Merge commit 'v2.6.29-rc2' into perfcounters/core
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
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/init_task.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
57
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
61
62 static void exit_mm(struct task_struct * tsk);
63
64 static inline int task_detached(struct task_struct *p)
65 {
66         return p->exit_signal == -1;
67 }
68
69 static void __unhash_process(struct task_struct *p)
70 {
71         nr_threads--;
72         detach_pid(p, PIDTYPE_PID);
73         if (thread_group_leader(p)) {
74                 detach_pid(p, PIDTYPE_PGID);
75                 detach_pid(p, PIDTYPE_SID);
76
77                 list_del_rcu(&p->tasks);
78                 __get_cpu_var(process_counts)--;
79         }
80         list_del_rcu(&p->thread_group);
81         list_del_init(&p->sibling);
82 }
83
84 /*
85  * This function expects the tasklist_lock write-locked.
86  */
87 static void __exit_signal(struct task_struct *tsk)
88 {
89         struct signal_struct *sig = tsk->signal;
90         struct sighand_struct *sighand;
91
92         BUG_ON(!sig);
93         BUG_ON(!atomic_read(&sig->count));
94
95         sighand = rcu_dereference(tsk->sighand);
96         spin_lock(&sighand->siglock);
97
98         posix_cpu_timers_exit(tsk);
99         if (atomic_dec_and_test(&sig->count))
100                 posix_cpu_timers_exit_group(tsk);
101         else {
102                 /*
103                  * If there is any task waiting for the group exit
104                  * then notify it:
105                  */
106                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
107                         wake_up_process(sig->group_exit_task);
108
109                 if (tsk == sig->curr_target)
110                         sig->curr_target = next_thread(tsk);
111                 /*
112                  * Accumulate here the counters for all threads but the
113                  * group leader as they die, so they can be added into
114                  * the process-wide totals when those are taken.
115                  * The group leader stays around as a zombie as long
116                  * as there are other threads.  When it gets reaped,
117                  * the exit.c code will add its counts into these totals.
118                  * We won't ever get here for the group leader, since it
119                  * will have been the last reference on the signal_struct.
120                  */
121                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
122                 sig->min_flt += tsk->min_flt;
123                 sig->maj_flt += tsk->maj_flt;
124                 sig->nvcsw += tsk->nvcsw;
125                 sig->nivcsw += tsk->nivcsw;
126                 sig->inblock += task_io_get_inblock(tsk);
127                 sig->oublock += task_io_get_oublock(tsk);
128                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
129                 sig = NULL; /* Marker for below. */
130         }
131
132         __unhash_process(tsk);
133
134         /*
135          * Do this under ->siglock, we can race with another thread
136          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
137          */
138         flush_sigqueue(&tsk->pending);
139
140         tsk->signal = NULL;
141         tsk->sighand = NULL;
142         spin_unlock(&sighand->siglock);
143
144         __cleanup_sighand(sighand);
145         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
146         if (sig) {
147                 flush_sigqueue(&sig->shared_pending);
148                 taskstats_tgid_free(sig);
149                 /*
150                  * Make sure ->signal can't go away under rq->lock,
151                  * see account_group_exec_runtime().
152                  */
153                 task_rq_unlock_wait(tsk);
154                 __cleanup_signal(sig);
155         }
156 }
157
158 static void delayed_put_task_struct(struct rcu_head *rhp)
159 {
160         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161
162 #ifdef CONFIG_PERF_COUNTERS
163         WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list));
164 #endif
165         trace_sched_process_free(tsk);
166         put_task_struct(tsk);
167 }
168
169
170 void release_task(struct task_struct * p)
171 {
172         struct task_struct *leader;
173         int zap_leader;
174 repeat:
175         tracehook_prepare_release_task(p);
176         /* don't need to get the RCU readlock here - the process is dead and
177          * can't be modifying its own credentials */
178         atomic_dec(&__task_cred(p)->user->processes);
179
180         proc_flush_task(p);
181         write_lock_irq(&tasklist_lock);
182         tracehook_finish_release_task(p);
183         __exit_signal(p);
184
185         /*
186          * If we are the last non-leader member of the thread
187          * group, and the leader is zombie, then notify the
188          * group leader's parent process. (if it wants notification.)
189          */
190         zap_leader = 0;
191         leader = p->group_leader;
192         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193                 BUG_ON(task_detached(leader));
194                 do_notify_parent(leader, leader->exit_signal);
195                 /*
196                  * If we were the last child thread and the leader has
197                  * exited already, and the leader's parent ignores SIGCHLD,
198                  * then we are the one who should release the leader.
199                  *
200                  * do_notify_parent() will have marked it self-reaping in
201                  * that case.
202                  */
203                 zap_leader = task_detached(leader);
204
205                 /*
206                  * This maintains the invariant that release_task()
207                  * only runs on a task in EXIT_DEAD, just for sanity.
208                  */
209                 if (zap_leader)
210                         leader->exit_state = EXIT_DEAD;
211         }
212
213         write_unlock_irq(&tasklist_lock);
214         release_thread(p);
215         call_rcu(&p->rcu, delayed_put_task_struct);
216
217         p = leader;
218         if (unlikely(zap_leader))
219                 goto repeat;
220 }
221
222 /*
223  * This checks not only the pgrp, but falls back on the pid if no
224  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
225  * without this...
226  *
227  * The caller must hold rcu lock or the tasklist lock.
228  */
229 struct pid *session_of_pgrp(struct pid *pgrp)
230 {
231         struct task_struct *p;
232         struct pid *sid = NULL;
233
234         p = pid_task(pgrp, PIDTYPE_PGID);
235         if (p == NULL)
236                 p = pid_task(pgrp, PIDTYPE_PID);
237         if (p != NULL)
238                 sid = task_session(p);
239
240         return sid;
241 }
242
243 /*
244  * Determine if a process group is "orphaned", according to the POSIX
245  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
246  * by terminal-generated stop signals.  Newly orphaned process groups are
247  * to receive a SIGHUP and a SIGCONT.
248  *
249  * "I ask you, have you ever known what it is to be an orphan?"
250  */
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 {
253         struct task_struct *p;
254
255         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256                 if ((p == ignored_task) ||
257                     (p->exit_state && thread_group_empty(p)) ||
258                     is_global_init(p->real_parent))
259                         continue;
260
261                 if (task_pgrp(p->real_parent) != pgrp &&
262                     task_session(p->real_parent) == task_session(p))
263                         return 0;
264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265
266         return 1;
267 }
268
269 int is_current_pgrp_orphaned(void)
270 {
271         int retval;
272
273         read_lock(&tasklist_lock);
274         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275         read_unlock(&tasklist_lock);
276
277         return retval;
278 }
279
280 static int has_stopped_jobs(struct pid *pgrp)
281 {
282         int retval = 0;
283         struct task_struct *p;
284
285         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286                 if (!task_is_stopped(p))
287                         continue;
288                 retval = 1;
289                 break;
290         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
291         return retval;
292 }
293
294 /*
295  * Check to see if any process groups have become orphaned as
296  * a result of our exiting, and if they have any stopped jobs,
297  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
298  */
299 static void
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 {
302         struct pid *pgrp = task_pgrp(tsk);
303         struct task_struct *ignored_task = tsk;
304
305         if (!parent)
306                  /* exit: our father is in a different pgrp than
307                   * we are and we were the only connection outside.
