~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/exit.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/sched/autogroup.h>
 10 #include <linux/sched/mm.h>
 11 #include <linux/sched/stat.h>
 12 #include <linux/sched/task.h>
 13 #include <linux/sched/task_stack.h>
 14 #include <linux/sched/cputime.h>
 15 #include <linux/interrupt.h>
 16 #include <linux/module.h>
 17 #include <linux/capability.h>
 18 #include <linux/completion.h>
 19 #include <linux/personality.h>
 20 #include <linux/tty.h>
 21 #include <linux/iocontext.h>
 22 #include <linux/key.h>
 23 #include <linux/cpu.h>
 24 #include <linux/acct.h>
 25 #include <linux/tsacct_kern.h>
 26 #include <linux/file.h>
 27 #include <linux/fdtable.h>
 28 #include <linux/freezer.h>
 29 #include <linux/binfmts.h>
 30 #include <linux/nsproxy.h>
 31 #include <linux/pid_namespace.h>
 32 #include <linux/ptrace.h>
 33 #include <linux/profile.h>
 34 #include <linux/mount.h>
 35 #include <linux/proc_fs.h>
 36 #include <linux/kthread.h>
 37 #include <linux/mempolicy.h>
 38 #include <linux/taskstats_kern.h>
 39 #include <linux/delayacct.h>
 40 #include <linux/cgroup.h>
 41 #include <linux/syscalls.h>
 42 #include <linux/signal.h>
 43 #include <linux/posix-timers.h>
 44 #include <linux/cn_proc.h>
 45 #include <linux/mutex.h>
 46 #include <linux/futex.h>
 47 #include <linux/pipe_fs_i.h>
 48 #include <linux/audit.h> /* for audit_free() */
 49 #include <linux/resource.h>
 50 #include <linux/blkdev.h>
 51 #include <linux/task_io_accounting_ops.h>
 52 #include <linux/tracehook.h>
 53 #include <linux/fs_struct.h>
 54 #include <linux/userfaultfd_k.h>
 55 #include <linux/init_task.h>
 56 #include <linux/perf_event.h>
 57 #include <trace/events/sched.h>
 58 #include <linux/hw_breakpoint.h>
 59 #include <linux/oom.h>
 60 #include <linux/writeback.h>
 61 #include <linux/shm.h>
 62 #include <linux/kcov.h>
 63 #include <linux/random.h>
 64 #include <linux/rcuwait.h>
 65 
 66 #include <linux/uaccess.h>
 67 #include <asm/unistd.h>
 68 #include <asm/pgtable.h>
 69 #include <asm/mmu_context.h>
 70 
 71 static void __unhash_process(struct task_struct *p, bool group_dead)
 72 {
 73         nr_threads--;
 74         detach_pid(p, PIDTYPE_PID);
 75         if (group_dead) {
 76                 detach_pid(p, PIDTYPE_PGID);
 77                 detach_pid(p, PIDTYPE_SID);
 78 
 79                 list_del_rcu(&p->tasks);
 80                 list_del_init(&p->sibling);
 81                 __this_cpu_dec(process_counts);
 82         }
 83         list_del_rcu(&p->thread_group);
 84         list_del_rcu(&p->thread_node);
 85 }
 86 
 87 /*
 88  * This function expects the tasklist_lock write-locked.
 89  */
 90 static void __exit_signal(struct task_struct *tsk)
 91 {
 92         struct signal_struct *sig = tsk->signal;
 93         bool group_dead = thread_group_leader(tsk);
 94         struct sighand_struct *sighand;
 95         struct tty_struct *uninitialized_var(tty);
 96         u64 utime, stime;
 97 
 98         sighand = rcu_dereference_check(tsk->sighand,
 99                                         lockdep_tasklist_lock_is_held());
100         spin_lock(&sighand->siglock);
101 
102 #ifdef CONFIG_POSIX_TIMERS
103         posix_cpu_timers_exit(tsk);
104         if (group_dead) {
105                 posix_cpu_timers_exit_group(tsk);
106         } else {
107                 /*
108                  * This can only happen if the caller is de_thread().
109                  * FIXME: this is the temporary hack, we should teach
110                  * posix-cpu-timers to handle this case correctly.
111                  */
112                 if (unlikely(has_group_leader_pid(tsk)))
113                         posix_cpu_timers_exit_group(tsk);
114         }
115 #endif
116 
117         if (group_dead) {
118                 tty = sig->tty;
119                 sig->tty = NULL;
120         } else {
121                 /*
122                  * If there is any task waiting for the group exit
123                  * then notify it:
124                  */
125                 if (sig->notify_count > 0 && !--sig->notify_count)
126                         wake_up_process(sig->group_exit_task);
127 
128                 if (tsk == sig->curr_target)
129                         sig->curr_target = next_thread(tsk);
130         }
131 
132         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
133                               sizeof(unsigned long long));
134 
135         /*
136          * Accumulate here the counters for all threads as they die. We could
137          * skip the group leader because it is the last user of signal_struct,
138          * but we want to avoid the race with thread_group_cputime() which can
139          * see the empty ->thread_head list.
140          */
141         task_cputime(tsk, &utime, &stime);
142         write_seqlock(&sig->stats_lock);
143         sig->utime += utime;
144         sig->stime += stime;
145         sig->gtime += task_gtime(tsk);
146         sig->min_flt += tsk->min_flt;
147         sig->maj_flt += tsk->maj_flt;
148         sig->nvcsw += tsk->nvcsw;
149         sig->nivcsw += tsk->nivcsw;
150         sig->inblock += task_io_get_inblock(tsk);
151         sig->oublock += task_io_get_oublock(tsk);
152         task_io_accounting_add(&sig->ioac, &tsk->ioac);
153         sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
154         sig->nr_threads--;
155         __unhash_process(tsk, group_dead);
156         write_sequnlock(&sig->stats_lock);
157 
158         /*
159          * Do this under ->siglock, we can race with another thread
160          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
161          */
162         flush_sigqueue(&tsk->pending);
163         tsk->sighand = NULL;
164         spin_unlock(&sighand->siglock);
165 
166         __cleanup_sighand(sighand);
167         clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
168         if (group_dead) {
169                 flush_sigqueue(&sig->shared_pending);
170                 tty_kref_put(tty);
171         }
172 }
173 
174 static void delayed_put_task_struct(struct rcu_head *rhp)
175 {
176         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
177 
178         perf_event_delayed_put(tsk);
179         trace_sched_process_free(tsk);
180         put_task_struct(tsk);
181 }
182 
183 
184 void release_task(struct task_struct *p)
185 {
186         struct task_struct *leader;
187         int zap_leader;
188 repeat:
189         /* don't need to get the RCU readlock here - the process is dead and
190          * can't be modifying its own credentials. But shut RCU-lockdep up */
191         rcu_read_lock();
192         atomic_dec(&__task_cred(p)->user->processes);
193         rcu_read_unlock();
194 
195         proc_flush_task(p);
196 
197         write_lock_irq(&tasklist_lock);
198         ptrace_release_task(p);
199         __exit_signal(p);
200 
201         /*
202          * If we are the last non-leader member of the thread
203          * group, and the leader is zombie, then notify the
204          * group leader's parent process. (if it wants notification.)
