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 exit_mm(struct task_struct *tsk);
64 static void __unhash_process(struct task_struct *p, bool group_dead)
67 detach_pid(p, PIDTYPE_PID);
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
76 list_del_rcu(&p->thread_group);
77 list_del_rcu(&p->thread_node);
81 * This function expects the tasklist_lock write-locked.
83 static void __exit_signal(struct task_struct *tsk)
85 struct signal_struct *sig = tsk->signal;
86 bool group_dead = thread_group_leader(tsk);
87 struct sighand_struct *sighand;
88 struct tty_struct *uninitialized_var(tty);
89 cputime_t utime, stime;
91 sighand = rcu_dereference_check(tsk->sighand,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand->siglock);
95 posix_cpu_timers_exit(tsk);
97 posix_cpu_timers_exit_group(tsk);
102 * This can only happen if the caller is de_thread().
103 * FIXME: this is the temporary hack, we should teach
104 * posix-cpu-timers to handle this case correctly.
106 if (unlikely(has_group_leader_pid(tsk)))
107 posix_cpu_timers_exit_group(tsk);
110 * If there is any task waiting for the group exit
113 if (sig->notify_count > 0 && !--sig->notify_count)
114 wake_up_process(sig->group_exit_task);
116 if (tsk == sig->curr_target)
117 sig->curr_target = next_thread(tsk);
121 * Accumulate here the counters for all threads as they die. We could
122 * skip the group leader because it is the last user of signal_struct,
123 * but we want to avoid the race with thread_group_cputime() which can
124 * see the empty ->thread_head list.
126 task_cputime(tsk, &utime, &stime);
127 write_seqlock(&sig->stats_lock);
130 sig->gtime += task_gtime(tsk);
131 sig->min_flt += tsk->min_flt;
132 sig->maj_flt += tsk->maj_flt;
133 sig->nvcsw += tsk->nvcsw;
134 sig->nivcsw += tsk->nivcsw;
135 sig->inblock += task_io_get_inblock(tsk);
136 sig->oublock += task_io_get_oublock(tsk);
137 task_io_accounting_add(&sig->ioac, &tsk->ioac);
138 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
140 __unhash_process(tsk, group_dead);
141 write_sequnlock(&sig->stats_lock);
144 * Do this under ->siglock, we can race with another thread
145 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
147 flush_sigqueue(&tsk->pending);
149 spin_unlock(&sighand->siglock);
151 __cleanup_sighand(sighand);
152 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
154 flush_sigqueue(&sig->shared_pending);
159 static void delayed_put_task_struct(struct rcu_head *rhp)
161 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
163 perf_event_delayed_put(tsk);
164 trace_sched_process_free(tsk);
165 put_task_struct(tsk);
169 void release_task(struct task_struct *p)
171 struct task_struct *leader;
174 /* don't need to get the RCU readlock here - the process is dead and
175 * can't be modifying its own credentials. But shut RCU-lockdep up */
177 atomic_dec(&__task_cred(p)->user->processes);
182 write_lock_irq(&tasklist_lock);
183 ptrace_release_task(p);
187 * If we are the last non-leader member of the thread
188 * group, and the leader is zombie, then notify the
189 * group leader's parent process. (if it wants notification.)
192 leader = p->group_leader;
193 if (leader != p && thread_group_empty(leader)
194 && leader->exit_state == EXIT_ZOMBIE) {
196 * If we were the last child thread and the leader has
197 * exited already, and the leader's parent ignores SIGCHLD,
198 * then we are the one who should release the leader.
200 zap_leader = do_notify_parent(leader, leader->exit_signal);
202 leader->exit_state = EXIT_DEAD;
205 write_unlock_irq(&tasklist_lock);
207 call_rcu(&p->rcu, delayed_put_task_struct);
210 if (unlikely(zap_leader))
215 * Determine if a process group is "orphaned", according to the POSIX
216 * definition in 2.2.2.52. Orphaned process groups are not to be affected
217 * by terminal-generated stop signals. Newly orphaned process groups are
218 * to receive a SIGHUP and a SIGCONT.
220 * "I ask you, have you ever known what it is to be an orphan?"
