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);
119 * Accumulate here the counters for all threads but the
120 * group leader as they die, so they can be added into
121 * the process-wide totals when those are taken.
122 * The group leader stays around as a zombie as long
123 * as there are other threads. When it gets reaped,
124 * the exit.c code will add its counts into these totals.
125 * We won't ever get here for the group leader, since it
126 * will have been the last reference on the signal_struct.
128 task_cputime(tsk, &utime, &stime);
131 sig->gtime += task_gtime(tsk);
132 sig->min_flt += tsk->min_flt;
133 sig->maj_flt += tsk->maj_flt;
134 sig->nvcsw += tsk->nvcsw;
135 sig->nivcsw += tsk->nivcsw;
136 sig->inblock += task_io_get_inblock(tsk);
137 sig->oublock += task_io_get_oublock(tsk);
138 task_io_accounting_add(&sig->ioac, &tsk->ioac);
139 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
143 __unhash_process(tsk, group_dead);
146 * Do this under ->siglock, we can race with another thread
147 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
149 flush_sigqueue(&tsk->pending);
151 spin_unlock(&sighand->siglock);
153 __cleanup_sighand(sighand);
154 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
156 flush_sigqueue(&sig->shared_pending);
161 static void delayed_put_task_struct(struct rcu_head *rhp)
163 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
165 perf_event_delayed_put(tsk);
166 trace_sched_process_free(tsk);
167 put_task_struct(tsk);
171 void release_task(struct task_struct * p)
173 struct task_struct *leader;
176 /* don't need to get the RCU readlock here - the process is dead and
177 * can't be modifying its own credentials. But shut RCU-lockdep up */
179 atomic_dec(&__task_cred(p)->user->processes);
184 write_lock_irq(&tasklist_lock);
185 ptrace_release_task(p);
189 * If we are the last non-leader member of the thread
190 * group, and the leader is zombie, then notify the
191 * group leader's parent process. (if it wants notification.)
194 leader = p->group_leader;
195 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
197 * If we were the last child thread and the leader has
198 * exited already, and the leader's parent ignores SIGCHLD,
199 * then we are the one who should release the leader.
201 zap_leader = do_notify_parent(leader, leader->exit_signal);
203 leader->exit_state = EXIT_DEAD;
206 write_unlock_irq(&tasklist_lock);
208 call_rcu(&p->rcu, delayed_put_task_struct);
211 if (unlikely(zap_leader))
216 * This checks not only the pgrp, but falls back on the pid if no
217 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
220 * The caller must hold rcu lock or the tasklist lock.
222 struct pid *session_of_pgrp(struct pid *pgrp)
224 struct task_struct *p;
225 struct pid *sid = NULL;
227 p = pid_task(pgrp, PIDTYPE_PGID);
229 p = pid_task(pgrp, PIDTYPE_PID);
231 sid = task_session(p);
237 * Determine if a process group is "orphaned", according to the POSIX
238 * definition in 2.2.2.52. Orphaned process groups are not to be affected
239 * by terminal-generated stop signals. Newly orphaned process groups are
240 * to receive a SIGHUP and a SIGCONT.
242 * "I ask you, have you ever known what it is to be an orphan?"
244 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
246 struct task_struct *p;
248 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249 if ((p == ignored_task) ||
250 (p->exit_state && thread_group_empty(p)) ||
251 is_global_init(p->real_parent))
254 if (task_pgrp(p->real_parent) != pgrp &&
255 task_session(p->real_parent) == task_session(p))
257 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
262 int is_current_pgrp_orphaned(void)
266 read_lock(&tasklist_lock);
267 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
268 read_unlock(&tasklist_lock);
273 static bool has_stopped_jobs(struct pid *pgrp)
275 struct task_struct *p;
277 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
278 if (p->signal->flags & SIGNAL_STOP_STOPPED)
280 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
286 * Check to see if any process groups have become orphaned as
287 * a result of our exiting, and if they have any stopped jobs,
288 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
291 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
293 struct pid *pgrp = task_pgrp(tsk);
294 struct task_struct *ignored_task = tsk;
297 /* exit: our father is in a different pgrp than
298 * we are and we were the only connection outside.