205          */
206         zap_leader = 0;
207         leader = p->group_leader;
208         if (leader != p && thread_group_empty(leader)
209                         && leader->exit_state == EXIT_ZOMBIE) {
210                 /*
211                  * If we were the last child thread and the leader has
212                  * exited already, and the leader's parent ignores SIGCHLD,
213                  * then we are the one who should release the leader.
214                  */
215                 zap_leader = do_notify_parent(leader, leader->exit_signal);
216                 if (zap_leader)
217                         leader->exit_state = EXIT_DEAD;
218         }
219 
220         write_unlock_irq(&tasklist_lock);
221         release_thread(p);
222         call_rcu(&p->rcu, delayed_put_task_struct);
223 
224         p = leader;
225         if (unlikely(zap_leader))
226                 goto repeat;
227 }
228 
229 /*
230  * Note that if this function returns a valid task_struct pointer (!NULL)
231  * task->usage must remain >0 for the duration of the RCU critical section.
232  */
233 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
234 {
235         struct sighand_struct *sighand;
236         struct task_struct *task;
237 
238         /*
239          * We need to verify that release_task() was not called and thus
240          * delayed_put_task_struct() can't run and drop the last reference
241          * before rcu_read_unlock(). We check task->sighand != NULL,
242          * but we can read the already freed and reused memory.
243          */
244 retry:
245         task = rcu_dereference(*ptask);
246         if (!task)
247                 return NULL;
248 
249         probe_kernel_address(&task->sighand, sighand);
250 
251         /*
252          * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
253          * was already freed we can not miss the preceding update of this
254          * pointer.
255          */
256         smp_rmb();
257         if (unlikely(task != READ_ONCE(*ptask)))
258                 goto retry;
259 
260         /*
261          * We've re-checked that "task == *ptask", now we have two different
262          * cases:
263          *
264          * 1. This is actually the same task/task_struct. In this case
265          *    sighand != NULL tells us it is still alive.
266          *
267          * 2. This is another task which got the same memory for task_struct.
268          *    We can't know this of course, and we can not trust
269          *    sighand != NULL.
270          *
271          *    In this case we actually return a random value, but this is
272          *    correct.
273          *
274          *    If we return NULL - we can pretend that we actually noticed that
275          *    *ptask was updated when the previous task has exited. Or pretend
276          *    that probe_slab_address(&sighand) reads NULL.
277          *
278          *    If we return the new task (because sighand is not NULL for any
279          *    reason) - this is fine too. This (new) task can't go away before
280          *    another gp pass.
281          *
282          *    And note: We could even eliminate the false positive if re-read
283          *    task->sighand once again to avoid the falsely NULL. But this case
284          *    is very unlikely so we don't care.
285          */
286         if (!sighand)
287                 return NULL;
288 
289         return task;
290 }
291 
292 void rcuwait_wake_up(struct rcuwait *w)
293 {
294         struct task_struct *task;
295 
296         rcu_read_lock();
297 
298         /*
299          * Order condition vs @task, such that everything prior to the load
300          * of @task is visible. This is the condition as to why the user called
301          * rcuwait_trywake() in the first place. Pairs with set_current_state()
302          * barrier (A) in rcuwait_wait_event().
303          *
304          *    WAIT                WAKE
305          *    [S] tsk = current   [S] cond = true
306          *        MB (A)              MB (B)
307          *    [L] cond            [L] tsk
308          */
309         smp_rmb(); /* (B) */
310 
311         /*
312          * Avoid using task_rcu_dereference() magic as long as we are careful,
313          * see comment in rcuwait_wait_event() regarding ->exit_state.
314          */
315         task = rcu_dereference(w->task);
316         if (task)
317                 wake_up_process(task);
318         rcu_read_unlock();
319 }
320 
321 struct task_struct *try_get_task_struct(struct task_struct **ptask)
322 {
323         struct task_struct *task;
324 
325         rcu_read_lock();
326         task = task_rcu_dereference(ptask);
327         if (task)
328                 get_task_struct(task);
329         rcu_read_unlock();
330 
331         return task;
332 }
333 
334 /*
335  * Determine if a process group is "orphaned", according to the POSIX
336  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
337  * by terminal-generated stop signals.  Newly orphaned process groups are
338  * to receive a SIGHUP and a SIGCONT.
339  *
340  * "I ask you, have you ever known what it is to be an orphan?"
341  */
342 static int will_become_orphaned_pgrp(struct pid *pgrp,
343                                         struct task_struct *ignored_task)
344 {
345         struct task_struct *p;
346 
347         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
348                 if ((p == ignored_task) ||
349                     (p->exit_state && thread_group_empty(p)) ||
350                     is_global_init(p->real_parent))
351                         continue;
352 
353                 if (task_pgrp(p->real_parent) != pgrp &&
354                     task_session(p->real_parent) == task_session(p))
355                         return 0;
356         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
357 
358         return 1;
359 }
360 
361 int is_current_pgrp_orphaned(void)
362 {
363         int retval;
364 
365         read_lock(&tasklist_lock);
366         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
367         read_unlock(&tasklist_lock);
368 
369         return retval;
370 }
371 
372 static bool has_stopped_jobs(struct pid *pgrp)
373 {
374         struct task_struct *p;
375 
376         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
377                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
378                         return true;
379         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
380 
381         return false;
382 }
383 
384 /*
385  * Check to see if any process groups have become orphaned as
386  * a result of our exiting, and if they have any stopped jobs,
387  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
388  */
389 static void
390 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
391 {
392         struct pid *pgrp = task_pgrp(tsk);
393         struct task_struct *ignored_task = tsk;
394 
395         if (!parent)
396                 /* exit: our father is in a different pgrp than
397                  * we are and we were the only connection outside.
398                  */
399                 parent = tsk->real_parent;
400         else
401                 /* reparent: our child is in a different pgrp than
402                  * we are, and it was the only connection outside.
403                  */
404                 ignored_task = NULL;
405 
406         if (task_pgrp(parent) != pgrp &&
407             task_session(parent) == task_session(tsk) &&
408             will_become_orphaned_pgrp(pgrp, ignored_task) &&
409             has_stopped_jobs(pgrp)) {
410                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
411                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
412         }
413 }
414 
415 #ifdef CONFIG_MEMCG
416 /*
417  * A task is exiting.   If it owned this mm, find a new owner for the mm.
418  */
419 void mm_update_next_owner(struct mm_struct *mm)
420 {
421         struct task_struct *c, *g, *p = current;
422 
423 retry:
424         /*
425          * If the exiting or execing task is not the owner, it's
426          * someone else's problem.
427          */
428         if (mm->owner != p)
429                 return;
430         /*
431          * The current owner is exiting/execing and there are no other
432          * candidates.  Do not leave the mm pointing to a possibly
433          * freed task structure.
