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);
458 clear_thread_flag(TIF_MEMDIE);
462 * When we die, we re-parent all our children, and try to:
463 * 1. give them to another thread in our thread group, if such a member exists
464 * 2. give it to the first ancestor process which prctl'd itself as a
465 * child_subreaper for its children (like a service manager)
466 * 3. give it to the init process (PID 1) in our pid namespace
468 static struct task_struct *find_new_reaper(struct task_struct *father)
469 __releases(&tasklist_lock)
470 __acquires(&tasklist_lock)
472 struct pid_namespace *pid_ns = task_active_pid_ns(father);
473 struct task_struct *thread;
476 while_each_thread(father, thread) {
477 if (thread->flags & PF_EXITING)
479 if (unlikely(pid_ns->child_reaper == father))
480 pid_ns->child_reaper = thread;
484 if (unlikely(pid_ns->child_reaper == father)) {
485 write_unlock_irq(&tasklist_lock);
486 if (unlikely(pid_ns == &init_pid_ns)) {
487 panic("Attempted to kill init! exitcode=0x%08x\n",
488 father->signal->group_exit_code ?:
492 zap_pid_ns_processes(pid_ns);
493 write_lock_irq(&tasklist_lock);
494 } else if (father->signal->has_child_subreaper) {
495 struct task_struct *reaper;
498 * Find the first ancestor marked as child_subreaper.
499 * Note that the code below checks same_thread_group(reaper,
500 * pid_ns->child_reaper). This is what we need to DTRT in a
501 * PID namespace. However we still need the check above, see
502 * http://marc.info/?l=linux-kernel&m=131385460420380
504 for (reaper = father->real_parent;
505 reaper != &init_task;
506 reaper = reaper->real_parent) {
507 if (same_thread_group(reaper, pid_ns->child_reaper))
509 if (!reaper->signal->is_child_subreaper)
513 if (!(thread->flags & PF_EXITING))
515 } while_each_thread(reaper, thread);
519 return pid_ns->child_reaper;
523 * Any that need to be release_task'd are put on the @dead list.
525 static void reparent_leader(struct task_struct *father, struct task_struct *p,
526 struct list_head *dead)
528 list_move_tail(&p->sibling, &p->real_parent->children);
530 if (p->exit_state == EXIT_DEAD)
533 * If this is a threaded reparent there is no need to
534 * notify anyone anything has happened.
536 if (same_thread_group(p->real_parent, father))
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_move_tail(&p->sibling, dead);
551 kill_orphaned_pgrp(p, father);
554 static void forget_original_parent(struct task_struct *father)
556 struct task_struct *p, *n, *reaper;
557 LIST_HEAD(dead_children);
559 write_lock_irq(&tasklist_lock);
561 * Note that exit_ptrace() and find_new_reaper() might
562 * drop tasklist_lock and reacquire it.
565 reaper = find_new_reaper(father);
567 list_for_each_entry_safe(p, n, &father->children, sibling) {
568 struct task_struct *t = p;
570 t->real_parent = reaper;
571 if (t->parent == father) {
573 t->parent = t->real_parent;
575 if (t->pdeath_signal)
576 group_send_sig_info(t->pdeath_signal,
578 } while_each_thread(p, t);
579 reparent_leader(father, p, &dead_children);
581 write_unlock_irq(&tasklist_lock);
583 BUG_ON(!list_empty(&father->children));
585 list_for_each_entry_safe(p, n, &dead_children, sibling) {
586 list_del_init(&p->sibling);
592 * Send signals to all our closest relatives so that they know
593 * to properly mourn us..
595 static void exit_notify(struct task_struct *tsk, int group_dead)
600 * This does two things:
602 * A. Make init inherit all the child processes
603 * B. Check to see if any process groups have become orphaned
604 * as a result of our exiting, and if they have any stopped
605 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
607 forget_original_parent(tsk);
609 write_lock_irq(&tasklist_lock);
611 kill_orphaned_pgrp(tsk->group_leader, NULL);
613 if (unlikely(tsk->ptrace)) {
614 int sig = thread_group_leader(tsk) &&
615 thread_group_empty(tsk) &&
616 !ptrace_reparented(tsk) ?
