1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/blkdev.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/tracehook.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/pgtable.h>
70 #include <asm/mmu_context.h>
72 static void __unhash_process(struct task_struct *p, bool group_dead)
75 detach_pid(p, PIDTYPE_PID);
77 detach_pid(p, PIDTYPE_TGID);
78 detach_pid(p, PIDTYPE_PGID);
79 detach_pid(p, PIDTYPE_SID);
81 list_del_rcu(&p->tasks);
82 list_del_init(&p->sibling);
83 __this_cpu_dec(process_counts);
85 list_del_rcu(&p->thread_group);
86 list_del_rcu(&p->thread_node);
90 * This function expects the tasklist_lock write-locked.
92 static void __exit_signal(struct task_struct *tsk)
94 struct signal_struct *sig = tsk->signal;
95 bool group_dead = thread_group_leader(tsk);
96 struct sighand_struct *sighand;
97 struct tty_struct *uninitialized_var(tty);
100 sighand = rcu_dereference_check(tsk->sighand,
101 lockdep_tasklist_lock_is_held());
102 spin_lock(&sighand->siglock);
104 #ifdef CONFIG_POSIX_TIMERS
105 posix_cpu_timers_exit(tsk);
107 posix_cpu_timers_exit_group(tsk);
110 * This can only happen if the caller is de_thread().
111 * FIXME: this is the temporary hack, we should teach
112 * posix-cpu-timers to handle this case correctly.
114 if (unlikely(has_group_leader_pid(tsk)))
115 posix_cpu_timers_exit_group(tsk);
124 * If there is any task waiting for the group exit
127 if (sig->notify_count > 0 && !--sig->notify_count)
128 wake_up_process(sig->group_exit_task);
130 if (tsk == sig->curr_target)
131 sig->curr_target = next_thread(tsk);
134 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
135 sizeof(unsigned long long));
138 * Accumulate here the counters for all threads as they die. We could
139 * skip the group leader because it is the last user of signal_struct,
140 * but we want to avoid the race with thread_group_cputime() which can
141 * see the empty ->thread_head list.
143 task_cputime(tsk, &utime, &stime);
144 write_seqlock(&sig->stats_lock);
147 sig->gtime += task_gtime(tsk);
148 sig->min_flt += tsk->min_flt;
149 sig->maj_flt += tsk->maj_flt;
150 sig->nvcsw += tsk->nvcsw;
151 sig->nivcsw += tsk->nivcsw;
152 sig->inblock += task_io_get_inblock(tsk);
153 sig->oublock += task_io_get_oublock(tsk);
154 task_io_accounting_add(&sig->ioac, &tsk->ioac);
155 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
157 __unhash_process(tsk, group_dead);
158 write_sequnlock(&sig->stats_lock);
161 * Do this under ->siglock, we can race with another thread
162 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
164 flush_sigqueue(&tsk->pending);
166 spin_unlock(&sighand->siglock);
168 __cleanup_sighand(sighand);
169 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
171 flush_sigqueue(&sig->shared_pending);
176 static void delayed_put_task_struct(struct rcu_head *rhp)
178 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
180 perf_event_delayed_put(tsk);
181 trace_sched_process_free(tsk);
182 put_task_struct(tsk);
185 void put_task_struct_rcu_user(struct task_struct *task)
187 if (refcount_dec_and_test(&task->rcu_users))
188 call_rcu(&task->rcu, delayed_put_task_struct);
191 void release_task(struct task_struct *p)
193 struct task_struct *leader;
196 /* don't need to get the RCU readlock here - the process is dead and
197 * can't be modifying its own credentials. But shut RCU-lockdep up */
199 atomic_dec(&__task_cred(p)->user->processes);
205 write_lock_irq(&tasklist_lock);
206 ptrace_release_task(p);
210 * If we are the last non-leader member of the thread
211 * group, and the leader is zombie, then notify the
212 * group leader's parent process. (if it wants notification.)
215 leader = p->group_leader;
216 if (leader != p && thread_group_empty(leader)
217 && leader->exit_state == EXIT_ZOMBIE) {
219 * If we were the last child thread and the leader has
220 * exited already, and the leader's parent ignores SIGCHLD,
221 * then we are the one who should release the leader.
223 zap_leader = do_notify_parent(leader, leader->exit_signal);
225 leader->exit_state = EXIT_DEAD;
228 write_unlock_irq(&tasklist_lock);
230 put_task_struct_rcu_user(p);
233 if (unlikely(zap_leader))
237 void rcuwait_wake_up(struct rcuwait *w)
239 struct task_struct *task;
244 * Order condition vs @task, such that everything prior to the load
245 * of @task is visible. This is the condition as to why the user called
246 * rcuwait_trywake() in the first place. Pairs with set_current_state()
247 * barrier (A) in rcuwait_wait_event().
250 * [S] tsk = current [S] cond = true
256 task = rcu_dereference(w->task);
258 wake_up_process(task);
263 * Determine if a process group is "orphaned", according to the POSIX
264 * definition in 2.2.2.52. Orphaned process groups are not to be affected
265 * by terminal-generated stop signals. Newly orphaned process groups are
266 * to receive a SIGHUP and a SIGCONT.
268 * "I ask you, have you ever known what it is to be an orphan?"
270 static int will_become_orphaned_pgrp(struct pid *pgrp,
271 struct task_struct *ignored_task)
273 struct task_struct *p;
275 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
276 if ((p == ignored_task) ||
277 (p->exit_state && thread_group_empty(p)) ||
278 is_global_init(p->real_parent))
281 if (task_pgrp(p->real_parent) != pgrp &&
282 task_session(p->real_parent) == task_session(p))
284 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
289 int is_current_pgrp_orphaned(void)
293 read_lock(&tasklist_lock);
294 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
295 read_unlock(&tasklist_lock);
300 static bool has_stopped_jobs(struct pid *pgrp)
302 struct task_struct *p;
304 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
305 if (p->signal->flags & SIGNAL_STOP_STOPPED)
307 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
313 * Check to see if any process groups have become orphaned as
314 * a result of our exiting, and if they have any stopped jobs,
315 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
318 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
320 struct pid *pgrp = task_pgrp(tsk);
321 struct task_struct *ignored_task = tsk;
324 /* exit: our father is in a different pgrp than
325 * we are and we were the only connection outside.
327 parent = tsk->real_parent;
329 /* reparent: our child is in a different pgrp than
330 * we are, and it was the only connection outside.
334 if (task_pgrp(parent) != pgrp &&
335 task_session(parent) == task_session(tsk) &&
336 will_become_orphaned_pgrp(pgrp, ignored_task) &&
337 has_stopped_jobs(pgrp)) {
338 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
339 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
345 * A task is exiting. If it owned this mm, find a new owner for the mm.
347 void mm_update_next_owner(struct mm_struct *mm)
349 struct task_struct *c, *g, *p = current;
353 * If the exiting or execing task is not the owner, it's
354 * someone else's problem.
359 * The current owner is exiting/execing and there are no other
360 * candidates. Do not leave the mm pointing to a possibly
361 * freed task structure.
363 if (atomic_read(&mm->mm_users) <= 1) {
364 WRITE_ONCE(mm->owner, NULL);
368 read_lock(&tasklist_lock);
370 * Search in the children
372 list_for_each_entry(c, &p->children, sibling) {
374 goto assign_new_owner;
378 * Search in the siblings
380 list_for_each_entry(c, &p->real_parent->children, sibling) {
382 goto assign_new_owner;
386 * Search through everything else, we should not get here often.
388 for_each_process(g) {
389 if (g->flags & PF_KTHREAD)
391 for_each_thread(g, c) {
393 goto assign_new_owner;
398 read_unlock(&tasklist_lock);
400 * We found no owner yet mm_users > 1: this implies that we are
401 * most likely racing with swapoff (try_to_unuse()) or /proc or
402 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
404 WRITE_ONCE(mm->owner, NULL);
411 * The task_lock protects c->mm from changing.
412 * We always want mm->owner->mm == mm
416 * Delay read_unlock() till we have the task_lock()
417 * to ensure that c does not slip away underneath us
419 read_unlock(&tasklist_lock);
425 WRITE_ONCE(mm->owner, c);
429 #endif /* CONFIG_MEMCG */
432 * Turn us into a lazy TLB process if we
435 static void exit_mm(void)
437 struct mm_struct *mm = current->mm;
438 struct core_state *core_state;
440 exit_mm_release(current, mm);
445 * Serialize with any possible pending coredump.
446 * We must hold mmap_sem around checking core_state
447 * and clearing tsk->mm. The core-inducing thread
448 * will increment ->nr_threads for each thread in the
449 * group with ->mm != NULL.
451 down_read(&mm->mmap_sem);
452 core_state = mm->core_state;
454 struct core_thread self;
456 up_read(&mm->mmap_sem);
459 if (self.task->flags & PF_SIGNALED)
460 self.next = xchg(&core_state->dumper.next, &self);
464 * Implies mb(), the result of xchg() must be visible
465 * to core_state->dumper.
