wait_task_zombie: don't fight with non-existing race with a dying ptracee
[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
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
52
53 extern void sem_exit (void);
54
55 static void exit_mm(struct task_struct * tsk);
56
57 static void __unhash_process(struct task_struct *p)
58 {
59         nr_threads--;
60         detach_pid(p, PIDTYPE_PID);
61         if (thread_group_leader(p)) {
62                 detach_pid(p, PIDTYPE_PGID);
63                 detach_pid(p, PIDTYPE_SID);
64
65                 list_del_rcu(&p->tasks);
66                 __get_cpu_var(process_counts)--;
67         }
68         list_del_rcu(&p->thread_group);
69         remove_parent(p);
70 }
71
72 /*
73  * This function expects the tasklist_lock write-locked.
74  */
75 static void __exit_signal(struct task_struct *tsk)
76 {
77         struct signal_struct *sig = tsk->signal;
78         struct sighand_struct *sighand;
79
80         BUG_ON(!sig);
81         BUG_ON(!atomic_read(&sig->count));
82
83         rcu_read_lock();
84         sighand = rcu_dereference(tsk->sighand);
85         spin_lock(&sighand->siglock);
86
87         posix_cpu_timers_exit(tsk);
88         if (atomic_dec_and_test(&sig->count))
89                 posix_cpu_timers_exit_group(tsk);
90         else {
91                 /*
92                  * If there is any task waiting for the group exit
93                  * then notify it:
94                  */
95                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
96                         wake_up_process(sig->group_exit_task);
97                         sig->group_exit_task = NULL;
98                 }
99                 if (tsk == sig->curr_target)
100                         sig->curr_target = next_thread(tsk);
101                 /*
102                  * Accumulate here the counters for all threads but the
103                  * group leader as they die, so they can be added into
104                  * the process-wide totals when those are taken.
105                  * The group leader stays around as a zombie as long
106                  * as there are other threads.  When it gets reaped,
107                  * the exit.c code will add its counts into these totals.
108                  * We won't ever get here for the group leader, since it
109                  * will have been the last reference on the signal_struct.
110                  */
111                 sig->utime = cputime_add(sig->utime, tsk->utime);
112                 sig->stime = cputime_add(sig->stime, tsk->stime);
113                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
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 void
596 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
597 {
598         if (p->pdeath_signal)
599                 /* We already hold the tasklist_lock here.  */
600                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
601
602         /* Move the child from its dying parent to the new one.  */
603         if (unlikely(traced)) {
604                 /* Preserve ptrace links if someone else is tracing this child.  */
605                 list_del_init(&p->ptrace_list);
606                 if (p->parent != p->real_parent)
607                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
608         } else {
609                 /* If this child is being traced, then we're the one tracing it
610                  * anyway, so let go of it.
611                  */
612                 p->ptrace = 0;
613                 remove_parent(p);
614                 p->parent = p->real_parent;
615                 add_parent(p);
616
617                 if (p->state == TASK_TRACED) {
618                         /*
619                          * If it was at a trace stop, turn it into
620                          * a normal stop since it's no longer being
621                          * traced.
622                          */
623                         ptrace_untrace(p);
624                 }
625         }
626
627         /* If this is a threaded reparent there is no need to
628          * notify anyone anything has happened.
629          */
630         if (p->real_parent->group_leader == father->group_leader)
631                 return;
632
633         /* We don't want people slaying init.  */
634         if (p->exit_signal != -1)
635                 p->exit_signal = SIGCHLD;
636
637         /* If we'd notified the old parent about this child's death,
638          * also notify the new parent.
639          */
640         if (!traced && p->exit_state == EXIT_ZOMBIE &&
641             p->exit_signal != -1 && thread_group_empty(p))
642                 do_notify_parent(p, p->exit_signal);
643
644         /*
645          * process group orphan check
646          * Case ii: Our child is in a different pgrp
647          * than we are, and it was the only connection
648          * outside, so the child pgrp is now orphaned.
649          */
650         if ((task_pgrp(p) != task_pgrp(father)) &&
651             (task_session(p) == task_session(father))) {
652                 struct pid *pgrp = task_pgrp(p);
653
654                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
655                     has_stopped_jobs(pgrp)) {
656                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
657                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
658                 }
659         }
660 }
661
662 /*
663  * When we die, we re-parent all our children.
664  * Try to give them to another thread in our thread
665  * group, and if no such member exists, give it to
666  * the child reaper process (ie "init") in our pid
667  * space.
668  */
669 static void
670 forget_original_parent(struct task_struct *father, struct list_head *to_release)
671 {
672         struct task_struct *p, *reaper = father;
673         struct list_head *_p, *_n;
674
675         do {
676                 reaper = next_thread(reaper);
677                 if (reaper == father) {
678                         reaper = child_reaper(father);
679                         break;
680                 }
681         } while (reaper->exit_state);
682
683         /*
684          * There are only two places where our children can be:
685          *
686          * - in our child list
687          * - in our ptraced child list
688          *
689          * Search them and reparent children.
