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>
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 * Determine if a process group is "orphaned", according to the POSIX
214 * definition in 2.2.2.52. Orphaned process groups are not to be affected
215 * by terminal-generated stop signals. Newly orphaned process groups are
216 * to receive a SIGHUP and a SIGCONT.
218 * "I ask you, have you ever known what it is to be an orphan?"
220 static int will_become_orphaned_pgrp(struct pid *pgrp,
221 struct task_struct *ignored_task)
223 struct task_struct *p;
225 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
226 if ((p == ignored_task) ||
227 (p->exit_state && thread_group_empty(p)) ||
228 is_global_init(p->real_parent))
231 if (task_pgrp(p->real_parent) != pgrp &&
232 task_session(p->real_parent) == task_session(p))
234 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
239 int is_current_pgrp_orphaned(void)
243 read_lock(&tasklist_lock);
244 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
245 read_unlock(&tasklist_lock);
250 static bool has_stopped_jobs(struct pid *pgrp)
252 struct task_struct *p;
254 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
255 if (p->signal->flags & SIGNAL_STOP_STOPPED)
257 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
263 * Check to see if any process groups have become orphaned as
264 * a result of our exiting, and if they have any stopped jobs,
265 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
268 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
270 struct pid *pgrp = task_pgrp(tsk);
271 struct task_struct *ignored_task = tsk;
274 /* exit: our father is in a different pgrp than
275 * we are and we were the only connection outside.
277 parent = tsk->real_parent;
279 /* reparent: our child is in a different pgrp than
280 * we are, and it was the only connection outside.
284 if (task_pgrp(parent) != pgrp &&
285 task_session(parent) == task_session(tsk) &&
286 will_become_orphaned_pgrp(pgrp, ignored_task) &&
287 has_stopped_jobs(pgrp)) {
288 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
289 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
295 * A task is exiting. If it owned this mm, find a new owner for the mm.
297 void mm_update_next_owner(struct mm_struct *mm)
299 struct task_struct *c, *g, *p = current;
303 * If the exiting or execing task is not the owner, it's
304 * someone else's problem.
309 * The current owner is exiting/execing and there are no other
310 * candidates. Do not leave the mm pointing to a possibly
311 * freed task structure.
313 if (atomic_read(&mm->mm_users) <= 1) {
318 read_lock(&tasklist_lock);
320 * Search in the children
322 list_for_each_entry(c, &p->children, sibling) {
324 goto assign_new_owner;
328 * Search in the siblings
330 list_for_each_entry(c, &p->real_parent->children, sibling) {
332 goto assign_new_owner;
336 * Search through everything else, we should not get here often.
338 for_each_process(g) {
339 if (g->flags & PF_KTHREAD)
341 for_each_thread(g, c) {
343 goto assign_new_owner;
348 read_unlock(&tasklist_lock);
350 * We found no owner yet mm_users > 1: this implies that we are
351 * most likely racing with swapoff (try_to_unuse()) or /proc or
352 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
361 * The task_lock protects c->mm from changing.
362 * We always want mm->owner->mm == mm
366 * Delay read_unlock() till we have the task_lock()
367 * to ensure that c does not slip away underneath us
369 read_unlock(&tasklist_lock);
379 #endif /* CONFIG_MEMCG */
382 * Turn us into a lazy TLB process if we
385 static void exit_mm(struct task_struct *tsk)
387 struct mm_struct *mm = tsk->mm;
388 struct core_state *core_state;
395 * Serialize with any possible pending coredump.
396 * We must hold mmap_sem around checking core_state
397 * and clearing tsk->mm. The core-inducing thread
398 * will increment ->nr_threads for each thread in the
399 * group with ->mm != NULL.
401 down_read(&mm->mmap_sem);
402 core_state = mm->core_state;
404 struct core_thread self;
406 up_read(&mm->mmap_sem);
409 self.next = xchg(&core_state->dumper.next, &self);
411 * Implies mb(), the result of xchg() must be visible
412 * to core_state->dumper.
