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/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.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/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/fs_struct.h>
50 #include <linux/init_task.h>
51 #include <trace/sched.h>
53 #include <asm/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/pgtable.h>
56 #include <asm/mmu_context.h>
57 #include "cred-internals.h"
59 DEFINE_TRACE(sched_process_free);
60 DEFINE_TRACE(sched_process_exit);
61 DEFINE_TRACE(sched_process_wait);
63 static void exit_mm(struct task_struct * tsk);
65 static inline int task_detached(struct task_struct *p)
67 return p->exit_signal == -1;
70 static void __unhash_process(struct task_struct *p)
73 detach_pid(p, PIDTYPE_PID);
74 if (thread_group_leader(p)) {
75 detach_pid(p, PIDTYPE_PGID);
76 detach_pid(p, PIDTYPE_SID);
78 list_del_rcu(&p->tasks);
79 __get_cpu_var(process_counts)--;
81 list_del_rcu(&p->thread_group);
82 list_del_init(&p->sibling);
86 * This function expects the tasklist_lock write-locked.
88 static void __exit_signal(struct task_struct *tsk)
90 struct signal_struct *sig = tsk->signal;
91 struct sighand_struct *sighand;
94 BUG_ON(!atomic_read(&sig->count));
96 sighand = rcu_dereference(tsk->sighand);
97 spin_lock(&sighand->siglock);
99 posix_cpu_timers_exit(tsk);
100 if (atomic_dec_and_test(&sig->count))
101 posix_cpu_timers_exit_group(tsk);
104 * If there is any task waiting for the group exit
107 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
108 wake_up_process(sig->group_exit_task);
110 if (tsk == sig->curr_target)
111 sig->curr_target = next_thread(tsk);
113 * Accumulate here the counters for all threads but the
114 * group leader as they die, so they can be added into
115 * the process-wide totals when those are taken.
116 * The group leader stays around as a zombie as long
117 * as there are other threads. When it gets reaped,
118 * the exit.c code will add its counts into these totals.
119 * We won't ever get here for the group leader, since it
120 * will have been the last reference on the signal_struct.
122 sig->utime = cputime_add(sig->utime, task_utime(tsk));
123 sig->stime = cputime_add(sig->stime, task_stime(tsk));
124 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
125 sig->min_flt += tsk->min_flt;
126 sig->maj_flt += tsk->maj_flt;
127 sig->nvcsw += tsk->nvcsw;
128 sig->nivcsw += tsk->nivcsw;
129 sig->inblock += task_io_get_inblock(tsk);
130 sig->oublock += task_io_get_oublock(tsk);
131 task_io_accounting_add(&sig->ioac, &tsk->ioac);
132 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
133 sig = NULL; /* Marker for below. */
136 __unhash_process(tsk);
139 * Do this under ->siglock, we can race with another thread
140 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
142 flush_sigqueue(&tsk->pending);
146 spin_unlock(&sighand->siglock);
148 __cleanup_sighand(sighand);
149 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
151 flush_sigqueue(&sig->shared_pending);
152 taskstats_tgid_free(sig);
154 * Make sure ->signal can't go away under rq->lock,
155 * see account_group_exec_runtime().
157 task_rq_unlock_wait(tsk);
158 __cleanup_signal(sig);
162 static void delayed_put_task_struct(struct rcu_head *rhp)
164 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
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 tracehook_prepare_release_task(p);
177 /* don't need to get the RCU readlock here - the process is dead and
178 * can't be modifying its own credentials */
179 atomic_dec(&__task_cred(p)->user->processes);
182 write_lock_irq(&tasklist_lock);
183 tracehook_finish_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) && leader->exit_state == EXIT_ZOMBIE) {
194 BUG_ON(task_detached(leader));
195 do_notify_parent(leader, leader->exit_signal);
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 * do_notify_parent() will have marked it self-reaping in
204 zap_leader = task_detached(leader);
207 * This maintains the invariant that release_task()
208 * only runs on a task in EXIT_DEAD, just for sanity.
211 leader->exit_state = EXIT_DEAD;
214 write_unlock_irq(&tasklist_lock);
216 call_rcu(&p->rcu, delayed_put_task_struct);
219 if (unlikely(zap_leader))
224 * This checks not only the pgrp, but falls back on the pid if no
225 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
228 * The caller must hold rcu lock or the tasklist lock.
230 struct pid *session_of_pgrp(struct pid *pgrp)
232 struct task_struct *p;
233 struct pid *sid = NULL;
235 p = pid_task(pgrp, PIDTYPE_PGID);
237 p = pid_task(pgrp, PIDTYPE_PID);
239 sid = task_session(p);
245 * Determine if a process group is "orphaned", according to the POSIX
246 * definition in 2.2.2.52. Orphaned process groups are not to be affected
247 * by terminal-generated stop signals. Newly orphaned process groups are
248 * to receive a SIGHUP and a SIGCONT.
250 * "I ask you, have you ever known what it is to be an orphan?"