222 static int will_become_orphaned_pgrp(struct pid *pgrp,
223 struct task_struct *ignored_task)
225 struct task_struct *p;
227 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
228 if ((p == ignored_task) ||
229 (p->exit_state && thread_group_empty(p)) ||
230 is_global_init(p->real_parent))
233 if (task_pgrp(p->real_parent) != pgrp &&
234 task_session(p->real_parent) == task_session(p))
236 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
241 int is_current_pgrp_orphaned(void)
245 read_lock(&tasklist_lock);
246 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
247 read_unlock(&tasklist_lock);
252 static bool has_stopped_jobs(struct pid *pgrp)
254 struct task_struct *p;
256 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
257 if (p->signal->flags & SIGNAL_STOP_STOPPED)
259 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265 * Check to see if any process groups have become orphaned as
266 * a result of our exiting, and if they have any stopped jobs,
267 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
270 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
272 struct pid *pgrp = task_pgrp(tsk);
273 struct task_struct *ignored_task = tsk;
276 /* exit: our father is in a different pgrp than
277 * we are and we were the only connection outside.
279 parent = tsk->real_parent;
281 /* reparent: our child is in a different pgrp than
282 * we are, and it was the only connection outside.
286 if (task_pgrp(parent) != pgrp &&
287 task_session(parent) == task_session(tsk) &&
288 will_become_orphaned_pgrp(pgrp, ignored_task) &&
289 has_stopped_jobs(pgrp)) {
290 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
291 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
297 * A task is exiting. If it owned this mm, find a new owner for the mm.
299 void mm_update_next_owner(struct mm_struct *mm)
301 struct task_struct *c, *g, *p = current;
305 * If the exiting or execing task is not the owner, it's
306 * someone else's problem.
311 * The current owner is exiting/execing and there are no other
312 * candidates. Do not leave the mm pointing to a possibly
313 * freed task structure.
315 if (atomic_read(&mm->mm_users) <= 1) {
320 read_lock(&tasklist_lock);
322 * Search in the children
324 list_for_each_entry(c, &p->children, sibling) {
326 goto assign_new_owner;
330 * Search in the siblings
332 list_for_each_entry(c, &p->real_parent->children, sibling) {
334 goto assign_new_owner;
338 * Search through everything else, we should not get here often.
340 for_each_process(g) {
341 if (g->flags & PF_KTHREAD)
343 for_each_thread(g, c) {
345 goto assign_new_owner;
350 read_unlock(&tasklist_lock);
352 * We found no owner yet mm_users > 1: this implies that we are
353 * most likely racing with swapoff (try_to_unuse()) or /proc or
354 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
363 * The task_lock protects c->mm from changing.
364 * We always want mm->owner->mm == mm
368 * Delay read_unlock() till we have the task_lock()
369 * to ensure that c does not slip away underneath us
371 read_unlock(&tasklist_lock);
381 #endif /* CONFIG_MEMCG */
384 * Turn us into a lazy TLB process if we
387 static void exit_mm(struct task_struct *tsk)
389 struct mm_struct *mm = tsk->mm;
390 struct core_state *core_state;
392 exit_mm_release(tsk, mm);
397 * Serialize with any possible pending coredump.
398 * We must hold mmap_sem around checking core_state
399 * and clearing tsk->mm. The core-inducing thread
400 * will increment ->nr_threads for each thread in the
401 * group with ->mm != NULL.
403 down_read(&mm->mmap_sem);
404 core_state = mm->core_state;
406 struct core_thread self;
408 up_read(&mm->mmap_sem);
411 if (self.task->flags & PF_SIGNALED)
412 self.next = xchg(&core_state->dumper.next, &self);
416 * Implies mb(), the result of xchg() must be visible
417 * to core_state->dumper.
419 if (atomic_dec_and_test(&core_state->nr_threads))
420 complete(&core_state->startup);
423 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
424 if (!self.task) /* see coredump_finish() */
426 freezable_schedule();
428 __set_task_state(tsk, TASK_RUNNING);
429 down_read(&mm->mmap_sem);
431 atomic_inc(&mm->mm_count);
432 BUG_ON(mm != tsk->active_mm);
433 /* more a memory barrier than a real lock */
436 up_read(&mm->mmap_sem);
437 enter_lazy_tlb(mm, current);
439 mm_update_next_owner(mm);
441 if (test_thread_flag(TIF_MEMDIE))
445 static struct task_struct *find_alive_thread(struct task_struct *p)
447 struct task_struct *t;
449 for_each_thread(p, t) {
450 if (!(t->flags & PF_EXITING))
456 static struct task_struct *find_child_reaper(struct task_struct *father,
457 struct list_head *dead)
458 __releases(&tasklist_lock)
459 __acquires(&tasklist_lock)
461 struct pid_namespace *pid_ns = task_active_pid_ns(father);
462 struct task_struct *reaper = pid_ns->child_reaper;
463 struct task_struct *p, *n;
465 if (likely(reaper != father))
468 reaper = find_alive_thread(father);
470 pid_ns->child_reaper = reaper;
474 write_unlock_irq(&tasklist_lock);
475 if (unlikely(pid_ns == &init_pid_ns)) {
476 panic("Attempted to kill init! exitcode=0x%08x\n",
477 father->signal->group_exit_code ?: father->exit_code);
480 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
481 list_del_init(&p->ptrace_entry);
485 zap_pid_ns_processes(pid_ns);
486 write_lock_irq(&tasklist_lock);
492 * When we die, we re-parent all our children, and try to:
493 * 1. give them to another thread in our thread group, if such a member exists
494 * 2. give it to the first ancestor process which prctl'd itself as a
495 * child_subreaper for its children (like a service manager)
496 * 3. give it to the init process (PID 1) in our pid namespace
498 static struct task_struct *find_new_reaper(struct task_struct *father,
499 struct task_struct *child_reaper)
501 struct task_struct *thread, *reaper;
503 thread = find_alive_thread(father);
507 if (father->signal->has_child_subreaper) {
509 * Find the first ->is_child_subreaper ancestor in our pid_ns.