434          */
435         if (atomic_read(&mm->mm_users) <= 1) {
436                 mm->owner = NULL;
437                 return;
438         }
439 
440         read_lock(&tasklist_lock);
441         /*
442          * Search in the children
443          */
444         list_for_each_entry(c, &p->children, sibling) {
445                 if (c->mm == mm)
446                         goto assign_new_owner;
447         }
448 
449         /*
450          * Search in the siblings
451          */
452         list_for_each_entry(c, &p->real_parent->children, sibling) {
453                 if (c->mm == mm)
454                         goto assign_new_owner;
455         }
456 
457         /*
458          * Search through everything else, we should not get here often.
459          */
460         for_each_process(g) {
461                 if (g->flags & PF_KTHREAD)
462                         continue;
463                 for_each_thread(g, c) {
464                         if (c->mm == mm)
465                                 goto assign_new_owner;
466                         if (c->mm)
467                                 break;
468                 }
469         }
470         read_unlock(&tasklist_lock);
471         /*
472          * We found no owner yet mm_users > 1: this implies that we are
473          * most likely racing with swapoff (try_to_unuse()) or /proc or
474          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
475          */
476         mm->owner = NULL;
477         return;
478 
479 assign_new_owner:
480         BUG_ON(c == p);
481         get_task_struct(c);
482         /*
483          * The task_lock protects c->mm from changing.
484          * We always want mm->owner->mm == mm
485          */
486         task_lock(c);
487         /*
488          * Delay read_unlock() till we have the task_lock()
489          * to ensure that c does not slip away underneath us
490          */
491         read_unlock(&tasklist_lock);
492         if (c->mm != mm) {
493                 task_unlock(c);
494                 put_task_struct(c);
495                 goto retry;
496         }
497         mm->owner = c;
498         task_unlock(c);
499         put_task_struct(c);
500 }
501 #endif /* CONFIG_MEMCG */
502 
503 /*
504  * Turn us into a lazy TLB process if we
505  * aren't already..
506  */
507 static void exit_mm(void)
508 {
509         struct mm_struct *mm = current->mm;
510         struct core_state *core_state;
511 
512         mm_release(current, mm);
513         if (!mm)
514                 return;
515         sync_mm_rss(mm);
516         /*
517          * Serialize with any possible pending coredump.
518          * We must hold mmap_sem around checking core_state
519          * and clearing tsk->mm.  The core-inducing thread
520          * will increment ->nr_threads for each thread in the
521          * group with ->mm != NULL.
522          */
523         down_read(&mm->mmap_sem);
524         core_state = mm->core_state;
525         if (core_state) {
526                 struct core_thread self;
527 
528                 up_read(&mm->mmap_sem);
529 
530                 self.task = current;
531                 self.next = xchg(&core_state->dumper.next, &self);
532                 /*
533                  * Implies mb(), the result of xchg() must be visible
534                  * to core_state->dumper.
535                  */
536                 if (atomic_dec_and_test(&core_state->nr_threads))
537                         complete(&core_state->startup);
538 
539                 for (;;) {
540                         set_current_state(TASK_UNINTERRUPTIBLE);
541                         if (!self.task) /* see coredump_finish() */
542                                 break;
543                         freezable_schedule();
544                 }
545                 __set_current_state(TASK_RUNNING);
546                 down_read(&mm->mmap_sem);
547         }
548         mmgrab(mm);
549         BUG_ON(mm != current->active_mm);
550         /* more a memory barrier than a real lock */
551         task_lock(current);
552         current->mm = NULL;
553         up_read(&mm->mmap_sem);
554         enter_lazy_tlb(mm, current);
555         task_unlock(current);
556         mm_update_next_owner(mm);
557         mmput(mm);
558         if (test_thread_flag(TIF_MEMDIE))
559                 exit_oom_victim();
560 }
561 
562 static struct task_struct *find_alive_thread(struct task_struct *p)
563 {
564         struct task_struct *t;
565 
566         for_each_thread(p, t) {
567                 if (!(t->flags & PF_EXITING))
568                         return t;
569         }
570         return NULL;
571 }
572 
573 static struct task_struct *find_child_reaper(struct task_struct *father)
574         __releases(&tasklist_lock)
575         __acquires(&tasklist_lock)
576 {
577         struct pid_namespace *pid_ns = task_active_pid_ns(father);
578         struct task_struct *reaper = pid_ns->child_reaper;
579 
580         if (likely(reaper != father))
581                 return reaper;
582 
583         reaper = find_alive_thread(father);
584         if (reaper) {
585                 pid_ns->child_reaper = reaper;
586                 return reaper;
587         }
588 
589         write_unlock_irq(&tasklist_lock);
590         if (unlikely(pid_ns == &init_pid_ns)) {
591                 panic("Attempted to kill init! exitcode=0x%08x\n",
592                         father->signal->group_exit_code ?: father->exit_code);
593         }
594         zap_pid_ns_processes(pid_ns);
595         write_lock_irq(&tasklist_lock);
596 
597         return father;
598 }
599 
600 /*
601  * When we die, we re-parent all our children, and try to:
602  * 1. give them to another thread in our thread group, if such a member exists
603  * 2. give it to the first ancestor process which prctl'd itself as a
604  *    child_subreaper for its children (like a service manager)
605  * 3. give it to the init process (PID 1) in our pid namespace
606  */
607 static struct task_struct *find_new_reaper(struct task_struct *father,
608                                            struct task_struct *child_reaper)
609 {
610         struct task_struct *thread, *reaper;
611 
612         thread = find_alive_thread(father);
613         if (thread)
614                 return thread;
615 
616         if (father->signal->has_child_subreaper) {
617                 unsigned int ns_level = task_pid(father)->level;
618                 /*
619                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
620                  * We can't check reaper != child_reaper to ensure we do not
621                  * cross the namespaces, the exiting parent could be injected
622                  * by setns() + fork().
623                  * We check pid->level, this is slightly more efficient than
624                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
625                  */
626                 for (reaper = father->real_parent;
627                      task_pid(reaper)->level == ns_level;
628                      reaper = reaper->real_parent) {
629                         if (reaper == &init_task)
630                                 break;
631                         if (!reaper->signal->is_child_subreaper)
632                                 continue;
633                         thread = find_alive_thread(reaper);
634                         if (thread)
635                                 return thread;
636                 }
637         }
638 
639         return child_reaper;
640 }
641 
642 /*
643 * Any that need to be release_task'd are put on the @dead list.
