projects/linux·blob v2.6.30-rc4 · refs

1734 lines · 43.1 KiB

1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/interrupt.h>
10#include <linux/module.h>
11#include <linux/capability.h>
12#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
15#include <linux/mnt_namespace.h>
16#include <linux/iocontext.h>
17#include <linux/key.h>
18#include <linux/security.h>
19#include <linux/cpu.h>
20#include <linux/acct.h>
21#include <linux/tsacct_kern.h>
22#include <linux/file.h>
23#include <linux/fdtable.h>
24#include <linux/binfmts.h>
25#include <linux/nsproxy.h>
26#include <linux/pid_namespace.h>
27#include <linux/ptrace.h>
28#include <linux/profile.h>
29#include <linux/mount.h>
30#include <linux/proc_fs.h>
31#include <linux/kthread.h>
32#include <linux/mempolicy.h>
33#include <linux/taskstats_kern.h>
34#include <linux/delayacct.h>
35#include <linux/freezer.h>
36#include <linux/cgroup.h>
37#include <linux/syscalls.h>
38#include <linux/signal.h>
39#include <linux/posix-timers.h>
40#include <linux/cn_proc.h>
41#include <linux/mutex.h>
42#include <linux/futex.h>
43#include <linux/pipe_fs_i.h>
44#include <linux/audit.h> /* for audit_free() */
45#include <linux/resource.h>
46#include <linux/blkdev.h>
47#include <linux/task_io_accounting_ops.h>
48#include <linux/tracehook.h>
49#include <linux/fs_struct.h>
50#include <linux/init_task.h>
51#include <trace/sched.h>
52
53#include <asm/uaccess.h>
54#include <asm/unistd.h>
55#include <asm/pgtable.h>
56#include <asm/mmu_context.h>
57#include "cred-internals.h"
58
59DEFINE_TRACE(sched_process_free);
60DEFINE_TRACE(sched_process_exit);
61DEFINE_TRACE(sched_process_wait);
62
63static void exit_mm(struct task_struct * tsk);
64
65static void __unhash_process(struct task_struct *p)
66{
67 nr_threads--;
68 detach_pid(p, PIDTYPE_PID);
69 if (thread_group_leader(p)) {
70 detach_pid(p, PIDTYPE_PGID);
71 detach_pid(p, PIDTYPE_SID);
72
73 list_del_rcu(&p->tasks);
74 __get_cpu_var(process_counts)--;
75 }
76 list_del_rcu(&p->thread_group);
77 list_del_init(&p->sibling);
78}
79
80/*
81 * This function expects the tasklist_lock write-locked.
82 */
83static void __exit_signal(struct task_struct *tsk)
84{
85 struct signal_struct *sig = tsk->signal;
86 struct sighand_struct *sighand;
87
88 BUG_ON(!sig);
89 BUG_ON(!atomic_read(&sig->count));
90
91 sighand = rcu_dereference(tsk->sighand);
92 spin_lock(&sighand->siglock);
93
94 posix_cpu_timers_exit(tsk);
95 if (atomic_dec_and_test(&sig->count))
96 posix_cpu_timers_exit_group(tsk);
97 else {
98 /*
99 * If there is any task waiting for the group exit
100 * then notify it:
101 */
102 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
103 wake_up_process(sig->group_exit_task);
104
105 if (tsk == sig->curr_target)
106 sig->curr_target = next_thread(tsk);
107 /*
108 * Accumulate here the counters for all threads but the
109 * group leader as they die, so they can be added into
110 * the process-wide totals when those are taken.
111 * The group leader stays around as a zombie as long
112 * as there are other threads. When it gets reaped,
113 * the exit.c code will add its counts into these totals.
114 * We won't ever get here for the group leader, since it
115 * will have been the last reference on the signal_struct.
116 */
117 sig->utime = cputime_add(sig->utime, task_utime(tsk));
118 sig->stime = cputime_add(sig->stime, task_stime(tsk));
119 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
120 sig->min_flt += tsk->min_flt;
121 sig->maj_flt += tsk->maj_flt;
122 sig->nvcsw += tsk->nvcsw;
123 sig->nivcsw += tsk->nivcsw;
124 sig->inblock += task_io_get_inblock(tsk);
125 sig->oublock += task_io_get_oublock(tsk);
126 task_io_accounting_add(&sig->ioac, &tsk->ioac);
127 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
128 sig = NULL; /* Marker for below. */
129 }
130
131 __unhash_process(tsk);
132
133 /*
134 * Do this under ->siglock, we can race with another thread
135 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136 */
137 flush_sigqueue(&tsk->pending);
138
139 tsk->signal = NULL;
140 tsk->sighand = NULL;
141 spin_unlock(&sighand->siglock);
142
143 __cleanup_sighand(sighand);
144 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
145 if (sig) {
146 flush_sigqueue(&sig->shared_pending);
147 taskstats_tgid_free(sig);
148 /*
149 * Make sure ->signal can't go away under rq->lock,
150 * see account_group_exec_runtime().
151 */
152 task_rq_unlock_wait(tsk);
153 __cleanup_signal(sig);
154 }
155}
156
157static void delayed_put_task_struct(struct rcu_head *rhp)
158{
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160
161 trace_sched_process_free(tsk);
162 put_task_struct(tsk);
163}
164
165
166void release_task(struct task_struct * p)
167{
168 struct task_struct *leader;
169 int zap_leader;
170repeat:
171 tracehook_prepare_release_task(p);
172 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials */
174 atomic_dec(&__task_cred(p)->user->processes);
175
176 proc_flush_task(p);
177 write_lock_irq(&tasklist_lock);
178 tracehook_finish_release_task(p);
179 __exit_signal(p);
180
181 /*
182 * If we are the last non-leader member of the thread
183 * group, and the leader is zombie, then notify the
184 * group leader's parent process. (if it wants notification.)
185 */
186 zap_leader = 0;
187 leader = p->group_leader;
188 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
189 BUG_ON(task_detached(leader));
190 do_notify_parent(leader, leader->exit_signal);
191 /*
192 * If we were the last child thread and the leader has
193 * exited already, and the leader's parent ignores SIGCHLD,
194 * then we are the one who should release the leader.
195 *
196 * do_notify_parent() will have marked it self-reaping in
197 * that case.
198 */
199 zap_leader = task_detached(leader);
200
201 /*
202 * This maintains the invariant that release_task()
203 * only runs on a task in EXIT_DEAD, just for sanity.