308                   */
309                 parent = tsk->real_parent;
310         else
311                 /* reparent: our child is in a different pgrp than
312                  * we are, and it was the only connection outside.
313                  */
314                 ignored_task = NULL;
315
316         if (task_pgrp(parent) != pgrp &&
317             task_session(parent) == task_session(tsk) &&
318             will_become_orphaned_pgrp(pgrp, ignored_task) &&
319             has_stopped_jobs(pgrp)) {
320                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
322         }
323 }
324
325 /**
326  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327  *
328  * If a kernel thread is launched as a result of a system call, or if
329  * it ever exits, it should generally reparent itself to kthreadd so it
330  * isn't in the way of other processes and is correctly cleaned up on exit.
331  *
332  * The various task state such as scheduling policy and priority may have
333  * been inherited from a user process, so we reset them to sane values here.
334  *
335  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336  */
337 static void reparent_to_kthreadd(void)
338 {
339         write_lock_irq(&tasklist_lock);
340
341         ptrace_unlink(current);
342         /* Reparent to init */
343         current->real_parent = current->parent = kthreadd_task;
344         list_move_tail(&current->sibling, &current->real_parent->children);
345
346         /* Set the exit signal to SIGCHLD so we signal init on exit */
347         current->exit_signal = SIGCHLD;
348
349         if (task_nice(current) < 0)
350                 set_user_nice(current, 0);
351         /* cpus_allowed? */
352         /* rt_priority? */
353         /* signals? */
354         memcpy(current->signal->rlim, init_task.signal->rlim,
355                sizeof(current->signal->rlim));
356
357         atomic_inc(&init_cred.usage);
358         commit_creds(&init_cred);
359         write_unlock_irq(&tasklist_lock);
360 }
361
362 void __set_special_pids(struct pid *pid)
363 {
364         struct task_struct *curr = current->group_leader;
365         pid_t nr = pid_nr(pid);
366
367         if (task_session(curr) != pid) {
368                 change_pid(curr, PIDTYPE_SID, pid);
369                 set_task_session(curr, nr);
370         }
371         if (task_pgrp(curr) != pid) {
372                 change_pid(curr, PIDTYPE_PGID, pid);
373                 set_task_pgrp(curr, nr);
374         }
375 }
376
377 static void set_special_pids(struct pid *pid)
378 {
379         write_lock_irq(&tasklist_lock);
380         __set_special_pids(pid);
381         write_unlock_irq(&tasklist_lock);
382 }
383
384 /*
385  * Let kernel threads use this to say that they
386  * allow a certain signal (since daemonize() will
387  * have disabled all of them by default).
388  */
389 int allow_signal(int sig)
390 {
391         if (!valid_signal(sig) || sig < 1)
392                 return -EINVAL;
393
394         spin_lock_irq(&current->sighand->siglock);
395         sigdelset(&current->blocked, sig);
396         if (!current->mm) {
397                 /* Kernel threads handle their own signals.
398                    Let the signal code know it'll be handled, so
399                    that they don't get converted to SIGKILL or
400                    just silently dropped */
401                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
402         }
403         recalc_sigpending();
404         spin_unlock_irq(&current->sighand->siglock);
405         return 0;
406 }
407
408 EXPORT_SYMBOL(allow_signal);
409
410 int disallow_signal(int sig)
411 {
412         if (!valid_signal(sig) || sig < 1)
413                 return -EINVAL;
414
415         spin_lock_irq(&current->sighand->siglock);
416         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
417         recalc_sigpending();
418         spin_unlock_irq(&current->sighand->siglock);
419         return 0;
420 }
421
422 EXPORT_SYMBOL(disallow_signal);
423
424 /*
425  *      Put all the gunge required to become a kernel thread without
426  *      attached user resources in one place where it belongs.
427  */
428
429 void daemonize(const char *name, ...)
430 {
431         va_list args;
432         struct fs_struct *fs;
433         sigset_t blocked;
434
435         va_start(args, name);
436         vsnprintf(current->comm, sizeof(current->comm), name, args);
437         va_end(args);
438
439         /*
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.
443          */
444         exit_mm(current);
445         /*
446          * We don't want to have TIF_FREEZE set if the system-wide hibernation
447          * or suspend transition begins right now.
448          */
449         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
450
451         if (current->nsproxy != &init_nsproxy) {
452                 get_nsproxy(&init_nsproxy);
453                 switch_task_namespaces(current, &init_nsproxy);
454         }
455         set_special_pids(&init_struct_pid);
456         proc_clear_tty(current);
457
458         /* Block and flush all signals */
459         sigfillset(&blocked);
460         sigprocmask(SIG_BLOCK, &blocked, NULL);
461         flush_signals(current);
462
463         /* Become as one with the init task */
464
465         exit_fs(current);       /* current->fs->count--; */
466         fs = init_task.fs;
467         current->fs = fs;
468         atomic_inc(&fs->count);
469
470         exit_files(current);
471         current->files = init_task.files;
472         atomic_inc(&current->files->count);
473
474         reparent_to_kthreadd();
475 }
476
477 EXPORT_SYMBOL(daemonize);
478
479 static void close_files(struct files_struct * files)
480 {
481         int i, j;
482         struct fdtable *fdt;
483
484         j = 0;
485
486         /*
487          * It is safe to dereference the fd table without RCU or
488          * ->file_lock because this is the last reference to the
489          * files structure.
490          */
491         fdt = files_fdtable(files);
492         for (;;) {
493                 unsigned long set;
494                 i = j * __NFDBITS;
495                 if (i >= fdt->max_fds)
496                         break;
497                 set = fdt->open_fds->fds_bits[j++];
498                 while (set) {
499                         if (set & 1) {
500                                 struct file * file = xchg(&fdt->fd[i], NULL);
501                                 if (file) {
502                                         filp_close(file, files);
503                                         cond_resched();
504                                 }
505                         }
506                         i++;
507                         set >>= 1;
508                 }
509         }
510 }
511
512 struct files_struct *get_files_struct(struct task_struct *task)
513 {
514         struct files_struct *files;
515
516         task_lock(task);
517         files = task->files;
518         if (files)
519                 atomic_inc(&files->count);
520         task_unlock(task);
521
522         return files;
523 }
524
525 void put_files_struct(struct files_struct *files)
526 {
527         struct fdtable *fdt;
528
529         if (atomic_dec_and_test(&files->count)) {
530                 close_files(files);
531                 /*
532                  * Free the fd and fdset arrays if we expanded them.
533                  * If the fdtable was embedded, pass files for freeing
534                  * at the end of the RCU grace period. Otherwise,
535                  * you can free files immediately.