644  */
645 static void reparent_leader(struct task_struct *father, struct task_struct *p,
646                                 struct list_head *dead)
647 {
648         if (unlikely(p->exit_state == EXIT_DEAD))
649                 return;
650 
651         /* We don't want people slaying init. */
652         p->exit_signal = SIGCHLD;
653 
654         /* If it has exited notify the new parent about this child's death. */
655         if (!p->ptrace &&
656             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
657                 if (do_notify_parent(p, p->exit_signal)) {
658                         p->exit_state = EXIT_DEAD;
659                         list_add(&p->ptrace_entry, dead);
660                 }
661         }
662 
663         kill_orphaned_pgrp(p, father);
664 }
665 
666 /*
667  * This does two things:
668  *
669  * A.  Make init inherit all the child processes
670  * B.  Check to see if any process groups have become orphaned
671  *      as a result of our exiting, and if they have any stopped
672  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
673  */
674 static void forget_original_parent(struct task_struct *father,
675                                         struct list_head *dead)
676 {
677         struct task_struct *p, *t, *reaper;
678 
679         if (unlikely(!list_empty(&father->ptraced)))
680                 exit_ptrace(father, dead);
681 
682         /* Can drop and reacquire tasklist_lock */
683         reaper = find_child_reaper(father);
684         if (list_empty(&father->children))
685                 return;
686 
687         reaper = find_new_reaper(father, reaper);
688         list_for_each_entry(p, &father->children, sibling) {
689                 for_each_thread(p, t) {
690                         t->real_parent = reaper;
691                         BUG_ON((!t->ptrace) != (t->parent == father));
692                         if (likely(!t->ptrace))
693                                 t->parent = t->real_parent;
694                         if (t->pdeath_signal)
695                                 group_send_sig_info(t->pdeath_signal,
696                                                     SEND_SIG_NOINFO, t);
697                 }
698                 /*
699                  * If this is a threaded reparent there is no need to
700                  * notify anyone anything has happened.
701                  */
702                 if (!same_thread_group(reaper, father))
703                         reparent_leader(father, p, dead);
704         }
705         list_splice_tail_init(&father->children, &reaper->children);
706 }
707 
708 /*
709  * Send signals to all our closest relatives so that they know
710  * to properly mourn us..
711  */
712 static void exit_notify(struct task_struct *tsk, int group_dead)
713 {
714         bool autoreap;
715         struct task_struct *p, *n;
716         LIST_HEAD(dead);
717 
718         write_lock_irq(&tasklist_lock);
719         forget_original_parent(tsk, &dead);
720 
721         if (group_dead)
722                 kill_orphaned_pgrp(tsk->group_leader, NULL);
723 
724         if (unlikely(tsk->ptrace)) {
725                 int sig = thread_group_leader(tsk) &&
726                                 thread_group_empty(tsk) &&
727                                 !ptrace_reparented(tsk) ?
728                         tsk->exit_signal : SIGCHLD;
729                 autoreap = do_notify_parent(tsk, sig);
730         } else if (thread_group_leader(tsk)) {
731                 autoreap = thread_group_empty(tsk) &&
732                         do_notify_parent(tsk, tsk->exit_signal);
733         } else {
734                 autoreap = true;
735         }
736 
737         tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
738         if (tsk->exit_state == EXIT_DEAD)
739                 list_add(&tsk->ptrace_entry, &dead);
740 
741         /* mt-exec, de_thread() is waiting for group leader */
742         if (unlikely(tsk->signal->notify_count < 0))
743                 wake_up_process(tsk->signal->group_exit_task);
744         write_unlock_irq(&tasklist_lock);
745 
746         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
747                 list_del_init(&p->ptrace_entry);
748                 release_task(p);
749         }
750 }
751 
752 #ifdef CONFIG_DEBUG_STACK_USAGE
753 static void check_stack_usage(void)
754 {
755         static DEFINE_SPINLOCK(low_water_lock);
756         static int lowest_to_date = THREAD_SIZE;
757         unsigned long free;
758 
759         free = stack_not_used(current);
760 
761         if (free >= lowest_to_date)
762                 return;
763 
764         spin_lock(&low_water_lock);
765         if (free < lowest_to_date) {
766                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
767                         current->comm, task_pid_nr(current), free);
768                 lowest_to_date = free;
769         }
770         spin_unlock(&low_water_lock);
771 }
772 #else
773 static inline void check_stack_usage(void) {}
774 #endif
775 
776 void __noreturn do_exit(long code)
777 {
778         struct task_struct *tsk = current;
779         int group_dead;
780         TASKS_RCU(int tasks_rcu_i);
781 
782         profile_task_exit(tsk);
783         kcov_task_exit(tsk);
784 
785         WARN_ON(blk_needs_flush_plug(tsk));
786 
787         if (unlikely(in_interrupt()))
788                 panic("Aiee, killing interrupt handler!");
789         if (unlikely(!tsk->pid))
790                 panic("Attempted to kill the idle task!");
791 
792         /*
793          * If do_exit is called because this processes oopsed, it's possible
794          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
795          * continuing. Amongst other possible reasons, this is to prevent
796          * mm_release()->clear_child_tid() from writing to a user-controlled
797          * kernel address.
798          */
799         set_fs(USER_DS);
800 
801         ptrace_event(PTRACE_EVENT_EXIT, code);
802 
803         validate_creds_for_do_exit(tsk);
804 
805         /*
806          * We're taking recursive faults here in do_exit. Safest is to just
807          * leave this task alone and wait for reboot.
808          */
809         if (unlikely(tsk->flags & PF_EXITING)) {
810                 pr_alert("Fixing recursive fault but reboot is needed!\n");
811                 /*
812                  * We can do this unlocked here. The futex code uses
813                  * this flag just to verify whether the pi state
814                  * cleanup has been done or not. In the worst case it
815                  * loops once more. We pretend that the cleanup was
816                  * done as there is no way to return. Either the
817                  * OWNER_DIED bit is set by now or we push the blocked
818                  * task into the wait for ever nirwana as well.
819                  */
820                 tsk->flags |= PF_EXITPIDONE;
821                 set_current_state(TASK_UNINTERRUPTIBLE);
822                 schedule();
823         }
824 
825         exit_signals(tsk);  /* sets PF_EXITING */
826         /*
827          * Ensure that all new tsk->pi_lock acquisitions must observe
828          * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
829          */
830         smp_mb();
831         /*
832          * Ensure that we must observe the pi_state in exit_mm() ->
833          * mm_release() -> exit_pi_state_list().
834          */
835         raw_spin_unlock_wait(&tsk->pi_lock);
836 
837         if (unlikely(in_atomic())) {
838                 pr_info("note: %s[%d] exited with preempt_count %d\n",
839                         current->comm, task_pid_nr(current),
840                         preempt_count());
841                 preempt_count_set(PREEMPT_ENABLED);
842         }
843 
844         /* sync mm's RSS info before statistics gathering */
845         if (tsk->mm)
846                 sync_mm_rss(tsk->mm);
847         acct_update_integrals(tsk);
848         group_dead = atomic_dec_and_test(&tsk->signal->live);
849         if (group_dead) {
850 #ifdef CONFIG_POSIX_TIMERS
851                 hrtimer_cancel(&tsk->signal->real_timer);
852                 exit_itimers(tsk->signal);
853 #endif
854                 if (tsk->mm)
855                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
856         }
857         acct_collect(code, group_dead);
858         if (group_dead)
859                 tty_audit_exit();
860         audit_free(tsk);
861 
862         tsk->exit_code = code;
863         taskstats_exit(tsk, group_dead);
864 
865         exit_mm();
866 
867         if (group_dead)
868                 acct_process();
869         trace_sched_process_exit(tsk);
870 
871         exit_sem(tsk);
872         exit_shm(tsk);
873         exit_files(tsk);
874         exit_fs(tsk);
875         if (group_dead)
876                 disassociate_ctty(1);
877         exit_task_namespaces(tsk);
878         exit_task_work(tsk);
879         exit_thread(tsk);
880 
881         /*
882          * Flush inherited counters to the parent - before the parent
883          * gets woken up by child-exit notifications.