510 * We start from father to ensure we can not look into another
511 * namespace, this is safe because all its threads are dead.
513 for (reaper = father;
514 !same_thread_group(reaper, child_reaper);
515 reaper = reaper->real_parent) {
516 /* call_usermodehelper() descendants need this check */
517 if (reaper == &init_task)
519 if (!reaper->signal->is_child_subreaper)
521 thread = find_alive_thread(reaper);
531 * Any that need to be release_task'd are put on the @dead list.
533 static void reparent_leader(struct task_struct *father, struct task_struct *p,
534 struct list_head *dead)
536 if (unlikely(p->exit_state == EXIT_DEAD))
539 /* We don't want people slaying init. */
540 p->exit_signal = SIGCHLD;
542 /* If it has exited notify the new parent about this child's death. */
544 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
545 if (do_notify_parent(p, p->exit_signal)) {
546 p->exit_state = EXIT_DEAD;
547 list_add(&p->ptrace_entry, dead);
551 kill_orphaned_pgrp(p, father);
555 * This does two things:
557 * A. Make init inherit all the child processes
558 * B. Check to see if any process groups have become orphaned
559 * as a result of our exiting, and if they have any stopped
560 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
562 static void forget_original_parent(struct task_struct *father,
563 struct list_head *dead)
565 struct task_struct *p, *t, *reaper;
567 if (unlikely(!list_empty(&father->ptraced)))
568 exit_ptrace(father, dead);
570 /* Can drop and reacquire tasklist_lock */
571 reaper = find_child_reaper(father, dead);
572 if (list_empty(&father->children))
575 reaper = find_new_reaper(father, reaper);
576 list_for_each_entry(p, &father->children, sibling) {
577 for_each_thread(p, t) {
578 t->real_parent = reaper;
579 BUG_ON((!t->ptrace) != (t->parent == father));
580 if (likely(!t->ptrace))
581 t->parent = t->real_parent;
582 if (t->pdeath_signal)
583 group_send_sig_info(t->pdeath_signal,
587 * If this is a threaded reparent there is no need to
588 * notify anyone anything has happened.
590 if (!same_thread_group(reaper, father))
591 reparent_leader(father, p, dead);
593 list_splice_tail_init(&father->children, &reaper->children);
597 * Send signals to all our closest relatives so that they know
598 * to properly mourn us..
600 static void exit_notify(struct task_struct *tsk, int group_dead)
603 struct task_struct *p, *n;
606 write_lock_irq(&tasklist_lock);
607 forget_original_parent(tsk, &dead);
610 kill_orphaned_pgrp(tsk->group_leader, NULL);
612 if (unlikely(tsk->ptrace)) {
613 int sig = thread_group_leader(tsk) &&
614 thread_group_empty(tsk) &&
615 !ptrace_reparented(tsk) ?