204 */
205 if (zap_leader)
206 leader->exit_state = EXIT_DEAD;
207 }
208
209 write_unlock_irq(&tasklist_lock);
210 release_thread(p);
211 call_rcu(&p->rcu, delayed_put_task_struct);
212
213 p = leader;
214 if (unlikely(zap_leader))
215 goto repeat;
216}
217
218/*
219 * This checks not only the pgrp, but falls back on the pid if no
220 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
221 * without this...
222 *
223 * The caller must hold rcu lock or the tasklist lock.
224 */
225struct pid *session_of_pgrp(struct pid *pgrp)
226{
227 struct task_struct *p;
228 struct pid *sid = NULL;
229
230 p = pid_task(pgrp, PIDTYPE_PGID);
231 if (p == NULL)
232 p = pid_task(pgrp, PIDTYPE_PID);
233 if (p != NULL)
234 sid = task_session(p);
235
236 return sid;
237}
238
239/*
240 * Determine if a process group is "orphaned", according to the POSIX
241 * definition in 2.2.2.52. Orphaned process groups are not to be affected
242 * by terminal-generated stop signals. Newly orphaned process groups are
243 * to receive a SIGHUP and a SIGCONT.
244 *
245 * "I ask you, have you ever known what it is to be an orphan?"
246 */
247static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
248{
249 struct task_struct *p;
250
251 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
252 if ((p == ignored_task) ||
253 (p->exit_state && thread_group_empty(p)) ||
254 is_global_init(p->real_parent))
255 continue;
256
257 if (task_pgrp(p->real_parent) != pgrp &&
258 task_session(p->real_parent) == task_session(p))
259 return 0;
260 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
261
262 return 1;
263}
264
265int is_current_pgrp_orphaned(void)
266{
267 int retval;
268
269 read_lock(&tasklist_lock);
270 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
271 read_unlock(&tasklist_lock);
272
273 return retval;
274}
275
276static int has_stopped_jobs(struct pid *pgrp)
277{
278 int retval = 0;
279 struct task_struct *p;
280
281 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
282 if (!task_is_stopped(p))
283 continue;
284 retval = 1;
285 break;
286 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
287 return retval;
288}
289
290/*
291 * Check to see if any process groups have become orphaned as
292 * a result of our exiting, and if they have any stopped jobs,
293 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
294 */
295static void
296kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
297{
298 struct pid *pgrp = task_pgrp(tsk);
299 struct task_struct *ignored_task = tsk;
300
301 if (!parent)
302 /* exit: our father is in a different pgrp than
303 * we are and we were the only connection outside.
304 */
305 parent = tsk->real_parent;
306 else
307 /* reparent: our child is in a different pgrp than
308 * we are, and it was the only connection outside.
309 */
310 ignored_task = NULL;
311
312 if (task_pgrp(parent) != pgrp &&
313 task_session(parent) == task_session(tsk) &&
314 will_become_orphaned_pgrp(pgrp, ignored_task) &&
315 has_stopped_jobs(pgrp)) {
316 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
317 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
318 }
319}
320
321/**
322 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
323 *
324 * If a kernel thread is launched as a result of a system call, or if
325 * it ever exits, it should generally reparent itself to kthreadd so it
326 * isn't in the way of other processes and is correctly cleaned up on exit.
327 *
328 * The various task state such as scheduling policy and priority may have
329 * been inherited from a user process, so we reset them to sane values here.
330 *
331 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
332 */
333static void reparent_to_kthreadd(void)
334{
335 write_lock_irq(&tasklist_lock);
336
337 ptrace_unlink(current);
338 /* Reparent to init */
339 current->real_parent = current->parent = kthreadd_task;
340 list_move_tail(&current->sibling, &current->real_parent->children);
341
342 /* Set the exit signal to SIGCHLD so we signal init on exit */
343 current->exit_signal = SIGCHLD;
344
345 if (task_nice(current) < 0)
346 set_user_nice(current, 0);
347 /* cpus_allowed? */
348 /* rt_priority? */
349 /* signals? */
350 memcpy(current->signal->rlim, init_task.signal->rlim,
351 sizeof(current->signal->rlim));
352
353 atomic_inc(&init_cred.usage);
354 commit_creds(&init_cred);
355 write_unlock_irq(&tasklist_lock);
356}
357
358void __set_special_pids(struct pid *pid)
359{
360 struct task_struct *curr = current->group_leader;
361
362 if (task_session(curr) != pid)
363 change_pid(curr, PIDTYPE_SID, pid);
364
365 if (task_pgrp(curr) != pid)
366 change_pid(curr, PIDTYPE_PGID, pid);
367}
368
369static void set_special_pids(struct pid *pid)
370{
371 write_lock_irq(&tasklist_lock);
372 __set_special_pids(pid);
373 write_unlock_irq(&tasklist_lock);
374}
375
376/*
377 * Let kernel threads use this to say that they
378 * allow a certain signal (since daemonize() will
379 * have disabled all of them by default).
380 */
381int allow_signal(int sig)
382{
383 if (!valid_signal(sig) || sig < 1)
384 return -EINVAL;
385
386 spin_lock_irq(&current->sighand->siglock);
387 sigdelset(&current->blocked, sig);
388 if (!current->mm) {
389 /* Kernel threads handle their own signals.
390 Let the signal code know it'll be handled, so
391 that they don't get converted to SIGKILL or
392 just silently dropped */
393 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
394 }
395 recalc_sigpending();
396 spin_unlock_irq(&current->sighand->siglock);
397 return 0;
398}
399
400EXPORT_SYMBOL(allow_signal);
401
402int disallow_signal(int sig)
403{
404 if (!valid_signal(sig) || sig < 1)
405 return -EINVAL;
406
407 spin_lock_irq(&current->sighand->siglock);
408 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
409 recalc_sigpending();
410 spin_unlock_irq(&current->sighand->siglock);
411 return 0;
412}
413
414EXPORT_SYMBOL(disallow_signal);
415
416/*
417 * Put all the gunge required to become a kernel thread without
418 * attached user resources in one place where it belongs.
419 */
420
421void daemonize(const char *name, ...)
422{
423 va_list args;
424 sigset_t blocked;
425
426 va_start(args, name);
427 vsnprintf(current->comm, sizeof(current->comm), name, args);
428 va_end(args);
429
430 /*
431 * If we were started as result of loading a module, close all of the
432 * user space pages. We don't need them, and if we didn't close them
433 * they would be locked into memory.
434 */
435 exit_mm(current);
436 /*
437 * We don't want to have TIF_FREEZE set if the system-wide hibernation
438 * or suspend transition begins right now.