884          *
885          * because of cgroup mode, must be called before cgroup_exit()
886          */
887         perf_event_exit_task(tsk);
888 
889         sched_autogroup_exit_task(tsk);
890         cgroup_exit(tsk);
891 
892         /*
893          * FIXME: do that only when needed, using sched_exit tracepoint
894          */
895         flush_ptrace_hw_breakpoint(tsk);
896 
897         TASKS_RCU(preempt_disable());
898         TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
899         TASKS_RCU(preempt_enable());
900         exit_notify(tsk, group_dead);
901         proc_exit_connector(tsk);
902         mpol_put_task_policy(tsk);
903 #ifdef CONFIG_FUTEX
904         if (unlikely(current->pi_state_cache))
905                 kfree(current->pi_state_cache);
906 #endif
907         /*
908          * Make sure we are holding no locks:
909          */
910         debug_check_no_locks_held();
911         /*
912          * We can do this unlocked here. The futex code uses this flag
913          * just to verify whether the pi state cleanup has been done
914          * or not. In the worst case it loops once more.
915          */
916         tsk->flags |= PF_EXITPIDONE;
917 
918         if (tsk->io_context)
919                 exit_io_context(tsk);
920 
921         if (tsk->splice_pipe)
922                 free_pipe_info(tsk->splice_pipe);
923 
924         if (tsk->task_frag.page)
925                 put_page(tsk->task_frag.page);
926 
927         validate_creds_for_do_exit(tsk);
928 
929         check_stack_usage();
930         preempt_disable();
931         if (tsk->nr_dirtied)
932                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
933         exit_rcu();
934         TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
935 
936         do_task_dead();
937 }
938 EXPORT_SYMBOL_GPL(do_exit);
939 
940 void complete_and_exit(struct completion *comp, long code)
941 {
942         if (comp)
943                 complete(comp);
944 
945         do_exit(code);
946 }
947 EXPORT_SYMBOL(complete_and_exit);
948 
949 SYSCALL_DEFINE1(exit, int, error_code)
950 {
951         do_exit((error_code&0xff)<<8);
952 }
953 
954 /*
955  * Take down every thread in the group.  This is called by fatal signals
956  * as well as by sys_exit_group (below).
957  */
958 void
959 do_group_exit(int exit_code)
960 {
961         struct signal_struct *sig = current->signal;
962 
963         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
964 
965         if (signal_group_exit(sig))
966                 exit_code = sig->group_exit_code;
967         else if (!thread_group_empty(current)) {
968                 struct sighand_struct *const sighand = current->sighand;
969 
970                 spin_lock_irq(&sighand->siglock);
971                 if (signal_group_exit(sig))
972                         /* Another thread got here before we took the lock.  */
973                         exit_code = sig->group_exit_code;
974                 else {
975                         sig->group_exit_code = exit_code;
976                         sig->flags = SIGNAL_GROUP_EXIT;
977                         zap_other_threads(current);
978                 }
979                 spin_unlock_irq(&sighand->siglock);
980         }
981 
982         do_exit(exit_code);
983         /* NOTREACHED */
984 }
985 
986 /*
987  * this kills every thread in the thread group. Note that any externally
988  * wait4()-ing process will get the correct exit code - even if this
989  * thread is not the thread group leader.
990  */
991 SYSCALL_DEFINE1(exit_group, int, error_code)
992 {
993         do_group_exit((error_code & 0xff) << 8);
994         /* NOTREACHED */
995         return 0;
996 }
997 
998 struct wait_opts {
999         enum pid_type           wo_type;
1000         int                     wo_flags;
1001         struct pid              *wo_pid;
1002 
1003         struct siginfo __user   *wo_info;
1004         int __user              *wo_stat;
1005         struct rusage __user    *wo_rusage;
1006 
1007         wait_queue_t            child_wait;
1008         int                     notask_error;
1009 };
1010 
1011 static inline
1012 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1013 {
1014         if (type != PIDTYPE_PID)
1015                 task = task->group_leader;
1016         return task->pids[type].pid;
1017 }
1018 
1019 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1020 {
1021         return  wo->wo_type == PIDTYPE_MAX ||
1022                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1023 }
1024 
1025 static int
1026 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1027 {
1028         if (!eligible_pid(wo, p))
1029                 return 0;
1030 
1031         /*
1032          * Wait for all children (clone and not) if __WALL is set or
1033          * if it is traced by us.
1034          */
1035         if (ptrace || (wo->wo_flags & __WALL))
1036                 return 1;
1037 
1038         /*
1039          * Otherwise, wait for clone children *only* if __WCLONE is set;
1040          * otherwise, wait for non-clone children *only*.
1041          *
1042          * Note: a "clone" child here is one that reports to its parent
1043          * using a signal other than SIGCHLD, or a non-leader thread which
1044          * we can only see if it is traced by us.
1045          */
1046         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1047                 return 0;
1048 
1049         return 1;
1050 }
1051 
1052 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1053                                 pid_t pid, uid_t uid, int why, int status)
1054 {
1055         struct siginfo __user *infop;
1056         int retval = wo->wo_rusage
1057                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1058 
1059         put_task_struct(p);
1060         infop = wo->wo_info;
1061         if (infop) {
1062                 if (!retval)
1063                         retval = put_user(SIGCHLD, &infop->si_signo);
1064                 if (!retval)
1065                         retval = put_user(0, &infop->si_errno);
1066                 if (!retval)
1067                         retval = put_user((short)why, &infop->si_code);
1068                 if (!retval)
1069                         retval = put_user(pid, &infop->si_pid);
1070                 if (!retval)
1071                         retval = put_user(uid, &infop->si_uid);
1072                 if (!retval)
1073                         retval = put_user(status, &infop->si_status);
1074         }
1075         if (!retval)
1076                 retval = pid;
1077         return retval;
1078 }
1079 
1080 /*
1081  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1082  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1083  * the lock and this task is uninteresting.  If we return nonzero, we have
1084  * released the lock and the system call should return.