300 parent = tsk->real_parent;
302 /* reparent: our child is in a different pgrp than
303 * we are, and it was the only connection outside.
307 if (task_pgrp(parent) != pgrp &&
308 task_session(parent) == task_session(tsk) &&
309 will_become_orphaned_pgrp(pgrp, ignored_task) &&
310 has_stopped_jobs(pgrp)) {
311 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
312 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
318 * A task is exiting. If it owned this mm, find a new owner for the mm.
320 void mm_update_next_owner(struct mm_struct *mm)
322 struct task_struct *c, *g, *p = current;
326 * If the exiting or execing task is not the owner, it's
327 * someone else's problem.
332 * The current owner is exiting/execing and there are no other
333 * candidates. Do not leave the mm pointing to a possibly
334 * freed task structure.
336 if (atomic_read(&mm->mm_users) <= 1) {
341 read_lock(&tasklist_lock);
343 * Search in the children
345 list_for_each_entry(c, &p->children, sibling) {
347 goto assign_new_owner;
351 * Search in the siblings
353 list_for_each_entry(c, &p->real_parent->children, sibling) {
355 goto assign_new_owner;
359 * Search through everything else, we should not get here often.
361 for_each_process(g) {
362 if (g->flags & PF_KTHREAD)
364 for_each_thread(g, c) {
366 goto assign_new_owner;
371 read_unlock(&tasklist_lock);
373 * We found no owner yet mm_users > 1: this implies that we are
374 * most likely racing with swapoff (try_to_unuse()) or /proc or
375 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
384 * The task_lock protects c->mm from changing.
385 * We always want mm->owner->mm == mm
389 * Delay read_unlock() till we have the task_lock()
390 * to ensure that c does not slip away underneath us
392 read_unlock(&tasklist_lock);
402 #endif /* CONFIG_MEMCG */
405 * Turn us into a lazy TLB process if we
408 static void exit_mm(struct task_struct * tsk)
410 struct mm_struct *mm = tsk->mm;
411 struct core_state *core_state;
418 * Serialize with any possible pending coredump.
419 * We must hold mmap_sem around checking core_state
420 * and clearing tsk->mm. The core-inducing thread
421 * will increment ->nr_threads for each thread in the
422 * group with ->mm != NULL.
424 down_read(&mm->mmap_sem);
425 core_state = mm->core_state;
427 struct core_thread self;
428 up_read(&mm->mmap_sem);
431 self.next = xchg(&core_state->dumper.next, &self);
433 * Implies mb(), the result of xchg() must be visible
434 * to core_state->dumper.
436 if (atomic_dec_and_test(&core_state->nr_threads))
437 complete(&core_state->startup);
440 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
441 if (!self.task) /* see coredump_finish() */
443 freezable_schedule();
445 __set_task_state(tsk, TASK_RUNNING);
446 down_read(&mm->mmap_sem);
448 atomic_inc(&mm->mm_count);
449 BUG_ON(mm != tsk->active_mm);
450 /* more a memory barrier than a real lock */
453 up_read(&mm->mmap_sem);
454 enter_lazy_tlb(mm, current);
456 mm_update_next_owner(mm);
461 * When we die, we re-parent all our children, and try to:
462 * 1. give them to another thread in our thread group, if such a member exists
463 * 2. give it to the first ancestor process which prctl'd itself as a
464 * child_subreaper for its children (like a service manager)
465 * 3. give it to the init process (PID 1) in our pid namespace
467 static struct task_struct *find_new_reaper(struct task_struct *father)
468 __releases(&tasklist_lock)
469 __acquires(&tasklist_lock)
471 struct pid_namespace *pid_ns = task_active_pid_ns(father);
472 struct task_struct *thread;
475 while_each_thread(father, thread) {
476 if (thread->flags & PF_EXITING)
478 if (unlikely(pid_ns->child_reaper == father))
479 pid_ns->child_reaper = thread;
483 if (unlikely(pid_ns->child_reaper == father)) {
484 write_unlock_irq(&tasklist_lock);
485 if (unlikely(pid_ns == &init_pid_ns)) {
486 panic("Attempted to kill init! exitcode=0x%08x\n",
487 father->signal->group_exit_code ?:
491 zap_pid_ns_processes(pid_ns);
492 write_lock_irq(&tasklist_lock);
493 } else if (father->signal->has_child_subreaper) {
494 struct task_struct *reaper;
497 * Find the first ancestor marked as child_subreaper.
