[NET]: Make socket creation namespace safe.
[linux-2.6] / 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/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.h>
34 #include <linux/cpuset.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/freezer.h>
48
49 #include <asm/uaccess.h>
50 #include <asm/unistd.h>
51 #include <asm/pgtable.h>
52 #include <asm/mmu_context.h>
53
54 extern void sem_exit (void);
55
56 static void exit_mm(struct task_struct * tsk);
57
58 static void __unhash_process(struct task_struct *p)
59 {
60         nr_threads--;
61         detach_pid(p, PIDTYPE_PID);
62         if (thread_group_leader(p)) {
63                 detach_pid(p, PIDTYPE_PGID);
64                 detach_pid(p, PIDTYPE_SID);
65
66                 list_del_rcu(&p->tasks);
67                 __get_cpu_var(process_counts)--;
68         }
69         list_del_rcu(&p->thread_group);
70         remove_parent(p);
71 }
72
73 /*
74  * This function expects the tasklist_lock write-locked.
75  */
76 static void __exit_signal(struct task_struct *tsk)
77 {
78         struct signal_struct *sig = tsk->signal;
79         struct sighand_struct *sighand;
80
81         BUG_ON(!sig);
82         BUG_ON(!atomic_read(&sig->count));
83
84         rcu_read_lock();
85         sighand = rcu_dereference(tsk->sighand);
86         spin_lock(&sighand->siglock);
87
88         posix_cpu_timers_exit(tsk);
89         if (atomic_dec_and_test(&sig->count))
90                 posix_cpu_timers_exit_group(tsk);
91         else {
92                 /*
93                  * If there is any task waiting for the group exit
94                  * then notify it:
95                  */
96                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
97                         wake_up_process(sig->group_exit_task);
98                         sig->group_exit_task = NULL;
99                 }
100                 if (tsk == sig->curr_target)
101                         sig->curr_target = next_thread(tsk);
102                 /*
103                  * Accumulate here the counters for all threads but the
104                  * group leader as they die, so they can be added into
105                  * the process-wide totals when those are taken.
106                  * The group leader stays around as a zombie as long
107                  * as there are other threads.  When it gets reaped,
108                  * the exit.c code will add its counts into these totals.
109                  * We won't ever get here for the group leader, since it
110                  * will have been the last reference on the signal_struct.
111                  */
112                 sig->utime = cputime_add(sig->utime, tsk->utime);
113                 sig->stime = cputime_add(sig->stime, tsk->stime);
114                 sig->min_flt += tsk->min_flt;
115                 sig->maj_flt += tsk->maj_flt;
116                 sig->nvcsw += tsk->nvcsw;
117                 sig->nivcsw += tsk->nivcsw;
118                 sig->inblock += task_io_get_inblock(tsk);
119                 sig->oublock += task_io_get_oublock(tsk);
120                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
121                 sig = NULL; /* Marker for below. */
122         }
123
124         __unhash_process(tsk);
125
126         tsk->signal = NULL;
127         tsk->sighand = NULL;
128         spin_unlock(&sighand->siglock);
129         rcu_read_unlock();
130
131         __cleanup_sighand(sighand);
132         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
133         flush_sigqueue(&tsk->pending);
134         if (sig) {
135                 flush_sigqueue(&sig->shared_pending);
136                 taskstats_tgid_free(sig);
137                 __cleanup_signal(sig);
138         }
139 }
140
141 static void delayed_put_task_struct(struct rcu_head *rhp)
142 {
143         put_task_struct(container_of(rhp, struct task_struct, rcu));
144 }
145
146 void release_task(struct task_struct * p)
147 {
148         struct task_struct *leader;
149         int zap_leader;
150 repeat:
151         atomic_dec(&p->user->processes);
152         write_lock_irq(&tasklist_lock);
153         ptrace_unlink(p);
154         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
155         __exit_signal(p);
156
157         /*
158          * If we are the last non-leader member of the thread
159          * group, and the leader is zombie, then notify the
160          * group leader's parent process. (if it wants notification.)
161          */
162         zap_leader = 0;
163         leader = p->group_leader;
164         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
165                 BUG_ON(leader->exit_signal == -1);
166                 do_notify_parent(leader, leader->exit_signal);
167                 /*
168                  * If we were the last child thread and the leader has
169                  * exited already, and the leader's parent ignores SIGCHLD,
170                  * then we are the one who should release the leader.
171                  *
172                  * do_notify_parent() will have marked it self-reaping in
173                  * that case.
174                  */
175                 zap_leader = (leader->exit_signal == -1);
176         }
177
178         write_unlock_irq(&tasklist_lock);
179         proc_flush_task(p);
180         release_thread(p);
181         call_rcu(&p->rcu, delayed_put_task_struct);
182
183         p = leader;
184         if (unlikely(zap_leader))
185                 goto repeat;
186 }
187
188 /*
189  * This checks not only the pgrp, but falls back on the pid if no
190  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
191  * without this...
192  *
193  * The caller must hold rcu lock or the tasklist lock.
194  */
195 struct pid *session_of_pgrp(struct pid *pgrp)
196 {
197         struct task_struct *p;
198         struct pid *sid = NULL;
199
200         p = pid_task(pgrp, PIDTYPE_PGID);
201         if (p == NULL)
202                 p = pid_task(pgrp, PIDTYPE_PID);
203         if (p != NULL)
204                 sid = task_session(p);
205
206         return sid;
207 }
208
209 /*
210  * Determine if a process group is "orphaned", according to the POSIX
211  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
212  * by terminal-generated stop signals.  Newly orphaned process groups are
213  * to receive a SIGHUP and a SIGCONT.
214  *
215  * "I ask you, have you ever known what it is to be an orphan?"
216  */
217 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
218 {
219         struct task_struct *p;
220         int ret = 1;
221
222         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
223                 if (p == ignored_task
224                                 || p->exit_state
225                                 || is_init(p->real_parent))
226                         continue;
227                 if (task_pgrp(p->real_parent) != pgrp &&
228                     task_session(p->real_parent) == task_session(p)) {
229                         ret = 0;
230                         break;
231                 }
232         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
233         return ret;     /* (sighing) "Often!" */
234 }
235
236 int is_current_pgrp_orphaned(void)
237 {
238         int retval;
239
240         read_lock(&tasklist_lock);
241         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
242         read_unlock(&tasklist_lock);
243
244         return retval;
245 }
246
247 static int has_stopped_jobs(struct pid *pgrp)
248 {
249         int retval = 0;
250         struct task_struct *p;
251
252         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253                 if (p->state != TASK_STOPPED)
254                         continue;
255                 retval = 1;
256                 break;
257         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258         return retval;
259 }
260
261 /**
262  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
263  *
264  * If a kernel thread is launched as a result of a system call, or if
265  * it ever exits, it should generally reparent itself to kthreadd so it
266  * isn't in the way of other processes and is correctly cleaned up on exit.