690          */
691         list_for_each_safe(_p, _n, &father->children) {
692                 int ptrace;
693                 p = list_entry(_p, struct task_struct, sibling);
694
695                 ptrace = p->ptrace;
696
697                 /* if father isn't the real parent, then ptrace must be enabled */
698                 BUG_ON(father != p->real_parent && !ptrace);
699
700                 if (father == p->real_parent) {
701                         /* reparent with a reaper, real father it's us */
702                         p->real_parent = reaper;
703                         reparent_thread(p, father, 0);
704                 } else {
705                         /* reparent ptraced task to its real parent */
706                         __ptrace_unlink (p);
707                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
708                             thread_group_empty(p))
709                                 do_notify_parent(p, p->exit_signal);
710                 }
711
712                 /*
713                  * if the ptraced child is a zombie with exit_signal == -1
714                  * we must collect it before we exit, or it will remain
715                  * zombie forever since we prevented it from self-reap itself
716                  * while it was being traced by us, to be able to see it in wait4.
717                  */
718                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
719                         list_add(&p->ptrace_list, to_release);
720         }
721         list_for_each_safe(_p, _n, &father->ptrace_children) {
722                 p = list_entry(_p, struct task_struct, ptrace_list);
723                 p->real_parent = reaper;
724                 reparent_thread(p, father, 1);
725         }
726 }
727
728 /*
729  * Send signals to all our closest relatives so that they know
730  * to properly mourn us..
731  */
732 static void exit_notify(struct task_struct *tsk)
733 {
734         int state;
735         struct task_struct *t;
736         struct list_head ptrace_dead, *_p, *_n;
737         struct pid *pgrp;
738
739         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
740             && !thread_group_empty(tsk)) {
741                 /*
742                  * This occurs when there was a race between our exit
743                  * syscall and a group signal choosing us as the one to
744                  * wake up.  It could be that we are the only thread
745                  * alerted to check for pending signals, but another thread
746                  * should be woken now to take the signal since we will not.
747                  * Now we'll wake all the threads in the group just to make
748                  * sure someone gets all the pending signals.
749                  */
750                 spin_lock_irq(&tsk->sighand->siglock);
751                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
752                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
753                                 recalc_sigpending_and_wake(t);
754                 spin_unlock_irq(&tsk->sighand->siglock);
755         }
756
757         write_lock_irq(&tasklist_lock);
758
759         /*
760          * This does two things:
761          *
762          * A.  Make init inherit all the child processes
763          * B.  Check to see if any process groups have become orphaned
764          *      as a result of our exiting, and if they have any stopped
765          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
766          */
767
768         INIT_LIST_HEAD(&ptrace_dead);
769         forget_original_parent(tsk, &ptrace_dead);
770         BUG_ON(!list_empty(&tsk->children));
771         BUG_ON(!list_empty(&tsk->ptrace_children));
772
773         /*
774          * Check to see if any process groups have become orphaned
775          * as a result of our exiting, and if they have any stopped
776          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
777          *
778          * Case i: Our father is in a different pgrp than we are
779          * and we were the only connection outside, so our pgrp
780          * is about to become orphaned.
781          */
782         t = tsk->real_parent;
783
784         pgrp = task_pgrp(tsk);
785         if ((task_pgrp(t) != pgrp) &&
786             (task_session(t) == task_session(tsk)) &&
787             will_become_orphaned_pgrp(pgrp, tsk) &&
788             has_stopped_jobs(pgrp)) {
789                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
790                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
791         }
792
793         /* Let father know we died
794          *
795          * Thread signals are configurable, but you aren't going to use
796          * that to send signals to arbitary processes. 
797          * That stops right now.
798          *
799          * If the parent exec id doesn't match the exec id we saved
800          * when we started then we know the parent has changed security
801          * domain.
802          *
803          * If our self_exec id doesn't match our parent_exec_id then
804          * we have changed execution domain as these two values started
805          * the same after a fork.
806          */
807         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
808             ( tsk->parent_exec_id != t->self_exec_id  ||
809               tsk->self_exec_id != tsk->parent_exec_id)
810             && !capable(CAP_KILL))
811                 tsk->exit_signal = SIGCHLD;
812
813
814         /* If something other than our normal parent is ptracing us, then
815          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
816          * only has special meaning to our real parent.
