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>
56 #include <linux/kcov.h>
57 #include <linux/cpufreq_times.h>
59 #include "sched/tune.h"
61 #include <asm/uaccess.h>
62 #include <asm/unistd.h>
63 #include <asm/pgtable.h>
64 #include <asm/mmu_context.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);
125 * Accumulate here the counters for all threads as they die. We could
126 * skip the group leader because it is the last user of signal_struct,
127 * but we want to avoid the race with thread_group_cputime() which can
128 * see the empty ->thread_head list.
130 task_cputime(tsk, &utime, &stime);
131 write_seqlock(&sig->stats_lock);
134 sig->gtime += task_gtime(tsk);
135 sig->min_flt += tsk->min_flt;
136 sig->maj_flt += tsk->maj_flt;
137 sig->nvcsw += tsk->nvcsw;
138 sig->nivcsw += tsk->nivcsw;
139 sig->inblock += task_io_get_inblock(tsk);
140 sig->oublock += task_io_get_oublock(tsk);
141 task_io_accounting_add(&sig->ioac, &tsk->ioac);
142 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
144 __unhash_process(tsk, group_dead);
145 write_sequnlock(&sig->stats_lock);
148 * Do this under ->siglock, we can race with another thread
149 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
151 flush_sigqueue(&tsk->pending);
153 spin_unlock(&sighand->siglock);
155 __cleanup_sighand(sighand);
156 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
158 flush_sigqueue(&sig->shared_pending);
163 static void delayed_put_task_struct(struct rcu_head *rhp)
165 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
167 perf_event_delayed_put(tsk);
168 trace_sched_process_free(tsk);
169 put_task_struct(tsk);
173 void release_task(struct task_struct *p)
175 struct task_struct *leader;
177 #ifdef CONFIG_CPU_FREQ_TIMES
178 cpufreq_task_times_exit(p);
181 /* don't need to get the RCU readlock here - the process is dead and
182 * can't be modifying its own credentials. But shut RCU-lockdep up */
184 atomic_dec(&__task_cred(p)->user->processes);
189 write_lock_irq(&tasklist_lock);
190 ptrace_release_task(p);
194 * If we are the last non-leader member of the thread
195 * group, and the leader is zombie, then notify the
196 * group leader's parent process. (if it wants notification.)
199 leader = p->group_leader;
200 if (leader != p && thread_group_empty(leader)
201 && leader->exit_state == EXIT_ZOMBIE) {
203 * If we were the last child thread and the leader has
204 * exited already, and the leader's parent ignores SIGCHLD,
205 * then we are the one who should release the leader.
207 zap_leader = do_notify_parent(leader, leader->exit_signal);
209 leader->exit_state = EXIT_DEAD;
212 write_unlock_irq(&tasklist_lock);
214 call_rcu(&p->rcu, delayed_put_task_struct);
217 if (unlikely(zap_leader))
222 * Determine if a process group is "orphaned", according to the POSIX
223 * definition in 2.2.2.52. Orphaned process groups are not to be affected
224 * by terminal-generated stop signals. Newly orphaned process groups are
225 * to receive a SIGHUP and a SIGCONT.
227 * "I ask you, have you ever known what it is to be an orphan?"
229 static int will_become_orphaned_pgrp(struct pid *pgrp,
230 struct task_struct *ignored_task)
232 struct task_struct *p;
234 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
235 if ((p == ignored_task) ||
236 (p->exit_state && thread_group_empty(p)) ||
237 is_global_init(p->real_parent))
240 if (task_pgrp(p->real_parent) != pgrp &&
241 task_session(p->real_parent) == task_session(p))
243 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
248 int is_current_pgrp_orphaned(void)
252 read_lock(&tasklist_lock);
253 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
254 read_unlock(&tasklist_lock);
259 static bool has_stopped_jobs(struct pid *pgrp)
261 struct task_struct *p;
263 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
264 if (p->signal->flags & SIGNAL_STOP_STOPPED)
266 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
272 * Check to see if any process groups have become orphaned as
273 * a result of our exiting, and if they have any stopped jobs,
274 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
277 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
279 struct pid *pgrp = task_pgrp(tsk);
280 struct task_struct *ignored_task = tsk;
283 /* exit: our father is in a different pgrp than
284 * we are and we were the only connection outside.
286 parent = tsk->real_parent;
288 /* reparent: our child is in a different pgrp than
289 * we are, and it was the only connection outside.