267  *
268  * The various task state such as scheduling policy and priority may have
269  * been inherited from a user process, so we reset them to sane values here.
270  *
271  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
272  */
273 static void reparent_to_kthreadd(void)
274 {
275         write_lock_irq(&tasklist_lock);
276
277         ptrace_unlink(current);
278         /* Reparent to init */
279         remove_parent(current);
280         current->real_parent = current->parent = kthreadd_task;
281         add_parent(current);
282
283         /* Set the exit signal to SIGCHLD so we signal init on exit */
284         current->exit_signal = SIGCHLD;
285
286         if (task_nice(current) < 0)
287                 set_user_nice(current, 0);
288         /* cpus_allowed? */
289         /* rt_priority? */
290         /* signals? */
291         security_task_reparent_to_init(current);
292         memcpy(current->signal->rlim, init_task.signal->rlim,
293                sizeof(current->signal->rlim));
294         atomic_inc(&(INIT_USER->__count));
295         write_unlock_irq(&tasklist_lock);
296         switch_uid(INIT_USER);
297 }
298
299 void __set_special_pids(pid_t session, pid_t pgrp)
300 {
301         struct task_struct *curr = current->group_leader;
302
303         if (process_session(curr) != session) {
304                 detach_pid(curr, PIDTYPE_SID);
305                 set_signal_session(curr->signal, session);
306                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
307         }
308         if (process_group(curr) != pgrp) {
309                 detach_pid(curr, PIDTYPE_PGID);
310                 curr->signal->pgrp = pgrp;
311                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
312         }
313 }
314
315 static void set_special_pids(pid_t session, pid_t pgrp)
316 {
317         write_lock_irq(&tasklist_lock);
318         __set_special_pids(session, pgrp);
319         write_unlock_irq(&tasklist_lock);
320 }
321
322 /*
323  * Let kernel threads use this to say that they
324  * allow a certain signal (since daemonize() will
325  * have disabled all of them by default).
326  */
327 int allow_signal(int sig)
328 {
329         if (!valid_signal(sig) || sig < 1)
330                 return -EINVAL;
331
332         spin_lock_irq(&current->sighand->siglock);
333         sigdelset(&current->blocked, sig);
334         if (!current->mm) {
335                 /* Kernel threads handle their own signals.
336                    Let the signal code know it'll be handled, so
337                    that they don't get converted to SIGKILL or
338                    just silently dropped */
339                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
340         }
341         recalc_sigpending();
342         spin_unlock_irq(&current->sighand->siglock);
343         return 0;
344 }
345
346 EXPORT_SYMBOL(allow_signal);
347
348 int disallow_signal(int sig)
349 {
350         if (!valid_signal(sig) || sig < 1)
351                 return -EINVAL;
352
353         spin_lock_irq(&current->sighand->siglock);
354         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
355         recalc_sigpending();
356         spin_unlock_irq(&current->sighand->siglock);
357         return 0;
358 }
359
360 EXPORT_SYMBOL(disallow_signal);
361
362 /*
363  *      Put all the gunge required to become a kernel thread without
364  *      attached user resources in one place where it belongs.
365  */
366
367 void daemonize(const char *name, ...)
368 {
369         va_list args;
370         struct fs_struct *fs;
371         sigset_t blocked;
372
373         va_start(args, name);
374         vsnprintf(current->comm, sizeof(current->comm), name, args);
375         va_end(args);
376
377         /*
378          * If we were started as result of loading a module, close all of the
379          * user space pages.  We don't need them, and if we didn't close them
380          * they would be locked into memory.
381          */
382         exit_mm(current);
383         /*
384          * We don't want to have TIF_FREEZE set if the system-wide hibernation
385          * or suspend transition begins right now.
386          */
387         current->flags |= PF_NOFREEZE;
388
389         set_special_pids(1, 1);
390         proc_clear_tty(current);
391
392         /* Block and flush all signals */
393         sigfillset(&blocked);
394         sigprocmask(SIG_BLOCK, &blocked, NULL);
395         flush_signals(current);
396
397         /* Become as one with the init task */
398
399         exit_fs(current);       /* current->fs->count--; */
400         fs = init_task.fs;
401         current->fs = fs;
402         atomic_inc(&fs->count);
403
404         exit_task_namespaces(current);
405         current->nsproxy = init_task.nsproxy;
406         get_task_namespaces(current);
407
408         exit_files(current);
409         current->files = init_task.files;
410         atomic_inc(&current->files->count);
411
412         reparent_to_kthreadd();
413 }
414
415 EXPORT_SYMBOL(daemonize);
416
417 static void close_files(struct files_struct * files)
418 {
419         int i, j;
420         struct fdtable *fdt;
421
422         j = 0;
423
424         /*
425          * It is safe to dereference the fd table without RCU or
426          * ->file_lock because this is the last reference to the
427          * files structure.
428          */
429         fdt = files_fdtable(files);
430         for (;;) {
431                 unsigned long set;
432                 i = j * __NFDBITS;
433                 if (i >= fdt->max_fds)
434                         break;
435                 set = fdt->open_fds->fds_bits[j++];
436                 while (set) {
437                         if (set & 1) {
438                                 struct file * file = xchg(&fdt->fd[i], NULL);
439                                 if (file) {
440                                         filp_close(file, files);
441                                         cond_resched();
442                                 }
443                         }
444                         i++;
445                         set >>= 1;
446                 }
447         }
448 }
449
450 struct files_struct *get_files_struct(struct task_struct *task)
451 {
452         struct files_struct *files;
453
454         task_lock(task);
455         files = task->files;
456         if (files)
457                 atomic_inc(&files->count);
458         task_unlock(task);
459
460         return files;
461 }
462
463 void fastcall put_files_struct(struct files_struct *files)
464 {
465         struct fdtable *fdt;
466
467         if (atomic_dec_and_test(&files->count)) {
468                 close_files(files);
469                 /*
470                  * Free the fd and fdset arrays if we expanded them.
471                  * If the fdtable was embedded, pass files for freeing
472                  * at the end of the RCU grace period. Otherwise,
473                  * you can free files immediately.