439 */
440 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
441
442 if (current->nsproxy != &init_nsproxy) {
443 get_nsproxy(&init_nsproxy);
444 switch_task_namespaces(current, &init_nsproxy);
445 }
446 set_special_pids(&init_struct_pid);
447 proc_clear_tty(current);
448
449 /* Block and flush all signals */
450 sigfillset(&blocked);
451 sigprocmask(SIG_BLOCK, &blocked, NULL);
452 flush_signals(current);
453
454 /* Become as one with the init task */
455
456 daemonize_fs_struct();
457 exit_files(current);
458 current->files = init_task.files;
459 atomic_inc(&current->files->count);
460
461 reparent_to_kthreadd();
462}
463
464EXPORT_SYMBOL(daemonize);
465
466static void close_files(struct files_struct * files)
467{
468 int i, j;
469 struct fdtable *fdt;
470
471 j = 0;
472
473 /*
474 * It is safe to dereference the fd table without RCU or
475 * ->file_lock because this is the last reference to the
476 * files structure.
477 */
478 fdt = files_fdtable(files);
479 for (;;) {
480 unsigned long set;
481 i = j * __NFDBITS;
482 if (i >= fdt->max_fds)
483 break;
484 set = fdt->open_fds->fds_bits[j++];
485 while (set) {
486 if (set & 1) {
487 struct file * file = xchg(&fdt->fd[i], NULL);
488 if (file) {
489 filp_close(file, files);
490 cond_resched();
491 }
492 }
493 i++;
494 set >>= 1;
495 }
496 }
497}
498
499struct files_struct *get_files_struct(struct task_struct *task)
500{
501 struct files_struct *files;
502
503 task_lock(task);
504 files = task->files;
505 if (files)
506 atomic_inc(&files->count);
507 task_unlock(task);
508
509 return files;
510}
511
512void put_files_struct(struct files_struct *files)
513{
514 struct fdtable *fdt;
515
516 if (atomic_dec_and_test(&files->count)) {
517 close_files(files);
518 /*
519 * Free the fd and fdset arrays if we expanded them.
520 * If the fdtable was embedded, pass files for freeing
521 * at the end of the RCU grace period. Otherwise,
522 * you can free files immediately.
523 */
524 fdt = files_fdtable(files);
525 if (fdt != &files->fdtab)
526 kmem_cache_free(files_cachep, files);
527 free_fdtable(fdt);
528 }
529}
530
531void reset_files_struct(struct files_struct *files)
532{
533 struct task_struct *tsk = current;
534 struct files_struct *old;
535
536 old = tsk->files;
537 task_lock(tsk);
538 tsk->files = files;
539 task_unlock(tsk);
540 put_files_struct(old);
541}
542
543void exit_files(struct task_struct *tsk)
544{
545 struct files_struct * files = tsk->files;
546
547 if (files) {
548 task_lock(tsk);
549 tsk->files = NULL;
550 task_unlock(tsk);
551 put_files_struct(files);
552 }
553}
554
555#ifdef CONFIG_MM_OWNER
556/*
557 * Task p is exiting and it owned mm, lets find a new owner for it
558 */
559static inline int
560mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
561{
562 /*
563 * If there are other users of the mm and the owner (us) is exiting
564 * we need to find a new owner to take on the responsibility.
565 */
566 if (atomic_read(&mm->mm_users) <= 1)
567 return 0;
568 if (mm->owner != p)
569 return 0;
570 return 1;
571}
572
573void mm_update_next_owner(struct mm_struct *mm)
574{
575 struct task_struct *c, *g, *p = current;
576
577retry:
578 if (!mm_need_new_owner(mm, p))
579 return;
580
581 read_lock(&tasklist_lock);
582 /*
583 * Search in the children
584 */
585 list_for_each_entry(c, &p->children, sibling) {
586 if (c->mm == mm)
587 goto assign_new_owner;
588 }
589
590 /*
591 * Search in the siblings
592 */
593 list_for_each_entry(c, &p->parent->children, sibling) {
594 if (c->mm == mm)
595 goto assign_new_owner;
596 }
597
598 /*
599 * Search through everything else. We should not get
600 * here often
601 */
602 do_each_thread(g, c) {
603 if (c->mm == mm)
604 goto assign_new_owner;
605 } while_each_thread(g, c);
606
607 read_unlock(&tasklist_lock);
608 /*
609 * We found no owner yet mm_users > 1: this implies that we are
610 * most likely racing with swapoff (try_to_unuse()) or /proc or
611 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
612 */
613 mm->owner = NULL;
614 return;
615
616assign_new_owner:
617 BUG_ON(c == p);
618 get_task_struct(c);
619 /*
620 * The task_lock protects c->mm from changing.
621 * We always want mm->owner->mm == mm
622 */
623 task_lock(c);
624 /*
625 * Delay read_unlock() till we have the task_lock()
626 * to ensure that c does not slip away underneath us
627 */
628 read_unlock(&tasklist_lock);
629 if (c->mm != mm) {
630 task_unlock(c);
631 put_task_struct(c);
632 goto retry;
633 }
634 mm->owner = c;
635 task_unlock(c);
636 put_task_struct(c);
637}
638#endif /* CONFIG_MM_OWNER */
639
640/*
641 * Turn us into a lazy TLB process if we
642 * aren't already..
643 */
644static void exit_mm(struct task_struct * tsk)
645{
646 struct mm_struct *mm = tsk->mm;
647 struct core_state *core_state;
648
649 mm_release(tsk, mm);
650 if (!mm)
651 return;
652 /*
653 * Serialize with any possible pending coredump.
654 * We must hold mmap_sem around checking core_state
655 * and clearing tsk->mm. The core-inducing thread
656 * will increment ->nr_threads for each thread in the
657 * group with ->mm != NULL.
658 */
659 down_read(&mm->mmap_sem);
660 core_state = mm->core_state;
661 if (core_state) {
662 struct core_thread self;
663 up_read(&mm->mmap_sem);
664
665 self.task = tsk;
666 self.next = xchg(&core_state->dumper.next, &self);
667 /*
668 * Implies mb(), the result of xchg() must be visible
669 * to core_state->dumper.
