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);
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 perf_event_delayed_put(tsk);
172 trace_sched_process_free(tsk);
173 put_task_struct(tsk);
176 void put_task_struct_rcu_user(struct task_struct *task)
178 if (refcount_dec_and_test(&task->rcu_users))
179 call_rcu(&task->rcu, delayed_put_task_struct);
182 void release_task(struct task_struct *p)
184 struct task_struct *leader;
185 struct pid *thread_pid;
188 /* don't need to get the RCU readlock here - the process is dead and
189 * can't be modifying its own credentials. But shut RCU-lockdep up */
191 atomic_dec(&__task_cred(p)->user->processes);
196 write_lock_irq(&tasklist_lock);
197 ptrace_release_task(p);
198 thread_pid = get_pid(p->thread_pid);
202 * If we are the last non-leader member of the thread
203 * group, and the leader is zombie, then notify the
204 * group leader's parent process. (if it wants notification.)
207 leader = p->group_leader;
208 if (leader != p && thread_group_empty(leader)
209 && leader->exit_state == EXIT_ZOMBIE) {
211 * If we were the last child thread and the leader has
212 * exited already, and the leader's parent ignores SIGCHLD,
213 * then we are the one who should release the leader.
215 zap_leader = do_notify_parent(leader, leader->exit_signal);
217 leader->exit_state = EXIT_DEAD;
220 write_unlock_irq(&tasklist_lock);
221 proc_flush_pid(thread_pid);
224 put_task_struct_rcu_user(p);
227 if (unlikely(zap_leader))
231 void rcuwait_wake_up(struct rcuwait *w)
233 struct task_struct *task;
238 * Order condition vs @task, such that everything prior to the load
239 * of @task is visible. This is the condition as to why the user called
240 * rcuwait_trywake() in the first place. Pairs with set_current_state()
241 * barrier (A) in rcuwait_wait_event().
244 * [S] tsk = current [S] cond = true
250 task = rcu_dereference(w->task);
252 wake_up_process(task);
255 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
258 * Determine if a process group is "orphaned", according to the POSIX
259 * definition in 2.2.2.52. Orphaned process groups are not to be affected
260 * by terminal-generated stop signals. Newly orphaned process groups are
261 * to receive a SIGHUP and a SIGCONT.
263 * "I ask you, have you ever known what it is to be an orphan?"
265 static int will_become_orphaned_pgrp(struct pid *pgrp,
266 struct task_struct *ignored_task)
268 struct task_struct *p;
270 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
271 if ((p == ignored_task) ||
272 (p->exit_state && thread_group_empty(p)) ||
273 is_global_init(p->real_parent))
276 if (task_pgrp(p->real_parent) != pgrp &&
277 task_session(p->real_parent) == task_session(p))
279 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
284 int is_current_pgrp_orphaned(void)
288 read_lock(&tasklist_lock);
289 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
290 read_unlock(&tasklist_lock);
295 static bool has_stopped_jobs(struct pid *pgrp)
297 struct task_struct *p;
299 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
300 if (p->signal->flags & SIGNAL_STOP_STOPPED)
302 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
308 * Check to see if any process groups have become orphaned as
309 * a result of our exiting, and if they have any stopped jobs,
310 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
313 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
315 struct pid *pgrp = task_pgrp(tsk);
316 struct task_struct *ignored_task = tsk;
319 /* exit: our father is in a different pgrp than
320 * we are and we were the only connection outside.
322 parent = tsk->real_parent;
324 /* reparent: our child is in a different pgrp than
325 * we are, and it was the only connection outside.
329 if (task_pgrp(parent) != pgrp &&
330 task_session(parent) == task_session(tsk) &&
331 will_become_orphaned_pgrp(pgrp, ignored_task) &&
332 has_stopped_jobs(pgrp)) {
333 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
334 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
340 * A task is exiting. If it owned this mm, find a new owner for the mm.
342 void mm_update_next_owner(struct mm_struct *mm)
344 struct task_struct *c, *g, *p = current;
348 * If the exiting or execing task is not the owner, it's
349 * someone else's problem.
354 * The current owner is exiting/execing and there are no other
355 * candidates. Do not leave the mm pointing to a possibly
356 * freed task structure.
