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/task_io_accounting_ops.h>
52 #include <linux/tracehook.h>
53 #include <linux/fs_struct.h>
54 #include <linux/init_task.h>
55 #include <linux/perf_event.h>
56 #include <trace/events/sched.h>
57 #include <linux/hw_breakpoint.h>
58 #include <linux/oom.h>
59 #include <linux/writeback.h>
60 #include <linux/shm.h>
61 #include <linux/kcov.h>
62 #include <linux/random.h>
63 #include <linux/rcuwait.h>
64 #include <linux/compat.h>
65 #include <linux/io_uring.h>
66 #include <linux/kprobes.h>
68 #include <linux/uaccess.h>
69 #include <asm/unistd.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 *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);
115 * If there is any task waiting for the group exit
118 if (sig->notify_count > 0 && !--sig->notify_count)
119 wake_up_process(sig->group_exit_task);
121 if (tsk == sig->curr_target)
122 sig->curr_target = next_thread(tsk);
125 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
126 sizeof(unsigned long long));
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
134 task_cputime(tsk, &utime, &stime);
135 write_seqlock(&sig->stats_lock);
138 sig->gtime += task_gtime(tsk);
139 sig->min_flt += tsk->min_flt;
140 sig->maj_flt += tsk->maj_flt;
141 sig->nvcsw += tsk->nvcsw;
142 sig->nivcsw += tsk->nivcsw;
143 sig->inblock += task_io_get_inblock(tsk);
144 sig->oublock += task_io_get_oublock(tsk);
145 task_io_accounting_add(&sig->ioac, &tsk->ioac);
146 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
148 __unhash_process(tsk, group_dead);
149 write_sequnlock(&sig->stats_lock);
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 flush_sigqueue(&tsk->pending);
157 spin_unlock(&sighand->siglock);
159 __cleanup_sighand(sighand);
160 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
162 flush_sigqueue(&sig->shared_pending);
167 static void delayed_put_task_struct(struct rcu_head *rhp)
169 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
171 kprobe_flush_task(tsk);
172 perf_event_delayed_put(tsk);
173 trace_sched_process_free(tsk);
174 put_task_struct(tsk);
177 void put_task_struct_rcu_user(struct task_struct *task)
179 if (refcount_dec_and_test(&task->rcu_users))
180 call_rcu(&task->rcu, delayed_put_task_struct);
183 void release_task(struct task_struct *p)
185 struct task_struct *leader;
186 struct pid *thread_pid;
189 /* don't need to get the RCU readlock here - the process is dead and
190 * can't be modifying its own credentials. But shut RCU-lockdep up */
192 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
197 write_lock_irq(&tasklist_lock);
198 ptrace_release_task(p);
199 thread_pid = get_pid(p->thread_pid);
203 * If we are the last non-leader member of the thread
204 * group, and the leader is zombie, then notify the
205 * group leader's parent process. (if it wants notification.)
208 leader = p->group_leader;
209 if (leader != p && thread_group_empty(leader)
210 && leader->exit_state == EXIT_ZOMBIE) {
212 * If we were the last child thread and the leader has
213 * exited already, and the leader's parent ignores SIGCHLD,
214 * then we are the one who should release the leader.
216 zap_leader = do_notify_parent(leader, leader->exit_signal);
218 leader->exit_state = EXIT_DEAD;
221 write_unlock_irq(&tasklist_lock);
222 seccomp_filter_release(p);
223 proc_flush_pid(thread_pid);
226 put_task_struct_rcu_user(p);
229 if (unlikely(zap_leader))
233 int rcuwait_wake_up(struct rcuwait *w)
236 struct task_struct *task;
241 * Order condition vs @task, such that everything prior to the load
242 * of @task is visible. This is the condition as to why the user called
243 * rcuwait_wake() in the first place. Pairs with set_current_state()
244 * barrier (A) in rcuwait_wait_event().
247 * [S] tsk = current [S] cond = true
253 task = rcu_dereference(w->task);
255 ret = wake_up_process(task);
260 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
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);
343 static void coredump_task_exit(struct task_struct *tsk)
345 struct core_state *core_state;
348 * Serialize with any possible pending coredump.
349 * We must hold siglock around checking core_state
350 * and setting PF_POSTCOREDUMP. The core-inducing thread
351 * will increment ->nr_threads for each thread in the
352 * group without PF_POSTCOREDUMP set.