616 tsk->exit_signal : SIGCHLD;
617 autoreap = do_notify_parent(tsk, sig);
618 } else if (thread_group_leader(tsk)) {
619 autoreap = thread_group_empty(tsk) &&
620 do_notify_parent(tsk, tsk->exit_signal);
625 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
626 if (tsk->exit_state == EXIT_DEAD)
627 list_add(&tsk->ptrace_entry, &dead);
629 /* mt-exec, de_thread() is waiting for group leader */
630 if (unlikely(tsk->signal->notify_count < 0))
631 wake_up_process(tsk->signal->group_exit_task);
632 write_unlock_irq(&tasklist_lock);
634 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
635 list_del_init(&p->ptrace_entry);
640 #ifdef CONFIG_DEBUG_STACK_USAGE
641 static void check_stack_usage(void)
643 static DEFINE_SPINLOCK(low_water_lock);
644 static int lowest_to_date = THREAD_SIZE;
647 free = stack_not_used(current);
649 if (free >= lowest_to_date)
652 spin_lock(&low_water_lock);
653 if (free < lowest_to_date) {
654 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
655 current->comm, task_pid_nr(current), free);
656 lowest_to_date = free;
658 spin_unlock(&low_water_lock);
661 static inline void check_stack_usage(void) {}
664 void do_exit(long code)
666 struct task_struct *tsk = current;
668 TASKS_RCU(int tasks_rcu_i);
670 profile_task_exit(tsk);
672 WARN_ON(blk_needs_flush_plug(tsk));
674 if (unlikely(in_interrupt()))
675 panic("Aiee, killing interrupt handler!");
676 if (unlikely(!tsk->pid))
677 panic("Attempted to kill the idle task!");
680 * If do_exit is called because this processes oopsed, it's possible
681 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
682 * continuing. Amongst other possible reasons, this is to prevent
683 * mm_release()->clear_child_tid() from writing to a user-controlled
688 ptrace_event(PTRACE_EVENT_EXIT, code);
690 validate_creds_for_do_exit(tsk);
693 * We're taking recursive faults here in do_exit. Safest is to just
694 * leave this task alone and wait for reboot.
696 if (unlikely(tsk->flags & PF_EXITING)) {
697 pr_alert("Fixing recursive fault but reboot is needed!\n");
698 futex_exit_recursive(tsk);
699 set_current_state(TASK_UNINTERRUPTIBLE);
703 exit_signals(tsk); /* sets PF_EXITING */
705 if (unlikely(in_atomic())) {
706 pr_info("note: %s[%d] exited with preempt_count %d\n",
707 current->comm, task_pid_nr(current),
709 preempt_count_set(PREEMPT_ENABLED);
712 /* sync mm's RSS info before statistics gathering */
714 sync_mm_rss(tsk->mm);
715 acct_update_integrals(tsk);
716 group_dead = atomic_dec_and_test(&tsk->signal->live);
718 hrtimer_cancel(&tsk->signal->real_timer);
719 exit_itimers(tsk->signal);
721 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
723 acct_collect(code, group_dead);
728 tsk->exit_code = code;
729 taskstats_exit(tsk, group_dead);
735 trace_sched_process_exit(tsk);
742 disassociate_ctty(1);
743 exit_task_namespaces(tsk);
748 * Flush inherited counters to the parent - before the parent
749 * gets woken up by child-exit notifications.
751 * because of cgroup mode, must be called before cgroup_exit()
753 perf_event_exit_task(tsk);
758 * FIXME: do that only when needed, using sched_exit tracepoint
760 flush_ptrace_hw_breakpoint(tsk);
762 TASKS_RCU(preempt_disable());
763 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
764 TASKS_RCU(preempt_enable());
765 exit_notify(tsk, group_dead);
766 proc_exit_connector(tsk);
769 mpol_put(tsk->mempolicy);
770 tsk->mempolicy = NULL;
774 if (unlikely(current->pi_state_cache))
775 kfree(current->pi_state_cache);
778 * Make sure we are holding no locks:
780 debug_check_no_locks_held();
783 exit_io_context(tsk);
785 if (tsk->splice_pipe)
786 free_pipe_info(tsk->splice_pipe);
788 if (tsk->task_frag.page)
789 put_page(tsk->task_frag.page);
791 validate_creds_for_do_exit(tsk);
796 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
798 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
801 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
802 * when the following two conditions become true.
803 * - There is race condition of mmap_sem (It is acquired by
805 * - SMI occurs before setting TASK_RUNINNG.
806 * (or hypervisor of virtual machine switches to other guest)
807 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
809 * To avoid it, we have to wait for releasing tsk->pi_lock which
810 * is held by try_to_wake_up()
813 raw_spin_unlock_wait(&tsk->pi_lock);
815 /* causes final put_task_struct in finish_task_switch(). */
816 tsk->state = TASK_DEAD;
817 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
820 /* Avoid "noreturn function does return". */
822 cpu_relax(); /* For when BUG is null */
824 EXPORT_SYMBOL_GPL(do_exit);
826 void complete_and_exit(struct completion *comp, long code)
833 EXPORT_SYMBOL(complete_and_exit);
835 SYSCALL_DEFINE1(exit, int, error_code)
837 do_exit((error_code&0xff)<<8);
841 * Take down every thread in the group. This is called by fatal signals
842 * as well as by sys_exit_group (below).