252 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
254 struct task_struct *p;
256 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
257 if ((p == ignored_task) ||
258 (p->exit_state && thread_group_empty(p)) ||
259 is_global_init(p->real_parent))
262 if (task_pgrp(p->real_parent) != pgrp &&
263 task_session(p->real_parent) == task_session(p))
265 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
270 int is_current_pgrp_orphaned(void)
274 read_lock(&tasklist_lock);
275 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
276 read_unlock(&tasklist_lock);
281 static int has_stopped_jobs(struct pid *pgrp)
284 struct task_struct *p;
286 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
287 if (!task_is_stopped(p))
291 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
296 * Check to see if any process groups have become orphaned as
297 * a result of our exiting, and if they have any stopped jobs,
298 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
301 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
303 struct pid *pgrp = task_pgrp(tsk);
304 struct task_struct *ignored_task = tsk;
307 /* exit: our father is in a different pgrp than
308 * we are and we were the only connection outside.
310 parent = tsk->real_parent;
312 /* reparent: our child is in a different pgrp than
313 * we are, and it was the only connection outside.
317 if (task_pgrp(parent) != pgrp &&
318 task_session(parent) == task_session(tsk) &&
319 will_become_orphaned_pgrp(pgrp, ignored_task) &&
320 has_stopped_jobs(pgrp)) {
321 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
322 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
327 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
329 * If a kernel thread is launched as a result of a system call, or if
330 * it ever exits, it should generally reparent itself to kthreadd so it
331 * isn't in the way of other processes and is correctly cleaned up on exit.
333 * The various task state such as scheduling policy and priority may have
334 * been inherited from a user process, so we reset them to sane values here.
336 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
338 static void reparent_to_kthreadd(void)
340 write_lock_irq(&tasklist_lock);
342 ptrace_unlink(current);
343 /* Reparent to init */
344 current->real_parent = current->parent = kthreadd_task;
345 list_move_tail(¤t->sibling, ¤t->real_parent->children);
347 /* Set the exit signal to SIGCHLD so we signal init on exit */
348 current->exit_signal = SIGCHLD;
350 if (task_nice(current) < 0)
351 set_user_nice(current, 0);
355 memcpy(current->signal->rlim, init_task.signal->rlim,
356 sizeof(current->signal->rlim));
358 atomic_inc(&init_cred.usage);
359 commit_creds(&init_cred);
360 write_unlock_irq(&tasklist_lock);
363 void __set_special_pids(struct pid *pid)
365 struct task_struct *curr = current->group_leader;
366 pid_t nr = pid_nr(pid);
368 if (task_session(curr) != pid) {
369 change_pid(curr, PIDTYPE_SID, pid);
370 set_task_session(curr, nr);
372 if (task_pgrp(curr) != pid) {
373 change_pid(curr, PIDTYPE_PGID, pid);
374 set_task_pgrp(curr, nr);
378 static void set_special_pids(struct pid *pid)
380 write_lock_irq(&tasklist_lock);
381 __set_special_pids(pid);
382 write_unlock_irq(&tasklist_lock);
386 * Let kernel threads use this to say that they
387 * allow a certain signal (since daemonize() will
388 * have disabled all of them by default).
390 int allow_signal(int sig)
392 if (!valid_signal(sig) || sig < 1)
395 spin_lock_irq(¤t->sighand->siglock);
396 sigdelset(¤t->blocked, sig);
398 /* Kernel threads handle their own signals.
399 Let the signal code know it'll be handled, so
400 that they don't get converted to SIGKILL or
401 just silently dropped */
402 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
405 spin_unlock_irq(¤t->sighand->siglock);
409 EXPORT_SYMBOL(allow_signal);
411 int disallow_signal(int sig)
413 if (!valid_signal(sig) || sig < 1)
416 spin_lock_irq(¤t->sighand->siglock);
417 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
419 spin_unlock_irq(¤t->sighand->siglock);
423 EXPORT_SYMBOL(disallow_signal);
426 * Put all the gunge required to become a kernel thread without
427 * attached user resources in one place where it belongs.
430 void daemonize(const char *name, ...)
435 va_start(args, name);
436 vsnprintf(current->comm, sizeof(current->comm), name, args);
440 * If we were started as result of loading a module, close all of the
441 * user space pages. We don't need them, and if we didn't close them
442 * they would be locked into memory.
446 * We don't want to have TIF_FREEZE set if the system-wide hibernation
447 * or suspend transition begins right now.
449 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
451 if (current->nsproxy != &init_nsproxy) {
452 get_nsproxy(&init_nsproxy);
453 switch_task_namespaces(current, &init_nsproxy);
455 set_special_pids(&init_struct_pid);
456 proc_clear_tty(current);
458 /* Block and flush all signals */
459 sigfillset(&blocked);
460 sigprocmask(SIG_BLOCK, &blocked, NULL);
461 flush_signals(current);
463 /* Become as one with the init task */
465 daemonize_fs_struct();
467 current->files = init_task.files;
468 atomic_inc(¤t->files->count);
470 reparent_to_kthreadd();
473 EXPORT_SYMBOL(daemonize);
475 static void close_files(struct files_struct * files)
483 * It is safe to dereference the fd table without RCU or
484 * ->file_lock because this is the last reference to the
487 fdt = files_fdtable(files);
491 if (i >= fdt->max_fds)
493 set = fdt->open_fds->fds_bits[j++];
496 struct file * file = xchg(&fdt->fd[i], NULL);
498 filp_close(file, files);
508 struct files_struct *get_files_struct(struct task_struct *task)
510 struct files_struct *files;
515 atomic_inc(&files->count);
521 void put_files_struct(struct files_struct *files)
525 if (atomic_dec_and_test(&files->count)) {
528 * Free the fd and fdset arrays if we expanded them.