474                  */
475                 fdt = files_fdtable(files);
476                 if (fdt != &files->fdtab)
477                         kmem_cache_free(files_cachep, files);
478                 free_fdtable(fdt);
479         }
480 }
481
482 EXPORT_SYMBOL(put_files_struct);
483
484 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
485 {
486         struct files_struct *old;
487
488         old = tsk->files;
489         task_lock(tsk);
490         tsk->files = files;
491         task_unlock(tsk);
492         put_files_struct(old);
493 }
494 EXPORT_SYMBOL(reset_files_struct);
495
496 static inline void __exit_files(struct task_struct *tsk)
497 {
498         struct files_struct * files = tsk->files;
499
500         if (files) {
501                 task_lock(tsk);
502                 tsk->files = NULL;
503                 task_unlock(tsk);
504                 put_files_struct(files);
505         }
506 }
507
508 void exit_files(struct task_struct *tsk)
509 {
510         __exit_files(tsk);
511 }
512
513 static inline void __put_fs_struct(struct fs_struct *fs)
514 {
515         /* No need to hold fs->lock if we are killing it */
516         if (atomic_dec_and_test(&fs->count)) {
517                 dput(fs->root);
518                 mntput(fs->rootmnt);
519                 dput(fs->pwd);
520                 mntput(fs->pwdmnt);
521                 if (fs->altroot) {
522                         dput(fs->altroot);
523                         mntput(fs->altrootmnt);
524                 }
525                 kmem_cache_free(fs_cachep, fs);
526         }
527 }
528
529 void put_fs_struct(struct fs_struct *fs)
530 {
531         __put_fs_struct(fs);
532 }
533
534 static inline void __exit_fs(struct task_struct *tsk)
535 {
536         struct fs_struct * fs = tsk->fs;
537
538         if (fs) {
539                 task_lock(tsk);
540                 tsk->fs = NULL;
541                 task_unlock(tsk);
542                 __put_fs_struct(fs);
543         }
544 }
545
546 void exit_fs(struct task_struct *tsk)
547 {
548         __exit_fs(tsk);
549 }
550
551 EXPORT_SYMBOL_GPL(exit_fs);
552
553 /*
554  * Turn us into a lazy TLB process if we
555  * aren't already..
556  */
557 static void exit_mm(struct task_struct * tsk)
558 {
559         struct mm_struct *mm = tsk->mm;
560
561         mm_release(tsk, mm);
562         if (!mm)
563                 return;
564         /*
565          * Serialize with any possible pending coredump.
566          * We must hold mmap_sem around checking core_waiters
567          * and clearing tsk->mm.  The core-inducing thread
568          * will increment core_waiters for each thread in the
569          * group with ->mm != NULL.
570          */
571         down_read(&mm->mmap_sem);
572         if (mm->core_waiters) {
573                 up_read(&mm->mmap_sem);
574                 down_write(&mm->mmap_sem);
575                 if (!--mm->core_waiters)
576                         complete(mm->core_startup_done);
577                 up_write(&mm->mmap_sem);
578
579                 wait_for_completion(&mm->core_done);
580                 down_read(&mm->mmap_sem);
581         }
582         atomic_inc(&mm->mm_count);
583         BUG_ON(mm != tsk->active_mm);
584         /* more a memory barrier than a real lock */
585         task_lock(tsk);
586         tsk->mm = NULL;
587         up_read(&mm->mmap_sem);
588         enter_lazy_tlb(mm, current);
589         /* We don't want this task to be frozen prematurely */
590         clear_freeze_flag(tsk);
591         task_unlock(tsk);
592         mmput(mm);
593 }
594
595 static inline void
596 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
597 {
598         /*
599          * Make sure we're not reparenting to ourselves and that
600          * the parent is not a zombie.
601          */
602         BUG_ON(p == reaper || reaper->exit_state);
603         p->real_parent = reaper;
604 }
605
606 static void
607 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
608 {
609         if (p->pdeath_signal)
610                 /* We already hold the tasklist_lock here.  */
611                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
612
613         /* Move the child from its dying parent to the new one.  */
614         if (unlikely(traced)) {
615                 /* Preserve ptrace links if someone else is tracing this child.  */
616                 list_del_init(&p->ptrace_list);
617                 if (p->parent != p->real_parent)
618                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
619         } else {
620                 /* If this child is being traced, then we're the one tracing it
621                  * anyway, so let go of it.
622                  */
623                 p->ptrace = 0;
624                 remove_parent(p);
625                 p->parent = p->real_parent;
626                 add_parent(p);
627
628                 if (p->state == TASK_TRACED) {
629                         /*
630                          * If it was at a trace stop, turn it into
631                          * a normal stop since it's no longer being
632                          * traced.
633                          */
634                         ptrace_untrace(p);
635                 }
636         }
637
638         /* If this is a threaded reparent there is no need to
639          * notify anyone anything has happened.
640          */
641         if (p->real_parent->group_leader == father->group_leader)
642                 return;
643
644         /* We don't want people slaying init.  */
645         if (p->exit_signal != -1)
646                 p->exit_signal = SIGCHLD;
647
648         /* If we'd notified the old parent about this child's death,
649          * also notify the new parent.
650          */
651         if (!traced && p->exit_state == EXIT_ZOMBIE &&
652             p->exit_signal != -1 && thread_group_empty(p))
653                 do_notify_parent(p, p->exit_signal);
654
655         /*
656          * process group orphan check
657          * Case ii: Our child is in a different pgrp
658          * than we are, and it was the only connection
659          * outside, so the child pgrp is now orphaned.
660          */
661         if ((task_pgrp(p) != task_pgrp(father)) &&
662             (task_session(p) == task_session(father))) {
663                 struct pid *pgrp = task_pgrp(p);
664
665                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
666                     has_stopped_jobs(pgrp)) {
667                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
668                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
669                 }
670         }
671 }
672
673 /*
674  * When we die, we re-parent all our children.
675  * Try to give them to another thread in our thread
676  * group, and if no such member exists, give it to
677  * the child reaper process (ie "init") in our pid
678  * space.
679  */
680 static void
681 forget_original_parent(struct task_struct *father, struct list_head *to_release)
682 {
683         struct task_struct *p, *reaper = father;
684         struct list_head *_p, *_n;
685
686         do {
687                 reaper = next_thread(reaper);
688                 if (reaper == father) {
689                         reaper = child_reaper(father);
690                         break;
691                 }
692         } while (reaper->exit_state);
693
694         /*
695          * There are only two places where our children can be:
696          *
697          * - in our child list
698          * - in our ptraced child list
699          *
700          * Search them and reparent children.