354 spin_lock_irq(&tsk->sighand->siglock);
355 tsk->flags |= PF_POSTCOREDUMP;
356 core_state = tsk->signal->core_state;
357 spin_unlock_irq(&tsk->sighand->siglock);
359 struct core_thread self;
362 if (self.task->flags & PF_SIGNALED)
363 self.next = xchg(&core_state->dumper.next, &self);
367 * Implies mb(), the result of xchg() must be visible
368 * to core_state->dumper.
370 if (atomic_dec_and_test(&core_state->nr_threads))
371 complete(&core_state->startup);
374 set_current_state(TASK_UNINTERRUPTIBLE);
375 if (!self.task) /* see coredump_finish() */
377 freezable_schedule();
379 __set_current_state(TASK_RUNNING);
385 * A task is exiting. If it owned this mm, find a new owner for the mm.
387 void mm_update_next_owner(struct mm_struct *mm)
389 struct task_struct *c, *g, *p = current;
393 * If the exiting or execing task is not the owner, it's
394 * someone else's problem.
399 * The current owner is exiting/execing and there are no other
400 * candidates. Do not leave the mm pointing to a possibly
401 * freed task structure.
403 if (atomic_read(&mm->mm_users) <= 1) {
404 WRITE_ONCE(mm->owner, NULL);
408 read_lock(&tasklist_lock);
410 * Search in the children
412 list_for_each_entry(c, &p->children, sibling) {
414 goto assign_new_owner;
418 * Search in the siblings
420 list_for_each_entry(c, &p->real_parent->children, sibling) {
422 goto assign_new_owner;
426 * Search through everything else, we should not get here often.
428 for_each_process(g) {
429 if (g->flags & PF_KTHREAD)
431 for_each_thread(g, c) {
433 goto assign_new_owner;
438 read_unlock(&tasklist_lock);
440 * We found no owner yet mm_users > 1: this implies that we are
441 * most likely racing with swapoff (try_to_unuse()) or /proc or
442 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
444 WRITE_ONCE(mm->owner, NULL);
451 * The task_lock protects c->mm from changing.
452 * We always want mm->owner->mm == mm
456 * Delay read_unlock() till we have the task_lock()
457 * to ensure that c does not slip away underneath us
459 read_unlock(&tasklist_lock);
465 WRITE_ONCE(mm->owner, c);
469 #endif /* CONFIG_MEMCG */
472 * Turn us into a lazy TLB process if we
475 static void exit_mm(void)
477 struct mm_struct *mm = current->mm;
479 exit_mm_release(current, mm);
485 BUG_ON(mm != current->active_mm);
486 /* more a memory barrier than a real lock */
489 * When a thread stops operating on an address space, the loop
490 * in membarrier_private_expedited() may not observe that
491 * tsk->mm, and the loop in membarrier_global_expedited() may
492 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
493 * rq->membarrier_state, so those would not issue an IPI.
494 * Membarrier requires a memory barrier after accessing
495 * user-space memory, before clearing tsk->mm or the
496 * rq->membarrier_state.
498 smp_mb__after_spinlock();
501 membarrier_update_current_mm(NULL);
502 enter_lazy_tlb(mm, current);
504 task_unlock(current);
505 mmap_read_unlock(mm);
506 mm_update_next_owner(mm);
508 if (test_thread_flag(TIF_MEMDIE))
512 static struct task_struct *find_alive_thread(struct task_struct *p)
514 struct task_struct *t;
516 for_each_thread(p, t) {
517 if (!(t->flags & PF_EXITING))
523 static struct task_struct *find_child_reaper(struct task_struct *father,
524 struct list_head *dead)
525 __releases(&tasklist_lock)
526 __acquires(&tasklist_lock)
528 struct pid_namespace *pid_ns = task_active_pid_ns(father);
529 struct task_struct *reaper = pid_ns->child_reaper;
530 struct task_struct *p, *n;
532 if (likely(reaper != father))
535 reaper = find_alive_thread(father);
537 pid_ns->child_reaper = reaper;
541 write_unlock_irq(&tasklist_lock);
543 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
544 list_del_init(&p->ptrace_entry);
548 zap_pid_ns_processes(pid_ns);
549 write_lock_irq(&tasklist_lock);
555 * When we die, we re-parent all our children, and try to:
556 * 1. give them to another thread in our thread group, if such a member exists
557 * 2. give it to the first ancestor process which prctl'd itself as a
558 * child_subreaper for its children (like a service manager)
559 * 3. give it to the init process (PID 1) in our pid namespace
561 static struct task_struct *find_new_reaper(struct task_struct *father,
562 struct task_struct *child_reaper)
564 struct task_struct *thread, *reaper;
566 thread = find_alive_thread(father);
570 if (father->signal->has_child_subreaper) {
571 unsigned int ns_level = task_pid(father)->level;
573 * Find the first ->is_child_subreaper ancestor in our pid_ns.
