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/mmu_context.h>
71 static void __unhash_process(struct task_struct *p, bool group_dead)
74 detach_pid(p, PIDTYPE_PID);
76 detach_pid(p, PIDTYPE_TGID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
89 * This function expects the tasklist_lock write-locked.
91 static void __exit_signal(struct task_struct *tsk)
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *tty;
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
103 #ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
106 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);
124 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
125 sizeof(unsigned long long));
128 * Accumulate here the counters for all threads as they die. We could
129 * skip the group leader because it is the last user of signal_struct,
130 * but we want to avoid the race with thread_group_cputime() which can
131 * see the empty ->thread_head list.
133 task_cputime(tsk, &utime, &stime);
134 write_seqlock(&sig->stats_lock);
137 sig->gtime += task_gtime(tsk);
138 sig->min_flt += tsk->min_flt;
139 sig->maj_flt += tsk->maj_flt;
140 sig->nvcsw += tsk->nvcsw;
141 sig->nivcsw += tsk->nivcsw;
142 sig->inblock += task_io_get_inblock(tsk);
143 sig->oublock += task_io_get_oublock(tsk);
144 task_io_accounting_add(&sig->ioac, &tsk->ioac);
145 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
147 __unhash_process(tsk, group_dead);
148 write_sequnlock(&sig->stats_lock);
151 * Do this under ->siglock, we can race with another thread
152 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
154 flush_sigqueue(&tsk->pending);
156 spin_unlock(&sighand->siglock);
158 __cleanup_sighand(sighand);
159 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
161 flush_sigqueue(&sig->shared_pending);
166 static void delayed_put_task_struct(struct rcu_head *rhp)
168 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
170 perf_event_delayed_put(tsk);
171 trace_sched_process_free(tsk);
172 put_task_struct(tsk);
175 void put_task_struct_rcu_user(struct task_struct *task)
177 if (refcount_dec_and_test(&task->rcu_users))
178 call_rcu(&task->rcu, delayed_put_task_struct);
181 void release_task(struct task_struct *p)
183 struct task_struct *leader;
184 struct pid *thread_pid;
187 /* don't need to get the RCU readlock here - the process is dead and
188 * can't be modifying its own credentials. But shut RCU-lockdep up */
190 atomic_dec(&__task_cred(p)->user->processes);
195 write_lock_irq(&tasklist_lock);
196 ptrace_release_task(p);
197 thread_pid = get_pid(p->thread_pid);
201 * If we are the last non-leader member of the thread
202 * group, and the leader is zombie, then notify the
203 * group leader's parent process. (if it wants notification.)
206 leader = p->group_leader;
207 if (leader != p && thread_group_empty(leader)
208 && leader->exit_state == EXIT_ZOMBIE) {
210 * If we were the last child thread and the leader has
211 * exited already, and the leader's parent ignores SIGCHLD,
212 * then we are the one who should release the leader.
214 zap_leader = do_notify_parent(leader, leader->exit_signal);
216 leader->exit_state = EXIT_DEAD;
219 write_unlock_irq(&tasklist_lock);
220 seccomp_filter_release(p);
221 proc_flush_pid(thread_pid);
224 put_task_struct_rcu_user(p);
227 if (unlikely(zap_leader))
231 int rcuwait_wake_up(struct rcuwait *w)
234 struct task_struct *task;
239 * Order condition vs @task, such that everything prior to the load
240 * of @task is visible. This is the condition as to why the user called
241 * rcuwait_wake() in the first place. Pairs with set_current_state()
242 * barrier (A) in rcuwait_wait_event().
245 * [S] tsk = current [S] cond = true
251 task = rcu_dereference(w->task);
253 ret = wake_up_process(task);
258 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
261 * Determine if a process group is "orphaned", according to the POSIX
262 * definition in 2.2.2.52. Orphaned process groups are not to be affected
263 * by terminal-generated stop signals. Newly orphaned process groups are
264 * to receive a SIGHUP and a SIGCONT.
266 * "I ask you, have you ever known what it is to be an orphan?"