498 * Note that the code below checks same_thread_group(reaper,
499 * pid_ns->child_reaper). This is what we need to DTRT in a
500 * PID namespace. However we still need the check above, see
501 * http://marc.info/?l=linux-kernel&m=131385460420380
503 for (reaper = father->real_parent;
504 reaper != &init_task;
505 reaper = reaper->real_parent) {
506 if (same_thread_group(reaper, pid_ns->child_reaper))
508 if (!reaper->signal->is_child_subreaper)
512 if (!(thread->flags & PF_EXITING))
514 } while_each_thread(reaper, thread);
518 return pid_ns->child_reaper;
522 * Any that need to be release_task'd are put on the @dead list.
524 static void reparent_leader(struct task_struct *father, struct task_struct *p,
525 struct list_head *dead)
527 list_move_tail(&p->sibling, &p->real_parent->children);
529 if (p->exit_state == EXIT_DEAD)
532 * If this is a threaded reparent there is no need to
533 * notify anyone anything has happened.
535 if (same_thread_group(p->real_parent, father))
538 /* We don't want people slaying init. */
539 p->exit_signal = SIGCHLD;
541 /* If it has exited notify the new parent about this child's death. */
543 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
544 if (do_notify_parent(p, p->exit_signal)) {
545 p->exit_state = EXIT_DEAD;
546 list_move_tail(&p->sibling, dead);
550 kill_orphaned_pgrp(p, father);
553 static void forget_original_parent(struct task_struct *father)
555 struct task_struct *p, *n, *reaper;
556 LIST_HEAD(dead_children);
558 write_lock_irq(&tasklist_lock);
560 * Note that exit_ptrace() and find_new_reaper() might
561 * drop tasklist_lock and reacquire it.
564 reaper = find_new_reaper(father);
566 list_for_each_entry_safe(p, n, &father->children, sibling) {
567 struct task_struct *t = p;
569 t->real_parent = reaper;
570 if (t->parent == father) {
572 t->parent = t->real_parent;
574 if (t->pdeath_signal)
575 group_send_sig_info(t->pdeath_signal,
577 } while_each_thread(p, t);
578 reparent_leader(father, p, &dead_children);
580 write_unlock_irq(&tasklist_lock);
582 BUG_ON(!list_empty(&father->children));
584 list_for_each_entry_safe(p, n, &dead_children, sibling) {
585 list_del_init(&p->sibling);
591 * Send signals to all our closest relatives so that they know
592 * to properly mourn us..
594 static void exit_notify(struct task_struct *tsk, int group_dead)
599 * This does two things:
601 * A. Make init inherit all the child processes
602 * B. Check to see if any process groups have become orphaned
603 * as a result of our exiting, and if they have any stopped
604 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
606 forget_original_parent(tsk);
608 write_lock_irq(&tasklist_lock);
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;
627 /* mt-exec, de_thread() is waiting for group leader */
628 if (unlikely(tsk->signal->notify_count < 0))
629 wake_up_process(tsk->signal->group_exit_task);
630 write_unlock_irq(&tasklist_lock);
632 /* If the process is dead, release it - nobody will wait for it */
637 #ifdef CONFIG_DEBUG_STACK_USAGE
638 static void check_stack_usage(void)
640 static DEFINE_SPINLOCK(low_water_lock);
641 static int lowest_to_date = THREAD_SIZE;
644 free = stack_not_used(current);
646 if (free >= lowest_to_date)
649 spin_lock(&low_water_lock);
650 if (free < lowest_to_date) {
651 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
653 current->comm, task_pid_nr(current), free);
654 lowest_to_date = free;
656 spin_unlock(&low_water_lock);
659 static inline void check_stack_usage(void) {}
662 void do_exit(long code)
664 struct task_struct *tsk = current;
667 profile_task_exit(tsk);
669 WARN_ON(blk_needs_flush_plug(tsk));
671 if (unlikely(in_interrupt()))
672 panic("Aiee, killing interrupt handler!");
673 if (unlikely(!tsk->pid))
674 panic("Attempted to kill the idle task!");
677 * If do_exit is called because this processes oopsed, it's possible
678 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
679 * continuing. Amongst other possible reasons, this is to prevent
680 * mm_release()->clear_child_tid() from writing to a user-controlled
685 ptrace_event(PTRACE_EVENT_EXIT, code);
687 validate_creds_for_do_exit(tsk);
690 * We're taking recursive faults here in do_exit. Safest is to just
691 * leave this task alone and wait for reboot.