670 */
671 if (atomic_dec_and_test(&core_state->nr_threads))
672 complete(&core_state->startup);
673
674 for (;;) {
675 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
676 if (!self.task) /* see coredump_finish() */
677 break;
678 schedule();
679 }
680 __set_task_state(tsk, TASK_RUNNING);
681 down_read(&mm->mmap_sem);
682 }
683 atomic_inc(&mm->mm_count);
684 BUG_ON(mm != tsk->active_mm);
685 /* more a memory barrier than a real lock */
686 task_lock(tsk);
687 tsk->mm = NULL;
688 up_read(&mm->mmap_sem);
689 enter_lazy_tlb(mm, current);
690 /* We don't want this task to be frozen prematurely */
691 clear_freeze_flag(tsk);
692 task_unlock(tsk);
693 mm_update_next_owner(mm);
694 mmput(mm);
695}
696
697/*
698 * When we die, we re-parent all our children.
699 * Try to give them to another thread in our thread
700 * group, and if no such member exists, give it to
701 * the child reaper process (ie "init") in our pid
702 * space.
703 */
704static struct task_struct *find_new_reaper(struct task_struct *father)
705{
706 struct pid_namespace *pid_ns = task_active_pid_ns(father);
707 struct task_struct *thread;
708
709 thread = father;
710 while_each_thread(father, thread) {
711 if (thread->flags & PF_EXITING)
712 continue;
713 if (unlikely(pid_ns->child_reaper == father))
714 pid_ns->child_reaper = thread;
715 return thread;
716 }
717
718 if (unlikely(pid_ns->child_reaper == father)) {
719 write_unlock_irq(&tasklist_lock);
720 if (unlikely(pid_ns == &init_pid_ns))
721 panic("Attempted to kill init!");
722
723 zap_pid_ns_processes(pid_ns);
724 write_lock_irq(&tasklist_lock);
725 /*
726 * We can not clear ->child_reaper or leave it alone.
727 * There may by stealth EXIT_DEAD tasks on ->children,
728 * forget_original_parent() must move them somewhere.
729 */
730 pid_ns->child_reaper = init_pid_ns.child_reaper;
731 }
732
733 return pid_ns->child_reaper;
734}
735
736/*
737* Any that need to be release_task'd are put on the @dead list.
738 */
739static void reparent_thread(struct task_struct *father, struct task_struct *p,
740 struct list_head *dead)
741{
742 if (p->pdeath_signal)
743 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
744
745 list_move_tail(&p->sibling, &p->real_parent->children);
746
747 if (task_detached(p))
748 return;
749 /*
750 * If this is a threaded reparent there is no need to
751 * notify anyone anything has happened.
752 */
753 if (same_thread_group(p->real_parent, father))
754 return;
755
756 /* We don't want people slaying init. */
757 p->exit_signal = SIGCHLD;
758
759 /* If it has exited notify the new parent about this child's death. */
760 if (!p->ptrace &&
761 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
762 do_notify_parent(p, p->exit_signal);
763 if (task_detached(p)) {
764 p->exit_state = EXIT_DEAD;
765 list_move_tail(&p->sibling, dead);
766 }
767 }
768
769 kill_orphaned_pgrp(p, father);
770}
771
772static void forget_original_parent(struct task_struct *father)
773{
774 struct task_struct *p, *n, *reaper;
775 LIST_HEAD(dead_children);
776
777 exit_ptrace(father);
778
779 write_lock_irq(&tasklist_lock);
780 reaper = find_new_reaper(father);
781
782 list_for_each_entry_safe(p, n, &father->children, sibling) {
783 p->real_parent = reaper;
784 if (p->parent == father) {
785 BUG_ON(p->ptrace);
786 p->parent = p->real_parent;
787 }
788 reparent_thread(father, p, &dead_children);
789 }
790 write_unlock_irq(&tasklist_lock);
791
792 BUG_ON(!list_empty(&father->children));
793
794 list_for_each_entry_safe(p, n, &dead_children, sibling) {
795 list_del_init(&p->sibling);
796 release_task(p);
797 }
798}
799
800/*
801 * Send signals to all our closest relatives so that they know
802 * to properly mourn us..
803 */
804static void exit_notify(struct task_struct *tsk, int group_dead)
805{
806 int signal;
807 void *cookie;
808
809 /*
810 * This does two things:
811 *
812 * A. Make init inherit all the child processes
813 * B. Check to see if any process groups have become orphaned
814 * as a result of our exiting, and if they have any stopped
815 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
816 */
817 forget_original_parent(tsk);
818 exit_task_namespaces(tsk);
819
820 write_lock_irq(&tasklist_lock);
821 if (group_dead)
822 kill_orphaned_pgrp(tsk->group_leader, NULL);
823
824 /* Let father know we died
825 *
826 * Thread signals are configurable, but you aren't going to use
827 * that to send signals to arbitary processes.
828 * That stops right now.
829 *
830 * If the parent exec id doesn't match the exec id we saved
831 * when we started then we know the parent has changed security
832 * domain.
833 *
834 * If our self_exec id doesn't match our parent_exec_id then
835 * we have changed execution domain as these two values started
836 * the same after a fork.
837 */
838 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
839 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
840 tsk->self_exec_id != tsk->parent_exec_id))
841 tsk->exit_signal = SIGCHLD;
842
843 signal = tracehook_notify_death(tsk, &cookie, group_dead);
844 if (signal >= 0)
845 signal = do_notify_parent(tsk, signal);
846
847 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
848
849 /* mt-exec, de_thread() is waiting for us */
850 if (thread_group_leader(tsk) &&
851 tsk->signal->group_exit_task &&
852 tsk->signal->notify_count < 0)
853 wake_up_process(tsk->signal->group_exit_task);
854
855 write_unlock_irq(&tasklist_lock);
856
857 tracehook_report_death(tsk, signal, cookie, group_dead);
858
859 /* If the process is dead, release it - nobody will wait for it */
860 if (signal == DEATH_REAP)
861 release_task(tsk);
862}
863
864#ifdef CONFIG_DEBUG_STACK_USAGE
865static void check_stack_usage(void)
866{
867 static DEFINE_SPINLOCK(low_water_lock);
868 static int lowest_to_date = THREAD_SIZE;
869 unsigned long free;
870
871 free = stack_not_used(current);
872
873 if (free >= lowest_to_date)
874 return;
875
876 spin_lock(&low_water_lock);
877 if (free < lowest_to_date) {
878 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
879 "left\n",
880 current->comm, free);
881 lowest_to_date = free;
882 }
883 spin_unlock(&low_water_lock);
884}
885#else
886static inline void check_stack_usage(void) {}
887#endif
888
889NORET_TYPE void do_exit(long code)
890{
891 struct task_struct *tsk = current;
892 int group_dead;
893
894 profile_task_exit(tsk);
895
896 WARN_ON(atomic_read(&tsk->fs_excl));
897
898 if (unlikely(in_interrupt()))
899 panic("Aiee, killing interrupt handler!");
900 if (unlikely(!tsk->pid))
901 panic("Attempted to kill the idle task!");
902
903 tracehook_report_exit(&code);
904
905 /*
906 * We're taking recursive faults here in do_exit. Safest is to just
907 * leave this task alone and wait for reboot.