617 tsk->exit_signal : SIGCHLD;
618 autoreap = do_notify_parent(tsk, sig);
619 } else if (thread_group_leader(tsk)) {
620 autoreap = thread_group_empty(tsk) &&
621 do_notify_parent(tsk, tsk->exit_signal);
626 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
628 /* mt-exec, de_thread() is waiting for group leader */
629 if (unlikely(tsk->signal->notify_count < 0))
630 wake_up_process(tsk->signal->group_exit_task);
631 write_unlock_irq(&tasklist_lock);
633 /* If the process is dead, release it - nobody will wait for it */
638 #ifdef CONFIG_DEBUG_STACK_USAGE
639 static void check_stack_usage(void)
641 static DEFINE_SPINLOCK(low_water_lock);
642 static int lowest_to_date = THREAD_SIZE;
645 free = stack_not_used(current);
647 if (free >= lowest_to_date)
650 spin_lock(&low_water_lock);
651 if (free < lowest_to_date) {
652 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
654 current->comm, task_pid_nr(current), free);
655 lowest_to_date = free;
657 spin_unlock(&low_water_lock);
660 static inline void check_stack_usage(void) {}
663 void do_exit(long code)
665 struct task_struct *tsk = current;
668 profile_task_exit(tsk);
670 WARN_ON(blk_needs_flush_plug(tsk));
672 if (unlikely(in_interrupt()))
673 panic("Aiee, killing interrupt handler!");
674 if (unlikely(!tsk->pid))
675 panic("Attempted to kill the idle task!");
678 * If do_exit is called because this processes oopsed, it's possible
679 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
680 * continuing. Amongst other possible reasons, this is to prevent
681 * mm_release()->clear_child_tid() from writing to a user-controlled
686 ptrace_event(PTRACE_EVENT_EXIT, code);
688 validate_creds_for_do_exit(tsk);
691 * We're taking recursive faults here in do_exit. Safest is to just
692 * leave this task alone and wait for reboot.
694 if (unlikely(tsk->flags & PF_EXITING)) {
696 "Fixing recursive fault but reboot is needed!\n");
698 * We can do this unlocked here. The futex code uses
699 * this flag just to verify whether the pi state
700 * cleanup has been done or not. In the worst case it
701 * loops once more. We pretend that the cleanup was
702 * done as there is no way to return. Either the
703 * OWNER_DIED bit is set by now or we push the blocked
704 * task into the wait for ever nirwana as well.
706 tsk->flags |= PF_EXITPIDONE;
707 set_current_state(TASK_UNINTERRUPTIBLE);
711 exit_signals(tsk); /* sets PF_EXITING */
713 * tsk->flags are checked in the futex code to protect against
714 * an exiting task cleaning up the robust pi futexes.
717 raw_spin_unlock_wait(&tsk->pi_lock);
719 if (unlikely(in_atomic()))
720 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
721 current->comm, task_pid_nr(current),
724 acct_update_integrals(tsk);
725 /* sync mm's RSS info before statistics gathering */
727 sync_mm_rss(tsk->mm);
728 group_dead = atomic_dec_and_test(&tsk->signal->live);
730 hrtimer_cancel(&tsk->signal->real_timer);
731 exit_itimers(tsk->signal);
733 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
735 acct_collect(code, group_dead);
740 tsk->exit_code = code;
741 taskstats_exit(tsk, group_dead);
747 trace_sched_process_exit(tsk);
754 disassociate_ctty(1);
755 exit_task_namespaces(tsk);
760 * Flush inherited counters to the parent - before the parent
761 * gets woken up by child-exit notifications.