268 static int will_become_orphaned_pgrp(struct pid *pgrp,
269 struct task_struct *ignored_task)
271 struct task_struct *p;
273 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274 if ((p == ignored_task) ||
275 (p->exit_state && thread_group_empty(p)) ||
276 is_global_init(p->real_parent))
279 if (task_pgrp(p->real_parent) != pgrp &&
280 task_session(p->real_parent) == task_session(p))
282 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
287 int is_current_pgrp_orphaned(void)
291 read_lock(&tasklist_lock);
292 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
293 read_unlock(&tasklist_lock);
298 static bool has_stopped_jobs(struct pid *pgrp)
300 struct task_struct *p;
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if (p->signal->flags & SIGNAL_STOP_STOPPED)
305 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
311 * Check to see if any process groups have become orphaned as
312 * a result of our exiting, and if they have any stopped jobs,
313 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
316 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
318 struct pid *pgrp = task_pgrp(tsk);
319 struct task_struct *ignored_task = tsk;
322 /* exit: our father is in a different pgrp than
323 * we are and we were the only connection outside.
325 parent = tsk->real_parent;
327 /* reparent: our child is in a different pgrp than
328 * we are, and it was the only connection outside.
332 if (task_pgrp(parent) != pgrp &&
333 task_session(parent) == task_session(tsk) &&
334 will_become_orphaned_pgrp(pgrp, ignored_task) &&
335 has_stopped_jobs(pgrp)) {
336 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
337 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
343 * A task is exiting. If it owned this mm, find a new owner for the mm.
345 void mm_update_next_owner(struct mm_struct *mm)
347 struct task_struct *c, *g, *p = current;
351 * If the exiting or execing task is not the owner, it's
352 * someone else's problem.
357 * The current owner is exiting/execing and there are no other
358 * candidates. Do not leave the mm pointing to a possibly
359 * freed task structure.
361 if (atomic_read(&mm->mm_users) <= 1) {
362 WRITE_ONCE(mm->owner, NULL);
366 read_lock(&tasklist_lock);
368 * Search in the children
370 list_for_each_entry(c, &p->children, sibling) {
372 goto assign_new_owner;
376 * Search in the siblings
378 list_for_each_entry(c, &p->real_parent->children, sibling) {
380 goto assign_new_owner;
384 * Search through everything else, we should not get here often.
386 for_each_process(g) {
387 if (g->flags & PF_KTHREAD)
389 for_each_thread(g, c) {
391 goto assign_new_owner;
396 read_unlock(&tasklist_lock);
398 * We found no owner yet mm_users > 1: this implies that we are
399 * most likely racing with swapoff (try_to_unuse()) or /proc or
400 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
402 WRITE_ONCE(mm->owner, NULL);
409 * The task_lock protects c->mm from changing.
410 * We always want mm->owner->mm == mm
414 * Delay read_unlock() till we have the task_lock()
415 * to ensure that c does not slip away underneath us
417 read_unlock(&tasklist_lock);
423 WRITE_ONCE(mm->owner, c);
427 #endif /* CONFIG_MEMCG */
430 * Turn us into a lazy TLB process if we
433 static void exit_mm(void)
435 struct mm_struct *mm = current->mm;
436 struct core_state *core_state;
438 exit_mm_release(current, mm);
443 * Serialize with any possible pending coredump.
444 * We must hold mmap_lock around checking core_state
445 * and clearing tsk->mm. The core-inducing thread
446 * will increment ->nr_threads for each thread in the
447 * group with ->mm != NULL.
450 core_state = mm->core_state;
452 struct core_thread self;
454 mmap_read_unlock(mm);
457 if (self.task->flags & PF_SIGNALED)
458 self.next = xchg(&core_state->dumper.next, &self);
462 * Implies mb(), the result of xchg() must be visible
463 * to core_state->dumper.
465 if (atomic_dec_and_test(&core_state->nr_threads))
466 complete(&core_state->startup);
469 set_current_state(TASK_UNINTERRUPTIBLE);
470 if (!self.task) /* see coredump_finish() */
472 freezable_schedule();
474 __set_current_state(TASK_RUNNING);
478 BUG_ON(mm != current->active_mm);
479 /* more a memory barrier than a real lock */
482 mmap_read_unlock(mm);
483 enter_lazy_tlb(mm, current);
484 task_unlock(current);
485 mm_update_next_owner(mm);
487 if (test_thread_flag(TIF_MEMDIE))
491 static struct task_struct *find_alive_thread(struct task_struct *p)
493 struct task_struct *t;
495 for_each_thread(p, t) {
496 if (!(t->flags & PF_EXITING))
502 static struct task_struct *find_child_reaper(struct task_struct *father,
503 struct list_head *dead)
504 __releases(&tasklist_lock)
505 __acquires(&tasklist_lock)
507 struct pid_namespace *pid_ns = task_active_pid_ns(father);
508 struct task_struct *reaper = pid_ns->child_reaper;
509 struct task_struct *p, *n;
511 if (likely(reaper != father))
514 reaper = find_alive_thread(father);
516 pid_ns->child_reaper = reaper;
520 write_unlock_irq(&tasklist_lock);
522 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
523 list_del_init(&p->ptrace_entry);
527 zap_pid_ns_processes(pid_ns);
528 write_lock_irq(&tasklist_lock);
534 * When we die, we re-parent all our children, and try to:
535 * 1. give them to another thread in our thread group, if such a member exists
536 * 2. give it to the first ancestor process which prctl'd itself as a
537 * child_subreaper for its children (like a service manager)
538 * 3. give it to the init process (PID 1) in our pid namespace
540 static struct task_struct *find_new_reaper(struct task_struct *father,
541 struct task_struct *child_reaper)
543 struct task_struct *thread, *reaper;
545 thread = find_alive_thread(father);
549 if (father->signal->has_child_subreaper) {
550 unsigned int ns_level = task_pid(father)->level;
552 * Find the first ->is_child_subreaper ancestor in our pid_ns.