358 if (atomic_read(&mm->mm_users) <= 1) {
359 WRITE_ONCE(mm->owner, NULL);
363 read_lock(&tasklist_lock);
365 * Search in the children
367 list_for_each_entry(c, &p->children, sibling) {
369 goto assign_new_owner;
373 * Search in the siblings
375 list_for_each_entry(c, &p->real_parent->children, sibling) {
377 goto assign_new_owner;
381 * Search through everything else, we should not get here often.
383 for_each_process(g) {
384 if (g->flags & PF_KTHREAD)
386 for_each_thread(g, c) {
388 goto assign_new_owner;
393 read_unlock(&tasklist_lock);
395 * We found no owner yet mm_users > 1: this implies that we are
396 * most likely racing with swapoff (try_to_unuse()) or /proc or
397 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
399 WRITE_ONCE(mm->owner, NULL);
406 * The task_lock protects c->mm from changing.
407 * We always want mm->owner->mm == mm
411 * Delay read_unlock() till we have the task_lock()
412 * to ensure that c does not slip away underneath us
414 read_unlock(&tasklist_lock);
420 WRITE_ONCE(mm->owner, c);
424 #endif /* CONFIG_MEMCG */
427 * Turn us into a lazy TLB process if we
430 static void exit_mm(void)
432 struct mm_struct *mm = current->mm;
433 struct core_state *core_state;
435 exit_mm_release(current, mm);
440 * Serialize with any possible pending coredump.
441 * We must hold mmap_sem around checking core_state
442 * and clearing tsk->mm. The core-inducing thread
443 * will increment ->nr_threads for each thread in the
444 * group with ->mm != NULL.
446 down_read(&mm->mmap_sem);
447 core_state = mm->core_state;
449 struct core_thread self;
451 up_read(&mm->mmap_sem);
454 self.next = xchg(&core_state->dumper.next, &self);
456 * Implies mb(), the result of xchg() must be visible
457 * to core_state->dumper.
459 if (atomic_dec_and_test(&core_state->nr_threads))
460 complete(&core_state->startup);
463 set_current_state(TASK_UNINTERRUPTIBLE);
464 if (!self.task) /* see coredump_finish() */
466 freezable_schedule();
468 __set_current_state(TASK_RUNNING);
469 down_read(&mm->mmap_sem);
472 BUG_ON(mm != current->active_mm);
473 /* more a memory barrier than a real lock */
476 up_read(&mm->mmap_sem);
477 enter_lazy_tlb(mm, current);
478 task_unlock(current);
479 mm_update_next_owner(mm);
481 if (test_thread_flag(TIF_MEMDIE))
485 static struct task_struct *find_alive_thread(struct task_struct *p)
487 struct task_struct *t;
489 for_each_thread(p, t) {
490 if (!(t->flags & PF_EXITING))
496 static struct task_struct *find_child_reaper(struct task_struct *father,
497 struct list_head *dead)
498 __releases(&tasklist_lock)
499 __acquires(&tasklist_lock)
501 struct pid_namespace *pid_ns = task_active_pid_ns(father);
502 struct task_struct *reaper = pid_ns->child_reaper;
503 struct task_struct *p, *n;
505 if (likely(reaper != father))
508 reaper = find_alive_thread(father);
510 pid_ns->child_reaper = reaper;
514 write_unlock_irq(&tasklist_lock);
516 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
517 list_del_init(&p->ptrace_entry);
521 zap_pid_ns_processes(pid_ns);
522 write_lock_irq(&tasklist_lock);
528 * When we die, we re-parent all our children, and try to:
529 * 1. give them to another thread in our thread group, if such a member exists
530 * 2. give it to the first ancestor process which prctl'd itself as a
531 * child_subreaper for its children (like a service manager)
532 * 3. give it to the init process (PID 1) in our pid namespace
534 static struct task_struct *find_new_reaper(struct task_struct *father,
535 struct task_struct *child_reaper)
537 struct task_struct *thread, *reaper;
539 thread = find_alive_thread(father);
543 if (father->signal->has_child_subreaper) {
544 unsigned int ns_level = task_pid(father)->level;
546 * Find the first ->is_child_subreaper ancestor in our pid_ns.