529 * If the fdtable was embedded, pass files for freeing
530 * at the end of the RCU grace period. Otherwise,
531 * you can free files immediately.
533 fdt = files_fdtable(files);
534 if (fdt != &files->fdtab)
535 kmem_cache_free(files_cachep, files);
540 void reset_files_struct(struct files_struct *files)
542 struct task_struct *tsk = current;
543 struct files_struct *old;
549 put_files_struct(old);
552 void exit_files(struct task_struct *tsk)
554 struct files_struct * files = tsk->files;
560 put_files_struct(files);
564 #ifdef CONFIG_MM_OWNER
566 * Task p is exiting and it owned mm, lets find a new owner for it
569 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
572 * If there are other users of the mm and the owner (us) is exiting
573 * we need to find a new owner to take on the responsibility.
575 if (atomic_read(&mm->mm_users) <= 1)
582 void mm_update_next_owner(struct mm_struct *mm)
584 struct task_struct *c, *g, *p = current;
587 if (!mm_need_new_owner(mm, p))
590 read_lock(&tasklist_lock);
592 * Search in the children
594 list_for_each_entry(c, &p->children, sibling) {
596 goto assign_new_owner;
600 * Search in the siblings
602 list_for_each_entry(c, &p->parent->children, sibling) {
604 goto assign_new_owner;
608 * Search through everything else. We should not get
611 do_each_thread(g, c) {
613 goto assign_new_owner;
614 } while_each_thread(g, c);
616 read_unlock(&tasklist_lock);
618 * We found no owner yet mm_users > 1: this implies that we are
619 * most likely racing with swapoff (try_to_unuse()) or /proc or
620 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
629 * The task_lock protects c->mm from changing.
630 * We always want mm->owner->mm == mm
634 * Delay read_unlock() till we have the task_lock()
635 * to ensure that c does not slip away underneath us
637 read_unlock(&tasklist_lock);
647 #endif /* CONFIG_MM_OWNER */
650 * Turn us into a lazy TLB process if we
653 static void exit_mm(struct task_struct * tsk)
655 struct mm_struct *mm = tsk->mm;
656 struct core_state *core_state;
662 * Serialize with any possible pending coredump.
663 * We must hold mmap_sem around checking core_state
664 * and clearing tsk->mm. The core-inducing thread
665 * will increment ->nr_threads for each thread in the
666 * group with ->mm != NULL.
668 down_read(&mm->mmap_sem);
669 core_state = mm->core_state;
671 struct core_thread self;
672 up_read(&mm->mmap_sem);
675 self.next = xchg(&core_state->dumper.next, &self);
677 * Implies mb(), the result of xchg() must be visible
678 * to core_state->dumper.
680 if (atomic_dec_and_test(&core_state->nr_threads))
681 complete(&core_state->startup);
684 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
685 if (!self.task) /* see coredump_finish() */
689 __set_task_state(tsk, TASK_RUNNING);
690 down_read(&mm->mmap_sem);
692 atomic_inc(&mm->mm_count);
693 BUG_ON(mm != tsk->active_mm);
694 /* more a memory barrier than a real lock */
697 up_read(&mm->mmap_sem);
698 enter_lazy_tlb(mm, current);
699 /* We don't want this task to be frozen prematurely */
700 clear_freeze_flag(tsk);
702 mm_update_next_owner(mm);
707 * Return nonzero if @parent's children should reap themselves.
709 * Called with write_lock_irq(&tasklist_lock) held.
711 static int ignoring_children(struct task_struct *parent)
714 struct sighand_struct *psig = parent->sighand;
716 spin_lock_irqsave(&psig->siglock, flags);
717 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
718 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
719 spin_unlock_irqrestore(&psig->siglock, flags);
724 * Detach all tasks we were using ptrace on.
725 * Any that need to be release_task'd are put on the @dead list.
727 * Called with write_lock(&tasklist_lock) held.
729 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
731 struct task_struct *p, *n;
734 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
737 if (p->exit_state != EXIT_ZOMBIE)
741 * If it's a zombie, our attachedness prevented normal
742 * parent notification or self-reaping. Do notification
743 * now if it would have happened earlier. If it should
744 * reap itself, add it to the @dead list. We can't call
745 * release_task() here because we already hold tasklist_lock.
747 * If it's our own child, there is no notification to do.
748 * But if our normal children self-reap, then this child
749 * was prevented by ptrace and we must reap it now.