693 if (unlikely(tsk->flags & PF_EXITING)) {
695 "Fixing recursive fault but reboot is needed!\n");
697 * We can do this unlocked here. The futex code uses
698 * this flag just to verify whether the pi state
699 * cleanup has been done or not. In the worst case it
700 * loops once more. We pretend that the cleanup was
701 * done as there is no way to return. Either the
702 * OWNER_DIED bit is set by now or we push the blocked
703 * task into the wait for ever nirwana as well.
705 tsk->flags |= PF_EXITPIDONE;
706 set_current_state(TASK_UNINTERRUPTIBLE);
710 exit_signals(tsk); /* sets PF_EXITING */
712 * tsk->flags are checked in the futex code to protect against
713 * an exiting task cleaning up the robust pi futexes.
716 raw_spin_unlock_wait(&tsk->pi_lock);
718 if (unlikely(in_atomic()))
719 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
720 current->comm, task_pid_nr(current),
723 acct_update_integrals(tsk);
724 /* sync mm's RSS info before statistics gathering */
726 sync_mm_rss(tsk->mm);
727 group_dead = atomic_dec_and_test(&tsk->signal->live);
729 hrtimer_cancel(&tsk->signal->real_timer);
730 exit_itimers(tsk->signal);
732 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
734 acct_collect(code, group_dead);
739 tsk->exit_code = code;
740 taskstats_exit(tsk, group_dead);
746 trace_sched_process_exit(tsk);
753 disassociate_ctty(1);
754 exit_task_namespaces(tsk);
759 * Flush inherited counters to the parent - before the parent
760 * gets woken up by child-exit notifications.
762 * because of cgroup mode, must be called before cgroup_exit()
764 perf_event_exit_task(tsk);
768 module_put(task_thread_info(tsk)->exec_domain->module);
771 * FIXME: do that only when needed, using sched_exit tracepoint
773 flush_ptrace_hw_breakpoint(tsk);
775 exit_notify(tsk, group_dead);
776 proc_exit_connector(tsk);
779 mpol_put(tsk->mempolicy);
780 tsk->mempolicy = NULL;
784 if (unlikely(current->pi_state_cache))
785 kfree(current->pi_state_cache);
788 * Make sure we are holding no locks:
790 debug_check_no_locks_held();
792 * We can do this unlocked here. The futex code uses this flag
793 * just to verify whether the pi state cleanup has been done
794 * or not. In the worst case it loops once more.
796 tsk->flags |= PF_EXITPIDONE;
799 exit_io_context(tsk);
801 if (tsk->splice_pipe)
802 free_pipe_info(tsk->splice_pipe);
804 if (tsk->task_frag.page)
805 put_page(tsk->task_frag.page);
807 validate_creds_for_do_exit(tsk);
812 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
816 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
817 * when the following two conditions become true.
818 * - There is race condition of mmap_sem (It is acquired by
820 * - SMI occurs before setting TASK_RUNINNG.
821 * (or hypervisor of virtual machine switches to other guest)
822 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
824 * To avoid it, we have to wait for releasing tsk->pi_lock which
825 * is held by try_to_wake_up()
828 raw_spin_unlock_wait(&tsk->pi_lock);
830 /* causes final put_task_struct in finish_task_switch(). */
831 tsk->state = TASK_DEAD;
832 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
835 /* Avoid "noreturn function does return". */
837 cpu_relax(); /* For when BUG is null */
840 EXPORT_SYMBOL_GPL(do_exit);
842 void complete_and_exit(struct completion *comp, long code)
850 EXPORT_SYMBOL(complete_and_exit);
852 SYSCALL_DEFINE1(exit, int, error_code)
854 do_exit((error_code&0xff)<<8);
858 * Take down every thread in the group. This is called by fatal signals
859 * as well as by sys_exit_group (below).