414 if (atomic_dec_and_test(&core_state->nr_threads))
415 complete(&core_state->startup);
418 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
419 if (!self.task) /* see coredump_finish() */
421 freezable_schedule();
423 __set_task_state(tsk, TASK_RUNNING);
424 down_read(&mm->mmap_sem);
426 atomic_inc(&mm->mm_count);
427 BUG_ON(mm != tsk->active_mm);
428 /* more a memory barrier than a real lock */
431 up_read(&mm->mmap_sem);
432 enter_lazy_tlb(mm, current);
434 mm_update_next_owner(mm);
436 if (test_thread_flag(TIF_MEMDIE))
440 static struct task_struct *find_alive_thread(struct task_struct *p)
442 struct task_struct *t;
444 for_each_thread(p, t) {
445 if (!(t->flags & PF_EXITING))
451 static struct task_struct *find_child_reaper(struct task_struct *father)
452 __releases(&tasklist_lock)
453 __acquires(&tasklist_lock)
455 struct pid_namespace *pid_ns = task_active_pid_ns(father);
456 struct task_struct *reaper = pid_ns->child_reaper;
458 if (likely(reaper != father))
461 reaper = find_alive_thread(father);
463 pid_ns->child_reaper = reaper;
467 write_unlock_irq(&tasklist_lock);
468 if (unlikely(pid_ns == &init_pid_ns)) {
469 panic("Attempted to kill init! exitcode=0x%08x\n",
470 father->signal->group_exit_code ?: father->exit_code);
472 zap_pid_ns_processes(pid_ns);
473 write_lock_irq(&tasklist_lock);
479 * When we die, we re-parent all our children, and try to:
480 * 1. give them to another thread in our thread group, if such a member exists
481 * 2. give it to the first ancestor process which prctl'd itself as a
482 * child_subreaper for its children (like a service manager)
483 * 3. give it to the init process (PID 1) in our pid namespace
485 static struct task_struct *find_new_reaper(struct task_struct *father,
486 struct task_struct *child_reaper)
488 struct task_struct *thread, *reaper;
490 thread = find_alive_thread(father);
494 if (father->signal->has_child_subreaper) {
496 * Find the first ->is_child_subreaper ancestor in our pid_ns.
497 * We start from father to ensure we can not look into another
498 * namespace, this is safe because all its threads are dead.
500 for (reaper = father;
501 !same_thread_group(reaper, child_reaper);
502 reaper = reaper->real_parent) {
503 /* call_usermodehelper() descendants need this check */
504 if (reaper == &init_task)
506 if (!reaper->signal->is_child_subreaper)
508 thread = find_alive_thread(reaper);
518 * Any that need to be release_task'd are put on the @dead list.
520 static void reparent_leader(struct task_struct *father, struct task_struct *p,
521 struct list_head *dead)
523 if (unlikely(p->exit_state == EXIT_DEAD))
526 /* We don't want people slaying init. */
527 p->exit_signal = SIGCHLD;
529 /* If it has exited notify the new parent about this child's death. */
531 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
532 if (do_notify_parent(p, p->exit_signal)) {
533 p->exit_state = EXIT_DEAD;
534 list_add(&p->ptrace_entry, dead);
538 kill_orphaned_pgrp(p, father);
542 * This does two things:
544 * A. Make init inherit all the child processes
545 * B. Check to see if any process groups have become orphaned
546 * as a result of our exiting, and if they have any stopped
547 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
549 static void forget_original_parent(struct task_struct *father,
550 struct list_head *dead)
552 struct task_struct *p, *t, *reaper;
554 if (unlikely(!list_empty(&father->ptraced)))
555 exit_ptrace(father, dead);
557 /* Can drop and reacquire tasklist_lock */
558 reaper = find_child_reaper(father);
559 if (list_empty(&father->children))
562 reaper = find_new_reaper(father, reaper);
563 list_for_each_entry(p, &father->children, sibling) {
564 for_each_thread(p, t) {
565 t->real_parent = reaper;
566 BUG_ON((!t->ptrace) != (t->parent == father));
567 if (likely(!t->ptrace))
568 t->parent = t->real_parent;
569 if (t->pdeath_signal)
570 group_send_sig_info(t->pdeath_signal,
574 * If this is a threaded reparent there is no need to
575 * notify anyone anything has happened.