547 * We can't check reaper != child_reaper to ensure we do not
548 * cross the namespaces, the exiting parent could be injected
549 * by setns() + fork().
550 * We check pid->level, this is slightly more efficient than
551 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
553 for (reaper = father->real_parent;
554 task_pid(reaper)->level == ns_level;
555 reaper = reaper->real_parent) {
556 if (reaper == &init_task)
558 if (!reaper->signal->is_child_subreaper)
560 thread = find_alive_thread(reaper);
570 * Any that need to be release_task'd are put on the @dead list.
572 static void reparent_leader(struct task_struct *father, struct task_struct *p,
573 struct list_head *dead)
575 if (unlikely(p->exit_state == EXIT_DEAD))
578 /* We don't want people slaying init. */
579 p->exit_signal = SIGCHLD;
581 /* If it has exited notify the new parent about this child's death. */
583 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
584 if (do_notify_parent(p, p->exit_signal)) {
585 p->exit_state = EXIT_DEAD;
586 list_add(&p->ptrace_entry, dead);
590 kill_orphaned_pgrp(p, father);
594 * This does two things:
596 * A. Make init inherit all the child processes
597 * B. Check to see if any process groups have become orphaned
598 * as a result of our exiting, and if they have any stopped
599 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
601 static void forget_original_parent(struct task_struct *father,
602 struct list_head *dead)
604 struct task_struct *p, *t, *reaper;
606 if (unlikely(!list_empty(&father->ptraced)))
607 exit_ptrace(father, dead);
609 /* Can drop and reacquire tasklist_lock */
610 reaper = find_child_reaper(father, dead);
611 if (list_empty(&father->children))
614 reaper = find_new_reaper(father, reaper);
615 list_for_each_entry(p, &father->children, sibling) {
616 for_each_thread(p, t) {
617 RCU_INIT_POINTER(t->real_parent, reaper);
618 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
619 if (likely(!t->ptrace))
620 t->parent = t->real_parent;
621 if (t->pdeath_signal)
622 group_send_sig_info(t->pdeath_signal,
627 * If this is a threaded reparent there is no need to
628 * notify anyone anything has happened.
630 if (!same_thread_group(reaper, father))
631 reparent_leader(father, p, dead);
633 list_splice_tail_init(&father->children, &reaper->children);
637 * Send signals to all our closest relatives so that they know
638 * to properly mourn us..
640 static void exit_notify(struct task_struct *tsk, int group_dead)
643 struct task_struct *p, *n;
646 write_lock_irq(&tasklist_lock);
647 forget_original_parent(tsk, &dead);
650 kill_orphaned_pgrp(tsk->group_leader, NULL);
652 tsk->exit_state = EXIT_ZOMBIE;
653 if (unlikely(tsk->ptrace)) {
654 int sig = thread_group_leader(tsk) &&
655 thread_group_empty(tsk) &&
656 !ptrace_reparented(tsk) ?
657 tsk->exit_signal : SIGCHLD;
658 autoreap = do_notify_parent(tsk, sig);
659 } else if (thread_group_leader(tsk)) {
660 autoreap = thread_group_empty(tsk) &&
661 do_notify_parent(tsk, tsk->exit_signal);
667 tsk->exit_state = EXIT_DEAD;
668 list_add(&tsk->ptrace_entry, &dead);
671 /* mt-exec, de_thread() is waiting for group leader */
672 if (unlikely(tsk->signal->notify_count < 0))
673 wake_up_process(tsk->signal->group_exit_task);
674 write_unlock_irq(&tasklist_lock);
676 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
677 list_del_init(&p->ptrace_entry);
682 #ifdef CONFIG_DEBUG_STACK_USAGE
683 static void check_stack_usage(void)
685 static DEFINE_SPINLOCK(low_water_lock);
686 static int lowest_to_date = THREAD_SIZE;
689 free = stack_not_used(current);
691 if (free >= lowest_to_date)
694 spin_lock(&low_water_lock);
695 if (free < lowest_to_date) {
696 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
697 current->comm, task_pid_nr(current), free);
698 lowest_to_date = free;
700 spin_unlock(&low_water_lock);
703 static inline void check_stack_usage(void) {}
706 void __noreturn do_exit(long code)
708 struct task_struct *tsk = current;
711 profile_task_exit(tsk);
714 WARN_ON(blk_needs_flush_plug(tsk));
716 if (unlikely(in_interrupt()))
717 panic("Aiee, killing interrupt handler!");
718 if (unlikely(!tsk->pid))
719 panic("Attempted to kill the idle task!");
722 * If do_exit is called because this processes oopsed, it's possible
723 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
724 * continuing. Amongst other possible reasons, this is to prevent
725 * mm_release()->clear_child_tid() from writing to a user-controlled
730 ptrace_event(PTRACE_EVENT_EXIT, code);
732 validate_creds_for_do_exit(tsk);
735 * We're taking recursive faults here in do_exit. Safest is to just
736 * leave this task alone and wait for reboot.