845 do_group_exit(int exit_code)
847 struct signal_struct *sig = current->signal;
849 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
851 if (signal_group_exit(sig))
852 exit_code = sig->group_exit_code;
853 else if (!thread_group_empty(current)) {
854 struct sighand_struct *const sighand = current->sighand;
856 spin_lock_irq(&sighand->siglock);
857 if (signal_group_exit(sig))
858 /* Another thread got here before we took the lock. */
859 exit_code = sig->group_exit_code;
861 sig->group_exit_code = exit_code;
862 sig->flags = SIGNAL_GROUP_EXIT;
863 zap_other_threads(current);
865 spin_unlock_irq(&sighand->siglock);
873 * this kills every thread in the thread group. Note that any externally
874 * wait4()-ing process will get the correct exit code - even if this
875 * thread is not the thread group leader.
877 SYSCALL_DEFINE1(exit_group, int, error_code)
879 do_group_exit((error_code & 0xff) << 8);
885 enum pid_type wo_type;
889 struct siginfo __user *wo_info;
891 struct rusage __user *wo_rusage;
893 wait_queue_t child_wait;
898 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
900 if (type != PIDTYPE_PID)
901 task = task->group_leader;
902 return task->pids[type].pid;
905 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
907 return wo->wo_type == PIDTYPE_MAX ||
908 task_pid_type(p, wo->wo_type) == wo->wo_pid;
912 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
914 if (!eligible_pid(wo, p))
918 * Wait for all children (clone and not) if __WALL is set or
919 * if it is traced by us.
921 if (ptrace || (wo->wo_flags & __WALL))
925 * Otherwise, wait for clone children *only* if __WCLONE is set;
926 * otherwise, wait for non-clone children *only*.
928 * Note: a "clone" child here is one that reports to its parent
929 * using a signal other than SIGCHLD, or a non-leader thread which
930 * we can only see if it is traced by us.
932 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
938 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
939 pid_t pid, uid_t uid, int why, int status)
941 struct siginfo __user *infop;
942 int retval = wo->wo_rusage
943 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
949 retval = put_user(SIGCHLD, &infop->si_signo);
951 retval = put_user(0, &infop->si_errno);
953 retval = put_user((short)why, &infop->si_code);
955 retval = put_user(pid, &infop->si_pid);
957 retval = put_user(uid, &infop->si_uid);
959 retval = put_user(status, &infop->si_status);
967 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
968 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
969 * the lock and this task is uninteresting. If we return nonzero, we have
970 * released the lock and the system call should return.
972 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
974 int state, retval, status;
975 pid_t pid = task_pid_vnr(p);
976 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
977 struct siginfo __user *infop;
979 if (!likely(wo->wo_flags & WEXITED))
982 if (unlikely(wo->wo_flags & WNOWAIT)) {
983 int exit_code = p->exit_code;
987 read_unlock(&tasklist_lock);
988 sched_annotate_sleep();
990 if ((exit_code & 0x7f) == 0) {
992 status = exit_code >> 8;
994 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
995 status = exit_code & 0x7f;
997 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1000 * Move the task's state to DEAD/TRACE, only one thread can do this.
1002 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1003 EXIT_TRACE : EXIT_DEAD;
1004 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1007 * We own this thread, nobody else can reap it.
1009 read_unlock(&tasklist_lock);
1010 sched_annotate_sleep();
1013 * Check thread_group_leader() to exclude the traced sub-threads.
1015 if (state == EXIT_DEAD && thread_group_leader(p)) {
1016 struct signal_struct *sig = p->signal;
1017 struct signal_struct *psig = current->signal;
1018 unsigned long maxrss;
1019 cputime_t tgutime, tgstime;
1022 * The resource counters for the group leader are in its
1023 * own task_struct. Those for dead threads in the group
1024 * are in its signal_struct, as are those for the child
1025 * processes it has previously reaped. All these
1026 * accumulate in the parent's signal_struct c* fields.
1028 * We don't bother to take a lock here to protect these
1029 * p->signal fields because the whole thread group is dead
1030 * and nobody can change them.
1032 * psig->stats_lock also protects us from our sub-theads
1033 * which can reap other children at the same time. Until
1034 * we change k_getrusage()-like users to rely on this lock
1035 * we have to take ->siglock as well.