763 * because of cgroup mode, must be called before cgroup_exit()
765 perf_event_exit_task(tsk);
769 module_put(task_thread_info(tsk)->exec_domain->module);
772 * FIXME: do that only when needed, using sched_exit tracepoint
774 flush_ptrace_hw_breakpoint(tsk);
776 exit_notify(tsk, group_dead);
777 proc_exit_connector(tsk);
780 mpol_put(tsk->mempolicy);
781 tsk->mempolicy = NULL;
785 if (unlikely(current->pi_state_cache))
786 kfree(current->pi_state_cache);
789 * Make sure we are holding no locks:
791 debug_check_no_locks_held();
793 * We can do this unlocked here. The futex code uses this flag
794 * just to verify whether the pi state cleanup has been done
795 * or not. In the worst case it loops once more.
797 tsk->flags |= PF_EXITPIDONE;
800 exit_io_context(tsk);
802 if (tsk->splice_pipe)
803 free_pipe_info(tsk->splice_pipe);
805 if (tsk->task_frag.page)
806 put_page(tsk->task_frag.page);
808 validate_creds_for_do_exit(tsk);
813 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
817 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
818 * when the following two conditions become true.
819 * - There is race condition of mmap_sem (It is acquired by
821 * - SMI occurs before setting TASK_RUNINNG.
822 * (or hypervisor of virtual machine switches to other guest)
823 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
825 * To avoid it, we have to wait for releasing tsk->pi_lock which
826 * is held by try_to_wake_up()
829 raw_spin_unlock_wait(&tsk->pi_lock);
831 /* causes final put_task_struct in finish_task_switch(). */
832 tsk->state = TASK_DEAD;
833 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
836 /* Avoid "noreturn function does return". */
838 cpu_relax(); /* For when BUG is null */
841 EXPORT_SYMBOL_GPL(do_exit);
843 void complete_and_exit(struct completion *comp, long code)
851 EXPORT_SYMBOL(complete_and_exit);
853 SYSCALL_DEFINE1(exit, int, error_code)
855 do_exit((error_code&0xff)<<8);
859 * Take down every thread in the group. This is called by fatal signals
860 * as well as by sys_exit_group (below).
863 do_group_exit(int exit_code)
865 struct signal_struct *sig = current->signal;
867 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
869 if (signal_group_exit(sig))
870 exit_code = sig->group_exit_code;
871 else if (!thread_group_empty(current)) {
872 struct sighand_struct *const sighand = current->sighand;
873 spin_lock_irq(&sighand->siglock);
874 if (signal_group_exit(sig))
875 /* Another thread got here before we took the lock. */
876 exit_code = sig->group_exit_code;
878 sig->group_exit_code = exit_code;
879 sig->flags = SIGNAL_GROUP_EXIT;
880 zap_other_threads(current);
882 spin_unlock_irq(&sighand->siglock);
890 * this kills every thread in the thread group. Note that any externally
891 * wait4()-ing process will get the correct exit code - even if this
892 * thread is not the thread group leader.
894 SYSCALL_DEFINE1(exit_group, int, error_code)
896 do_group_exit((error_code & 0xff) << 8);
902 enum pid_type wo_type;
906 struct siginfo __user *wo_info;
908 struct rusage __user *wo_rusage;
910 wait_queue_t child_wait;
915 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
917 if (type != PIDTYPE_PID)
918 task = task->group_leader;
919 return task->pids[type].pid;
922 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
924 return wo->wo_type == PIDTYPE_MAX ||
925 task_pid_type(p, wo->wo_type) == wo->wo_pid;
928 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
930 if (!eligible_pid(wo, p))
932 /* Wait for all children (clone and not) if __WALL is set;
933 * otherwise, wait for clone children *only* if __WCLONE is
934 * set; otherwise, wait for non-clone children *only*. (Note:
935 * A "clone" child here is one that reports to its parent
936 * using a signal other than SIGCHLD.) */
937 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
938 && !(wo->wo_flags & __WALL))
944 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
945 pid_t pid, uid_t uid, int why, int status)
947 struct siginfo __user *infop;
948 int retval = wo->wo_rusage
949 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
955 retval = put_user(SIGCHLD, &infop->si_signo);
957 retval = put_user(0, &infop->si_errno);
959 retval = put_user((short)why, &infop->si_code);
961 retval = put_user(pid, &infop->si_pid);
963 retval = put_user(uid, &infop->si_uid);
965 retval = put_user(status, &infop->si_status);
973 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
974 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
975 * the lock and this task is uninteresting. If we return nonzero, we have
976 * released the lock and the system call should return.