738 if (unlikely(tsk->flags & PF_EXITING)) {
739 pr_alert("Fixing recursive fault but reboot is needed!\n");
740 futex_exit_recursive(tsk);
741 set_current_state(TASK_UNINTERRUPTIBLE);
745 exit_signals(tsk); /* sets PF_EXITING */
747 if (unlikely(in_atomic())) {
748 pr_info("note: %s[%d] exited with preempt_count %d\n",
749 current->comm, task_pid_nr(current),
751 preempt_count_set(PREEMPT_ENABLED);
754 /* sync mm's RSS info before statistics gathering */
756 sync_mm_rss(tsk->mm);
757 acct_update_integrals(tsk);
758 group_dead = atomic_dec_and_test(&tsk->signal->live);
761 * If the last thread of global init has exited, panic
762 * immediately to get a useable coredump.
764 if (unlikely(is_global_init(tsk)))
765 panic("Attempted to kill init! exitcode=0x%08x\n",
766 tsk->signal->group_exit_code ?: (int)code);
768 #ifdef CONFIG_POSIX_TIMERS
769 hrtimer_cancel(&tsk->signal->real_timer);
770 exit_itimers(tsk->signal);
773 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
775 acct_collect(code, group_dead);
780 tsk->exit_code = code;
781 taskstats_exit(tsk, group_dead);
787 trace_sched_process_exit(tsk);
794 disassociate_ctty(1);
795 exit_task_namespaces(tsk);
801 * Flush inherited counters to the parent - before the parent
802 * gets woken up by child-exit notifications.
804 * because of cgroup mode, must be called before cgroup_exit()
806 perf_event_exit_task(tsk);
808 sched_autogroup_exit_task(tsk);
812 * FIXME: do that only when needed, using sched_exit tracepoint
814 flush_ptrace_hw_breakpoint(tsk);
816 exit_tasks_rcu_start();
817 exit_notify(tsk, group_dead);
818 proc_exit_connector(tsk);
819 mpol_put_task_policy(tsk);
821 if (unlikely(current->pi_state_cache))
822 kfree(current->pi_state_cache);
825 * Make sure we are holding no locks:
827 debug_check_no_locks_held();
830 exit_io_context(tsk);
832 if (tsk->splice_pipe)
833 free_pipe_info(tsk->splice_pipe);
835 if (tsk->task_frag.page)
836 put_page(tsk->task_frag.page);
838 validate_creds_for_do_exit(tsk);
843 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
845 exit_tasks_rcu_finish();
847 lockdep_free_task(tsk);
850 EXPORT_SYMBOL_GPL(do_exit);
852 void complete_and_exit(struct completion *comp, long code)
859 EXPORT_SYMBOL(complete_and_exit);
861 SYSCALL_DEFINE1(exit, int, error_code)
863 do_exit((error_code&0xff)<<8);
867 * Take down every thread in the group. This is called by fatal signals
868 * as well as by sys_exit_group (below).