536                  */
537                 fdt = files_fdtable(files);
538                 if (fdt != &files->fdtab)
539                         kmem_cache_free(files_cachep, files);
540                 free_fdtable(fdt);
541         }
542 }
543
544 void reset_files_struct(struct files_struct *files)
545 {
546         struct task_struct *tsk = current;
547         struct files_struct *old;
548
549         old = tsk->files;
550         task_lock(tsk);
551         tsk->files = files;
552         task_unlock(tsk);
553         put_files_struct(old);
554 }
555
556 void exit_files(struct task_struct *tsk)
557 {
558         struct files_struct * files = tsk->files;
559
560         if (files) {
561                 task_lock(tsk);
562                 tsk->files = NULL;
563                 task_unlock(tsk);
564                 put_files_struct(files);
565         }
566 }
567
568 void put_fs_struct(struct fs_struct *fs)
569 {
570         /* No need to hold fs->lock if we are killing it */
571         if (atomic_dec_and_test(&fs->count)) {
572                 path_put(&fs->root);
573                 path_put(&fs->pwd);
574                 kmem_cache_free(fs_cachep, fs);
575         }
576 }
577
578 void exit_fs(struct task_struct *tsk)
579 {
580         struct fs_struct * fs = tsk->fs;
581
582         if (fs) {
583                 task_lock(tsk);
584                 tsk->fs = NULL;
585                 task_unlock(tsk);
586                 put_fs_struct(fs);
587         }
588 }
589
590 EXPORT_SYMBOL_GPL(exit_fs);
591
592 #ifdef CONFIG_MM_OWNER
593 /*
594  * Task p is exiting and it owned mm, lets find a new owner for it
595  */
596 static inline int
597 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
598 {
599         /*
600          * If there are other users of the mm and the owner (us) is exiting
601          * we need to find a new owner to take on the responsibility.
602          */
603         if (atomic_read(&mm->mm_users) <= 1)
604                 return 0;
605         if (mm->owner != p)
606                 return 0;
607         return 1;
608 }
609
610 void mm_update_next_owner(struct mm_struct *mm)
611 {
612         struct task_struct *c, *g, *p = current;
613
614 retry:
615         if (!mm_need_new_owner(mm, p))
616                 return;
617
618         read_lock(&tasklist_lock);
619         /*
620          * Search in the children
621          */
622         list_for_each_entry(c, &p->children, sibling) {
623                 if (c->mm == mm)
624                         goto assign_new_owner;
625         }
626
627         /*
628          * Search in the siblings
629          */
630         list_for_each_entry(c, &p->parent->children, sibling) {
631                 if (c->mm == mm)
632                         goto assign_new_owner;
633         }
634
635         /*
636          * Search through everything else. We should not get
637          * here often
638          */
639         do_each_thread(g, c) {
640                 if (c->mm == mm)
641                         goto assign_new_owner;
642         } while_each_thread(g, c);
643
644         read_unlock(&tasklist_lock);
645         /*
646          * We found no owner yet mm_users > 1: this implies that we are
647          * most likely racing with swapoff (try_to_unuse()) or /proc or
648          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
649          */
650         mm->owner = NULL;
651         return;
652
653 assign_new_owner:
654         BUG_ON(c == p);
655         get_task_struct(c);
656         /*
657          * The task_lock protects c->mm from changing.
658          * We always want mm->owner->mm == mm
659          */
660         task_lock(c);
661         /*
662          * Delay read_unlock() till we have the task_lock()
663          * to ensure that c does not slip away underneath us
664          */
665         read_unlock(&tasklist_lock);
666         if (c->mm != mm) {
667                 task_unlock(c);
668                 put_task_struct(c);
669                 goto retry;
670         }
671         mm->owner = c;
672         task_unlock(c);
673         put_task_struct(c);
674 }
675 #endif /* CONFIG_MM_OWNER */
676
677 /*
678  * Turn us into a lazy TLB process if we
679  * aren't already..
680  */
681 static void exit_mm(struct task_struct * tsk)
682 {
683         struct mm_struct *mm = tsk->mm;
684         struct core_state *core_state;
685
686         mm_release(tsk, mm);
687         if (!mm)
688                 return;
689         /*
690          * Serialize with any possible pending coredump.
691          * We must hold mmap_sem around checking core_state
692          * and clearing tsk->mm.  The core-inducing thread
693          * will increment ->nr_threads for each thread in the
694          * group with ->mm != NULL.
695          */
696         down_read(&mm->mmap_sem);
697         core_state = mm->core_state;
698         if (core_state) {
699                 struct core_thread self;
700                 up_read(&mm->mmap_sem);
701
702                 self.task = tsk;
703                 self.next = xchg(&core_state->dumper.next, &self);
704                 /*
705                  * Implies mb(), the result of xchg() must be visible
706                  * to core_state->dumper.
707                  */
708                 if (atomic_dec_and_test(&core_state->nr_threads))
709                         complete(&core_state->startup);
710
711                 for (;;) {
712                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
713                         if (!self.task) /* see coredump_finish() */
714                                 break;
715                         schedule();
716                 }
717                 __set_task_state(tsk, TASK_RUNNING);
718                 down_read(&mm->mmap_sem);
719         }
720         atomic_inc(&mm->mm_count);
721         BUG_ON(mm != tsk->active_mm);
722         /* more a memory barrier than a real lock */
723         task_lock(tsk);
724         tsk->mm = NULL;
725         up_read(&mm->mmap_sem);
726         enter_lazy_tlb(mm, current);
727         /* We don't want this task to be frozen prematurely */
728         clear_freeze_flag(tsk);
729         task_unlock(tsk);
730         mm_update_next_owner(mm);
731         mmput(mm);
732 }
733
734 /*
735  * Return nonzero if @parent's children should reap themselves.
736  *
737  * Called with write_lock_irq(&tasklist_lock) held.
738  */
739 static int ignoring_children(struct task_struct *parent)
740 {
741         int ret;
742         struct sighand_struct *psig = parent->sighand;
743         unsigned long flags;
744         spin_lock_irqsave(&psig->siglock, flags);
745         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
746                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
747         spin_unlock_irqrestore(&psig->siglock, flags);
748         return ret;
749 }
750
751 /*
752  * Detach all tasks we were using ptrace on.
753  * Any that need to be release_task'd are put on the @dead list.
754  *
755  * Called with write_lock(&tasklist_lock) held.
756  */
757 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
758 {
759         struct task_struct *p, *n;
760         int ign = -1;
761
762         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
763                 __ptrace_unlink(p);
764
765                 if (p->exit_state != EXIT_ZOMBIE)
766                         continue;
767
768                 /*
769                  * If it's a zombie, our attachedness prevented normal
770                  * parent notification or self-reaping.  Do notification
771                  * now if it would have happened earlier.  If it should
772                  * reap itself, add it to the @dead list.  We can't call
773                  * release_task() here because we already hold tasklist_lock.
774                  *
775                  * If it's our own child, there is no notification to do.