862 do_group_exit(int exit_code)
864 struct signal_struct *sig = current->signal;
866 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
868 if (signal_group_exit(sig))
869 exit_code = sig->group_exit_code;
870 else if (!thread_group_empty(current)) {
871 struct sighand_struct *const sighand = current->sighand;
872 spin_lock_irq(&sighand->siglock);
873 if (signal_group_exit(sig))
874 /* Another thread got here before we took the lock. */
875 exit_code = sig->group_exit_code;
877 sig->group_exit_code = exit_code;
878 sig->flags = SIGNAL_GROUP_EXIT;
879 zap_other_threads(current);
881 spin_unlock_irq(&sighand->siglock);
889 * this kills every thread in the thread group. Note that any externally
890 * wait4()-ing process will get the correct exit code - even if this
891 * thread is not the thread group leader.
893 SYSCALL_DEFINE1(exit_group, int, error_code)
895 do_group_exit((error_code & 0xff) << 8);
901 enum pid_type wo_type;
905 struct siginfo __user *wo_info;
907 struct rusage __user *wo_rusage;
909 wait_queue_t child_wait;
914 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
916 if (type != PIDTYPE_PID)
917 task = task->group_leader;
918 return task->pids[type].pid;
921 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
923 return wo->wo_type == PIDTYPE_MAX ||
924 task_pid_type(p, wo->wo_type) == wo->wo_pid;
927 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
929 if (!eligible_pid(wo, p))
931 /* Wait for all children (clone and not) if __WALL is set;
932 * otherwise, wait for clone children *only* if __WCLONE is
933 * set; otherwise, wait for non-clone children *only*. (Note:
934 * A "clone" child here is one that reports to its parent
935 * using a signal other than SIGCHLD.) */
936 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
937 && !(wo->wo_flags & __WALL))
943 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
944 pid_t pid, uid_t uid, int why, int status)
946 struct siginfo __user *infop;
947 int retval = wo->wo_rusage
948 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
954 retval = put_user(SIGCHLD, &infop->si_signo);
956 retval = put_user(0, &infop->si_errno);
958 retval = put_user((short)why, &infop->si_code);
960 retval = put_user(pid, &infop->si_pid);
962 retval = put_user(uid, &infop->si_uid);
964 retval = put_user(status, &infop->si_status);
972 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
973 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
974 * the lock and this task is uninteresting. If we return nonzero, we have
975 * released the lock and the system call should return.
977 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
980 int retval, status, traced;
981 pid_t pid = task_pid_vnr(p);
982 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
983 struct siginfo __user *infop;
985 if (!likely(wo->wo_flags & WEXITED))
988 if (unlikely(wo->wo_flags & WNOWAIT)) {
989 int exit_code = p->exit_code;
993 read_unlock(&tasklist_lock);
994 if ((exit_code & 0x7f) == 0) {
996 status = exit_code >> 8;
998 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
999 status = exit_code & 0x7f;
1001 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1004 traced = ptrace_reparented(p);
1006 * Move the task's state to DEAD/TRACE, only one thread can do this.
1008 state = traced && thread_group_leader(p) ? EXIT_TRACE : EXIT_DEAD;
1009 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1012 * It can be ptraced but not reparented, check
1013 * thread_group_leader() to filter out sub-threads.
1015 if (likely(!traced) && thread_group_leader(p)) {
1016 struct signal_struct *psig;
1017 struct signal_struct *sig;
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 they are only touched by
1030 * __exit_signal, which runs with tasklist_lock
1031 * write-locked anyway, and so is excluded here. We do
1032 * need to protect the access to parent->signal fields,
1033 * as other threads in the parent group can be right
1034 * here reaping other children at the same time.
