCONT.
246 * "I ask you, have you ever known what it is to be an orphan?"
248 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
250 struct task_struct *p;
252 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253 if ((p == ignored_task) ||
254 (p->exit_state && thread_group_empty(p)) ||
255 is_global_init(p->real_parent))
258 if (task_pgrp(p->real_parent) != pgrp &&
259 task_session(p->real_parent) == task_session(p))
261 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
266 int is_current_pgrp_orphaned(void)
270 read_lock(&tasklist_lock);
271 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
272 read_unlock(&tasklist_lock);
277 static bool has_stopped_jobs(struct pid *pgrp)
279 struct task_struct *p;
281 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
282 if (p->signal->flags & SIGNAL_STOP_STOPPED)
284 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
290 * Check to see if any process groups have become orphaned as
291 * a result of our exiting, and if they have any stopped jobs,
292 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
295 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
297 struct pid *pgrp = task_pgrp(tsk);
298 struct task_struct *ignored_task = tsk;
301 /* exit: our father is in a different pgrp than
302 * we are and we were the only connection outside.
304 parent = tsk->real_parent;
306 /* reparent: our child is in a different pgrp than
307 * we are, and it was the only connection outside.
311 if (task_pgrp(parent) != pgrp &&
312 task_session(parent) == task_session(tsk) &&
313 will_become_orphaned_pgrp(pgrp, ignored_task) &&
314 has_stopped_jobs(pgrp)) {
315 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
316 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
320 void __set_special_pids(struct pid *pid)
322 struct task_struct *curr = current->group_leader;
324 if (task_session(curr) != pid)
325 change_pid(curr, PIDTYPE_SID, pid);
327 if (task_pgrp(curr) != pid)
328 change_pid(curr, PIDTYPE_PGID, pid);
332 * Let kernel threads use this to say that they allow a certain signal.
333 * Must not be used if kthread was cloned with CLONE_SIGHAND.
335 int allow_signal(int sig)
337 if (!valid_signal(sig) || sig < 1)
340 spin_lock_irq(¤t->sighand->siglock);
341 /* This is only needed for daemonize()'ed kthreads */
342 sigdelset(¤t->blocked, sig);
344 * Kernel threads handle their own signals. Let the signal code
345 * know it'll be handled, so that they don't get converted to
346 * SIGKILL or just silently dropped.
348 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
350 spin_unlock_irq(¤t->sighand->siglock);
354 EXPORT_SYMBOL(allow_signal);
356 int disallow_signal(int sig)
358 if (!valid_signal(sig) || sig < 1)
361 spin_lock_irq(¤t->sighand->siglock);
362 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
364 spin_unlock_irq(¤t->sighand->siglock);
368 EXPORT_SYMBOL(disallow_signal);
370 #ifdef CONFIG_MM_OWNER
372 * A task is exiting. If it owned this mm, find a new owner for the mm.
374 void mm_update_next_owner(struct mm_struct *mm)
376 struct task_struct *c, *g, *p = current;
380 * If the exiting or execing task is not the owner, it's
381 * someone else's problem.
386 * The current owner is exiting/execing and there are no other
387 * candidates. Do not leave the mm pointing to a possibly
388 * freed task structure.
390 if (atomic_read(&mm->mm_users) <= 1) {
395 read_lock(&tasklist_lock);
397 * Search in the children
399 list_for_each_entry(c, &p->children, sibling) {
401 goto assign_new_owner;
405 * Search in the siblings
407 list_for_each_entry(c, &p->real_parent->children, sibling) {
409 goto assign_new_owner;
413 * Search through everything else. We should not get
416 do_each_thread(g, c) {
418 goto assign_new_owner;
419 } while_each_thread(g, c);
421 read_unlock(&tasklist_lock);
423 * We found no owner yet mm_users > 1: this implies that we are
424 * most likely racing with swapoff (try_to_unuse()) or /proc or
425 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
434 * The task_lock protects c->mm from changing.
435 * We always want mm->owner->mm == mm
439 * Delay read_unlock() till we have the task_lock()
440 * to ensure that c does not slip away underneath us
442 read_unlock(&tasklist_lock);
452 #endif /* CONFIG_MM_OWNER */
455 * Turn us into a lazy TLB process if we
458 static void exit_mm(struct task_struct * tsk)
460 struct mm_struct *mm = tsk->mm;
461 struct core_state *core_state;
468 * Serialize with any possible pending coredump.