577 if (!same_thread_group(reaper, father))
578 reparent_leader(father, p, dead);
580 list_splice_tail_init(&father->children, &reaper->children);
584 * Send signals to all our closest relatives so that they know
585 * to properly mourn us..
587 static void exit_notify(struct task_struct *tsk, int group_dead)
590 struct task_struct *p, *n;
593 write_lock_irq(&tasklist_lock);
594 forget_original_parent(tsk, &dead);
597 kill_orphaned_pgrp(tsk->group_leader, NULL);
599 if (unlikely(tsk->ptrace)) {
600 int sig = thread_group_leader(tsk) &&
601 thread_group_empty(tsk) &&
602 !ptrace_reparented(tsk) ?
603 tsk->exit_signal : SIGCHLD;
604 autoreap = do_notify_parent(tsk, sig);
605 } else if (thread_group_leader(tsk)) {
606 autoreap = thread_group_empty(tsk) &&
607 do_notify_parent(tsk, tsk->exit_signal);
612 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
613 if (tsk->exit_state == EXIT_DEAD)
614 list_add(&tsk->ptrace_entry, &dead);
616 /* mt-exec, de_thread() is waiting for group leader */
617 if (unlikely(tsk->signal->notify_count < 0))
618 wake_up_process(tsk->signal->group_exit_task);
619 write_unlock_irq(&tasklist_lock);
621 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
622 list_del_init(&p->ptrace_entry);
627 #ifdef CONFIG_DEBUG_STACK_USAGE
628 static void check_stack_usage(void)
630 static DEFINE_SPINLOCK(low_water_lock);
631 static int lowest_to_date = THREAD_SIZE;
634 free = stack_not_used(current);
636 if (free >= lowest_to_date)
639 spin_lock(&low_water_lock);
640 if (free < lowest_to_date) {
641 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
642 current->comm, task_pid_nr(current), free);
643 lowest_to_date = free;
645 spin_unlock(&low_water_lock);
648 static inline void check_stack_usage(void) {}
651 void do_exit(long code)
653 struct task_struct *tsk = current;
655 TASKS_RCU(int tasks_rcu_i);
657 profile_task_exit(tsk);
659 WARN_ON(blk_needs_flush_plug(tsk));
661 if (unlikely(in_interrupt()))
662 panic("Aiee, killing interrupt handler!");
663 if (unlikely(!tsk->pid))
664 panic("Attempted to kill the idle task!");
667 * If do_exit is called because this processes oopsed, it's possible
668 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
669 * continuing. Amongst other possible reasons, this is to prevent
670 * mm_release()->clear_child_tid() from writing to a user-controlled
675 ptrace_event(PTRACE_EVENT_EXIT, code);
677 validate_creds_for_do_exit(tsk);
680 * We're taking recursive faults here in do_exit. Safest is to just
681 * leave this task alone and wait for reboot.
683 if (unlikely(tsk->flags & PF_EXITING)) {
684 pr_alert("Fixing recursive fault but reboot is needed!\n");
686 * We can do this unlocked here. The futex code uses
687 * this flag just to verify whether the pi state
688 * cleanup has been done or not. In the worst case it
689 * loops once more. We pretend that the cleanup was
690 * done as there is no way to return. Either the
691 * OWNER_DIED bit is set by now or we push the blocked
692 * task into the wait for ever nirwana as well.
694 tsk->flags |= PF_EXITPIDONE;
695 set_current_state(TASK_UNINTERRUPTIBLE);
699 exit_signals(tsk); /* sets PF_EXITING */
701 * tsk->flags are checked in the futex code to protect against
702 * an exiting task cleaning up the robust pi futexes.