908 */
909 if (unlikely(tsk->flags & PF_EXITING)) {
910 printk(KERN_ALERT
911 "Fixing recursive fault but reboot is needed!\n");
912 /*
913 * We can do this unlocked here. The futex code uses
914 * this flag just to verify whether the pi state
915 * cleanup has been done or not. In the worst case it
916 * loops once more. We pretend that the cleanup was
917 * done as there is no way to return. Either the
918 * OWNER_DIED bit is set by now or we push the blocked
919 * task into the wait for ever nirwana as well.
920 */
921 tsk->flags |= PF_EXITPIDONE;
922 set_current_state(TASK_UNINTERRUPTIBLE);
923 schedule();
924 }
925
926 exit_irq_thread();
927
928 exit_signals(tsk); /* sets PF_EXITING */
929 /*
930 * tsk->flags are checked in the futex code to protect against
931 * an exiting task cleaning up the robust pi futexes.
932 */
933 smp_mb();
934 spin_unlock_wait(&tsk->pi_lock);
935
936 if (unlikely(in_atomic()))
937 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
938 current->comm, task_pid_nr(current),
939 preempt_count());
940
941 acct_update_integrals(tsk);
942
943 group_dead = atomic_dec_and_test(&tsk->signal->live);
944 if (group_dead) {
945 hrtimer_cancel(&tsk->signal->real_timer);
946 exit_itimers(tsk->signal);
947 }
948 acct_collect(code, group_dead);
949 if (group_dead)
950 tty_audit_exit();
951 if (unlikely(tsk->audit_context))
952 audit_free(tsk);
953
954 tsk->exit_code = code;
955 taskstats_exit(tsk, group_dead);
956
957 exit_mm(tsk);
958
959 if (group_dead)
960 acct_process();
961 trace_sched_process_exit(tsk);
962
963 exit_sem(tsk);
964 exit_files(tsk);
965 exit_fs(tsk);
966 check_stack_usage();
967 exit_thread();
968 cgroup_exit(tsk, 1);
969
970 if (group_dead && tsk->signal->leader)
971 disassociate_ctty(1);
972
973 module_put(task_thread_info(tsk)->exec_domain->module);
974 if (tsk->binfmt)
975 module_put(tsk->binfmt->module);
976
977 proc_exit_connector(tsk);
978 exit_notify(tsk, group_dead);
979#ifdef CONFIG_NUMA
980 mpol_put(tsk->mempolicy);
981 tsk->mempolicy = NULL;
982#endif
983#ifdef CONFIG_FUTEX
984 /*
985 * This must happen late, after the PID is not
986 * hashed anymore:
987 */
988 if (unlikely(!list_empty(&tsk->pi_state_list)))
989 exit_pi_state_list(tsk);
990 if (unlikely(current->pi_state_cache))
991 kfree(current->pi_state_cache);
992#endif
993 /*
994 * Make sure we are holding no locks:
995 */
996 debug_check_no_locks_held(tsk);
997 /*
998 * We can do this unlocked here. The futex code uses this flag
999 * just to verify whether the pi state cleanup has been done
1000 * or not. In the worst case it loops once more.
1001 */
1002 tsk->flags |= PF_EXITPIDONE;
1003
1004 if (tsk->io_context)
1005 exit_io_context();
1006
1007 if (tsk->splice_pipe)
1008 __free_pipe_info(tsk->splice_pipe);
1009
1010 preempt_disable();
1011 /* causes final put_task_struct in finish_task_switch(). */
1012 tsk->state = TASK_DEAD;
1013 schedule();
1014 BUG();
1015 /* Avoid "noreturn function does return". */
1016 for (;;)
1017 cpu_relax(); /* For when BUG is null */
1018}
1019
1020EXPORT_SYMBOL_GPL(do_exit);
1021
1022NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1023{
1024 if (comp)
1025 complete(comp);
1026
1027 do_exit(code);
1028}
1029
1030EXPORT_SYMBOL(complete_and_exit);
1031
1032SYSCALL_DEFINE1(exit, int, error_code)
1033{
1034 do_exit((error_code&0xff)<<8);
1035}
1036
1037/*
1038 * Take down every thread in the group. This is called by fatal signals
1039 * as well as by sys_exit_group (below).
1040 */
1041NORET_TYPE void
1042do_group_exit(int exit_code)
1043{
1044 struct signal_struct *sig = current->signal;
1045
1046 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1047
1048 if (signal_group_exit(sig))
1049 exit_code = sig->group_exit_code;
1050 else if (!thread_group_empty(current)) {
1051 struct sighand_struct *const sighand = current->sighand;
1052 spin_lock_irq(&sighand->siglock);
1053 if (signal_group_exit(sig))
1054 /* Another thread got here before we took the lock. */
1055 exit_code = sig->group_exit_code;
1056 else {
1057 sig->group_exit_code = exit_code;
1058 sig->flags = SIGNAL_GROUP_EXIT;
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 */
1073SYSCALL_DEFINE1(exit_group, int, error_code)
1074{
1075 do_group_exit((error_code & 0xff) << 8);
1076 /* NOTREACHED */
1077 return 0;
1078}
1079
1080static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1081{
1082 struct pid *pid = NULL;
1083 if (type == PIDTYPE_PID)
1084 pid = task->pids[type].pid;
1085 else if (type < PIDTYPE_MAX)
1086 pid = task->group_leader->pids[type].pid;
1087 return pid;
1088}
1089
1090static int eligible_child(enum pid_type type, struct pid *pid, int options,
1091 struct task_struct *p)
1092{
1093 int err;
1094
1095 if (type < PIDTYPE_MAX) {
1096 if (task_pid_type(p, type) != pid)
1097 return 0;
1098 }
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 err = security_task_wait(p);
1110 if (err)
1111 return err;
1112
1113 return 1;
1114}
1115
1116static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1117 int why, int status,
1118 struct siginfo __user *infop,
1119 struct rusage __user *rusagep)
1120{
1121 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1122
1123 put_task_struct(p);
1124 if (!retval)
1125 retval = put_user(SIGCHLD, &infop->si_signo);
1126 if (!retval)
1127 retval = put_user(0, &infop->si_errno);
1128 if (!retval)
1129 retval = put_user((short)why, &infop->si_code);
1130 if (!retval)
1131 retval = put_user(pid, &infop->si_pid);
1132 if (!retval)
1133 retval = put_user(uid, &infop->si_uid);
1134 if (!retval)
1135 retval = put_user(status, &infop->si_status);
1136 if (!retval)
1137 retval = pid;
1138 return retval;
1139}
1140
1141/*
1142 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1143 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1144 * the lock and this task is uninteresting. If we return nonzero, we have
1145 * released the lock and the system call should return.