553 * We can't check reaper != child_reaper to ensure we do not
554 * cross the namespaces, the exiting parent could be injected
555 * by setns() + fork().
556 * We check pid->level, this is slightly more efficient than
557 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
559 for (reaper = father->real_parent;
560 task_pid(reaper)->level == ns_level;
561 reaper = reaper->real_parent) {
562 if (reaper == &init_task)
564 if (!reaper->signal->is_child_subreaper)
566 thread = find_alive_thread(reaper);
576 * Any that need to be release_task'd are put on the @dead list.
578 static void reparent_leader(struct task_struct *father, struct task_struct *p,
579 struct list_head *dead)
581 if (unlikely(p->exit_state == EXIT_DEAD))
584 /* We don't want people slaying init. */
585 p->exit_signal = SIGCHLD;
587 /* If it has exited notify the new parent about this child's death. */
589 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
590 if (do_notify_parent(p, p->exit_signal)) {
591 p->exit_state = EXIT_DEAD;
592 list_add(&p->ptrace_entry, dead);
596 kill_orphaned_pgrp(p, father);
600 * This does two things:
602 * A. Make init inherit all the child processes
603 * B. Check to see if any process groups have become orphaned
604 * as a result of our exiting, and if they have any stopped
605 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
607 static void forget_original_parent(struct task_struct *father,
608 struct list_head *dead)
610 struct task_struct *p, *t, *reaper;
612 if (unlikely(!list_empty(&father->ptraced)))
613 exit_ptrace(father, dead);
615 /* Can drop and reacquire tasklist_lock */
616 reaper = find_child_reaper(father, dead);
617 if (list_empty(&father->children))
620 reaper = find_new_reaper(father, reaper);
621 list_for_each_entry(p, &father->children, sibling) {
622 for_each_thread(p, t) {
623 RCU_INIT_POINTER(t->real_parent, reaper);
624 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
625 if (likely(!t->ptrace))
626 t->parent = t->real_parent;
627 if (t->pdeath_signal)
628 group_send_sig_info(t->pdeath_signal,
633 * If this is a threaded reparent there is no need to
634 * notify anyone anything has happened.
636 if (!same_thread_group(reaper, father))
637 reparent_leader(father, p, dead);
639 list_splice_tail_init(&father->children, &reaper->children);
643 * Send signals to all our closest relatives so that they know
644 * to properly mourn us..
646 static void exit_notify(struct task_struct *tsk, int group_dead)
649 struct task_struct *p, *n;
652 write_lock_irq(&tasklist_lock);
653 forget_original_parent(tsk, &dead);
656 kill_orphaned_pgrp(tsk->group_leader, NULL);
658 tsk->exit_state = EXIT_ZOMBIE;
659 if (unlikely(tsk->ptrace)) {
660 int sig = thread_group_leader(tsk) &&
661 thread_group_empty(tsk) &&
662 !ptrace_reparented(tsk) ?