701          */
702         list_for_each_safe(_p, _n, &father->children) {
703                 int ptrace;
704                 p = list_entry(_p, struct task_struct, sibling);
705
706                 ptrace = p->ptrace;
707
708                 /* if father isn't the real parent, then ptrace must be enabled */
709                 BUG_ON(father != p->real_parent && !ptrace);
710
711                 if (father == p->real_parent) {
712                         /* reparent with a reaper, real father it's us */
713                         choose_new_parent(p, reaper);
714                         reparent_thread(p, father, 0);
715                 } else {
716                         /* reparent ptraced task to its real parent */
717                         __ptrace_unlink (p);
718                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
719                             thread_group_empty(p))
720                                 do_notify_parent(p, p->exit_signal);
721                 }
722
723                 /*
724                  * if the ptraced child is a zombie with exit_signal == -1
725                  * we must collect it before we exit, or it will remain
726                  * zombie forever since we prevented it from self-reap itself
727                  * while it was being traced by us, to be able to see it in wait4.
728                  */
729                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
730                         list_add(&p->ptrace_list, to_release);
731         }
732         list_for_each_safe(_p, _n, &father->ptrace_children) {
733                 p = list_entry(_p, struct task_struct, ptrace_list);
734                 choose_new_parent(p, reaper);
735                 reparent_thread(p, father, 1);
736         }
737 }
738
739 /*
740  * Send signals to all our closest relatives so that they know
741  * to properly mourn us..
742  */
743 static void exit_notify(struct task_struct *tsk)
744 {
745         int state;
746         struct task_struct *t;
747         struct list_head ptrace_dead, *_p, *_n;
748         struct pid *pgrp;
749
750         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
751             && !thread_group_empty(tsk)) {
752                 /*
753                  * This occurs when there was a race between our exit
754                  * syscall and a group signal choosing us as the one to
755                  * wake up.  It could be that we are the only thread
756                  * alerted to check for pending signals, but another thread
757                  * should be woken now to take the signal since we will not.
758                  * Now we'll wake all the threads in the group just to make
759                  * sure someone gets all the pending signals.
760                  */
761                 read_lock(&tasklist_lock);
762                 spin_lock_irq(&tsk->sighand->siglock);
763                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
764                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
765                                 recalc_sigpending_and_wake(t);
766                 spin_unlock_irq(&tsk->sighand->siglock);
767                 read_unlock(&tasklist_lock);
768         }
769
770         write_lock_irq(&tasklist_lock);
771
772         /*
773          * This does two things:
774          *
775          * A.  Make init inherit all the child processes
776          * B.  Check to see if any process groups have become orphaned
777          *      as a result of our exiting, and if they have any stopped
778          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
779          */
780
781         INIT_LIST_HEAD(&ptrace_dead);
782         forget_original_parent(tsk, &ptrace_dead);
783         BUG_ON(!list_empty(&tsk->children));
784         BUG_ON(!list_empty(&tsk->ptrace_children));
785
786         /*
787          * Check to see if any process groups have become orphaned
788          * as a result of our exiting, and if they have any stopped
789          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
790          *
791          * Case i: Our father is in a different pgrp than we are
792          * and we were the only connection outside, so our pgrp
793          * is about to become orphaned.
794          */
795          
796         t = tsk->real_parent;
797         
798         pgrp = task_pgrp(tsk);
799         if ((task_pgrp(t) != pgrp) &&
800             (task_session(t) == task_session(tsk)) &&
801             will_become_orphaned_pgrp(pgrp, tsk) &&
802             has_stopped_jobs(pgrp)) {
803                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
804                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
805         }
806
807         /* Let father know we died
808          *
809          * Thread signals are configurable, but you aren't going to use
810          * that to send signals to arbitary processes. 
811          * That stops right now.
812          *
813          * If the parent exec id doesn't match the exec id we saved
814          * when we started then we know the parent has changed security
815          * domain.
816          *
817          * If our self_exec id doesn't match our parent_exec_id then
818          * we have changed execution domain as these two values started
819          * the same after a fork.
820          */
821         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
822             ( tsk->parent_exec_id != t->self_exec_id  ||
823               tsk->self_exec_id != tsk->parent_exec_id)
824             && !capable(CAP_KILL))
825                 tsk->exit_signal = SIGCHLD;
826
827
828         /* If something other than our normal parent is ptracing us, then
829          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
830          * only has special meaning to our real parent.
831          */
832         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
833                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
834                 do_notify_parent(tsk, signal);
835         } else if (tsk->ptrace) {
836                 do_notify_parent(tsk, SIGCHLD);
837         }
838
839         state = EXIT_ZOMBIE;
840         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
841                 state = EXIT_DEAD;
842         tsk->exit_state = state;
843
844         write_unlock_irq(&tasklist_lock);
845
846         list_for_each_safe(_p, _n, &ptrace_dead) {
847                 list_del_init(_p);
848                 t = list_entry(_p, struct task_struct, ptrace_list);
849                 release_task(t);
850         }
851
852         /* If the process is dead, release it - nobody will wait for it */
853         if (state == EXIT_DEAD)
854                 release_task(tsk);
855 }
856
857 #ifdef CONFIG_DEBUG_STACK_USAGE
858 static void check_stack_usage(void)
859 {
860         static DEFINE_SPINLOCK(low_water_lock);
861         static int lowest_to_date = THREAD_SIZE;
862         unsigned long *n = end_of_stack(current);
863         unsigned long free;
864
865         while (*n == 0)
866                 n++;
867         free = (unsigned long)n - (unsigned long)end_of_stack(current);
868
869         if (free >= lowest_to_date)
870                 return;
871
872         spin_lock(&low_water_lock);
873         if (free < lowest_to_date) {
874                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
875                                 "left\n",
876                                 current->comm, free);
877                 lowest_to_date = free;
878         }
879         spin_unlock(&low_water_lock);
880 }
881 #else
882 static inline void check_stack_usage(void) {}
883 #endif
884
885 fastcall NORET_TYPE void do_exit(long code)
886 {
887         struct task_struct *tsk = current;
888         int group_dead;
889
890         profile_task_exit(tsk);
891
892         WARN_ON(atomic_read(&tsk->fs_excl));
893
894         if (unlikely(in_interrupt()))
895                 panic("Aiee, killing interrupt handler!");
896         if (unlikely(!tsk->pid))
897                 panic("Attempted to kill the idle task!");
898         if (unlikely(tsk == child_reaper(tsk))) {
899                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
900                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
901                 else
902                         panic("Attempted to kill init!");
903         }
904
905
906         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
907                 current->ptrace_message = code;
908                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
909         }
910
911         /*
912          * We're taking recursive faults here in do_exit. Safest is to just
913          * leave this task alone and wait for reboot.
914          */
915         if (unlikely(tsk->flags & PF_EXITING)) {
916                 printk(KERN_ALERT
917                         "Fixing recursive fault but reboot is needed!\n");
918                 /*
919                  * We can do this unlocked here. The futex code uses
920                  * this flag just to verify whether the pi state
921                  * cleanup has been done or not. In the worst case it
922                  * loops once more. We pretend that the cleanup was
923                  * done as there is no way to return. Either the
924                  * OWNER_DIED bit is set by now or we push the blocked
925                  * task into the wait for ever nirwana as well.