574 * We can't check reaper != child_reaper to ensure we do not
575 * cross the namespaces, the exiting parent could be injected
576 * by setns() + fork().
577 * We check pid->level, this is slightly more efficient than
578 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
580 for (reaper = father->real_parent;
581 task_pid(reaper)->level == ns_level;
582 reaper = reaper->real_parent) {
583 if (reaper == &init_task)
585 if (!reaper->signal->is_child_subreaper)
587 thread = find_alive_thread(reaper);
597 * Any that need to be release_task'd are put on the @dead list.
599 static void reparent_leader(struct task_struct *father, struct task_struct *p,
600 struct list_head *dead)
602 if (unlikely(p->exit_state == EXIT_DEAD))
605 /* We don't want people slaying init. */
606 p->exit_signal = SIGCHLD;
608 /* If it has exited notify the new parent about this child's death. */
610 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
611 if (do_notify_parent(p, p->exit_signal)) {
612 p->exit_state = EXIT_DEAD;
613 list_add(&p->ptrace_entry, dead);
617 kill_orphaned_pgrp(p, father);
621 * This does two things:
623 * A. Make init inherit all the child processes
624 * B. Check to see if any process groups have become orphaned
625 * as a result of our exiting, and if they have any stopped
626 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
628 static void forget_original_parent(struct task_struct *father,
629 struct list_head *dead)
631 struct task_struct *p, *t, *reaper;
633 if (unlikely(!list_empty(&father->ptraced)))
634 exit_ptrace(father, dead);
636 /* Can drop and reacquire tasklist_lock */
637 reaper = find_child_reaper(father, dead);
638 if (list_empty(&father->children))
641 reaper = find_new_reaper(father, reaper);
642 list_for_each_entry(p, &father->children, sibling) {
643 for_each_thread(p, t) {
644 RCU_INIT_POINTER(t->real_parent, reaper);
645 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
646 if (likely(!t->ptrace))
647 t->parent = t->real_parent;
648 if (t->pdeath_signal)
649 group_send_sig_info(t->pdeath_signal,
654 * If this is a threaded reparent there is no need to
655 * notify anyone anything has happened.
657 if (!same_thread_group(reaper, father))
658 reparent_leader(father, p, dead);
660 list_splice_tail_init(&father->children, &reaper->children);
664 * Send signals to all our closest relatives so that they know
665 * to properly mourn us..
667 static void exit_notify(struct task_struct *tsk, int group_dead)
670 struct task_struct *p, *n;
673 write_lock_irq(&tasklist_lock);
674 forget_original_parent(tsk, &dead);
677 kill_orphaned_pgrp(tsk->group_leader, NULL);
679 tsk->exit_state = EXIT_ZOMBIE;
680 if (unlikely(tsk->ptrace)) {
681 int sig = thread_group_leader(tsk) &&
682 thread_group_empty(tsk) &&
683 !ptrace_reparented(tsk) ?
684 tsk->exit_signal : SIGCHLD;
685 autoreap = do_notify_parent(tsk, sig);
686 } else if (thread_group_leader(tsk)) {
687 autoreap = thread_group_empty(tsk) &&
688 do_notify_parent(tsk, tsk->exit_signal);
694 tsk->exit_state = EXIT_DEAD;
695 list_add(&tsk->ptrace_entry, &dead);
698 /* mt-exec, de_thread() is waiting for group leader */
699 if (unlikely(tsk->signal->notify_count < 0))
700 wake_up_process(tsk->signal->group_exit_task);
701 write_unlock_irq(&tasklist_lock);
703 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
704 list_del_init(&p->ptrace_entry);
709 #ifdef CONFIG_DEBUG_STACK_USAGE
710 static void check_stack_usage(void)
712 static DEFINE_SPINLOCK(low_water_lock);
713 static int lowest_to_date = THREAD_SIZE;
716 free = stack_not_used(current);
718 if (free >= lowest_to_date)
721 spin_lock(&low_water_lock);
722 if (free < lowest_to_date) {
723 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
724 current->comm, task_pid_nr(current), free);
725 lowest_to_date = free;
727 spin_unlock(&low_water_lock);
730 static inline void check_stack_usage(void) {}
733 void __noreturn do_exit(long code)
735 struct task_struct *tsk = current;
739 * We can get here from a kernel oops, sometimes with preemption off.