817          */
818         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
819                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
820                 do_notify_parent(tsk, signal);
821         } else if (tsk->ptrace) {
822                 do_notify_parent(tsk, SIGCHLD);
823         }
824
825         state = EXIT_ZOMBIE;
826         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
827                 state = EXIT_DEAD;
828         tsk->exit_state = state;
829
830         write_unlock_irq(&tasklist_lock);
831
832         list_for_each_safe(_p, _n, &ptrace_dead) {
833                 list_del_init(_p);
834                 t = list_entry(_p, struct task_struct, ptrace_list);
835                 release_task(t);
836         }
837
838         /* If the process is dead, release it - nobody will wait for it */
839         if (state == EXIT_DEAD)
840                 release_task(tsk);
841 }
842
843 #ifdef CONFIG_DEBUG_STACK_USAGE
844 static void check_stack_usage(void)
845 {
846         static DEFINE_SPINLOCK(low_water_lock);
847         static int lowest_to_date = THREAD_SIZE;
848         unsigned long *n = end_of_stack(current);
849         unsigned long free;
850
851         while (*n == 0)
852                 n++;
853         free = (unsigned long)n - (unsigned long)end_of_stack(current);
854
855         if (free >= lowest_to_date)
856                 return;
857
858         spin_lock(&low_water_lock);
859         if (free < lowest_to_date) {
860                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
861                                 "left\n",
862                                 current->comm, free);
863                 lowest_to_date = free;
864         }
865         spin_unlock(&low_water_lock);
866 }
867 #else
868 static inline void check_stack_usage(void) {}
869 #endif
870
871 static inline void exit_child_reaper(struct task_struct *tsk)
872 {
873         if (likely(tsk->group_leader != child_reaper(tsk)))
874                 return;
875
876         panic("Attempted to kill init!");
877 }
878
879 fastcall NORET_TYPE void do_exit(long code)
880 {
881         struct task_struct *tsk = current;
882         int group_dead;
883
884         profile_task_exit(tsk);
885
886         WARN_ON(atomic_read(&tsk->fs_excl));
887
888         if (unlikely(in_interrupt()))
889                 panic("Aiee, killing interrupt handler!");
890         if (unlikely(!tsk->pid))
891                 panic("Attempted to kill the idle task!");
892
893         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
894                 current->ptrace_message = code;
895                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
896         }
897
898         /*
899          * We're taking recursive faults here in do_exit. Safest is to just
900          * leave this task alone and wait for reboot.
901          */
902         if (unlikely(tsk->flags & PF_EXITING)) {
903                 printk(KERN_ALERT
904                         "Fixing recursive fault but reboot is needed!\n");
905                 /*
906                  * We can do this unlocked here. The futex code uses
907                  * this flag just to verify whether the pi state
908                  * cleanup has been done or not. In the worst case it
909                  * loops once more. We pretend that the cleanup was
910                  * done as there is no way to return. Either the
911                  * OWNER_DIED bit is set by now or we push the blocked
912                  * task into the wait for ever nirwana as well.
913                  */
914                 tsk->flags |= PF_EXITPIDONE;
915                 if (tsk->io_context)
916                         exit_io_context();
917                 set_current_state(TASK_UNINTERRUPTIBLE);
918                 schedule();
919         }
920
921         tsk->flags |= PF_EXITING;
922         /*
923          * tsk->flags are checked in the futex code to protect against
924          * an exiting task cleaning up the robust pi futexes.
925          */
926         smp_mb();
927         spin_unlock_wait(&tsk->pi_lock);
928
929         if (unlikely(in_atomic()))
930                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
931                                 current->comm, current->pid,
932                                 preempt_count());
933
934         acct_update_integrals(tsk);
935         if (tsk->mm) {
936                 update_hiwater_rss(tsk->mm);
937                 update_hiwater_vm(tsk->mm);
938         }
939         group_dead = atomic_dec_and_test(&tsk->signal->live);
940         if (group_dead) {
941                 exit_child_reaper(tsk);
942                 hrtimer_cancel(&tsk->signal->real_timer);
943                 exit_itimers(tsk->signal);
944         }
945         acct_collect(code, group_dead);
946         if (unlikely(tsk->robust_list))
947                 exit_robust_list(tsk);
948 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
949         if (unlikely(tsk->compat_robust_list))
950                 compat_exit_robust_list(tsk);
951 #endif
952         if (group_dead)
953                 tty_audit_exit();
954         if (unlikely(tsk->audit_context))
955                 audit_free(tsk);
956
957         tsk->exit_code = code;
958         taskstats_exit(tsk, group_dead);
959
960         exit_mm(tsk);
961
962         if (group_dead)
963                 acct_process();
964         exit_sem(tsk);
965         __exit_files(tsk);
966         __exit_fs(tsk);
967         check_stack_usage();
968         exit_thread();
969         cpuset_exit(tsk);
970         exit_keys(tsk);
971
972         if (group_dead && tsk->signal->leader)
973                 disassociate_ctty(1);
974
975         module_put(task_thread_info(tsk)->exec_domain->module);
976         if (tsk->binfmt)
977                 module_put(tsk->binfmt->module);
978
979         proc_exit_connector(tsk);
980         exit_task_namespaces(tsk);
981         exit_notify(tsk);
982 #ifdef CONFIG_NUMA
983         mpol_free(tsk->mempolicy);
984         tsk->mempolicy = NULL;
985 #endif
986         /*
987          * This must happen late, after the PID is not
988          * hashed anymore:
989          */
990         if (unlikely(!list_empty(&tsk->pi_state_list)))
991                 exit_pi_state_list(tsk);
992         if (unlikely(current->pi_state_cache))
993                 kfree(current->pi_state_cache);
994         /*
995          * Make sure we are holding no locks:
996          */
997         debug_check_no_locks_held(tsk);
998         /*
999          * We can do this unlocked here. The futex code uses this flag
1000          * just to verify whether the pi state cleanup has been done
1001          * or not. In the worst case it loops once more.