871 do_group_exit(int exit_code)
873 struct signal_struct *sig = current->signal;
875 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
877 if (signal_group_exit(sig))
878 exit_code = sig->group_exit_code;
879 else if (!thread_group_empty(current)) {
880 struct sighand_struct *const sighand = current->sighand;
882 spin_lock_irq(&sighand->siglock);
883 if (signal_group_exit(sig))
884 /* Another thread got here before we took the lock. */
885 exit_code = sig->group_exit_code;
887 sig->group_exit_code = exit_code;
888 sig->flags = SIGNAL_GROUP_EXIT;
889 zap_other_threads(current);
891 spin_unlock_irq(&sighand->siglock);
899 * this kills every thread in the thread group. Note that any externally
900 * wait4()-ing process will get the correct exit code - even if this
901 * thread is not the thread group leader.
903 SYSCALL_DEFINE1(exit_group, int, error_code)
905 do_group_exit((error_code & 0xff) << 8);
918 enum pid_type wo_type;
922 struct waitid_info *wo_info;
924 struct rusage *wo_rusage;
926 wait_queue_entry_t child_wait;
930 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
932 return wo->wo_type == PIDTYPE_MAX ||
933 task_pid_type(p, wo->wo_type) == wo->wo_pid;
937 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
939 if (!eligible_pid(wo, p))
943 * Wait for all children (clone and not) if __WALL is set or
944 * if it is traced by us.
946 if (ptrace || (wo->wo_flags & __WALL))
950 * Otherwise, wait for clone children *only* if __WCLONE is set;
951 * otherwise, wait for non-clone children *only*.
953 * Note: a "clone" child here is one that reports to its parent
954 * using a signal other than SIGCHLD, or a non-leader thread which
955 * we can only see if it is traced by us.
957 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
964 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
965 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
966 * the lock and this task is uninteresting. If we return nonzero, we have
967 * released the lock and the system call should return.
969 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
972 pid_t pid = task_pid_vnr(p);
973 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
974 struct waitid_info *infop;
976 if (!likely(wo->wo_flags & WEXITED))
979 if (unlikely(wo->wo_flags & WNOWAIT)) {
980 status = p->exit_code;
982 read_unlock(&tasklist_lock);
983 sched_annotate_sleep();
985 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
990 * Move the task's state to DEAD/TRACE, only one thread can do this.
992 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
993 EXIT_TRACE : EXIT_DEAD;
994 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
997 * We own this thread, nobody else can reap it.
999 read_unlock(&tasklist_lock);
1000 sched_annotate_sleep();
1003 * Check thread_group_leader() to exclude the traced sub-threads.
1005 if (state == EXIT_DEAD && thread_group_leader(p)) {
1006 struct signal_struct *sig = p->signal;
1007 struct signal_struct *psig = current->signal;
1008 unsigned long maxrss;
1009 u64 tgutime, tgstime;
1012 * The resource counters for the group leader are in its
1013 * own task_struct. Those for dead threads in the group
1014 * are in its signal_struct, as are those for the child
1015 * processes it has previously reaped. All these
1016 * accumulate in the parent's signal_struct c* fields.
1018 * We don't bother to take a lock here to protect these
1019 * p->signal fields because the whole thread group is dead
1020 * and nobody can change them.
1022 * psig->stats_lock also protects us from our sub-theads
1023 * which can reap other children at the same time. Until
1024 * we change k_getrusage()-like users to rely on this lock
1025 * we have to take ->siglock as well.
1027 * We use thread_group_cputime_adjusted() to get times for
1028 * the thread group, which consolidates times for all threads
1029 * in the group including the group leader.