1085  */
1086 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1087 {
1088         int state, retval, status;
1089         pid_t pid = task_pid_vnr(p);
1090         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1091         struct siginfo __user *infop;
1092 
1093         if (!likely(wo->wo_flags & WEXITED))
1094                 return 0;
1095 
1096         if (unlikely(wo->wo_flags & WNOWAIT)) {
1097                 int exit_code = p->exit_code;
1098                 int why;
1099 
1100                 get_task_struct(p);
1101                 read_unlock(&tasklist_lock);
1102                 sched_annotate_sleep();
1103 
1104                 if ((exit_code & 0x7f) == 0) {
1105                         why = CLD_EXITED;
1106                         status = exit_code >> 8;
1107                 } else {
1108                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1109                         status = exit_code & 0x7f;
1110                 }
1111                 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1112         }
1113         /*
1114          * Move the task's state to DEAD/TRACE, only one thread can do this.
1115          */
1116         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1117                 EXIT_TRACE : EXIT_DEAD;
1118         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1119                 return 0;
1120         /*
1121          * We own this thread, nobody else can reap it.
1122          */
1123         read_unlock(&tasklist_lock);
1124         sched_annotate_sleep();
1125 
1126         /*
1127          * Check thread_group_leader() to exclude the traced sub-threads.
1128          */
1129         if (state == EXIT_DEAD && thread_group_leader(p)) {
1130                 struct signal_struct *sig = p->signal;
1131                 struct signal_struct *psig = current->signal;
1132                 unsigned long maxrss;
1133                 u64 tgutime, tgstime;
1134 
1135                 /*
1136                  * The resource counters for the group leader are in its
1137                  * own task_struct.  Those for dead threads in the group
1138                  * are in its signal_struct, as are those for the child
1139                  * processes it has previously reaped.  All these
1140                  * accumulate in the parent's signal_struct c* fields.
1141                  *
1142                  * We don't bother to take a lock here to protect these
1143                  * p->signal fields because the whole thread group is dead
1144                  * and nobody can change them.
1145                  *
1146                  * psig->stats_lock also protects us from our sub-theads
1147                  * which can reap other children at the same time. Until
1148                  * we change k_getrusage()-like users to rely on this lock
1149                  * we have to take ->siglock as well.
1150                  *
1151                  * We use thread_group_cputime_adjusted() to get times for
1152                  * the thread group, which consolidates times for all threads
1153                  * in the group including the group leader.
1154                  */
1155                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1156                 spin_lock_irq(&current->sighand->siglock);
1157                 write_seqlock(&psig->stats_lock);
1158                 psig->cutime += tgutime + sig->cutime;
1159                 psig->cstime += tgstime + sig->cstime;
1160                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1161                 psig->cmin_flt +=
1162                         p->min_flt + sig->min_flt + sig->cmin_flt;
1163                 psig->cmaj_flt +=
1164                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1165                 psig->cnvcsw +=
1166                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1167                 psig->cnivcsw +=
1168                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1169                 psig->cinblock +=
1170                         task_io_get_inblock(p) +
1171                         sig->inblock + sig->cinblock;
1172                 psig->coublock +=
1173                         task_io_get_oublock(p) +
1174                         sig->oublock + sig->coublock;
1175                 maxrss = max(sig->maxrss, sig->cmaxrss);
1176                 if (psig->cmaxrss < maxrss)
1177                         psig->cmaxrss = maxrss;
1178                 task_io_accounting_add(&psig->ioac, &p->ioac);
1179                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1180                 write_sequnlock(&psig->stats_lock);
1181                 spin_unlock_irq(&current->sighand->siglock);
1182         }
1183 
1184         retval = wo->wo_rusage
1185                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1186         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1187                 ? p->signal->group_exit_code : p->exit_code;
1188         if (!retval && wo->wo_stat)
1189                 retval = put_user(status, wo->wo_stat);
1190 
1191         infop = wo->wo_info;
1192         if (!retval && infop)
1193                 retval = put_user(SIGCHLD, &infop->si_signo);
1194         if (!retval && infop)
1195                 retval = put_user(0, &infop->si_errno);
1196         if (!retval && infop) {
1197                 int why;
1198 
1199                 if ((status & 0x7f) == 0) {
1200                         why = CLD_EXITED;
1201                         status >>= 8;
1202                 } else {
1203                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1204                         status &= 0x7f;
1205                 }
1206                 retval = put_user((short)why, &infop->si_code);
1207                 if (!retval)
1208                         retval = put_user(status, &infop->si_status);
1209         }
1210         if (!retval && infop)
1211                 retval = put_user(pid, &infop->si_pid);
1212         if (!retval && infop)
1213                 retval = put_user(uid, &infop->si_uid);
1214         if (!retval)
1215                 retval = pid;
1216 
1217         if (state == EXIT_TRACE) {
1218                 write_lock_irq(&tasklist_lock);
1219                 /* We dropped tasklist, ptracer could die and untrace */
1220                 ptrace_unlink(p);
1221 
1222                 /* If parent wants a zombie, don't release it now */
1223                 state = EXIT_ZOMBIE;
1224                 if (do_notify_parent(p, p->exit_signal))
1225                         state = EXIT_DEAD;
1226                 p->exit_state = state;
1227                 write_unlock_irq(&tasklist_lock);
1228         }
1229         if (state == EXIT_DEAD)
1230                 release_task(p);
1231 
1232         return retval;
1233 }
1234 
1235 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1236 {
1237         if (ptrace) {
1238                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1239                         return &p->exit_code;
1240         } else {
1241                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1242                         return &p->signal->group_exit_code;
1243         }
1244         return NULL;
1245 }
1246 
1247 /**
1248  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1249  * @wo: wait options
1250  * @ptrace: is the wait for ptrace
1251  * @p: task to wait for
1252  *
1253  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1254  *
1255  * CONTEXT:
1256  * read_lock(&tasklist_lock), which is released if return value is
1257  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1258  *
1259  * RETURNS:
1260  * 0 if wait condition didn't exist and search for other wait conditions
1261  * should continue.  Non-zero return, -errno on failure and @p's pid on
1262  * success, implies that tasklist_lock is released and wait condition
1263  * search should terminate.
1264  */
1265 static int wait_task_stopped(struct wait_opts *wo,
1266                                 int ptrace, struct task_struct *p)
1267 {
1268         struct siginfo __user *infop;
1269         int retval, exit_code, *p_code, why;
1270         uid_t uid = 0; /* unneeded, required by compiler */
1271         pid_t pid;
1272 
1273         /*
1274          * Traditionally we see ptrace'd stopped tasks regardless of options.
1275          */
1276         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1277                 return 0;
1278 
1279         if (!task_stopped_code(p, ptrace))
1280                 return 0;
1281 
1282         exit_code = 0;
1283         spin_lock_irq(&p->sighand->siglock);
1284 
1285         p_code = task_stopped_code(p, ptrace);
1286         if (unlikely(!p_code))
1287                 goto unlock_sig;
1288 
1289         exit_code = *p_code;
1290         if (!exit_code)
1291                 goto unlock_sig;
1292 
1293         if (!unlikely(wo->wo_flags & WNOWAIT))
1294                 *p_code = 0;
1295 
1296         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1297 unlock_sig:
1298         spin_unlock_irq(&p->sighand->siglock);
1299         if (!exit_code)
1300                 return 0;
1301 
1302         /*
1303          * Now we are pretty sure this task is interesting.