1036 * We use thread_group_cputime_adjusted() to get times for the thread
1037 * group, which consolidates times for all threads in the
1038 * group including the group leader.
1040 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1041 spin_lock_irq(&p->real_parent->sighand->siglock);
1042 psig = p->real_parent->signal;
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 spin_unlock_irq(&p->real_parent->sighand->siglock);
1070 * Now we are sure this task is interesting, and no other
1071 * thread can reap it because we its state == DEAD/TRACE.
1073 read_unlock(&tasklist_lock);
1075 retval = wo->wo_rusage
1076 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1077 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1078 ? p->signal->group_exit_code : p->exit_code;
1079 if (!retval && wo->wo_stat)
1080 retval = put_user(status, wo->wo_stat);
1082 infop = wo->wo_info;
1083 if (!retval && infop)
1084 retval = put_user(SIGCHLD, &infop->si_signo);
1085 if (!retval && infop)
1086 retval = put_user(0, &infop->si_errno);
1087 if (!retval && infop) {
1090 if ((status & 0x7f) == 0) {
1094 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1097 retval = put_user((short)why, &infop->si_code);
1099 retval = put_user(status, &infop->si_status);
1101 if (!retval && infop)
1102 retval = put_user(pid, &infop->si_pid);
1103 if (!retval && infop)
1104 retval = put_user(uid, &infop->si_uid);
1108 if (state == EXIT_TRACE) {
1109 write_lock_irq(&tasklist_lock);
1110 /* We dropped tasklist, ptracer could die and untrace */
1113 /* If parent wants a zombie, don't release it now */
1114 state = EXIT_ZOMBIE;
1115 if (do_notify_parent(p, p->exit_signal))
1117 p->exit_state = state;
1118 write_unlock_irq(&tasklist_lock);
1120 if (state == EXIT_DEAD)
1126 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1129 if (task_is_stopped_or_traced(p) &&
1130 !(p->jobctl & JOBCTL_LISTENING))
1131 return &p->exit_code;
1133 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1134 return &p->signal->group_exit_code;
1140 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1142 * @ptrace: is the wait for ptrace
1143 * @p: task to wait for
1145 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1148 * read_lock(&tasklist_lock), which is released if return value is
1149 * non-zero. Also, grabs and releases @p->sighand->siglock.
1152 * 0 if wait condition didn't exist and search for other wait conditions
1153 * should continue. Non-zero return, -errno on failure and @p's pid on
1154 * success, implies that tasklist_lock is released and wait condition
1155 * search should terminate.
1157 static int wait_task_stopped(struct wait_opts *wo,
1158 int ptrace, struct task_struct *p)
1160 struct siginfo __user *infop;
1161 int retval, exit_code, *p_code, why;
1162 uid_t uid = 0; /* unneeded, required by compiler */
1166 * Traditionally we see ptrace'd stopped tasks regardless of options.
1168 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1171 if (!task_stopped_code(p, ptrace))
1175 spin_lock_irq(&p->sighand->siglock);
1177 p_code = task_stopped_code(p, ptrace);
1178 if (unlikely(!p_code))
1181 exit_code = *p_code;
1185 if (!unlikely(wo->wo_flags & WNOWAIT))
1188 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1190 spin_unlock_irq(&p->sighand->siglock);
1195 * Now we are pretty sure this task is interesting.
1196 * Make sure it doesn't get reaped out from under us while we
1197 * give up the lock and then examine it below. We don't want to
1198 * keep holding onto the tasklist_lock while we call getrusage and
1199 * possibly take page faults for user memory.
1202 pid = task_pid_vnr(p);
1203 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1204 read_unlock(&tasklist_lock);
1206 if (unlikely(wo->wo_flags & WNOWAIT))
1207 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1209 retval = wo->wo_rusage
1210 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1211 if (!retval && wo->wo_stat)
1212 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1214 infop = wo->wo_info;
1215 if (!retval && infop)
1216 retval = put_user(SIGCHLD, &infop->si_signo);
1217 if (!retval && infop)
1218 retval = put_user(0, &infop->si_errno);
1219 if (!retval && infop)
1220 retval = put_user((short)why, &infop->si_code);
1221 if (!retval && infop)
1222 retval = put_user(exit_code, &infop->si_status);
1223 if (!retval && infop)
1224 retval = put_user(pid, &infop->si_pid);
1225 if (!retval && infop)
1226 retval = put_user(uid, &infop->si_uid);
1236 * Handle do_wait work for one task in a live, non-stopped state.