1031 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1032 spin_lock_irq(¤t->sighand->siglock);
1033 write_seqlock(&psig->stats_lock);
1034 psig->cutime += tgutime + sig->cutime;
1035 psig->cstime += tgstime + sig->cstime;
1036 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1038 p->min_flt + sig->min_flt + sig->cmin_flt;
1040 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1042 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1044 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1046 task_io_get_inblock(p) +
1047 sig->inblock + sig->cinblock;
1049 task_io_get_oublock(p) +
1050 sig->oublock + sig->coublock;
1051 maxrss = max(sig->maxrss, sig->cmaxrss);
1052 if (psig->cmaxrss < maxrss)
1053 psig->cmaxrss = maxrss;
1054 task_io_accounting_add(&psig->ioac, &p->ioac);
1055 task_io_accounting_add(&psig->ioac, &sig->ioac);
1056 write_sequnlock(&psig->stats_lock);
1057 spin_unlock_irq(¤t->sighand->siglock);
1061 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1062 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1063 ? p->signal->group_exit_code : p->exit_code;
1064 wo->wo_stat = status;
1066 if (state == EXIT_TRACE) {
1067 write_lock_irq(&tasklist_lock);
1068 /* We dropped tasklist, ptracer could die and untrace */
1071 /* If parent wants a zombie, don't release it now */
1072 state = EXIT_ZOMBIE;
1073 if (do_notify_parent(p, p->exit_signal))
1075 p->exit_state = state;
1076 write_unlock_irq(&tasklist_lock);
1078 if (state == EXIT_DEAD)
1082 infop = wo->wo_info;
1084 if ((status & 0x7f) == 0) {
1085 infop->cause = CLD_EXITED;
1086 infop->status = status >> 8;
1088 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1089 infop->status = status & 0x7f;
1098 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1101 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1102 return &p->exit_code;
1104 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1105 return &p->signal->group_exit_code;
1111 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1113 * @ptrace: is the wait for ptrace
1114 * @p: task to wait for
1116 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1119 * read_lock(&tasklist_lock), which is released if return value is
1120 * non-zero. Also, grabs and releases @p->sighand->siglock.
1123 * 0 if wait condition didn't exist and search for other wait conditions
1124 * should continue. Non-zero return, -errno on failure and @p's pid on
1125 * success, implies that tasklist_lock is released and wait condition
1126 * search should terminate.
1128 static int wait_task_stopped(struct wait_opts *wo,
1129 int ptrace, struct task_struct *p)
1131 struct waitid_info *infop;
1132 int exit_code, *p_code, why;
1133 uid_t uid = 0; /* unneeded, required by compiler */
1137 * Traditionally we see ptrace'd stopped tasks regardless of options.
1139 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1142 if (!task_stopped_code(p, ptrace))
1146 spin_lock_irq(&p->sighand->siglock);
1148 p_code = task_stopped_code(p, ptrace);
1149 if (unlikely(!p_code))
1152 exit_code = *p_code;
1156 if (!unlikely(wo->wo_flags & WNOWAIT))
1159 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1161 spin_unlock_irq(&p->sighand->siglock);
1166 * Now we are pretty sure this task is interesting.
1167 * Make sure it doesn't get reaped out from under us while we
1168 * give up the lock and then examine it below. We don't want to
1169 * keep holding onto the tasklist_lock while we call getrusage and
1170 * possibly take page faults for user memory.
1173 pid = task_pid_vnr(p);
1174 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1175 read_unlock(&tasklist_lock);
1176 sched_annotate_sleep();
1178 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1181 if (likely(!(wo->wo_flags & WNOWAIT)))
1182 wo->wo_stat = (exit_code << 8) | 0x7f;
1184 infop = wo->wo_info;
1187 infop->status = exit_code;
1195 * Handle do_wait work for one task in a live, non-stopped state.
1196 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1197 * the lock and this task is uninteresting. If we return nonzero, we have
1198 * released the lock and the system call should return.
1200 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1202 struct waitid_info *infop;
1206 if (!unlikely(wo->wo_flags & WCONTINUED))
1209 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1212 spin_lock_irq(&p->sighand->siglock);
1213 /* Re-check with the lock held. */
1214 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1215 spin_unlock_irq(&p->sighand->siglock);
1218 if (!unlikely(wo->wo_flags & WNOWAIT))
1219 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1220 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1221 spin_unlock_irq(&p->sighand->siglock);
1223 pid = task_pid_vnr(p);
1225 read_unlock(&tasklist_lock);
1226 sched_annotate_sleep();
1228 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1231 infop = wo->wo_info;
1233 wo->wo_stat = 0xffff;
1235 infop->cause = CLD_CONTINUED;
1238 infop->status = SIGCONT;
1244 * Consider @p for a wait by @parent.
1246 * -ECHILD should be in ->notask_error before the first call.
1247 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1248 * Returns zero if the search for a child should continue;
1249 * then ->notask_error is 0 if @p is an eligible child,
1252 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1253 struct task_struct *p)
1256 * We can race with wait_task_zombie() from another thread.
1257 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1258 * can't confuse the checks below.
1260 int exit_state = READ_ONCE(p->exit_state);
1263 if (unlikely(exit_state == EXIT_DEAD))
1266 ret = eligible_child(wo, ptrace, p);
1270 if (unlikely(exit_state == EXIT_TRACE)) {
1272 * ptrace == 0 means we are the natural parent. In this case
1273 * we should clear notask_error, debugger will notify us.
1275 if (likely(!ptrace))
1276 wo->notask_error = 0;
1280 if (likely(!ptrace) && unlikely(p->ptrace)) {
1282 * If it is traced by its real parent's group, just pretend
1283 * the caller is ptrace_do_wait() and reap this child if it
1286 * This also hides group stop state from real parent; otherwise
1287 * a single stop can be reported twice as group and ptrace stop.
1288 * If a ptracer wants to distinguish these two events for its
1289 * own children it should create a separate process which takes
1290 * the role of real parent.
1292 if (!ptrace_reparented(p))
1297 if (exit_state == EXIT_ZOMBIE) {
1298 /* we don't reap group leaders with subthreads */
1299 if (!delay_group_leader(p)) {
1301 * A zombie ptracee is only visible to its ptracer.
1302 * Notification and reaping will be cascaded to the
1303 * real parent when the ptracer detaches.
1305 if (unlikely(ptrace) || likely(!p->ptrace))
1306 return wait_task_zombie(wo, p);
1310 * Allow access to stopped/continued state via zombie by
1311 * falling through. Clearing of notask_error is complex.
1315 * If WEXITED is set, notask_error should naturally be
1316 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1317 * so, if there are live subthreads, there are events to
1318 * wait for. If all subthreads are dead, it's still safe
1319 * to clear - this function will be called again in finite
1320 * amount time once all the subthreads are released and
1321 * will then return without clearing.
1325 * Stopped state is per-task and thus can't change once the
1326 * target task dies. Only continued and exited can happen.
1327 * Clear notask_error if WCONTINUED | WEXITED.
1329 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1330 wo->notask_error = 0;
1333 * @p is alive and it's gonna stop, continue or exit, so
1334 * there always is something to wait for.
1336 wo->notask_error = 0;
1340 * Wait for stopped. Depending on @ptrace, different stopped state
1341 * is used and the two don't interact with each other.
1343 ret = wait_task_stopped(wo, ptrace, p);
1348 * Wait for continued. There's only one continued state and the
1349 * ptracer can consume it which can confuse the real parent. Don't
1350 * use WCONTINUED from ptracer. You don't need or want it.
1352 return wait_task_continued(wo, p);
1356 * Do the work of do_wait() for one thread in the group, @tsk.
1358 * -ECHILD should be in ->notask_error before the first call.
1359 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1360 * Returns zero if the search for a child should continue; then
1361 * ->notask_error is 0 if there were any eligible children,
1364 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1366 struct task_struct *p;
1368 list_for_each_entry(p, &tsk->children, sibling) {
1369 int ret = wait_consider_task(wo, 0, p);
1378 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1380 struct task_struct *p;
1382 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1383 int ret = wait_consider_task(wo, 1, p);
1392 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1393 int sync, void *key)
1395 struct wait_opts *wo = container_of(wait, struct wait_opts,
1397 struct task_struct *p = key;
1399 if (!eligible_pid(wo, p))
1402 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1405 return default_wake_function(wait, mode, sync, key);
1408 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1410 __wake_up_sync_key(&parent->signal->wait_chldexit,
1411 TASK_INTERRUPTIBLE, p);
1414 static long do_wait(struct wait_opts *wo)
1416 struct task_struct *tsk;
1419 trace_sched_process_wait(wo->wo_pid);
1421 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1422 wo->child_wait.private = current;
1423 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1426 * If there is nothing that can match our criteria, just get out.