1002          */
1003         tsk->flags |= PF_EXITPIDONE;
1004
1005         if (tsk->io_context)
1006                 exit_io_context();
1007
1008         if (tsk->splice_pipe)
1009                 __free_pipe_info(tsk->splice_pipe);
1010
1011         preempt_disable();
1012         /* causes final put_task_struct in finish_task_switch(). */
1013         tsk->state = TASK_DEAD;
1014
1015         schedule();
1016         BUG();
1017         /* Avoid "noreturn function does return".  */
1018         for (;;)
1019                 cpu_relax();    /* For when BUG is null */
1020 }
1021
1022 EXPORT_SYMBOL_GPL(do_exit);
1023
1024 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1025 {
1026         if (comp)
1027                 complete(comp);
1028
1029         do_exit(code);
1030 }
1031
1032 EXPORT_SYMBOL(complete_and_exit);
1033
1034 asmlinkage long sys_exit(int error_code)
1035 {
1036         do_exit((error_code&0xff)<<8);
1037 }
1038
1039 /*
1040  * Take down every thread in the group.  This is called by fatal signals
1041  * as well as by sys_exit_group (below).
1042  */
1043 NORET_TYPE void
1044 do_group_exit(int exit_code)
1045 {
1046         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1047
1048         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1049                 exit_code = current->signal->group_exit_code;
1050         else if (!thread_group_empty(current)) {
1051                 struct signal_struct *const sig = current->signal;
1052                 struct sighand_struct *const sighand = current->sighand;
1053                 spin_lock_irq(&sighand->siglock);
1054                 if (sig->flags & SIGNAL_GROUP_EXIT)
1055                         /* Another thread got here before we took the lock.  */
1056                         exit_code = sig->group_exit_code;
1057                 else {
1058                         sig->group_exit_code = exit_code;
1059                         zap_other_threads(current);
1060                 }
1061                 spin_unlock_irq(&sighand->siglock);
1062         }
1063
1064         do_exit(exit_code);
1065         /* NOTREACHED */
1066 }
1067
1068 /*
1069  * this kills every thread in the thread group. Note that any externally
1070  * wait4()-ing process will get the correct exit code - even if this
1071  * thread is not the thread group leader.
1072  */
1073 asmlinkage void sys_exit_group(int error_code)
1074 {
1075         do_group_exit((error_code & 0xff) << 8);
1076 }
1077
1078 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1079 {
1080         int err;
1081
1082         if (pid > 0) {
1083                 if (p->pid != pid)
1084                         return 0;
1085         } else if (!pid) {
1086                 if (process_group(p) != process_group(current))
1087                         return 0;
1088         } else if (pid != -1) {
1089                 if (process_group(p) != -pid)
1090                         return 0;
1091         }
1092
1093         /*
1094          * Do not consider detached threads that are
1095          * not ptraced:
1096          */
1097         if (p->exit_signal == -1 && !p->ptrace)
1098                 return 0;
1099
1100         /* Wait for all children (clone and not) if __WALL is set;
1101          * otherwise, wait for clone children *only* if __WCLONE is
1102          * set; otherwise, wait for non-clone children *only*.  (Note:
1103          * A "clone" child here is one that reports to its parent
1104          * using a signal other than SIGCHLD.) */
1105         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1106             && !(options & __WALL))
1107                 return 0;
1108         /*
1109          * Do not consider thread group leaders that are
1110          * in a non-empty thread group:
1111          */
1112         if (delay_group_leader(p))
1113                 return 2;
1114
1115         err = security_task_wait(p);
1116         if (err)
1117                 return err;
1118
1119         return 1;
1120 }
1121
1122 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1123                                int why, int status,
1124                                struct siginfo __user *infop,
1125                                struct rusage __user *rusagep)
1126 {
1127         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1128
1129         put_task_struct(p);
1130         if (!retval)
1131                 retval = put_user(SIGCHLD, &infop->si_signo);
1132         if (!retval)
1133                 retval = put_user(0, &infop->si_errno);
1134         if (!retval)
1135                 retval = put_user((short)why, &infop->si_code);
1136         if (!retval)
1137                 retval = put_user(pid, &infop->si_pid);
1138         if (!retval)
1139                 retval = put_user(uid, &infop->si_uid);
1140         if (!retval)
1141                 retval = put_user(status, &infop->si_status);
1142         if (!retval)
1143                 retval = pid;
1144         return retval;
1145 }
1146
1147 /*
1148  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1149  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1150  * the lock and this task is uninteresting.  If we return nonzero, we have
1151  * released the lock and the system call should return.