740 * Start by checking for critical errors.
741 * Then fix up important state like USER_DS and preemption.
742 * Then do everything else.
745 WARN_ON(blk_needs_flush_plug(tsk));
747 if (unlikely(in_interrupt()))
748 panic("Aiee, killing interrupt handler!");
749 if (unlikely(!tsk->pid))
750 panic("Attempted to kill the idle task!");
753 * If do_exit is called because this processes oopsed, it's possible
754 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
755 * continuing. Amongst other possible reasons, this is to prevent
756 * mm_release()->clear_child_tid() from writing to a user-controlled
759 force_uaccess_begin();
761 if (unlikely(in_atomic())) {
762 pr_info("note: %s[%d] exited with preempt_count %d\n",
763 current->comm, task_pid_nr(current),
765 preempt_count_set(PREEMPT_ENABLED);
768 profile_task_exit(tsk);
771 coredump_task_exit(tsk);
772 ptrace_event(PTRACE_EVENT_EXIT, code);
774 validate_creds_for_do_exit(tsk);
777 * We're taking recursive faults here in do_exit. Safest is to just
778 * leave this task alone and wait for reboot.
780 if (unlikely(tsk->flags & PF_EXITING)) {
781 pr_alert("Fixing recursive fault but reboot is needed!\n");
782 futex_exit_recursive(tsk);
783 set_current_state(TASK_UNINTERRUPTIBLE);
787 io_uring_files_cancel();
788 exit_signals(tsk); /* sets PF_EXITING */
790 /* sync mm's RSS info before statistics gathering */
792 sync_mm_rss(tsk->mm);
793 acct_update_integrals(tsk);
794 group_dead = atomic_dec_and_test(&tsk->signal->live);
797 * If the last thread of global init has exited, panic
798 * immediately to get a useable coredump.
800 if (unlikely(is_global_init(tsk)))
801 panic("Attempted to kill init! exitcode=0x%08x\n",
802 tsk->signal->group_exit_code ?: (int)code);
804 #ifdef CONFIG_POSIX_TIMERS
805 hrtimer_cancel(&tsk->signal->real_timer);
806 exit_itimers(tsk->signal);
809 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
811 acct_collect(code, group_dead);
816 tsk->exit_code = code;
817 taskstats_exit(tsk, group_dead);
823 trace_sched_process_exit(tsk);
830 disassociate_ctty(1);
831 exit_task_namespaces(tsk);
836 * Flush inherited counters to the parent - before the parent
837 * gets woken up by child-exit notifications.
839 * because of cgroup mode, must be called before cgroup_exit()
841 perf_event_exit_task(tsk);
843 sched_autogroup_exit_task(tsk);
847 * FIXME: do that only when needed, using sched_exit tracepoint
849 flush_ptrace_hw_breakpoint(tsk);
851 exit_tasks_rcu_start();
852 exit_notify(tsk, group_dead);
853 proc_exit_connector(tsk);
854 mpol_put_task_policy(tsk);
856 if (unlikely(current->pi_state_cache))
857 kfree(current->pi_state_cache);
860 * Make sure we are holding no locks:
862 debug_check_no_locks_held();
865 exit_io_context(tsk);
867 if (tsk->splice_pipe)
868 free_pipe_info(tsk->splice_pipe);
870 if (tsk->task_frag.page)
871 put_page(tsk->task_frag.page);
873 validate_creds_for_do_exit(tsk);
878 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
880 exit_tasks_rcu_finish();
882 lockdep_free_task(tsk);
885 EXPORT_SYMBOL_GPL(do_exit);
887 void complete_and_exit(struct completion *comp, long code)
894 EXPORT_SYMBOL(complete_and_exit);
896 SYSCALL_DEFINE1(exit, int, error_code)
898 do_exit((error_code&0xff)<<8);
902 * Take down every thread in the group. This is called by fatal signals
903 * as well as by sys_exit_group (below).
906 do_group_exit(int exit_code)
908 struct signal_struct *sig = current->signal;
910 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
912 if (signal_group_exit(sig))
913 exit_code = sig->group_exit_code;
914 else if (!thread_group_empty(current)) {
915 struct sighand_struct *const sighand = current->sighand;
917 spin_lock_irq(&sighand->siglock);
918 if (signal_group_exit(sig))
919 /* Another thread got here before we took the lock. */
920 exit_code = sig->group_exit_code;
922 sig->group_exit_code = exit_code;
923 sig->flags = SIGNAL_GROUP_EXIT;
924 zap_other_threads(current);
926 spin_unlock_irq(&sighand->siglock);
934 * this kills every thread in the thread group. Note that any externally
935 * wait4()-ing process will get the correct exit code - even if this
936 * thread is not the thread group leader.