926                  */
927                 tsk->flags |= PF_EXITPIDONE;
928                 if (tsk->io_context)
929                         exit_io_context();
930                 set_current_state(TASK_UNINTERRUPTIBLE);
931                 schedule();
932         }
933
934         /*
935          * tsk->flags are checked in the futex code to protect against
936          * an exiting task cleaning up the robust pi futexes.
937          */
938         spin_lock_irq(&tsk->pi_lock);
939         tsk->flags |= PF_EXITING;
940         spin_unlock_irq(&tsk->pi_lock);
941
942         if (unlikely(in_atomic()))
943                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
944                                 current->comm, current->pid,
945                                 preempt_count());
946
947         acct_update_integrals(tsk);
948         if (tsk->mm) {
949                 update_hiwater_rss(tsk->mm);
950                 update_hiwater_vm(tsk->mm);
951         }
952         group_dead = atomic_dec_and_test(&tsk->signal->live);
953         if (group_dead) {
954                 hrtimer_cancel(&tsk->signal->real_timer);
955                 exit_itimers(tsk->signal);
956         }
957         acct_collect(code, group_dead);
958         if (unlikely(tsk->robust_list))
959                 exit_robust_list(tsk);
960 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
961         if (unlikely(tsk->compat_robust_list))
962                 compat_exit_robust_list(tsk);
963 #endif
964         if (group_dead)
965                 tty_audit_exit();
966         if (unlikely(tsk->audit_context))
967                 audit_free(tsk);
968
969         tsk->exit_code = code;
970         taskstats_exit(tsk, group_dead);
971
972         exit_mm(tsk);
973
974         if (group_dead)
975                 acct_process();
976         exit_sem(tsk);
977         __exit_files(tsk);
978         __exit_fs(tsk);
979         check_stack_usage();
980         exit_thread();
981         cpuset_exit(tsk);
982         exit_keys(tsk);
983
984         if (group_dead && tsk->signal->leader)
985                 disassociate_ctty(1);
986
987         module_put(task_thread_info(tsk)->exec_domain->module);
988         if (tsk->binfmt)
989                 module_put(tsk->binfmt->module);
990
991         proc_exit_connector(tsk);
992         exit_task_namespaces(tsk);
993         exit_notify(tsk);
994 #ifdef CONFIG_NUMA
995         mpol_free(tsk->mempolicy);
996         tsk->mempolicy = NULL;
997 #endif
998         /*
999          * This must happen late, after the PID is not
1000          * hashed anymore:
1001          */
1002         if (unlikely(!list_empty(&tsk->pi_state_list)))
1003                 exit_pi_state_list(tsk);
1004         if (unlikely(current->pi_state_cache))
1005                 kfree(current->pi_state_cache);
1006         /*
1007          * Make sure we are holding no locks:
1008          */
1009         debug_check_no_locks_held(tsk);
1010         /*
1011          * We can do this unlocked here. The futex code uses this flag
1012          * just to verify whether the pi state cleanup has been done
1013          * or not. In the worst case it loops once more.
1014          */
1015         tsk->flags |= PF_EXITPIDONE;
1016
1017         if (tsk->io_context)
1018                 exit_io_context();
1019
1020         if (tsk->splice_pipe)
1021                 __free_pipe_info(tsk->splice_pipe);
1022
1023         preempt_disable();
1024         /* causes final put_task_struct in finish_task_switch(). */
1025         tsk->state = TASK_DEAD;
1026
1027         schedule();
1028         BUG();
1029         /* Avoid "noreturn function does return".  */
1030         for (;;)
1031                 cpu_relax();    /* For when BUG is null */
1032 }
1033
1034 EXPORT_SYMBOL_GPL(do_exit);
1035
1036 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1037 {
1038         if (comp)
1039                 complete(comp);
1040
1041         do_exit(code);
1042 }
1043
1044 EXPORT_SYMBOL(complete_and_exit);
1045
1046 asmlinkage long sys_exit(int error_code)
1047 {
1048         do_exit((error_code&0xff)<<8);
1049 }
1050
1051 /*
1052  * Take down every thread in the group.  This is called by fatal signals
1053  * as well as by sys_exit_group (below).
1054  */
1055 NORET_TYPE void
1056 do_group_exit(int exit_code)
1057 {
1058         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1059
1060         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1061                 exit_code = current->signal->group_exit_code;
1062         else if (!thread_group_empty(current)) {
1063                 struct signal_struct *const sig = current->signal;
1064                 struct sighand_struct *const sighand = current->sighand;
1065                 spin_lock_irq(&sighand->siglock);
1066                 if (sig->flags & SIGNAL_GROUP_EXIT)
1067                         /* Another thread got here before we took the lock.  */
1068                         exit_code = sig->group_exit_code;
1069                 else {
1070                         sig->group_exit_code = exit_code;
1071                         zap_other_threads(current);
1072                 }
1073                 spin_unlock_irq(&sighand->siglock);
1074         }
1075
1076         do_exit(exit_code);
1077         /* NOTREACHED */
1078 }
1079
1080 /*
1081  * this kills every thread in the thread group. Note that any externally
1082  * wait4()-ing process will get the correct exit code - even if this
1083  * thread is not the thread group leader.
1084  */
1085 asmlinkage void sys_exit_group(int error_code)
1086 {
1087         do_group_exit((error_code & 0xff) << 8);
1088 }
1089
1090 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1091 {
1092         int err;
1093
1094         if (pid > 0) {
1095                 if (p->pid != pid)
1096                         return 0;
1097         } else if (!pid) {
1098                 if (process_group(p) != process_group(current))
1099                         return 0;
1100         } else if (pid != -1) {
1101                 if (process_group(p) != -pid)
1102                         return 0;
1103         }
1104
1105         /*
1106          * Do not consider detached threads that are
1107          * not ptraced:
1108          */
1109         if (p->exit_signal == -1 && !p->ptrace)
1110                 return 0;
1111
1112         /* Wait for all children (clone and not) if __WALL is set;
1113          * otherwise, wait for clone children *only* if __WCLONE is
1114          * set; otherwise, wait for non-clone children *only*.  (Note:
1115          * A "clone" child here is one that reports to its parent
1116          * using a signal other than SIGCHLD.) */
1117         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1118             && !(options & __WALL))
1119                 return 0;
1120         /*
1121          * Do not consider thread group leaders that are
1122          * in a non-empty thread group:
1123          */
1124         if (delay_group_leader(p))
1125                 return 2;
1126
1127         err = security_task_wait(p);
1128         if (err)
1129                 return err;
1130
1131         return 1;
1132 }
1133
1134 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1135                                int why, int status,
1136                                struct siginfo __user *infop,
1137                                struct rusage __user *rusagep)
1138 {
1139         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1140
1141         put_task_struct(p);
1142         if (!retval)
1143                 retval = put_user(SIGCHLD, &infop->si_signo);
1144         if (!retval)
1145                 retval = put_user(0, &infop->si_errno);
1146         if (!retval)
1147                 retval = put_user((short)why, &infop->si_code);
1148         if (!retval)
1149                 retval = put_user(pid, &infop->si_pid);
1150         if (!retval)
1151                 retval = put_user(uid, &infop->si_uid);
1152         if (!retval)
1153                 retval = put_user(status, &infop->si_status);
1154         if (!retval)
1155                 retval = pid;
1156         return retval;
1157 }
1158
1159 /*
1160  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1161  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1162  * the lock and this task is uninteresting.  If we return nonzero, we have
1163  * released the lock and the system call should return.