1146 */
1147static int wait_task_zombie(struct task_struct *p, int options,
1148 struct siginfo __user *infop,
1149 int __user *stat_addr, struct rusage __user *ru)
1150{
1151 unsigned long state;
1152 int retval, status, traced;
1153 pid_t pid = task_pid_vnr(p);
1154 uid_t uid = __task_cred(p)->uid;
1155
1156 if (!likely(options & WEXITED))
1157 return 0;
1158
1159 if (unlikely(options & WNOWAIT)) {
1160 int exit_code = p->exit_code;
1161 int why, status;
1162
1163 get_task_struct(p);
1164 read_unlock(&tasklist_lock);
1165 if ((exit_code & 0x7f) == 0) {
1166 why = CLD_EXITED;
1167 status = exit_code >> 8;
1168 } else {
1169 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1170 status = exit_code & 0x7f;
1171 }
1172 return wait_noreap_copyout(p, pid, uid, why,
1173 status, infop, ru);
1174 }
1175
1176 /*
1177 * Try to move the task's state to DEAD
1178 * only one thread is allowed to do this:
1179 */
1180 state = xchg(&p->exit_state, EXIT_DEAD);
1181 if (state != EXIT_ZOMBIE) {
1182 BUG_ON(state != EXIT_DEAD);
1183 return 0;
1184 }
1185
1186 traced = ptrace_reparented(p);
1187
1188 if (likely(!traced)) {
1189 struct signal_struct *psig;
1190 struct signal_struct *sig;
1191 struct task_cputime cputime;
1192
1193 /*
1194 * The resource counters for the group leader are in its
1195 * own task_struct. Those for dead threads in the group
1196 * are in its signal_struct, as are those for the child
1197 * processes it has previously reaped. All these
1198 * accumulate in the parent's signal_struct c* fields.
1199 *
1200 * We don't bother to take a lock here to protect these
1201 * p->signal fields, because they are only touched by
1202 * __exit_signal, which runs with tasklist_lock
1203 * write-locked anyway, and so is excluded here. We do
1204 * need to protect the access to p->parent->signal fields,
1205 * as other threads in the parent group can be right
1206 * here reaping other children at the same time.
1207 *
1208 * We use thread_group_cputime() to get times for the thread
1209 * group, which consolidates times for all threads in the
1210 * group including the group leader.
1211 */
1212 thread_group_cputime(p, &cputime);
1213 spin_lock_irq(&p->parent->sighand->siglock);
1214 psig = p->parent->signal;
1215 sig = p->signal;
1216 psig->cutime =
1217 cputime_add(psig->cutime,
1218 cputime_add(cputime.utime,
1219 sig->cutime));
1220 psig->cstime =
1221 cputime_add(psig->cstime,
1222 cputime_add(cputime.stime,
1223 sig->cstime));
1224 psig->cgtime =
1225 cputime_add(psig->cgtime,
1226 cputime_add(p->gtime,
1227 cputime_add(sig->gtime,
1228 sig->cgtime)));
1229 psig->cmin_flt +=
1230 p->min_flt + sig->min_flt + sig->cmin_flt;
1231 psig->cmaj_flt +=
1232 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1233 psig->cnvcsw +=
1234 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1235 psig->cnivcsw +=
1236 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1237 psig->cinblock +=
1238 task_io_get_inblock(p) +
1239 sig->inblock + sig->cinblock;
1240 psig->coublock +=
1241 task_io_get_oublock(p) +
1242 sig->oublock + sig->coublock;
1243 task_io_accounting_add(&psig->ioac, &p->ioac);
1244 task_io_accounting_add(&psig->ioac, &sig->ioac);
1245 spin_unlock_irq(&p->parent->sighand->siglock);
1246 }
1247
1248 /*
1249 * Now we are sure this task is interesting, and no other
1250 * thread can reap it because we set its state to EXIT_DEAD.
1251 */
1252 read_unlock(&tasklist_lock);
1253
1254 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1255 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1256 ? p->signal->group_exit_code : p->exit_code;
1257 if (!retval && stat_addr)
1258 retval = put_user(status, stat_addr);
1259 if (!retval && infop)
1260 retval = put_user(SIGCHLD, &infop->si_signo);
1261 if (!retval && infop)
1262 retval = put_user(0, &infop->si_errno);
1263 if (!retval && infop) {
1264 int why;
1265
1266 if ((status & 0x7f) == 0) {
1267 why = CLD_EXITED;
1268 status >>= 8;
1269 } else {
1270 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1271 status &= 0x7f;
1272 }
1273 retval = put_user((short)why, &infop->si_code);
1274 if (!retval)
1275 retval = put_user(status, &infop->si_status);
1276 }
1277 if (!retval && infop)
1278 retval = put_user(pid, &infop->si_pid);
1279 if (!retval && infop)
1280 retval = put_user(uid, &infop->si_uid);
1281 if (!retval)
1282 retval = pid;
1283
1284 if (traced) {
1285 write_lock_irq(&tasklist_lock);
1286 /* We dropped tasklist, ptracer could die and untrace */
1287 ptrace_unlink(p);
1288 /*
1289 * If this is not a detached task, notify the parent.
1290 * If it's still not detached after that, don't release
1291 * it now.
1292 */
1293 if (!task_detached(p)) {
1294 do_notify_parent(p, p->exit_signal);
1295 if (!task_detached(p)) {
1296 p->exit_state = EXIT_ZOMBIE;
1297 p = NULL;
1298 }
1299 }
1300 write_unlock_irq(&tasklist_lock);
1301 }
1302 if (p != NULL)
1303 release_task(p);
1304
1305 return retval;
1306}
1307
1308static int *task_stopped_code(struct task_struct *p, bool ptrace)
1309{
1310 if (ptrace) {
1311 if (task_is_stopped_or_traced(p))
1312 return &p->exit_code;
1313 } else {
1314 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1315 return &p->signal->group_exit_code;
1316 }
1317 return NULL;
1318}
1319
1320/*
1321 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1322 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1323 * the lock and this task is uninteresting. If we return nonzero, we have
1324 * released the lock and the system call should return.