705 raw_spin_unlock_wait(&tsk->pi_lock);
707 if (unlikely(in_atomic())) {
708 pr_info("note: %s[%d] exited with preempt_count %d\n",
709 current->comm, task_pid_nr(current),
711 preempt_count_set(PREEMPT_ENABLED);
714 /* sync mm's RSS info before statistics gathering */
716 sync_mm_rss(tsk->mm);
717 acct_update_integrals(tsk);
718 group_dead = atomic_dec_and_test(&tsk->signal->live);
720 hrtimer_cancel(&tsk->signal->real_timer);
721 exit_itimers(tsk->signal);
723 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
725 acct_collect(code, group_dead);
730 tsk->exit_code = code;
731 taskstats_exit(tsk, group_dead);
737 trace_sched_process_exit(tsk);
744 disassociate_ctty(1);
745 exit_task_namespaces(tsk);
750 * Flush inherited counters to the parent - before the parent
751 * gets woken up by child-exit notifications.
753 * because of cgroup mode, must be called before cgroup_exit()
755 perf_event_exit_task(tsk);
760 * FIXME: do that only when needed, using sched_exit tracepoint
762 flush_ptrace_hw_breakpoint(tsk);
764 TASKS_RCU(preempt_disable());
765 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
766 TASKS_RCU(preempt_enable());
767 exit_notify(tsk, group_dead);
768 proc_exit_connector(tsk);
771 mpol_put(tsk->mempolicy);
772 tsk->mempolicy = NULL;
776 if (unlikely(current->pi_state_cache))
777 kfree(current->pi_state_cache);
780 * Make sure we are holding no locks:
782 debug_check_no_locks_held();
784 * We can do this unlocked here. The futex code uses this flag
785 * just to verify whether the pi state cleanup has been done
786 * or not. In the worst case it loops once more.
788 tsk->flags |= PF_EXITPIDONE;
791 exit_io_context(tsk);
793 if (tsk->splice_pipe)
794 free_pipe_info(tsk->splice_pipe);
796 if (tsk->task_frag.page)
797 put_page(tsk->task_frag.page);
799 validate_creds_for_do_exit(tsk);
804 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
806 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
809 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
810 * when the following two conditions become true.
811 * - There is race condition of mmap_sem (It is acquired by
813 * - SMI occurs before setting TASK_RUNINNG.
814 * (or hypervisor of virtual machine switches to other guest)
815 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
817 * To avoid it, we have to wait for releasing tsk->pi_lock which
818 * is held by try_to_wake_up()
821 raw_spin_unlock_wait(&tsk->pi_lock);
823 /* causes final put_task_struct in finish_task_switch(). */
824 tsk->state = TASK_DEAD;
825 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
828 /* Avoid "noreturn function does return". */
830 cpu_relax(); /* For when BUG is null */
832 EXPORT_SYMBOL_GPL(do_exit);
834 void complete_and_exit(struct completion *comp, long code)
841 EXPORT_SYMBOL(complete_and_exit);
843 SYSCALL_DEFINE1(exit, int, error_code)
845 do_exit((error_code&0xff)<<8);
849 * Take down every thread in the group. This is called by fatal signals
850 * as well as by sys_exit_group (below).
853 do_group_exit(int exit_code)
855 struct signal_struct *sig = current->signal;
857 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
859 if (signal_group_exit(sig))
860 exit_code = sig->group_exit_code;
861 else if (!thread_group_empty(current)) {
862 struct sighand_struct *const sighand = current->sighand;
864 spin_lock_irq(&sighand->siglock);
865 if (signal_group_exit(sig))
866 /* Another thread got here before we took the lock. */
867 exit_code = sig->group_exit_code;
869 sig->group_exit_code = exit_code;
870 sig->flags = SIGNAL_GROUP_EXIT;
871 zap_other_threads(current);
873 spin_unlock_irq(&sighand->siglock);
881 * this kills every thread in the thread group. Note that any externally
882 * wait4()-ing process will get the correct exit code - even if this
883 * thread is not the thread group leader.