1152  */
1153 static int wait_task_zombie(struct task_struct *p, int noreap,
1154                             struct siginfo __user *infop,
1155                             int __user *stat_addr, struct rusage __user *ru)
1156 {
1157         unsigned long state;
1158         int retval, status, traced;
1159
1160         if (unlikely(noreap)) {
1161                 pid_t pid = p->pid;
1162                 uid_t uid = p->uid;
1163                 int exit_code = p->exit_code;
1164                 int why, status;
1165
1166                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1167                         return 0;
1168                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1169                         return 0;
1170                 get_task_struct(p);
1171                 read_unlock(&tasklist_lock);
1172                 if ((exit_code & 0x7f) == 0) {
1173                         why = CLD_EXITED;
1174                         status = exit_code >> 8;
1175                 } else {
1176                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1177                         status = exit_code & 0x7f;
1178                 }
1179                 return wait_noreap_copyout(p, pid, uid, why,
1180                                            status, infop, ru);
1181         }
1182
1183         /*
1184          * Try to move the task's state to DEAD
1185          * only one thread is allowed to do this:
1186          */
1187         state = xchg(&p->exit_state, EXIT_DEAD);
1188         if (state != EXIT_ZOMBIE) {
1189                 BUG_ON(state != EXIT_DEAD);
1190                 return 0;
1191         }
1192
1193         /* traced means p->ptrace, but not vice versa */
1194         traced = (p->real_parent != p->parent);
1195
1196         if (likely(!traced)) {
1197                 struct signal_struct *psig;
1198                 struct signal_struct *sig;
1199
1200                 /*
1201                  * The resource counters for the group leader are in its
1202                  * own task_struct.  Those for dead threads in the group
1203                  * are in its signal_struct, as are those for the child
1204                  * processes it has previously reaped.  All these
1205                  * accumulate in the parent's signal_struct c* fields.
1206                  *
1207                  * We don't bother to take a lock here to protect these
1208                  * p->signal fields, because they are only touched by
1209                  * __exit_signal, which runs with tasklist_lock
1210                  * write-locked anyway, and so is excluded here.  We do
1211                  * need to protect the access to p->parent->signal fields,
1212                  * as other threads in the parent group can be right
1213                  * here reaping other children at the same time.
1214                  */
1215                 spin_lock_irq(&p->parent->sighand->siglock);
1216                 psig = p->parent->signal;
1217                 sig = p->signal;
1218                 psig->cutime =
1219                         cputime_add(psig->cutime,
1220                         cputime_add(p->utime,
1221                         cputime_add(sig->utime,
1222                                     sig->cutime)));
1223                 psig->cstime =
1224                         cputime_add(psig->cstime,
1225                         cputime_add(p->stime,
1226                         cputime_add(sig->stime,
1227                                     sig->cstime)));
1228                 psig->cgtime =
1229                         cputime_add(psig->cgtime,
1230                         cputime_add(p->gtime,
1231                         cputime_add(sig->gtime,
1232                                     sig->cgtime)));
1233                 psig->cmin_flt +=
1234                         p->min_flt + sig->min_flt + sig->cmin_flt;
1235                 psig->cmaj_flt +=
1236                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1237                 psig->cnvcsw +=
1238                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1239                 psig->cnivcsw +=
1240                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1241                 psig->cinblock +=
1242                         task_io_get_inblock(p) +
1243                         sig->inblock + sig->cinblock;
1244                 psig->coublock +=
1245                         task_io_get_oublock(p) +
1246                         sig->oublock + sig->coublock;
1247                 spin_unlock_irq(&p->parent->sighand->siglock);
1248         }
1249
1250         /*
1251          * Now we are sure this task is interesting, and no other
1252          * thread can reap it because we set its state to EXIT_DEAD.