978 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
981 int retval, status, traced;
982 pid_t pid = task_pid_vnr(p);
983 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
984 struct siginfo __user *infop;
986 if (!likely(wo->wo_flags & WEXITED))
989 if (unlikely(wo->wo_flags & WNOWAIT)) {
990 int exit_code = p->exit_code;
994 read_unlock(&tasklist_lock);
995 if ((exit_code & 0x7f) == 0) {
997 status = exit_code >> 8;
999 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1000 status = exit_code & 0x7f;
1002 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1005 traced = ptrace_reparented(p);
1007 * Move the task's state to DEAD/TRACE, only one thread can do this.
1009 state = traced && thread_group_leader(p) ? EXIT_TRACE : EXIT_DEAD;
1010 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1013 * It can be ptraced but not reparented, check
1014 * thread_group_leader() to filter out sub-threads.
1016 if (likely(!traced) && thread_group_leader(p)) {
1017 struct signal_struct *psig;
1018 struct signal_struct *sig;
1019 unsigned long maxrss;
1020 cputime_t tgutime, tgstime;
1023 * The resource counters for the group leader are in its
1024 * own task_struct. Those for dead threads in the group
1025 * are in its signal_struct, as are those for the child
1026 * processes it has previously reaped. All these
1027 * accumulate in the parent's signal_struct c* fields.
1029 * We don't bother to take a lock here to protect these
1030 * p->signal fields, because they are only touched by
1031 * __exit_signal, which runs with tasklist_lock
1032 * write-locked anyway, and so is excluded here. We do
1033 * need to protect the access to parent->signal fields,
1034 * as other threads in the parent group can be right
1035 * here reaping other children at the same time.
1037 * We use thread_group_cputime_adjusted() to get times for the thread
1038 * group, which consolidates times for all threads in the
1039 * group including the group leader.
1041 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1042 spin_lock_irq(&p->real_parent->sighand->siglock);
1043 psig = p->real_parent->signal;
1045 psig->cutime += tgutime + sig->cutime;
1046 psig->cstime += tgstime + sig->cstime;
1047 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1049 p->min_flt + sig->min_flt + sig->cmin_flt;
1051 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1053 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1055 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1057 task_io_get_inblock(p) +
1058 sig->inblock + sig->cinblock;
1060 task_io_get_oublock(p) +
1061 sig->oublock + sig->coublock;
1062 maxrss = max(sig->maxrss, sig->cmaxrss);
1063 if (psig->cmaxrss < maxrss)
1064 psig->cmaxrss = maxrss;
1065 task_io_accounting_add(&psig->ioac, &p->ioac);
1066 task_io_accounting_add(&psig->ioac, &sig->ioac);
1067 spin_unlock_irq(&p->real_parent->sighand->siglock);
1071 * Now we are sure this task is interesting, and no other
1072 * thread can reap it because we its state == DEAD/TRACE.