776                  * But if our normal children self-reap, then this child
777                  * was prevented by ptrace and we must reap it now.
778                  */
779                 if (!task_detached(p) && thread_group_empty(p)) {
780                         if (!same_thread_group(p->real_parent, parent))
781                                 do_notify_parent(p, p->exit_signal);
782                         else {
783                                 if (ign < 0)
784                                         ign = ignoring_children(parent);
785                                 if (ign)
786                                         p->exit_signal = -1;
787                         }
788                 }
789
790                 if (task_detached(p)) {
791                         /*
792                          * Mark it as in the process of being reaped.
793                          */
794                         p->exit_state = EXIT_DEAD;
795                         list_add(&p->ptrace_entry, dead);
796                 }
797         }
798 }
799
800 /*
801  * Finish up exit-time ptrace cleanup.
802  *
803  * Called without locks.
804  */
805 static void ptrace_exit_finish(struct task_struct *parent,
806                                struct list_head *dead)
807 {
808         struct task_struct *p, *n;
809
810         BUG_ON(!list_empty(&parent->ptraced));
811
812         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
813                 list_del_init(&p->ptrace_entry);
814                 release_task(p);
815         }
816 }
817
818 static void reparent_thread(struct task_struct *p, struct task_struct *father)
819 {
820         if (p->pdeath_signal)
821                 /* We already hold the tasklist_lock here.  */
822                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
823
824         list_move_tail(&p->sibling, &p->real_parent->children);
825
826         /* If this is a threaded reparent there is no need to
827          * notify anyone anything has happened.
828          */
829         if (same_thread_group(p->real_parent, father))
830                 return;
831
832         /* We don't want people slaying init.  */
833         if (!task_detached(p))
834                 p->exit_signal = SIGCHLD;
835
836         /* If we'd notified the old parent about this child's death,
837          * also notify the new parent.
838          */
839         if (!ptrace_reparented(p) &&
840             p->exit_state == EXIT_ZOMBIE &&
841             !task_detached(p) && thread_group_empty(p))
842                 do_notify_parent(p, p->exit_signal);
843
844         kill_orphaned_pgrp(p, father);
845 }
846
847 /*
848  * When we die, we re-parent all our children.
849  * Try to give them to another thread in our thread
850  * group, and if no such member exists, give it to
851  * the child reaper process (ie "init") in our pid
852  * space.
853  */
854 static struct task_struct *find_new_reaper(struct task_struct *father)
855 {
856         struct pid_namespace *pid_ns = task_active_pid_ns(father);
857         struct task_struct *thread;
858
859         thread = father;
860         while_each_thread(father, thread) {
861                 if (thread->flags & PF_EXITING)
862                         continue;
863                 if (unlikely(pid_ns->child_reaper == father))
864                         pid_ns->child_reaper = thread;
865                 return thread;
866         }
867
868         if (unlikely(pid_ns->child_reaper == father)) {
869                 write_unlock_irq(&tasklist_lock);
870                 if (unlikely(pid_ns == &init_pid_ns))
871                         panic("Attempted to kill init!");
872
873                 zap_pid_ns_processes(pid_ns);
874                 write_lock_irq(&tasklist_lock);
875                 /*
876                  * We can not clear ->child_reaper or leave it alone.
877                  * There may by stealth EXIT_DEAD tasks on ->children,
878                  * forget_original_parent() must move them somewhere.
879                  */
880                 pid_ns->child_reaper = init_pid_ns.child_reaper;
881         }
882
883         return pid_ns->child_reaper;
884 }
885
886 static void forget_original_parent(struct task_struct *father)
887 {
888         struct task_struct *p, *n, *reaper;
889         LIST_HEAD(ptrace_dead);
890
891         write_lock_irq(&tasklist_lock);
892         reaper = find_new_reaper(father);
893         /*
894          * First clean up ptrace if we were using it.
895          */
896         ptrace_exit(father, &ptrace_dead);
897
898         list_for_each_entry_safe(p, n, &father->children, sibling) {
899                 p->real_parent = reaper;
900                 if (p->parent == father) {
901                         BUG_ON(p->ptrace);
902                         p->parent = p->real_parent;
903                 }
904                 reparent_thread(p, father);
905         }
906
907         write_unlock_irq(&tasklist_lock);
908         BUG_ON(!list_empty(&father->children));
909
910         ptrace_exit_finish(father, &ptrace_dead);
911 }
912
913 /*
914  * Send signals to all our closest relatives so that they know
915  * to properly mourn us..
916  */
917 static void exit_notify(struct task_struct *tsk, int group_dead)
918 {
919         int signal;
920         void *cookie;
921
922         /*
923          * This does two things:
924          *
925          * A.  Make init inherit all the child processes
926          * B.  Check to see if any process groups have become orphaned
927          *      as a result of our exiting, and if they have any stopped
928          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
929          */
930         forget_original_parent(tsk);
931         exit_task_namespaces(tsk);
932
933         write_lock_irq(&tasklist_lock);
934         if (group_dead)
935                 kill_orphaned_pgrp(tsk->group_leader, NULL);
936
937         /* Let father know we died
938          *
939          * Thread signals are configurable, but you aren't going to use
940          * that to send signals to arbitary processes.
941          * That stops right now.
942          *
943          * If the parent exec id doesn't match the exec id we saved
944          * when we started then we know the parent has changed security
945          * domain.
946          *
947          * If our self_exec id doesn't match our parent_exec_id then
948          * we have changed execution domain as these two values started
949          * the same after a fork.
950          */
951         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
952             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
953              tsk->self_exec_id != tsk->parent_exec_id) &&
954             !capable(CAP_KILL))
955                 tsk->exit_signal = SIGCHLD;
956
957         signal = tracehook_notify_death(tsk, &cookie, group_dead);
958         if (signal >= 0)
959                 signal = do_notify_parent(tsk, signal);
960
961         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
962
963         /* mt-exec, de_thread() is waiting for us */
964         if (thread_group_leader(tsk) &&
965             tsk->signal->group_exit_task &&
966             tsk->signal->notify_count < 0)
967                 wake_up_process(tsk->signal->group_exit_task);
968
969         write_unlock_irq(&tasklist_lock);
970
971         tracehook_report_death(tsk, signal, cookie, group_dead);
972
973         /* If the process is dead, release it - nobody will wait for it */
974         if (signal == DEATH_REAP)
975                 release_task(tsk);
976 }
977
978 #ifdef CONFIG_DEBUG_STACK_USAGE
979 static void check_stack_usage(void)
980 {
981         static DEFINE_SPINLOCK(low_water_lock);
982         static int lowest_to_date = THREAD_SIZE;
983         unsigned long *n = end_of_stack(current);
984         unsigned long free;
985
986         while (*n == 0)
987                 n++;
988         free = (unsigned long)n - (unsigned long)end_of_stack(current);
989
990         if (free >= lowest_to_date)
991                 return;
992
993         spin_lock(&low_water_lock);
994         if (free < lowest_to_date) {
995                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
996                                 "left\n",
997                                 current->comm, free);
998                 lowest_to_date = free;
999         }
1000         spin_unlock(&low_water_lock);
1001 }
1002 #else
1003 static inline void check_stack_usage(void) {}
1004 #endif
1005
1006 NORET_TYPE void do_exit(long code)
1007 {
1008         struct task_struct *tsk = current;
1009         int group_dead;
1010
1011         profile_task_exit(tsk);
1012
1013         WARN_ON(atomic_read(&tsk->fs_excl));
1014
1015         if (unlikely(in_interrupt()))
1016                 panic("Aiee, killing interrupt handler!");
1017         if (unlikely(!tsk->pid))
1018                 panic("Attempted to kill the idle task!");
1019
1020         tracehook_report_exit(&code);
1021
1022         /*
1023          * We're taking recursive faults here in do_exit. Safest is to just
1024          * leave this task alone and wait for reboot.