751 if (!task_detached(p) && thread_group_empty(p)) {
752 if (!same_thread_group(p->real_parent, parent))
753 do_notify_parent(p, p->exit_signal);
756 ign = ignoring_children(parent);
762 if (task_detached(p)) {
764 * Mark it as in the process of being reaped.
766 p->exit_state = EXIT_DEAD;
767 list_add(&p->ptrace_entry, dead);
773 * Finish up exit-time ptrace cleanup.
775 * Called without locks.
777 static void ptrace_exit_finish(struct task_struct *parent,
778 struct list_head *dead)
780 struct task_struct *p, *n;
782 BUG_ON(!list_empty(&parent->ptraced));
784 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
785 list_del_init(&p->ptrace_entry);
790 static void reparent_thread(struct task_struct *p, struct task_struct *father)
792 if (p->pdeath_signal)
793 /* We already hold the tasklist_lock here. */
794 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
796 list_move_tail(&p->sibling, &p->real_parent->children);
798 /* If this is a threaded reparent there is no need to
799 * notify anyone anything has happened.
801 if (same_thread_group(p->real_parent, father))
804 /* We don't want people slaying init. */
805 if (!task_detached(p))
806 p->exit_signal = SIGCHLD;
808 /* If we'd notified the old parent about this child's death,
809 * also notify the new parent.
811 if (!ptrace_reparented(p) &&
812 p->exit_state == EXIT_ZOMBIE &&
813 !task_detached(p) && thread_group_empty(p))
814 do_notify_parent(p, p->exit_signal);
816 kill_orphaned_pgrp(p, father);
820 * When we die, we re-parent all our children.
821 * Try to give them to another thread in our thread
822 * group, and if no such member exists, give it to
823 * the child reaper process (ie "init") in our pid
826 static struct task_struct *find_new_reaper(struct task_struct *father)
828 struct pid_namespace *pid_ns = task_active_pid_ns(father);
829 struct task_struct *thread;
832 while_each_thread(father, thread) {
833 if (thread->flags & PF_EXITING)
835 if (unlikely(pid_ns->child_reaper == father))
836 pid_ns->child_reaper = thread;
840 if (unlikely(pid_ns->child_reaper == father)) {
841 write_unlock_irq(&tasklist_lock);
842 if (unlikely(pid_ns == &init_pid_ns))
843 panic("Attempted to kill init!");
845 zap_pid_ns_processes(pid_ns);
846 write_lock_irq(&tasklist_lock);
848 * We can not clear ->child_reaper or leave it alone.
849 * There may by stealth EXIT_DEAD tasks on ->children,
850 * forget_original_parent() must move them somewhere.
852 pid_ns->child_reaper = init_pid_ns.child_reaper;
855 return pid_ns->child_reaper;
858 static void forget_original_parent(struct task_struct *father)
860 struct task_struct *p, *n, *reaper;
861 LIST_HEAD(ptrace_dead);
863 write_lock_irq(&tasklist_lock);
864 reaper = find_new_reaper(father);
866 * First clean up ptrace if we were using it.
868 ptrace_exit(father, &ptrace_dead);
870 list_for_each_entry_safe(p, n, &father->children, sibling) {
871 p->real_parent = reaper;
872 if (p->parent == father) {
874 p->parent = p->real_parent;
876 reparent_thread(p, father);
879 write_unlock_irq(&tasklist_lock);
880 BUG_ON(!list_empty(&father->children));
882 ptrace_exit_finish(father, &ptrace_dead);
886 * Send signals to all our closest relatives so that they know
887 * to properly mourn us..
889 static void exit_notify(struct task_struct *tsk, int group_dead)
895 * This does two things:
897 * A. Make init inherit all the child processes
898 * B. Check to see if any process groups have become orphaned
899 * as a result of our exiting, and if they have any stopped
900 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
902 forget_original_parent(tsk);
903 exit_task_namespaces(tsk);
905 write_lock_irq(&tasklist_lock);
907 kill_orphaned_pgrp(tsk->group_leader, NULL);
909 /* Let father know we died
911 * Thread signals are configurable, but you aren't going to use
912 * that to send signals to arbitary processes.
913 * That stops right now.
915 * If the parent exec id doesn't match the exec id we saved
916 * when we started then we know the parent has changed security
919 * If our self_exec id doesn't match our parent_exec_id then
920 * we have changed execution domain as these two values started
921 * the same after a fork.