469 * We must hold mmap_sem around checking core_state
470 * and clearing tsk->mm. The core-inducing thread
471 * will increment ->nr_threads for each thread in the
472 * group with ->mm != NULL.
474 down_read(&mm->mmap_sem);
475 core_state = mm->core_state;
477 struct core_thread self;
478 up_read(&mm->mmap_sem);
481 self.next = xchg(&core_state->dumper.next, &self);
483 * Implies mb(), the result of xchg() must be visible
484 * to core_state->dumper.
486 if (atomic_dec_and_test(&core_state->nr_threads))
487 complete(&core_state->startup);
490 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
491 if (!self.task) /* see coredump_finish() */
493 freezable_schedule();
495 __set_task_state(tsk, TASK_RUNNING);
496 down_read(&mm->mmap_sem);
498 atomic_inc(&mm->mm_count);
499 BUG_ON(mm != tsk->active_mm);
500 /* more a memory barrier than a real lock */
503 up_read(&mm->mmap_sem);
504 enter_lazy_tlb(mm, current);
506 mm_update_next_owner(mm);
511 * When we die, we re-parent all our children, and try to:
512 * 1. give them to another thread in our thread group, if such a member exists
513 * 2. give it to the first ancestor process which prctl'd itself as a
514 * child_subreaper for its children (like a service manager)
515 * 3. give it to the init process (PID 1) in our pid namespace
517 static struct task_struct *find_new_reaper(struct task_struct *father)
518 __releases(&tasklist_lock)
519 __acquires(&tasklist_lock)
521 struct pid_namespace *pid_ns = task_active_pid_ns(father);
522 struct task_struct *thread;
525 while_each_thread(father, thread) {
526 if (thread->flags & PF_EXITING)
528 if (unlikely(pid_ns->child_reaper == father))
529 pid_ns->child_reaper = thread;
533 if (unlikely(pid_ns->child_reaper == father)) {
534 write_unlock_irq(&tasklist_lock);
535 if (unlikely(pid_ns == &init_pid_ns)) {
536 panic("Attempted to kill init! exitcode=0x%08x\n",
537 father->signal->group_exit_code ?:
541 zap_pid_ns_processes(pid_ns);
542 write_lock_irq(&tasklist_lock);
543 } else if (father->signal->has_child_subreaper) {
544 struct task_struct *reaper;
547 * Find the first ancestor marked as child_subreaper.
548 * Note that the code below checks same_thread_group(reaper,
549 * pid_ns->child_reaper). This is what we need to DTRT in a
550 * PID namespace. However we still need the check above, see
551 * http://marc.info/?l=linux-kernel&m=131385460420380
553 for (reaper = father->real_parent;
554 reaper != &init_task;
555 reaper = reaper->real_parent) {
556 if (same_thread_group(reaper, pid_ns->child_reaper))
558 if (!reaper->signal->is_child_subreaper)
562 if (!(thread->flags & PF_EXITING))
564 } while_each_thread(reaper, thread);
568 return pid_ns->child_reaper;
572 * Any that need to be release_task'd are put on the @dead list.
574 static void reparent_leader(struct task_struct *father, struct task_struct *p,
575 struct list_head *dead)
577 list_move_tail(&p->sibling, &p->real_parent->children);
579 * If this is a threaded reparent there is no need to
580 * notify anyone anything has happened.
582 if (same_thread_group(p->real_parent, father))
586 * We don't want people slaying init.
588 * Note: we do this even if it is EXIT_DEAD, wait_task_zombie()
589 * can change ->exit_state to EXIT_ZOMBIE. If this is the final
590 * state, do_notify_parent() was already called and ->exit_signal
593 p->exit_signal = SIGCHLD;
595 if (p->exit_state == EXIT_DEAD)
598 /* If it has exited notify the new parent about this child's death. */
600 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
601 if (do_notify_parent(p, p->exit_signal)) {
602 p->exit_state = EXIT_DEAD;
603 list_move_tail(&p->sibling, dead);
607 kill_orphaned_pgrp(p, father);
610 static void forget_original_parent(struct task_struct *father)
612 struct task_struct *p, *n, *reaper;
613 LIST_HEAD(dead_children);
615 write_lock_irq(&tasklist_lock);
617 * Note that exit_ptrace() and find_new_reaper() might
618 * drop tasklist_lock and reacquire it.