1025          */
1026         if (unlikely(tsk->flags & PF_EXITING)) {
1027                 printk(KERN_ALERT
1028                         "Fixing recursive fault but reboot is needed!\n");
1029                 /*
1030                  * We can do this unlocked here. The futex code uses
1031                  * this flag just to verify whether the pi state
1032                  * cleanup has been done or not. In the worst case it
1033                  * loops once more. We pretend that the cleanup was
1034                  * done as there is no way to return. Either the
1035                  * OWNER_DIED bit is set by now or we push the blocked
1036                  * task into the wait for ever nirwana as well.
1037                  */
1038                 tsk->flags |= PF_EXITPIDONE;
1039                 set_current_state(TASK_UNINTERRUPTIBLE);
1040                 schedule();
1041         }
1042
1043         exit_signals(tsk);  /* sets PF_EXITING */
1044         /*
1045          * tsk->flags are checked in the futex code to protect against
1046          * an exiting task cleaning up the robust pi futexes.
1047          */
1048         smp_mb();
1049         spin_unlock_wait(&tsk->pi_lock);
1050
1051         if (unlikely(in_atomic()))
1052                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1053                                 current->comm, task_pid_nr(current),
1054                                 preempt_count());
1055
1056         acct_update_integrals(tsk);
1057
1058         group_dead = atomic_dec_and_test(&tsk->signal->live);
1059         if (group_dead) {
1060                 hrtimer_cancel(&tsk->signal->real_timer);
1061                 exit_itimers(tsk->signal);
1062         }
1063         acct_collect(code, group_dead);
1064         if (group_dead)
1065                 tty_audit_exit();
1066         if (unlikely(tsk->audit_context))
1067                 audit_free(tsk);
1068
1069         tsk->exit_code = code;
1070         taskstats_exit(tsk, group_dead);
1071
1072         exit_mm(tsk);
1073
1074         if (group_dead)
1075                 acct_process();
1076         trace_sched_process_exit(tsk);
1077
1078         exit_sem(tsk);
1079         exit_files(tsk);
1080         exit_fs(tsk);
1081         check_stack_usage();
1082         exit_thread();
1083         cgroup_exit(tsk, 1);
1084
1085         if (group_dead && tsk->signal->leader)
1086                 disassociate_ctty(1);
1087
1088         module_put(task_thread_info(tsk)->exec_domain->module);
1089         if (tsk->binfmt)
1090                 module_put(tsk->binfmt->module);
1091
1092         proc_exit_connector(tsk);
1093         exit_notify(tsk, group_dead);
1094 #ifdef CONFIG_NUMA
1095         mpol_put(tsk->mempolicy);
1096         tsk->mempolicy = NULL;
1097 #endif
1098 #ifdef CONFIG_FUTEX
1099         if (unlikely(!list_empty(&tsk->pi_state_list)))
1100                 exit_pi_state_list(tsk);
1101         if (unlikely(current->pi_state_cache))
1102                 kfree(current->pi_state_cache);
1103 #endif
1104         /*
1105          * Make sure we are holding no locks:
1106          */
1107         debug_check_no_locks_held(tsk);
1108         /*
1109          * We can do this unlocked here. The futex code uses this flag
1110          * just to verify whether the pi state cleanup has been done
1111          * or not. In the worst case it loops once more.
1112          */
1113         tsk->flags |= PF_EXITPIDONE;
1114
1115         if (tsk->io_context)
1116                 exit_io_context();
1117
1118         if (tsk->splice_pipe)
1119                 __free_pipe_info(tsk->splice_pipe);
1120
1121         preempt_disable();
1122         /* causes final put_task_struct in finish_task_switch(). */
1123         tsk->state = TASK_DEAD;
1124         schedule();
1125         BUG();
1126         /* Avoid "noreturn function does return".  */
1127         for (;;)
1128                 cpu_relax();    /* For when BUG is null */
1129 }
1130
1131 EXPORT_SYMBOL_GPL(do_exit);
1132
1133 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1134 {
1135         if (comp)
1136                 complete(comp);
1137
1138         do_exit(code);
1139 }
1140
1141 EXPORT_SYMBOL(complete_and_exit);
1142
1143 SYSCALL_DEFINE1(exit, int, error_code)
1144 {
1145         do_exit((error_code&0xff)<<8);
1146 }
1147
1148 /*
1149  * Take down every thread in the group.  This is called by fatal signals
1150  * as well as by sys_exit_group (below).
1151  */
1152 NORET_TYPE void
1153 do_group_exit(int exit_code)
1154 {
1155         struct signal_struct *sig = current->signal;
1156
1157         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1158
1159         if (signal_group_exit(sig))
1160                 exit_code = sig->group_exit_code;
1161         else if (!thread_group_empty(current)) {
1162                 struct sighand_struct *const sighand = current->sighand;
1163                 spin_lock_irq(&sighand->siglock);
1164                 if (signal_group_exit(sig))
1165                         /* Another thread got here before we took the lock.  */
1166                         exit_code = sig->group_exit_code;
1167                 else {
1168                         sig->group_exit_code = exit_code;
1169                         sig->flags = SIGNAL_GROUP_EXIT;
1170                         zap_other_threads(current);
1171                 }
1172                 spin_unlock_irq(&sighand->siglock);
1173         }
1174
1175         do_exit(exit_code);
1176         /* NOTREACHED */
1177 }
1178
1179 /*
1180  * this kills every thread in the thread group. Note that any externally
1181  * wait4()-ing process will get the correct exit code - even if this
1182  * thread is not the thread group leader.