923 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
924 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
925 tsk->self_exec_id != tsk->parent_exec_id) &&
927 tsk->exit_signal = SIGCHLD;
929 signal = tracehook_notify_death(tsk, &cookie, group_dead);
931 signal = do_notify_parent(tsk, signal);
933 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
935 /* mt-exec, de_thread() is waiting for us */
936 if (thread_group_leader(tsk) &&
937 tsk->signal->group_exit_task &&
938 tsk->signal->notify_count < 0)
939 wake_up_process(tsk->signal->group_exit_task);
941 write_unlock_irq(&tasklist_lock);
943 tracehook_report_death(tsk, signal, cookie, group_dead);
945 /* If the process is dead, release it - nobody will wait for it */
946 if (signal == DEATH_REAP)
950 #ifdef CONFIG_DEBUG_STACK_USAGE
951 static void check_stack_usage(void)
953 static DEFINE_SPINLOCK(low_water_lock);
954 static int lowest_to_date = THREAD_SIZE;
957 free = stack_not_used(current);
959 if (free >= lowest_to_date)
962 spin_lock(&low_water_lock);
963 if (free < lowest_to_date) {
964 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
966 current->comm, free);
967 lowest_to_date = free;
969 spin_unlock(&low_water_lock);
972 static inline void check_stack_usage(void) {}
975 NORET_TYPE void do_exit(long code)
977 struct task_struct *tsk = current;
980 profile_task_exit(tsk);
982 WARN_ON(atomic_read(&tsk->fs_excl));
984 if (unlikely(in_interrupt()))
985 panic("Aiee, killing interrupt handler!");
986 if (unlikely(!tsk->pid))
987 panic("Attempted to kill the idle task!");
989 tracehook_report_exit(&code);
992 * We're taking recursive faults here in do_exit. Safest is to just
993 * leave this task alone and wait for reboot.
995 if (unlikely(tsk->flags & PF_EXITING)) {
997 "Fixing recursive fault but reboot is needed!\n");
999 * We can do this unlocked here. The futex code uses
1000 * this flag just to verify whether the pi state
1001 * cleanup has been done or not. In the worst case it
1002 * loops once more. We pretend that the cleanup was
1003 * done as there is no way to return. Either the
1004 * OWNER_DIED bit is set by now or we push the blocked
1005 * task into the wait for ever nirwana as well.
1007 tsk->flags |= PF_EXITPIDONE;
1008 set_current_state(TASK_UNINTERRUPTIBLE);
1012 exit_signals(tsk); /* sets PF_EXITING */
1014 * tsk->flags are checked in the futex code to protect against
1015 * an exiting task cleaning up the robust pi futexes.
1018 spin_unlock_wait(&tsk->pi_lock);
1020 if (unlikely(in_atomic()))
1021 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1022 current->comm, task_pid_nr(current),
1025 acct_update_integrals(tsk);
1027 group_dead = atomic_dec_and_test(&tsk->signal->live);
1029 hrtimer_cancel(&tsk->signal->real_timer);
1030 exit_itimers(tsk->signal);
1032 acct_collect(code, group_dead);
1035 if (unlikely(tsk->audit_context))
1038 tsk->exit_code = code;
1039 taskstats_exit(tsk, group_dead);
1045 trace_sched_process_exit(tsk);
1050 check_stack_usage();
1052 cgroup_exit(tsk, 1);
1054 if (group_dead && tsk->signal->leader)
1055 disassociate_ctty(1);
1057 module_put(task_thread_info(tsk)->exec_domain->module);
1059 module_put(tsk->binfmt->module);
1061 proc_exit_connector(tsk);
1062 exit_notify(tsk, group_dead);
1064 mpol_put(tsk->mempolicy);
1065 tsk->mempolicy = NULL;
1069 * This must happen late, after the PID is not
1072 if (unlikely(!list_empty(&tsk->pi_state_list)))
1073 exit_pi_state_list(tsk);
1074 if (unlikely(current->pi_state_cache))
1075 kfree(current->pi_state_cache);
1078 * Make sure we are holding no locks:
1080 debug_check_no_locks_held(tsk);
1082 * We can do this unlocked here. The futex code uses this flag
1083 * just to verify whether the pi state cleanup has been done
1084 * or not. In the worst case it loops once more.
1086 tsk->flags |= PF_EXITPIDONE;
1088 if (tsk->io_context)
1091 if (tsk->splice_pipe)
1092 __free_pipe_info(tsk->splice_pipe);
1095 /* causes final put_task_struct in finish_task_switch(). */
1096 tsk->state = TASK_DEAD;
1099 /* Avoid "noreturn function does return". */
1101 cpu_relax(); /* For when BUG is null */
1104 EXPORT_SYMBOL_GPL(do_exit);
1106 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1114 EXPORT_SYMBOL(complete_and_exit);
1116 SYSCALL_DEFINE1(exit, int, error_code)
1118 do_exit((error_code&0xff)<<8);
1122 * Take down every thread in the group. This is called by fatal signals
1123 * as well as by sys_exit_group (below).
1126 do_group_exit(int exit_code)
1128 struct signal_struct *sig = current->signal;
1130 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1132 if (signal_group_exit(sig))
1133 exit_code = sig->group_exit_code;
1134 else if (!thread_group_empty(current)) {
1135 struct sighand_struct *const sighand = current->sighand;
1136 spin_lock_irq(&sighand->siglock);
1137 if (signal_group_exit(sig))
1138 /* Another thread got here before we took the lock. */
1139 exit_code = sig->group_exit_code;
1141 sig->group_exit_code = exit_code;
1142 sig->flags = SIGNAL_GROUP_EXIT;
1143 zap_other_threads(current);
1145 spin_unlock_irq(&sighand->siglock);
1153 * this kills every thread in the thread group. Note that any externally
1154 * wait4()-ing process will get the correct exit code - even if this
1155 * thread is not the thread group leader.
