4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/freezer.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
56 #include <linux/kcov.h>
58 #include <asm/uaccess.h>
59 #include <asm/unistd.h>
60 #include <asm/pgtable.h>
61 #include <asm/mmu_context.h>
63 static void __unhash_process(struct task_struct *p, bool group_dead)
66 detach_pid(p, PIDTYPE_PID);
68 detach_pid(p, PIDTYPE_PGID);
69 detach_pid(p, PIDTYPE_SID);
71 list_del_rcu(&p->tasks);
72 list_del_init(&p->sibling);
73 __this_cpu_dec(process_counts);
75 list_del_rcu(&p->thread_group);
76 list_del_rcu(&p->thread_node);
80 * This function expects the tasklist_lock write-locked.
82 static void __exit_signal(struct task_struct *tsk)
84 struct signal_struct *sig = tsk->signal;
85 bool group_dead = thread_group_leader(tsk);
86 struct sighand_struct *sighand;
87 struct tty_struct *uninitialized_var(tty);
88 cputime_t utime, stime;
90 sighand = rcu_dereference_check(tsk->sighand,
91 lockdep_tasklist_lock_is_held());
92 spin_lock(&sighand->siglock);
94 posix_cpu_timers_exit(tsk);
96 posix_cpu_timers_exit_group(tsk);
101 * This can only happen if the caller is de_thread().
102 * FIXME: this is the temporary hack, we should teach
103 * posix-cpu-timers to handle this case correctly.
105 if (unlikely(has_group_leader_pid(tsk)))
106 posix_cpu_timers_exit_group(tsk);
109 * If there is any task waiting for the group exit
112 if (sig->notify_count > 0 && !--sig->notify_count)
113 wake_up_process(sig->group_exit_task);
115 if (tsk == sig->curr_target)
116 sig->curr_target = next_thread(tsk);
120 * Accumulate here the counters for all threads as they die. We could
121 * skip the group leader because it is the last user of signal_struct,
122 * but we want to avoid the race with thread_group_cputime() which can
123 * see the empty ->thread_head list.
125 task_cputime(tsk, &utime, &stime);
126 write_seqlock(&sig->stats_lock);
129 sig->gtime += task_gtime(tsk);
130 sig->min_flt += tsk->min_flt;
131 sig->maj_flt += tsk->maj_flt;
132 sig->nvcsw += tsk->nvcsw;
133 sig->nivcsw += tsk->nivcsw;
134 sig->inblock += task_io_get_inblock(tsk);
135 sig->oublock += task_io_get_oublock(tsk);
136 task_io_accounting_add(&sig->ioac, &tsk->ioac);
137 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
139 __unhash_process(tsk, group_dead);
140 write_sequnlock(&sig->stats_lock);
143 * Do this under ->siglock, we can race with another thread
144 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
146 flush_sigqueue(&tsk->pending);
148 spin_unlock(&sighand->siglock);
150 __cleanup_sighand(sighand);
151 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
153 flush_sigqueue(&sig->shared_pending);
158 static void delayed_put_task_struct(struct rcu_head *rhp)
160 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
162 perf_event_delayed_put(tsk);
163 trace_sched_process_free(tsk);
164 put_task_struct(tsk);
168 void release_task(struct task_struct *p)
170 struct task_struct *leader;
173 /* don't need to get the RCU readlock here - the process is dead and
174 * can't be modifying its own credentials. But shut RCU-lockdep up */
176 atomic_dec(&__task_cred(p)->user->processes);
181 write_lock_irq(&tasklist_lock);
182 ptrace_release_task(p);
186 * If we are the last non-leader member of the thread
187 * group, and the leader is zombie, then notify the
188 * group leader's parent process. (if it wants notification.)
191 leader = p->group_leader;
192 if (leader != p && thread_group_empty(leader)
193 && leader->exit_state == EXIT_ZOMBIE) {
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
199 zap_leader = do_notify_parent(leader, leader->exit_signal);
201 leader->exit_state = EXIT_DEAD;
204 write_unlock_irq(&tasklist_lock);
206 call_rcu(&p->rcu, delayed_put_task_struct);
209 if (unlikely(zap_leader))
214 * Note that if this function returns a valid task_struct pointer (!NULL)
215 * task->usage must remain >0 for the duration of the RCU critical section.
217 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
219 struct sighand_struct *sighand;
220 struct task_struct *task;
223 * We need to verify that release_task() was not called and thus
224 * delayed_put_task_struct() can't run and drop the last reference
225 * before rcu_read_unlock(). We check task->sighand != NULL,
226 * but we can read the already freed and reused memory.
229 task = rcu_dereference(*ptask);
233 probe_kernel_address(&task->sighand, sighand);
236 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
237 * was already freed we can not miss the preceding update of this
241 if (unlikely(task != READ_ONCE(*ptask)))
245 * We've re-checked that "task == *ptask", now we have two different
248 * 1. This is actually the same task/task_struct. In this case
249 * sighand != NULL tells us it is still alive.
251 * 2. This is another task which got the same memory for task_struct.
252 * We can't know this of course, and we can not trust
255 * In this case we actually return a random value, but this is
258 * If we return NULL - we can pretend that we actually noticed that
259 * *ptask was updated when the previous task has exited. Or pretend
260 * that probe_slab_address(&sighand) reads NULL.
262 * If we return the new task (because sighand is not NULL for any
263 * reason) - this is fine too. This (new) task can't go away before
266 * And note: We could even eliminate the false positive if re-read
267 * task->sighand once again to avoid the falsely NULL. But this case
268 * is very unlikely so we don't care.
276 struct task_struct *try_get_task_struct(struct task_struct **ptask)
278 struct task_struct *task;
281 task = task_rcu_dereference(ptask);
283 get_task_struct(task);
290 * Determine if a process group is "orphaned", according to the POSIX
291 * definition in 2.2.2.52. Orphaned process groups are not to be affected
292 * by terminal-generated stop signals. Newly orphaned process groups are
293 * to receive a SIGHUP and a SIGCONT.
295 * "I ask you, have you ever known what it is to be an orphan?"
297 static int will_become_orphaned_pgrp(struct pid *pgrp,
298 struct task_struct *ignored_task)
300 struct task_struct *p;
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if ((p == ignored_task) ||
304 (p->exit_state && thread_group_empty(p)) ||
305 is_global_init(p->real_parent))
308 if (task_pgrp(p->real_parent) != pgrp &&
309 task_session(p->real_parent) == task_session(p))
311 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
316 int is_current_pgrp_orphaned(void)
320 read_lock(&tasklist_lock);
321 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
322 read_unlock(&tasklist_lock);
327 static bool has_stopped_jobs(struct pid *pgrp)
329 struct task_struct *p;
331 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 if (p->signal->flags & SIGNAL_STOP_STOPPED)
334 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
340 * Check to see if any process groups have become orphaned as
341 * a result of our exiting, and if they have any stopped jobs,
342 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
345 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
347 struct pid *pgrp = task_pgrp(tsk);
348 struct task_struct *ignored_task = tsk;
351 /* exit: our father is in a different pgrp than
352 * we are and we were the only connection outside.
354 parent = tsk->real_parent;
356 /* reparent: our child is in a different pgrp than
357 * we are, and it was the only connection outside.
361 if (task_pgrp(parent) != pgrp &&
362 task_session(parent) == task_session(tsk) &&
363 will_become_orphaned_pgrp(pgrp, ignored_task) &&
364 has_stopped_jobs(pgrp)) {
365 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
366 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
372 * A task is exiting. If it owned this mm, find a new owner for the mm.
374 void mm_update_next_owner(struct mm_struct *mm)
376 struct task_struct *c, *g, *p = current;
380 * If the exiting or execing task is not the owner, it's
381 * someone else's problem.
386 * The current owner is exiting/execing and there are no other
387 * candidates. Do not leave the mm pointing to a possibly
388 * freed task structure.
390 if (atomic_read(&mm->mm_users) <= 1) {
395 read_lock(&tasklist_lock);
397 * Search in the children
399 list_for_each_entry(c, &p->children, sibling) {
401 goto assign_new_owner;
405 * Search in the siblings
407 list_for_each_entry(c, &p->real_parent->children, sibling) {
409 goto assign_new_owner;
413 * Search through everything else, we should not get here often.
415 for_each_process(g) {
416 if (g->flags & PF_KTHREAD)
418 for_each_thread(g, c) {
420 goto assign_new_owner;
425 read_unlock(&tasklist_lock);
427 * We found no owner yet mm_users > 1: this implies that we are
428 * most likely racing with swapoff (try_to_unuse()) or /proc or
429 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
438 * The task_lock protects c->mm from changing.
439 * We always want mm->owner->mm == mm
443 * Delay read_unlock() till we have the task_lock()
444 * to ensure that c does not slip away underneath us
446 read_unlock(&tasklist_lock);
456 #endif /* CONFIG_MEMCG */
459 * Turn us into a lazy TLB process if we
462 static void exit_mm(struct task_struct *tsk)
464 struct mm_struct *mm = tsk->mm;
465 struct core_state *core_state;
467 exit_mm_release(tsk, mm);
472 * Serialize with any possible pending coredump.
473 * We must hold mmap_sem around checking core_state
474 * and clearing tsk->mm. The core-inducing thread
475 * will increment ->nr_threads for each thread in the
476 * group with ->mm != NULL.
478 down_read(&mm->mmap_sem);
479 core_state = mm->core_state;
481 struct core_thread self;
483 up_read(&mm->mmap_sem);
486 if (self.task->flags & PF_SIGNALED)
487 self.next = xchg(&core_state->dumper.next, &self);
491 * Implies mb(), the result of xchg() must be visible
492 * to core_state->dumper.
494 if (atomic_dec_and_test(&core_state->nr_threads))
495 complete(&core_state->startup);
498 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
499 if (!self.task) /* see coredump_finish() */
501 freezable_schedule();
503 __set_task_state(tsk, TASK_RUNNING);
504 down_read(&mm->mmap_sem);
506 atomic_inc(&mm->mm_count);
507 BUG_ON(mm != tsk->active_mm);
508 /* more a memory barrier than a real lock */
511 up_read(&mm->mmap_sem);
512 enter_lazy_tlb(mm, current);
514 mm_update_next_owner(mm);
516 if (test_thread_flag(TIF_MEMDIE))
520 static struct task_struct *find_alive_thread(struct task_struct *p)
522 struct task_struct *t;
524 for_each_thread(p, t) {
525 if (!(t->flags & PF_EXITING))
531 static struct task_struct *find_child_reaper(struct task_struct *father,
532 struct list_head *dead)
533 __releases(&tasklist_lock)
534 __acquires(&tasklist_lock)
536 struct pid_namespace *pid_ns = task_active_pid_ns(father);
537 struct task_struct *reaper = pid_ns->child_reaper;
538 struct task_struct *p, *n;
540 if (likely(reaper != father))
543 reaper = find_alive_thread(father);
545 pid_ns->child_reaper = reaper;
549 write_unlock_irq(&tasklist_lock);
550 if (unlikely(pid_ns == &init_pid_ns)) {
551 panic("Attempted to kill init! exitcode=0x%08x\n",
552 father->signal->group_exit_code ?: father->exit_code);
555 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
556 list_del_init(&p->ptrace_entry);
560 zap_pid_ns_processes(pid_ns);
561 write_lock_irq(&tasklist_lock);
567 * When we die, we re-parent all our children, and try to:
568 * 1. give them to another thread in our thread group, if such a member exists
569 * 2. give it to the first ancestor process which prctl'd itself as a
570 * child_subreaper for its children (like a service manager)
571 * 3. give it to the init process (PID 1) in our pid namespace
573 static struct task_struct *find_new_reaper(struct task_struct *father,
574 struct task_struct *child_reaper)
576 struct task_struct *thread, *reaper;
578 thread = find_alive_thread(father);
582 if (father->signal->has_child_subreaper) {
584 * Find the first ->is_child_subreaper ancestor in our pid_ns.
585 * We start from father to ensure we can not look into another
586 * namespace, this is safe because all its threads are dead.
588 for (reaper = father;
589 !same_thread_group(reaper, child_reaper);
590 reaper = reaper->real_parent) {
591 /* call_usermodehelper() descendants need this check */
592 if (reaper == &init_task)
594 if (!reaper->signal->is_child_subreaper)
596 thread = find_alive_thread(reaper);
606 * Any that need to be release_task'd are put on the @dead list.
608 static void reparent_leader(struct task_struct *father, struct task_struct *p,
609 struct list_head *dead)
611 if (unlikely(p->exit_state == EXIT_DEAD))
614 /* We don't want people slaying init. */
615 p->exit_signal = SIGCHLD;
617 /* If it has exited notify the new parent about this child's death. */
619 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
620 if (do_notify_parent(p, p->exit_signal)) {
621 p->exit_state = EXIT_DEAD;
622 list_add(&p->ptrace_entry, dead);
626 kill_orphaned_pgrp(p, father);
630 * This does two things:
632 * A. Make init inherit all the child processes
633 * B. Check to see if any process groups have become orphaned
634 * as a result of our exiting, and if they have any stopped
635 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
637 static void forget_original_parent(struct task_struct *father,
638 struct list_head *dead)
640 struct task_struct *p, *t, *reaper;
642 if (unlikely(!list_empty(&father->ptraced)))
643 exit_ptrace(father, dead);
645 /* Can drop and reacquire tasklist_lock */
646 reaper = find_child_reaper(father, dead);
647 if (list_empty(&father->children))
650 reaper = find_new_reaper(father, reaper);
651 list_for_each_entry(p, &father->children, sibling) {
652 for_each_thread(p, t) {
653 t->real_parent = reaper;
654 BUG_ON((!t->ptrace) != (t->parent == father));
655 if (likely(!t->ptrace))
656 t->parent = t->real_parent;
657 if (t->pdeath_signal)
658 group_send_sig_info(t->pdeath_signal,
662 * If this is a threaded reparent there is no need to
663 * notify anyone anything has happened.
665 if (!same_thread_group(reaper, father))
666 reparent_leader(father, p, dead);
668 list_splice_tail_init(&father->children, &reaper->children);
672 * Send signals to all our closest relatives so that they know
673 * to properly mourn us..
675 static void exit_notify(struct task_struct *tsk, int group_dead)
678 struct task_struct *p, *n;
681 write_lock_irq(&tasklist_lock);
682 forget_original_parent(tsk, &dead);
685 kill_orphaned_pgrp(tsk->group_leader, NULL);
687 if (unlikely(tsk->ptrace)) {
688 int sig = thread_group_leader(tsk) &&
689 thread_group_empty(tsk) &&
690 !ptrace_reparented(tsk) ?
691 tsk->exit_signal : SIGCHLD;
692 autoreap = do_notify_parent(tsk, sig);
693 } else if (thread_group_leader(tsk)) {
694 autoreap = thread_group_empty(tsk) &&
695 do_notify_parent(tsk, tsk->exit_signal);
700 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
701 if (tsk->exit_state == EXIT_DEAD)
702 list_add(&tsk->ptrace_entry, &dead);
704 /* mt-exec, de_thread() is waiting for group leader */
705 if (unlikely(tsk->signal->notify_count < 0))
706 wake_up_process(tsk->signal->group_exit_task);
707 write_unlock_irq(&tasklist_lock);
709 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
710 list_del_init(&p->ptrace_entry);
715 #ifdef CONFIG_DEBUG_STACK_USAGE
716 static void check_stack_usage(void)
718 static DEFINE_SPINLOCK(low_water_lock);
719 static int lowest_to_date = THREAD_SIZE;
722 free = stack_not_used(current);
724 if (free >= lowest_to_date)
727 spin_lock(&low_water_lock);
728 if (free < lowest_to_date) {
729 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
730 current->comm, task_pid_nr(current), free);
731 lowest_to_date = free;
733 spin_unlock(&low_water_lock);
736 static inline void check_stack_usage(void) {}
739 void __noreturn do_exit(long code)
741 struct task_struct *tsk = current;
743 TASKS_RCU(int tasks_rcu_i);
746 * We can get here from a kernel oops, sometimes with preemption off.
747 * Start by checking for critical errors.
748 * Then fix up important state like USER_DS and preemption.
749 * Then do everything else.
752 WARN_ON(blk_needs_flush_plug(tsk));
754 if (unlikely(in_interrupt()))
755 panic("Aiee, killing interrupt handler!");
756 if (unlikely(!tsk->pid))
757 panic("Attempted to kill the idle task!");
760 * If do_exit is called because this processes oopsed, it's possible
761 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
762 * continuing. Amongst other possible reasons, this is to prevent
763 * mm_release()->clear_child_tid() from writing to a user-controlled
768 if (unlikely(in_atomic())) {
769 pr_info("note: %s[%d] exited with preempt_count %d\n",
770 current->comm, task_pid_nr(current),
772 preempt_count_set(PREEMPT_ENABLED);
775 profile_task_exit(tsk);
778 ptrace_event(PTRACE_EVENT_EXIT, code);
780 validate_creds_for_do_exit(tsk);
783 * We're taking recursive faults here in do_exit. Safest is to just
784 * leave this task alone and wait for reboot.
786 if (unlikely(tsk->flags & PF_EXITING)) {
787 pr_alert("Fixing recursive fault but reboot is needed!\n");
788 futex_exit_recursive(tsk);
789 set_current_state(TASK_UNINTERRUPTIBLE);
793 exit_signals(tsk); /* sets PF_EXITING */
795 /* sync mm's RSS info before statistics gathering */
797 sync_mm_rss(tsk->mm);
798 acct_update_integrals(tsk);
799 group_dead = atomic_dec_and_test(&tsk->signal->live);
801 hrtimer_cancel(&tsk->signal->real_timer);
802 exit_itimers(tsk->signal);
804 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
806 acct_collect(code, group_dead);
811 tsk->exit_code = code;
812 taskstats_exit(tsk, group_dead);
818 trace_sched_process_exit(tsk);
825 disassociate_ctty(1);
826 exit_task_namespaces(tsk);
831 * Flush inherited counters to the parent - before the parent
832 * gets woken up by child-exit notifications.
834 * because of cgroup mode, must be called before cgroup_exit()
836 perf_event_exit_task(tsk);
838 sched_autogroup_exit_task(tsk);
842 * FIXME: do that only when needed, using sched_exit tracepoint
844 flush_ptrace_hw_breakpoint(tsk);
846 TASKS_RCU(preempt_disable());
847 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
848 TASKS_RCU(preempt_enable());
849 exit_notify(tsk, group_dead);
850 proc_exit_connector(tsk);
851 mpol_put_task_policy(tsk);
853 if (unlikely(current->pi_state_cache))
854 kfree(current->pi_state_cache);
857 * Make sure we are holding no locks:
859 debug_check_no_locks_held();
862 exit_io_context(tsk);
864 if (tsk->splice_pipe)
865 free_pipe_info(tsk->splice_pipe);
867 if (tsk->task_frag.page)
868 put_page(tsk->task_frag.page);
870 validate_creds_for_do_exit(tsk);
875 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
877 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
881 EXPORT_SYMBOL_GPL(do_exit);
883 void complete_and_exit(struct completion *comp, long code)
890 EXPORT_SYMBOL(complete_and_exit);
892 SYSCALL_DEFINE1(exit, int, error_code)
894 do_exit((error_code&0xff)<<8);
898 * Take down every thread in the group. This is called by fatal signals
899 * as well as by sys_exit_group (below).
902 do_group_exit(int exit_code)
904 struct signal_struct *sig = current->signal;
906 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
908 if (signal_group_exit(sig))
909 exit_code = sig->group_exit_code;
910 else if (!thread_group_empty(current)) {
911 struct sighand_struct *const sighand = current->sighand;
913 spin_lock_irq(&sighand->siglock);
914 if (signal_group_exit(sig))
915 /* Another thread got here before we took the lock. */
916 exit_code = sig->group_exit_code;
918 sig->group_exit_code = exit_code;
919 sig->flags = SIGNAL_GROUP_EXIT;
920 zap_other_threads(current);
922 spin_unlock_irq(&sighand->siglock);
930 * this kills every thread in the thread group. Note that any externally
931 * wait4()-ing process will get the correct exit code - even if this
932 * thread is not the thread group leader.
934 SYSCALL_DEFINE1(exit_group, int, error_code)
936 do_group_exit((error_code & 0xff) << 8);
942 enum pid_type wo_type;
946 struct siginfo __user *wo_info;
948 struct rusage __user *wo_rusage;
950 wait_queue_t child_wait;
955 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
957 if (type != PIDTYPE_PID)
958 task = task->group_leader;
959 return task->pids[type].pid;
962 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
964 return wo->wo_type == PIDTYPE_MAX ||
965 task_pid_type(p, wo->wo_type) == wo->wo_pid;
969 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
971 if (!eligible_pid(wo, p))
975 * Wait for all children (clone and not) if __WALL is set or
976 * if it is traced by us.
978 if (ptrace || (wo->wo_flags & __WALL))
982 * Otherwise, wait for clone children *only* if __WCLONE is set;
983 * otherwise, wait for non-clone children *only*.
985 * Note: a "clone" child here is one that reports to its parent
986 * using a signal other than SIGCHLD, or a non-leader thread which
987 * we can only see if it is traced by us.
989 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
995 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
996 pid_t pid, uid_t uid, int why, int status)
998 struct siginfo __user *infop;
999 int retval = wo->wo_rusage
1000 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1003 infop = wo->wo_info;
1006 retval = put_user(SIGCHLD, &infop->si_signo);
1008 retval = put_user(0, &infop->si_errno);
1010 retval = put_user((short)why, &infop->si_code);
1012 retval = put_user(pid, &infop->si_pid);
1014 retval = put_user(uid, &infop->si_uid);
1016 retval = put_user(status, &infop->si_status);
1024 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1025 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1026 * the lock and this task is uninteresting. If we return nonzero, we have
1027 * released the lock and the system call should return.
1029 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1031 int state, retval, status;
1032 pid_t pid = task_pid_vnr(p);
1033 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1034 struct siginfo __user *infop;
1036 if (!likely(wo->wo_flags & WEXITED))
1039 if (unlikely(wo->wo_flags & WNOWAIT)) {
1040 int exit_code = p->exit_code;
1044 read_unlock(&tasklist_lock);
1045 sched_annotate_sleep();
1047 if ((exit_code & 0x7f) == 0) {
1049 status = exit_code >> 8;
1051 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1052 status = exit_code & 0x7f;
1054 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1057 * Move the task's state to DEAD/TRACE, only one thread can do this.
1059 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1060 EXIT_TRACE : EXIT_DEAD;
1061 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1064 * We own this thread, nobody else can reap it.
1066 read_unlock(&tasklist_lock);
1067 sched_annotate_sleep();
1070 * Check thread_group_leader() to exclude the traced sub-threads.
1072 if (state == EXIT_DEAD && thread_group_leader(p)) {
1073 struct signal_struct *sig = p->signal;
1074 struct signal_struct *psig = current->signal;
1075 unsigned long maxrss;
1076 cputime_t tgutime, tgstime;
1079 * The resource counters for the group leader are in its
1080 * own task_struct. Those for dead threads in the group
1081 * are in its signal_struct, as are those for the child
1082 * processes it has previously reaped. All these
1083 * accumulate in the parent's signal_struct c* fields.
1085 * We don't bother to take a lock here to protect these
1086 * p->signal fields because the whole thread group is dead
1087 * and nobody can change them.
1089 * psig->stats_lock also protects us from our sub-theads
1090 * which can reap other children at the same time. Until
1091 * we change k_getrusage()-like users to rely on this lock
1092 * we have to take ->siglock as well.
1094 * We use thread_group_cputime_adjusted() to get times for
1095 * the thread group, which consolidates times for all threads
1096 * in the group including the group leader.
1098 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1099 spin_lock_irq(¤t->sighand->siglock);
1100 write_seqlock(&psig->stats_lock);
1101 psig->cutime += tgutime + sig->cutime;
1102 psig->cstime += tgstime + sig->cstime;
1103 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1105 p->min_flt + sig->min_flt + sig->cmin_flt;
1107 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1109 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1111 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1113 task_io_get_inblock(p) +
1114 sig->inblock + sig->cinblock;
1116 task_io_get_oublock(p) +
1117 sig->oublock + sig->coublock;
1118 maxrss = max(sig->maxrss, sig->cmaxrss);
1119 if (psig->cmaxrss < maxrss)
1120 psig->cmaxrss = maxrss;
1121 task_io_accounting_add(&psig->ioac, &p->ioac);
1122 task_io_accounting_add(&psig->ioac, &sig->ioac);
1123 write_sequnlock(&psig->stats_lock);
1124 spin_unlock_irq(¤t->sighand->siglock);
1127 retval = wo->wo_rusage
1128 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1129 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1130 ? p->signal->group_exit_code : p->exit_code;
1131 if (!retval && wo->wo_stat)
1132 retval = put_user(status, wo->wo_stat);
1134 infop = wo->wo_info;
1135 if (!retval && infop)
1136 retval = put_user(SIGCHLD, &infop->si_signo);
1137 if (!retval && infop)
1138 retval = put_user(0, &infop->si_errno);
1139 if (!retval && infop) {
1142 if ((status & 0x7f) == 0) {
1146 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1149 retval = put_user((short)why, &infop->si_code);
1151 retval = put_user(status, &infop->si_status);
1153 if (!retval && infop)
1154 retval = put_user(pid, &infop->si_pid);
1155 if (!retval && infop)
1156 retval = put_user(uid, &infop->si_uid);
1160 if (state == EXIT_TRACE) {
1161 write_lock_irq(&tasklist_lock);
1162 /* We dropped tasklist, ptracer could die and untrace */
1165 /* If parent wants a zombie, don't release it now */
1166 state = EXIT_ZOMBIE;
1167 if (do_notify_parent(p, p->exit_signal))
1169 p->exit_state = state;
1170 write_unlock_irq(&tasklist_lock);
1172 if (state == EXIT_DEAD)
1178 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1181 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1182 return &p->exit_code;
1184 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1185 return &p->signal->group_exit_code;
1191 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1193 * @ptrace: is the wait for ptrace
1194 * @p: task to wait for
1196 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1199 * read_lock(&tasklist_lock), which is released if return value is
1200 * non-zero. Also, grabs and releases @p->sighand->siglock.
1203 * 0 if wait condition didn't exist and search for other wait conditions
1204 * should continue. Non-zero return, -errno on failure and @p's pid on
1205 * success, implies that tasklist_lock is released and wait condition
1206 * search should terminate.
1208 static int wait_task_stopped(struct wait_opts *wo,
1209 int ptrace, struct task_struct *p)
1211 struct siginfo __user *infop;
1212 int retval, exit_code, *p_code, why;
1213 uid_t uid = 0; /* unneeded, required by compiler */
1217 * Traditionally we see ptrace'd stopped tasks regardless of options.
1219 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1222 if (!task_stopped_code(p, ptrace))
1226 spin_lock_irq(&p->sighand->siglock);
1228 p_code = task_stopped_code(p, ptrace);
1229 if (unlikely(!p_code))
1232 exit_code = *p_code;
1236 if (!unlikely(wo->wo_flags & WNOWAIT))
1239 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1241 spin_unlock_irq(&p->sighand->siglock);
1246 * Now we are pretty sure this task is interesting.
1247 * Make sure it doesn't get reaped out from under us while we
1248 * give up the lock and then examine it below. We don't want to
1249 * keep holding onto the tasklist_lock while we call getrusage and
1250 * possibly take page faults for user memory.
1253 pid = task_pid_vnr(p);
1254 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1255 read_unlock(&tasklist_lock);
1256 sched_annotate_sleep();
1258 if (unlikely(wo->wo_flags & WNOWAIT))
1259 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1261 retval = wo->wo_rusage
1262 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1263 if (!retval && wo->wo_stat)
1264 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1266 infop = wo->wo_info;
1267 if (!retval && infop)
1268 retval = put_user(SIGCHLD, &infop->si_signo);
1269 if (!retval && infop)
1270 retval = put_user(0, &infop->si_errno);
1271 if (!retval && infop)
1272 retval = put_user((short)why, &infop->si_code);
1273 if (!retval && infop)
1274 retval = put_user(exit_code, &infop->si_status);
1275 if (!retval && infop)
1276 retval = put_user(pid, &infop->si_pid);
1277 if (!retval && infop)
1278 retval = put_user(uid, &infop->si_uid);
1288 * Handle do_wait work for one task in a live, non-stopped state.
1289 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1290 * the lock and this task is uninteresting. If we return nonzero, we have
1291 * released the lock and the system call should return.
1293 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1299 if (!unlikely(wo->wo_flags & WCONTINUED))
1302 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1305 spin_lock_irq(&p->sighand->siglock);
1306 /* Re-check with the lock held. */
1307 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1308 spin_unlock_irq(&p->sighand->siglock);
1311 if (!unlikely(wo->wo_flags & WNOWAIT))
1312 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1313 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1314 spin_unlock_irq(&p->sighand->siglock);
1316 pid = task_pid_vnr(p);
1318 read_unlock(&tasklist_lock);
1319 sched_annotate_sleep();
1322 retval = wo->wo_rusage
1323 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1325 if (!retval && wo->wo_stat)
1326 retval = put_user(0xffff, wo->wo_stat);
1330 retval = wait_noreap_copyout(wo, p, pid, uid,
1331 CLD_CONTINUED, SIGCONT);
1332 BUG_ON(retval == 0);
1339 * Consider @p for a wait by @parent.
1341 * -ECHILD should be in ->notask_error before the first call.
1342 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1343 * Returns zero if the search for a child should continue;
1344 * then ->notask_error is 0 if @p is an eligible child,
1345 * or another error from security_task_wait(), or still -ECHILD.
1347 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1348 struct task_struct *p)
1351 * We can race with wait_task_zombie() from another thread.
1352 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1353 * can't confuse the checks below.
1355 int exit_state = ACCESS_ONCE(p->exit_state);
1358 if (unlikely(exit_state == EXIT_DEAD))
1361 ret = eligible_child(wo, ptrace, p);
1365 ret = security_task_wait(p);
1366 if (unlikely(ret < 0)) {
1368 * If we have not yet seen any eligible child,
1369 * then let this error code replace -ECHILD.
1370 * A permission error will give the user a clue
1371 * to look for security policy problems, rather
1372 * than for mysterious wait bugs.
1374 if (wo->notask_error)
1375 wo->notask_error = ret;
1379 if (unlikely(exit_state == EXIT_TRACE)) {
1381 * ptrace == 0 means we are the natural parent. In this case
1382 * we should clear notask_error, debugger will notify us.
1384 if (likely(!ptrace))
1385 wo->notask_error = 0;
1389 if (likely(!ptrace) && unlikely(p->ptrace)) {
1391 * If it is traced by its real parent's group, just pretend
1392 * the caller is ptrace_do_wait() and reap this child if it
1395 * This also hides group stop state from real parent; otherwise
1396 * a single stop can be reported twice as group and ptrace stop.
1397 * If a ptracer wants to distinguish these two events for its
1398 * own children it should create a separate process which takes
1399 * the role of real parent.
1401 if (!ptrace_reparented(p))
1406 if (exit_state == EXIT_ZOMBIE) {
1407 /* we don't reap group leaders with subthreads */
1408 if (!delay_group_leader(p)) {
1410 * A zombie ptracee is only visible to its ptracer.
1411 * Notification and reaping will be cascaded to the
1412 * real parent when the ptracer detaches.
1414 if (unlikely(ptrace) || likely(!p->ptrace))
1415 return wait_task_zombie(wo, p);
1419 * Allow access to stopped/continued state via zombie by
1420 * falling through. Clearing of notask_error is complex.
1424 * If WEXITED is set, notask_error should naturally be
1425 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1426 * so, if there are live subthreads, there are events to
1427 * wait for. If all subthreads are dead, it's still safe
1428 * to clear - this function will be called again in finite
1429 * amount time once all the subthreads are released and
1430 * will then return without clearing.
1434 * Stopped state is per-task and thus can't change once the
1435 * target task dies. Only continued and exited can happen.
1436 * Clear notask_error if WCONTINUED | WEXITED.
1438 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1439 wo->notask_error = 0;
1442 * @p is alive and it's gonna stop, continue or exit, so
1443 * there always is something to wait for.
1445 wo->notask_error = 0;
1449 * Wait for stopped. Depending on @ptrace, different stopped state
1450 * is used and the two don't interact with each other.
1452 ret = wait_task_stopped(wo, ptrace, p);
1457 * Wait for continued. There's only one continued state and the
1458 * ptracer can consume it which can confuse the real parent. Don't
1459 * use WCONTINUED from ptracer. You don't need or want it.
1461 return wait_task_continued(wo, p);
1465 * Do the work of do_wait() for one thread in the group, @tsk.
1467 * -ECHILD should be in ->notask_error before the first call.
1468 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1469 * Returns zero if the search for a child should continue; then
1470 * ->notask_error is 0 if there were any eligible children,
1471 * or another error from security_task_wait(), or still -ECHILD.
1473 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1475 struct task_struct *p;
1477 list_for_each_entry(p, &tsk->children, sibling) {
1478 int ret = wait_consider_task(wo, 0, p);
1487 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1489 struct task_struct *p;
1491 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1492 int ret = wait_consider_task(wo, 1, p);
1501 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1502 int sync, void *key)
1504 struct wait_opts *wo = container_of(wait, struct wait_opts,
1506 struct task_struct *p = key;
1508 if (!eligible_pid(wo, p))
1511 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1514 return default_wake_function(wait, mode, sync, key);
1517 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1519 __wake_up_sync_key(&parent->signal->wait_chldexit,
1520 TASK_INTERRUPTIBLE, 1, p);
1523 static long do_wait(struct wait_opts *wo)
1525 struct task_struct *tsk;
1528 trace_sched_process_wait(wo->wo_pid);
1530 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1531 wo->child_wait.private = current;
1532 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1535 * If there is nothing that can match our criteria, just get out.
1536 * We will clear ->notask_error to zero if we see any child that
1537 * might later match our criteria, even if we are not able to reap
1540 wo->notask_error = -ECHILD;
1541 if ((wo->wo_type < PIDTYPE_MAX) &&
1542 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1545 set_current_state(TASK_INTERRUPTIBLE);
1546 read_lock(&tasklist_lock);
1549 retval = do_wait_thread(wo, tsk);
1553 retval = ptrace_do_wait(wo, tsk);
1557 if (wo->wo_flags & __WNOTHREAD)
1559 } while_each_thread(current, tsk);
1560 read_unlock(&tasklist_lock);
1563 retval = wo->notask_error;
1564 if (!retval && !(wo->wo_flags & WNOHANG)) {
1565 retval = -ERESTARTSYS;
1566 if (!signal_pending(current)) {
1572 __set_current_state(TASK_RUNNING);
1573 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1577 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1578 infop, int, options, struct rusage __user *, ru)
1580 struct wait_opts wo;
1581 struct pid *pid = NULL;
1585 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1586 __WNOTHREAD|__WCLONE|__WALL))
1588 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1601 type = PIDTYPE_PGID;
1609 if (type < PIDTYPE_MAX)
1610 pid = find_get_pid(upid);
1614 wo.wo_flags = options;
1624 * For a WNOHANG return, clear out all the fields
1625 * we would set so the user can easily tell the
1629 ret = put_user(0, &infop->si_signo);
1631 ret = put_user(0, &infop->si_errno);
1633 ret = put_user(0, &infop->si_code);
1635 ret = put_user(0, &infop->si_pid);
1637 ret = put_user(0, &infop->si_uid);
1639 ret = put_user(0, &infop->si_status);
1646 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1647 int, options, struct rusage __user *, ru)
1649 struct wait_opts wo;
1650 struct pid *pid = NULL;
1654 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1655 __WNOTHREAD|__WCLONE|__WALL))
1658 /* -INT_MIN is not defined */
1659 if (upid == INT_MIN)
1664 else if (upid < 0) {
1665 type = PIDTYPE_PGID;
1666 pid = find_get_pid(-upid);
1667 } else if (upid == 0) {
1668 type = PIDTYPE_PGID;
1669 pid = get_task_pid(current, PIDTYPE_PGID);
1670 } else /* upid > 0 */ {
1672 pid = find_get_pid(upid);
1677 wo.wo_flags = options | WEXITED;
1679 wo.wo_stat = stat_addr;
1687 #ifdef __ARCH_WANT_SYS_WAITPID
1690 * sys_waitpid() remains for compatibility. waitpid() should be
1691 * implemented by calling sys_wait4() from libc.a.
1693 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1695 return sys_wait4(pid, stat_addr, options, NULL);