1183  */
1184 SYSCALL_DEFINE1(exit_group, int, error_code)
1185 {
1186         do_group_exit((error_code & 0xff) << 8);
1187         /* NOTREACHED */
1188         return 0;
1189 }
1190
1191 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1192 {
1193         struct pid *pid = NULL;
1194         if (type == PIDTYPE_PID)
1195                 pid = task->pids[type].pid;
1196         else if (type < PIDTYPE_MAX)
1197                 pid = task->group_leader->pids[type].pid;
1198         return pid;
1199 }
1200
1201 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1202                           struct task_struct *p)
1203 {
1204         int err;
1205
1206         if (type < PIDTYPE_MAX) {
1207                 if (task_pid_type(p, type) != pid)
1208                         return 0;
1209         }
1210
1211         /* Wait for all children (clone and not) if __WALL is set;
1212          * otherwise, wait for clone children *only* if __WCLONE is
1213          * set; otherwise, wait for non-clone children *only*.  (Note:
1214          * A "clone" child here is one that reports to its parent
1215          * using a signal other than SIGCHLD.) */
1216         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1217             && !(options & __WALL))
1218                 return 0;
1219
1220         err = security_task_wait(p);
1221         if (err)
1222                 return err;
1223
1224         return 1;
1225 }
1226
1227 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1228                                int why, int status,
1229                                struct siginfo __user *infop,
1230                                struct rusage __user *rusagep)
1231 {
1232         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1233
1234         put_task_struct(p);
1235         if (!retval)
1236                 retval = put_user(SIGCHLD, &infop->si_signo);
1237         if (!retval)
1238                 retval = put_user(0, &infop->si_errno);
1239         if (!retval)
1240                 retval = put_user((short)why, &infop->si_code);
1241         if (!retval)
1242                 retval = put_user(pid, &infop->si_pid);
1243         if (!retval)
1244                 retval = put_user(uid, &infop->si_uid);
1245         if (!retval)
1246                 retval = put_user(status, &infop->si_status);
1247         if (!retval)
1248                 retval = pid;
1249         return retval;
1250 }
1251
1252 /*
1253  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1254  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1255  * the lock and this task is uninteresting.  If we return nonzero, we have
1256  * released the lock and the system call should return.
1257  */
1258 static int wait_task_zombie(struct task_struct *p, int options,
1259                             struct siginfo __user *infop,
1260                             int __user *stat_addr, struct rusage __user *ru)
1261 {
1262         unsigned long state;
1263         int retval, status, traced;
1264         pid_t pid = task_pid_vnr(p);
1265         uid_t uid = __task_cred(p)->uid;
1266
1267         if (!likely(options & WEXITED))
1268                 return 0;
1269
1270         if (unlikely(options & WNOWAIT)) {
1271                 int exit_code = p->exit_code;
1272                 int why, status;
1273
1274                 get_task_struct(p);
1275                 read_unlock(&tasklist_lock);
1276                 if ((exit_code & 0x7f) == 0) {
1277                         why = CLD_EXITED;
1278                         status = exit_code >> 8;
1279                 } else {
1280                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1281                         status = exit_code & 0x7f;
1282                 }
1283                 return wait_noreap_copyout(p, pid, uid, why,
1284                                            status, infop, ru);
1285         }
1286
1287         /*
1288          * Try to move the task's state to DEAD
1289          * only one thread is allowed to do this:
1290          */
1291         state = xchg(&p->exit_state, EXIT_DEAD);
1292         if (state != EXIT_ZOMBIE) {
1293                 BUG_ON(state != EXIT_DEAD);
1294                 return 0;
1295         }
1296
1297         traced = ptrace_reparented(p);
1298
1299         if (likely(!traced)) {
1300                 struct signal_struct *psig;
1301                 struct signal_struct *sig;
1302                 struct task_cputime cputime;
1303
1304                 /*
1305                  * The resource counters for the group leader are in its
1306                  * own task_struct.  Those for dead threads in the group
1307                  * are in its signal_struct, as are those for the child
1308                  * processes it has previously reaped.  All these
1309                  * accumulate in the parent's signal_struct c* fields.
1310                  *
1311                  * We don't bother to take a lock here to protect these
1312                  * p->signal fields, because they are only touched by
1313                  * __exit_signal, which runs with tasklist_lock
1314                  * write-locked anyway, and so is excluded here.  We do
1315                  * need to protect the access to p->parent->signal fields,
1316                  * as other threads in the parent group can be right
1317                  * here reaping other children at the same time.
1318                  *
1319                  * We use thread_group_cputime() to get times for the thread
1320                  * group, which consolidates times for all threads in the
1321                  * group including the group leader.
1322                  */
1323                 thread_group_cputime(p, &cputime);
1324                 spin_lock_irq(&p->parent->sighand->siglock);
1325                 psig = p->parent->signal;
1326                 sig = p->signal;
1327                 psig->cutime =
1328                         cputime_add(psig->cutime,
1329                         cputime_add(cputime.utime,
1330                                     sig->cutime));
1331                 psig->cstime =
1332                         cputime_add(psig->cstime,
1333                         cputime_add(cputime.stime,
1334                                     sig->cstime));
1335                 psig->cgtime =
1336                         cputime_add(psig->cgtime,
1337                         cputime_add(p->gtime,
1338                         cputime_add(sig->gtime,
1339                                     sig->cgtime)));
1340                 psig->cmin_flt +=
1341                         p->min_flt + sig->min_flt + sig->cmin_flt;
1342                 psig->cmaj_flt +=
1343                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1344                 psig->cnvcsw +=
1345                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1346                 psig->cnivcsw +=
1347                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1348                 psig->cinblock +=
1349                         task_io_get_inblock(p) +
1350                         sig->inblock + sig->cinblock;
1351                 psig->coublock +=
1352                         task_io_get_oublock(p) +
1353                         sig->oublock + sig->coublock;
1354                 task_io_accounting_add(&psig->ioac, &p->ioac);
1355                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1356                 spin_unlock_irq(&p->parent->sighand->siglock);
1357         }
1358
1359         /*
1360          * Now we are sure this task is interesting, and no other
1361          * thread can reap it because we set its state to EXIT_DEAD.
1362          */
1363         read_unlock(&tasklist_lock);
1364
1365         /*
1366          * Flush inherited counters to the parent - before the parent
1367          * gets woken up by child-exit notifications.
1368          */
1369         perf_counter_exit_task(p);
1370
1371         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1372         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1373                 ? p->signal->group_exit_code : p->exit_code;
1374         if (!retval && stat_addr)
1375                 retval = put_user(status, stat_addr);
1376         if (!retval && infop)
1377                 retval = put_user(SIGCHLD, &infop->si_signo);
1378         if (!retval && infop)
1379                 retval = put_user(0, &infop->si_errno);
1380         if (!retval && infop) {
1381                 int why;
1382
1383                 if ((status & 0x7f) == 0) {
1384                         why = CLD_EXITED;
1385                         status >>= 8;
1386                 } else {
1387                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1388                         status &= 0x7f;
1389                 }
1390                 retval = put_user((short)why, &infop->si_code);
1391                 if (!retval)
1392                         retval = put_user(status, &infop->si_status);
1393         }
1394         if (!retval && infop)
1395                 retval = put_user(pid, &infop->si_pid);
1396         if (!retval && infop)
1397                 retval = put_user(uid, &infop->si_uid);
1398         if (!retval)
1399                 retval = pid;
1400
1401         if (traced) {
1402                 write_lock_irq(&tasklist_lock);
1403                 /* We dropped tasklist, ptracer could die and untrace */
1404                 ptrace_unlink(p);
1405                 /*
1406                  * If this is not a detached task, notify the parent.