1074 read_unlock(&tasklist_lock);
1076 retval = wo->wo_rusage
1077 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1078 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1079 ? p->signal->group_exit_code : p->exit_code;
1080 if (!retval && wo->wo_stat)
1081 retval = put_user(status, wo->wo_stat);
1083 infop = wo->wo_info;
1084 if (!retval && infop)
1085 retval = put_user(SIGCHLD, &infop->si_signo);
1086 if (!retval && infop)
1087 retval = put_user(0, &infop->si_errno);
1088 if (!retval && infop) {
1091 if ((status & 0x7f) == 0) {
1095 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1098 retval = put_user((short)why, &infop->si_code);
1100 retval = put_user(status, &infop->si_status);
1102 if (!retval && infop)
1103 retval = put_user(pid, &infop->si_pid);
1104 if (!retval && infop)
1105 retval = put_user(uid, &infop->si_uid);
1109 if (state == EXIT_TRACE) {
1110 write_lock_irq(&tasklist_lock);
1111 /* We dropped tasklist, ptracer could die and untrace */
1114 /* If parent wants a zombie, don't release it now */
1115 state = EXIT_ZOMBIE;
1116 if (do_notify_parent(p, p->exit_signal))
1118 p->exit_state = state;
1119 write_unlock_irq(&tasklist_lock);
1121 if (state == EXIT_DEAD)
1127 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1130 if (task_is_stopped_or_traced(p) &&
1131 !(p->jobctl & JOBCTL_LISTENING))
1132 return &p->exit_code;
1134 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1135 return &p->signal->group_exit_code;
1141 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1143 * @ptrace: is the wait for ptrace
1144 * @p: task to wait for
1146 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1149 * read_lock(&tasklist_lock), which is released if return value is
1150 * non-zero. Also, grabs and releases @p->sighand->siglock.
1153 * 0 if wait condition didn't exist and search for other wait conditions
1154 * should continue. Non-zero return, -errno on failure and @p's pid on
1155 * success, implies that tasklist_lock is released and wait condition
1156 * search should terminate.
1158 static int wait_task_stopped(struct wait_opts *wo,
1159 int ptrace, struct task_struct *p)
1161 struct siginfo __user *infop;
1162 int retval, exit_code, *p_code, why;
1163 uid_t uid = 0; /* unneeded, required by compiler */
1167 * Traditionally we see ptrace'd stopped tasks regardless of options.
1169 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1172 if (!task_stopped_code(p, ptrace))
1176 spin_lock_irq(&p->sighand->siglock);
1178 p_code = task_stopped_code(p, ptrace);
1179 if (unlikely(!p_code))
1182 exit_code = *p_code;
1186 if (!unlikely(wo->wo_flags & WNOWAIT))
1189 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1191 spin_unlock_irq(&p->sighand->siglock);
1196 * Now we are pretty sure this task is interesting.
1197 * Make sure it doesn't get reaped out from under us while we
1198 * give up the lock and then examine it below. We don't want to
1199 * keep holding onto the tasklist_lock while we call getrusage and
1200 * possibly take page faults for user memory.
1203 pid = task_pid_vnr(p);
1204 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1205 read_unlock(&tasklist_lock);
1207 if (unlikely(wo->wo_flags & WNOWAIT))
1208 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1210 retval = wo->wo_rusage
1211 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1212 if (!retval && wo->wo_stat)
1213 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1215 infop = wo->wo_info;
1216 if (!retval && infop)
1217 retval = put_user(SIGCHLD, &infop->si_signo);
1218 if (!retval && infop)
1219 retval = put_user(0, &infop->si_errno);
1220 if (!retval && infop)
1221 retval = put_user((short)why, &infop->si_code);
1222 if (!retval && infop)
1223 retval = put_user(exit_code, &infop->si_status);
1224 if (!retval && infop)
1225 retval = put_user(pid, &infop->si_pid);
1226 if (!retval && infop)
1227 retval = put_user(uid, &infop->si_uid);
1237 * Handle do_wait work for one task in a live, non-stopped state.
1238 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1239 * the lock and this task is uninteresting. If we return nonzero, we have
1240 * released the lock and the system call should return.
1242 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1248 if (!unlikely(wo->wo_flags & WCONTINUED))
1251 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1254 spin_lock_irq(&p->sighand->siglock);
1255 /* Re-check with the lock held. */
1256 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1257 spin_unlock_irq(&p->sighand->siglock);
1260 if (!unlikely(wo->wo_flags & WNOWAIT))
1261 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1262 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1263 spin_unlock_irq(&p->sighand->siglock);
1265 pid = task_pid_vnr(p);
1267 read_unlock(&tasklist_lock);
1270 retval = wo->wo_rusage
1271 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1273 if (!retval && wo->wo_stat)
1274 retval = put_user(0xffff, wo->wo_stat);
1278 retval = wait_noreap_copyout(wo, p, pid, uid,
1279 CLD_CONTINUED, SIGCONT);
1280 BUG_ON(retval == 0);
1287 * Consider @p for a wait by @parent.