1157 SYSCALL_DEFINE1(exit_group, int, error_code)
1159 do_group_exit((error_code & 0xff) << 8);
1164 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1166 struct pid *pid = NULL;
1167 if (type == PIDTYPE_PID)
1168 pid = task->pids[type].pid;
1169 else if (type < PIDTYPE_MAX)
1170 pid = task->group_leader->pids[type].pid;
1174 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1175 struct task_struct *p)
1179 if (type < PIDTYPE_MAX) {
1180 if (task_pid_type(p, type) != pid)
1184 /* Wait for all children (clone and not) if __WALL is set;
1185 * otherwise, wait for clone children *only* if __WCLONE is
1186 * set; otherwise, wait for non-clone children *only*. (Note:
1187 * A "clone" child here is one that reports to its parent
1188 * using a signal other than SIGCHLD.) */
1189 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1190 && !(options & __WALL))
1193 err = security_task_wait(p);
1200 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1201 int why, int status,
1202 struct siginfo __user *infop,
1203 struct rusage __user *rusagep)
1205 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1209 retval = put_user(SIGCHLD, &infop->si_signo);
1211 retval = put_user(0, &infop->si_errno);
1213 retval = put_user((short)why, &infop->si_code);
1215 retval = put_user(pid, &infop->si_pid);
1217 retval = put_user(uid, &infop->si_uid);
1219 retval = put_user(status, &infop->si_status);
1226 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1227 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1228 * the lock and this task is uninteresting. If we return nonzero, we have
1229 * released the lock and the system call should return.
1231 static int wait_task_zombie(struct task_struct *p, int options,
1232 struct siginfo __user *infop,
1233 int __user *stat_addr, struct rusage __user *ru)
1235 unsigned long state;
1236 int retval, status, traced;
1237 pid_t pid = task_pid_vnr(p);
1238 uid_t uid = __task_cred(p)->uid;
1240 if (!likely(options & WEXITED))
1243 if (unlikely(options & WNOWAIT)) {
1244 int exit_code = p->exit_code;
1248 read_unlock(&tasklist_lock);
1249 if ((exit_code & 0x7f) == 0) {
1251 status = exit_code >> 8;
1253 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1254 status = exit_code & 0x7f;
1256 return wait_noreap_copyout(p, pid, uid, why,
1261 * Try to move the task's state to DEAD
1262 * only one thread is allowed to do this:
1264 state = xchg(&p->exit_state, EXIT_DEAD);
1265 if (state != EXIT_ZOMBIE) {
1266 BUG_ON(state != EXIT_DEAD);
1270 traced = ptrace_reparented(p);
1272 if (likely(!traced)) {
1273 struct signal_struct *psig;
1274 struct signal_struct *sig;
1275 struct task_cputime cputime;
1278 * The resource counters for the group leader are in its
1279 * own task_struct. Those for dead threads in the group
1280 * are in its signal_struct, as are those for the child
1281 * processes it has previously reaped. All these
1282 * accumulate in the parent's signal_struct c* fields.
1284 * We don't bother to take a lock here to protect these
1285 * p->signal fields, because they are only touched by
1286 * __exit_signal, which runs with tasklist_lock
1287 * write-locked anyway, and so is excluded here. We do
1288 * need to protect the access to p->parent->signal fields,
1289 * as other threads in the parent group can be right
1290 * here reaping other children at the same time.
1292 * We use thread_group_cputime() to get times for the thread
1293 * group, which consolidates times for all threads in the
1294 * group including the group leader.
1296 thread_group_cputime(p, &cputime);
1297 spin_lock_irq(&p->parent->sighand->siglock);
1298 psig = p->parent->signal;
1301 cputime_add(psig->cutime,
1302 cputime_add(cputime.utime,
1305 cputime_add(psig->cstime,
1306 cputime_add(cputime.stime,
1309 cputime_add(psig->cgtime,
1310 cputime_add(p->gtime,
1311 cputime_add(sig->gtime,
1314 p->min_flt + sig->min_flt + sig->cmin_flt;
1316 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1318 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1320 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1322 task_io_get_inblock(p) +
1323 sig->inblock + sig->cinblock;
1325 task_io_get_oublock(p) +
1326 sig->oublock + sig->coublock;
1327 task_io_accounting_add(&psig->ioac, &p->ioac);
1328 task_io_accounting_add(&psig->ioac, &sig->ioac);
1329 spin_unlock_irq(&p->parent->sighand->siglock);
1333 * Now we are sure this task is interesting, and no other
1334 * thread can reap it because we set its state to EXIT_DEAD.
1336 read_unlock(&tasklist_lock);
1338 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1339 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1340 ? p->signal->group_exit_code : p->exit_code;
1341 if (!retval && stat_addr)
1342 retval = put_user(status, stat_addr);
1343 if (!retval && infop)
1344 retval = put_user(SIGCHLD, &infop->si_signo);
1345 if (!retval && infop)
1346 retval = put_user(0, &infop->si_errno);
1347 if (!retval && infop) {
1350 if ((status & 0x7f) == 0) {
1354 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1357 retval = put_user((short)why, &infop->si_code);
1359 retval = put_user(status, &infop->si_status);
1361 if (!retval && infop)
1362 retval = put_user(pid, &infop->si_pid);
1363 if (!retval && infop)
1364 retval = put_user(uid, &infop->si_uid);
1369 write_lock_irq(&tasklist_lock);
1370 /* We dropped tasklist, ptracer could die and untrace */
1373 * If this is not a detached task, notify the parent.
1374 * If it's still not detached after that, don't release
1377 if (!task_detached(p)) {
1378 do_notify_parent(p, p->exit_signal);
1379 if (!task_detached(p)) {
1380 p->exit_state = EXIT_ZOMBIE;
1384 write_unlock_irq(&tasklist_lock);
1393 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1394 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1395 * the lock and this task is uninteresting. If we return nonzero, we have
1396 * released the lock and the system call should return.
1398 static int wait_task_stopped(int ptrace, struct task_struct *p,
1399 int options, struct siginfo __user *infop,
1400 int __user *stat_addr, struct rusage __user *ru)
1402 int retval, exit_code, why;
1403 uid_t uid = 0; /* unneeded, required by compiler */
1406 if (!(options & WUNTRACED))
1410 spin_lock_irq(&p->sighand->siglock);
1412 if (unlikely(!task_is_stopped_or_traced(p)))
1415 if (!ptrace && p->signal->group_stop_count > 0)
1417 * A group stop is in progress and this is the group leader.
1418 * We won't report until all threads have stopped.
1422 exit_code = p->exit_code;
1426 if (!unlikely(options & WNOWAIT))
1429 /* don't need the RCU readlock here as we're holding a spinlock */
1430 uid = __task_cred(p)->uid;
1432 spin_unlock_irq(&p->sighand->siglock);
1437 * Now we are pretty sure this task is interesting.
1438 * Make sure it doesn't get reaped out from under us while we
1439 * give up the lock and then examine it below. We don't want to
1440 * keep holding onto the tasklist_lock while we call getrusage and
1441 * possibly take page faults for user memory.
1444 pid = task_pid_vnr(p);
1445 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1446 read_unlock(&tasklist_lock);
1448 if (unlikely(options & WNOWAIT))
1449 return wait_noreap_copyout(p, pid, uid,
1453 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1454 if (!retval && stat_addr)
1455 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1456 if (!retval && infop)
1457 retval = put_user(SIGCHLD, &infop->si_signo);
1458 if (!retval && infop)
1459 retval = put_user(0, &infop->si_errno);
1460 if (!retval && infop)
1461 retval = put_user((short)why, &infop->si_code);
1462 if (!retval && infop)
1463 retval = put_user(exit_code, &infop->si_status);
1464 if (!retval && infop)
1465 retval = put_user(pid, &infop->si_pid);
1466 if (!retval && infop)
1467 retval = put_user(uid, &infop->si_uid);
1477 * Handle do_wait work for one task in a live, non-stopped state.
1478 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1479 * the lock and this task is uninteresting. If we return nonzero, we have
1480 * released the lock and the system call should return.
1482 static int wait_task_continued(struct task_struct *p, int options,
1483 struct siginfo __user *infop,
1484 int __user *stat_addr, struct rusage __user *ru)
1490 if (!unlikely(options & WCONTINUED))
1493 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1496 spin_lock_irq(&p->sighand->siglock);
1497 /* Re-check with the lock held. */
1498 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1499 spin_unlock_irq(&p->sighand->siglock);
1502 if (!unlikely(options & WNOWAIT))
1503 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1504 uid = __task_cred(p)->uid;
1505 spin_unlock_irq(&p->sighand->siglock);
1507 pid = task_pid_vnr(p);
1509 read_unlock(&tasklist_lock);
1512 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1514 if (!retval && stat_addr)
1515 retval = put_user(0xffff, stat_addr);
1519 retval = wait_noreap_copyout(p, pid, uid,
1520 CLD_CONTINUED, SIGCONT,
1522 BUG_ON(retval == 0);
1529 * Consider @p for a wait by @parent.
1531 * -ECHILD should be in *@notask_error before the first call.
1532 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1533 * Returns zero if the search for a child should continue;
1534 * then *@notask_error is 0 if @p is an eligible child,
1535 * or another error from security_task_wait(), or still -ECHILD.
1537 static int wait_consider_task(struct task_struct *parent, int ptrace,
1538 struct task_struct *p, int *notask_error,
1539 enum pid_type type, struct pid *pid, int options,
1540 struct siginfo __user *infop,
1541 int __user *stat_addr, struct rusage __user *ru)
1543 int ret = eligible_child(type, pid, options, p);
1547 if (unlikely(ret < 0)) {
1549 * If we have not yet seen any eligible child,
1550 * then let this error code replace -ECHILD.
1551 * A permission error will give the user a clue
1552 * to look for security policy problems, rather
1553 * than for mysterious wait bugs.