663 tsk->exit_signal : SIGCHLD;
664 autoreap = do_notify_parent(tsk, sig);
665 } else if (thread_group_leader(tsk)) {
666 autoreap = thread_group_empty(tsk) &&
667 do_notify_parent(tsk, tsk->exit_signal);
673 tsk->exit_state = EXIT_DEAD;
674 list_add(&tsk->ptrace_entry, &dead);
677 /* mt-exec, de_thread() is waiting for group leader */
678 if (unlikely(tsk->signal->notify_count < 0))
679 wake_up_process(tsk->signal->group_exit_task);
680 write_unlock_irq(&tasklist_lock);
682 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
683 list_del_init(&p->ptrace_entry);
688 #ifdef CONFIG_DEBUG_STACK_USAGE
689 static void check_stack_usage(void)
691 static DEFINE_SPINLOCK(low_water_lock);
692 static int lowest_to_date = THREAD_SIZE;
695 free = stack_not_used(current);
697 if (free >= lowest_to_date)
700 spin_lock(&low_water_lock);
701 if (free < lowest_to_date) {
702 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
703 current->comm, task_pid_nr(current), free);
704 lowest_to_date = free;
706 spin_unlock(&low_water_lock);
709 static inline void check_stack_usage(void) {}
712 void __noreturn do_exit(long code)
714 struct task_struct *tsk = current;
718 * We can get here from a kernel oops, sometimes with preemption off.
719 * Start by checking for critical errors.
720 * Then fix up important state like USER_DS and preemption.
721 * Then do everything else.
724 WARN_ON(blk_needs_flush_plug(tsk));
726 if (unlikely(in_interrupt()))
727 panic("Aiee, killing interrupt handler!");
728 if (unlikely(!tsk->pid))
729 panic("Attempted to kill the idle task!");
732 * If do_exit is called because this processes oopsed, it's possible
733 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
734 * continuing. Amongst other possible reasons, this is to prevent
735 * mm_release()->clear_child_tid() from writing to a user-controlled
738 force_uaccess_begin();
740 if (unlikely(in_atomic())) {
741 pr_info("note: %s[%d] exited with preempt_count %d\n",
742 current->comm, task_pid_nr(current),
744 preempt_count_set(PREEMPT_ENABLED);
747 profile_task_exit(tsk);
750 ptrace_event(PTRACE_EVENT_EXIT, code);
752 validate_creds_for_do_exit(tsk);
755 * We're taking recursive faults here in do_exit. Safest is to just
756 * leave this task alone and wait for reboot.
758 if (unlikely(tsk->flags & PF_EXITING)) {
759 pr_alert("Fixing recursive fault but reboot is needed!\n");
760 futex_exit_recursive(tsk);
761 set_current_state(TASK_UNINTERRUPTIBLE);
765 exit_signals(tsk); /* sets PF_EXITING */
767 /* sync mm's RSS info before statistics gathering */
769 sync_mm_rss(tsk->mm);
770 acct_update_integrals(tsk);
771 group_dead = atomic_dec_and_test(&tsk->signal->live);
774 * If the last thread of global init has exited, panic
775 * immediately to get a useable coredump.
777 if (unlikely(is_global_init(tsk)))
778 panic("Attempted to kill init! exitcode=0x%08x\n",
779 tsk->signal->group_exit_code ?: (int)code);
781 #ifdef CONFIG_POSIX_TIMERS
782 hrtimer_cancel(&tsk->signal->real_timer);
783 exit_itimers(tsk->signal);
786 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
788 acct_collect(code, group_dead);
793 tsk->exit_code = code;
794 taskstats_exit(tsk, group_dead);
800 trace_sched_process_exit(tsk);
807 disassociate_ctty(1);
808 exit_task_namespaces(tsk);
813 * Flush inherited counters to the parent - before the parent
814 * gets woken up by child-exit notifications.
816 * because of cgroup mode, must be called before cgroup_exit()
818 perf_event_exit_task(tsk);
820 sched_autogroup_exit_task(tsk);
824 * FIXME: do that only when needed, using sched_exit tracepoint
826 flush_ptrace_hw_breakpoint(tsk);
828 exit_tasks_rcu_start();
829 exit_notify(tsk, group_dead);
830 proc_exit_connector(tsk);
831 mpol_put_task_policy(tsk);
833 if (unlikely(current->pi_state_cache))
834 kfree(current->pi_state_cache);
837 * Make sure we are holding no locks:
839 debug_check_no_locks_held();
842 exit_io_context(tsk);
844 if (tsk->splice_pipe)
845 free_pipe_info(tsk->splice_pipe);
847 if (tsk->task_frag.page)
848 put_page(tsk->task_frag.page);
850 validate_creds_for_do_exit(tsk);
855 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
857 exit_tasks_rcu_finish();
859 lockdep_free_task(tsk);
862 EXPORT_SYMBOL_GPL(do_exit);
864 void complete_and_exit(struct completion *comp, long code)
871 EXPORT_SYMBOL(complete_and_exit);
873 SYSCALL_DEFINE1(exit, int, error_code)
875 do_exit((error_code&0xff)<<8);
879 * Take down every thread in the group. This is called by fatal signals
880 * as well as by sys_exit_group (below).