885 SYSCALL_DEFINE1(exit_group, int, error_code)
887 do_group_exit((error_code & 0xff) << 8);
893 enum pid_type wo_type;
897 struct siginfo __user *wo_info;
899 struct rusage __user *wo_rusage;
901 wait_queue_t child_wait;
906 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
908 if (type != PIDTYPE_PID)
909 task = task->group_leader;
910 return task->pids[type].pid;
913 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
915 return wo->wo_type == PIDTYPE_MAX ||
916 task_pid_type(p, wo->wo_type) == wo->wo_pid;
919 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
921 if (!eligible_pid(wo, p))
923 /* Wait for all children (clone and not) if __WALL is set;
924 * otherwise, wait for clone children *only* if __WCLONE is
925 * set; otherwise, wait for non-clone children *only*. (Note:
926 * A "clone" child here is one that reports to its parent
927 * using a signal other than SIGCHLD.) */
928 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
929 && !(wo->wo_flags & __WALL))
935 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
936 pid_t pid, uid_t uid, int why, int status)
938 struct siginfo __user *infop;
939 int retval = wo->wo_rusage
940 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
946 retval = put_user(SIGCHLD, &infop->si_signo);
948 retval = put_user(0, &infop->si_errno);
950 retval = put_user((short)why, &infop->si_code);
952 retval = put_user(pid, &infop->si_pid);
954 retval = put_user(uid, &infop->si_uid);
956 retval = put_user(status, &infop->si_status);
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)
971 int state, retval, status;
972 pid_t pid = task_pid_vnr(p);
973 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
974 struct siginfo __user *infop;
976 if (!likely(wo->wo_flags & WEXITED))
979 if (unlikely(wo->wo_flags & WNOWAIT)) {
980 int exit_code = p->exit_code;
984 read_unlock(&tasklist_lock);
985 sched_annotate_sleep();
987 if ((exit_code & 0x7f) == 0) {
989 status = exit_code >> 8;
991 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
992 status = exit_code & 0x7f;
994 return wait_noreap_copyout(wo, p, pid, uid, why, status);
997 * Move the task's state to DEAD/TRACE, only one thread can do this.
999 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1000 EXIT_TRACE : EXIT_DEAD;
1001 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1004 * We own this thread, nobody else can reap it.
1006 read_unlock(&tasklist_lock);
1007 sched_annotate_sleep();
1010 * Check thread_group_leader() to exclude the traced sub-threads.
1012 if (state == EXIT_DEAD && thread_group_leader(p)) {
1013 struct signal_struct *sig = p->signal;
1014 struct signal_struct *psig = current->signal;
1015 unsigned long maxrss;
1016 cputime_t tgutime, tgstime;
1019 * The resource counters for the group leader are in its
1020 * own task_struct. Those for dead threads in the group
1021 * are in its signal_struct, as are those for the child
1022 * processes it has previously reaped. All these
1023 * accumulate in the parent's signal_struct c* fields.
1025 * We don't bother to take a lock here to protect these
1026 * p->signal fields because the whole thread group is dead
1027 * and nobody can change them.
1029 * psig->stats_lock also protects us from our sub-theads
1030 * which can reap other children at the same time. Until
1031 * we change k_getrusage()-like users to rely on this lock
1032 * we have to take ->siglock as well.
1034 * We use thread_group_cputime_adjusted() to get times for
1035 * the thread group, which consolidates times for all threads
1036 * in the group including the group leader.