621 reaper = find_new_reaper(father);
623 list_for_each_entry_safe(p, n, &father->children, sibling) {
624 struct task_struct *t = p;
626 t->real_parent = reaper;
627 if (t->parent == father) {
629 t->parent = t->real_parent;
631 if (t->pdeath_signal)
632 group_send_sig_info(t->pdeath_signal,
634 } while_each_thread(p, t);
635 reparent_leader(father, p, &dead_children);
637 write_unlock_irq(&tasklist_lock);
639 BUG_ON(!list_empty(&father->children));
641 list_for_each_entry_safe(p, n, &dead_children, sibling) {
642 list_del_init(&p->sibling);
648 * Send signals to all our closest relatives so that they know
649 * to properly mourn us..
651 static void exit_notify(struct task_struct *tsk, int group_dead)
656 * This does two things:
658 * A. Make init inherit all the child processes
659 * B. Check to see if any process groups have become orphaned
660 * as a result of our exiting, and if they have any stopped
661 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
663 forget_original_parent(tsk);
665 write_lock_irq(&tasklist_lock);
667 kill_orphaned_pgrp(tsk->group_leader, NULL);
669 if (unlikely(tsk->ptrace)) {
670 int sig = thread_group_leader(tsk) &&
671 thread_group_empty(tsk) &&
672 !ptrace_reparented(tsk) ?
673 tsk->exit_signal : SIGCHLD;
674 autoreap = do_notify_parent(tsk, sig);
675 } else if (thread_group_leader(tsk)) {
676 autoreap = thread_group_empty(tsk) &&
677 do_notify_parent(tsk, tsk->exit_signal);
682 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
684 /* mt-exec, de_thread() is waiting for group leader */
685 if (unlikely(tsk->signal->notify_count < 0))
686 wake_up_process(tsk->signal->group_exit_task);
687 write_unlock_irq(&tasklist_lock);
689 /* If the process is dead, release it - nobody will wait for it */
694 #ifdef CONFIG_DEBUG_STACK_USAGE
695 static void check_stack_usage(void)
697 static DEFINE_SPINLOCK(low_water_lock);
698 static int lowest_to_date = THREAD_SIZE;
701 free = stack_not_used(current);
703 if (free >= lowest_to_date)
706 spin_lock(&low_water_lock);
707 if (free < lowest_to_date) {
708 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
710 current->comm, task_pid_nr(current), free);
711 lowest_to_date = free;
713 spin_unlock(&low_water_lock);
716 static inline void check_stack_usage(void) {}
719 void do_exit(long code)
721 struct task_struct *tsk = current;
724 #if defined(CONFIG_SYSTEM_LOAD_ANALYZER)
725 store_killed_task(tsk);
728 profile_task_exit(tsk);
730 WARN_ON(blk_needs_flush_plug(tsk));
732 if (unlikely(in_interrupt()))
733 panic("Aiee, killing interrupt handler!");
734 if (unlikely(!tsk->pid))
735 panic("Attempted to kill the idle task!");
738 * If do_exit is called because this processes oopsed, it's possible
739 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
740 * continuing. Amongst other possible reasons, this is to prevent
741 * mm_release()->clear_child_tid() from writing to a user-controlled
746 ptrace_event(PTRACE_EVENT_EXIT, code);
748 validate_creds_for_do_exit(tsk);
751 * We're taking recursive faults here in do_exit. Safest is to just
752 * leave this task alone and wait for reboot.
754 if (unlikely(tsk->flags & PF_EXITING)) {
756 "Fixing recursive fault but reboot is needed!\n");
758 * We can do this unlocked here. The futex code uses
759 * this flag just to verify whether the pi state
760 * cleanup has been done or not. In the worst case it
761 * loops once more. We pretend that the cleanup was
762 * done as there is no way to return. Either the
763 * OWNER_DIED bit is set by now or we push the blocked
764 * task into the wait for ever nirwana as well.