1427 * We will clear ->notask_error to zero if we see any child that
1428 * might later match our criteria, even if we are not able to reap
1431 wo->notask_error = -ECHILD;
1432 if ((wo->wo_type < PIDTYPE_MAX) &&
1433 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1436 set_current_state(TASK_INTERRUPTIBLE);
1437 read_lock(&tasklist_lock);
1440 retval = do_wait_thread(wo, tsk);
1444 retval = ptrace_do_wait(wo, tsk);
1448 if (wo->wo_flags & __WNOTHREAD)
1450 } while_each_thread(current, tsk);
1451 read_unlock(&tasklist_lock);
1454 retval = wo->notask_error;
1455 if (!retval && !(wo->wo_flags & WNOHANG)) {
1456 retval = -ERESTARTSYS;
1457 if (!signal_pending(current)) {
1463 __set_current_state(TASK_RUNNING);
1464 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1468 static struct pid *pidfd_get_pid(unsigned int fd)
1475 return ERR_PTR(-EBADF);
1477 pid = pidfd_pid(f.file);
1485 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1486 int options, struct rusage *ru)
1488 struct wait_opts wo;
1489 struct pid *pid = NULL;
1493 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1494 __WNOTHREAD|__WCLONE|__WALL))
1496 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1508 pid = find_get_pid(upid);
1511 type = PIDTYPE_PGID;
1516 pid = find_get_pid(upid);
1518 pid = get_task_pid(current, PIDTYPE_PGID);
1525 pid = pidfd_get_pid(upid);
1527 return PTR_ERR(pid);
1535 wo.wo_flags = options;
1544 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1545 infop, int, options, struct rusage __user *, ru)
1548 struct waitid_info info = {.status = 0};
1549 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1555 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1561 if (!user_access_begin(infop, sizeof(*infop)))
1564 unsafe_put_user(signo, &infop->si_signo, Efault);
1565 unsafe_put_user(0, &infop->si_errno, Efault);
1566 unsafe_put_user(info.cause, &infop->si_code, Efault);
1567 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1568 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1569 unsafe_put_user(info.status, &infop->si_status, Efault);
1577 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1580 struct wait_opts wo;
1581 struct pid *pid = NULL;
1585 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1586 __WNOTHREAD|__WCLONE|__WALL))
1589 /* -INT_MIN is not defined */
1590 if (upid == INT_MIN)
1595 else if (upid < 0) {
1596 type = PIDTYPE_PGID;
1597 pid = find_get_pid(-upid);
1598 } else if (upid == 0) {
1599 type = PIDTYPE_PGID;
1600 pid = get_task_pid(current, PIDTYPE_PGID);
1601 } else /* upid > 0 */ {
1603 pid = find_get_pid(upid);
1608 wo.wo_flags = options | WEXITED;
1614 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1620 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1621 int, options, struct rusage __user *, ru)
1624 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1627 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1633 #ifdef __ARCH_WANT_SYS_WAITPID
1636 * sys_waitpid() remains for compatibility. waitpid() should be
1637 * implemented by calling sys_wait4() from libc.a.
1639 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1641 return kernel_wait4(pid, stat_addr, options, NULL);
1646 #ifdef CONFIG_COMPAT
1647 COMPAT_SYSCALL_DEFINE4(wait4,
1649 compat_uint_t __user *, stat_addr,
1651 struct compat_rusage __user *, ru)
1654 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1656 if (ru && put_compat_rusage(&r, ru))
1662 COMPAT_SYSCALL_DEFINE5(waitid,
1663 int, which, compat_pid_t, pid,
1664 struct compat_siginfo __user *, infop, int, options,
1665 struct compat_rusage __user *, uru)
1668 struct waitid_info info = {.status = 0};
1669 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1675 /* kernel_waitid() overwrites everything in ru */
1676 if (COMPAT_USE_64BIT_TIME)
1677 err = copy_to_user(uru, &ru, sizeof(ru));
1679 err = put_compat_rusage(&ru, uru);
1688 if (!user_access_begin(infop, sizeof(*infop)))
1691 unsafe_put_user(signo, &infop->si_signo, Efault);
1692 unsafe_put_user(0, &infop->si_errno, Efault);
1693 unsafe_put_user(info.cause, &infop->si_code, Efault);
1694 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1695 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1696 unsafe_put_user(info.status, &infop->si_status, Efault);
1705 __weak void abort(void)
1709 /* if that doesn't kill us, halt */
1710 panic("Oops failed to kill thread");
1712 EXPORT_SYMBOL(abort);