293 if (task_pgrp(parent) != pgrp &&
294 task_session(parent) == task_session(tsk) &&
295 will_become_orphaned_pgrp(pgrp, ignored_task) &&
296 has_stopped_jobs(pgrp)) {
297 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
298 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
304 * A task is exiting. If it owned this mm, find a new owner for the mm.
306 void mm_update_next_owner(struct mm_struct *mm)
308 struct task_struct *c, *g, *p = current;
312 * If the exiting or execing task is not the owner, it's
313 * someone else's problem.
318 * The current owner is exiting/execing and there are no other
319 * candidates. Do not leave the mm pointing to a possibly
320 * freed task structure.
322 if (atomic_read(&mm->mm_users) <= 1) {
327 read_lock(&tasklist_lock);
329 * Search in the children
331 list_for_each_entry(c, &p->children, sibling) {
333 goto assign_new_owner;
337 * Search in the siblings
339 list_for_each_entry(c, &p->real_parent->children, sibling) {
341 goto assign_new_owner;
345 * Search through everything else, we should not get here often.
347 for_each_process(g) {
348 if (g->flags & PF_KTHREAD)
350 for_each_thread(g, c) {
352 goto assign_new_owner;
357 read_unlock(&tasklist_lock);
359 * We found no owner yet mm_users > 1: this implies that we are
360 * most likely racing with swapoff (try_to_unuse()) or /proc or
361 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
370 * The task_lock protects c->mm from changing.
371 * We always want mm->owner->mm == mm
375 * Delay read_unlock() till we have the task_lock()
376 * to ensure that c does not slip away underneath us
378 read_unlock(&tasklist_lock);
388 #endif /* CONFIG_MEMCG */
391 * Turn us into a lazy TLB process if we
394 static void exit_mm(struct task_struct *tsk)
396 struct mm_struct *mm = tsk->mm;
397 struct core_state *core_state;
404 * Serialize with any possible pending coredump.
405 * We must hold mmap_sem around checking core_state
406 * and clearing tsk->mm. The core-inducing thread
407 * will increment ->nr_threads for each thread in the
408 * group with ->mm != NULL.
410 down_read(&mm->mmap_sem);
411 core_state = mm->core_state;
413 struct core_thread self;
415 up_read(&mm->mmap_sem);
418 self.next = xchg(&core_state->dumper.next, &self);
420 * Implies mb(), the result of xchg() must be visible
421 * to core_state->dumper.
423 if (atomic_dec_and_test(&core_state->nr_threads))
424 complete(&core_state->startup);
427 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
428 if (!self.task) /* see coredump_finish() */
430 freezable_schedule();
432 __set_task_state(tsk, TASK_RUNNING);
433 down_read(&mm->mmap_sem);
435 atomic_inc(&mm->mm_count);
436 BUG_ON(mm != tsk->active_mm);
437 /* more a memory barrier than a real lock */
440 up_read(&mm->mmap_sem);
441 enter_lazy_tlb(mm, current);
443 mm_update_next_owner(mm);
445 if (test_thread_flag(TIF_MEMDIE))
449 static struct task_struct *find_alive_thread(struct task_struct *p)
451 struct task_struct *t;
453 for_each_thread(p, t) {
454 if (!(t->flags & PF_EXITING))
460 static struct task_struct *find_child_reaper(struct task_struct *father)
461 __releases(&tasklist_lock)
462 __acquires(&tasklist_lock)
464 struct pid_namespace *pid_ns = task_active_pid_ns(father);
465 struct task_struct *reaper = pid_ns->child_reaper;
467 if (likely(reaper != father))
470 reaper = find_alive_thread(father);
472 pid_ns->child_reaper = reaper;
476 write_unlock_irq(&tasklist_lock);
477 if (unlikely(pid_ns == &init_pid_ns)) {
478 panic("Attempted to kill init! exitcode=0x%08x\n",
479 father->signal->group_exit_code ?: father->exit_code);
481 zap_pid_ns_processes(pid_ns);
482 write_lock_irq(&tasklist_lock);
488 * When we die, we re-parent all our children, and try to:
489 * 1. give them to another thread in our thread group, if such a member exists
490 * 2. give it to the first ancestor process which prctl'd itself as a
491 * child_subreaper for its children (like a service manager)
492 * 3. give it to the init process (PID 1) in our pid namespace
494 static struct task_struct *find_new_reaper(struct task_struct *father,
495 struct task_struct *child_reaper)
497 struct task_struct *thread, *reaper;
499 thread = find_alive_thread(father);
503 if (father->signal->has_child_subreaper) {
505 * Find the first ->is_child_subreaper ancestor in our pid_ns.