883 do_group_exit(int exit_code)
885 struct signal_struct *sig = current->signal;
887 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
889 if (signal_group_exit(sig))
890 exit_code = sig->group_exit_code;
891 else if (!thread_group_empty(current)) {
892 struct sighand_struct *const sighand = current->sighand;
894 spin_lock_irq(&sighand->siglock);
895 if (signal_group_exit(sig))
896 /* Another thread got here before we took the lock. */
897 exit_code = sig->group_exit_code;
899 sig->group_exit_code = exit_code;
900 sig->flags = SIGNAL_GROUP_EXIT;
901 zap_other_threads(current);
903 spin_unlock_irq(&sighand->siglock);
911 * this kills every thread in the thread group. Note that any externally
912 * wait4()-ing process will get the correct exit code - even if this
913 * thread is not the thread group leader.
915 SYSCALL_DEFINE1(exit_group, int, error_code)
917 do_group_exit((error_code & 0xff) << 8);
930 enum pid_type wo_type;
934 struct waitid_info *wo_info;
936 struct rusage *wo_rusage;
938 wait_queue_entry_t child_wait;
942 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
944 return wo->wo_type == PIDTYPE_MAX ||
945 task_pid_type(p, wo->wo_type) == wo->wo_pid;
949 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
951 if (!eligible_pid(wo, p))
955 * Wait for all children (clone and not) if __WALL is set or
956 * if it is traced by us.
958 if (ptrace || (wo->wo_flags & __WALL))
962 * Otherwise, wait for clone children *only* if __WCLONE is set;
963 * otherwise, wait for non-clone children *only*.
965 * Note: a "clone" child here is one that reports to its parent
966 * using a signal other than SIGCHLD, or a non-leader thread which
967 * we can only see if it is traced by us.
969 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
976 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
977 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
978 * the lock and this task is uninteresting. If we return nonzero, we have
979 * released the lock and the system call should return.
981 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
984 pid_t pid = task_pid_vnr(p);
985 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
986 struct waitid_info *infop;
988 if (!likely(wo->wo_flags & WEXITED))
991 if (unlikely(wo->wo_flags & WNOWAIT)) {
992 status = p->exit_code;
994 read_unlock(&tasklist_lock);
995 sched_annotate_sleep();
997 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1002 * Move the task's state to DEAD/TRACE, only one thread can do this.
1004 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1005 EXIT_TRACE : EXIT_DEAD;
1006 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1009 * We own this thread, nobody else can reap it.
1011 read_unlock(&tasklist_lock);
1012 sched_annotate_sleep();
1015 * Check thread_group_leader() to exclude the traced sub-threads.
1017 if (state == EXIT_DEAD && thread_group_leader(p)) {
1018 struct signal_struct *sig = p->signal;
1019 struct signal_struct *psig = current->signal;
1020 unsigned long maxrss;
1021 u64 tgutime, tgstime;
1024 * The resource counters for the group leader are in its
1025 * own task_struct. Those for dead threads in the group
1026 * are in its signal_struct, as are those for the child
1027 * processes it has previously reaped. All these
1028 * accumulate in the parent's signal_struct c* fields.
1030 * We don't bother to take a lock here to protect these
1031 * p->signal fields because the whole thread group is dead
1032 * and nobody can change them.
1034 * psig->stats_lock also protects us from our sub-theads
1035 * which can reap other children at the same time. Until
1036 * we change k_getrusage()-like users to rely on this lock
1037 * we have to take ->siglock as well.
1039 * We use thread_group_cputime_adjusted() to get times for
1040 * the thread group, which consolidates times for all threads
1041 * in the group including the group leader.