1253          */
1254         read_unlock(&tasklist_lock);
1255
1256         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1257         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1258                 ? p->signal->group_exit_code : p->exit_code;
1259         if (!retval && stat_addr)
1260                 retval = put_user(status, stat_addr);
1261         if (!retval && infop)
1262                 retval = put_user(SIGCHLD, &infop->si_signo);
1263         if (!retval && infop)
1264                 retval = put_user(0, &infop->si_errno);
1265         if (!retval && infop) {
1266                 int why;
1267
1268                 if ((status & 0x7f) == 0) {
1269                         why = CLD_EXITED;
1270                         status >>= 8;
1271                 } else {
1272                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1273                         status &= 0x7f;
1274                 }
1275                 retval = put_user((short)why, &infop->si_code);
1276                 if (!retval)
1277                         retval = put_user(status, &infop->si_status);
1278         }
1279         if (!retval && infop)
1280                 retval = put_user(p->pid, &infop->si_pid);
1281         if (!retval && infop)
1282                 retval = put_user(p->uid, &infop->si_uid);
1283         if (!retval)
1284                 retval = p->pid;
1285
1286         if (traced) {
1287                 write_lock_irq(&tasklist_lock);
1288                 /* We dropped tasklist, ptracer could die and untrace */
1289                 ptrace_unlink(p);
1290                 /*
1291                  * If this is not a detached task, notify the parent.
1292                  * If it's still not detached after that, don't release
1293                  * it now.
1294                  */
1295                 if (p->exit_signal != -1) {
1296                         do_notify_parent(p, p->exit_signal);
1297                         if (p->exit_signal != -1) {
1298                                 p->exit_state = EXIT_ZOMBIE;
1299                                 p = NULL;
1300                         }
1301                 }
1302                 write_unlock_irq(&tasklist_lock);
1303         }
1304         if (p != NULL)
1305                 release_task(p);
1306
1307         return retval;
1308 }
1309
1310 /*
1311  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1312  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1313  * the lock and this task is uninteresting.  If we return nonzero, we have
1314  * released the lock and the system call should return.
1315  */
1316 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1317                              int noreap, struct siginfo __user *infop,
1318                              int __user *stat_addr, struct rusage __user *ru)
1319 {
1320         int retval, exit_code;
1321
1322         if (!p->exit_code)
1323                 return 0;
1324         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1325             p->signal && p->signal->group_stop_count > 0)
1326                 /*
1327                  * A group stop is in progress and this is the group leader.
1328                  * We won't report until all threads have stopped.
1329                  */
1330                 return 0;
1331
1332         /*
1333          * Now we are pretty sure this task is interesting.
1334          * Make sure it doesn't get reaped out from under us while we
1335          * give up the lock and then examine it below.  We don't want to
1336          * keep holding onto the tasklist_lock while we call getrusage and
1337          * possibly take page faults for user memory.
1338          */
1339         get_task_struct(p);
1340         read_unlock(&tasklist_lock);
1341
1342         if (unlikely(noreap)) {
1343                 pid_t pid = p->pid;
1344                 uid_t uid = p->uid;
1345                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1346
1347                 exit_code = p->exit_code;
1348                 if (unlikely(!exit_code) ||
1349                     unlikely(p->state & TASK_TRACED))
1350                         goto bail_ref;
1351                 return wait_noreap_copyout(p, pid, uid,
1352                                            why, (exit_code << 8) | 0x7f,
1353                                            infop, ru);
1354         }
1355
1356         write_lock_irq(&tasklist_lock);
1357
1358         /*
1359          * This uses xchg to be atomic with the thread resuming and setting
1360          * it.  It must also be done with the write lock held to prevent a
1361          * race with the EXIT_ZOMBIE case.
1362          */
1363         exit_code = xchg(&p->exit_code, 0);
1364         if (unlikely(p->exit_state)) {
1365                 /*
1366                  * The task resumed and then died.  Let the next iteration
1367                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1368                  * already be zero here if it resumed and did _exit(0).
1369                  * The task itself is dead and won't touch exit_code again;
1370                  * other processors in this function are locked out.
1371                  */
1372                 p->exit_code = exit_code;
1373                 exit_code = 0;
1374         }
1375         if (unlikely(exit_code == 0)) {
1376                 /*
1377                  * Another thread in this function got to it first, or it
1378                  * resumed, or it resumed and then died.
1379                  */
1380                 write_unlock_irq(&tasklist_lock);
1381 bail_ref:
1382                 put_task_struct(p);
1383                 /*
1384                  * We are returning to the wait loop without having successfully
1385                  * removed the process and having released the lock. We cannot
1386                  * continue, since the "p" task pointer is potentially stale.
1387                  *
1388                  * Return -EAGAIN, and do_wait() will restart the loop from the
1389                  * beginning. Do _not_ re-acquire the lock.
1390                  */
1391                 return -EAGAIN;
1392         }
1393
1394         /* move to end of parent's list to avoid starvation */
1395         remove_parent(p);
1396         add_parent(p);
1397
1398         write_unlock_irq(&tasklist_lock);
1399
1400         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1401         if (!retval && stat_addr)
1402                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1403         if (!retval && infop)
1404                 retval = put_user(SIGCHLD, &infop->si_signo);
1405         if (!retval && infop)
1406                 retval = put_user(0, &infop->si_errno);
1407         if (!retval && infop)
1408                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1409                                           ? CLD_TRAPPED : CLD_STOPPED),
1410                                   &infop->si_code);
1411         if (!retval && infop)
1412                 retval = put_user(exit_code, &infop->si_status);
1413         if (!retval && infop)
1414                 retval = put_user(p->pid, &infop->si_pid);
1415         if (!retval && infop)
1416                 retval = put_user(p->uid, &infop->si_uid);
1417         if (!retval)
1418                 retval = p->pid;
1419         put_task_struct(p);
1420
1421         BUG_ON(!retval);
1422         return retval;
1423 }
1424
1425 /*
1426  * Handle do_wait work for one task in a live, non-stopped state.