1237 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1238 * the lock and this task is uninteresting. If we return nonzero, we have
1239 * released the lock and the system call should return.
1241 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1247 if (!unlikely(wo->wo_flags & WCONTINUED))
1250 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1253 spin_lock_irq(&p->sighand->siglock);
1254 /* Re-check with the lock held. */
1255 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1256 spin_unlock_irq(&p->sighand->siglock);
1259 if (!unlikely(wo->wo_flags & WNOWAIT))
1260 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1261 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1262 spin_unlock_irq(&p->sighand->siglock);
1264 pid = task_pid_vnr(p);
1266 read_unlock(&tasklist_lock);
1269 retval = wo->wo_rusage
1270 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1272 if (!retval && wo->wo_stat)
1273 retval = put_user(0xffff, wo->wo_stat);
1277 retval = wait_noreap_copyout(wo, p, pid, uid,
1278 CLD_CONTINUED, SIGCONT);
1279 BUG_ON(retval == 0);
1286 * Consider @p for a wait by @parent.
1288 * -ECHILD should be in ->notask_error before the first call.
1289 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1290 * Returns zero if the search for a child should continue;
1291 * then ->notask_error is 0 if @p is an eligible child,
1292 * or another error from security_task_wait(), or still -ECHILD.
1294 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1295 struct task_struct *p)
1299 if (unlikely(p->exit_state == EXIT_DEAD))
1302 ret = eligible_child(wo, p);
1306 ret = security_task_wait(p);
1307 if (unlikely(ret < 0)) {
1309 * If we have not yet seen any eligible child,
1310 * then let this error code replace -ECHILD.
1311 * A permission error will give the user a clue
1312 * to look for security policy problems, rather
1313 * than for mysterious wait bugs.
1315 if (wo->notask_error)
1316 wo->notask_error = ret;
1320 if (unlikely(p->exit_state == EXIT_TRACE)) {
1322 * ptrace == 0 means we are the natural parent. In this case
1323 * we should clear notask_error, debugger will notify us.
1325 if (likely(!ptrace))
1326 wo->notask_error = 0;
1330 if (likely(!ptrace) && unlikely(p->ptrace)) {
1332 * If it is traced by its real parent's group, just pretend
1333 * the caller is ptrace_do_wait() and reap this child if it
1336 * This also hides group stop state from real parent; otherwise
1337 * a single stop can be reported twice as group and ptrace stop.
1338 * If a ptracer wants to distinguish these two events for its
1339 * own children it should create a separate process which takes
1340 * the role of real parent.
1342 if (!ptrace_reparented(p))
1347 if (p->exit_state == EXIT_ZOMBIE) {
1348 /* we don't reap group leaders with subthreads */
1349 if (!delay_group_leader(p)) {
1351 * A zombie ptracee is only visible to its ptracer.
1352 * Notification and reaping will be cascaded to the
1353 * real parent when the ptracer detaches.
1355 if (unlikely(ptrace) || likely(!p->ptrace))
1356 return wait_task_zombie(wo, p);
1360 * Allow access to stopped/continued state via zombie by
1361 * falling through. Clearing of notask_error is complex.
1365 * If WEXITED is set, notask_error should naturally be
1366 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1367 * so, if there are live subthreads, there are events to
1368 * wait for. If all subthreads are dead, it's still safe
1369 * to clear - this function will be called again in finite
1370 * amount time once all the subthreads are released and
1371 * will then return without clearing.
1375 * Stopped state is per-task and thus can't change once the
1376 * target task dies. Only continued and exited can happen.
1377 * Clear notask_error if WCONTINUED | WEXITED.
1379 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1380 wo->notask_error = 0;
1383 * @p is alive and it's gonna stop, continue or exit, so
1384 * there always is something to wait for.