766 tsk->flags |= PF_EXITPIDONE;
767 set_current_state(TASK_UNINTERRUPTIBLE);
771 exit_signals(tsk); /* sets PF_EXITING */
773 * tsk->flags are checked in the futex code to protect against
774 * an exiting task cleaning up the robust pi futexes.
777 raw_spin_unlock_wait(&tsk->pi_lock);
779 if (unlikely(in_atomic()))
780 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
781 current->comm, task_pid_nr(current),
784 acct_update_integrals(tsk);
785 /* sync mm's RSS info before statistics gathering */
787 sync_mm_rss(tsk->mm);
788 group_dead = atomic_dec_and_test(&tsk->signal->live);
790 hrtimer_cancel(&tsk->signal->real_timer);
791 exit_itimers(tsk->signal);
793 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
795 acct_collect(code, group_dead);
800 tsk->exit_code = code;
801 taskstats_exit(tsk, group_dead);
807 trace_sched_process_exit(tsk);
814 disassociate_ctty(1);
815 exit_task_namespaces(tsk);
821 * Flush inherited counters to the parent - before the parent
822 * gets woken up by child-exit notifications.
824 * because of cgroup mode, must be called before cgroup_exit()
826 perf_event_exit_task(tsk);
830 module_put(task_thread_info(tsk)->exec_domain->module);
832 proc_exit_connector(tsk);
834 * FIXME: do that only when needed, using sched_exit tracepoint
836 ptrace_put_breakpoints(tsk);
838 exit_notify(tsk, group_dead);
841 mpol_put(tsk->mempolicy);
842 tsk->mempolicy = NULL;
846 if (unlikely(current->pi_state_cache))
847 kfree(current->pi_state_cache);
850 * Make sure we are holding no locks:
852 debug_check_no_locks_held();
854 * We can do this unlocked here. The futex code uses this flag
855 * just to verify whether the pi state cleanup has been done
856 * or not. In the worst case it loops once more.
858 tsk->flags |= PF_EXITPIDONE;
861 exit_io_context(tsk);
863 if (tsk->splice_pipe)
864 free_pipe_info(tsk->splice_pipe);
866 if (tsk->task_frag.page)
867 put_page(tsk->task_frag.page);
869 validate_creds_for_do_exit(tsk);
873 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
877 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
878 * when the following two conditions become true.
879 * - There is race condition of mmap_sem (It is acquired by
881 * - SMI occurs before setting TASK_RUNINNG.
882 * (or hypervisor of virtual machine switches to other guest)
883 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
885 * To avoid it, we have to wait for releasing tsk->pi_lock which
886 * is held by try_to_wake_up()
889 raw_spin_unlock_wait(&tsk->pi_lock);
891 /* causes final put_task_struct in finish_task_switch(). */
892 tsk->state = TASK_DEAD;
893 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
896 /* Avoid "noreturn function does return". */
898 cpu_relax(); /* For when BUG is null */
901 EXPORT_SYMBOL_GPL(do_exit);
903 void complete_and_exit(struct completion *comp, long code)
911 EXPORT_SYMBOL(complete_and_exit);
913 SYSCALL_DEFINE1(exit, int, error_code)
915 do_exit((error_code&0xff)<<8);
919 * Take down every thread in the group. This is called by fatal signals
920 * as well as by sys_exit_group (below).
923 do_group_exit(int exit_code)
925 struct signal_struct *sig = current->signal;
927 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
929 if (signal_group_exit(sig))
930 exit_code = sig->group_exit_code;
931 else if (!thread_group_empty(current)) {
932 struct sighand_struct *const sighand = current->sighand;
933 spin_lock_irq(&sighand->siglock);
934 if (signal_group_exit(sig))
935 /* Another thread got here before we took the lock. */
936 exit_code = sig->group_exit_code;
938 sig->group_exit_code = exit_code;
939 sig->flags = SIGNAL_GROUP_EXIT;
940 zap_other_threads(current);
942 spin_unlock_irq(&sighand->siglock);
950 * this kills every thread in the thread group. Note that any externally
951 * wait4()-ing process will get the correct exit code - even if this
952 * thread is not the thread group leader.