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_continued(struct task_struct *p, int noreap,
1432                                struct siginfo __user *infop,
1433                                int __user *stat_addr, struct rusage __user *ru)
1434 {
1435         int retval;
1436         pid_t pid;
1437         uid_t uid;
1438
1439         if (unlikely(!p->signal))
1440                 return 0;
1441
1442         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1443                 return 0;
1444
1445         spin_lock_irq(&p->sighand->siglock);
1446         /* Re-check with the lock held.  */
1447         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1448                 spin_unlock_irq(&p->sighand->siglock);
1449                 return 0;
1450         }
1451         if (!noreap)
1452                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1453         spin_unlock_irq(&p->sighand->siglock);
1454
1455         pid = p->pid;
1456         uid = p->uid;
1457         get_task_struct(p);
1458         read_unlock(&tasklist_lock);
1459
1460         if (!infop) {
1461                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1462                 put_task_struct(p);
1463                 if (!retval && stat_addr)
1464                         retval = put_user(0xffff, stat_addr);
1465                 if (!retval)
1466                         retval = p->pid;
1467         } else {
1468                 retval = wait_noreap_copyout(p, pid, uid,
1469                                              CLD_CONTINUED, SIGCONT,
1470                                              infop, ru);
1471                 BUG_ON(retval == 0);
1472         }
1473
1474         return retval;
1475 }
1476
1477
1478 static inline int my_ptrace_child(struct task_struct *p)
1479 {
1480         if (!(p->ptrace & PT_PTRACED))
1481                 return 0;
1482         if (!(p->ptrace & PT_ATTACHED))
1483                 return 1;
1484         /*
1485          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1486          * we are the attacher.  If we are the real parent, this is a race
1487          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1488          * which we have to switch the parent links, but has already set
1489          * the flags in p->ptrace.
1490          */
1491         return (p->parent != p->real_parent);
1492 }
1493
1494 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1495                     int __user *stat_addr, struct rusage __user *ru)
1496 {
1497         DECLARE_WAITQUEUE(wait, current);
1498         struct task_struct *tsk;
1499         int flag, retval;
1500         int allowed, denied;
1501
1502         add_wait_queue(&current->signal->wait_chldexit,&wait);
1503 repeat:
1504         /*
1505          * We will set this flag if we see any child that might later
1506          * match our criteria, even if we are not able to reap it yet.
1507          */
1508         flag = 0;
1509         allowed = denied = 0;
1510         current->state = TASK_INTERRUPTIBLE;
1511         read_lock(&tasklist_lock);
1512         tsk = current;
1513         do {
1514                 struct task_struct *p;
1515                 struct list_head *_p;
1516                 int ret;
1517
1518                 list_for_each(_p,&tsk->children) {
1519                         p = list_entry(_p, struct task_struct, sibling);
1520
1521                         ret = eligible_child(pid, options, p);
1522                         if (!ret)
1523                                 continue;
1524
1525                         if (unlikely(ret < 0)) {
1526                                 denied = ret;
1527                                 continue;
1528                         }
1529                         allowed = 1;
1530
1531                         switch (p->state) {
1532                         case TASK_TRACED:
1533                                 /*
1534                                  * When we hit the race with PTRACE_ATTACH,
1535                                  * we will not report this child.  But the
1536                                  * race means it has not yet been moved to
1537                                  * our ptrace_children list, so we need to
1538                                  * set the flag here to avoid a spurious ECHILD
1539                                  * when the race happens with the only child.
1540                                  */
1541                                 flag = 1;
1542                                 if (!my_ptrace_child(p))
1543                                         continue;
1544                                 /*FALLTHROUGH*/
1545                         case TASK_STOPPED:
1546                                 /*
1547                                  * It's stopped now, so it might later
1548                                  * continue, exit, or stop again.