1164  */
1165 static int wait_task_zombie(struct task_struct *p, int noreap,
1166                             struct siginfo __user *infop,
1167                             int __user *stat_addr, struct rusage __user *ru)
1168 {
1169         unsigned long state;
1170         int retval;
1171         int status;
1172
1173         if (unlikely(noreap)) {
1174                 pid_t pid = p->pid;
1175                 uid_t uid = p->uid;
1176                 int exit_code = p->exit_code;
1177                 int why, status;
1178
1179                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1180                         return 0;
1181                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1182                         return 0;
1183                 get_task_struct(p);
1184                 read_unlock(&tasklist_lock);
1185                 if ((exit_code & 0x7f) == 0) {
1186                         why = CLD_EXITED;
1187                         status = exit_code >> 8;
1188                 } else {
1189                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1190                         status = exit_code & 0x7f;
1191                 }
1192                 return wait_noreap_copyout(p, pid, uid, why,
1193                                            status, infop, ru);
1194         }
1195
1196         /*
1197          * Try to move the task's state to DEAD
1198          * only one thread is allowed to do this:
1199          */
1200         state = xchg(&p->exit_state, EXIT_DEAD);
1201         if (state != EXIT_ZOMBIE) {
1202                 BUG_ON(state != EXIT_DEAD);
1203                 return 0;
1204         }
1205         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1206                 /*
1207                  * This can only happen in a race with a ptraced thread
1208                  * dying on another processor.
1209                  */
1210                 return 0;
1211         }
1212
1213         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1214                 struct signal_struct *psig;
1215                 struct signal_struct *sig;
1216
1217                 /*
1218                  * The resource counters for the group leader are in its
1219                  * own task_struct.  Those for dead threads in the group
1220                  * are in its signal_struct, as are those for the child
1221                  * processes it has previously reaped.  All these
1222                  * accumulate in the parent's signal_struct c* fields.
1223                  *
1224                  * We don't bother to take a lock here to protect these
1225                  * p->signal fields, because they are only touched by
1226                  * __exit_signal, which runs with tasklist_lock
1227                  * write-locked anyway, and so is excluded here.  We do
1228                  * need to protect the access to p->parent->signal fields,
1229                  * as other threads in the parent group can be right
1230                  * here reaping other children at the same time.
1231                  */
1232                 spin_lock_irq(&p->parent->sighand->siglock);
1233                 psig = p->parent->signal;
1234                 sig = p->signal;
1235                 psig->cutime =
1236                         cputime_add(psig->cutime,
1237                         cputime_add(p->utime,
1238                         cputime_add(sig->utime,
1239                                     sig->cutime)));
1240                 psig->cstime =
1241                         cputime_add(psig->cstime,
1242                         cputime_add(p->stime,
1243                         cputime_add(sig->stime,
1244                                     sig->cstime)));
1245                 psig->cmin_flt +=
1246                         p->min_flt + sig->min_flt + sig->cmin_flt;
1247                 psig->cmaj_flt +=
1248                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1249                 psig->cnvcsw +=
1250                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1251                 psig->cnivcsw +=
1252                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1253                 psig->cinblock +=
1254                         task_io_get_inblock(p) +
1255                         sig->inblock + sig->cinblock;
1256                 psig->coublock +=
1257                         task_io_get_oublock(p) +
1258                         sig->oublock + sig->coublock;
1259                 spin_unlock_irq(&p->parent->sighand->siglock);
1260         }
1261
1262         /*
1263          * Now we are sure this task is interesting, and no other
1264          * thread can reap it because we set its state to EXIT_DEAD.
1265          */
1266         read_unlock(&tasklist_lock);
1267
1268         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1269         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1270                 ? p->signal->group_exit_code : p->exit_code;
1271         if (!retval && stat_addr)
1272                 retval = put_user(status, stat_addr);
1273         if (!retval && infop)
1274                 retval = put_user(SIGCHLD, &infop->si_signo);
1275         if (!retval && infop)
1276                 retval = put_user(0, &infop->si_errno);
1277         if (!retval && infop) {
1278                 int why;
1279
1280                 if ((status & 0x7f) == 0) {
1281                         why = CLD_EXITED;
1282                         status >>= 8;
1283                 } else {
1284                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1285                         status &= 0x7f;
1286                 }
1287                 retval = put_user((short)why, &infop->si_code);
1288                 if (!retval)
1289                         retval = put_user(status, &infop->si_status);
1290         }
1291         if (!retval && infop)
1292                 retval = put_user(p->pid, &infop->si_pid);
1293         if (!retval && infop)
1294                 retval = put_user(p->uid, &infop->si_uid);
1295         if (retval) {
1296                 // TODO: is this safe?
1297                 p->exit_state = EXIT_ZOMBIE;
1298                 return retval;
1299         }
1300         retval = p->pid;
1301         if (p->real_parent != p->parent) {
1302                 write_lock_irq(&tasklist_lock);
1303                 /* Double-check with lock held.  */
1304                 if (p->real_parent != p->parent) {
1305                         __ptrace_unlink(p);
1306                         // TODO: is this safe?
1307                         p->exit_state = EXIT_ZOMBIE;
1308                         /*
1309                          * If this is not a detached task, notify the parent.
1310                          * If it's still not detached after that, don't release
1311                          * it now.
1312                          */
1313                         if (p->exit_signal != -1) {
1314                                 do_notify_parent(p, p->exit_signal);
1315                                 if (p->exit_signal != -1)
1316                                         p = NULL;
1317                         }
1318                 }
1319                 write_unlock_irq(&tasklist_lock);
1320         }
1321         if (p != NULL)
1322                 release_task(p);
1323         BUG_ON(!retval);
1324         return retval;
1325 }
1326
1327 /*
1328  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1329  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1330  * the lock and this task is uninteresting.  If we return nonzero, we have
1331  * released the lock and the system call should return.