467 if (atomic_dec_and_test(&core_state->nr_threads))
468 complete(&core_state->startup);
471 set_current_state(TASK_UNINTERRUPTIBLE);
472 if (!self.task) /* see coredump_finish() */
474 freezable_schedule();
476 __set_current_state(TASK_RUNNING);
477 down_read(&mm->mmap_sem);
480 BUG_ON(mm != current->active_mm);
481 /* more a memory barrier than a real lock */
484 up_read(&mm->mmap_sem);
485 enter_lazy_tlb(mm, current);
486 task_unlock(current);
487 mm_update_next_owner(mm);
489 if (test_thread_flag(TIF_MEMDIE))
493 static struct task_struct *find_alive_thread(struct task_struct *p)
495 struct task_struct *t;
497 for_each_thread(p, t) {
498 if (!(t->flags & PF_EXITING))
504 static struct task_struct *find_child_reaper(struct task_struct *father,
505 struct list_head *dead)
506 __releases(&tasklist_lock)
507 __acquires(&tasklist_lock)
509 struct pid_namespace *pid_ns = task_active_pid_ns(father);
510 struct task_struct *reaper = pid_ns->child_reaper;
511 struct task_struct *p, *n;
513 if (likely(reaper != father))
516 reaper = find_alive_thread(father);
518 pid_ns->child_reaper = reaper;
522 write_unlock_irq(&tasklist_lock);
524 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
525 list_del_init(&p->ptrace_entry);
529 zap_pid_ns_processes(pid_ns);
530 write_lock_irq(&tasklist_lock);
536 * When we die, we re-parent all our children, and try to:
537 * 1. give them to another thread in our thread group, if such a member exists
538 * 2. give it to the first ancestor process which prctl'd itself as a
539 * child_subreaper for its children (like a service manager)
540 * 3. give it to the init process (PID 1) in our pid namespace
542 static struct task_struct *find_new_reaper(struct task_struct *father,
543 struct task_struct *child_reaper)
545 struct task_struct *thread, *reaper;
547 thread = find_alive_thread(father);
551 if (father->signal->has_child_subreaper) {
552 unsigned int ns_level = task_pid(father)->level;
554 * Find the first ->is_child_subreaper ancestor in our pid_ns.
555 * We can't check reaper != child_reaper to ensure we do not
556 * cross the namespaces, the exiting parent could be injected
557 * by setns() + fork().
558 * We check pid->level, this is slightly more efficient than
559 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
561 for (reaper = father->real_parent;
562 task_pid(reaper)->level == ns_level;
563 reaper = reaper->real_parent) {
564 if (reaper == &init_task)
566 if (!reaper->signal->is_child_subreaper)
568 thread = find_alive_thread(reaper);
578 * Any that need to be release_task'd are put on the @dead list.
580 static void reparent_leader(struct task_struct *father, struct task_struct *p,
581 struct list_head *dead)
583 if (unlikely(p->exit_state == EXIT_DEAD))
586 /* We don't want people slaying init. */
587 p->exit_signal = SIGCHLD;
589 /* If it has exited notify the new parent about this child's death. */
591 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
592 if (do_notify_parent(p, p->exit_signal)) {
593 p->exit_state = EXIT_DEAD;
594 list_add(&p->ptrace_entry, dead);
598 kill_orphaned_pgrp(p, father);
602 * This does two things:
604 * A. Make init inherit all the child processes
605 * B. Check to see if any process groups have become orphaned
606 * as a result of our exiting, and if they have any stopped
607 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
609 static void forget_original_parent(struct task_struct *father,
610 struct list_head *dead)
612 struct task_struct *p, *t, *reaper;
614 if (unlikely(!list_empty(&father->ptraced)))
615 exit_ptrace(father, dead);
617 /* Can drop and reacquire tasklist_lock */
618 reaper = find_child_reaper(father, dead);
619 if (list_empty(&father->children))
622 reaper = find_new_reaper(father, reaper);
623 list_for_each_entry(p, &father->children, sibling) {
624 for_each_thread(p, t) {
625 t->real_parent = reaper;
626 BUG_ON((!t->ptrace) != (t->parent == father));
627 if (likely(!t->ptrace))
628 t->parent = t->real_parent;
629 if (t->pdeath_signal)
630 group_send_sig_info(t->pdeath_signal,
635 * If this is a threaded reparent there is no need to
636 * notify anyone anything has happened.
638 if (!same_thread_group(reaper, father))
639 reparent_leader(father, p, dead);
641 list_splice_tail_init(&father->children, &reaper->children);
645 * Send signals to all our closest relatives so that they know
646 * to properly mourn us..
648 static void exit_notify(struct task_struct *tsk, int group_dead)
651 struct task_struct *p, *n;
654 write_lock_irq(&tasklist_lock);
655 forget_original_parent(tsk, &dead);
658 kill_orphaned_pgrp(tsk->group_leader, NULL);
660 tsk->exit_state = EXIT_ZOMBIE;
661 if (unlikely(tsk->ptrace)) {
662 int sig = thread_group_leader(tsk) &&
663 thread_group_empty(tsk) &&
664 !ptrace_reparented(tsk) ?
665 tsk->exit_signal : SIGCHLD;
666 autoreap = do_notify_parent(tsk, sig);
667 } else if (thread_group_leader(tsk)) {
668 autoreap = thread_group_empty(tsk) &&
669 do_notify_parent(tsk, tsk->exit_signal);
675 tsk->exit_state = EXIT_DEAD;
676 list_add(&tsk->ptrace_entry, &dead);
679 /* mt-exec, de_thread() is waiting for group leader */
680 if (unlikely(tsk->signal->notify_count < 0))
681 wake_up_process(tsk->signal->group_exit_task);
682 write_unlock_irq(&tasklist_lock);
684 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
685 list_del_init(&p->ptrace_entry);
690 #ifdef CONFIG_DEBUG_STACK_USAGE
691 static void check_stack_usage(void)
693 static DEFINE_SPINLOCK(low_water_lock);
694 static int lowest_to_date = THREAD_SIZE;
697 free = stack_not_used(current);
699 if (free >= lowest_to_date)
702 spin_lock(&low_water_lock);
703 if (free < lowest_to_date) {
704 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
705 current->comm, task_pid_nr(current), free);
706 lowest_to_date = free;
708 spin_unlock(&low_water_lock);
711 static inline void check_stack_usage(void) {}
714 void __noreturn do_exit(long code)
716 struct task_struct *tsk = current;
720 * We can get here from a kernel oops, sometimes with preemption off.
721 * Start by checking for critical errors.
722 * Then fix up important state like USER_DS and preemption.
723 * Then do everything else.
726 WARN_ON(blk_needs_flush_plug(tsk));
728 if (unlikely(in_interrupt()))
729 panic("Aiee, killing interrupt handler!");
730 if (unlikely(!tsk->pid))
731 panic("Attempted to kill the idle task!");
734 * If do_exit is called because this processes oopsed, it's possible
735 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
736 * continuing. Amongst other possible reasons, this is to prevent
737 * mm_release()->clear_child_tid() from writing to a user-controlled
742 if (unlikely(in_atomic())) {
743 pr_info("note: %s[%d] exited with preempt_count %d\n",
744 current->comm, task_pid_nr(current),
746 preempt_count_set(PREEMPT_ENABLED);
749 profile_task_exit(tsk);
752 ptrace_event(PTRACE_EVENT_EXIT, code);
754 validate_creds_for_do_exit(tsk);
757 * We're taking recursive faults here in do_exit. Safest is to just
758 * leave this task alone and wait for reboot.
760 if (unlikely(tsk->flags & PF_EXITING)) {
761 pr_alert("Fixing recursive fault but reboot is needed!\n");
762 futex_exit_recursive(tsk);
763 set_current_state(TASK_UNINTERRUPTIBLE);
767 exit_signals(tsk); /* sets PF_EXITING */
769 /* sync mm's RSS info before statistics gathering */
771 sync_mm_rss(tsk->mm);
772 acct_update_integrals(tsk);
773 group_dead = atomic_dec_and_test(&tsk->signal->live);
776 * If the last thread of global init has exited, panic
777 * immediately to get a useable coredump.
779 if (unlikely(is_global_init(tsk)))
780 panic("Attempted to kill init! exitcode=0x%08x\n",
781 tsk->signal->group_exit_code ?: (int)code);
783 #ifdef CONFIG_POSIX_TIMERS
784 hrtimer_cancel(&tsk->signal->real_timer);
785 exit_itimers(tsk->signal);
788 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
790 acct_collect(code, group_dead);
795 tsk->exit_code = code;
796 taskstats_exit(tsk, group_dead);
802 trace_sched_process_exit(tsk);
809 disassociate_ctty(1);
810 exit_task_namespaces(tsk);
816 * Flush inherited counters to the parent - before the parent
817 * gets woken up by child-exit notifications.
819 * because of cgroup mode, must be called before cgroup_exit()
821 perf_event_exit_task(tsk);
823 sched_autogroup_exit_task(tsk);
827 * FIXME: do that only when needed, using sched_exit tracepoint
829 flush_ptrace_hw_breakpoint(tsk);
831 exit_tasks_rcu_start();
832 exit_notify(tsk, group_dead);
833 proc_exit_connector(tsk);
834 mpol_put_task_policy(tsk);
836 if (unlikely(current->pi_state_cache))
837 kfree(current->pi_state_cache);
840 * Make sure we are holding no locks:
842 debug_check_no_locks_held();
845 exit_io_context(tsk);
847 if (tsk->splice_pipe)
848 free_pipe_info(tsk->splice_pipe);
850 if (tsk->task_frag.page)
851 put_page(tsk->task_frag.page);
853 validate_creds_for_do_exit(tsk);
858 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
860 exit_tasks_rcu_finish();
862 lockdep_free_task(tsk);
865 EXPORT_SYMBOL_GPL(do_exit);
867 void complete_and_exit(struct completion *comp, long code)
874 EXPORT_SYMBOL(complete_and_exit);
876 SYSCALL_DEFINE1(exit, int, error_code)
878 do_exit((error_code&0xff)<<8);
882 * Take down every thread in the group. This is called by fatal signals
883 * as well as by sys_exit_group (below).
886 do_group_exit(int exit_code)
888 struct signal_struct *sig = current->signal;
890 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
892 if (signal_group_exit(sig))
893 exit_code = sig->group_exit_code;
894 else if (!thread_group_empty(current)) {
895 struct sighand_struct *const sighand = current->sighand;
897 spin_lock_irq(&sighand->siglock);
898 if (signal_group_exit(sig))
899 /* Another thread got here before we took the lock. */
900 exit_code = sig->group_exit_code;
902 sig->group_exit_code = exit_code;
903 sig->flags = SIGNAL_GROUP_EXIT;
904 zap_other_threads(current);
906 spin_unlock_irq(&sighand->siglock);
914 * this kills every thread in the thread group. Note that any externally
915 * wait4()-ing process will get the correct exit code - even if this
916 * thread is not the thread group leader.
918 SYSCALL_DEFINE1(exit_group, int, error_code)
920 do_group_exit((error_code & 0xff) << 8);
933 enum pid_type wo_type;
937 struct waitid_info *wo_info;
939 struct rusage *wo_rusage;
941 wait_queue_entry_t child_wait;
945 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
947 return wo->wo_type == PIDTYPE_MAX ||
948 task_pid_type(p, wo->wo_type) == wo->wo_pid;
952 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
954 if (!eligible_pid(wo, p))
958 * Wait for all children (clone and not) if __WALL is set or
959 * if it is traced by us.
961 if (ptrace || (wo->wo_flags & __WALL))
965 * Otherwise, wait for clone children *only* if __WCLONE is set;
966 * otherwise, wait for non-clone children *only*.
968 * Note: a "clone" child here is one that reports to its parent
969 * using a signal other than SIGCHLD, or a non-leader thread which
970 * we can only see if it is traced by us.
972 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
979 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
980 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
981 * the lock and this task is uninteresting. If we return nonzero, we have
982 * released the lock and the system call should return.
984 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
987 pid_t pid = task_pid_vnr(p);
988 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
989 struct waitid_info *infop;
991 if (!likely(wo->wo_flags & WEXITED))
994 if (unlikely(wo->wo_flags & WNOWAIT)) {
995 status = p->exit_code;
997 read_unlock(&tasklist_lock);
998 sched_annotate_sleep();
1000 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1005 * Move the task's state to DEAD/TRACE, only one thread can do this.
1007 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1008 EXIT_TRACE : EXIT_DEAD;
1009 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1012 * We own this thread, nobody else can reap it.
1014 read_unlock(&tasklist_lock);
1015 sched_annotate_sleep();
1018 * Check thread_group_leader() to exclude the traced sub-threads.
1020 if (state == EXIT_DEAD && thread_group_leader(p)) {
1021 struct signal_struct *sig = p->signal;
1022 struct signal_struct *psig = current->signal;
1023 unsigned long maxrss;
1024 u64 tgutime, tgstime;
1027 * The resource counters for the group leader are in its
1028 * own task_struct. Those for dead threads in the group
1029 * are in its signal_struct, as are those for the child
1030 * processes it has previously reaped. All these
1031 * accumulate in the parent's signal_struct c* fields.
1033 * We don't bother to take a lock here to protect these
1034 * p->signal fields because the whole thread group is dead
1035 * and nobody can change them.
1037 * psig->stats_lock also protects us from our sub-theads
1038 * which can reap other children at the same time. Until
1039 * we change k_getrusage()-like users to rely on this lock
1040 * we have to take ->siglock as well.
1042 * We use thread_group_cputime_adjusted() to get times for
1043 * the thread group, which consolidates times for all threads
1044 * in the group including the group leader.
1046 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1047 spin_lock_irq(¤t->sighand->siglock);
1048 write_seqlock(&psig->stats_lock);
1049 psig->cutime += tgutime + sig->cutime;
1050 psig->cstime += tgstime + sig->cstime;
1051 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1053 p->min_flt + sig->min_flt + sig->cmin_flt;
1055 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1057 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1059 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1061 task_io_get_inblock(p) +
1062 sig->inblock + sig->cinblock;
1064 task_io_get_oublock(p) +
1065 sig->oublock + sig->coublock;
1066 maxrss = max(sig->maxrss, sig->cmaxrss);
1067 if (psig->cmaxrss < maxrss)
1068 psig->cmaxrss = maxrss;
1069 task_io_accounting_add(&psig->ioac, &p->ioac);
1070 task_io_accounting_add(&psig->ioac, &sig->ioac);
1071 write_sequnlock(&psig->stats_lock);
1072 spin_unlock_irq(¤t->sighand->siglock);
1076 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1077 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1078 ? p->signal->group_exit_code : p->exit_code;
1079 wo->wo_stat = status;
1081 if (state == EXIT_TRACE) {
1082 write_lock_irq(&tasklist_lock);
1083 /* We dropped tasklist, ptracer could die and untrace */
1086 /* If parent wants a zombie, don't release it now */
1087 state = EXIT_ZOMBIE;
1088 if (do_notify_parent(p, p->exit_signal))
1090 p->exit_state = state;
1091 write_unlock_irq(&tasklist_lock);
1093 if (state == EXIT_DEAD)
1097 infop = wo->wo_info;
1099 if ((status & 0x7f) == 0) {
1100 infop->cause = CLD_EXITED;
1101 infop->status = status >> 8;
1103 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1104 infop->status = status & 0x7f;
1113 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1116 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1117 return &p->exit_code;
1119 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1120 return &p->signal->group_exit_code;
1126 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1128 * @ptrace: is the wait for ptrace
1129 * @p: task to wait for
1131 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1134 * read_lock(&tasklist_lock), which is released if return value is
1135 * non-zero. Also, grabs and releases @p->sighand->siglock.
1138 * 0 if wait condition didn't exist and search for other wait conditions
1139 * should continue. Non-zero return, -errno on failure and @p's pid on
1140 * success, implies that tasklist_lock is released and wait condition
1141 * search should terminate.
1143 static int wait_task_stopped(struct wait_opts *wo,
1144 int ptrace, struct task_struct *p)
1146 struct waitid_info *infop;
1147 int exit_code, *p_code, why;
1148 uid_t uid = 0; /* unneeded, required by compiler */
1152 * Traditionally we see ptrace'd stopped tasks regardless of options.
1154 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1157 if (!task_stopped_code(p, ptrace))
1161 spin_lock_irq(&p->sighand->siglock);
1163 p_code = task_stopped_code(p, ptrace);
1164 if (unlikely(!p_code))
1167 exit_code = *p_code;
1171 if (!unlikely(wo->wo_flags & WNOWAIT))
1174 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1176 spin_unlock_irq(&p->sighand->siglock);
1181 * Now we are pretty sure this task is interesting.
1182 * Make sure it doesn't get reaped out from under us while we
1183 * give up the lock and then examine it below. We don't want to
1184 * keep holding onto the tasklist_lock while we call getrusage and
1185 * possibly take page faults for user memory.
1188 pid = task_pid_vnr(p);
1189 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1190 read_unlock(&tasklist_lock);
1191 sched_annotate_sleep();
1193 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1196 if (likely(!(wo->wo_flags & WNOWAIT)))
1197 wo->wo_stat = (exit_code << 8) | 0x7f;
1199 infop = wo->wo_info;
1202 infop->status = exit_code;
1210 * Handle do_wait work for one task in a live, non-stopped state.
1211 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1212 * the lock and this task is uninteresting. If we return nonzero, we have
1213 * released the lock and the system call should return.