1038 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1039 spin_lock_irq(¤t->sighand->siglock);
1040 write_seqlock(&psig->stats_lock);
1041 psig->cutime += tgutime + sig->cutime;
1042 psig->cstime += tgstime + sig->cstime;
1043 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1045 p->min_flt + sig->min_flt + sig->cmin_flt;
1047 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1049 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1051 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1053 task_io_get_inblock(p) +
1054 sig->inblock + sig->cinblock;
1056 task_io_get_oublock(p) +
1057 sig->oublock + sig->coublock;
1058 maxrss = max(sig->maxrss, sig->cmaxrss);
1059 if (psig->cmaxrss < maxrss)
1060 psig->cmaxrss = maxrss;
1061 task_io_accounting_add(&psig->ioac, &p->ioac);
1062 task_io_accounting_add(&psig->ioac, &sig->ioac);
1063 write_sequnlock(&psig->stats_lock);
1064 spin_unlock_irq(¤t->sighand->siglock);
1067 retval = wo->wo_rusage
1068 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1069 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1070 ? p->signal->group_exit_code : p->exit_code;
1071 if (!retval && wo->wo_stat)
1072 retval = put_user(status, wo->wo_stat);
1074 infop = wo->wo_info;
1075 if (!retval && infop)
1076 retval = put_user(SIGCHLD, &infop->si_signo);
1077 if (!retval && infop)
1078 retval = put_user(0, &infop->si_errno);
1079 if (!retval && infop) {
1082 if ((status & 0x7f) == 0) {
1086 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1089 retval = put_user((short)why, &infop->si_code);
1091 retval = put_user(status, &infop->si_status);
1093 if (!retval && infop)
1094 retval = put_user(pid, &infop->si_pid);
1095 if (!retval && infop)
1096 retval = put_user(uid, &infop->si_uid);
1100 if (state == EXIT_TRACE) {
1101 write_lock_irq(&tasklist_lock);
1102 /* We dropped tasklist, ptracer could die and untrace */
1105 /* If parent wants a zombie, don't release it now */
1106 state = EXIT_ZOMBIE;
1107 if (do_notify_parent(p, p->exit_signal))
1109 p->exit_state = state;
1110 write_unlock_irq(&tasklist_lock);
1112 if (state == EXIT_DEAD)
1118 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1121 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1122 return &p->exit_code;
1124 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1125 return &p->signal->group_exit_code;
1131 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1133 * @ptrace: is the wait for ptrace
1134 * @p: task to wait for
1136 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1139 * read_lock(&tasklist_lock), which is released if return value is
1140 * non-zero. Also, grabs and releases @p->sighand->siglock.
1143 * 0 if wait condition didn't exist and search for other wait conditions
1144 * should continue. Non-zero return, -errno on failure and @p's pid on
1145 * success, implies that tasklist_lock is released and wait condition
1146 * search should terminate.
1148 static int wait_task_stopped(struct wait_opts *wo,
1149 int ptrace, struct task_struct *p)
1151 struct siginfo __user *infop;
1152 int retval, exit_code, *p_code, why;
1153 uid_t uid = 0; /* unneeded, required by compiler */
1157 * Traditionally we see ptrace'd stopped tasks regardless of options.
1159 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1162 if (!task_stopped_code(p, ptrace))
1166 spin_lock_irq(&p->sighand->siglock);
1168 p_code = task_stopped_code(p, ptrace);
1169 if (unlikely(!p_code))
1172 exit_code = *p_code;
1176 if (!unlikely(wo->wo_flags & WNOWAIT))
1179 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1181 spin_unlock_irq(&p->sighand->siglock);
1186 * Now we are pretty sure this task is interesting.
1187 * Make sure it doesn't get reaped out from under us while we
1188 * give up the lock and then examine it below. We don't want to
1189 * keep holding onto the tasklist_lock while we call getrusage and
1190 * possibly take page faults for user memory.
1193 pid = task_pid_vnr(p);
1194 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1195 read_unlock(&tasklist_lock);
1196 sched_annotate_sleep();
1198 if (unlikely(wo->wo_flags & WNOWAIT))
1199 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1201 retval = wo->wo_rusage
1202 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1203 if (!retval && wo->wo_stat)
1204 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1206 infop = wo->wo_info;
1207 if (!retval && infop)
1208 retval = put_user(SIGCHLD, &infop->si_signo);
1209 if (!retval && infop)
1210 retval = put_user(0, &infop->si_errno);
1211 if (!retval && infop)
1212 retval = put_user((short)why, &infop->si_code);
1213 if (!retval && infop)
1214 retval = put_user(exit_code, &infop->si_status);
1215 if (!retval && infop)
1216 retval = put_user(pid, &infop->si_pid);
1217 if (!retval && infop)
1218 retval = put_user(uid, &infop->si_uid);
1228 * Handle do_wait work for one task in a live, non-stopped state.
1229 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1230 * the lock and this task is uninteresting. If we return nonzero, we have
1231 * released the lock and the system call should return.