1386 wo->notask_error = 0;
1390 * Wait for stopped. Depending on @ptrace, different stopped state
1391 * is used and the two don't interact with each other.
1393 ret = wait_task_stopped(wo, ptrace, p);
1398 * Wait for continued. There's only one continued state and the
1399 * ptracer can consume it which can confuse the real parent. Don't
1400 * use WCONTINUED from ptracer. You don't need or want it.
1402 return wait_task_continued(wo, p);
1406 * Do the work of do_wait() for one thread in the group, @tsk.
1408 * -ECHILD should be in ->notask_error before the first call.
1409 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1410 * Returns zero if the search for a child should continue; then
1411 * ->notask_error is 0 if there were any eligible children,
1412 * or another error from security_task_wait(), or still -ECHILD.
1414 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1416 struct task_struct *p;
1418 list_for_each_entry(p, &tsk->children, sibling) {
1419 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);
1440 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1441 int sync, void *key)
1443 struct wait_opts *wo = container_of(wait, struct wait_opts,
1445 struct task_struct *p = key;
1447 if (!eligible_pid(wo, p))
1450 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1453 return default_wake_function(wait, mode, sync, key);
1456 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1458 __wake_up_sync_key(&parent->signal->wait_chldexit,
1459 TASK_INTERRUPTIBLE, 1, p);
1462 static long do_wait(struct wait_opts *wo)
1464 struct task_struct *tsk;
1467 trace_sched_process_wait(wo->wo_pid);
1469 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1470 wo->child_wait.private = current;
1471 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1474 * If there is nothing that can match our critiera just get out.
1475 * We will clear ->notask_error to zero if we see any child that
1476 * might later match our criteria, even if we are not able to reap
1479 wo->notask_error = -ECHILD;
1480 if ((wo->wo_type < PIDTYPE_MAX) &&
1481 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1484 set_current_state(TASK_INTERRUPTIBLE);
1485 read_lock(&tasklist_lock);
1488 retval = do_wait_thread(wo, tsk);
1492 retval = ptrace_do_wait(wo, tsk);
1496 if (wo->wo_flags & __WNOTHREAD)
1498 } while_each_thread(current, tsk);
1499 read_unlock(&tasklist_lock);
1502 retval = wo->notask_error;
1503 if (!retval && !(wo->wo_flags & WNOHANG)) {
1504 retval = -ERESTARTSYS;
1505 if (!signal_pending(current)) {
1511 __set_current_state(TASK_RUNNING);
1512 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1516 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1517 infop, int, options, struct rusage __user *, ru)
1519 struct wait_opts wo;
1520 struct pid *pid = NULL;
1524 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1526 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1539 type = PIDTYPE_PGID;
1547 if (type < PIDTYPE_MAX)
1548 pid = find_get_pid(upid);
1552 wo.wo_flags = options;
1562 * For a WNOHANG return, clear out all the fields
1563 * we would set so the user can easily tell the
1567 ret = put_user(0, &infop->si_signo);
1569 ret = put_user(0, &infop->si_errno);
1571 ret = put_user(0, &infop->si_code);
1573 ret = put_user(0, &infop->si_pid);
1575 ret = put_user(0, &infop->si_uid);
1577 ret = put_user(0, &infop->si_status);
1584 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1585 int, options, struct rusage __user *, ru)
1587 struct wait_opts wo;
1588 struct pid *pid = NULL;
1592 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1593 __WNOTHREAD|__WCLONE|__WALL))
1598 else if (upid < 0) {
1599 type = PIDTYPE_PGID;
1600 pid = find_get_pid(-upid);
1601 } else if (upid == 0) {
1602 type = PIDTYPE_PGID;
1603 pid = get_task_pid(current, PIDTYPE_PGID);
1604 } else /* upid > 0 */ {
1606 pid = find_get_pid(upid);
1611 wo.wo_flags = options | WEXITED;
1613 wo.wo_stat = stat_addr;
1621 #ifdef __ARCH_WANT_SYS_WAITPID
1624 * sys_waitpid() remains for compatibility. waitpid() should be
1625 * implemented by calling sys_wait4() from libc.a.
1627 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1629 return sys_wait4(pid, stat_addr, options, NULL);