954 SYSCALL_DEFINE1(exit_group, int, error_code)
956 do_group_exit((error_code & 0xff) << 8);
962 enum pid_type wo_type;
966 struct siginfo __user *wo_info;
968 struct rusage __user *wo_rusage;
970 wait_queue_t child_wait;
975 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
977 if (type != PIDTYPE_PID)
978 task = task->group_leader;
979 return task->pids[type].pid;
982 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
984 return wo->wo_type == PIDTYPE_MAX ||
985 task_pid_type(p, wo->wo_type) == wo->wo_pid;
988 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
990 if (!eligible_pid(wo, p))
992 /* Wait for all children (clone and not) if __WALL is set;
993 * otherwise, wait for clone children *only* if __WCLONE is
994 * set; otherwise, wait for non-clone children *only*. (Note:
995 * A "clone" child here is one that reports to its parent
996 * using a signal other than SIGCHLD.) */
997 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
998 && !(wo->wo_flags & __WALL))
1004 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1005 pid_t pid, uid_t uid, int why, int status)
1007 struct siginfo __user *infop;
1008 int retval = wo->wo_rusage
1009 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1012 infop = wo->wo_info;
1015 retval = put_user(SIGCHLD, &infop->si_signo);
1017 retval = put_user(0, &infop->si_errno);
1019 retval = put_user((short)why, &infop->si_code);
1021 retval = put_user(pid, &infop->si_pid);
1023 retval = put_user(uid, &infop->si_uid);
1025 retval = put_user(status, &infop->si_status);
1033 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1034 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1035 * the lock and this task is uninteresting. If we return nonzero, we have
1036 * released the lock and the system call should return.
1038 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1040 unsigned long state;
1041 int retval, status, traced;
1042 pid_t pid = task_pid_vnr(p);
1043 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1044 struct siginfo __user *infop;
1046 if (!likely(wo->wo_flags & WEXITED))
1049 if (unlikely(wo->wo_flags & WNOWAIT)) {
1050 int exit_code = p->exit_code;
1054 read_unlock(&tasklist_lock);
1055 if ((exit_code & 0x7f) == 0) {
1057 status = exit_code >> 8;
1059 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1060 status = exit_code & 0x7f;
1062 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1066 * Try to move the task's state to DEAD
1067 * only one thread is allowed to do this:
1069 state = xchg(&p->exit_state, EXIT_DEAD);
1070 if (state != EXIT_ZOMBIE) {
1071 BUG_ON(state != EXIT_DEAD);
1075 traced = ptrace_reparented(p);
1077 * It can be ptraced but not reparented, check
1078 * thread_group_leader() to filter out sub-threads.
1080 if (likely(!traced) && thread_group_leader(p)) {
1081 struct signal_struct *psig;
1082 struct signal_struct *sig;
1083 unsigned long maxrss;
1084 cputime_t tgutime, tgstime;
1087 * The resource counters for the group leader are in its
1088 * own task_struct. Those for dead threads in the group
1089 * are in its signal_struct, as are those for the child
1090 * processes it has previously reaped. All these
1091 * accumulate in the parent's signal_struct c* fields.
1093 * We don't bother to take a lock here to protect these
1094 * p->signal fields, because they are only touched by
1095 * __exit_signal, which runs with tasklist_lock
1096 * write-locked anyway, and so is excluded here. We do
1097 * need to protect the access to parent->signal fields,
1098 * as other threads in the parent group can be right
1099 * here reaping other children at the same time.
1101 * We use thread_group_cputime_adjusted() to get times for the thread
1102 * group, which consolidates times for all threads in the
1103 * group including the group leader.