1233 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1239 if (!unlikely(wo->wo_flags & WCONTINUED))
1242 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1245 spin_lock_irq(&p->sighand->siglock);
1246 /* Re-check with the lock held. */
1247 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1248 spin_unlock_irq(&p->sighand->siglock);
1251 if (!unlikely(wo->wo_flags & WNOWAIT))
1252 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1253 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1254 spin_unlock_irq(&p->sighand->siglock);
1256 pid = task_pid_vnr(p);
1258 read_unlock(&tasklist_lock);
1259 sched_annotate_sleep();
1262 retval = wo->wo_rusage
1263 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1265 if (!retval && wo->wo_stat)
1266 retval = put_user(0xffff, wo->wo_stat);
1270 retval = wait_noreap_copyout(wo, p, pid, uid,
1271 CLD_CONTINUED, SIGCONT);
1272 BUG_ON(retval == 0);
1279 * Consider @p for a wait by @parent.
1281 * -ECHILD should be in ->notask_error before the first call.
1282 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1283 * Returns zero if the search for a child should continue;
1284 * then ->notask_error is 0 if @p is an eligible child,
1285 * or another error from security_task_wait(), or still -ECHILD.
1287 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1288 struct task_struct *p)
1291 * We can race with wait_task_zombie() from another thread.
1292 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1293 * can't confuse the checks below.
1295 int exit_state = ACCESS_ONCE(p->exit_state);
1298 if (unlikely(exit_state == EXIT_DEAD))
1301 ret = eligible_child(wo, p);
1305 ret = security_task_wait(p);
1306 if (unlikely(ret < 0)) {
1308 * If we have not yet seen any eligible child,
1309 * then let this error code replace -ECHILD.
1310 * A permission error will give the user a clue
1311 * to look for security policy problems, rather
1312 * than for mysterious wait bugs.
1314 if (wo->notask_error)
1315 wo->notask_error = ret;
1319 if (unlikely(exit_state == EXIT_TRACE)) {
1321 * ptrace == 0 means we are the natural parent. In this case
1322 * we should clear notask_error, debugger will notify us.
1324 if (likely(!ptrace))
1325 wo->notask_error = 0;
1329 if (likely(!ptrace) && unlikely(p->ptrace)) {
1331 * If it is traced by its real parent's group, just pretend
1332 * the caller is ptrace_do_wait() and reap this child if it
1335 * This also hides group stop state from real parent; otherwise
1336 * a single stop can be reported twice as group and ptrace stop.
1337 * If a ptracer wants to distinguish these two events for its
1338 * own children it should create a separate process which takes
1339 * the role of real parent.
1341 if (!ptrace_reparented(p))
1346 if (exit_state == EXIT_ZOMBIE) {
1347 /* we don't reap group leaders with subthreads */
1348 if (!delay_group_leader(p)) {
1350 * A zombie ptracee is only visible to its ptracer.
1351 * Notification and reaping will be cascaded to the
1352 * real parent when the ptracer detaches.
1354 if (unlikely(ptrace) || likely(!p->ptrace))
1355 return wait_task_zombie(wo, p);
1359 * Allow access to stopped/continued state via zombie by
1360 * falling through. Clearing of notask_error is complex.
1364 * If WEXITED is set, notask_error should naturally be
1365 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1366 * so, if there are live subthreads, there are events to
1367 * wait for. If all subthreads are dead, it's still safe
1368 * to clear - this function will be called again in finite
1369 * amount time once all the subthreads are released and
1370 * will then return without clearing.
1374 * Stopped state is per-task and thus can't change once the
1375 * target task dies. Only continued and exited can happen.
1376 * Clear notask_error if WCONTINUED | WEXITED.
1378 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1379 wo->notask_error = 0;
1382 * @p is alive and it's gonna stop, continue or exit, so
1383 * there always is something to wait for.
1385 wo->notask_error = 0;
1389 * Wait for stopped. Depending on @ptrace, different stopped state
1390 * is used and the two don't interact with each other.
1392 ret = wait_task_stopped(wo, ptrace, p);
1397 * Wait for continued. There's only one continued state and the
1398 * ptracer can consume it which can confuse the real parent. Don't
1399 * use WCONTINUED from ptracer. You don't need or want it.