1407                  * If it's still not detached after that, don't release
1408                  * it now.
1409                  */
1410                 if (!task_detached(p)) {
1411                         do_notify_parent(p, p->exit_signal);
1412                         if (!task_detached(p)) {
1413                                 p->exit_state = EXIT_ZOMBIE;
1414                                 p = NULL;
1415                         }
1416                 }
1417                 write_unlock_irq(&tasklist_lock);
1418         }
1419         if (p != NULL)
1420                 release_task(p);
1421
1422         return retval;
1423 }
1424
1425 /*
1426  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1427  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1428  * the lock and this task is uninteresting.  If we return nonzero, we have
1429  * released the lock and the system call should return.
1430  */
1431 static int wait_task_stopped(int ptrace, struct task_struct *p,
1432                              int options, struct siginfo __user *infop,
1433                              int __user *stat_addr, struct rusage __user *ru)
1434 {
1435         int retval, exit_code, why;
1436         uid_t uid = 0; /* unneeded, required by compiler */
1437         pid_t pid;
1438
1439         if (!(options & WUNTRACED))
1440                 return 0;
1441
1442         exit_code = 0;
1443         spin_lock_irq(&p->sighand->siglock);
1444
1445         if (unlikely(!task_is_stopped_or_traced(p)))
1446                 goto unlock_sig;
1447
1448         if (!ptrace && p->signal->group_stop_count > 0)
1449                 /*
1450                  * A group stop is in progress and this is the group leader.
1451                  * We won't report until all threads have stopped.
1452                  */
1453                 goto unlock_sig;
1454
1455         exit_code = p->exit_code;
1456         if (!exit_code)
1457                 goto unlock_sig;
1458
1459         if (!unlikely(options & WNOWAIT))
1460                 p->exit_code = 0;
1461
1462         /* don't need the RCU readlock here as we're holding a spinlock */
1463         uid = __task_cred(p)->uid;
1464 unlock_sig:
1465         spin_unlock_irq(&p->sighand->siglock);
1466         if (!exit_code)
1467                 return 0;
1468
1469         /*
1470          * Now we are pretty sure this task is interesting.
1471          * Make sure it doesn't get reaped out from under us while we
1472          * give up the lock and then examine it below.  We don't want to
1473          * keep holding onto the tasklist_lock while we call getrusage and
1474          * possibly take page faults for user memory.
1475          */
1476         get_task_struct(p);
1477         pid = task_pid_vnr(p);
1478         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1479         read_unlock(&tasklist_lock);
1480
1481         if (unlikely(options & WNOWAIT))
1482                 return wait_noreap_copyout(p, pid, uid,
1483                                            why, exit_code,
1484                                            infop, ru);
1485
1486         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1487         if (!retval && stat_addr)
1488                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1489         if (!retval && infop)
1490                 retval = put_user(SIGCHLD, &infop->si_signo);
1491         if (!retval && infop)
1492                 retval = put_user(0, &infop->si_errno);
1493         if (!retval && infop)
1494                 retval = put_user((short)why, &infop->si_code);
1495         if (!retval && infop)
1496                 retval = put_user(exit_code, &infop->si_status);
1497         if (!retval && infop)
1498                 retval = put_user(pid, &infop->si_pid);
1499         if (!retval && infop)
1500                 retval = put_user(uid, &infop->si_uid);
1501         if (!retval)
1502                 retval = pid;
1503         put_task_struct(p);
1504
1505         BUG_ON(!retval);
1506         return retval;
1507 }
1508
1509 /*
1510  * Handle do_wait work for one task in a live, non-stopped state.
1511  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1512  * the lock and this task is uninteresting.  If we return nonzero, we have
1513  * released the lock and the system call should return.
1514  */
1515 static int wait_task_continued(struct task_struct *p, int options,
1516                                struct siginfo __user *infop,
1517                                int __user *stat_addr, struct rusage __user *ru)
1518 {
1519         int retval;
1520         pid_t pid;
1521         uid_t uid;
1522
1523         if (!unlikely(options & WCONTINUED))
1524                 return 0;
1525
1526         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1527                 return 0;
1528
1529         spin_lock_irq(&p->sighand->siglock);
1530         /* Re-check with the lock held.  */
1531         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1532                 spin_unlock_irq(&p->sighand->siglock);
1533                 return 0;
1534         }
1535         if (!unlikely(options & WNOWAIT))
1536                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1537         uid = __task_cred(p)->uid;
1538         spin_unlock_irq(&p->sighand->siglock);
1539
1540         pid = task_pid_vnr(p);
1541         get_task_struct(p);
1542         read_unlock(&tasklist_lock);
1543
1544         if (!infop) {
1545                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1546                 put_task_struct(p);
1547                 if (!retval && stat_addr)
1548                         retval = put_user(0xffff, stat_addr);
1549                 if (!retval)
1550                         retval = pid;
1551         } else {
1552                 retval = wait_noreap_copyout(p, pid, uid,
1553                                              CLD_CONTINUED, SIGCONT,
1554                                              infop, ru);
1555                 BUG_ON(retval == 0);
1556         }
1557
1558         return retval;
1559 }
1560
1561 /*
1562  * Consider @p for a wait by @parent.
1563  *
1564  * -ECHILD should be in *@notask_error before the first call.
1565  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1566  * Returns zero if the search for a child should continue;
1567  * then *@notask_error is 0 if @p is an eligible child,
1568  * or another error from security_task_wait(), or still -ECHILD.
1569  */
1570 static int wait_consider_task(struct task_struct *parent, int ptrace,
1571                               struct task_struct *p, int *notask_error,
1572                               enum pid_type type, struct pid *pid, int options,
1573                               struct siginfo __user *infop,
1574                               int __user *stat_addr, struct rusage __user *ru)
1575 {
1576         int ret = eligible_child(type, pid, options, p);
1577         if (!ret)
1578                 return ret;
1579
1580         if (unlikely(ret < 0)) {
1581                 /*
1582                  * If we have not yet seen any eligible child,
1583                  * then let this error code replace -ECHILD.
1584                  * A permission error will give the user a clue
1585                  * to look for security policy problems, rather
1586                  * than for mysterious wait bugs.
1587                  */
1588                 if (*notask_error)
1589                         *notask_error = ret;
1590         }
1591
1592         if (likely(!ptrace) && unlikely(p->ptrace)) {
1593                 /*
1594                  * This child is hidden by ptrace.
1595                  * We aren't allowed to see it now, but eventually we will.
1596                  */
1597                 *notask_error = 0;
1598                 return 0;
1599         }
1600
1601         if (p->exit_state == EXIT_DEAD)
1602                 return 0;
1603
1604         /*
1605          * We don't reap group leaders with subthreads.
1606          */
1607         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1608                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1609
1610         /*
1611          * It's stopped or running now, so it might
1612          * later continue, exit, or stop again.