1037 * We use thread_group_cputime_adjusted() to get times for
1038 * the thread group, which consolidates times for all threads
1039 * in the group including the group leader.
1041 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1042 spin_lock_irq(¤t->sighand->siglock);
1043 write_seqlock(&psig->stats_lock);
1044 psig->cutime += tgutime + sig->cutime;
1045 psig->cstime += tgstime + sig->cstime;
1046 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1048 p->min_flt + sig->min_flt + sig->cmin_flt;
1050 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1052 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1054 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1056 task_io_get_inblock(p) +
1057 sig->inblock + sig->cinblock;
1059 task_io_get_oublock(p) +
1060 sig->oublock + sig->coublock;
1061 maxrss = max(sig->maxrss, sig->cmaxrss);
1062 if (psig->cmaxrss < maxrss)
1063 psig->cmaxrss = maxrss;
1064 task_io_accounting_add(&psig->ioac, &p->ioac);
1065 task_io_accounting_add(&psig->ioac, &sig->ioac);
1066 write_sequnlock(&psig->stats_lock);
1067 spin_unlock_irq(¤t->sighand->siglock);
1070 retval = wo->wo_rusage
1071 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1072 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1073 ? p->signal->group_exit_code : p->exit_code;
1074 if (!retval && wo->wo_stat)
1075 retval = put_user(status, wo->wo_stat);
1077 infop = wo->wo_info;
1078 if (!retval && infop)
1079 retval = put_user(SIGCHLD, &infop->si_signo);
1080 if (!retval && infop)
1081 retval = put_user(0, &infop->si_errno);
1082 if (!retval && infop) {
1085 if ((status & 0x7f) == 0) {
1089 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1092 retval = put_user((short)why, &infop->si_code);
1094 retval = put_user(status, &infop->si_status);
1096 if (!retval && infop)
1097 retval = put_user(pid, &infop->si_pid);
1098 if (!retval && infop)
1099 retval = put_user(uid, &infop->si_uid);
1103 if (state == EXIT_TRACE) {
1104 write_lock_irq(&tasklist_lock);
1105 /* We dropped tasklist, ptracer could die and untrace */
1108 /* If parent wants a zombie, don't release it now */
1109 state = EXIT_ZOMBIE;
1110 if (do_notify_parent(p, p->exit_signal))
1112 p->exit_state = state;
1113 write_unlock_irq(&tasklist_lock);
1115 if (state == EXIT_DEAD)
1121 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1124 if (task_is_stopped_or_traced(p) &&
1125 !(p->jobctl & JOBCTL_LISTENING))
1126 return &p->exit_code;
1128 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1129 return &p->signal->group_exit_code;
1135 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1137 * @ptrace: is the wait for ptrace
1138 * @p: task to wait for
1140 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1143 * read_lock(&tasklist_lock), which is released if return value is
1144 * non-zero. Also, grabs and releases @p->sighand->siglock.
1147 * 0 if wait condition didn't exist and search for other wait conditions
1148 * should continue. Non-zero return, -errno on failure and @p's pid on
1149 * success, implies that tasklist_lock is released and wait condition
1150 * search should terminate.
1152 static int wait_task_stopped(struct wait_opts *wo,
1153 int ptrace, struct task_struct *p)
1155 struct siginfo __user *infop;
1156 int retval, exit_code, *p_code, why;
1157 uid_t uid = 0; /* unneeded, required by compiler */
1161 * Traditionally we see ptrace'd stopped tasks regardless of options.
1163 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1166 if (!task_stopped_code(p, ptrace))
1170 spin_lock_irq(&p->sighand->siglock);
1172 p_code = task_stopped_code(p, ptrace);
1173 if (unlikely(!p_code))
1176 exit_code = *p_code;
1180 if (!unlikely(wo->wo_flags & WNOWAIT))
1183 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1185 spin_unlock_irq(&p->sighand->siglock);
1190 * Now we are pretty sure this task is interesting.
1191 * Make sure it doesn't get reaped out from under us while we
1192 * give up the lock and then examine it below. We don't want to
1193 * keep holding onto the tasklist_lock while we call getrusage and
1194 * possibly take page faults for user memory.