1549                                  */
1550                                 flag = 1;
1551                                 if (!(options & WUNTRACED) &&
1552                                     !my_ptrace_child(p))
1553                                         continue;
1554                                 retval = wait_task_stopped(p, ret == 2,
1555                                                            (options & WNOWAIT),
1556                                                            infop,
1557                                                            stat_addr, ru);
1558                                 if (retval == -EAGAIN)
1559                                         goto repeat;
1560                                 if (retval != 0) /* He released the lock.  */
1561                                         goto end;
1562                                 break;
1563                         default:
1564                         // case EXIT_DEAD:
1565                                 if (p->exit_state == EXIT_DEAD)
1566                                         continue;
1567                         // case EXIT_ZOMBIE:
1568                                 if (p->exit_state == EXIT_ZOMBIE) {
1569                                         /*
1570                                          * Eligible but we cannot release
1571                                          * it yet:
1572                                          */
1573                                         if (ret == 2)
1574                                                 goto check_continued;
1575                                         if (!likely(options & WEXITED))
1576                                                 continue;
1577                                         retval = wait_task_zombie(
1578                                                 p, (options & WNOWAIT),
1579                                                 infop, stat_addr, ru);
1580                                         /* He released the lock.  */
1581                                         if (retval != 0)
1582                                                 goto end;
1583                                         break;
1584                                 }
1585 check_continued:
1586                                 /*
1587                                  * It's running now, so it might later
1588                                  * exit, stop, or stop and then continue.
1589                                  */
1590                                 flag = 1;
1591                                 if (!unlikely(options & WCONTINUED))
1592                                         continue;
1593                                 retval = wait_task_continued(
1594                                         p, (options & WNOWAIT),
1595                                         infop, stat_addr, ru);
1596                                 if (retval != 0) /* He released the lock.  */
1597                                         goto end;
1598                                 break;
1599                         }
1600                 }
1601                 if (!flag) {
1602                         list_for_each(_p, &tsk->ptrace_children) {
1603                                 p = list_entry(_p, struct task_struct,
1604                                                 ptrace_list);
1605                                 if (!eligible_child(pid, options, p))
1606                                         continue;
1607                                 flag = 1;
1608                                 break;
1609                         }
1610                 }
1611                 if (options & __WNOTHREAD)
1612                         break;
1613                 tsk = next_thread(tsk);
1614                 BUG_ON(tsk->signal != current->signal);
1615         } while (tsk != current);
1616
1617         read_unlock(&tasklist_lock);
1618         if (flag) {
1619                 retval = 0;
1620                 if (options & WNOHANG)
1621                         goto end;
1622                 retval = -ERESTARTSYS;
1623                 if (signal_pending(current))
1624                         goto end;
1625                 schedule();
1626                 goto repeat;
1627         }
1628         retval = -ECHILD;
1629         if (unlikely(denied) && !allowed)
1630                 retval = denied;
1631 end:
1632         current->state = TASK_RUNNING;
1633         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1634         if (infop) {
1635                 if (retval > 0)
1636                 retval = 0;
1637                 else {
1638                         /*
1639                          * For a WNOHANG return, clear out all the fields
1640                          * we would set so the user can easily tell the
1641                          * difference.
1642                          */
1643                         if (!retval)
1644                                 retval = put_user(0, &infop->si_signo);
1645                         if (!retval)
1646                                 retval = put_user(0, &infop->si_errno);
1647                         if (!retval)
1648                                 retval = put_user(0, &infop->si_code);
1649                         if (!retval)
1650                                 retval = put_user(0, &infop->si_pid);
1651                         if (!retval)
1652                                 retval = put_user(0, &infop->si_uid);
1653                         if (!retval)
1654                                 retval = put_user(0, &infop->si_status);
1655                 }
1656         }
1657         return retval;
1658 }
1659
1660 asmlinkage long sys_waitid(int which, pid_t pid,
1661                            struct siginfo __user *infop, int options,
1662                            struct rusage __user *ru)
1663 {
1664         long ret;
1665
1666         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1667                 return -EINVAL;
1668         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1669                 return -EINVAL;
1670
1671         switch (which) {
1672         case P_ALL:
1673                 pid = -1;
1674                 break;
1675         case P_PID:
1676                 if (pid <= 0)
1677                         return -EINVAL;
1678                 break;
1679         case P_PGID:
1680                 if (pid <= 0)
1681                         return -EINVAL;
1682                 pid = -pid;
1683                 break;
1684         default:
1685                 return -EINVAL;
1686         }
1687
1688         ret = do_wait(pid, options, infop, NULL, ru);
1689
1690         /* avoid REGPARM breakage on x86: */
1691         prevent_tail_call(ret);
1692         return ret;
1693 }
1694
1695 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1696                           int options, struct rusage __user *ru)
1697 {
1698         long ret;
1699
1700         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1701                         __WNOTHREAD|__WCLONE|__WALL))
1702                 return -EINVAL;
1703         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1704
1705         /* avoid REGPARM breakage on x86: */
1706         prevent_tail_call(ret);
1707         return ret;
1708 }
1709
1710 #ifdef __ARCH_WANT_SYS_WAITPID
1711
1712 /*
1713  * sys_waitpid() remains for compatibility. waitpid() should be
1714  * implemented by calling sys_wait4() from libc.a.
1715  */
1716 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1717 {
1718         return sys_wait4(pid, stat_addr, options, NULL);
1719 }
1720
1721 #endif