1613          */
1614         *notask_error = 0;
1615
1616         if (task_is_stopped_or_traced(p))
1617                 return wait_task_stopped(ptrace, p, options,
1618                                          infop, stat_addr, ru);
1619
1620         return wait_task_continued(p, options, infop, stat_addr, ru);
1621 }
1622
1623 /*
1624  * Do the work of do_wait() for one thread in the group, @tsk.
1625  *
1626  * -ECHILD should be in *@notask_error before the first call.
1627  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1628  * Returns zero if the search for a child should continue; then
1629  * *@notask_error is 0 if there were any eligible children,
1630  * or another error from security_task_wait(), or still -ECHILD.
1631  */
1632 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1633                           enum pid_type type, struct pid *pid, int options,
1634                           struct siginfo __user *infop, int __user *stat_addr,
1635                           struct rusage __user *ru)
1636 {
1637         struct task_struct *p;
1638
1639         list_for_each_entry(p, &tsk->children, sibling) {
1640                 /*
1641                  * Do not consider detached threads.
1642                  */
1643                 if (!task_detached(p)) {
1644                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1645                                                      type, pid, options,
1646                                                      infop, stat_addr, ru);
1647                         if (ret)
1648                                 return ret;
1649                 }
1650         }
1651
1652         return 0;
1653 }
1654
1655 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1656                           enum pid_type type, struct pid *pid, int options,
1657                           struct siginfo __user *infop, int __user *stat_addr,
1658                           struct rusage __user *ru)
1659 {
1660         struct task_struct *p;
1661
1662         /*
1663          * Traditionally we see ptrace'd stopped tasks regardless of options.
1664          */
1665         options |= WUNTRACED;
1666
1667         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1668                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1669                                              type, pid, options,
1670                                              infop, stat_addr, ru);
1671                 if (ret)
1672                         return ret;
1673         }
1674
1675         return 0;
1676 }
1677
1678 static long do_wait(enum pid_type type, struct pid *pid, int options,
1679                     struct siginfo __user *infop, int __user *stat_addr,
1680                     struct rusage __user *ru)
1681 {
1682         DECLARE_WAITQUEUE(wait, current);
1683         struct task_struct *tsk;
1684         int retval;
1685
1686         trace_sched_process_wait(pid);
1687
1688         add_wait_queue(&current->signal->wait_chldexit,&wait);
1689 repeat:
1690         /*
1691          * If there is nothing that can match our critiera just get out.
1692          * We will clear @retval to zero if we see any child that might later
1693          * match our criteria, even if we are not able to reap it yet.
1694          */
1695         retval = -ECHILD;
1696         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1697                 goto end;
1698
1699         current->state = TASK_INTERRUPTIBLE;
1700         read_lock(&tasklist_lock);
1701         tsk = current;
1702         do {
1703                 int tsk_result = do_wait_thread(tsk, &retval,
1704                                                 type, pid, options,
1705                                                 infop, stat_addr, ru);
1706                 if (!tsk_result)
1707                         tsk_result = ptrace_do_wait(tsk, &retval,
1708                                                     type, pid, options,
1709                                                     infop, stat_addr, ru);
1710                 if (tsk_result) {
1711                         /*
1712                          * tasklist_lock is unlocked and we have a final result.
1713                          */
1714                         retval = tsk_result;
1715                         goto end;
1716                 }
1717
1718                 if (options & __WNOTHREAD)
1719                         break;
1720                 tsk = next_thread(tsk);
1721                 BUG_ON(tsk->signal != current->signal);
1722         } while (tsk != current);
1723         read_unlock(&tasklist_lock);
1724
1725         if (!retval && !(options & WNOHANG)) {
1726                 retval = -ERESTARTSYS;
1727                 if (!signal_pending(current)) {
1728                         schedule();
1729                         goto repeat;
1730                 }
1731         }
1732
1733 end:
1734         current->state = TASK_RUNNING;
1735         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1736         if (infop) {
1737                 if (retval > 0)
1738                         retval = 0;
1739                 else {
1740                         /*
1741                          * For a WNOHANG return, clear out all the fields
1742                          * we would set so the user can easily tell the
1743                          * difference.
1744                          */
1745                         if (!retval)
1746                                 retval = put_user(0, &infop->si_signo);
1747                         if (!retval)
1748                                 retval = put_user(0, &infop->si_errno);
1749                         if (!retval)
1750                                 retval = put_user(0, &infop->si_code);
1751                         if (!retval)
1752                                 retval = put_user(0, &infop->si_pid);
1753                         if (!retval)
1754                                 retval = put_user(0, &infop->si_uid);
1755                         if (!retval)
1756                                 retval = put_user(0, &infop->si_status);
1757                 }
1758         }
1759         return retval;
1760 }
1761
1762 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1763                 infop, int, options, struct rusage __user *, ru)
1764 {
1765         struct pid *pid = NULL;
1766         enum pid_type type;
1767         long ret;
1768
1769         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1770                 return -EINVAL;
1771         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1772                 return -EINVAL;
1773
1774         switch (which) {
1775         case P_ALL:
1776                 type = PIDTYPE_MAX;
1777                 break;
1778         case P_PID:
1779                 type = PIDTYPE_PID;
1780                 if (upid <= 0)
1781                         return -EINVAL;
1782                 break;
1783         case P_PGID:
1784                 type = PIDTYPE_PGID;
1785                 if (upid <= 0)
1786                         return -EINVAL;
1787                 break;
1788         default:
1789                 return -EINVAL;
1790         }
1791
1792         if (type < PIDTYPE_MAX)
1793                 pid = find_get_pid(upid);
1794         ret = do_wait(type, pid, options, infop, NULL, ru);
1795         put_pid(pid);
1796
1797         /* avoid REGPARM breakage on x86: */
1798         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1799         return ret;
1800 }
1801
1802 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1803                 int, options, struct rusage __user *, ru)
1804 {
1805         struct pid *pid = NULL;
1806         enum pid_type type;
1807         long ret;
1808
1809         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1810                         __WNOTHREAD|__WCLONE|__WALL))
1811                 return -EINVAL;
1812
1813         if (upid == -1)
1814                 type = PIDTYPE_MAX;
1815         else if (upid < 0) {
1816                 type = PIDTYPE_PGID;
1817                 pid = find_get_pid(-upid);
1818         } else if (upid == 0) {
1819                 type = PIDTYPE_PGID;
1820                 pid = get_pid(task_pgrp(current));
1821         } else /* upid > 0 */ {
1822                 type = PIDTYPE_PID;
1823                 pid = find_get_pid(upid);
1824         }
1825
1826         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1827         put_pid(pid);
1828
1829         /* avoid REGPARM breakage on x86: */
1830         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1831         return ret;
1832 }
1833
1834 #ifdef __ARCH_WANT_SYS_WAITPID
1835
1836 /*
1837  * sys_waitpid() remains for compatibility. waitpid() should be
1838  * implemented by calling sys_wait4() from libc.a.
1839  */
1840 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1841 {
1842         return sys_wait4(pid, stat_addr, options, NULL);
1843 }
1844
1845 #endif