1215 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1217 struct waitid_info *infop;
1221 if (!unlikely(wo->wo_flags & WCONTINUED))
1224 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1227 spin_lock_irq(&p->sighand->siglock);
1228 /* Re-check with the lock held. */
1229 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1230 spin_unlock_irq(&p->sighand->siglock);
1233 if (!unlikely(wo->wo_flags & WNOWAIT))
1234 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1235 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1236 spin_unlock_irq(&p->sighand->siglock);
1238 pid = task_pid_vnr(p);
1240 read_unlock(&tasklist_lock);
1241 sched_annotate_sleep();
1243 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1246 infop = wo->wo_info;
1248 wo->wo_stat = 0xffff;
1250 infop->cause = CLD_CONTINUED;
1253 infop->status = SIGCONT;
1259 * Consider @p for a wait by @parent.
1261 * -ECHILD should be in ->notask_error before the first call.
1262 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1263 * Returns zero if the search for a child should continue;
1264 * then ->notask_error is 0 if @p is an eligible child,
1267 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1268 struct task_struct *p)
1271 * We can race with wait_task_zombie() from another thread.
1272 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1273 * can't confuse the checks below.
1275 int exit_state = READ_ONCE(p->exit_state);
1278 if (unlikely(exit_state == EXIT_DEAD))
1281 ret = eligible_child(wo, ptrace, p);
1285 if (unlikely(exit_state == EXIT_TRACE)) {
1287 * ptrace == 0 means we are the natural parent. In this case
1288 * we should clear notask_error, debugger will notify us.
1290 if (likely(!ptrace))
1291 wo->notask_error = 0;
1295 if (likely(!ptrace) && unlikely(p->ptrace)) {
1297 * If it is traced by its real parent's group, just pretend
1298 * the caller is ptrace_do_wait() and reap this child if it
1301 * This also hides group stop state from real parent; otherwise
1302 * a single stop can be reported twice as group and ptrace stop.
1303 * If a ptracer wants to distinguish these two events for its
1304 * own children it should create a separate process which takes
1305 * the role of real parent.
1307 if (!ptrace_reparented(p))
1312 if (exit_state == EXIT_ZOMBIE) {
1313 /* we don't reap group leaders with subthreads */
1314 if (!delay_group_leader(p)) {
1316 * A zombie ptracee is only visible to its ptracer.
1317 * Notification and reaping will be cascaded to the
1318 * real parent when the ptracer detaches.
1320 if (unlikely(ptrace) || likely(!p->ptrace))
1321 return wait_task_zombie(wo, p);
1325 * Allow access to stopped/continued state via zombie by
1326 * falling through. Clearing of notask_error is complex.
1330 * If WEXITED is set, notask_error should naturally be
1331 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1332 * so, if there are live subthreads, there are events to
1333 * wait for. If all subthreads are dead, it's still safe
1334 * to clear - this function will be called again in finite
1335 * amount time once all the subthreads are released and
1336 * will then return without clearing.
1340 * Stopped state is per-task and thus can't change once the
1341 * target task dies. Only continued and exited can happen.
1342 * Clear notask_error if WCONTINUED | WEXITED.
1344 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1345 wo->notask_error = 0;
1348 * @p is alive and it's gonna stop, continue or exit, so
1349 * there always is something to wait for.
1351 wo->notask_error = 0;
1355 * Wait for stopped. Depending on @ptrace, different stopped state
1356 * is used and the two don't interact with each other.
1358 ret = wait_task_stopped(wo, ptrace, p);
1363 * Wait for continued. There's only one continued state and the
1364 * ptracer can consume it which can confuse the real parent. Don't
1365 * use WCONTINUED from ptracer. You don't need or want it.
1367 return wait_task_continued(wo, p);
1371 * Do the work of do_wait() for one thread in the group, @tsk.
1373 * -ECHILD should be in ->notask_error before the first call.
1374 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1375 * Returns zero if the search for a child should continue; then
1376 * ->notask_error is 0 if there were any eligible children,
1379 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1381 struct task_struct *p;
1383 list_for_each_entry(p, &tsk->children, sibling) {
1384 int ret = wait_consider_task(wo, 0, p);
1393 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1395 struct task_struct *p;
1397 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1398 int ret = wait_consider_task(wo, 1, p);
1407 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1408 int sync, void *key)
1410 struct wait_opts *wo = container_of(wait, struct wait_opts,
1412 struct task_struct *p = key;
1414 if (!eligible_pid(wo, p))
1417 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1420 return default_wake_function(wait, mode, sync, key);
1423 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1425 __wake_up_sync_key(&parent->signal->wait_chldexit,
1426 TASK_INTERRUPTIBLE, 1, p);
1429 static long do_wait(struct wait_opts *wo)
1431 struct task_struct *tsk;
1434 trace_sched_process_wait(wo->wo_pid);
1436 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1437 wo->child_wait.private = current;
1438 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1441 * If there is nothing that can match our criteria, just get out.