1105 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1106 spin_lock_irq(&p->real_parent->sighand->siglock);
1107 psig = p->real_parent->signal;
1109 psig->cutime += tgutime + sig->cutime;
1110 psig->cstime += tgstime + sig->cstime;
1111 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1113 p->min_flt + sig->min_flt + sig->cmin_flt;
1115 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1117 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1119 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1121 task_io_get_inblock(p) +
1122 sig->inblock + sig->cinblock;
1124 task_io_get_oublock(p) +
1125 sig->oublock + sig->coublock;
1126 maxrss = max(sig->maxrss, sig->cmaxrss);
1127 if (psig->cmaxrss < maxrss)
1128 psig->cmaxrss = maxrss;
1129 task_io_accounting_add(&psig->ioac, &p->ioac);
1130 task_io_accounting_add(&psig->ioac, &sig->ioac);
1131 spin_unlock_irq(&p->real_parent->sighand->siglock);
1135 * Now we are sure this task is interesting, and no other
1136 * thread can reap it because we set its state to EXIT_DEAD.
1138 read_unlock(&tasklist_lock);
1140 retval = wo->wo_rusage
1141 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1142 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1143 ? p->signal->group_exit_code : p->exit_code;
1144 if (!retval && wo->wo_stat)
1145 retval = put_user(status, wo->wo_stat);
1147 infop = wo->wo_info;
1148 if (!retval && infop)
1149 retval = put_user(SIGCHLD, &infop->si_signo);
1150 if (!retval && infop)
1151 retval = put_user(0, &infop->si_errno);
1152 if (!retval && infop) {
1155 if ((status & 0x7f) == 0) {
1159 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1162 retval = put_user((short)why, &infop->si_code);
1164 retval = put_user(status, &infop->si_status);
1166 if (!retval && infop)
1167 retval = put_user(pid, &infop->si_pid);
1168 if (!retval && infop)
1169 retval = put_user(uid, &infop->si_uid);
1174 write_lock_irq(&tasklist_lock);
1175 /* We dropped tasklist, ptracer could die and untrace */
1178 * If this is not a sub-thread, notify the parent.
1179 * If parent wants a zombie, don't release it now.
1181 if (thread_group_leader(p) &&
1182 !do_notify_parent(p, p->exit_signal)) {
1183 p->exit_state = EXIT_ZOMBIE;
1186 write_unlock_irq(&tasklist_lock);
1194 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1197 if (task_is_stopped_or_traced(p) &&
1198 !(p->jobctl & JOBCTL_LISTENING))
1199 return &p->exit_code;
1201 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1202 return &p->signal->group_exit_code;
1208 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1210 * @ptrace: is the wait for ptrace
1211 * @p: task to wait for
1213 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1216 * read_lock(&tasklist_lock), which is released if return value is
1217 * non-zero. Also, grabs and releases @p->sighand->siglock.
1220 * 0 if wait condition didn't exist and search for other wait conditions
1221 * should continue. Non-zero return, -errno on failure and @p's pid on
1222 * success, implies that tasklist_lock is released and wait condition
1223 * search should terminate.
1225 static int wait_task_stopped(struct wait_opts *wo,
1226 int ptrace, struct task_struct *p)
1228 struct siginfo __user *infop;
1229 int retval, exit_code, *p_code, why;
1230 uid_t uid = 0; /* unneeded, required by compiler */
1234 * Traditionally we see ptrace'd stopped tasks regardless of options.
1236 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1239 if (!task_stopped_code(p, ptrace))
1243 spin_lock_irq(&p->sighand->siglock);
1245 p_code = task_stopped_code(p, ptrace);
1246 if (unlikely(!p_code))
1249 exit_code = *p_code;
1253 if (!unlikely(wo->wo_flags & WNOWAIT))
1256 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1258 spin_unlock_irq(&p->sighand->siglock);
1263 * Now we are pretty sure this task is interesting.
1264 * Make sure it doesn't get reaped out from under us while we
1265 * give up the lock and then examine it below. We don't want to
1266 * keep holding onto the tasklist_lock while we call getrusage and
1267 * possibly take page faults for user memory.
1270 pid = task_pid_vnr(p);
1271 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1272 read_unlock(&tasklist_lock);
1274 if (unlikely(wo->wo_flags & WNOWAIT))
1275 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1277 retval = wo->wo_rusage
1278 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1279 if (!retval && wo->wo_stat)
1280 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1282 infop = wo->wo_info;
1283 if (!retval && infop)
1284 retval = put_user(SIGCHLD, &infop->si_signo);
1285 if (!retval && infop)
1286 retval = put_user(0, &infop->si_errno);
1287 if (!retval && infop)
1288 retval = put_user((short)why, &infop->si_code);
1289 if (!retval && infop)
1290 retval = put_user(exit_code, &infop->si_status);
1291 if (!retval && infop)
1292 retval = put_user(pid, &infop->si_pid);
1293 if (!retval && infop)
1294 retval = put_user(uid, &infop->si_uid);
1304 * Handle do_wait work for one task in a live, non-stopped state.