1325 */
1326static int wait_task_stopped(int ptrace, struct task_struct *p,
1327 int options, struct siginfo __user *infop,
1328 int __user *stat_addr, struct rusage __user *ru)
1329{
1330 int retval, exit_code, *p_code, why;
1331 uid_t uid = 0; /* unneeded, required by compiler */
1332 pid_t pid;
1333
1334 if (!(options & WUNTRACED))
1335 return 0;
1336
1337 exit_code = 0;
1338 spin_lock_irq(&p->sighand->siglock);
1339
1340 p_code = task_stopped_code(p, ptrace);
1341 if (unlikely(!p_code))
1342 goto unlock_sig;
1343
1344 exit_code = *p_code;
1345 if (!exit_code)
1346 goto unlock_sig;
1347
1348 if (!unlikely(options & WNOWAIT))
1349 *p_code = 0;
1350
1351 /* don't need the RCU readlock here as we're holding a spinlock */
1352 uid = __task_cred(p)->uid;
1353unlock_sig:
1354 spin_unlock_irq(&p->sighand->siglock);
1355 if (!exit_code)
1356 return 0;
1357
1358 /*
1359 * Now we are pretty sure this task is interesting.
1360 * Make sure it doesn't get reaped out from under us while we
1361 * give up the lock and then examine it below. We don't want to
1362 * keep holding onto the tasklist_lock while we call getrusage and
1363 * possibly take page faults for user memory.
1364 */
1365 get_task_struct(p);
1366 pid = task_pid_vnr(p);
1367 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1368 read_unlock(&tasklist_lock);
1369
1370 if (unlikely(options & WNOWAIT))
1371 return wait_noreap_copyout(p, pid, uid,
1372 why, exit_code,
1373 infop, ru);
1374
1375 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1376 if (!retval && stat_addr)
1377 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1378 if (!retval && infop)
1379 retval = put_user(SIGCHLD, &infop->si_signo);
1380 if (!retval && infop)
1381 retval = put_user(0, &infop->si_errno);
1382 if (!retval && infop)
1383 retval = put_user((short)why, &infop->si_code);
1384 if (!retval && infop)
1385 retval = put_user(exit_code, &infop->si_status);
1386 if (!retval && infop)
1387 retval = put_user(pid, &infop->si_pid);
1388 if (!retval && infop)
1389 retval = put_user(uid, &infop->si_uid);
1390 if (!retval)
1391 retval = pid;
1392 put_task_struct(p);
1393
1394 BUG_ON(!retval);
1395 return retval;
1396}
1397
1398/*
1399 * Handle do_wait work for one task in a live, non-stopped state.
1400 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1401 * the lock and this task is uninteresting. If we return nonzero, we have
1402 * released the lock and the system call should return.
1403 */
1404static int wait_task_continued(struct task_struct *p, int options,
1405 struct siginfo __user *infop,
1406 int __user *stat_addr, struct rusage __user *ru)
1407{
1408 int retval;
1409 pid_t pid;
1410 uid_t uid;
1411
1412 if (!unlikely(options & WCONTINUED))
1413 return 0;
1414
1415 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1416 return 0;
1417
1418 spin_lock_irq(&p->sighand->siglock);
1419 /* Re-check with the lock held. */
1420 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1421 spin_unlock_irq(&p->sighand->siglock);
1422 return 0;
1423 }
1424 if (!unlikely(options & WNOWAIT))
1425 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1426 uid = __task_cred(p)->uid;
1427 spin_unlock_irq(&p->sighand->siglock);
1428
1429 pid = task_pid_vnr(p);
1430 get_task_struct(p);
1431 read_unlock(&tasklist_lock);
1432
1433 if (!infop) {
1434 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1435 put_task_struct(p);
1436 if (!retval && stat_addr)
1437 retval = put_user(0xffff, stat_addr);
1438 if (!retval)
1439 retval = pid;
1440 } else {
1441 retval = wait_noreap_copyout(p, pid, uid,
1442 CLD_CONTINUED, SIGCONT,
1443 infop, ru);
1444 BUG_ON(retval == 0);
1445 }
1446
1447 return retval;
1448}
1449
1450/*
1451 * Consider @p for a wait by @parent.
1452 *
1453 * -ECHILD should be in *@notask_error before the first call.
1454 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1455 * Returns zero if the search for a child should continue;
1456 * then *@notask_error is 0 if @p is an eligible child,
1457 * or another error from security_task_wait(), or still -ECHILD.
1458 */
1459static int wait_consider_task(struct task_struct *parent, int ptrace,
1460 struct task_struct *p, int *notask_error,
1461 enum pid_type type, struct pid *pid, int options,
1462 struct siginfo __user *infop,
1463 int __user *stat_addr, struct rusage __user *ru)
1464{
1465 int ret = eligible_child(type, pid, options, p);
1466 if (!ret)
1467 return ret;
1468
1469 if (unlikely(ret < 0)) {
1470 /*
1471 * If we have not yet seen any eligible child,
1472 * then let this error code replace -ECHILD.
1473 * A permission error will give the user a clue
1474 * to look for security policy problems, rather
1475 * than for mysterious wait bugs.
1476 */
1477 if (*notask_error)
1478 *notask_error = ret;
1479 }
1480
1481 if (likely(!ptrace) && unlikely(p->ptrace)) {
1482 /*
1483 * This child is hidden by ptrace.
1484 * We aren't allowed to see it now, but eventually we will.
1485 */
1486 *notask_error = 0;
1487 return 0;
1488 }
1489
1490 if (p->exit_state == EXIT_DEAD)
1491 return 0;
1492
1493 /*
1494 * We don't reap group leaders with subthreads.
1495 */
1496 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1497 return wait_task_zombie(p, options, infop, stat_addr, ru);
1498
1499 /*
1500 * It's stopped or running now, so it might
1501 * later continue, exit, or stop again.
1502 */
1503 *notask_error = 0;
1504
1505 if (task_stopped_code(p, ptrace))
1506 return wait_task_stopped(ptrace, p, options,
1507 infop, stat_addr, ru);
1508
1509 return wait_task_continued(p, options, infop, stat_addr, ru);
1510}
1511
1512/*
1513 * Do the work of do_wait() for one thread in the group, @tsk.
1514 *
1515 * -ECHILD should be in *@notask_error before the first call.