1401 return wait_task_continued(wo, p);
1405 * Do the work of do_wait() for one thread in the group, @tsk.
1407 * -ECHILD should be in ->notask_error before the first call.
1408 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1409 * Returns zero if the search for a child should continue; then
1410 * ->notask_error is 0 if there were any eligible children,
1411 * or another error from security_task_wait(), or still -ECHILD.
1413 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1415 struct task_struct *p;
1417 list_for_each_entry(p, &tsk->children, sibling) {
1418 int ret = wait_consider_task(wo, 0, p);
1427 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1429 struct task_struct *p;
1431 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1432 int ret = wait_consider_task(wo, 1, p);
1441 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1442 int sync, void *key)
1444 struct wait_opts *wo = container_of(wait, struct wait_opts,
1446 struct task_struct *p = key;
1448 if (!eligible_pid(wo, p))
1451 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1454 return default_wake_function(wait, mode, sync, key);
1457 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1459 __wake_up_sync_key(&parent->signal->wait_chldexit,
1460 TASK_INTERRUPTIBLE, 1, p);
1463 static long do_wait(struct wait_opts *wo)
1465 struct task_struct *tsk;
1468 trace_sched_process_wait(wo->wo_pid);
1470 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1471 wo->child_wait.private = current;
1472 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1475 * If there is nothing that can match our criteria, just get out.
1476 * We will clear ->notask_error to zero if we see any child that
1477 * might later match our criteria, even if we are not able to reap
1480 wo->notask_error = -ECHILD;
1481 if ((wo->wo_type < PIDTYPE_MAX) &&
1482 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1485 set_current_state(TASK_INTERRUPTIBLE);
1486 read_lock(&tasklist_lock);
1489 retval = do_wait_thread(wo, tsk);
1493 retval = ptrace_do_wait(wo, tsk);
1497 if (wo->wo_flags & __WNOTHREAD)
1499 } while_each_thread(current, tsk);
1500 read_unlock(&tasklist_lock);
1503 retval = wo->notask_error;
1504 if (!retval && !(wo->wo_flags & WNOHANG)) {
1505 retval = -ERESTARTSYS;
1506 if (!signal_pending(current)) {
1512 __set_current_state(TASK_RUNNING);
1513 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1517 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1518 infop, int, options, struct rusage __user *, ru)
1520 struct wait_opts wo;
1521 struct pid *pid = NULL;
1525 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1527 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1540 type = PIDTYPE_PGID;
1548 if (type < PIDTYPE_MAX)
1549 pid = find_get_pid(upid);
1553 wo.wo_flags = options;
1563 * For a WNOHANG return, clear out all the fields
1564 * we would set so the user can easily tell the
1568 ret = put_user(0, &infop->si_signo);
1570 ret = put_user(0, &infop->si_errno);
1572 ret = put_user(0, &infop->si_code);
1574 ret = put_user(0, &infop->si_pid);
1576 ret = put_user(0, &infop->si_uid);
1578 ret = put_user(0, &infop->si_status);
1585 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1586 int, options, struct rusage __user *, ru)
1588 struct wait_opts wo;
1589 struct pid *pid = NULL;
1593 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1594 __WNOTHREAD|__WCLONE|__WALL))
1599 else if (upid < 0) {
1600 type = PIDTYPE_PGID;
1601 pid = find_get_pid(-upid);
1602 } else if (upid == 0) {
1603 type = PIDTYPE_PGID;
1604 pid = get_task_pid(current, PIDTYPE_PGID);
1605 } else /* upid > 0 */ {
1607 pid = find_get_pid(upid);
1612 wo.wo_flags = options | WEXITED;
1614 wo.wo_stat = stat_addr;
1622 #ifdef __ARCH_WANT_SYS_WAITPID
1625 * sys_waitpid() remains for compatibility. waitpid() should be
1626 * implemented by calling sys_wait4() from libc.a.
1628 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1630 return sys_wait4(pid, stat_addr, options, NULL);