1305 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1306 * the lock and this task is uninteresting. If we return nonzero, we have
1307 * released the lock and the system call should return.
1309 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1315 if (!unlikely(wo->wo_flags & WCONTINUED))
1318 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1321 spin_lock_irq(&p->sighand->siglock);
1322 /* Re-check with the lock held. */
1323 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1324 spin_unlock_irq(&p->sighand->siglock);
1327 if (!unlikely(wo->wo_flags & WNOWAIT))
1328 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1329 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1330 spin_unlock_irq(&p->sighand->siglock);
1332 pid = task_pid_vnr(p);
1334 read_unlock(&tasklist_lock);
1337 retval = wo->wo_rusage
1338 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1340 if (!retval && wo->wo_stat)
1341 retval = put_user(0xffff, wo->wo_stat);
1345 retval = wait_noreap_copyout(wo, p, pid, uid,
1346 CLD_CONTINUED, SIGCONT);
1347 BUG_ON(retval == 0);
1354 * Consider @p for a wait by @parent.
1356 * -ECHILD should be in ->notask_error before the first call.
1357 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1358 * Returns zero if the search for a child should continue;
1359 * then ->notask_error is 0 if @p is an eligible child,
1360 * or another error from security_task_wait(), or still -ECHILD.
1362 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1363 struct task_struct *p)
1365 int ret = eligible_child(wo, p);
1369 ret = security_task_wait(p);
1370 if (unlikely(ret < 0)) {
1372 * If we have not yet seen any eligible child,
1373 * then let this error code replace -ECHILD.
1374 * A permission error will give the user a clue
1375 * to look for security policy problems, rather
1376 * than for mysterious wait bugs.
1378 if (wo->notask_error)
1379 wo->notask_error = ret;
1383 /* dead body doesn't have much to contribute */
1384 if (unlikely(p->exit_state == EXIT_DEAD)) {
1386 * But do not ignore this task until the tracer does
1387 * wait_task_zombie()->do_notify_parent().
1389 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1390 wo->notask_error = 0;
1395 if (p->exit_state == EXIT_ZOMBIE) {
1397 * A zombie ptracee is only visible to its ptracer.
1398 * Notification and reaping will be cascaded to the real
1399 * parent when the ptracer detaches.
1401 if (likely(!ptrace) && unlikely(p->ptrace)) {
1402 /* it will become visible, clear notask_error */
1403 wo->notask_error = 0;
1407 /* we don't reap group leaders with subthreads */
1408 if (!delay_group_leader(p))
1409 return wait_task_zombie(wo, p);
1412 * Allow access to stopped/continued state via zombie by
1413 * falling through. Clearing of notask_error is complex.
1417 * If WEXITED is set, notask_error should naturally be
1418 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1419 * so, if there are live subthreads, there are events to
1420 * wait for. If all subthreads are dead, it's still safe
1421 * to clear - this function will be called again in finite
1422 * amount time once all the subthreads are released and
1423 * will then return without clearing.
1427 * Stopped state is per-task and thus can't change once the
1428 * target task dies. Only continued and exited can happen.
1429 * Clear notask_error if WCONTINUED | WEXITED.
1431 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1432 wo->notask_error = 0;
1435 * If @p is ptraced by a task in its real parent's group,
1436 * hide group stop/continued state when looking at @p as
1437 * the real parent; otherwise, a single stop can be
1438 * reported twice as group and ptrace stops.
1440 * If a ptracer wants to distinguish the two events for its
1441 * own children, it should create a separate process which
1442 * takes the role of real parent.
1444 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1448 * @p is alive and it's gonna stop, continue or exit, so
1449 * there always is something to wait for.
1451 wo->notask_error = 0;
1455 * Wait for stopped. Depending on @ptrace, different stopped state
1456 * is used and the two don't interact with each other.
1458 ret = wait_task_stopped(wo, ptrace, p);
1463 * Wait for continued. There's only one continued state and the
1464 * ptracer can consume it which can confuse the real parent. Don't
1465 * use WCONTINUED from ptracer. You don't need or want it.
1467 return wait_task_continued(wo, p);
1471 * Do the work of do_wait() for one thread in the group, @tsk.
1473 * -ECHILD should be in ->notask_error before the first call.
1474 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1475 * Returns zero if the search for a child should continue; then
1476 * ->notask_error is 0 if there were any eligible children,
1477 * or another error from security_task_wait(), or still -ECHILD.
1479 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1481 struct task_struct *p;
1483 list_for_each_entry(p, &tsk->children, sibling) {
1484 int ret = wait_consider_task(wo, 0, p);
1492 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1494 struct task_struct *p;
1496 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1497 int ret = wait_consider_task(wo, 1, p);
1505 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1506 int sync, void *key)
1508 struct wait_opts *wo = container_of(wait, struct wait_opts,
1510 struct task_struct *p = key;
1512 if (!eligible_pid(wo, p))
1515 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1518 return default_wake_function(wait, mode, sync, key);
1521 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1523 __wake_up_sync_key(&parent->signal->wait_chldexit,
1524 TASK_INTERRUPTIBLE, 1, p);
1527 static long do_wait(struct wait_opts *wo)
1529 struct task_struct *tsk;
1532 trace_sched_process_wait(wo->wo_pid);
1534 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1535 wo->child_wait.private = current;
1536 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1539 * If there is nothing that can match our critiera just get out.
1540 * We will clear ->notask_error to zero if we see any child that
1541 * might later match our criteria, even if we are not able to reap
1544 wo->notask_error = -ECHILD;
1545 if ((wo->wo_type < PIDTYPE_MAX) &&
1546 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1549 set_current_state(TASK_INTERRUPTIBLE);
1550 read_lock(&tasklist_lock);
1553 retval = do_wait_thread(wo, tsk);
1557 retval = ptrace_do_wait(wo, tsk);
1561 if (wo->wo_flags & __WNOTHREAD)
1563 } while_each_thread(current, tsk);
1564 read_unlock(&tasklist_lock);
1567 retval = wo->notask_error;
1568 if (!retval && !(wo->wo_flags & WNOHANG)) {
1569 retval = -ERESTARTSYS;
1570 if (!signal_pending(current)) {
1576 __set_current_state(TASK_RUNNING);
1577 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1581 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1582 infop, int, options, struct rusage __user *, ru)
1584 struct wait_opts wo;
1585 struct pid *pid = NULL;
1589 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1591 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1604 type = PIDTYPE_PGID;
1612 if (type < PIDTYPE_MAX)
1613 pid = find_get_pid(upid);
1617 wo.wo_flags = options;
1627 * For a WNOHANG return, clear out all the fields
1628 * we would set so the user can easily tell the
1632 ret = put_user(0, &infop->si_signo);
1634 ret = put_user(0, &infop->si_errno);
1636 ret = put_user(0, &infop->si_code);
1638 ret = put_user(0, &infop->si_pid);
1640 ret = put_user(0, &infop->si_uid);
1642 ret = put_user(0, &infop->si_status);
1649 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1650 int, options, struct rusage __user *, ru)
1652 struct wait_opts wo;
1653 struct pid *pid = NULL;
1657 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1658 __WNOTHREAD|__WCLONE|__WALL))
1663 else if (upid < 0) {
1664 type = PIDTYPE_PGID;
1665 pid = find_get_pid(-upid);
1666 } else if (upid == 0) {
1667 type = PIDTYPE_PGID;
1668 pid = get_task_pid(current, PIDTYPE_PGID);
1669 } else /* upid > 0 */ {
1671 pid = find_get_pid(upid);
1676 wo.wo_flags = options | WEXITED;
1678 wo.wo_stat = stat_addr;
1686 #ifdef __ARCH_WANT_SYS_WAITPID
1689 * sys_waitpid() remains for compatibility. waitpid() should be
1690 * implemented by calling sys_wait4() from libc.a.
1692 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1694 return sys_wait4(pid, stat_addr, options, NULL);