1304          * Make sure it doesn't get reaped out from under us while we
1305          * give up the lock and then examine it below.  We don't want to
1306          * keep holding onto the tasklist_lock while we call getrusage and
1307          * possibly take page faults for user memory.
1308          */
1309         get_task_struct(p);
1310         pid = task_pid_vnr(p);
1311         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1312         read_unlock(&tasklist_lock);
1313         sched_annotate_sleep();
1314 
1315         if (unlikely(wo->wo_flags & WNOWAIT))
1316                 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1317 
1318         retval = wo->wo_rusage
1319                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1320         if (!retval && wo->wo_stat)
1321                 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1322 
1323         infop = wo->wo_info;
1324         if (!retval && infop)
1325                 retval = put_user(SIGCHLD, &infop->si_signo);
1326         if (!retval && infop)
1327                 retval = put_user(0, &infop->si_errno);
1328         if (!retval && infop)
1329                 retval = put_user((short)why, &infop->si_code);
1330         if (!retval && infop)
1331                 retval = put_user(exit_code, &infop->si_status);
1332         if (!retval && infop)
1333                 retval = put_user(pid, &infop->si_pid);
1334         if (!retval && infop)
1335                 retval = put_user(uid, &infop->si_uid);
1336         if (!retval)
1337                 retval = pid;
1338         put_task_struct(p);
1339 
1340         BUG_ON(!retval);
1341         return retval;
1342 }
1343 
1344 /*
1345  * Handle do_wait work for one task in a live, non-stopped state.
1346  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1347  * the lock and this task is uninteresting.  If we return nonzero, we have
1348  * released the lock and the system call should return.
1349  */
1350 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1351 {
1352         int retval;
1353         pid_t pid;
1354         uid_t uid;
1355 
1356         if (!unlikely(wo->wo_flags & WCONTINUED))
1357                 return 0;
1358 
1359         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1360                 return 0;
1361 
1362         spin_lock_irq(&p->sighand->siglock);
1363         /* Re-check with the lock held.  */
1364         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1365                 spin_unlock_irq(&p->sighand->siglock);
1366                 return 0;
1367         }
1368         if (!unlikely(wo->wo_flags & WNOWAIT))
1369                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1370         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1371         spin_unlock_irq(&p->sighand->siglock);
1372 
1373         pid = task_pid_vnr(p);
1374         get_task_struct(p);
1375         read_unlock(&tasklist_lock);
1376         sched_annotate_sleep();
1377 
1378         if (!wo->wo_info) {
1379                 retval = wo->wo_rusage
1380                         ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1381                 put_task_struct(p);
1382                 if (!retval && wo->wo_stat)
1383                         retval = put_user(0xffff, wo->wo_stat);
1384                 if (!retval)
1385                         retval = pid;
1386         } else {
1387                 retval = wait_noreap_copyout(wo, p, pid, uid,
1388                                              CLD_CONTINUED, SIGCONT);
1389                 BUG_ON(retval == 0);
1390         }
1391 
1392         return retval;
1393 }
1394 
1395 /*
1396  * Consider @p for a wait by @parent.
1397  *
1398  * -ECHILD should be in ->notask_error before the first call.
1399  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1400  * Returns zero if the search for a child should continue;
1401  * then ->notask_error is 0 if @p is an eligible child,
1402  * or still -ECHILD.
1403  */
1404 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1405                                 struct task_struct *p)
1406 {
1407         /*
1408          * We can race with wait_task_zombie() from another thread.
1409          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1410          * can't confuse the checks below.
1411          */
1412         int exit_state = ACCESS_ONCE(p->exit_state);
1413         int ret;
1414 
1415         if (unlikely(exit_state == EXIT_DEAD))
1416                 return 0;
1417 
1418         ret = eligible_child(wo, ptrace, p);
1419         if (!ret)
1420                 return ret;
1421 
1422         if (unlikely(exit_state == EXIT_TRACE)) {
1423                 /*
1424                  * ptrace == 0 means we are the natural parent. In this case
1425                  * we should clear notask_error, debugger will notify us.
1426                  */
1427                 if (likely(!ptrace))
1428                         wo->notask_error = 0;
1429                 return 0;
1430         }
1431 
1432         if (likely(!ptrace) && unlikely(p->ptrace)) {
1433                 /*
1434                  * If it is traced by its real parent's group, just pretend
1435                  * the caller is ptrace_do_wait() and reap this child if it
1436                  * is zombie.
1437                  *
1438                  * This also hides group stop state from real parent; otherwise
1439                  * a single stop can be reported twice as group and ptrace stop.
1440                  * If a ptracer wants to distinguish these two events for its
1441                  * own children it should create a separate process which takes
1442                  * the role of real parent.
1443                  */
1444                 if (!ptrace_reparented(p))
1445                         ptrace = 1;
1446         }
1447 
1448         /* slay zombie? */
1449         if (exit_state == EXIT_ZOMBIE) {
1450                 /* we don't reap group leaders with subthreads */
1451                 if (!delay_group_leader(p)) {
1452                         /*
1453                          * A zombie ptracee is only visible to its ptracer.
1454                          * Notification and reaping will be cascaded to the
1455                          * real parent when the ptracer detaches.
1456                          */
1457                         if (unlikely(ptrace) || likely(!p->ptrace))
1458                                 return wait_task_zombie(wo, p);
1459                 }
1460 
1461                 /*
1462                  * Allow access to stopped/continued state via zombie by
1463                  * falling through.  Clearing of notask_error is complex.
1464                  *
1465                  * When !@ptrace:
1466                  *
1467                  * If WEXITED is set, notask_error should naturally be
1468                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1469                  * so, if there are live subthreads, there are events to
1470                  * wait for.  If all subthreads are dead, it's still safe
1471                  * to clear - this function will be called again in finite
1472                  * amount time once all the subthreads are released and
1473                  * will then return without clearing.
1474                  *
1475                  * When @ptrace:
1476                  *
1477                  * Stopped state is per-task and thus can't change once the
1478                  * target task dies.  Only continued and exited can happen.
1479                  * Clear notask_error if WCONTINUED | WEXITED.
1480                  */
1481                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1482                         wo->notask_error = 0;
1483         } else {
1484                 /*
1485                  * @p is alive and it's gonna stop, continue or exit, so
1486                  * there always is something to wait for.
1487                  */
1488                 wo->notask_error = 0;
1489         }
1490 
1491         /*
1492          * Wait for stopped.  Depending on @ptrace, different stopped state
1493          * is used and the two don't interact with each other.
1494          */
1495         ret = wait_task_stopped(wo, ptrace, p);
1496         if (ret)
1497                 return ret;
1498 
1499         /*
1500          * Wait for continued.  There's only one continued state and the
1501          * ptracer can consume it which can confuse the real parent.  Don't
1502          * use WCONTINUED from ptracer.  You don't need or want it.