1332  */
1333 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1334                              int noreap, struct siginfo __user *infop,
1335                              int __user *stat_addr, struct rusage __user *ru)
1336 {
1337         int retval, exit_code;
1338
1339         if (!p->exit_code)
1340                 return 0;
1341         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1342             p->signal && p->signal->group_stop_count > 0)
1343                 /*
1344                  * A group stop is in progress and this is the group leader.
1345                  * We won't report until all threads have stopped.
1346                  */
1347                 return 0;
1348
1349         /*
1350          * Now we are pretty sure this task is interesting.
1351          * Make sure it doesn't get reaped out from under us while we
1352          * give up the lock and then examine it below.  We don't want to
1353          * keep holding onto the tasklist_lock while we call getrusage and
1354          * possibly take page faults for user memory.
1355          */
1356         get_task_struct(p);
1357         read_unlock(&tasklist_lock);
1358
1359         if (unlikely(noreap)) {
1360                 pid_t pid = p->pid;
1361                 uid_t uid = p->uid;
1362                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1363
1364                 exit_code = p->exit_code;
1365                 if (unlikely(!exit_code) ||
1366                     unlikely(p->state & TASK_TRACED))
1367                         goto bail_ref;
1368                 return wait_noreap_copyout(p, pid, uid,
1369                                            why, (exit_code << 8) | 0x7f,
1370                                            infop, ru);
1371         }
1372
1373         write_lock_irq(&tasklist_lock);
1374
1375         /*
1376          * This uses xchg to be atomic with the thread resuming and setting
1377          * it.  It must also be done with the write lock held to prevent a
1378          * race with the EXIT_ZOMBIE case.
1379          */
1380         exit_code = xchg(&p->exit_code, 0);
1381         if (unlikely(p->exit_state)) {
1382                 /*
1383                  * The task resumed and then died.  Let the next iteration
1384                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1385                  * already be zero here if it resumed and did _exit(0).
1386                  * The task itself is dead and won't touch exit_code again;
1387                  * other processors in this function are locked out.
1388                  */
1389                 p->exit_code = exit_code;
1390                 exit_code = 0;
1391         }
1392         if (unlikely(exit_code == 0)) {
1393                 /*
1394                  * Another thread in this function got to it first, or it
1395                  * resumed, or it resumed and then died.
1396                  */
1397                 write_unlock_irq(&tasklist_lock);
1398 bail_ref:
1399                 put_task_struct(p);
1400                 /*
1401                  * We are returning to the wait loop without having successfully
1402                  * removed the process and having released the lock. We cannot
1403                  * continue, since the "p" task pointer is potentially stale.
1404                  *
1405                  * Return -EAGAIN, and do_wait() will restart the loop from the
1406                  * beginning. Do _not_ re-acquire the lock.
1407                  */
1408                 return -EAGAIN;
1409         }
1410
1411         /* move to end of parent's list to avoid starvation */
1412         remove_parent(p);
1413         add_parent(p);
1414
1415         write_unlock_irq(&tasklist_lock);
1416
1417         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1418         if (!retval && stat_addr)
1419                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1420         if (!retval && infop)
1421                 retval = put_user(SIGCHLD, &infop->si_signo);
1422         if (!retval && infop)
1423                 retval = put_user(0, &infop->si_errno);
1424         if (!retval && infop)
1425                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1426                                           ? CLD_TRAPPED : CLD_STOPPED),
1427                                   &infop->si_code);
1428         if (!retval && infop)
1429                 retval = put_user(exit_code, &infop->si_status);
1430         if (!retval && infop)
1431                 retval = put_user(p->pid, &infop->si_pid);
1432         if (!retval && infop)
1433                 retval = put_user(p->uid, &infop->si_uid);
1434         if (!retval)
1435                 retval = p->pid;
1436         put_task_struct(p);
1437
1438         BUG_ON(!retval);
1439         return retval;
1440 }
1441
1442 /*
1443  * Handle do_wait work for one task in a live, non-stopped state.
1444  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1445  * the lock and this task is uninteresting.  If we return nonzero, we have
1446  * released the lock and the system call should return.
1447  */
1448 static int wait_task_continued(struct task_struct *p, int noreap,
1449                                struct siginfo __user *infop,
1450                                int __user *stat_addr, struct rusage __user *ru)
1451 {
1452         int retval;
1453         pid_t pid;
1454         uid_t uid;
1455
1456         if (unlikely(!p->signal))
1457                 return 0;
1458
1459         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1460                 return 0;
1461
1462         spin_lock_irq(&p->sighand->siglock);
1463         /* Re-check with the lock held.  */
1464         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1465                 spin_unlock_irq(&p->sighand->siglock);
1466                 return 0;
1467         }
1468         if (!noreap)
1469                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1470         spin_unlock_irq(&p->sighand->siglock);
1471
1472         pid = p->pid;
1473         uid = p->uid;
1474         get_task_struct(p);
1475         read_unlock(&tasklist_lock);
1476
1477         if (!infop) {
1478                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1479                 put_task_struct(p);
1480                 if (!retval && stat_addr)
1481                         retval = put_user(0xffff, stat_addr);
1482                 if (!retval)
1483                         retval = p->pid;
1484         } else {
1485                 retval = wait_noreap_copyout(p, pid, uid,
1486                                              CLD_CONTINUED, SIGCONT,
1487                                              infop, ru);
1488                 BUG_ON(retval == 0);
1489         }
1490
1491         return retval;
1492 }
1493
1494
1495 static inline int my_ptrace_child(struct task_struct *p)
1496 {
1497         if (!(p->ptrace & PT_PTRACED))
1498                 return 0;
1499         if (!(p->ptrace & PT_ATTACHED))
1500                 return 1;
1501         /*
1502          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1503          * we are the attacher.  If we are the real parent, this is a race
1504          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1505          * which we have to switch the parent links, but has already set
1506          * the flags in p->ptrace.
1507          */
1508         return (p->parent != p->real_parent);
1509 }
1510
1511 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1512                     int __user *stat_addr, struct rusage __user *ru)
1513 {
1514         DECLARE_WAITQUEUE(wait, current);
1515         struct task_struct *tsk;
1516         int flag, retval;
1517         int allowed, denied;
1518
1519         add_wait_queue(&current->signal->wait_chldexit,&wait);
1520 repeat:
1521         /*
1522          * We will set this flag if we see any child that might later
1523          * match our criteria, even if we are not able to reap it yet.