1197 pid = task_pid_vnr(p);
1198 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1199 read_unlock(&tasklist_lock);
1200 sched_annotate_sleep();
1202 if (unlikely(wo->wo_flags & WNOWAIT))
1203 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1205 retval = wo->wo_rusage
1206 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1207 if (!retval && wo->wo_stat)
1208 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1210 infop = wo->wo_info;
1211 if (!retval && infop)
1212 retval = put_user(SIGCHLD, &infop->si_signo);
1213 if (!retval && infop)
1214 retval = put_user(0, &infop->si_errno);
1215 if (!retval && infop)
1216 retval = put_user((short)why, &infop->si_code);
1217 if (!retval && infop)
1218 retval = put_user(exit_code, &infop->si_status);
1219 if (!retval && infop)
1220 retval = put_user(pid, &infop->si_pid);
1221 if (!retval && infop)
1222 retval = put_user(uid, &infop->si_uid);
1232 * Handle do_wait work for one task in a live, non-stopped state.
1233 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1234 * the lock and this task is uninteresting. If we return nonzero, we have
1235 * released the lock and the system call should return.
1237 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1243 if (!unlikely(wo->wo_flags & WCONTINUED))
1246 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1249 spin_lock_irq(&p->sighand->siglock);
1250 /* Re-check with the lock held. */
1251 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1252 spin_unlock_irq(&p->sighand->siglock);
1255 if (!unlikely(wo->wo_flags & WNOWAIT))
1256 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1257 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1258 spin_unlock_irq(&p->sighand->siglock);
1260 pid = task_pid_vnr(p);
1262 read_unlock(&tasklist_lock);
1263 sched_annotate_sleep();
1266 retval = wo->wo_rusage
1267 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1269 if (!retval && wo->wo_stat)
1270 retval = put_user(0xffff, wo->wo_stat);
1274 retval = wait_noreap_copyout(wo, p, pid, uid,
1275 CLD_CONTINUED, SIGCONT);
1276 BUG_ON(retval == 0);
1283 * Consider @p for a wait by @parent.
1285 * -ECHILD should be in ->notask_error before the first call.
1286 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1287 * Returns zero if the search for a child should continue;
1288 * then ->notask_error is 0 if @p is an eligible child,
1289 * or another error from security_task_wait(), or still -ECHILD.
1291 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1292 struct task_struct *p)
1295 * We can race with wait_task_zombie() from another thread.
1296 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1297 * can't confuse the checks below.
1299 int exit_state = ACCESS_ONCE(p->exit_state);
1302 if (unlikely(exit_state == EXIT_DEAD))
1305 ret = eligible_child(wo, ptrace, p);
1309 ret = security_task_wait(p);
1310 if (unlikely(ret < 0)) {
1312 * If we have not yet seen any eligible child,
1313 * then let this error code replace -ECHILD.
1314 * A permission error will give the user a clue
1315 * to look for security policy problems, rather
1316 * than for mysterious wait bugs.
1318 if (wo->notask_error)
1319 wo->notask_error = ret;
1323 if (unlikely(exit_state == EXIT_TRACE)) {
1325 * ptrace == 0 means we are the natural parent. In this case
1326 * we should clear notask_error, debugger will notify us.
1328 if (likely(!ptrace))
1329 wo->notask_error = 0;
1333 if (likely(!ptrace) && unlikely(p->ptrace)) {
1335 * If it is traced by its real parent's group, just pretend
1336 * the caller is ptrace_do_wait() and reap this child if it
1339 * This also hides group stop state from real parent; otherwise
1340 * a single stop can be reported twice as group and ptrace stop.
1341 * If a ptracer wants to distinguish these two events for its
1342 * own children it should create a separate process which takes
1343 * the role of real parent.
1345 if (!ptrace_reparented(p))
1350 if (exit_state == EXIT_ZOMBIE) {
1351 /* we don't reap group leaders with subthreads */
1352 if (!delay_group_leader(p)) {
1354 * A zombie ptracee is only visible to its ptracer.
1355 * Notification and reaping will be cascaded to the
1356 * real parent when the ptracer detaches.
1358 if (unlikely(ptrace) || likely(!p->ptrace))
1359 return wait_task_zombie(wo, p);
1363 * Allow access to stopped/continued state via zombie by
1364 * falling through. Clearing of notask_error is complex.
1368 * If WEXITED is set, notask_error should naturally be
1369 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1370 * so, if there are live subthreads, there are events to
1371 * wait for. If all subthreads are dead, it's still safe
1372 * to clear - this function will be called again in finite
1373 * amount time once all the subthreads are released and
1374 * will then return without clearing.
1378 * Stopped state is per-task and thus can't change once the
1379 * target task dies. Only continued and exited can happen.
1380 * Clear notask_error if WCONTINUED | WEXITED.