1556 *notask_error = ret;
1559 if (likely(!ptrace) && unlikely(p->ptrace)) {
1561 * This child is hidden by ptrace.
1562 * We aren't allowed to see it now, but eventually we will.
1568 if (p->exit_state == EXIT_DEAD)
1572 * We don't reap group leaders with subthreads.
1574 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1575 return wait_task_zombie(p, options, infop, stat_addr, ru);
1578 * It's stopped or running now, so it might
1579 * later continue, exit, or stop again.
1583 if (task_is_stopped_or_traced(p))
1584 return wait_task_stopped(ptrace, p, options,
1585 infop, stat_addr, ru);
1587 return wait_task_continued(p, options, infop, stat_addr, ru);
1591 * Do the work of do_wait() for one thread in the group, @tsk.
1593 * -ECHILD should be in *@notask_error before the first call.
1594 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1595 * Returns zero if the search for a child should continue; then
1596 * *@notask_error is 0 if there were any eligible children,
1597 * or another error from security_task_wait(), or still -ECHILD.
1599 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1600 enum pid_type type, struct pid *pid, int options,
1601 struct siginfo __user *infop, int __user *stat_addr,
1602 struct rusage __user *ru)
1604 struct task_struct *p;
1606 list_for_each_entry(p, &tsk->children, sibling) {
1608 * Do not consider detached threads.
1610 if (!task_detached(p)) {
1611 int ret = wait_consider_task(tsk, 0, p, notask_error,
1613 infop, stat_addr, ru);
1622 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1623 enum pid_type type, struct pid *pid, int options,
1624 struct siginfo __user *infop, int __user *stat_addr,
1625 struct rusage __user *ru)
1627 struct task_struct *p;
1630 * Traditionally we see ptrace'd stopped tasks regardless of options.
1632 options |= WUNTRACED;
1634 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1635 int ret = wait_consider_task(tsk, 1, p, notask_error,
1637 infop, stat_addr, ru);
1645 static long do_wait(enum pid_type type, struct pid *pid, int options,
1646 struct siginfo __user *infop, int __user *stat_addr,
1647 struct rusage __user *ru)
1649 DECLARE_WAITQUEUE(wait, current);
1650 struct task_struct *tsk;
1653 trace_sched_process_wait(pid);
1655 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1658 * If there is nothing that can match our critiera just get out.
1659 * We will clear @retval to zero if we see any child that might later
1660 * match our criteria, even if we are not able to reap it yet.
1663 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1666 current->state = TASK_INTERRUPTIBLE;
1667 read_lock(&tasklist_lock);
1670 int tsk_result = do_wait_thread(tsk, &retval,
1672 infop, stat_addr, ru);
1674 tsk_result = ptrace_do_wait(tsk, &retval,
1676 infop, stat_addr, ru);
1679 * tasklist_lock is unlocked and we have a final result.
1681 retval = tsk_result;
1685 if (options & __WNOTHREAD)
1687 tsk = next_thread(tsk);
1688 BUG_ON(tsk->signal != current->signal);
1689 } while (tsk != current);
1690 read_unlock(&tasklist_lock);
1692 if (!retval && !(options & WNOHANG)) {
1693 retval = -ERESTARTSYS;
1694 if (!signal_pending(current)) {
1701 current->state = TASK_RUNNING;
1702 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1708 * For a WNOHANG return, clear out all the fields
1709 * we would set so the user can easily tell the
1713 retval = put_user(0, &infop->si_signo);
1715 retval = put_user(0, &infop->si_errno);
1717 retval = put_user(0, &infop->si_code);
1719 retval = put_user(0, &infop->si_pid);
1721 retval = put_user(0, &infop->si_uid);
1723 retval = put_user(0, &infop->si_status);
1729 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1730 infop, int, options, struct rusage __user *, ru)
1732 struct pid *pid = NULL;
1736 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1738 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1751 type = PIDTYPE_PGID;
1759 if (type < PIDTYPE_MAX)
1760 pid = find_get_pid(upid);
1761 ret = do_wait(type, pid, options, infop, NULL, ru);
1764 /* avoid REGPARM breakage on x86: */
1765 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1769 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1770 int, options, struct rusage __user *, ru)
1772 struct pid *pid = NULL;
1776 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1777 __WNOTHREAD|__WCLONE|__WALL))
1782 else if (upid < 0) {
1783 type = PIDTYPE_PGID;
1784 pid = find_get_pid(-upid);
1785 } else if (upid == 0) {
1786 type = PIDTYPE_PGID;
1787 pid = get_pid(task_pgrp(current));
1788 } else /* upid > 0 */ {
1790 pid = find_get_pid(upid);
1793 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1796 /* avoid REGPARM breakage on x86: */
1797 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1801 #ifdef __ARCH_WANT_SYS_WAITPID
1804 * sys_waitpid() remains for compatibility. waitpid() should be
1805 * implemented by calling sys_wait4() from libc.a.
1807 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1809 return sys_wait4(pid, stat_addr, options, NULL);