1516 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1517 * Returns zero if the search for a child should continue; then
1518 * *@notask_error is 0 if there were any eligible children,
1519 * or another error from security_task_wait(), or still -ECHILD.
1520 */
1521static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1522 enum pid_type type, struct pid *pid, int options,
1523 struct siginfo __user *infop, int __user *stat_addr,
1524 struct rusage __user *ru)
1525{
1526 struct task_struct *p;
1527
1528 list_for_each_entry(p, &tsk->children, sibling) {
1529 /*
1530 * Do not consider detached threads.
1531 */
1532 if (!task_detached(p)) {
1533 int ret = wait_consider_task(tsk, 0, p, notask_error,
1534 type, pid, options,
1535 infop, stat_addr, ru);
1536 if (ret)
1537 return ret;
1538 }
1539 }
1540
1541 return 0;
1542}
1543
1544static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1545 enum pid_type type, struct pid *pid, int options,
1546 struct siginfo __user *infop, int __user *stat_addr,
1547 struct rusage __user *ru)
1548{
1549 struct task_struct *p;
1550
1551 /*
1552 * Traditionally we see ptrace'd stopped tasks regardless of options.
1553 */
1554 options |= WUNTRACED;
1555
1556 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1557 int ret = wait_consider_task(tsk, 1, p, notask_error,
1558 type, pid, options,
1559 infop, stat_addr, ru);
1560 if (ret)
1561 return ret;
1562 }
1563
1564 return 0;
1565}
1566
1567static long do_wait(enum pid_type type, struct pid *pid, int options,
1568 struct siginfo __user *infop, int __user *stat_addr,
1569 struct rusage __user *ru)
1570{
1571 DECLARE_WAITQUEUE(wait, current);
1572 struct task_struct *tsk;
1573 int retval;
1574
1575 trace_sched_process_wait(pid);
1576
1577 add_wait_queue(&current->signal->wait_chldexit,&wait);
1578repeat:
1579 /*
1580 * If there is nothing that can match our critiera just get out.
1581 * We will clear @retval to zero if we see any child that might later
1582 * match our criteria, even if we are not able to reap it yet.
1583 */
1584 retval = -ECHILD;
1585 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1586 goto end;
1587
1588 current->state = TASK_INTERRUPTIBLE;
1589 read_lock(&tasklist_lock);
1590 tsk = current;
1591 do {
1592 int tsk_result = do_wait_thread(tsk, &retval,
1593 type, pid, options,
1594 infop, stat_addr, ru);
1595 if (!tsk_result)
1596 tsk_result = ptrace_do_wait(tsk, &retval,
1597 type, pid, options,
1598 infop, stat_addr, ru);
1599 if (tsk_result) {
1600 /*
1601 * tasklist_lock is unlocked and we have a final result.
1602 */
1603 retval = tsk_result;
1604 goto end;
1605 }
1606
1607 if (options & __WNOTHREAD)
1608 break;
1609 tsk = next_thread(tsk);
1610 BUG_ON(tsk->signal != current->signal);
1611 } while (tsk != current);
1612 read_unlock(&tasklist_lock);
1613
1614 if (!retval && !(options & WNOHANG)) {
1615 retval = -ERESTARTSYS;
1616 if (!signal_pending(current)) {
1617 schedule();
1618 goto repeat;
1619 }
1620 }
1621
1622end:
1623 current->state = TASK_RUNNING;
1624 remove_wait_queue(&current->signal->wait_chldexit,&wait);
1625 if (infop) {
1626 if (retval > 0)
1627 retval = 0;
1628 else {
1629 /*
1630 * For a WNOHANG return, clear out all the fields
1631 * we would set so the user can easily tell the
1632 * difference.
1633 */
1634 if (!retval)
1635 retval = put_user(0, &infop->si_signo);
1636 if (!retval)
1637 retval = put_user(0, &infop->si_errno);
1638 if (!retval)
1639 retval = put_user(0, &infop->si_code);
1640 if (!retval)
1641 retval = put_user(0, &infop->si_pid);
1642 if (!retval)
1643 retval = put_user(0, &infop->si_uid);
1644 if (!retval)
1645 retval = put_user(0, &infop->si_status);
1646 }
1647 }
1648 return retval;
1649}
1650
1651SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1652 infop, int, options, struct rusage __user *, ru)
1653{
1654 struct pid *pid = NULL;
1655 enum pid_type type;
1656 long ret;
1657
1658 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1659 return -EINVAL;
1660 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1661 return -EINVAL;
1662
1663 switch (which) {
1664 case P_ALL:
1665 type = PIDTYPE_MAX;
1666 break;
1667 case P_PID:
1668 type = PIDTYPE_PID;
1669 if (upid <= 0)
1670 return -EINVAL;
1671 break;
1672 case P_PGID:
1673 type = PIDTYPE_PGID;
1674 if (upid <= 0)
1675 return -EINVAL;
1676 break;
1677 default:
1678 return -EINVAL;
1679 }
1680
1681 if (type < PIDTYPE_MAX)
1682 pid = find_get_pid(upid);
1683 ret = do_wait(type, pid, options, infop, NULL, ru);
1684 put_pid(pid);
1685
1686 /* avoid REGPARM breakage on x86: */
1687 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1688 return ret;
1689}
1690
1691SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1692 int, options, struct rusage __user *, ru)
1693{
1694 struct pid *pid = NULL;
1695 enum pid_type type;
1696 long ret;
1697
1698 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1699 __WNOTHREAD|__WCLONE|__WALL))
1700 return -EINVAL;
1701
1702 if (upid == -1)
1703 type = PIDTYPE_MAX;
1704 else if (upid < 0) {
1705 type = PIDTYPE_PGID;
1706 pid = find_get_pid(-upid);
1707 } else if (upid == 0) {
1708 type = PIDTYPE_PGID;
1709 pid = get_task_pid(current, PIDTYPE_PGID);
1710 } else /* upid > 0 */ {
1711 type = PIDTYPE_PID;
1712 pid = find_get_pid(upid);
1713 }
1714
1715 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1716 put_pid(pid);
1717
1718 /* avoid REGPARM breakage on x86: */
1719 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1720 return ret;
1721}
1722
1723#ifdef __ARCH_WANT_SYS_WAITPID
1724
1725/*
1726 * sys_waitpid() remains for compatibility. waitpid() should be
1727 * implemented by calling sys_wait4() from libc.a.
1728 */
1729SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1730{
1731 return sys_wait4(pid, stat_addr, options, NULL);
1732}
1733
1734#endif