1503          */
1504         return wait_task_continued(wo, p);
1505 }
1506 
1507 /*
1508  * Do the work of do_wait() for one thread in the group, @tsk.
1509  *
1510  * -ECHILD should be in ->notask_error before the first call.
1511  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1512  * Returns zero if the search for a child should continue; then
1513  * ->notask_error is 0 if there were any eligible children,
1514  * or still -ECHILD.
1515  */
1516 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1517 {
1518         struct task_struct *p;
1519 
1520         list_for_each_entry(p, &tsk->children, sibling) {
1521                 int ret = wait_consider_task(wo, 0, p);
1522 
1523                 if (ret)
1524                         return ret;
1525         }
1526 
1527         return 0;
1528 }
1529 
1530 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1531 {
1532         struct task_struct *p;
1533 
1534         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1535                 int ret = wait_consider_task(wo, 1, p);
1536 
1537                 if (ret)
1538                         return ret;
1539         }
1540 
1541         return 0;
1542 }
1543 
1544 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1545                                 int sync, void *key)
1546 {
1547         struct wait_opts *wo = container_of(wait, struct wait_opts,
1548                                                 child_wait);
1549         struct task_struct *p = key;
1550 
1551         if (!eligible_pid(wo, p))
1552                 return 0;
1553 
1554         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1555                 return 0;
1556 
1557         return default_wake_function(wait, mode, sync, key);
1558 }
1559 
1560 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1561 {
1562         __wake_up_sync_key(&parent->signal->wait_chldexit,
1563                                 TASK_INTERRUPTIBLE, 1, p);
1564 }
1565 
1566 static long do_wait(struct wait_opts *wo)
1567 {
1568         struct task_struct *tsk;
1569         int retval;
1570 
1571         trace_sched_process_wait(wo->wo_pid);
1572 
1573         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1574         wo->child_wait.private = current;
1575         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1576 repeat:
1577         /*
1578          * If there is nothing that can match our criteria, just get out.
1579          * We will clear ->notask_error to zero if we see any child that
1580          * might later match our criteria, even if we are not able to reap
1581          * it yet.
1582          */
1583         wo->notask_error = -ECHILD;
1584         if ((wo->wo_type < PIDTYPE_MAX) &&
1585            (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1586                 goto notask;
1587 
1588         set_current_state(TASK_INTERRUPTIBLE);
1589         read_lock(&tasklist_lock);
1590         tsk = current;
1591         do {
1592                 retval = do_wait_thread(wo, tsk);
1593                 if (retval)
1594                         goto end;
1595 
1596                 retval = ptrace_do_wait(wo, tsk);
1597                 if (retval)
1598                         goto end;
1599 
1600                 if (wo->wo_flags & __WNOTHREAD)
1601                         break;
1602         } while_each_thread(current, tsk);
1603         read_unlock(&tasklist_lock);
1604 
1605 notask:
1606         retval = wo->notask_error;
1607         if (!retval && !(wo->wo_flags & WNOHANG)) {
1608                 retval = -ERESTARTSYS;
1609                 if (!signal_pending(current)) {
1610                         schedule();
1611                         goto repeat;
1612                 }
1613         }
1614 end:
1615         __set_current_state(TASK_RUNNING);
1616         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1617         return retval;
1618 }
1619 
1620 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1621                 infop, int, options, struct rusage __user *, ru)
1622 {
1623         struct wait_opts wo;
1624         struct pid *pid = NULL;
1625         enum pid_type type;
1626         long ret;
1627 
1628         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1629                         __WNOTHREAD|__WCLONE|__WALL))
1630                 return -EINVAL;
1631         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1632                 return -EINVAL;
1633 
1634         switch (which) {
1635         case P_ALL:
1636                 type = PIDTYPE_MAX;
1637                 break;
1638         case P_PID:
1639                 type = PIDTYPE_PID;
1640                 if (upid <= 0)
1641                         return -EINVAL;
1642                 break;
1643         case P_PGID:
1644                 type = PIDTYPE_PGID;
1645                 if (upid <= 0)
1646                         return -EINVAL;
1647                 break;
1648         default:
1649                 return -EINVAL;
1650         }
1651 
1652         if (type < PIDTYPE_MAX)
1653                 pid = find_get_pid(upid);
1654 
1655         wo.wo_type      = type;
1656         wo.wo_pid       = pid;
1657         wo.wo_flags     = options;
1658         wo.wo_info      = infop;
1659         wo.wo_stat      = NULL;
1660         wo.wo_rusage    = ru;
1661         ret = do_wait(&wo);
1662 
1663         if (ret > 0) {
1664                 ret = 0;
1665         } else if (infop) {
1666                 /*
1667                  * For a WNOHANG return, clear out all the fields
1668                  * we would set so the user can easily tell the
1669                  * difference.
1670                  */
1671                 if (!ret)
1672                         ret = put_user(0, &infop->si_signo);
1673                 if (!ret)
1674                         ret = put_user(0, &infop->si_errno);
1675                 if (!ret)
1676                         ret = put_user(0, &infop->si_code);
1677                 if (!ret)
1678                         ret = put_user(0, &infop->si_pid);
1679                 if (!ret)
1680                         ret = put_user(0, &infop->si_uid);
1681                 if (!ret)
1682                         ret = put_user(0, &infop->si_status);
1683         }
1684 
1685         put_pid(pid);
1686         return ret;
1687 }
1688 
1689 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1690                 int, options, struct rusage __user *, ru)
1691 {
1692         struct wait_opts wo;
1693         struct pid *pid = NULL;
1694         enum pid_type type;
1695         long ret;
1696 
1697         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1698                         __WNOTHREAD|__WCLONE|__WALL))
1699                 return -EINVAL;
1700 
1701         if (upid == -1)
1702                 type = PIDTYPE_MAX;
1703         else if (upid < 0) {
1704                 type = PIDTYPE_PGID;
1705                 pid = find_get_pid(-upid);
1706         } else if (upid == 0) {
1707                 type = PIDTYPE_PGID;
1708                 pid = get_task_pid(current, PIDTYPE_PGID);
1709         } else /* upid > 0 */ {
1710                 type = PIDTYPE_PID;
1711                 pid = find_get_pid(upid);
1712         }
1713 
1714         wo.wo_type      = type;
1715         wo.wo_pid       = pid;
1716         wo.wo_flags     = options | WEXITED;
1717         wo.wo_info      = NULL;
1718         wo.wo_stat      = stat_addr;
1719         wo.wo_rusage    = ru;
1720         ret = do_wait(&wo);
1721         put_pid(pid);
1722 
1723         return ret;
1724 }
1725 
1726 #ifdef __ARCH_WANT_SYS_WAITPID
1727 
1728 /*
1729  * sys_waitpid() remains for compatibility. waitpid() should be
1730  * implemented by calling sys_wait4() from libc.a.
1731  */
1732 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1733 {
1734         return sys_wait4(pid, stat_addr, options, NULL);
1735 }
1736 
1737 #endif
1738 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp