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/cpu.h>
18 #include <linux/acct.h>
19 #include <linux/tsacct_kern.h>
20 #include <linux/file.h>
21 #include <linux/fdtable.h>
22 #include <linux/freezer.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/cgroup.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/pipe_fs_i.h>
42 #include <linux/audit.h> /* for audit_free() */
43 #include <linux/resource.h>
44 #include <linux/blkdev.h>
45 #include <linux/task_io_accounting_ops.h>
46 #include <linux/tracehook.h>
47 #include <linux/fs_struct.h>
48 #include <linux/init_task.h>
49 #include <linux/perf_event.h>
50 #include <trace/events/sched.h>
51 #include <linux/hw_breakpoint.h>
52 #include <linux/oom.h>
53 #include <linux/writeback.h>
54 #include <linux/shm.h>
55 #include <linux/kcov.h>
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
62 static void __unhash_process(struct task_struct *p, bool group_dead)
65 detach_pid(p, PIDTYPE_PID);
67 detach_pid(p, PIDTYPE_PGID);
68 detach_pid(p, PIDTYPE_SID);
70 list_del_rcu(&p->tasks);
71 list_del_init(&p->sibling);
72 __this_cpu_dec(process_counts);
74 list_del_rcu(&p->thread_group);
75 list_del_rcu(&p->thread_node);
79 * This function expects the tasklist_lock write-locked.
81 static void __exit_signal(struct task_struct *tsk)
83 struct signal_struct *sig = tsk->signal;
84 bool group_dead = thread_group_leader(tsk);
85 struct sighand_struct *sighand;
86 struct tty_struct *uninitialized_var(tty);
87 cputime_t utime, stime;
89 sighand = rcu_dereference_check(tsk->sighand,
90 lockdep_tasklist_lock_is_held());
91 spin_lock(&sighand->siglock);
93 posix_cpu_timers_exit(tsk);
95 posix_cpu_timers_exit_group(tsk);
100 * This can only happen if the caller is de_thread().
101 * FIXME: this is the temporary hack, we should teach
102 * posix-cpu-timers to handle this case correctly.
104 if (unlikely(has_group_leader_pid(tsk)))
105 posix_cpu_timers_exit_group(tsk);
108 * If there is any task waiting for the group exit
111 if (sig->notify_count > 0 && !--sig->notify_count)
112 wake_up_process(sig->group_exit_task);
114 if (tsk == sig->curr_target)
115 sig->curr_target = next_thread(tsk);
119 * Accumulate here the counters for all threads as they die. We could
120 * skip the group leader because it is the last user of signal_struct,
121 * but we want to avoid the race with thread_group_cputime() which can
122 * see the empty ->thread_head list.
124 task_cputime(tsk, &utime, &stime);
125 write_seqlock(&sig->stats_lock);
128 sig->gtime += task_gtime(tsk);
129 sig->min_flt += tsk->min_flt;
130 sig->maj_flt += tsk->maj_flt;
131 sig->nvcsw += tsk->nvcsw;
132 sig->nivcsw += tsk->nivcsw;
133 sig->inblock += task_io_get_inblock(tsk);
134 sig->oublock += task_io_get_oublock(tsk);
135 task_io_accounting_add(&sig->ioac, &tsk->ioac);
136 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138 __unhash_process(tsk, group_dead);
139 write_sequnlock(&sig->stats_lock);
142 * Do this under ->siglock, we can race with another thread
143 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145 flush_sigqueue(&tsk->pending);
147 spin_unlock(&sighand->siglock);
149 __cleanup_sighand(sighand);
150 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
152 flush_sigqueue(&sig->shared_pending);
157 static void delayed_put_task_struct(struct rcu_head *rhp)
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 perf_event_delayed_put(tsk);
162 trace_sched_process_free(tsk);
163 put_task_struct(tsk);
167 void release_task(struct task_struct *p)
169 struct task_struct *leader;
172 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials. But shut RCU-lockdep up */
175 atomic_dec(&__task_cred(p)->user->processes);
180 write_lock_irq(&tasklist_lock);
181 ptrace_release_task(p);
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader)
192 && leader->exit_state == EXIT_ZOMBIE) {
194 * If we were the last child thread and the leader has
195 * exited already, and the leader's parent ignores SIGCHLD,
196 * then we are the one who should release the leader.
198 zap_leader = do_notify_parent(leader, leader->exit_signal);
200 leader->exit_state = EXIT_DEAD;
203 write_unlock_irq(&tasklist_lock);
205 call_rcu(&p->rcu, delayed_put_task_struct);
208 if (unlikely(zap_leader))
213 * Note that if this function returns a valid task_struct pointer (!NULL)
214 * task->usage must remain >0 for the duration of the RCU critical section.
216 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
218 struct sighand_struct *sighand;
219 struct task_struct *task;
222 * We need to verify that release_task() was not called and thus
223 * delayed_put_task_struct() can't run and drop the last reference
224 * before rcu_read_unlock(). We check task->sighand != NULL,
225 * but we can read the already freed and reused memory.
228 task = rcu_dereference(*ptask);
232 probe_kernel_address(&task->sighand, sighand);
235 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
236 * was already freed we can not miss the preceding update of this
240 if (unlikely(task != READ_ONCE(*ptask)))
244 * We've re-checked that "task == *ptask", now we have two different
247 * 1. This is actually the same task/task_struct. In this case
248 * sighand != NULL tells us it is still alive.
250 * 2. This is another task which got the same memory for task_struct.
251 * We can't know this of course, and we can not trust
254 * In this case we actually return a random value, but this is
257 * If we return NULL - we can pretend that we actually noticed that
258 * *ptask was updated when the previous task has exited. Or pretend
259 * that probe_slab_address(&sighand) reads NULL.
261 * If we return the new task (because sighand is not NULL for any
262 * reason) - this is fine too. This (new) task can't go away before
265 * And note: We could even eliminate the false positive if re-read
266 * task->sighand once again to avoid the falsely NULL. But this case
267 * is very unlikely so we don't care.
275 struct task_struct *try_get_task_struct(struct task_struct **ptask)
277 struct task_struct *task;
280 task = task_rcu_dereference(ptask);
282 get_task_struct(task);
289 * Determine if a process group is "orphaned", according to the POSIX
290 * definition in 2.2.2.52. Orphaned process groups are not to be affected
291 * by terminal-generated stop signals. Newly orphaned process groups are
292 * to receive a SIGHUP and a SIGCONT.
294 * "I ask you, have you ever known what it is to be an orphan?"
296 static int will_become_orphaned_pgrp(struct pid *pgrp,
297 struct task_struct *ignored_task)
299 struct task_struct *p;
301 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
302 if ((p == ignored_task) ||
303 (p->exit_state && thread_group_empty(p)) ||
304 is_global_init(p->real_parent))
307 if (task_pgrp(p->real_parent) != pgrp &&
308 task_session(p->real_parent) == task_session(p))
310 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
315 int is_current_pgrp_orphaned(void)
319 read_lock(&tasklist_lock);
320 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
321 read_unlock(&tasklist_lock);
326 static bool has_stopped_jobs(struct pid *pgrp)
328 struct task_struct *p;
330 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
331 if (p->signal->flags & SIGNAL_STOP_STOPPED)
333 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
339 * Check to see if any process groups have become orphaned as
340 * a result of our exiting, and if they have any stopped jobs,
341 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
344 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
346 struct pid *pgrp = task_pgrp(tsk);
347 struct task_struct *ignored_task = tsk;
350 /* exit: our father is in a different pgrp than
351 * we are and we were the only connection outside.
353 parent = tsk->real_parent;
355 /* reparent: our child is in a different pgrp than
356 * we are, and it was the only connection outside.
360 if (task_pgrp(parent) != pgrp &&
361 task_session(parent) == task_session(tsk) &&
362 will_become_orphaned_pgrp(pgrp, ignored_task) &&
363 has_stopped_jobs(pgrp)) {
364 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
365 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
371 * A task is exiting. If it owned this mm, find a new owner for the mm.
373 void mm_update_next_owner(struct mm_struct *mm)
375 struct task_struct *c, *g, *p = current;
379 * If the exiting or execing task is not the owner, it's
380 * someone else's problem.
385 * The current owner is exiting/execing and there are no other
386 * candidates. Do not leave the mm pointing to a possibly
387 * freed task structure.
389 if (atomic_read(&mm->mm_users) <= 1) {
394 read_lock(&tasklist_lock);
396 * Search in the children
398 list_for_each_entry(c, &p->children, sibling) {
400 goto assign_new_owner;
404 * Search in the siblings
406 list_for_each_entry(c, &p->real_parent->children, sibling) {
408 goto assign_new_owner;
412 * Search through everything else, we should not get here often.
414 for_each_process(g) {
415 if (g->flags & PF_KTHREAD)
417 for_each_thread(g, c) {
419 goto assign_new_owner;
424 read_unlock(&tasklist_lock);
426 * We found no owner yet mm_users > 1: this implies that we are
427 * most likely racing with swapoff (try_to_unuse()) or /proc or
428 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
437 * The task_lock protects c->mm from changing.
438 * We always want mm->owner->mm == mm
442 * Delay read_unlock() till we have the task_lock()
443 * to ensure that c does not slip away underneath us
445 read_unlock(&tasklist_lock);
455 #endif /* CONFIG_MEMCG */
458 * Turn us into a lazy TLB process if we
461 static void exit_mm(struct task_struct *tsk)
463 struct mm_struct *mm = tsk->mm;
464 struct core_state *core_state;
466 exit_mm_release(tsk, mm);
471 * Serialize with any possible pending coredump.
472 * We must hold mmap_sem around checking core_state
473 * and clearing tsk->mm. The core-inducing thread
474 * will increment ->nr_threads for each thread in the
475 * group with ->mm != NULL.
477 down_read(&mm->mmap_sem);
478 core_state = mm->core_state;
480 struct core_thread self;
482 up_read(&mm->mmap_sem);
485 if (self.task->flags & PF_SIGNALED)
486 self.next = xchg(&core_state->dumper.next, &self);
490 * Implies mb(), the result of xchg() must be visible
491 * to core_state->dumper.
493 if (atomic_dec_and_test(&core_state->nr_threads))
494 complete(&core_state->startup);
497 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
498 if (!self.task) /* see coredump_finish() */
500 freezable_schedule();
502 __set_task_state(tsk, TASK_RUNNING);
503 down_read(&mm->mmap_sem);
505 atomic_inc(&mm->mm_count);
506 BUG_ON(mm != tsk->active_mm);
507 /* more a memory barrier than a real lock */
510 up_read(&mm->mmap_sem);
511 enter_lazy_tlb(mm, current);
513 mm_update_next_owner(mm);
515 if (test_thread_flag(TIF_MEMDIE))
519 static struct task_struct *find_alive_thread(struct task_struct *p)
521 struct task_struct *t;
523 for_each_thread(p, t) {
524 if (!(t->flags & PF_EXITING))
530 static struct task_struct *find_child_reaper(struct task_struct *father,
531 struct list_head *dead)
532 __releases(&tasklist_lock)
533 __acquires(&tasklist_lock)
535 struct pid_namespace *pid_ns = task_active_pid_ns(father);
536 struct task_struct *reaper = pid_ns->child_reaper;
537 struct task_struct *p, *n;
539 if (likely(reaper != father))
542 reaper = find_alive_thread(father);
544 pid_ns->child_reaper = reaper;
548 write_unlock_irq(&tasklist_lock);
549 if (unlikely(pid_ns == &init_pid_ns)) {
550 panic("Attempted to kill init! exitcode=0x%08x\n",
551 father->signal->group_exit_code ?: father->exit_code);
554 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
555 list_del_init(&p->ptrace_entry);
559 zap_pid_ns_processes(pid_ns);
560 write_lock_irq(&tasklist_lock);
566 * When we die, we re-parent all our children, and try to:
567 * 1. give them to another thread in our thread group, if such a member exists
568 * 2. give it to the first ancestor process which prctl'd itself as a
569 * child_subreaper for its children (like a service manager)
570 * 3. give it to the init process (PID 1) in our pid namespace
572 static struct task_struct *find_new_reaper(struct task_struct *father,
573 struct task_struct *child_reaper)
575 struct task_struct *thread, *reaper;
577 thread = find_alive_thread(father);
581 if (father->signal->has_child_subreaper) {
583 * Find the first ->is_child_subreaper ancestor in our pid_ns.
584 * We start from father to ensure we can not look into another
585 * namespace, this is safe because all its threads are dead.
587 for (reaper = father;
588 !same_thread_group(reaper, child_reaper);
589 reaper = reaper->real_parent) {
590 /* call_usermodehelper() descendants need this check */
591 if (reaper == &init_task)
593 if (!reaper->signal->is_child_subreaper)
595 thread = find_alive_thread(reaper);
605 * Any that need to be release_task'd are put on the @dead list.
607 static void reparent_leader(struct task_struct *father, struct task_struct *p,
608 struct list_head *dead)
610 if (unlikely(p->exit_state == EXIT_DEAD))
613 /* We don't want people slaying init. */
614 p->exit_signal = SIGCHLD;
616 /* If it has exited notify the new parent about this child's death. */
618 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
619 if (do_notify_parent(p, p->exit_signal)) {
620 p->exit_state = EXIT_DEAD;
621 list_add(&p->ptrace_entry, dead);
625 kill_orphaned_pgrp(p, father);
629 * This does two things:
631 * A. Make init inherit all the child processes
632 * B. Check to see if any process groups have become orphaned
633 * as a result of our exiting, and if they have any stopped
634 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
636 static void forget_original_parent(struct task_struct *father,
637 struct list_head *dead)
639 struct task_struct *p, *t, *reaper;
641 if (unlikely(!list_empty(&father->ptraced)))
642 exit_ptrace(father, dead);
644 /* Can drop and reacquire tasklist_lock */
645 reaper = find_child_reaper(father, dead);
646 if (list_empty(&father->children))
649 reaper = find_new_reaper(father, reaper);
650 list_for_each_entry(p, &father->children, sibling) {
651 for_each_thread(p, t) {
652 t->real_parent = reaper;
653 BUG_ON((!t->ptrace) != (t->parent == father));
654 if (likely(!t->ptrace))
655 t->parent = t->real_parent;
656 if (t->pdeath_signal)
657 group_send_sig_info(t->pdeath_signal,
661 * If this is a threaded reparent there is no need to
662 * notify anyone anything has happened.
664 if (!same_thread_group(reaper, father))
665 reparent_leader(father, p, dead);
667 list_splice_tail_init(&father->children, &reaper->children);
671 * Send signals to all our closest relatives so that they know
672 * to properly mourn us..
674 static void exit_notify(struct task_struct *tsk, int group_dead)
677 struct task_struct *p, *n;
680 write_lock_irq(&tasklist_lock);
681 forget_original_parent(tsk, &dead);
684 kill_orphaned_pgrp(tsk->group_leader, NULL);
686 if (unlikely(tsk->ptrace)) {
687 int sig = thread_group_leader(tsk) &&
688 thread_group_empty(tsk) &&
689 !ptrace_reparented(tsk) ?
690 tsk->exit_signal : SIGCHLD;
691 autoreap = do_notify_parent(tsk, sig);
692 } else if (thread_group_leader(tsk)) {
693 autoreap = thread_group_empty(tsk) &&
694 do_notify_parent(tsk, tsk->exit_signal);
699 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
700 if (tsk->exit_state == EXIT_DEAD)
701 list_add(&tsk->ptrace_entry, &dead);
703 /* mt-exec, de_thread() is waiting for group leader */
704 if (unlikely(tsk->signal->notify_count < 0))
705 wake_up_process(tsk->signal->group_exit_task);
706 write_unlock_irq(&tasklist_lock);
708 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
709 list_del_init(&p->ptrace_entry);
714 #ifdef CONFIG_DEBUG_STACK_USAGE
715 static void check_stack_usage(void)
717 static DEFINE_SPINLOCK(low_water_lock);
718 static int lowest_to_date = THREAD_SIZE;
721 free = stack_not_used(current);
723 if (free >= lowest_to_date)
726 spin_lock(&low_water_lock);
727 if (free < lowest_to_date) {
728 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
729 current->comm, task_pid_nr(current), free);
730 lowest_to_date = free;
732 spin_unlock(&low_water_lock);
735 static inline void check_stack_usage(void) {}
738 void __noreturn do_exit(long code)
740 struct task_struct *tsk = current;
742 TASKS_RCU(int tasks_rcu_i);
745 * We can get here from a kernel oops, sometimes with preemption off.
746 * Start by checking for critical errors.
747 * Then fix up important state like USER_DS and preemption.
748 * Then do everything else.
751 WARN_ON(blk_needs_flush_plug(tsk));
753 if (unlikely(in_interrupt()))
754 panic("Aiee, killing interrupt handler!");
755 if (unlikely(!tsk->pid))
756 panic("Attempted to kill the idle task!");
759 * If do_exit is called because this processes oopsed, it's possible
760 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
761 * continuing. Amongst other possible reasons, this is to prevent
762 * mm_release()->clear_child_tid() from writing to a user-controlled
767 if (unlikely(in_atomic())) {
768 pr_info("note: %s[%d] exited with preempt_count %d\n",
769 current->comm, task_pid_nr(current),
771 preempt_count_set(PREEMPT_ENABLED);
774 profile_task_exit(tsk);
777 ptrace_event(PTRACE_EVENT_EXIT, code);
779 validate_creds_for_do_exit(tsk);
782 * We're taking recursive faults here in do_exit. Safest is to just
783 * leave this task alone and wait for reboot.
785 if (unlikely(tsk->flags & PF_EXITING)) {
786 pr_alert("Fixing recursive fault but reboot is needed!\n");
787 futex_exit_recursive(tsk);
788 set_current_state(TASK_UNINTERRUPTIBLE);
792 exit_signals(tsk); /* sets PF_EXITING */
794 /* sync mm's RSS info before statistics gathering */
796 sync_mm_rss(tsk->mm);
797 acct_update_integrals(tsk);
798 group_dead = atomic_dec_and_test(&tsk->signal->live);
800 hrtimer_cancel(&tsk->signal->real_timer);
801 exit_itimers(tsk->signal);
803 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
805 acct_collect(code, group_dead);
810 tsk->exit_code = code;
811 taskstats_exit(tsk, group_dead);
817 trace_sched_process_exit(tsk);
824 disassociate_ctty(1);
825 exit_task_namespaces(tsk);
830 * Flush inherited counters to the parent - before the parent
831 * gets woken up by child-exit notifications.
833 * because of cgroup mode, must be called before cgroup_exit()
835 perf_event_exit_task(tsk);
837 sched_autogroup_exit_task(tsk);
841 * FIXME: do that only when needed, using sched_exit tracepoint
843 flush_ptrace_hw_breakpoint(tsk);
845 TASKS_RCU(preempt_disable());
846 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
847 TASKS_RCU(preempt_enable());
848 exit_notify(tsk, group_dead);
849 proc_exit_connector(tsk);
850 mpol_put_task_policy(tsk);
852 if (unlikely(current->pi_state_cache))
853 kfree(current->pi_state_cache);
856 * Make sure we are holding no locks:
858 debug_check_no_locks_held();
861 exit_io_context(tsk);
863 if (tsk->splice_pipe)
864 free_pipe_info(tsk->splice_pipe);
866 if (tsk->task_frag.page)
867 put_page(tsk->task_frag.page);
869 validate_creds_for_do_exit(tsk);
874 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
876 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
880 EXPORT_SYMBOL_GPL(do_exit);
882 void complete_and_exit(struct completion *comp, long code)
889 EXPORT_SYMBOL(complete_and_exit);
891 SYSCALL_DEFINE1(exit, int, error_code)
893 do_exit((error_code&0xff)<<8);
897 * Take down every thread in the group. This is called by fatal signals
898 * as well as by sys_exit_group (below).
901 do_group_exit(int exit_code)
903 struct signal_struct *sig = current->signal;
905 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
907 if (signal_group_exit(sig))
908 exit_code = sig->group_exit_code;
909 else if (!thread_group_empty(current)) {
910 struct sighand_struct *const sighand = current->sighand;
912 spin_lock_irq(&sighand->siglock);
913 if (signal_group_exit(sig))
914 /* Another thread got here before we took the lock. */
915 exit_code = sig->group_exit_code;
917 sig->group_exit_code = exit_code;
918 sig->flags = SIGNAL_GROUP_EXIT;
919 zap_other_threads(current);
921 spin_unlock_irq(&sighand->siglock);
929 * this kills every thread in the thread group. Note that any externally
930 * wait4()-ing process will get the correct exit code - even if this
931 * thread is not the thread group leader.
933 SYSCALL_DEFINE1(exit_group, int, error_code)
935 do_group_exit((error_code & 0xff) << 8);
941 enum pid_type wo_type;
945 struct siginfo __user *wo_info;
947 struct rusage __user *wo_rusage;
949 wait_queue_t child_wait;
954 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
956 if (type != PIDTYPE_PID)
957 task = task->group_leader;
958 return task->pids[type].pid;
961 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
963 return wo->wo_type == PIDTYPE_MAX ||
964 task_pid_type(p, wo->wo_type) == wo->wo_pid;
968 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
970 if (!eligible_pid(wo, p))
974 * Wait for all children (clone and not) if __WALL is set or
975 * if it is traced by us.
977 if (ptrace || (wo->wo_flags & __WALL))
981 * Otherwise, wait for clone children *only* if __WCLONE is set;
982 * otherwise, wait for non-clone children *only*.
984 * Note: a "clone" child here is one that reports to its parent
985 * using a signal other than SIGCHLD, or a non-leader thread which
986 * we can only see if it is traced by us.
988 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
994 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
995 pid_t pid, uid_t uid, int why, int status)
997 struct siginfo __user *infop;
998 int retval = wo->wo_rusage
999 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1002 infop = wo->wo_info;
1005 retval = put_user(SIGCHLD, &infop->si_signo);
1007 retval = put_user(0, &infop->si_errno);
1009 retval = put_user((short)why, &infop->si_code);
1011 retval = put_user(pid, &infop->si_pid);
1013 retval = put_user(uid, &infop->si_uid);
1015 retval = put_user(status, &infop->si_status);
1023 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1024 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1025 * the lock and this task is uninteresting. If we return nonzero, we have
1026 * released the lock and the system call should return.
1028 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1030 int state, retval, status;
1031 pid_t pid = task_pid_vnr(p);
1032 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1033 struct siginfo __user *infop;
1035 if (!likely(wo->wo_flags & WEXITED))
1038 if (unlikely(wo->wo_flags & WNOWAIT)) {
1039 int exit_code = p->exit_code;
1043 read_unlock(&tasklist_lock);
1044 sched_annotate_sleep();
1046 if ((exit_code & 0x7f) == 0) {
1048 status = exit_code >> 8;
1050 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1051 status = exit_code & 0x7f;
1053 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1056 * Move the task's state to DEAD/TRACE, only one thread can do this.
1058 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1059 EXIT_TRACE : EXIT_DEAD;
1060 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1063 * We own this thread, nobody else can reap it.
1065 read_unlock(&tasklist_lock);
1066 sched_annotate_sleep();
1069 * Check thread_group_leader() to exclude the traced sub-threads.
1071 if (state == EXIT_DEAD && thread_group_leader(p)) {
1072 struct signal_struct *sig = p->signal;
1073 struct signal_struct *psig = current->signal;
1074 unsigned long maxrss;
1075 cputime_t tgutime, tgstime;
1078 * The resource counters for the group leader are in its
1079 * own task_struct. Those for dead threads in the group
1080 * are in its signal_struct, as are those for the child
1081 * processes it has previously reaped. All these
1082 * accumulate in the parent's signal_struct c* fields.
1084 * We don't bother to take a lock here to protect these
1085 * p->signal fields because the whole thread group is dead
1086 * and nobody can change them.
1088 * psig->stats_lock also protects us from our sub-theads
1089 * which can reap other children at the same time. Until
1090 * we change k_getrusage()-like users to rely on this lock
1091 * we have to take ->siglock as well.
1093 * We use thread_group_cputime_adjusted() to get times for
1094 * the thread group, which consolidates times for all threads
1095 * in the group including the group leader.
1097 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1098 spin_lock_irq(¤t->sighand->siglock);
1099 write_seqlock(&psig->stats_lock);
1100 psig->cutime += tgutime + sig->cutime;
1101 psig->cstime += tgstime + sig->cstime;
1102 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1104 p->min_flt + sig->min_flt + sig->cmin_flt;
1106 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1108 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1110 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1112 task_io_get_inblock(p) +
1113 sig->inblock + sig->cinblock;
1115 task_io_get_oublock(p) +
1116 sig->oublock + sig->coublock;
1117 maxrss = max(sig->maxrss, sig->cmaxrss);
1118 if (psig->cmaxrss < maxrss)
1119 psig->cmaxrss = maxrss;
1120 task_io_accounting_add(&psig->ioac, &p->ioac);
1121 task_io_accounting_add(&psig->ioac, &sig->ioac);
1122 write_sequnlock(&psig->stats_lock);
1123 spin_unlock_irq(¤t->sighand->siglock);
1126 retval = wo->wo_rusage
1127 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1128 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1129 ? p->signal->group_exit_code : p->exit_code;
1130 if (!retval && wo->wo_stat)
1131 retval = put_user(status, wo->wo_stat);
1133 infop = wo->wo_info;
1134 if (!retval && infop)
1135 retval = put_user(SIGCHLD, &infop->si_signo);
1136 if (!retval && infop)
1137 retval = put_user(0, &infop->si_errno);
1138 if (!retval && infop) {
1141 if ((status & 0x7f) == 0) {
1145 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1148 retval = put_user((short)why, &infop->si_code);
1150 retval = put_user(status, &infop->si_status);
1152 if (!retval && infop)
1153 retval = put_user(pid, &infop->si_pid);
1154 if (!retval && infop)
1155 retval = put_user(uid, &infop->si_uid);
1159 if (state == EXIT_TRACE) {
1160 write_lock_irq(&tasklist_lock);
1161 /* We dropped tasklist, ptracer could die and untrace */
1164 /* If parent wants a zombie, don't release it now */
1165 state = EXIT_ZOMBIE;
1166 if (do_notify_parent(p, p->exit_signal))
1168 p->exit_state = state;
1169 write_unlock_irq(&tasklist_lock);
1171 if (state == EXIT_DEAD)
1177 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1180 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1181 return &p->exit_code;
1183 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1184 return &p->signal->group_exit_code;
1190 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1192 * @ptrace: is the wait for ptrace
1193 * @p: task to wait for
1195 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1198 * read_lock(&tasklist_lock), which is released if return value is
1199 * non-zero. Also, grabs and releases @p->sighand->siglock.
1202 * 0 if wait condition didn't exist and search for other wait conditions
1203 * should continue. Non-zero return, -errno on failure and @p's pid on
1204 * success, implies that tasklist_lock is released and wait condition
1205 * search should terminate.
1207 static int wait_task_stopped(struct wait_opts *wo,
1208 int ptrace, struct task_struct *p)
1210 struct siginfo __user *infop;
1211 int retval, exit_code, *p_code, why;
1212 uid_t uid = 0; /* unneeded, required by compiler */
1216 * Traditionally we see ptrace'd stopped tasks regardless of options.
1218 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1221 if (!task_stopped_code(p, ptrace))
1225 spin_lock_irq(&p->sighand->siglock);
1227 p_code = task_stopped_code(p, ptrace);
1228 if (unlikely(!p_code))
1231 exit_code = *p_code;
1235 if (!unlikely(wo->wo_flags & WNOWAIT))
1238 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1240 spin_unlock_irq(&p->sighand->siglock);
1245 * Now we are pretty sure this task is interesting.
1246 * Make sure it doesn't get reaped out from under us while we
1247 * give up the lock and then examine it below. We don't want to
1248 * keep holding onto the tasklist_lock while we call getrusage and
1249 * possibly take page faults for user memory.
1252 pid = task_pid_vnr(p);
1253 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1254 read_unlock(&tasklist_lock);
1255 sched_annotate_sleep();
1257 if (unlikely(wo->wo_flags & WNOWAIT))
1258 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1260 retval = wo->wo_rusage
1261 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1262 if (!retval && wo->wo_stat)
1263 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1265 infop = wo->wo_info;
1266 if (!retval && infop)
1267 retval = put_user(SIGCHLD, &infop->si_signo);
1268 if (!retval && infop)
1269 retval = put_user(0, &infop->si_errno);
1270 if (!retval && infop)
1271 retval = put_user((short)why, &infop->si_code);
1272 if (!retval && infop)
1273 retval = put_user(exit_code, &infop->si_status);
1274 if (!retval && infop)
1275 retval = put_user(pid, &infop->si_pid);
1276 if (!retval && infop)
1277 retval = put_user(uid, &infop->si_uid);
1287 * Handle do_wait work for one task in a live, non-stopped state.
1288 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1289 * the lock and this task is uninteresting. If we return nonzero, we have
1290 * released the lock and the system call should return.
1292 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1298 if (!unlikely(wo->wo_flags & WCONTINUED))
1301 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1304 spin_lock_irq(&p->sighand->siglock);
1305 /* Re-check with the lock held. */
1306 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1307 spin_unlock_irq(&p->sighand->siglock);
1310 if (!unlikely(wo->wo_flags & WNOWAIT))
1311 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1312 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1313 spin_unlock_irq(&p->sighand->siglock);
1315 pid = task_pid_vnr(p);
1317 read_unlock(&tasklist_lock);
1318 sched_annotate_sleep();
1321 retval = wo->wo_rusage
1322 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1324 if (!retval && wo->wo_stat)
1325 retval = put_user(0xffff, wo->wo_stat);
1329 retval = wait_noreap_copyout(wo, p, pid, uid,
1330 CLD_CONTINUED, SIGCONT);
1331 BUG_ON(retval == 0);
1338 * Consider @p for a wait by @parent.
1340 * -ECHILD should be in ->notask_error before the first call.
1341 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1342 * Returns zero if the search for a child should continue;
1343 * then ->notask_error is 0 if @p is an eligible child,
1346 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1347 struct task_struct *p)
1350 * We can race with wait_task_zombie() from another thread.
1351 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1352 * can't confuse the checks below.
1354 int exit_state = ACCESS_ONCE(p->exit_state);
1357 if (unlikely(exit_state == EXIT_DEAD))
1360 ret = eligible_child(wo, ptrace, p);
1364 if (unlikely(exit_state == EXIT_TRACE)) {
1366 * ptrace == 0 means we are the natural parent. In this case
1367 * we should clear notask_error, debugger will notify us.
1369 if (likely(!ptrace))
1370 wo->notask_error = 0;
1374 if (likely(!ptrace) && unlikely(p->ptrace)) {
1376 * If it is traced by its real parent's group, just pretend
1377 * the caller is ptrace_do_wait() and reap this child if it
1380 * This also hides group stop state from real parent; otherwise
1381 * a single stop can be reported twice as group and ptrace stop.
1382 * If a ptracer wants to distinguish these two events for its
1383 * own children it should create a separate process which takes
1384 * the role of real parent.
1386 if (!ptrace_reparented(p))
1391 if (exit_state == EXIT_ZOMBIE) {
1392 /* we don't reap group leaders with subthreads */
1393 if (!delay_group_leader(p)) {
1395 * A zombie ptracee is only visible to its ptracer.
1396 * Notification and reaping will be cascaded to the
1397 * real parent when the ptracer detaches.
1399 if (unlikely(ptrace) || likely(!p->ptrace))
1400 return wait_task_zombie(wo, p);
1404 * Allow access to stopped/continued state via zombie by
1405 * falling through. Clearing of notask_error is complex.
1409 * If WEXITED is set, notask_error should naturally be
1410 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1411 * so, if there are live subthreads, there are events to
1412 * wait for. If all subthreads are dead, it's still safe
1413 * to clear - this function will be called again in finite
1414 * amount time once all the subthreads are released and
1415 * will then return without clearing.
1419 * Stopped state is per-task and thus can't change once the
1420 * target task dies. Only continued and exited can happen.
1421 * Clear notask_error if WCONTINUED | WEXITED.
1423 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1424 wo->notask_error = 0;
1427 * @p is alive and it's gonna stop, continue or exit, so
1428 * there always is something to wait for.
1430 wo->notask_error = 0;
1434 * Wait for stopped. Depending on @ptrace, different stopped state
1435 * is used and the two don't interact with each other.
1437 ret = wait_task_stopped(wo, ptrace, p);
1442 * Wait for continued. There's only one continued state and the
1443 * ptracer can consume it which can confuse the real parent. Don't
1444 * use WCONTINUED from ptracer. You don't need or want it.
1446 return wait_task_continued(wo, p);
1450 * Do the work of do_wait() for one thread in the group, @tsk.
1452 * -ECHILD should be in ->notask_error before the first call.
1453 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1454 * Returns zero if the search for a child should continue; then
1455 * ->notask_error is 0 if there were any eligible children,
1458 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1460 struct task_struct *p;
1462 list_for_each_entry(p, &tsk->children, sibling) {
1463 int ret = wait_consider_task(wo, 0, p);
1472 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1474 struct task_struct *p;
1476 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1477 int ret = wait_consider_task(wo, 1, p);
1486 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1487 int sync, void *key)
1489 struct wait_opts *wo = container_of(wait, struct wait_opts,
1491 struct task_struct *p = key;
1493 if (!eligible_pid(wo, p))
1496 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1499 return default_wake_function(wait, mode, sync, key);
1502 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1504 __wake_up_sync_key(&parent->signal->wait_chldexit,
1505 TASK_INTERRUPTIBLE, 1, p);
1508 static long do_wait(struct wait_opts *wo)
1510 struct task_struct *tsk;
1513 trace_sched_process_wait(wo->wo_pid);
1515 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1516 wo->child_wait.private = current;
1517 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1520 * If there is nothing that can match our criteria, just get out.
1521 * We will clear ->notask_error to zero if we see any child that
1522 * might later match our criteria, even if we are not able to reap
1525 wo->notask_error = -ECHILD;
1526 if ((wo->wo_type < PIDTYPE_MAX) &&
1527 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1530 set_current_state(TASK_INTERRUPTIBLE);
1531 read_lock(&tasklist_lock);
1534 retval = do_wait_thread(wo, tsk);
1538 retval = ptrace_do_wait(wo, tsk);
1542 if (wo->wo_flags & __WNOTHREAD)
1544 } while_each_thread(current, tsk);
1545 read_unlock(&tasklist_lock);
1548 retval = wo->notask_error;
1549 if (!retval && !(wo->wo_flags & WNOHANG)) {
1550 retval = -ERESTARTSYS;
1551 if (!signal_pending(current)) {
1557 __set_current_state(TASK_RUNNING);
1558 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1562 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1563 infop, int, options, struct rusage __user *, ru)
1565 struct wait_opts wo;
1566 struct pid *pid = NULL;
1570 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1571 __WNOTHREAD|__WCLONE|__WALL))
1573 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1586 type = PIDTYPE_PGID;
1594 if (type < PIDTYPE_MAX)
1595 pid = find_get_pid(upid);
1599 wo.wo_flags = options;
1609 * For a WNOHANG return, clear out all the fields
1610 * we would set so the user can easily tell the
1614 ret = put_user(0, &infop->si_signo);
1616 ret = put_user(0, &infop->si_errno);
1618 ret = put_user(0, &infop->si_code);
1620 ret = put_user(0, &infop->si_pid);
1622 ret = put_user(0, &infop->si_uid);
1624 ret = put_user(0, &infop->si_status);
1631 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1632 int, options, struct rusage __user *, ru)
1634 struct wait_opts wo;
1635 struct pid *pid = NULL;
1639 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1640 __WNOTHREAD|__WCLONE|__WALL))
1643 /* -INT_MIN is not defined */
1644 if (upid == INT_MIN)
1649 else if (upid < 0) {
1650 type = PIDTYPE_PGID;
1651 pid = find_get_pid(-upid);
1652 } else if (upid == 0) {
1653 type = PIDTYPE_PGID;
1654 pid = get_task_pid(current, PIDTYPE_PGID);
1655 } else /* upid > 0 */ {
1657 pid = find_get_pid(upid);
1662 wo.wo_flags = options | WEXITED;
1664 wo.wo_stat = stat_addr;
1672 #ifdef __ARCH_WANT_SYS_WAITPID
1675 * sys_waitpid() remains for compatibility. waitpid() should be
1676 * implemented by calling sys_wait4() from libc.a.
1678 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1680 return sys_wait4(pid, stat_addr, options, NULL);