1442 * We will clear ->notask_error to zero if we see any child that
1443 * might later match our criteria, even if we are not able to reap
1446 wo->notask_error = -ECHILD;
1447 if ((wo->wo_type < PIDTYPE_MAX) &&
1448 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1451 set_current_state(TASK_INTERRUPTIBLE);
1452 read_lock(&tasklist_lock);
1455 retval = do_wait_thread(wo, tsk);
1459 retval = ptrace_do_wait(wo, tsk);
1463 if (wo->wo_flags & __WNOTHREAD)
1465 } while_each_thread(current, tsk);
1466 read_unlock(&tasklist_lock);
1469 retval = wo->notask_error;
1470 if (!retval && !(wo->wo_flags & WNOHANG)) {
1471 retval = -ERESTARTSYS;
1472 if (!signal_pending(current)) {
1478 __set_current_state(TASK_RUNNING);
1479 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1483 static struct pid *pidfd_get_pid(unsigned int fd)
1490 return ERR_PTR(-EBADF);
1492 pid = pidfd_pid(f.file);
1500 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1501 int options, struct rusage *ru)
1503 struct wait_opts wo;
1504 struct pid *pid = NULL;
1508 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1509 __WNOTHREAD|__WCLONE|__WALL))
1511 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1523 pid = find_get_pid(upid);
1526 type = PIDTYPE_PGID;
1531 pid = find_get_pid(upid);
1533 pid = get_task_pid(current, PIDTYPE_PGID);
1540 pid = pidfd_get_pid(upid);
1542 return PTR_ERR(pid);
1550 wo.wo_flags = options;
1559 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1560 infop, int, options, struct rusage __user *, ru)
1563 struct waitid_info info = {.status = 0};
1564 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1570 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1576 if (!user_access_begin(infop, sizeof(*infop)))
1579 unsafe_put_user(signo, &infop->si_signo, Efault);
1580 unsafe_put_user(0, &infop->si_errno, Efault);
1581 unsafe_put_user(info.cause, &infop->si_code, Efault);
1582 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1583 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1584 unsafe_put_user(info.status, &infop->si_status, Efault);
1592 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1595 struct wait_opts wo;
1596 struct pid *pid = NULL;
1600 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1601 __WNOTHREAD|__WCLONE|__WALL))
1604 /* -INT_MIN is not defined */
1605 if (upid == INT_MIN)
1610 else if (upid < 0) {
1611 type = PIDTYPE_PGID;
1612 pid = find_get_pid(-upid);
1613 } else if (upid == 0) {
1614 type = PIDTYPE_PGID;
1615 pid = get_task_pid(current, PIDTYPE_PGID);
1616 } else /* upid > 0 */ {
1618 pid = find_get_pid(upid);
1623 wo.wo_flags = options | WEXITED;
1629 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1635 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1636 int, options, struct rusage __user *, ru)
1639 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1642 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1648 #ifdef __ARCH_WANT_SYS_WAITPID
1651 * sys_waitpid() remains for compatibility. waitpid() should be
1652 * implemented by calling sys_wait4() from libc.a.
1654 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1656 return kernel_wait4(pid, stat_addr, options, NULL);
1661 #ifdef CONFIG_COMPAT
1662 COMPAT_SYSCALL_DEFINE4(wait4,
1664 compat_uint_t __user *, stat_addr,
1666 struct compat_rusage __user *, ru)
1669 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1671 if (ru && put_compat_rusage(&r, ru))
1677 COMPAT_SYSCALL_DEFINE5(waitid,
1678 int, which, compat_pid_t, pid,
1679 struct compat_siginfo __user *, infop, int, options,
1680 struct compat_rusage __user *, uru)
1683 struct waitid_info info = {.status = 0};
1684 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1690 /* kernel_waitid() overwrites everything in ru */
1691 if (COMPAT_USE_64BIT_TIME)
1692 err = copy_to_user(uru, &ru, sizeof(ru));
1694 err = put_compat_rusage(&ru, uru);
1703 if (!user_access_begin(infop, sizeof(*infop)))
1706 unsafe_put_user(signo, &infop->si_signo, Efault);
1707 unsafe_put_user(0, &infop->si_errno, Efault);
1708 unsafe_put_user(info.cause, &infop->si_code, Efault);
1709 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1710 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1711 unsafe_put_user(info.status, &infop->si_status, Efault);
1720 __weak void abort(void)
1724 /* if that doesn't kill us, halt */
1725 panic("Oops failed to kill thread");
1727 EXPORT_SYMBOL(abort);