506 * We start from father to ensure we can not look into another
507 * namespace, this is safe because all its threads are dead.
509 for (reaper = father;
510 !same_thread_group(reaper, child_reaper);
511 reaper = reaper->real_parent) {
512 /* call_usermodehelper() descendants need this check */
513 if (reaper == &init_task)
515 if (!reaper->signal->is_child_subreaper)
517 thread = find_alive_thread(reaper);
527 * Any that need to be release_task'd are put on the @dead list.
529 static void reparent_leader(struct task_struct *father, struct task_struct *p,
530 struct list_head *dead)
532 if (unlikely(p->exit_state == EXIT_DEAD))
535 /* We don't want people slaying init. */
536 p->exit_signal = SIGCHLD;
538 /* If it has exited notify the new parent about this child's death. */
540 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
541 if (do_notify_parent(p, p->exit_signal)) {
542 p->exit_state = EXIT_DEAD;
543 list_add(&p->ptrace_entry, dead);
547 kill_orphaned_pgrp(p, father);
551 * This does two things:
553 * A. Make init inherit all the child processes
554 * B. Check to see if any process groups have become orphaned
555 * as a result of our exiting, and if they have any stopped
556 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
558 static void forget_original_parent(struct task_struct *father,
559 struct list_head *dead)
561 struct task_struct *p, *t, *reaper;
563 if (unlikely(!list_empty(&father->ptraced)))
564 exit_ptrace(father, dead);
566 /* Can drop and reacquire tasklist_lock */
567 reaper = find_child_reaper(father);
568 if (list_empty(&father->children))
571 reaper = find_new_reaper(father, reaper);
572 list_for_each_entry(p, &father->children, sibling) {
573 for_each_thread(p, t) {
574 t->real_parent = reaper;
575 BUG_ON((!t->ptrace) != (t->parent == father));
576 if (likely(!t->ptrace))
577 t->parent = t->real_parent;
578 if (t->pdeath_signal)
579 group_send_sig_info(t->pdeath_signal,
583 * If this is a threaded reparent there is no need to
584 * notify anyone anything has happened.
586 if (!same_thread_group(reaper, father))
587 reparent_leader(father, p, dead);
589 list_splice_tail_init(&father->children, &reaper->children);
593 * Send signals to all our closest relatives so that they know
594 * to properly mourn us..
596 static void exit_notify(struct task_struct *tsk, int group_dead)
599 struct task_struct *p, *n;
602 write_lock_irq(&tasklist_lock);
603 forget_original_parent(tsk, &dead);
606 kill_orphaned_pgrp(tsk->group_leader, NULL);
608 if (unlikely(tsk->ptrace)) {
609 int sig = thread_group_leader(tsk) &&
610 thread_group_empty(tsk) &&
611 !ptrace_reparented(tsk) ?
612 tsk->exit_signal : SIGCHLD;
613 autoreap = do_notify_parent(tsk, sig);
614 } else if (thread_group_leader(tsk)) {
615 autoreap = thread_group_empty(tsk) &&
616 do_notify_parent(tsk, tsk->exit_signal);
621 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
622 if (tsk->exit_state == EXIT_DEAD)
623 list_add(&tsk->ptrace_entry, &dead);
625 /* mt-exec, de_thread() is waiting for group leader */
626 if (unlikely(tsk->signal->notify_count < 0))
627 wake_up_process(tsk->signal->group_exit_task);
628 write_unlock_irq(&tasklist_lock);
630 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
631 list_del_init(&p->ptrace_entry);
636 #ifdef CONFIG_DEBUG_STACK_USAGE
637 static void check_stack_usage(void)
639 static DEFINE_SPINLOCK(low_water_lock);
640 static int lowest_to_date = THREAD_SIZE;
643 free = stack_not_used(current);
645 if (free >= lowest_to_date)
648 spin_lock(&low_water_lock);
649 if (free < lowest_to_date) {
650 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
651 current->comm, task_pid_nr(current), free);
652 lowest_to_date = free;
654 spin_unlock(&low_water_lock);
657 static inline void check_stack_usage(void) {}
660 void do_exit(long code)
662 struct task_struct *tsk = current;
664 TASKS_RCU(int tasks_rcu_i);
666 profile_task_exit(tsk);
669 WARN_ON(blk_needs_flush_plug(tsk));
671 if (unlikely(in_interrupt()))
672 panic("Aiee, killing interrupt handler!");
673 if (unlikely(!tsk->pid))
674 panic("Attempted to kill the idle task!");
677 * If do_exit is called because this processes oopsed, it's possible
678 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
679 * continuing. Amongst other possible reasons, this is to prevent
680 * mm_release()->clear_child_tid() from writing to a user-controlled
685 ptrace_event(PTRACE_EVENT_EXIT, code);
687 validate_creds_for_do_exit(tsk);
690 * We're taking recursive faults here in do_exit. Safest is to just
691 * leave this task alone and wait for reboot.