1289 * -ECHILD should be in ->notask_error before the first call.
1290 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1291 * Returns zero if the search for a child should continue;
1292 * then ->notask_error is 0 if @p is an eligible child,
1293 * or another error from security_task_wait(), or still -ECHILD.
1295 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1296 struct task_struct *p)
1300 if (unlikely(p->exit_state == EXIT_DEAD))
1303 ret = eligible_child(wo, p);
1307 ret = security_task_wait(p);
1308 if (unlikely(ret < 0)) {
1310 * If we have not yet seen any eligible child,
1311 * then let this error code replace -ECHILD.
1312 * A permission error will give the user a clue
1313 * to look for security policy problems, rather
1314 * than for mysterious wait bugs.
1316 if (wo->notask_error)
1317 wo->notask_error = ret;
1321 if (unlikely(p->exit_state == EXIT_TRACE)) {
1323 * ptrace == 0 means we are the natural parent. In this case
1324 * we should clear notask_error, debugger will notify us.
1326 if (likely(!ptrace))
1327 wo->notask_error = 0;
1331 if (likely(!ptrace) && unlikely(p->ptrace)) {
1333 * If it is traced by its real parent's group, just pretend
1334 * the caller is ptrace_do_wait() and reap this child if it
1337 * This also hides group stop state from real parent; otherwise
1338 * a single stop can be reported twice as group and ptrace stop.
1339 * If a ptracer wants to distinguish these two events for its
1340 * own children it should create a separate process which takes
1341 * the role of real parent.
1343 if (!ptrace_reparented(p))
1348 if (p->exit_state == EXIT_ZOMBIE) {
1349 /* we don't reap group leaders with subthreads */
1350 if (!delay_group_leader(p)) {
1352 * A zombie ptracee is only visible to its ptracer.
1353 * Notification and reaping will be cascaded to the
1354 * real parent when the ptracer detaches.
1356 if (unlikely(ptrace) || likely(!p->ptrace))
1357 return wait_task_zombie(wo, p);
1361 * Allow access to stopped/continued state via zombie by
1362 * falling through. Clearing of notask_error is complex.
1366 * If WEXITED is set, notask_error should naturally be
1367 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1368 * so, if there are live subthreads, there are events to
1369 * wait for. If all subthreads are dead, it's still safe
1370 * to clear - this function will be called again in finite
1371 * amount time once all the subthreads are released and
1372 * will then return without clearing.
1376 * Stopped state is per-task and thus can't change once the
1377 * target task dies. Only continued and exited can happen.
1378 * Clear notask_error if WCONTINUED | WEXITED.
1380 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1381 wo->notask_error = 0;
1384 * @p is alive and it's gonna stop, continue or exit, so
1385 * there always is something to wait for.
1387 wo->notask_error = 0;
1391 * Wait for stopped. Depending on @ptrace, different stopped state
1392 * is used and the two don't interact with each other.
1394 ret = wait_task_stopped(wo, ptrace, p);
1399 * Wait for continued. There's only one continued state and the
1400 * ptracer can consume it which can confuse the real parent. Don't
1401 * use WCONTINUED from ptracer. You don't need or want it.
1403 return wait_task_continued(wo, p);
1407 * Do the work of do_wait() for one thread in the group, @tsk.
1409 * -ECHILD should be in ->notask_error before the first call.
1410 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1411 * Returns zero if the search for a child should continue; then
1412 * ->notask_error is 0 if there were any eligible children,
1413 * or another error from security_task_wait(), or still -ECHILD.
1415 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1417 struct task_struct *p;
1419 list_for_each_entry(p, &tsk->children, sibling) {
1420 int ret = wait_consider_task(wo, 0, p);
1428 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1430 struct task_struct *p;
1432 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1433 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 critiera 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);