1043 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1044 spin_lock_irq(¤t->sighand->siglock);
1045 write_seqlock(&psig->stats_lock);
1046 psig->cutime += tgutime + sig->cutime;
1047 psig->cstime += tgstime + sig->cstime;
1048 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1050 p->min_flt + sig->min_flt + sig->cmin_flt;
1052 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1054 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1056 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1058 task_io_get_inblock(p) +
1059 sig->inblock + sig->cinblock;
1061 task_io_get_oublock(p) +
1062 sig->oublock + sig->coublock;
1063 maxrss = max(sig->maxrss, sig->cmaxrss);
1064 if (psig->cmaxrss < maxrss)
1065 psig->cmaxrss = maxrss;
1066 task_io_accounting_add(&psig->ioac, &p->ioac);
1067 task_io_accounting_add(&psig->ioac, &sig->ioac);
1068 write_sequnlock(&psig->stats_lock);
1069 spin_unlock_irq(¤t->sighand->siglock);
1073 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1074 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1075 ? p->signal->group_exit_code : p->exit_code;
1076 wo->wo_stat = status;
1078 if (state == EXIT_TRACE) {
1079 write_lock_irq(&tasklist_lock);
1080 /* We dropped tasklist, ptracer could die and untrace */
1083 /* If parent wants a zombie, don't release it now */
1084 state = EXIT_ZOMBIE;
1085 if (do_notify_parent(p, p->exit_signal))
1087 p->exit_state = state;
1088 write_unlock_irq(&tasklist_lock);
1090 if (state == EXIT_DEAD)
1094 infop = wo->wo_info;
1096 if ((status & 0x7f) == 0) {
1097 infop->cause = CLD_EXITED;
1098 infop->status = status >> 8;
1100 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1101 infop->status = status & 0x7f;
1110 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1113 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1114 return &p->exit_code;
1116 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1117 return &p->signal->group_exit_code;
1123 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1125 * @ptrace: is the wait for ptrace
1126 * @p: task to wait for
1128 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1131 * read_lock(&tasklist_lock), which is released if return value is
1132 * non-zero. Also, grabs and releases @p->sighand->siglock.
1135 * 0 if wait condition didn't exist and search for other wait conditions
1136 * should continue. Non-zero return, -errno on failure and @p's pid on
1137 * success, implies that tasklist_lock is released and wait condition
1138 * search should terminate.
1140 static int wait_task_stopped(struct wait_opts *wo,
1141 int ptrace, struct task_struct *p)
1143 struct waitid_info *infop;
1144 int exit_code, *p_code, why;
1145 uid_t uid = 0; /* unneeded, required by compiler */
1149 * Traditionally we see ptrace'd stopped tasks regardless of options.
1151 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1154 if (!task_stopped_code(p, ptrace))
1158 spin_lock_irq(&p->sighand->siglock);
1160 p_code = task_stopped_code(p, ptrace);
1161 if (unlikely(!p_code))
1164 exit_code = *p_code;
1168 if (!unlikely(wo->wo_flags & WNOWAIT))
1171 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1173 spin_unlock_irq(&p->sighand->siglock);
1178 * Now we are pretty sure this task is interesting.
1179 * Make sure it doesn't get reaped out from under us while we
1180 * give up the lock and then examine it below. We don't want to
1181 * keep holding onto the tasklist_lock while we call getrusage and
1182 * possibly take page faults for user memory.
1185 pid = task_pid_vnr(p);
1186 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1187 read_unlock(&tasklist_lock);
1188 sched_annotate_sleep();
1190 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1193 if (likely(!(wo->wo_flags & WNOWAIT)))
1194 wo->wo_stat = (exit_code << 8) | 0x7f;
1196 infop = wo->wo_info;
1199 infop->status = exit_code;
1207 * Handle do_wait work for one task in a live, non-stopped state.
1208 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1209 * the lock and this task is uninteresting. If we return nonzero, we have
1210 * released the lock and the system call should return.
1212 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1214 struct waitid_info *infop;
1218 if (!unlikely(wo->wo_flags & WCONTINUED))
1221 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1224 spin_lock_irq(&p->sighand->siglock);
1225 /* Re-check with the lock held. */
1226 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1227 spin_unlock_irq(&p->sighand->siglock);
1230 if (!unlikely(wo->wo_flags & WNOWAIT))
1231 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1232 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1233 spin_unlock_irq(&p->sighand->siglock);
1235 pid = task_pid_vnr(p);
1237 read_unlock(&tasklist_lock);
1238 sched_annotate_sleep();
1240 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1243 infop = wo->wo_info;
1245 wo->wo_stat = 0xffff;
1247 infop->cause = CLD_CONTINUED;
1250 infop->status = SIGCONT;
1256 * Consider @p for a wait by @parent.
1258 * -ECHILD should be in ->notask_error before the first call.
1259 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1260 * Returns zero if the search for a child should continue;
1261 * then ->notask_error is 0 if @p is an eligible child,
1264 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1265 struct task_struct *p)
1268 * We can race with wait_task_zombie() from another thread.
1269 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1270 * can't confuse the checks below.
1272 int exit_state = READ_ONCE(p->exit_state);
1275 if (unlikely(exit_state == EXIT_DEAD))
1278 ret = eligible_child(wo, ptrace, p);
1282 if (unlikely(exit_state == EXIT_TRACE)) {
1284 * ptrace == 0 means we are the natural parent. In this case
1285 * we should clear notask_error, debugger will notify us.