938 SYSCALL_DEFINE1(exit_group, int, error_code)
940 do_group_exit((error_code & 0xff) << 8);
953 enum pid_type wo_type;
957 struct waitid_info *wo_info;
959 struct rusage *wo_rusage;
961 wait_queue_entry_t child_wait;
965 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
967 return wo->wo_type == PIDTYPE_MAX ||
968 task_pid_type(p, wo->wo_type) == wo->wo_pid;
972 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
974 if (!eligible_pid(wo, p))
978 * Wait for all children (clone and not) if __WALL is set or
979 * if it is traced by us.
981 if (ptrace || (wo->wo_flags & __WALL))
985 * Otherwise, wait for clone children *only* if __WCLONE is set;
986 * otherwise, wait for non-clone children *only*.
988 * Note: a "clone" child here is one that reports to its parent
989 * using a signal other than SIGCHLD, or a non-leader thread which
990 * we can only see if it is traced by us.
992 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
999 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1000 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1001 * the lock and this task is uninteresting. If we return nonzero, we have
1002 * released the lock and the system call should return.
1004 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1007 pid_t pid = task_pid_vnr(p);
1008 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1009 struct waitid_info *infop;
1011 if (!likely(wo->wo_flags & WEXITED))
1014 if (unlikely(wo->wo_flags & WNOWAIT)) {
1015 status = p->exit_code;
1017 read_unlock(&tasklist_lock);
1018 sched_annotate_sleep();
1020 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1025 * Move the task's state to DEAD/TRACE, only one thread can do this.
1027 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1028 EXIT_TRACE : EXIT_DEAD;
1029 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1032 * We own this thread, nobody else can reap it.
1034 read_unlock(&tasklist_lock);
1035 sched_annotate_sleep();
1038 * Check thread_group_leader() to exclude the traced sub-threads.
1040 if (state == EXIT_DEAD && thread_group_leader(p)) {
1041 struct signal_struct *sig = p->signal;
1042 struct signal_struct *psig = current->signal;
1043 unsigned long maxrss;
1044 u64 tgutime, tgstime;
1047 * The resource counters for the group leader are in its
1048 * own task_struct. Those for dead threads in the group
1049 * are in its signal_struct, as are those for the child
1050 * processes it has previously reaped. All these
1051 * accumulate in the parent's signal_struct c* fields.
1053 * We don't bother to take a lock here to protect these
1054 * p->signal fields because the whole thread group is dead
1055 * and nobody can change them.
1057 * psig->stats_lock also protects us from our sub-theads
1058 * which can reap other children at the same time. Until
1059 * we change k_getrusage()-like users to rely on this lock
1060 * we have to take ->siglock as well.
1062 * We use thread_group_cputime_adjusted() to get times for
1063 * the thread group, which consolidates times for all threads
1064 * in the group including the group leader.
1066 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1067 spin_lock_irq(¤t->sighand->siglock);
1068 write_seqlock(&psig->stats_lock);
1069 psig->cutime += tgutime + sig->cutime;
1070 psig->cstime += tgstime + sig->cstime;
1071 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1073 p->min_flt + sig->min_flt + sig->cmin_flt;
1075 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1077 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1079 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1081 task_io_get_inblock(p) +
1082 sig->inblock + sig->cinblock;
1084 task_io_get_oublock(p) +
1085 sig->oublock + sig->coublock;
1086 maxrss = max(sig->maxrss, sig->cmaxrss);
1087 if (psig->cmaxrss < maxrss)
1088 psig->cmaxrss = maxrss;
1089 task_io_accounting_add(&psig->ioac, &p->ioac);
1090 task_io_accounting_add(&psig->ioac, &sig->ioac);
1091 write_sequnlock(&psig->stats_lock);
1092 spin_unlock_irq(¤t->sighand->siglock);
1096 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1097 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1098 ? p->signal->group_exit_code : p->exit_code;
1099 wo->wo_stat = status;
1101 if (state == EXIT_TRACE) {
1102 write_lock_irq(&tasklist_lock);
1103 /* We dropped tasklist, ptracer could die and untrace */
1106 /* If parent wants a zombie, don't release it now */
1107 state = EXIT_ZOMBIE;
1108 if (do_notify_parent(p, p->exit_signal))
1110 p->exit_state = state;
1111 write_unlock_irq(&tasklist_lock);
1113 if (state == EXIT_DEAD)
1117 infop = wo->wo_info;
1119 if ((status & 0x7f) == 0) {
1120 infop->cause = CLD_EXITED;
1121 infop->status = status >> 8;
1123 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1124 infop->status = status & 0x7f;
1133 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1136 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1137 return &p->exit_code;
1139 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1140 return &p->signal->group_exit_code;
1146 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1148 * @ptrace: is the wait for ptrace
1149 * @p: task to wait for
1151 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1154 * read_lock(&tasklist_lock), which is released if return value is
1155 * non-zero. Also, grabs and releases @p->sighand->siglock.
1158 * 0 if wait condition didn't exist and search for other wait conditions
1159 * should continue. Non-zero return, -errno on failure and @p's pid on
1160 * success, implies that tasklist_lock is released and wait condition
1161 * search should terminate.
1163 static int wait_task_stopped(struct wait_opts *wo,
1164 int ptrace, struct task_struct *p)
1166 struct waitid_info *infop;
1167 int exit_code, *p_code, why;
1168 uid_t uid = 0; /* unneeded, required by compiler */
1172 * Traditionally we see ptrace'd stopped tasks regardless of options.
1174 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1177 if (!task_stopped_code(p, ptrace))
1181 spin_lock_irq(&p->sighand->siglock);
1183 p_code = task_stopped_code(p, ptrace);
1184 if (unlikely(!p_code))
1187 exit_code = *p_code;
1191 if (!unlikely(wo->wo_flags & WNOWAIT))
1194 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1196 spin_unlock_irq(&p->sighand->siglock);
1201 * Now we are pretty sure this task is interesting.
1202 * Make sure it doesn't get reaped out from under us while we
1203 * give up the lock and then examine it below. We don't want to
1204 * keep holding onto the tasklist_lock while we call getrusage and
1205 * possibly take page faults for user memory.
1208 pid = task_pid_vnr(p);
1209 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1210 read_unlock(&tasklist_lock);
1211 sched_annotate_sleep();
1213 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1216 if (likely(!(wo->wo_flags & WNOWAIT)))
1217 wo->wo_stat = (exit_code << 8) | 0x7f;
1219 infop = wo->wo_info;
1222 infop->status = exit_code;
1230 * Handle do_wait work for one task in a live, non-stopped state.
1231 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1232 * the lock and this task is uninteresting. If we return nonzero, we have
1233 * released the lock and the system call should return.
1235 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1237 struct waitid_info *infop;
1241 if (!unlikely(wo->wo_flags & WCONTINUED))
1244 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1247 spin_lock_irq(&p->sighand->siglock);
1248 /* Re-check with the lock held. */
1249 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1250 spin_unlock_irq(&p->sighand->siglock);
1253 if (!unlikely(wo->wo_flags & WNOWAIT))
1254 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1255 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1256 spin_unlock_irq(&p->sighand->siglock);
1258 pid = task_pid_vnr(p);
1260 read_unlock(&tasklist_lock);
1261 sched_annotate_sleep();
1263 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1266 infop = wo->wo_info;
1268 wo->wo_stat = 0xffff;
1270 infop->cause = CLD_CONTINUED;
1273 infop->status = SIGCONT;
1279 * Consider @p for a wait by @parent.
1281 * -ECHILD should be in ->notask_error before the first call.
1282 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1283 * Returns zero if the search for a child should continue;
1284 * then ->notask_error is 0 if @p is an eligible child,
1287 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1288 struct task_struct *p)
1291 * We can race with wait_task_zombie() from another thread.
1292 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1293 * can't confuse the checks below.
1295 int exit_state = READ_ONCE(p->exit_state);
1298 if (unlikely(exit_state == EXIT_DEAD))
1301 ret = eligible_child(wo, ptrace, p);
1305 if (unlikely(exit_state == EXIT_TRACE)) {
1307 * ptrace == 0 means we are the natural parent. In this case
1308 * we should clear notask_error, debugger will notify us.
1310 if (likely(!ptrace))
1311 wo->notask_error = 0;
1315 if (likely(!ptrace) && unlikely(p->ptrace)) {
1317 * If it is traced by its real parent's group, just pretend
1318 * the caller is ptrace_do_wait() and reap this child if it
1321 * This also hides group stop state from real parent; otherwise
1322 * a single stop can be reported twice as group and ptrace stop.
1323 * If a ptracer wants to distinguish these two events for its
1324 * own children it should create a separate process which takes
1325 * the role of real parent.
1327 if (!ptrace_reparented(p))
1332 if (exit_state == EXIT_ZOMBIE) {
1333 /* we don't reap group leaders with subthreads */
1334 if (!delay_group_leader(p)) {
1336 * A zombie ptracee is only visible to its ptracer.
1337 * Notification and reaping will be cascaded to the
1338 * real parent when the ptracer detaches.
1340 if (unlikely(ptrace) || likely(!p->ptrace))
1341 return wait_task_zombie(wo, p);
1345 * Allow access to stopped/continued state via zombie by
1346 * falling through. Clearing of notask_error is complex.
1350 * If WEXITED is set, notask_error should naturally be
1351 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1352 * so, if there are live subthreads, there are events to
1353 * wait for. If all subthreads are dead, it's still safe
1354 * to clear - this function will be called again in finite
1355 * amount time once all the subthreads are released and
1356 * will then return without clearing.
1360 * Stopped state is per-task and thus can't change once the
1361 * target task dies. Only continued and exited can happen.
1362 * Clear notask_error if WCONTINUED | WEXITED.
1364 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1365 wo->notask_error = 0;
1368 * @p is alive and it's gonna stop, continue or exit, so
1369 * there always is something to wait for.
1371 wo->notask_error = 0;
1375 * Wait for stopped. Depending on @ptrace, different stopped state
1376 * is used and the two don't interact with each other.
1378 ret = wait_task_stopped(wo, ptrace, p);
1383 * Wait for continued. There's only one continued state and the
1384 * ptracer can consume it which can confuse the real parent. Don't
1385 * use WCONTINUED from ptracer. You don't need or want it.
1387 return wait_task_continued(wo, p);
1391 * Do the work of do_wait() for one thread in the group, @tsk.
1393 * -ECHILD should be in ->notask_error before the first call.
1394 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1395 * Returns zero if the search for a child should continue; then
1396 * ->notask_error is 0 if there were any eligible children,
1399 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1401 struct task_struct *p;
1403 list_for_each_entry(p, &tsk->children, sibling) {
1404 int ret = wait_consider_task(wo, 0, p);
1413 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1415 struct task_struct *p;
1417 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1418 int ret = wait_consider_task(wo, 1, p);
1427 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1428 int sync, void *key)
1430 struct wait_opts *wo = container_of(wait, struct wait_opts,
1432 struct task_struct *p = key;
1434 if (!eligible_pid(wo, p))
1437 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1440 return default_wake_function(wait, mode, sync, key);
1443 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1445 __wake_up_sync_key(&parent->signal->wait_chldexit,
1446 TASK_INTERRUPTIBLE, p);
1449 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1450 struct task_struct *target)
1452 struct task_struct *parent =
1453 !ptrace ? target->real_parent : target->parent;
1455 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1456 same_thread_group(current, parent));
1460 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1461 * and tracee lists to find the target task.
1463 static int do_wait_pid(struct wait_opts *wo)
1466 struct task_struct *target;
1470 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1471 if (target && is_effectively_child(wo, ptrace, target)) {
1472 retval = wait_consider_task(wo, ptrace, target);
1478 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1479 if (target && target->ptrace &&
1480 is_effectively_child(wo, ptrace, target)) {
1481 retval = wait_consider_task(wo, ptrace, target);
1489 static long do_wait(struct wait_opts *wo)
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 || !pid_has_task(wo->wo_pid, wo->wo_type)))
1510 set_current_state(TASK_INTERRUPTIBLE);
1511 read_lock(&tasklist_lock);
1513 if (wo->wo_type == PIDTYPE_PID) {
1514 retval = do_wait_pid(wo);
1518 struct task_struct *tsk = current;
1521 retval = do_wait_thread(wo, tsk);
1525 retval = ptrace_do_wait(wo, tsk);
1529 if (wo->wo_flags & __WNOTHREAD)
1531 } while_each_thread(current, tsk);
1533 read_unlock(&tasklist_lock);
1536 retval = wo->notask_error;
1537 if (!retval && !(wo->wo_flags & WNOHANG)) {
1538 retval = -ERESTARTSYS;
1539 if (!signal_pending(current)) {
1545 __set_current_state(TASK_RUNNING);
1546 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1550 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1551 int options, struct rusage *ru)
1553 struct wait_opts wo;
1554 struct pid *pid = NULL;
1557 unsigned int f_flags = 0;
1559 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1560 __WNOTHREAD|__WCLONE|__WALL))
1562 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1574 pid = find_get_pid(upid);
1577 type = PIDTYPE_PGID;
1582 pid = find_get_pid(upid);
1584 pid = get_task_pid(current, PIDTYPE_PGID);
1591 pid = pidfd_get_pid(upid, &f_flags);
1593 return PTR_ERR(pid);
1602 wo.wo_flags = options;
1605 if (f_flags & O_NONBLOCK)
1606 wo.wo_flags |= WNOHANG;
1609 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1616 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1617 infop, int, options, struct rusage __user *, ru)
1620 struct waitid_info info = {.status = 0};
1621 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1627 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1633 if (!user_write_access_begin(infop, sizeof(*infop)))
1636 unsafe_put_user(signo, &infop->si_signo, Efault);
1637 unsafe_put_user(0, &infop->si_errno, Efault);
1638 unsafe_put_user(info.cause, &infop->si_code, Efault);
1639 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1640 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1641 unsafe_put_user(info.status, &infop->si_status, Efault);
1642 user_write_access_end();
1645 user_write_access_end();
1649 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1652 struct wait_opts wo;
1653 struct pid *pid = NULL;
1657 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1658 __WNOTHREAD|__WCLONE|__WALL))
1661 /* -INT_MIN is not defined */
1662 if (upid == INT_MIN)
1667 else if (upid < 0) {
1668 type = PIDTYPE_PGID;
1669 pid = find_get_pid(-upid);
1670 } else if (upid == 0) {
1671 type = PIDTYPE_PGID;
1672 pid = get_task_pid(current, PIDTYPE_PGID);
1673 } else /* upid > 0 */ {
1675 pid = find_get_pid(upid);
1680 wo.wo_flags = options | WEXITED;
1686 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1692 int kernel_wait(pid_t pid, int *stat)
1694 struct wait_opts wo = {
1695 .wo_type = PIDTYPE_PID,
1696 .wo_pid = find_get_pid(pid),
1697 .wo_flags = WEXITED,
1702 if (ret > 0 && wo.wo_stat)
1708 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1709 int, options, struct rusage __user *, ru)
1712 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1715 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1721 #ifdef __ARCH_WANT_SYS_WAITPID
1724 * sys_waitpid() remains for compatibility. waitpid() should be
1725 * implemented by calling sys_wait4() from libc.a.
1727 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1729 return kernel_wait4(pid, stat_addr, options, NULL);
1734 #ifdef CONFIG_COMPAT
1735 COMPAT_SYSCALL_DEFINE4(wait4,
1737 compat_uint_t __user *, stat_addr,
1739 struct compat_rusage __user *, ru)
1742 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1744 if (ru && put_compat_rusage(&r, ru))
1750 COMPAT_SYSCALL_DEFINE5(waitid,
1751 int, which, compat_pid_t, pid,
1752 struct compat_siginfo __user *, infop, int, options,
1753 struct compat_rusage __user *, uru)
1756 struct waitid_info info = {.status = 0};
1757 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1763 /* kernel_waitid() overwrites everything in ru */
1764 if (COMPAT_USE_64BIT_TIME)
1765 err = copy_to_user(uru, &ru, sizeof(ru));
1767 err = put_compat_rusage(&ru, uru);
1776 if (!user_write_access_begin(infop, sizeof(*infop)))
1779 unsafe_put_user(signo, &infop->si_signo, Efault);
1780 unsafe_put_user(0, &infop->si_errno, Efault);
1781 unsafe_put_user(info.cause, &infop->si_code, Efault);
1782 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1783 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1784 unsafe_put_user(info.status, &infop->si_status, Efault);
1785 user_write_access_end();
1788 user_write_access_end();
1794 * thread_group_exited - check that a thread group has exited
1795 * @pid: tgid of thread group to be checked.
1797 * Test if the thread group represented by tgid has exited (all
1798 * threads are zombies, dead or completely gone).
1800 * Return: true if the thread group has exited. false otherwise.
1802 bool thread_group_exited(struct pid *pid)
1804 struct task_struct *task;
1808 task = pid_task(pid, PIDTYPE_PID);
1810 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1815 EXPORT_SYMBOL(thread_group_exited);
1817 __weak void abort(void)
1821 /* if that doesn't kill us, halt */
1822 panic("Oops failed to kill thread");
1824 EXPORT_SYMBOL(abort);