1524          */
1525         flag = 0;
1526         allowed = denied = 0;
1527         current->state = TASK_INTERRUPTIBLE;
1528         read_lock(&tasklist_lock);
1529         tsk = current;
1530         do {
1531                 struct task_struct *p;
1532                 struct list_head *_p;
1533                 int ret;
1534
1535                 list_for_each(_p,&tsk->children) {
1536                         p = list_entry(_p, struct task_struct, sibling);
1537
1538                         ret = eligible_child(pid, options, p);
1539                         if (!ret)
1540                                 continue;
1541
1542                         if (unlikely(ret < 0)) {
1543                                 denied = ret;
1544                                 continue;
1545                         }
1546                         allowed = 1;
1547
1548                         switch (p->state) {
1549                         case TASK_TRACED:
1550                                 /*
1551                                  * When we hit the race with PTRACE_ATTACH,
1552                                  * we will not report this child.  But the
1553                                  * race means it has not yet been moved to
1554                                  * our ptrace_children list, so we need to
1555                                  * set the flag here to avoid a spurious ECHILD
1556                                  * when the race happens with the only child.
1557                                  */
1558                                 flag = 1;
1559                                 if (!my_ptrace_child(p))
1560                                         continue;
1561                                 /*FALLTHROUGH*/
1562                         case TASK_STOPPED:
1563                                 /*
1564                                  * It's stopped now, so it might later
1565                                  * continue, exit, or stop again.
1566                                  */
1567                                 flag = 1;
1568                                 if (!(options & WUNTRACED) &&
1569                                     !my_ptrace_child(p))
1570                                         continue;
1571                                 retval = wait_task_stopped(p, ret == 2,
1572                                                            (options & WNOWAIT),
1573                                                            infop,
1574                                                            stat_addr, ru);
1575                                 if (retval == -EAGAIN)
1576                                         goto repeat;
1577                                 if (retval != 0) /* He released the lock.  */
1578                                         goto end;
1579                                 break;
1580                         default:
1581                         // case EXIT_DEAD:
1582                                 if (p->exit_state == EXIT_DEAD)
1583                                         continue;
1584                         // case EXIT_ZOMBIE:
1585                                 if (p->exit_state == EXIT_ZOMBIE) {
1586                                         /*
1587                                          * Eligible but we cannot release
1588                                          * it yet:
1589                                          */
1590                                         if (ret == 2)
1591                                                 goto check_continued;
1592                                         if (!likely(options & WEXITED))
1593                                                 continue;
1594                                         retval = wait_task_zombie(
1595                                                 p, (options & WNOWAIT),
1596                                                 infop, stat_addr, ru);
1597                                         /* He released the lock.  */
1598                                         if (retval != 0)
1599                                                 goto end;
1600                                         break;
1601                                 }
1602 check_continued:
1603                                 /*
1604                                  * It's running now, so it might later
1605                                  * exit, stop, or stop and then continue.
1606                                  */
1607                                 flag = 1;
1608                                 if (!unlikely(options & WCONTINUED))
1609                                         continue;
1610                                 retval = wait_task_continued(
1611                                         p, (options & WNOWAIT),
1612                                         infop, stat_addr, ru);
1613                                 if (retval != 0) /* He released the lock.  */
1614                                         goto end;
1615                                 break;
1616                         }
1617                 }
1618                 if (!flag) {
1619                         list_for_each(_p, &tsk->ptrace_children) {
1620                                 p = list_entry(_p, struct task_struct,
1621                                                 ptrace_list);
1622                                 if (!eligible_child(pid, options, p))
1623                                         continue;
1624                                 flag = 1;
1625                                 break;
1626                         }
1627                 }
1628                 if (options & __WNOTHREAD)
1629                         break;
1630                 tsk = next_thread(tsk);
1631                 BUG_ON(tsk->signal != current->signal);
1632         } while (tsk != current);
1633
1634         read_unlock(&tasklist_lock);
1635         if (flag) {
1636                 retval = 0;
1637                 if (options & WNOHANG)
1638                         goto end;
1639                 retval = -ERESTARTSYS;
1640                 if (signal_pending(current))
1641                         goto end;
1642                 schedule();
1643                 goto repeat;
1644         }
1645         retval = -ECHILD;
1646         if (unlikely(denied) && !allowed)
1647                 retval = denied;
1648 end:
1649         current->state = TASK_RUNNING;
1650         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1651         if (infop) {
1652                 if (retval > 0)
1653                 retval = 0;
1654                 else {
1655                         /*
1656                          * For a WNOHANG return, clear out all the fields
1657                          * we would set so the user can easily tell the
1658                          * difference.
1659                          */
1660                         if (!retval)
1661                                 retval = put_user(0, &infop->si_signo);
1662                         if (!retval)
1663                                 retval = put_user(0, &infop->si_errno);
1664                         if (!retval)
1665                                 retval = put_user(0, &infop->si_code);
1666                         if (!retval)
1667                                 retval = put_user(0, &infop->si_pid);
1668                         if (!retval)
1669                                 retval = put_user(0, &infop->si_uid);
1670                         if (!retval)
1671                                 retval = put_user(0, &infop->si_status);
1672                 }
1673         }
1674         return retval;
1675 }
1676
1677 asmlinkage long sys_waitid(int which, pid_t pid,
1678                            struct siginfo __user *infop, int options,
1679                            struct rusage __user *ru)
1680 {
1681         long ret;
1682
1683         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1684                 return -EINVAL;
1685         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1686                 return -EINVAL;
1687
1688         switch (which) {
1689         case P_ALL:
1690                 pid = -1;
1691                 break;
1692         case P_PID:
1693                 if (pid <= 0)
1694                         return -EINVAL;
1695                 break;
1696         case P_PGID:
1697                 if (pid <= 0)
1698                         return -EINVAL;
1699                 pid = -pid;
1700                 break;
1701         default:
1702                 return -EINVAL;
1703         }
1704
1705         ret = do_wait(pid, options, infop, NULL, ru);
1706
1707         /* avoid REGPARM breakage on x86: */
1708         prevent_tail_call(ret);
1709         return ret;
1710 }
1711
1712 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1713                           int options, struct rusage __user *ru)
1714 {
1715         long ret;
1716
1717         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1718                         __WNOTHREAD|__WCLONE|__WALL))
1719                 return -EINVAL;
1720         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1721
1722         /* avoid REGPARM breakage on x86: */
1723         prevent_tail_call(ret);
1724         return ret;
1725 }
1726
1727 #ifdef __ARCH_WANT_SYS_WAITPID
1728
1729 /*
1730  * sys_waitpid() remains for compatibility. waitpid() should be
1731  * implemented by calling sys_wait4() from libc.a.
1732  */
1733 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1734 {
1735         return sys_wait4(pid, stat_addr, options, NULL);
1736 }
1737
1738 #endif