1382 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1383 wo->notask_error = 0;
1386 * @p is alive and it's gonna stop, continue or exit, so
1387 * there always is something to wait for.
1389 wo->notask_error = 0;
1393 * Wait for stopped. Depending on @ptrace, different stopped state
1394 * is used and the two don't interact with each other.
1396 ret = wait_task_stopped(wo, ptrace, p);
1401 * Wait for continued. There's only one continued state and the
1402 * ptracer can consume it which can confuse the real parent. Don't
1403 * use WCONTINUED from ptracer. You don't need or want it.
1405 return wait_task_continued(wo, p);
1409 * Do the work of do_wait() for one thread in the group, @tsk.
1411 * -ECHILD should be in ->notask_error before the first call.
1412 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1413 * Returns zero if the search for a child should continue; then
1414 * ->notask_error is 0 if there were any eligible children,
1415 * or another error from security_task_wait(), or still -ECHILD.
1417 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1419 struct task_struct *p;
1421 list_for_each_entry(p, &tsk->children, sibling) {
1422 int ret = wait_consider_task(wo, 0, p);
1431 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1433 struct task_struct *p;
1435 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1436 int ret = wait_consider_task(wo, 1, p);
1445 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1446 int sync, void *key)
1448 struct wait_opts *wo = container_of(wait, struct wait_opts,
1450 struct task_struct *p = key;
1452 if (!eligible_pid(wo, p))
1455 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1458 return default_wake_function(wait, mode, sync, key);
1461 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1463 __wake_up_sync_key(&parent->signal->wait_chldexit,
1464 TASK_INTERRUPTIBLE, 1, p);
1467 static long do_wait(struct wait_opts *wo)
1469 struct task_struct *tsk;
1472 trace_sched_process_wait(wo->wo_pid);
1474 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1475 wo->child_wait.private = current;
1476 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1479 * If there is nothing that can match our criteria, just get out.
1480 * We will clear ->notask_error to zero if we see any child that
1481 * might later match our criteria, even if we are not able to reap
1484 wo->notask_error = -ECHILD;
1485 if ((wo->wo_type < PIDTYPE_MAX) &&
1486 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1489 set_current_state(TASK_INTERRUPTIBLE);
1490 read_lock(&tasklist_lock);
1493 retval = do_wait_thread(wo, tsk);
1497 retval = ptrace_do_wait(wo, tsk);
1501 if (wo->wo_flags & __WNOTHREAD)
1503 } while_each_thread(current, tsk);
1504 read_unlock(&tasklist_lock);
1507 retval = wo->notask_error;
1508 if (!retval && !(wo->wo_flags & WNOHANG)) {
1509 retval = -ERESTARTSYS;
1510 if (!signal_pending(current)) {
1516 __set_current_state(TASK_RUNNING);
1517 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1521 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1522 infop, int, options, struct rusage __user *, ru)
1524 struct wait_opts wo;
1525 struct pid *pid = NULL;
1529 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1531 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1544 type = PIDTYPE_PGID;
1552 if (type < PIDTYPE_MAX)
1553 pid = find_get_pid(upid);
1557 wo.wo_flags = options;
1567 * For a WNOHANG return, clear out all the fields
1568 * we would set so the user can easily tell the
1572 ret = put_user(0, &infop->si_signo);
1574 ret = put_user(0, &infop->si_errno);
1576 ret = put_user(0, &infop->si_code);
1578 ret = put_user(0, &infop->si_pid);
1580 ret = put_user(0, &infop->si_uid);
1582 ret = put_user(0, &infop->si_status);
1589 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1590 int, options, struct rusage __user *, ru)
1592 struct wait_opts wo;
1593 struct pid *pid = NULL;
1597 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1598 __WNOTHREAD|__WCLONE|__WALL))
1601 /* -INT_MIN is not defined */
1602 if (upid == INT_MIN)
1607 else if (upid < 0) {
1608 type = PIDTYPE_PGID;
1609 pid = find_get_pid(-upid);
1610 } else if (upid == 0) {
1611 type = PIDTYPE_PGID;
1612 pid = get_task_pid(current, PIDTYPE_PGID);
1613 } else /* upid > 0 */ {
1615 pid = find_get_pid(upid);
1620 wo.wo_flags = options | WEXITED;
1622 wo.wo_stat = stat_addr;
1630 #ifdef __ARCH_WANT_SYS_WAITPID
1633 * sys_waitpid() remains for compatibility. waitpid() should be
1634 * implemented by calling sys_wait4() from libc.a.
1636 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1638 return sys_wait4(pid, stat_addr, options, NULL);