1287 if (likely(!ptrace))
1288 wo->notask_error = 0;
1292 if (likely(!ptrace) && unlikely(p->ptrace)) {
1294 * If it is traced by its real parent's group, just pretend
1295 * the caller is ptrace_do_wait() and reap this child if it
1298 * This also hides group stop state from real parent; otherwise
1299 * a single stop can be reported twice as group and ptrace stop.
1300 * If a ptracer wants to distinguish these two events for its
1301 * own children it should create a separate process which takes
1302 * the role of real parent.
1304 if (!ptrace_reparented(p))
1309 if (exit_state == EXIT_ZOMBIE) {
1310 /* we don't reap group leaders with subthreads */
1311 if (!delay_group_leader(p)) {
1313 * A zombie ptracee is only visible to its ptracer.
1314 * Notification and reaping will be cascaded to the
1315 * real parent when the ptracer detaches.
1317 if (unlikely(ptrace) || likely(!p->ptrace))
1318 return wait_task_zombie(wo, p);
1322 * Allow access to stopped/continued state via zombie by
1323 * falling through. Clearing of notask_error is complex.
1327 * If WEXITED is set, notask_error should naturally be
1328 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1329 * so, if there are live subthreads, there are events to
1330 * wait for. If all subthreads are dead, it's still safe
1331 * to clear - this function will be called again in finite
1332 * amount time once all the subthreads are released and
1333 * will then return without clearing.
1337 * Stopped state is per-task and thus can't change once the
1338 * target task dies. Only continued and exited can happen.
1339 * Clear notask_error if WCONTINUED | WEXITED.
1341 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1342 wo->notask_error = 0;
1345 * @p is alive and it's gonna stop, continue or exit, so
1346 * there always is something to wait for.
1348 wo->notask_error = 0;
1352 * Wait for stopped. Depending on @ptrace, different stopped state
1353 * is used and the two don't interact with each other.
1355 ret = wait_task_stopped(wo, ptrace, p);
1360 * Wait for continued. There's only one continued state and the
1361 * ptracer can consume it which can confuse the real parent. Don't
1362 * use WCONTINUED from ptracer. You don't need or want it.
1364 return wait_task_continued(wo, p);
1368 * Do the work of do_wait() for one thread in the group, @tsk.
1370 * -ECHILD should be in ->notask_error before the first call.
1371 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1372 * Returns zero if the search for a child should continue; then
1373 * ->notask_error is 0 if there were any eligible children,
1376 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1378 struct task_struct *p;
1380 list_for_each_entry(p, &tsk->children, sibling) {
1381 int ret = wait_consider_task(wo, 0, p);
1390 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1392 struct task_struct *p;
1394 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1395 int ret = wait_consider_task(wo, 1, p);
1404 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1405 int sync, void *key)
1407 struct wait_opts *wo = container_of(wait, struct wait_opts,
1409 struct task_struct *p = key;
1411 if (!eligible_pid(wo, p))
1414 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1417 return default_wake_function(wait, mode, sync, key);
1420 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1422 __wake_up_sync_key(&parent->signal->wait_chldexit,
1423 TASK_INTERRUPTIBLE, p);
1426 static long do_wait(struct wait_opts *wo)
1428 struct task_struct *tsk;
1431 trace_sched_process_wait(wo->wo_pid);
1433 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1434 wo->child_wait.private = current;
1435 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1438 * If there is nothing that can match our criteria, just get out.
1439 * We will clear ->notask_error to zero if we see any child that
1440 * might later match our criteria, even if we are not able to reap
1443 wo->notask_error = -ECHILD;
1444 if ((wo->wo_type < PIDTYPE_MAX) &&
1445 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1448 set_current_state(TASK_INTERRUPTIBLE);
1449 read_lock(&tasklist_lock);
1452 retval = do_wait_thread(wo, tsk);
1456 retval = ptrace_do_wait(wo, tsk);
1460 if (wo->wo_flags & __WNOTHREAD)
1462 } while_each_thread(current, tsk);
1463 read_unlock(&tasklist_lock);
1466 retval = wo->notask_error;
1467 if (!retval && !(wo->wo_flags & WNOHANG)) {
1468 retval = -ERESTARTSYS;
1469 if (!signal_pending(current)) {
1475 __set_current_state(TASK_RUNNING);
1476 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1480 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1481 int options, struct rusage *ru)
1483 struct wait_opts wo;
1484 struct pid *pid = NULL;
1487 unsigned int f_flags = 0;
1489 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1490 __WNOTHREAD|__WCLONE|__WALL))
1492 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1504 pid = find_get_pid(upid);
1507 type = PIDTYPE_PGID;
1512 pid = find_get_pid(upid);
1514 pid = get_task_pid(current, PIDTYPE_PGID);
1521 pid = pidfd_get_pid(upid, &f_flags);
1523 return PTR_ERR(pid);
1532 wo.wo_flags = options;
1535 if (f_flags & O_NONBLOCK)
1536 wo.wo_flags |= WNOHANG;
1539 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1546 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1547 infop, int, options, struct rusage __user *, ru)
1550 struct waitid_info info = {.status = 0};
1551 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1557 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1563 if (!user_write_access_begin(infop, sizeof(*infop)))
1566 unsafe_put_user(signo, &infop->si_signo, Efault);
1567 unsafe_put_user(0, &infop->si_errno, Efault);
1568 unsafe_put_user(info.cause, &infop->si_code, Efault);
1569 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1570 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1571 unsafe_put_user(info.status, &infop->si_status, Efault);
1572 user_write_access_end();
1575 user_write_access_end();
1579 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1582 struct wait_opts wo;
1583 struct pid *pid = NULL;
1587 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1588 __WNOTHREAD|__WCLONE|__WALL))
1591 /* -INT_MIN is not defined */
1592 if (upid == INT_MIN)
1597 else if (upid < 0) {
1598 type = PIDTYPE_PGID;
1599 pid = find_get_pid(-upid);
1600 } else if (upid == 0) {
1601 type = PIDTYPE_PGID;
1602 pid = get_task_pid(current, PIDTYPE_PGID);
1603 } else /* upid > 0 */ {
1605 pid = find_get_pid(upid);
1610 wo.wo_flags = options | WEXITED;
1616 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1622 int kernel_wait(pid_t pid, int *stat)
1624 struct wait_opts wo = {
1625 .wo_type = PIDTYPE_PID,
1626 .wo_pid = find_get_pid(pid),
1627 .wo_flags = WEXITED,
1632 if (ret > 0 && wo.wo_stat)
1638 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1639 int, options, struct rusage __user *, ru)
1642 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1645 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1651 #ifdef __ARCH_WANT_SYS_WAITPID
1654 * sys_waitpid() remains for compatibility. waitpid() should be
1655 * implemented by calling sys_wait4() from libc.a.
1657 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1659 return kernel_wait4(pid, stat_addr, options, NULL);
1664 #ifdef CONFIG_COMPAT
1665 COMPAT_SYSCALL_DEFINE4(wait4,
1667 compat_uint_t __user *, stat_addr,
1669 struct compat_rusage __user *, ru)
1672 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1674 if (ru && put_compat_rusage(&r, ru))
1680 COMPAT_SYSCALL_DEFINE5(waitid,
1681 int, which, compat_pid_t, pid,
1682 struct compat_siginfo __user *, infop, int, options,
1683 struct compat_rusage __user *, uru)
1686 struct waitid_info info = {.status = 0};
1687 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1693 /* kernel_waitid() overwrites everything in ru */
1694 if (COMPAT_USE_64BIT_TIME)
1695 err = copy_to_user(uru, &ru, sizeof(ru));
1697 err = put_compat_rusage(&ru, uru);
1706 if (!user_write_access_begin(infop, sizeof(*infop)))
1709 unsafe_put_user(signo, &infop->si_signo, Efault);
1710 unsafe_put_user(0, &infop->si_errno, Efault);
1711 unsafe_put_user(info.cause, &infop->si_code, Efault);
1712 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1713 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1714 unsafe_put_user(info.status, &infop->si_status, Efault);
1715 user_write_access_end();
1718 user_write_access_end();
1724 * thread_group_exited - check that a thread group has exited
1725 * @pid: tgid of thread group to be checked.
1727 * Test if the thread group represented by tgid has exited (all
1728 * threads are zombies, dead or completely gone).
1730 * Return: true if the thread group has exited. false otherwise.
1732 bool thread_group_exited(struct pid *pid)
1734 struct task_struct *task;
1738 task = pid_task(pid, PIDTYPE_PID);
1740 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1745 EXPORT_SYMBOL(thread_group_exited);
1747 __weak void abort(void)
1751 /* if that doesn't kill us, halt */
1752 panic("Oops failed to kill thread");
1754 EXPORT_SYMBOL(abort);