693 if (unlikely(tsk->flags & PF_EXITING)) {
694 pr_alert("Fixing recursive fault but reboot is needed!\n");
696 * We can do this unlocked here. The futex code uses
697 * this flag just to verify whether the pi state
698 * cleanup has been done or not. In the worst case it
699 * loops once more. We pretend that the cleanup was
700 * done as there is no way to return. Either the
701 * OWNER_DIED bit is set by now or we push the blocked
702 * task into the wait for ever nirwana as well.
704 tsk->flags |= PF_EXITPIDONE;
705 set_current_state(TASK_UNINTERRUPTIBLE);
709 exit_signals(tsk); /* sets PF_EXITING */
711 schedtune_exit_task(tsk);
714 * tsk->flags are checked in the futex code to protect against
715 * an exiting task cleaning up the robust pi futexes.
718 raw_spin_unlock_wait(&tsk->pi_lock);
720 if (unlikely(in_atomic())) {
721 pr_info("note: %s[%d] exited with preempt_count %d\n",
722 current->comm, task_pid_nr(current),
724 preempt_count_set(PREEMPT_ENABLED);
727 /* sync mm's RSS info before statistics gathering */
729 sync_mm_rss(tsk->mm);
730 acct_update_integrals(tsk);
731 group_dead = atomic_dec_and_test(&tsk->signal->live);
733 hrtimer_cancel(&tsk->signal->real_timer);
734 exit_itimers(tsk->signal);
736 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
738 acct_collect(code, group_dead);
743 tsk->exit_code = code;
744 taskstats_exit(tsk, group_dead);
750 trace_sched_process_exit(tsk);
757 disassociate_ctty(1);
758 exit_task_namespaces(tsk);
763 * Flush inherited counters to the parent - before the parent
764 * gets woken up by child-exit notifications.
766 * because of cgroup mode, must be called before cgroup_exit()
768 perf_event_exit_task(tsk);
773 * FIXME: do that only when needed, using sched_exit tracepoint
775 flush_ptrace_hw_breakpoint(tsk);
777 TASKS_RCU(preempt_disable());
778 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
779 TASKS_RCU(preempt_enable());
780 exit_notify(tsk, group_dead);
781 proc_exit_connector(tsk);
784 mpol_put(tsk->mempolicy);
785 tsk->mempolicy = NULL;
789 if (unlikely(current->pi_state_cache))
790 kfree(current->pi_state_cache);
793 * Make sure we are holding no locks:
795 debug_check_no_locks_held();
797 * We can do this unlocked here. The futex code uses this flag
798 * just to verify whether the pi state cleanup has been done
799 * or not. In the worst case it loops once more.
801 tsk->flags |= PF_EXITPIDONE;
804 exit_io_context(tsk);
806 if (tsk->splice_pipe)
807 free_pipe_info(tsk->splice_pipe);
809 if (tsk->task_frag.page)
810 put_page(tsk->task_frag.page);
812 validate_creds_for_do_exit(tsk);
817 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
819 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
822 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
823 * when the following two conditions become true.
824 * - There is race condition of mmap_sem (It is acquired by
826 * - SMI occurs before setting TASK_RUNINNG.
827 * (or hypervisor of virtual machine switches to other guest)
828 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
830 * To avoid it, we have to wait for releasing tsk->pi_lock which
831 * is held by try_to_wake_up()
834 raw_spin_unlock_wait(&tsk->pi_lock);
836 /* causes final put_task_struct in finish_task_switch(). */
837 tsk->state = TASK_DEAD;
838 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
841 /* Avoid "noreturn function does return". */
843 cpu_relax(); /* For when BUG is null */
845 EXPORT_SYMBOL_GPL(do_exit);
847 void complete_and_exit(struct completion *comp, long code)
854 EXPORT_SYMBOL(complete_and_exit);
856 SYSCALL_DEFINE1(exit, int, error_code)
858 do_exit((error_code&0xff)<<8);
862 * Take down every thread in the group. This is called by fatal signals
863 * as well as by sys_exit_group (below).
866 do_group_exit(int exit_code)
868 struct signal_struct *sig = current->signal;
870 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
872 if (signal_group_exit(sig))
873 exit_code = sig->group_exit_code;
874 else if (!thread_group_empty(current)) {
875 struct sighand_struct *const sighand = current->sighand;
877 spin_lock_irq(&sighand->siglock);
878 if (signal_group_exit(sig))
879 /* Another thread got here before we took the lock. */
880 exit_code = sig->group_exit_code;
882 sig->group_exit_code = exit_code;
883 sig->flags = SIGNAL_GROUP_EXIT;
884 zap_other_threads(current);
886 spin_unlock_irq(&sighand->siglock);
894 * this kills every thread in the thread group. Note that any externally
895 * wait4()-ing process will get the correct exit code - even if this
896 * thread is not the thread group leader.
898 SYSCALL_DEFINE1(exit_group, int, error_code)
900 do_group_exit((error_code & 0xff) << 8);
906 enum pid_type wo_type;
910 struct siginfo __user *wo_info;
912 struct rusage __user *wo_rusage;
914 wait_queue_t child_wait;
919 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
921 if (type != PIDTYPE_PID)
922 task = task->group_leader;
923 return task->pids[type].pid;
926 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
928 return wo->wo_type == PIDTYPE_MAX ||
929 task_pid_type(p, wo->wo_type) == wo->wo_pid;
933 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
935 if (!eligible_pid(wo, p))
939 * Wait for all children (clone and not) if __WALL is set or
940 * if it is traced by us.
942 if (ptrace || (wo->wo_flags & __WALL))
946 * Otherwise, wait for clone children *only* if __WCLONE is set;
947 * otherwise, wait for non-clone children *only*.
949 * Note: a "clone" child here is one that reports to its parent
950 * using a signal other than SIGCHLD, or a non-leader thread which
951 * we can only see if it is traced by us.
953 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
959 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
960 pid_t pid, uid_t uid, int why, int status)
962 struct siginfo __user *infop;
963 int retval = wo->wo_rusage
964 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
970 retval = put_user(SIGCHLD, &infop->si_signo);
972 retval = put_user(0, &infop->si_errno);
974 retval = put_user((short)why, &infop->si_code);
976 retval = put_user(pid, &infop->si_pid);
978 retval = put_user(uid, &infop->si_uid);
980 retval = put_user(status, &infop->si_status);
988 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
989 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
990 * the lock and this task is uninteresting. If we return nonzero, we have
991 * released the lock and the system call should return.
993 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
995 int state, retval, status;
996 pid_t pid = task_pid_vnr(p);
997 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
998 struct siginfo __user *infop;
1000 if (!likely(wo->wo_flags & WEXITED))
1003 if (unlikely(wo->wo_flags & WNOWAIT)) {
1004 int exit_code = p->exit_code;
1008 read_unlock(&tasklist_lock);
1009 sched_annotate_sleep();
1011 if ((exit_code & 0x7f) == 0) {
1013 status = exit_code >> 8;
1015 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1016 status = exit_code & 0x7f;
1018 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1021 * Move the task's state to DEAD/TRACE, only one thread can do this.
1023 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1024 EXIT_TRACE : EXIT_DEAD;
1025 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1028 * We own this thread, nobody else can reap it.
1030 read_unlock(&tasklist_lock);
1031 sched_annotate_sleep();
1034 * Check thread_group_leader() to exclude the traced sub-threads.
1036 if (state == EXIT_DEAD && thread_group_leader(p)) {
1037 struct signal_struct *sig = p->signal;
1038 struct signal_struct *psig = current->signal;
1039 unsigned long maxrss;
1040 cputime_t tgutime, tgstime;
1043 * The resource counters for the group leader are in its
1044 * own task_struct. Those for dead threads in the group
1045 * are in its signal_struct, as are those for the child
1046 * processes it has previously reaped. All these
1047 * accumulate in the parent's signal_struct c* fields.
1049 * We don't bother to take a lock here to protect these
1050 * p->signal fields because the whole thread group is dead
1051 * and nobody can change them.
1053 * psig->stats_lock also protects us from our sub-theads
1054 * which can reap other children at the same time. Until
1055 * we change k_getrusage()-like users to rely on this lock
1056 * we have to take ->siglock as well.
1058 * We use thread_group_cputime_adjusted() to get times for
1059 * the thread group, which consolidates times for all threads
1060 * in the group including the group leader.
1062 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1063 spin_lock_irq(¤t->sighand->siglock);
1064 write_seqlock(&psig->stats_lock);
1065 psig->cutime += tgutime + sig->cutime;
1066 psig->cstime += tgstime + sig->cstime;
1067 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1069 p->min_flt + sig->min_flt + sig->cmin_flt;
1071 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1073 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1075 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1077 task_io_get_inblock(p) +
1078 sig->inblock + sig->cinblock;
1080 task_io_get_oublock(p) +
1081 sig->oublock + sig->coublock;
1082 maxrss = max(sig->maxrss, sig->cmaxrss);
1083 if (psig->cmaxrss < maxrss)
1084 psig->cmaxrss = maxrss;
1085 task_io_accounting_add(&psig->ioac, &p->ioac);
1086 task_io_accounting_add(&psig->ioac, &sig->ioac);
1087 write_sequnlock(&psig->stats_lock);
1088 spin_unlock_irq(¤t->sighand->siglock);
1091 retval = wo->wo_rusage
1092 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1093 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1094 ? p->signal->group_exit_code : p->exit_code;
1095 if (!retval && wo->wo_stat)
1096 retval = put_user(status, wo->wo_stat);
1098 infop = wo->wo_info;
1099 if (!retval && infop)
1100 retval = put_user(SIGCHLD, &infop->si_signo);
1101 if (!retval && infop)
1102 retval = put_user(0, &infop->si_errno);
1103 if (!retval && infop) {
1106 if ((status & 0x7f) == 0) {
1110 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1113 retval = put_user((short)why, &infop->si_code);
1115 retval = put_user(status, &infop->si_status);
1117 if (!retval && infop)
1118 retval = put_user(pid, &infop->si_pid);
1119 if (!retval && infop)
1120 retval = put_user(uid, &infop->si_uid);
1124 if (state == EXIT_TRACE) {
1125 write_lock_irq(&tasklist_lock);
1126 /* We dropped tasklist, ptracer could die and untrace */
1129 /* If parent wants a zombie, don't release it now */
1130 state = EXIT_ZOMBIE;
1131 if (do_notify_parent(p, p->exit_signal))
1133 p->exit_state = state;
1134 write_unlock_irq(&tasklist_lock);
1136 if (state == EXIT_DEAD)
1142 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1145 if (task_is_stopped_or_traced(p) &&
1146 !(p->jobctl & JOBCTL_LISTENING))
1147 return &p->exit_code;
1149 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1150 return &p->signal->group_exit_code;
1156 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1158 * @ptrace: is the wait for ptrace
1159 * @p: task to wait for
1161 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1164 * read_lock(&tasklist_lock), which is released if return value is
1165 * non-zero. Also, grabs and releases @p->sighand->siglock.
1168 * 0 if wait condition didn't exist and search for other wait conditions
1169 * should continue. Non-zero return, -errno on failure and @p's pid on
1170 * success, implies that tasklist_lock is released and wait condition
1171 * search should terminate.
1173 static int wait_task_stopped(struct wait_opts *wo,
1174 int ptrace, struct task_struct *p)
1176 struct siginfo __user *infop;
1177 int retval, exit_code, *p_code, why;
1178 uid_t uid = 0; /* unneeded, required by compiler */
1182 * Traditionally we see ptrace'd stopped tasks regardless of options.
1184 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1187 if (!task_stopped_code(p, ptrace))
1191 spin_lock_irq(&p->sighand->siglock);
1193 p_code = task_stopped_code(p, ptrace);
1194 if (unlikely(!p_code))
1197 exit_code = *p_code;
1201 if (!unlikely(wo->wo_flags & WNOWAIT))
1204 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1206 spin_unlock_irq(&p->sighand->siglock);
1211 * Now we are pretty sure this task is interesting.
1212 * Make sure it doesn't get reaped out from under us while we
1213 * give up the lock and then examine it below. We don't want to
1214 * keep holding onto the tasklist_lock while we call getrusage and
1215 * possibly take page faults for user memory.
1218 pid = task_pid_vnr(p);
1219 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1220 read_unlock(&tasklist_lock);
1221 sched_annotate_sleep();
1223 if (unlikely(wo->wo_flags & WNOWAIT))
1224 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1226 retval = wo->wo_rusage
1227 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1228 if (!retval && wo->wo_stat)
1229 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1231 infop = wo->wo_info;
1232 if (!retval && infop)
1233 retval = put_user(SIGCHLD, &infop->si_signo);
1234 if (!retval && infop)
1235 retval = put_user(0, &infop->si_errno);
1236 if (!retval && infop)
1237 retval = put_user((short)why, &infop->si_code);
1238 if (!retval && infop)
1239 retval = put_user(exit_code, &infop->si_status);
1240 if (!retval && infop)
1241 retval = put_user(pid, &infop->si_pid);
1242 if (!retval && infop)
1243 retval = put_user(uid, &infop->si_uid);
1253 * Handle do_wait work for one task in a live, non-stopped state.
1254 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1255 * the lock and this task is uninteresting. If we return nonzero, we have
1256 * released the lock and the system call should return.
1258 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1264 if (!unlikely(wo->wo_flags & WCONTINUED))
1267 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1270 spin_lock_irq(&p->sighand->siglock);
1271 /* Re-check with the lock held. */
1272 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1273 spin_unlock_irq(&p->sighand->siglock);
1276 if (!unlikely(wo->wo_flags & WNOWAIT))
1277 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1278 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1279 spin_unlock_irq(&p->sighand->siglock);
1281 pid = task_pid_vnr(p);
1283 read_unlock(&tasklist_lock);
1284 sched_annotate_sleep();
1287 retval = wo->wo_rusage
1288 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1290 if (!retval && wo->wo_stat)
1291 retval = put_user(0xffff, wo->wo_stat);
1295 retval = wait_noreap_copyout(wo, p, pid, uid,
1296 CLD_CONTINUED, SIGCONT);
1297 BUG_ON(retval == 0);
1304 * Consider @p for a wait by @parent.
1306 * -ECHILD should be in ->notask_error before the first call.
1307 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1308 * Returns zero if the search for a child should continue;
1309 * then ->notask_error is 0 if @p is an eligible child,
1310 * or another error from security_task_wait(), or still -ECHILD.
1312 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1313 struct task_struct *p)
1316 * We can race with wait_task_zombie() from another thread.
1317 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1318 * can't confuse the checks below.
1320 int exit_state = ACCESS_ONCE(p->exit_state);
1323 if (unlikely(exit_state == EXIT_DEAD))
1326 ret = eligible_child(wo, ptrace, p);
1330 ret = security_task_wait(p);
1331 if (unlikely(ret < 0)) {
1333 * If we have not yet seen any eligible child,
1334 * then let this error code replace -ECHILD.
1335 * A permission error will give the user a clue
1336 * to look for security policy problems, rather
1337 * than for mysterious wait bugs.
1339 if (wo->notask_error)
1340 wo->notask_error = ret;
1344 if (unlikely(exit_state == EXIT_TRACE)) {
1346 * ptrace == 0 means we are the natural parent. In this case
1347 * we should clear notask_error, debugger will notify us.
1349 if (likely(!ptrace))
1350 wo->notask_error = 0;
1354 if (likely(!ptrace) && unlikely(p->ptrace)) {
1356 * If it is traced by its real parent's group, just pretend
1357 * the caller is ptrace_do_wait() and reap this child if it
1360 * This also hides group stop state from real parent; otherwise
1361 * a single stop can be reported twice as group and ptrace stop.
1362 * If a ptracer wants to distinguish these two events for its
1363 * own children it should create a separate process which takes
1364 * the role of real parent.
1366 if (!ptrace_reparented(p))
1371 if (exit_state == EXIT_ZOMBIE) {
1372 /* we don't reap group leaders with subthreads */
1373 if (!delay_group_leader(p)) {
1375 * A zombie ptracee is only visible to its ptracer.
1376 * Notification and reaping will be cascaded to the
1377 * real parent when the ptracer detaches.
1379 if (unlikely(ptrace) || likely(!p->ptrace))
1380 return wait_task_zombie(wo, p);
1384 * Allow access to stopped/continued state via zombie by
1385 * falling through. Clearing of notask_error is complex.
1389 * If WEXITED is set, notask_error should naturally be
1390 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1391 * so, if there are live subthreads, there are events to
1392 * wait for. If all subthreads are dead, it's still safe
1393 * to clear - this function will be called again in finite
1394 * amount time once all the subthreads are released and
1395 * will then return without clearing.
1399 * Stopped state is per-task and thus can't change once the
1400 * target task dies. Only continued and exited can happen.
1401 * Clear notask_error if WCONTINUED | WEXITED.
1403 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1404 wo->notask_error = 0;
1407 * @p is alive and it's gonna stop, continue or exit, so
1408 * there always is something to wait for.
1410 wo->notask_error = 0;
1414 * Wait for stopped. Depending on @ptrace, different stopped state
1415 * is used and the two don't interact with each other.
1417 ret = wait_task_stopped(wo, ptrace, p);
1422 * Wait for continued. There's only one continued state and the
1423 * ptracer can consume it which can confuse the real parent. Don't
1424 * use WCONTINUED from ptracer. You don't need or want it.
1426 return wait_task_continued(wo, p);
1430 * Do the work of do_wait() for one thread in the group, @tsk.
1432 * -ECHILD should be in ->notask_error before the first call.
1433 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1434 * Returns zero if the search for a child should continue; then
1435 * ->notask_error is 0 if there were any eligible children,
1436 * or another error from security_task_wait(), or still -ECHILD.
1438 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1440 struct task_struct *p;
1442 list_for_each_entry(p, &tsk->children, sibling) {
1443 int ret = wait_consider_task(wo, 0, p);
1452 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1454 struct task_struct *p;
1456 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1457 int ret = wait_consider_task(wo, 1, p);
1466 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1467 int sync, void *key)
1469 struct wait_opts *wo = container_of(wait, struct wait_opts,
1471 struct task_struct *p = key;
1473 if (!eligible_pid(wo, p))
1476 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1479 return default_wake_function(wait, mode, sync, key);
1482 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1484 __wake_up_sync_key(&parent->signal->wait_chldexit,
1485 TASK_INTERRUPTIBLE, 1, p);
1488 static long do_wait(struct wait_opts *wo)
1490 struct task_struct *tsk;
1493 trace_sched_process_wait(wo->wo_pid);
1495 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1496 wo->child_wait.private = current;
1497 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1500 * If there is nothing that can match our criteria, just get out.
1501 * We will clear ->notask_error to zero if we see any child that
1502 * might later match our criteria, even if we are not able to reap
1505 wo->notask_error = -ECHILD;
1506 if ((wo->wo_type < PIDTYPE_MAX) &&
1507 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1510 set_current_state(TASK_INTERRUPTIBLE);
1511 read_lock(&tasklist_lock);
1514 retval = do_wait_thread(wo, tsk);
1518 retval = ptrace_do_wait(wo, tsk);
1522 if (wo->wo_flags & __WNOTHREAD)
1524 } while_each_thread(current, tsk);
1525 read_unlock(&tasklist_lock);
1528 retval = wo->notask_error;
1529 if (!retval && !(wo->wo_flags & WNOHANG)) {
1530 retval = -ERESTARTSYS;
1531 if (!signal_pending(current)) {
1537 __set_current_state(TASK_RUNNING);
1538 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1542 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1543 infop, int, options, struct rusage __user *, ru)
1545 struct wait_opts wo;
1546 struct pid *pid = NULL;
1550 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1552 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1565 type = PIDTYPE_PGID;
1573 if (type < PIDTYPE_MAX)
1574 pid = find_get_pid(upid);
1578 wo.wo_flags = options;
1588 * For a WNOHANG return, clear out all the fields
1589 * we would set so the user can easily tell the
1593 ret = put_user(0, &infop->si_signo);
1595 ret = put_user(0, &infop->si_errno);
1597 ret = put_user(0, &infop->si_code);
1599 ret = put_user(0, &infop->si_pid);
1601 ret = put_user(0, &infop->si_uid);
1603 ret = put_user(0, &infop->si_status);
1610 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1611 int, options, struct rusage __user *, ru)
1613 struct wait_opts wo;
1614 struct pid *pid = NULL;
1618 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1619 __WNOTHREAD|__WCLONE|__WALL))
1624 else if (upid < 0) {
1625 type = PIDTYPE_PGID;
1626 pid = find_get_pid(-upid);
1627 } else if (upid == 0) {
1628 type = PIDTYPE_PGID;
1629 pid = get_task_pid(current, PIDTYPE_PGID);
1630 } else /* upid > 0 */ {
1632 pid = find_get_pid(upid);
1637 wo.wo_flags = options | WEXITED;
1639 wo.wo_stat = stat_addr;
1647 #ifdef __ARCH_WANT_SYS_WAITPID
1650 * sys_waitpid() remains for compatibility. waitpid() should be
1651 * implemented by calling sys_wait4() from libc.a.
1653 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1655 return sys_wait4(pid, stat_addr, options, NULL);